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 unsigned int min_timer_period_us
= 500;
111 module_param(min_timer_period_us
, uint
, S_IRUGO
| S_IWUSR
);
113 static bool __read_mostly kvmclock_periodic_sync
= true;
114 module_param(kvmclock_periodic_sync
, bool, S_IRUGO
);
116 bool __read_mostly kvm_has_tsc_control
;
117 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
118 u32 __read_mostly kvm_max_guest_tsc_khz
;
119 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
120 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits
;
121 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits
);
122 u64 __read_mostly kvm_max_tsc_scaling_ratio
;
123 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio
);
124 u64 __read_mostly kvm_default_tsc_scaling_ratio
;
125 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio
);
127 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
128 static u32 __read_mostly tsc_tolerance_ppm
= 250;
129 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
131 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
132 unsigned int __read_mostly lapic_timer_advance_ns
= 0;
133 module_param(lapic_timer_advance_ns
, uint
, S_IRUGO
| S_IWUSR
);
135 static bool __read_mostly vector_hashing
= true;
136 module_param(vector_hashing
, bool, S_IRUGO
);
138 #define KVM_NR_SHARED_MSRS 16
140 struct kvm_shared_msrs_global
{
142 u32 msrs
[KVM_NR_SHARED_MSRS
];
145 struct kvm_shared_msrs
{
146 struct user_return_notifier urn
;
148 struct kvm_shared_msr_values
{
151 } values
[KVM_NR_SHARED_MSRS
];
154 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
155 static struct kvm_shared_msrs __percpu
*shared_msrs
;
157 struct kvm_stats_debugfs_item debugfs_entries
[] = {
158 { "pf_fixed", VCPU_STAT(pf_fixed
) },
159 { "pf_guest", VCPU_STAT(pf_guest
) },
160 { "tlb_flush", VCPU_STAT(tlb_flush
) },
161 { "invlpg", VCPU_STAT(invlpg
) },
162 { "exits", VCPU_STAT(exits
) },
163 { "io_exits", VCPU_STAT(io_exits
) },
164 { "mmio_exits", VCPU_STAT(mmio_exits
) },
165 { "signal_exits", VCPU_STAT(signal_exits
) },
166 { "irq_window", VCPU_STAT(irq_window_exits
) },
167 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
168 { "halt_exits", VCPU_STAT(halt_exits
) },
169 { "halt_successful_poll", VCPU_STAT(halt_successful_poll
) },
170 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll
) },
171 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid
) },
172 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
173 { "hypercalls", VCPU_STAT(hypercalls
) },
174 { "request_irq", VCPU_STAT(request_irq_exits
) },
175 { "irq_exits", VCPU_STAT(irq_exits
) },
176 { "host_state_reload", VCPU_STAT(host_state_reload
) },
177 { "efer_reload", VCPU_STAT(efer_reload
) },
178 { "fpu_reload", VCPU_STAT(fpu_reload
) },
179 { "insn_emulation", VCPU_STAT(insn_emulation
) },
180 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
181 { "irq_injections", VCPU_STAT(irq_injections
) },
182 { "nmi_injections", VCPU_STAT(nmi_injections
) },
183 { "req_event", VCPU_STAT(req_event
) },
184 { "l1d_flush", VCPU_STAT(l1d_flush
) },
185 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
186 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
187 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
188 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
189 { "mmu_flooded", VM_STAT(mmu_flooded
) },
190 { "mmu_recycled", VM_STAT(mmu_recycled
) },
191 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
192 { "mmu_unsync", VM_STAT(mmu_unsync
) },
193 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
194 { "largepages", VM_STAT(lpages
) },
195 { "max_mmu_page_hash_collisions",
196 VM_STAT(max_mmu_page_hash_collisions
) },
200 u64 __read_mostly host_xcr0
;
202 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
204 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
207 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
208 vcpu
->arch
.apf
.gfns
[i
] = ~0;
211 static void kvm_on_user_return(struct user_return_notifier
*urn
)
214 struct kvm_shared_msrs
*locals
215 = container_of(urn
, struct kvm_shared_msrs
, urn
);
216 struct kvm_shared_msr_values
*values
;
220 * Disabling irqs at this point since the following code could be
221 * interrupted and executed through kvm_arch_hardware_disable()
223 local_irq_save(flags
);
224 if (locals
->registered
) {
225 locals
->registered
= false;
226 user_return_notifier_unregister(urn
);
228 local_irq_restore(flags
);
229 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
230 values
= &locals
->values
[slot
];
231 if (values
->host
!= values
->curr
) {
232 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
233 values
->curr
= values
->host
;
238 static void shared_msr_update(unsigned slot
, u32 msr
)
241 unsigned int cpu
= smp_processor_id();
242 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
244 /* only read, and nobody should modify it at this time,
245 * so don't need lock */
246 if (slot
>= shared_msrs_global
.nr
) {
247 printk(KERN_ERR
"kvm: invalid MSR slot!");
250 rdmsrl_safe(msr
, &value
);
251 smsr
->values
[slot
].host
= value
;
252 smsr
->values
[slot
].curr
= value
;
255 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
257 BUG_ON(slot
>= KVM_NR_SHARED_MSRS
);
258 shared_msrs_global
.msrs
[slot
] = msr
;
259 if (slot
>= shared_msrs_global
.nr
)
260 shared_msrs_global
.nr
= slot
+ 1;
262 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
264 static void kvm_shared_msr_cpu_online(void)
268 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
269 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
272 int kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
274 unsigned int cpu
= smp_processor_id();
275 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
278 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
280 smsr
->values
[slot
].curr
= value
;
281 err
= wrmsrl_safe(shared_msrs_global
.msrs
[slot
], value
);
285 if (!smsr
->registered
) {
286 smsr
->urn
.on_user_return
= kvm_on_user_return
;
287 user_return_notifier_register(&smsr
->urn
);
288 smsr
->registered
= true;
292 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
294 static void drop_user_return_notifiers(void)
296 unsigned int cpu
= smp_processor_id();
297 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
299 if (smsr
->registered
)
300 kvm_on_user_return(&smsr
->urn
);
303 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
305 return vcpu
->arch
.apic_base
;
307 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
309 int kvm_set_apic_base(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
311 u64 old_state
= vcpu
->arch
.apic_base
&
312 (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
);
313 u64 new_state
= msr_info
->data
&
314 (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
);
315 u64 reserved_bits
= ((~0ULL) << cpuid_maxphyaddr(vcpu
)) | 0x2ff |
316 (guest_cpuid_has(vcpu
, X86_FEATURE_X2APIC
) ? 0 : X2APIC_ENABLE
);
318 if ((msr_info
->data
& reserved_bits
) || new_state
== X2APIC_ENABLE
)
320 if (!msr_info
->host_initiated
&&
321 ((new_state
== MSR_IA32_APICBASE_ENABLE
&&
322 old_state
== (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
)) ||
323 (new_state
== (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
) &&
327 kvm_lapic_set_base(vcpu
, msr_info
->data
);
330 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
332 asmlinkage __visible
void kvm_spurious_fault(void)
334 /* Fault while not rebooting. We want the trace. */
337 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
339 #define EXCPT_BENIGN 0
340 #define EXCPT_CONTRIBUTORY 1
343 static int exception_class(int vector
)
353 return EXCPT_CONTRIBUTORY
;
360 #define EXCPT_FAULT 0
362 #define EXCPT_ABORT 2
363 #define EXCPT_INTERRUPT 3
365 static int exception_type(int vector
)
369 if (WARN_ON(vector
> 31 || vector
== NMI_VECTOR
))
370 return EXCPT_INTERRUPT
;
374 /* #DB is trap, as instruction watchpoints are handled elsewhere */
375 if (mask
& ((1 << DB_VECTOR
) | (1 << BP_VECTOR
) | (1 << OF_VECTOR
)))
378 if (mask
& ((1 << DF_VECTOR
) | (1 << MC_VECTOR
)))
381 /* Reserved exceptions will result in fault */
385 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
386 unsigned nr
, bool has_error
, u32 error_code
,
392 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
394 if (!vcpu
->arch
.exception
.pending
&& !vcpu
->arch
.exception
.injected
) {
396 if (has_error
&& !is_protmode(vcpu
))
400 * On vmentry, vcpu->arch.exception.pending is only
401 * true if an event injection was blocked by
402 * nested_run_pending. In that case, however,
403 * vcpu_enter_guest requests an immediate exit,
404 * and the guest shouldn't proceed far enough to
407 WARN_ON_ONCE(vcpu
->arch
.exception
.pending
);
408 vcpu
->arch
.exception
.injected
= true;
410 vcpu
->arch
.exception
.pending
= true;
411 vcpu
->arch
.exception
.injected
= false;
413 vcpu
->arch
.exception
.has_error_code
= has_error
;
414 vcpu
->arch
.exception
.nr
= nr
;
415 vcpu
->arch
.exception
.error_code
= error_code
;
419 /* to check exception */
420 prev_nr
= vcpu
->arch
.exception
.nr
;
421 if (prev_nr
== DF_VECTOR
) {
422 /* triple fault -> shutdown */
423 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
426 class1
= exception_class(prev_nr
);
427 class2
= exception_class(nr
);
428 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
429 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
431 * Generate double fault per SDM Table 5-5. Set
432 * exception.pending = true so that the double fault
433 * can trigger a nested vmexit.
435 vcpu
->arch
.exception
.pending
= true;
436 vcpu
->arch
.exception
.injected
= false;
437 vcpu
->arch
.exception
.has_error_code
= true;
438 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
439 vcpu
->arch
.exception
.error_code
= 0;
441 /* replace previous exception with a new one in a hope
442 that instruction re-execution will regenerate lost
447 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
449 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
451 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
453 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
455 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
457 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
459 int kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
462 kvm_inject_gp(vcpu
, 0);
464 return kvm_skip_emulated_instruction(vcpu
);
468 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
470 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
472 ++vcpu
->stat
.pf_guest
;
473 vcpu
->arch
.exception
.nested_apf
=
474 is_guest_mode(vcpu
) && fault
->async_page_fault
;
475 if (vcpu
->arch
.exception
.nested_apf
)
476 vcpu
->arch
.apf
.nested_apf_token
= fault
->address
;
478 vcpu
->arch
.cr2
= fault
->address
;
479 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
481 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
483 static bool kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
485 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
486 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
488 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
490 return fault
->nested_page_fault
;
493 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
495 atomic_inc(&vcpu
->arch
.nmi_queued
);
496 kvm_make_request(KVM_REQ_NMI
, vcpu
);
498 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
500 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
502 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
504 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
506 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
508 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
510 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
513 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
514 * a #GP and return false.
516 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
518 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
520 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
523 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
525 bool kvm_require_dr(struct kvm_vcpu
*vcpu
, int dr
)
527 if ((dr
!= 4 && dr
!= 5) || !kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
530 kvm_queue_exception(vcpu
, UD_VECTOR
);
533 EXPORT_SYMBOL_GPL(kvm_require_dr
);
536 * This function will be used to read from the physical memory of the currently
537 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
538 * can read from guest physical or from the guest's guest physical memory.
540 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
541 gfn_t ngfn
, void *data
, int offset
, int len
,
544 struct x86_exception exception
;
548 ngpa
= gfn_to_gpa(ngfn
);
549 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
, &exception
);
550 if (real_gfn
== UNMAPPED_GVA
)
553 real_gfn
= gpa_to_gfn(real_gfn
);
555 return kvm_vcpu_read_guest_page(vcpu
, real_gfn
, data
, offset
, len
);
557 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
559 static int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
560 void *data
, int offset
, int len
, u32 access
)
562 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
563 data
, offset
, len
, access
);
567 * Load the pae pdptrs. Return true is they are all valid.
569 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
571 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
572 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
575 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
577 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
578 offset
* sizeof(u64
), sizeof(pdpte
),
579 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
584 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
585 if ((pdpte
[i
] & PT_PRESENT_MASK
) &&
587 vcpu
->arch
.mmu
.guest_rsvd_check
.rsvd_bits_mask
[0][2])) {
594 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
595 __set_bit(VCPU_EXREG_PDPTR
,
596 (unsigned long *)&vcpu
->arch
.regs_avail
);
597 __set_bit(VCPU_EXREG_PDPTR
,
598 (unsigned long *)&vcpu
->arch
.regs_dirty
);
603 EXPORT_SYMBOL_GPL(load_pdptrs
);
605 bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
607 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
613 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
616 if (!test_bit(VCPU_EXREG_PDPTR
,
617 (unsigned long *)&vcpu
->arch
.regs_avail
))
620 gfn
= (kvm_read_cr3(vcpu
) & 0xffffffe0ul
) >> PAGE_SHIFT
;
621 offset
= (kvm_read_cr3(vcpu
) & 0xffffffe0ul
) & (PAGE_SIZE
- 1);
622 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
623 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
626 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
631 EXPORT_SYMBOL_GPL(pdptrs_changed
);
633 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
635 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
636 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
;
641 if (cr0
& 0xffffffff00000000UL
)
645 cr0
&= ~CR0_RESERVED_BITS
;
647 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
650 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
653 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
655 if ((vcpu
->arch
.efer
& EFER_LME
)) {
660 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
665 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
670 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
673 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
675 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
676 kvm_clear_async_pf_completion_queue(vcpu
);
677 kvm_async_pf_hash_reset(vcpu
);
680 if ((cr0
^ old_cr0
) & update_bits
)
681 kvm_mmu_reset_context(vcpu
);
683 if (((cr0
^ old_cr0
) & X86_CR0_CD
) &&
684 kvm_arch_has_noncoherent_dma(vcpu
->kvm
) &&
685 !kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
686 kvm_zap_gfn_range(vcpu
->kvm
, 0, ~0ULL);
690 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
692 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
694 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
696 EXPORT_SYMBOL_GPL(kvm_lmsw
);
698 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
700 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
701 !vcpu
->guest_xcr0_loaded
) {
702 /* kvm_set_xcr() also depends on this */
703 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
704 vcpu
->guest_xcr0_loaded
= 1;
708 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
710 if (vcpu
->guest_xcr0_loaded
) {
711 if (vcpu
->arch
.xcr0
!= host_xcr0
)
712 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
713 vcpu
->guest_xcr0_loaded
= 0;
717 static int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
720 u64 old_xcr0
= vcpu
->arch
.xcr0
;
723 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
724 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
726 if (!(xcr0
& XFEATURE_MASK_FP
))
728 if ((xcr0
& XFEATURE_MASK_YMM
) && !(xcr0
& XFEATURE_MASK_SSE
))
732 * Do not allow the guest to set bits that we do not support
733 * saving. However, xcr0 bit 0 is always set, even if the
734 * emulated CPU does not support XSAVE (see fx_init).
736 valid_bits
= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FP
;
737 if (xcr0
& ~valid_bits
)
740 if ((!(xcr0
& XFEATURE_MASK_BNDREGS
)) !=
741 (!(xcr0
& XFEATURE_MASK_BNDCSR
)))
744 if (xcr0
& XFEATURE_MASK_AVX512
) {
745 if (!(xcr0
& XFEATURE_MASK_YMM
))
747 if ((xcr0
& XFEATURE_MASK_AVX512
) != XFEATURE_MASK_AVX512
)
750 vcpu
->arch
.xcr0
= xcr0
;
752 if ((xcr0
^ old_xcr0
) & XFEATURE_MASK_EXTEND
)
753 kvm_update_cpuid(vcpu
);
757 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
759 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 ||
760 __kvm_set_xcr(vcpu
, index
, xcr
)) {
761 kvm_inject_gp(vcpu
, 0);
766 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
768 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
770 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
771 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
|
772 X86_CR4_SMEP
| X86_CR4_SMAP
| X86_CR4_PKE
;
774 if (cr4
& CR4_RESERVED_BITS
)
777 if (!guest_cpuid_has(vcpu
, X86_FEATURE_XSAVE
) && (cr4
& X86_CR4_OSXSAVE
))
780 if (!guest_cpuid_has(vcpu
, X86_FEATURE_SMEP
) && (cr4
& X86_CR4_SMEP
))
783 if (!guest_cpuid_has(vcpu
, X86_FEATURE_SMAP
) && (cr4
& X86_CR4_SMAP
))
786 if (!guest_cpuid_has(vcpu
, X86_FEATURE_FSGSBASE
) && (cr4
& X86_CR4_FSGSBASE
))
789 if (!guest_cpuid_has(vcpu
, X86_FEATURE_PKU
) && (cr4
& X86_CR4_PKE
))
792 if (!guest_cpuid_has(vcpu
, X86_FEATURE_LA57
) && (cr4
& X86_CR4_LA57
))
795 if (is_long_mode(vcpu
)) {
796 if (!(cr4
& X86_CR4_PAE
))
798 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
799 && ((cr4
^ old_cr4
) & pdptr_bits
)
800 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
804 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
805 if (!guest_cpuid_has(vcpu
, X86_FEATURE_PCID
))
808 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
809 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
813 if (kvm_x86_ops
->set_cr4(vcpu
, cr4
))
816 if (((cr4
^ old_cr4
) & pdptr_bits
) ||
817 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
818 kvm_mmu_reset_context(vcpu
);
820 if ((cr4
^ old_cr4
) & (X86_CR4_OSXSAVE
| X86_CR4_PKE
))
821 kvm_update_cpuid(vcpu
);
825 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
827 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
830 cr3
&= ~CR3_PCID_INVD
;
833 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
834 kvm_mmu_sync_roots(vcpu
);
835 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
839 if (is_long_mode(vcpu
) &&
840 (cr3
& rsvd_bits(cpuid_maxphyaddr(vcpu
), 63)))
842 else if (is_pae(vcpu
) && is_paging(vcpu
) &&
843 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
846 vcpu
->arch
.cr3
= cr3
;
847 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
848 kvm_mmu_new_cr3(vcpu
);
851 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
853 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
855 if (cr8
& CR8_RESERVED_BITS
)
857 if (lapic_in_kernel(vcpu
))
858 kvm_lapic_set_tpr(vcpu
, cr8
);
860 vcpu
->arch
.cr8
= cr8
;
863 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
865 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
867 if (lapic_in_kernel(vcpu
))
868 return kvm_lapic_get_cr8(vcpu
);
870 return vcpu
->arch
.cr8
;
872 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
874 static void kvm_update_dr0123(struct kvm_vcpu
*vcpu
)
878 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
879 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
880 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
881 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_RELOAD
;
885 static void kvm_update_dr6(struct kvm_vcpu
*vcpu
)
887 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
888 kvm_x86_ops
->set_dr6(vcpu
, vcpu
->arch
.dr6
);
891 static void kvm_update_dr7(struct kvm_vcpu
*vcpu
)
895 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
896 dr7
= vcpu
->arch
.guest_debug_dr7
;
898 dr7
= vcpu
->arch
.dr7
;
899 kvm_x86_ops
->set_dr7(vcpu
, dr7
);
900 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_BP_ENABLED
;
901 if (dr7
& DR7_BP_EN_MASK
)
902 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_BP_ENABLED
;
905 static u64
kvm_dr6_fixed(struct kvm_vcpu
*vcpu
)
907 u64 fixed
= DR6_FIXED_1
;
909 if (!guest_cpuid_has(vcpu
, X86_FEATURE_RTM
))
914 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
918 vcpu
->arch
.db
[dr
] = val
;
919 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
920 vcpu
->arch
.eff_db
[dr
] = val
;
925 if (val
& 0xffffffff00000000ULL
)
927 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | kvm_dr6_fixed(vcpu
);
928 kvm_update_dr6(vcpu
);
933 if (val
& 0xffffffff00000000ULL
)
935 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
936 kvm_update_dr7(vcpu
);
943 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
945 if (__kvm_set_dr(vcpu
, dr
, val
)) {
946 kvm_inject_gp(vcpu
, 0);
951 EXPORT_SYMBOL_GPL(kvm_set_dr
);
953 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
957 *val
= vcpu
->arch
.db
[dr
];
962 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
963 *val
= vcpu
->arch
.dr6
;
965 *val
= kvm_x86_ops
->get_dr6(vcpu
);
970 *val
= vcpu
->arch
.dr7
;
975 EXPORT_SYMBOL_GPL(kvm_get_dr
);
977 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
979 u32 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
983 err
= kvm_pmu_rdpmc(vcpu
, ecx
, &data
);
986 kvm_register_write(vcpu
, VCPU_REGS_RAX
, (u32
)data
);
987 kvm_register_write(vcpu
, VCPU_REGS_RDX
, data
>> 32);
990 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
993 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
994 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
996 * This list is modified at module load time to reflect the
997 * capabilities of the host cpu. This capabilities test skips MSRs that are
998 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
999 * may depend on host virtualization features rather than host cpu features.
1002 static u32 msrs_to_save
[] = {
1003 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1005 #ifdef CONFIG_X86_64
1006 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1008 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
,
1009 MSR_IA32_FEATURE_CONTROL
, MSR_IA32_BNDCFGS
, MSR_TSC_AUX
,
1010 MSR_IA32_SPEC_CTRL
, MSR_IA32_ARCH_CAPABILITIES
1013 static unsigned num_msrs_to_save
;
1015 static u32 emulated_msrs
[] = {
1016 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
1017 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
1018 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
1019 HV_X64_MSR_TIME_REF_COUNT
, HV_X64_MSR_REFERENCE_TSC
,
1020 HV_X64_MSR_TSC_FREQUENCY
, HV_X64_MSR_APIC_FREQUENCY
,
1021 HV_X64_MSR_CRASH_P0
, HV_X64_MSR_CRASH_P1
, HV_X64_MSR_CRASH_P2
,
1022 HV_X64_MSR_CRASH_P3
, HV_X64_MSR_CRASH_P4
, HV_X64_MSR_CRASH_CTL
,
1024 HV_X64_MSR_VP_INDEX
,
1025 HV_X64_MSR_VP_RUNTIME
,
1026 HV_X64_MSR_SCONTROL
,
1027 HV_X64_MSR_STIMER0_CONFIG
,
1028 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
1031 MSR_IA32_TSC_ADJUST
,
1032 MSR_IA32_TSCDEADLINE
,
1033 MSR_IA32_MISC_ENABLE
,
1034 MSR_IA32_MCG_STATUS
,
1036 MSR_IA32_MCG_EXT_CTL
,
1039 MSR_MISC_FEATURES_ENABLES
,
1040 MSR_AMD64_VIRT_SPEC_CTRL
,
1043 static unsigned num_emulated_msrs
;
1046 * List of msr numbers which are used to expose MSR-based features that
1047 * can be used by a hypervisor to validate requested CPU features.
1049 static u32 msr_based_features
[] = {
1052 static unsigned int num_msr_based_features
;
1054 static int do_get_msr_feature(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1056 struct kvm_msr_entry msr
;
1059 if (kvm_x86_ops
->get_msr_feature(&msr
))
1067 bool kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1069 if (efer
& efer_reserved_bits
)
1072 if (efer
& EFER_FFXSR
&& !guest_cpuid_has(vcpu
, X86_FEATURE_FXSR_OPT
))
1075 if (efer
& EFER_SVME
&& !guest_cpuid_has(vcpu
, X86_FEATURE_SVM
))
1080 EXPORT_SYMBOL_GPL(kvm_valid_efer
);
1082 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1084 u64 old_efer
= vcpu
->arch
.efer
;
1086 if (!kvm_valid_efer(vcpu
, efer
))
1090 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
1094 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
1096 kvm_x86_ops
->set_efer(vcpu
, efer
);
1098 /* Update reserved bits */
1099 if ((efer
^ old_efer
) & EFER_NX
)
1100 kvm_mmu_reset_context(vcpu
);
1105 void kvm_enable_efer_bits(u64 mask
)
1107 efer_reserved_bits
&= ~mask
;
1109 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
1112 * Writes msr value into into the appropriate "register".
1113 * Returns 0 on success, non-0 otherwise.
1114 * Assumes vcpu_load() was already called.
1116 int kvm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1118 switch (msr
->index
) {
1121 case MSR_KERNEL_GS_BASE
:
1124 if (is_noncanonical_address(msr
->data
, vcpu
))
1127 case MSR_IA32_SYSENTER_EIP
:
1128 case MSR_IA32_SYSENTER_ESP
:
1130 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1131 * non-canonical address is written on Intel but not on
1132 * AMD (which ignores the top 32-bits, because it does
1133 * not implement 64-bit SYSENTER).
1135 * 64-bit code should hence be able to write a non-canonical
1136 * value on AMD. Making the address canonical ensures that
1137 * vmentry does not fail on Intel after writing a non-canonical
1138 * value, and that something deterministic happens if the guest
1139 * invokes 64-bit SYSENTER.
1141 msr
->data
= get_canonical(msr
->data
, vcpu_virt_addr_bits(vcpu
));
1143 return kvm_x86_ops
->set_msr(vcpu
, msr
);
1145 EXPORT_SYMBOL_GPL(kvm_set_msr
);
1148 * Adapt set_msr() to msr_io()'s calling convention
1150 static int do_get_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1152 struct msr_data msr
;
1156 msr
.host_initiated
= true;
1157 r
= kvm_get_msr(vcpu
, &msr
);
1165 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1167 struct msr_data msr
;
1171 msr
.host_initiated
= true;
1172 return kvm_set_msr(vcpu
, &msr
);
1175 #ifdef CONFIG_X86_64
1176 struct pvclock_gtod_data
{
1179 struct { /* extract of a clocksource struct */
1192 static struct pvclock_gtod_data pvclock_gtod_data
;
1194 static void update_pvclock_gtod(struct timekeeper
*tk
)
1196 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
1199 boot_ns
= ktime_to_ns(ktime_add(tk
->tkr_mono
.base
, tk
->offs_boot
));
1201 write_seqcount_begin(&vdata
->seq
);
1203 /* copy pvclock gtod data */
1204 vdata
->clock
.vclock_mode
= tk
->tkr_mono
.clock
->archdata
.vclock_mode
;
1205 vdata
->clock
.cycle_last
= tk
->tkr_mono
.cycle_last
;
1206 vdata
->clock
.mask
= tk
->tkr_mono
.mask
;
1207 vdata
->clock
.mult
= tk
->tkr_mono
.mult
;
1208 vdata
->clock
.shift
= tk
->tkr_mono
.shift
;
1210 vdata
->boot_ns
= boot_ns
;
1211 vdata
->nsec_base
= tk
->tkr_mono
.xtime_nsec
;
1213 vdata
->wall_time_sec
= tk
->xtime_sec
;
1215 write_seqcount_end(&vdata
->seq
);
1219 void kvm_set_pending_timer(struct kvm_vcpu
*vcpu
)
1222 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1223 * vcpu_enter_guest. This function is only called from
1224 * the physical CPU that is running vcpu.
1226 kvm_make_request(KVM_REQ_PENDING_TIMER
, vcpu
);
1229 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
1233 struct pvclock_wall_clock wc
;
1234 struct timespec64 boot
;
1239 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1244 ++version
; /* first time write, random junk */
1248 if (kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
)))
1252 * The guest calculates current wall clock time by adding
1253 * system time (updated by kvm_guest_time_update below) to the
1254 * wall clock specified here. guest system time equals host
1255 * system time for us, thus we must fill in host boot time here.
1257 getboottime64(&boot
);
1259 if (kvm
->arch
.kvmclock_offset
) {
1260 struct timespec64 ts
= ns_to_timespec64(kvm
->arch
.kvmclock_offset
);
1261 boot
= timespec64_sub(boot
, ts
);
1263 wc
.sec
= (u32
)boot
.tv_sec
; /* overflow in 2106 guest time */
1264 wc
.nsec
= boot
.tv_nsec
;
1265 wc
.version
= version
;
1267 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1270 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1273 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
1275 do_shl32_div32(dividend
, divisor
);
1279 static void kvm_get_time_scale(uint64_t scaled_hz
, uint64_t base_hz
,
1280 s8
*pshift
, u32
*pmultiplier
)
1288 scaled64
= scaled_hz
;
1289 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
1294 tps32
= (uint32_t)tps64
;
1295 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
1296 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
1304 *pmultiplier
= div_frac(scaled64
, tps32
);
1306 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1307 __func__
, base_hz
, scaled_hz
, shift
, *pmultiplier
);
1310 #ifdef CONFIG_X86_64
1311 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
1314 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
1315 static unsigned long max_tsc_khz
;
1317 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1319 u64 v
= (u64
)khz
* (1000000 + ppm
);
1324 static int set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
, bool scale
)
1328 /* Guest TSC same frequency as host TSC? */
1330 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1334 /* TSC scaling supported? */
1335 if (!kvm_has_tsc_control
) {
1336 if (user_tsc_khz
> tsc_khz
) {
1337 vcpu
->arch
.tsc_catchup
= 1;
1338 vcpu
->arch
.tsc_always_catchup
= 1;
1341 WARN(1, "user requested TSC rate below hardware speed\n");
1346 /* TSC scaling required - calculate ratio */
1347 ratio
= mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits
,
1348 user_tsc_khz
, tsc_khz
);
1350 if (ratio
== 0 || ratio
>= kvm_max_tsc_scaling_ratio
) {
1351 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1356 vcpu
->arch
.tsc_scaling_ratio
= ratio
;
1360 static int kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
)
1362 u32 thresh_lo
, thresh_hi
;
1363 int use_scaling
= 0;
1365 /* tsc_khz can be zero if TSC calibration fails */
1366 if (user_tsc_khz
== 0) {
1367 /* set tsc_scaling_ratio to a safe value */
1368 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1372 /* Compute a scale to convert nanoseconds in TSC cycles */
1373 kvm_get_time_scale(user_tsc_khz
* 1000LL, NSEC_PER_SEC
,
1374 &vcpu
->arch
.virtual_tsc_shift
,
1375 &vcpu
->arch
.virtual_tsc_mult
);
1376 vcpu
->arch
.virtual_tsc_khz
= user_tsc_khz
;
1379 * Compute the variation in TSC rate which is acceptable
1380 * within the range of tolerance and decide if the
1381 * rate being applied is within that bounds of the hardware
1382 * rate. If so, no scaling or compensation need be done.
1384 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1385 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1386 if (user_tsc_khz
< thresh_lo
|| user_tsc_khz
> thresh_hi
) {
1387 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz
, thresh_lo
, thresh_hi
);
1390 return set_tsc_khz(vcpu
, user_tsc_khz
, use_scaling
);
1393 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1395 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1396 vcpu
->arch
.virtual_tsc_mult
,
1397 vcpu
->arch
.virtual_tsc_shift
);
1398 tsc
+= vcpu
->arch
.this_tsc_write
;
1402 static void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
1404 #ifdef CONFIG_X86_64
1406 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1407 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1409 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1410 atomic_read(&vcpu
->kvm
->online_vcpus
));
1413 * Once the masterclock is enabled, always perform request in
1414 * order to update it.
1416 * In order to enable masterclock, the host clocksource must be TSC
1417 * and the vcpus need to have matched TSCs. When that happens,
1418 * perform request to enable masterclock.
1420 if (ka
->use_master_clock
||
1421 (gtod
->clock
.vclock_mode
== VCLOCK_TSC
&& vcpus_matched
))
1422 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1424 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
1425 atomic_read(&vcpu
->kvm
->online_vcpus
),
1426 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
1430 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu
*vcpu
, s64 offset
)
1432 u64 curr_offset
= vcpu
->arch
.tsc_offset
;
1433 vcpu
->arch
.ia32_tsc_adjust_msr
+= offset
- curr_offset
;
1437 * Multiply tsc by a fixed point number represented by ratio.
1439 * The most significant 64-N bits (mult) of ratio represent the
1440 * integral part of the fixed point number; the remaining N bits
1441 * (frac) represent the fractional part, ie. ratio represents a fixed
1442 * point number (mult + frac * 2^(-N)).
1444 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1446 static inline u64
__scale_tsc(u64 ratio
, u64 tsc
)
1448 return mul_u64_u64_shr(tsc
, ratio
, kvm_tsc_scaling_ratio_frac_bits
);
1451 u64
kvm_scale_tsc(struct kvm_vcpu
*vcpu
, u64 tsc
)
1454 u64 ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1456 if (ratio
!= kvm_default_tsc_scaling_ratio
)
1457 _tsc
= __scale_tsc(ratio
, tsc
);
1461 EXPORT_SYMBOL_GPL(kvm_scale_tsc
);
1463 static u64
kvm_compute_tsc_offset(struct kvm_vcpu
*vcpu
, u64 target_tsc
)
1467 tsc
= kvm_scale_tsc(vcpu
, rdtsc());
1469 return target_tsc
- tsc
;
1472 u64
kvm_read_l1_tsc(struct kvm_vcpu
*vcpu
, u64 host_tsc
)
1474 return vcpu
->arch
.tsc_offset
+ kvm_scale_tsc(vcpu
, host_tsc
);
1476 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc
);
1478 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1480 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1481 vcpu
->arch
.tsc_offset
= offset
;
1484 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1486 struct kvm
*kvm
= vcpu
->kvm
;
1487 u64 offset
, ns
, elapsed
;
1488 unsigned long flags
;
1490 bool already_matched
;
1491 u64 data
= msr
->data
;
1492 bool synchronizing
= false;
1494 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1495 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1496 ns
= ktime_get_boot_ns();
1497 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1499 if (vcpu
->arch
.virtual_tsc_khz
) {
1500 if (data
== 0 && msr
->host_initiated
) {
1502 * detection of vcpu initialization -- need to sync
1503 * with other vCPUs. This particularly helps to keep
1504 * kvm_clock stable after CPU hotplug
1506 synchronizing
= true;
1508 u64 tsc_exp
= kvm
->arch
.last_tsc_write
+
1509 nsec_to_cycles(vcpu
, elapsed
);
1510 u64 tsc_hz
= vcpu
->arch
.virtual_tsc_khz
* 1000LL;
1512 * Special case: TSC write with a small delta (1 second)
1513 * of virtual cycle time against real time is
1514 * interpreted as an attempt to synchronize the CPU.
1516 synchronizing
= data
< tsc_exp
+ tsc_hz
&&
1517 data
+ tsc_hz
> tsc_exp
;
1522 * For a reliable TSC, we can match TSC offsets, and for an unstable
1523 * TSC, we add elapsed time in this computation. We could let the
1524 * compensation code attempt to catch up if we fall behind, but
1525 * it's better to try to match offsets from the beginning.
1527 if (synchronizing
&&
1528 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1529 if (!check_tsc_unstable()) {
1530 offset
= kvm
->arch
.cur_tsc_offset
;
1531 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1533 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1535 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1536 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1539 already_matched
= (vcpu
->arch
.this_tsc_generation
== kvm
->arch
.cur_tsc_generation
);
1542 * We split periods of matched TSC writes into generations.
1543 * For each generation, we track the original measured
1544 * nanosecond time, offset, and write, so if TSCs are in
1545 * sync, we can match exact offset, and if not, we can match
1546 * exact software computation in compute_guest_tsc()
1548 * These values are tracked in kvm->arch.cur_xxx variables.
1550 kvm
->arch
.cur_tsc_generation
++;
1551 kvm
->arch
.cur_tsc_nsec
= ns
;
1552 kvm
->arch
.cur_tsc_write
= data
;
1553 kvm
->arch
.cur_tsc_offset
= offset
;
1555 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1556 kvm
->arch
.cur_tsc_generation
, data
);
1560 * We also track th most recent recorded KHZ, write and time to
1561 * allow the matching interval to be extended at each write.
1563 kvm
->arch
.last_tsc_nsec
= ns
;
1564 kvm
->arch
.last_tsc_write
= data
;
1565 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1567 vcpu
->arch
.last_guest_tsc
= data
;
1569 /* Keep track of which generation this VCPU has synchronized to */
1570 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1571 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1572 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1574 if (!msr
->host_initiated
&& guest_cpuid_has(vcpu
, X86_FEATURE_TSC_ADJUST
))
1575 update_ia32_tsc_adjust_msr(vcpu
, offset
);
1577 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
1578 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1580 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1582 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
1583 } else if (!already_matched
) {
1584 kvm
->arch
.nr_vcpus_matched_tsc
++;
1587 kvm_track_tsc_matching(vcpu
);
1588 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1591 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1593 static inline void adjust_tsc_offset_guest(struct kvm_vcpu
*vcpu
,
1596 kvm_vcpu_write_tsc_offset(vcpu
, vcpu
->arch
.tsc_offset
+ adjustment
);
1599 static inline void adjust_tsc_offset_host(struct kvm_vcpu
*vcpu
, s64 adjustment
)
1601 if (vcpu
->arch
.tsc_scaling_ratio
!= kvm_default_tsc_scaling_ratio
)
1602 WARN_ON(adjustment
< 0);
1603 adjustment
= kvm_scale_tsc(vcpu
, (u64
) adjustment
);
1604 adjust_tsc_offset_guest(vcpu
, adjustment
);
1607 #ifdef CONFIG_X86_64
1609 static u64
read_tsc(void)
1611 u64 ret
= (u64
)rdtsc_ordered();
1612 u64 last
= pvclock_gtod_data
.clock
.cycle_last
;
1614 if (likely(ret
>= last
))
1618 * GCC likes to generate cmov here, but this branch is extremely
1619 * predictable (it's just a function of time and the likely is
1620 * very likely) and there's a data dependence, so force GCC
1621 * to generate a branch instead. I don't barrier() because
1622 * we don't actually need a barrier, and if this function
1623 * ever gets inlined it will generate worse code.
1629 static inline u64
vgettsc(u64
*cycle_now
)
1632 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1634 *cycle_now
= read_tsc();
1636 v
= (*cycle_now
- gtod
->clock
.cycle_last
) & gtod
->clock
.mask
;
1637 return v
* gtod
->clock
.mult
;
1640 static int do_monotonic_boot(s64
*t
, u64
*cycle_now
)
1642 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1648 seq
= read_seqcount_begin(>od
->seq
);
1649 mode
= gtod
->clock
.vclock_mode
;
1650 ns
= gtod
->nsec_base
;
1651 ns
+= vgettsc(cycle_now
);
1652 ns
>>= gtod
->clock
.shift
;
1653 ns
+= gtod
->boot_ns
;
1654 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1660 static int do_realtime(struct timespec
*ts
, u64
*cycle_now
)
1662 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1668 seq
= read_seqcount_begin(>od
->seq
);
1669 mode
= gtod
->clock
.vclock_mode
;
1670 ts
->tv_sec
= gtod
->wall_time_sec
;
1671 ns
= gtod
->nsec_base
;
1672 ns
+= vgettsc(cycle_now
);
1673 ns
>>= gtod
->clock
.shift
;
1674 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1676 ts
->tv_sec
+= __iter_div_u64_rem(ns
, NSEC_PER_SEC
, &ns
);
1682 /* returns true if host is using tsc clocksource */
1683 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, u64
*cycle_now
)
1685 /* checked again under seqlock below */
1686 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1689 return do_monotonic_boot(kernel_ns
, cycle_now
) == VCLOCK_TSC
;
1692 /* returns true if host is using tsc clocksource */
1693 static bool kvm_get_walltime_and_clockread(struct timespec
*ts
,
1696 /* checked again under seqlock below */
1697 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1700 return do_realtime(ts
, cycle_now
) == VCLOCK_TSC
;
1706 * Assuming a stable TSC across physical CPUS, and a stable TSC
1707 * across virtual CPUs, the following condition is possible.
1708 * Each numbered line represents an event visible to both
1709 * CPUs at the next numbered event.
1711 * "timespecX" represents host monotonic time. "tscX" represents
1714 * VCPU0 on CPU0 | VCPU1 on CPU1
1716 * 1. read timespec0,tsc0
1717 * 2. | timespec1 = timespec0 + N
1719 * 3. transition to guest | transition to guest
1720 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1721 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1722 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1724 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1727 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1729 * - 0 < N - M => M < N
1731 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1732 * always the case (the difference between two distinct xtime instances
1733 * might be smaller then the difference between corresponding TSC reads,
1734 * when updating guest vcpus pvclock areas).
1736 * To avoid that problem, do not allow visibility of distinct
1737 * system_timestamp/tsc_timestamp values simultaneously: use a master
1738 * copy of host monotonic time values. Update that master copy
1741 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1745 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
1747 #ifdef CONFIG_X86_64
1748 struct kvm_arch
*ka
= &kvm
->arch
;
1750 bool host_tsc_clocksource
, vcpus_matched
;
1752 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1753 atomic_read(&kvm
->online_vcpus
));
1756 * If the host uses TSC clock, then passthrough TSC as stable
1759 host_tsc_clocksource
= kvm_get_time_and_clockread(
1760 &ka
->master_kernel_ns
,
1761 &ka
->master_cycle_now
);
1763 ka
->use_master_clock
= host_tsc_clocksource
&& vcpus_matched
1764 && !ka
->backwards_tsc_observed
1765 && !ka
->boot_vcpu_runs_old_kvmclock
;
1767 if (ka
->use_master_clock
)
1768 atomic_set(&kvm_guest_has_master_clock
, 1);
1770 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
1771 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
1776 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
1778 kvm_make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
1781 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
1783 #ifdef CONFIG_X86_64
1785 struct kvm_vcpu
*vcpu
;
1786 struct kvm_arch
*ka
= &kvm
->arch
;
1788 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1789 kvm_make_mclock_inprogress_request(kvm
);
1790 /* no guest entries from this point */
1791 pvclock_update_vm_gtod_copy(kvm
);
1793 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1794 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1796 /* guest entries allowed */
1797 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1798 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS
, vcpu
);
1800 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1804 u64
get_kvmclock_ns(struct kvm
*kvm
)
1806 struct kvm_arch
*ka
= &kvm
->arch
;
1807 struct pvclock_vcpu_time_info hv_clock
;
1810 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1811 if (!ka
->use_master_clock
) {
1812 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1813 return ktime_get_boot_ns() + ka
->kvmclock_offset
;
1816 hv_clock
.tsc_timestamp
= ka
->master_cycle_now
;
1817 hv_clock
.system_time
= ka
->master_kernel_ns
+ ka
->kvmclock_offset
;
1818 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1820 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1823 if (__this_cpu_read(cpu_tsc_khz
)) {
1824 kvm_get_time_scale(NSEC_PER_SEC
, __this_cpu_read(cpu_tsc_khz
) * 1000LL,
1825 &hv_clock
.tsc_shift
,
1826 &hv_clock
.tsc_to_system_mul
);
1827 ret
= __pvclock_read_cycles(&hv_clock
, rdtsc());
1829 ret
= ktime_get_boot_ns() + ka
->kvmclock_offset
;
1836 static void kvm_setup_pvclock_page(struct kvm_vcpu
*v
)
1838 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1839 struct pvclock_vcpu_time_info guest_hv_clock
;
1841 if (unlikely(kvm_read_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1842 &guest_hv_clock
, sizeof(guest_hv_clock
))))
1845 /* This VCPU is paused, but it's legal for a guest to read another
1846 * VCPU's kvmclock, so we really have to follow the specification where
1847 * it says that version is odd if data is being modified, and even after
1850 * Version field updates must be kept separate. This is because
1851 * kvm_write_guest_cached might use a "rep movs" instruction, and
1852 * writes within a string instruction are weakly ordered. So there
1853 * are three writes overall.
1855 * As a small optimization, only write the version field in the first
1856 * and third write. The vcpu->pv_time cache is still valid, because the
1857 * version field is the first in the struct.
1859 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info
, version
) != 0);
1861 if (guest_hv_clock
.version
& 1)
1862 ++guest_hv_clock
.version
; /* first time write, random junk */
1864 vcpu
->hv_clock
.version
= guest_hv_clock
.version
+ 1;
1865 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1867 sizeof(vcpu
->hv_clock
.version
));
1871 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1872 vcpu
->hv_clock
.flags
|= (guest_hv_clock
.flags
& PVCLOCK_GUEST_STOPPED
);
1874 if (vcpu
->pvclock_set_guest_stopped_request
) {
1875 vcpu
->hv_clock
.flags
|= PVCLOCK_GUEST_STOPPED
;
1876 vcpu
->pvclock_set_guest_stopped_request
= false;
1879 trace_kvm_pvclock_update(v
->vcpu_id
, &vcpu
->hv_clock
);
1881 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1883 sizeof(vcpu
->hv_clock
));
1887 vcpu
->hv_clock
.version
++;
1888 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1890 sizeof(vcpu
->hv_clock
.version
));
1893 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1895 unsigned long flags
, tgt_tsc_khz
;
1896 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1897 struct kvm_arch
*ka
= &v
->kvm
->arch
;
1899 u64 tsc_timestamp
, host_tsc
;
1901 bool use_master_clock
;
1907 * If the host uses TSC clock, then passthrough TSC as stable
1910 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1911 use_master_clock
= ka
->use_master_clock
;
1912 if (use_master_clock
) {
1913 host_tsc
= ka
->master_cycle_now
;
1914 kernel_ns
= ka
->master_kernel_ns
;
1916 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1918 /* Keep irq disabled to prevent changes to the clock */
1919 local_irq_save(flags
);
1920 tgt_tsc_khz
= __this_cpu_read(cpu_tsc_khz
);
1921 if (unlikely(tgt_tsc_khz
== 0)) {
1922 local_irq_restore(flags
);
1923 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1926 if (!use_master_clock
) {
1928 kernel_ns
= ktime_get_boot_ns();
1931 tsc_timestamp
= kvm_read_l1_tsc(v
, host_tsc
);
1934 * We may have to catch up the TSC to match elapsed wall clock
1935 * time for two reasons, even if kvmclock is used.
1936 * 1) CPU could have been running below the maximum TSC rate
1937 * 2) Broken TSC compensation resets the base at each VCPU
1938 * entry to avoid unknown leaps of TSC even when running
1939 * again on the same CPU. This may cause apparent elapsed
1940 * time to disappear, and the guest to stand still or run
1943 if (vcpu
->tsc_catchup
) {
1944 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1945 if (tsc
> tsc_timestamp
) {
1946 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1947 tsc_timestamp
= tsc
;
1951 local_irq_restore(flags
);
1953 /* With all the info we got, fill in the values */
1955 if (kvm_has_tsc_control
)
1956 tgt_tsc_khz
= kvm_scale_tsc(v
, tgt_tsc_khz
);
1958 if (unlikely(vcpu
->hw_tsc_khz
!= tgt_tsc_khz
)) {
1959 kvm_get_time_scale(NSEC_PER_SEC
, tgt_tsc_khz
* 1000LL,
1960 &vcpu
->hv_clock
.tsc_shift
,
1961 &vcpu
->hv_clock
.tsc_to_system_mul
);
1962 vcpu
->hw_tsc_khz
= tgt_tsc_khz
;
1965 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1966 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1967 vcpu
->last_guest_tsc
= tsc_timestamp
;
1969 /* If the host uses TSC clocksource, then it is stable */
1971 if (use_master_clock
)
1972 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
1974 vcpu
->hv_clock
.flags
= pvclock_flags
;
1976 if (vcpu
->pv_time_enabled
)
1977 kvm_setup_pvclock_page(v
);
1978 if (v
== kvm_get_vcpu(v
->kvm
, 0))
1979 kvm_hv_setup_tsc_page(v
->kvm
, &vcpu
->hv_clock
);
1984 * kvmclock updates which are isolated to a given vcpu, such as
1985 * vcpu->cpu migration, should not allow system_timestamp from
1986 * the rest of the vcpus to remain static. Otherwise ntp frequency
1987 * correction applies to one vcpu's system_timestamp but not
1990 * So in those cases, request a kvmclock update for all vcpus.
1991 * We need to rate-limit these requests though, as they can
1992 * considerably slow guests that have a large number of vcpus.
1993 * The time for a remote vcpu to update its kvmclock is bound
1994 * by the delay we use to rate-limit the updates.
1997 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1999 static void kvmclock_update_fn(struct work_struct
*work
)
2002 struct delayed_work
*dwork
= to_delayed_work(work
);
2003 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2004 kvmclock_update_work
);
2005 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2006 struct kvm_vcpu
*vcpu
;
2008 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2009 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2010 kvm_vcpu_kick(vcpu
);
2014 static void kvm_gen_kvmclock_update(struct kvm_vcpu
*v
)
2016 struct kvm
*kvm
= v
->kvm
;
2018 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
2019 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
,
2020 KVMCLOCK_UPDATE_DELAY
);
2023 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2025 static void kvmclock_sync_fn(struct work_struct
*work
)
2027 struct delayed_work
*dwork
= to_delayed_work(work
);
2028 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2029 kvmclock_sync_work
);
2030 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2032 if (!kvmclock_periodic_sync
)
2035 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
, 0);
2036 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
2037 KVMCLOCK_SYNC_PERIOD
);
2040 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
2042 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2043 unsigned bank_num
= mcg_cap
& 0xff;
2046 case MSR_IA32_MCG_STATUS
:
2047 vcpu
->arch
.mcg_status
= data
;
2049 case MSR_IA32_MCG_CTL
:
2050 if (!(mcg_cap
& MCG_CTL_P
))
2052 if (data
!= 0 && data
!= ~(u64
)0)
2054 vcpu
->arch
.mcg_ctl
= data
;
2057 if (msr
>= MSR_IA32_MC0_CTL
&&
2058 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2059 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2060 /* only 0 or all 1s can be written to IA32_MCi_CTL
2061 * some Linux kernels though clear bit 10 in bank 4 to
2062 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2063 * this to avoid an uncatched #GP in the guest
2065 if ((offset
& 0x3) == 0 &&
2066 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
2068 vcpu
->arch
.mce_banks
[offset
] = data
;
2076 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
2078 struct kvm
*kvm
= vcpu
->kvm
;
2079 int lm
= is_long_mode(vcpu
);
2080 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
2081 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
2082 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
2083 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
2084 u32 page_num
= data
& ~PAGE_MASK
;
2085 u64 page_addr
= data
& PAGE_MASK
;
2090 if (page_num
>= blob_size
)
2093 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
2098 if (kvm_vcpu_write_guest(vcpu
, page_addr
, page
, PAGE_SIZE
))
2107 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
2109 gpa_t gpa
= data
& ~0x3f;
2111 /* Bits 3:5 are reserved, Should be zero */
2115 vcpu
->arch
.apf
.msr_val
= data
;
2117 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
2118 kvm_clear_async_pf_completion_queue(vcpu
);
2119 kvm_async_pf_hash_reset(vcpu
);
2123 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
,
2127 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
2128 vcpu
->arch
.apf
.delivery_as_pf_vmexit
= data
& KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT
;
2129 kvm_async_pf_wakeup_all(vcpu
);
2133 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
2135 vcpu
->arch
.pv_time_enabled
= false;
2138 static void record_steal_time(struct kvm_vcpu
*vcpu
)
2140 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
2143 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2144 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
2147 vcpu
->arch
.st
.steal
.preempted
= 0;
2149 if (vcpu
->arch
.st
.steal
.version
& 1)
2150 vcpu
->arch
.st
.steal
.version
+= 1; /* first time write, random junk */
2152 vcpu
->arch
.st
.steal
.version
+= 1;
2154 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2155 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2159 vcpu
->arch
.st
.steal
.steal
+= current
->sched_info
.run_delay
-
2160 vcpu
->arch
.st
.last_steal
;
2161 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
2163 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2164 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2168 vcpu
->arch
.st
.steal
.version
+= 1;
2170 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2171 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2174 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2177 u32 msr
= msr_info
->index
;
2178 u64 data
= msr_info
->data
;
2181 case MSR_AMD64_NB_CFG
:
2182 case MSR_IA32_UCODE_REV
:
2183 case MSR_IA32_UCODE_WRITE
:
2184 case MSR_VM_HSAVE_PA
:
2185 case MSR_AMD64_PATCH_LOADER
:
2186 case MSR_AMD64_BU_CFG2
:
2187 case MSR_AMD64_DC_CFG
:
2191 return set_efer(vcpu
, data
);
2193 data
&= ~(u64
)0x40; /* ignore flush filter disable */
2194 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
2195 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
2196 data
&= ~(u64
)0x40000; /* ignore Mc status write enable */
2198 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
2203 case MSR_FAM10H_MMIO_CONF_BASE
:
2205 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
2210 case MSR_IA32_DEBUGCTLMSR
:
2212 /* We support the non-activated case already */
2214 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
2215 /* Values other than LBR and BTF are vendor-specific,
2216 thus reserved and should throw a #GP */
2219 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2222 case 0x200 ... 0x2ff:
2223 return kvm_mtrr_set_msr(vcpu
, msr
, data
);
2224 case MSR_IA32_APICBASE
:
2225 return kvm_set_apic_base(vcpu
, msr_info
);
2226 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2227 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
2228 case MSR_IA32_TSCDEADLINE
:
2229 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
2231 case MSR_IA32_TSC_ADJUST
:
2232 if (guest_cpuid_has(vcpu
, X86_FEATURE_TSC_ADJUST
)) {
2233 if (!msr_info
->host_initiated
) {
2234 s64 adj
= data
- vcpu
->arch
.ia32_tsc_adjust_msr
;
2235 adjust_tsc_offset_guest(vcpu
, adj
);
2237 vcpu
->arch
.ia32_tsc_adjust_msr
= data
;
2240 case MSR_IA32_MISC_ENABLE
:
2241 vcpu
->arch
.ia32_misc_enable_msr
= data
;
2243 case MSR_IA32_SMBASE
:
2244 if (!msr_info
->host_initiated
)
2246 vcpu
->arch
.smbase
= data
;
2248 case MSR_KVM_WALL_CLOCK_NEW
:
2249 case MSR_KVM_WALL_CLOCK
:
2250 vcpu
->kvm
->arch
.wall_clock
= data
;
2251 kvm_write_wall_clock(vcpu
->kvm
, data
);
2253 case MSR_KVM_SYSTEM_TIME_NEW
:
2254 case MSR_KVM_SYSTEM_TIME
: {
2255 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
2257 kvmclock_reset(vcpu
);
2259 if (vcpu
->vcpu_id
== 0 && !msr_info
->host_initiated
) {
2260 bool tmp
= (msr
== MSR_KVM_SYSTEM_TIME
);
2262 if (ka
->boot_vcpu_runs_old_kvmclock
!= tmp
)
2263 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
2265 ka
->boot_vcpu_runs_old_kvmclock
= tmp
;
2268 vcpu
->arch
.time
= data
;
2269 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2271 /* we verify if the enable bit is set... */
2275 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2276 &vcpu
->arch
.pv_time
, data
& ~1ULL,
2277 sizeof(struct pvclock_vcpu_time_info
)))
2278 vcpu
->arch
.pv_time_enabled
= false;
2280 vcpu
->arch
.pv_time_enabled
= true;
2284 case MSR_KVM_ASYNC_PF_EN
:
2285 if (kvm_pv_enable_async_pf(vcpu
, data
))
2288 case MSR_KVM_STEAL_TIME
:
2290 if (unlikely(!sched_info_on()))
2293 if (data
& KVM_STEAL_RESERVED_MASK
)
2296 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2297 data
& KVM_STEAL_VALID_BITS
,
2298 sizeof(struct kvm_steal_time
)))
2301 vcpu
->arch
.st
.msr_val
= data
;
2303 if (!(data
& KVM_MSR_ENABLED
))
2306 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2309 case MSR_KVM_PV_EOI_EN
:
2310 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
2314 case MSR_IA32_MCG_CTL
:
2315 case MSR_IA32_MCG_STATUS
:
2316 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2317 return set_msr_mce(vcpu
, msr
, data
);
2319 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2320 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2321 pr
= true; /* fall through */
2322 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2323 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2324 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2325 return kvm_pmu_set_msr(vcpu
, msr_info
);
2327 if (pr
|| data
!= 0)
2328 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
2329 "0x%x data 0x%llx\n", msr
, data
);
2331 case MSR_K7_CLK_CTL
:
2333 * Ignore all writes to this no longer documented MSR.
2334 * Writes are only relevant for old K7 processors,
2335 * all pre-dating SVM, but a recommended workaround from
2336 * AMD for these chips. It is possible to specify the
2337 * affected processor models on the command line, hence
2338 * the need to ignore the workaround.
2341 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2342 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2343 case HV_X64_MSR_CRASH_CTL
:
2344 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2345 return kvm_hv_set_msr_common(vcpu
, msr
, data
,
2346 msr_info
->host_initiated
);
2347 case MSR_IA32_BBL_CR_CTL3
:
2348 /* Drop writes to this legacy MSR -- see rdmsr
2349 * counterpart for further detail.
2351 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data 0x%llx\n", msr
, data
);
2353 case MSR_AMD64_OSVW_ID_LENGTH
:
2354 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2356 vcpu
->arch
.osvw
.length
= data
;
2358 case MSR_AMD64_OSVW_STATUS
:
2359 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2361 vcpu
->arch
.osvw
.status
= data
;
2363 case MSR_PLATFORM_INFO
:
2364 if (!msr_info
->host_initiated
||
2365 data
& ~MSR_PLATFORM_INFO_CPUID_FAULT
||
2366 (!(data
& MSR_PLATFORM_INFO_CPUID_FAULT
) &&
2367 cpuid_fault_enabled(vcpu
)))
2369 vcpu
->arch
.msr_platform_info
= data
;
2371 case MSR_MISC_FEATURES_ENABLES
:
2372 if (data
& ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
||
2373 (data
& MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
&&
2374 !supports_cpuid_fault(vcpu
)))
2376 vcpu
->arch
.msr_misc_features_enables
= data
;
2379 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
2380 return xen_hvm_config(vcpu
, data
);
2381 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2382 return kvm_pmu_set_msr(vcpu
, msr_info
);
2384 vcpu_debug_ratelimited(vcpu
, "unhandled wrmsr: 0x%x data 0x%llx\n",
2388 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data 0x%llx\n",
2395 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
2399 * Reads an msr value (of 'msr_index') into 'pdata'.
2400 * Returns 0 on success, non-0 otherwise.
2401 * Assumes vcpu_load() was already called.
2403 int kvm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
2405 return kvm_x86_ops
->get_msr(vcpu
, msr
);
2407 EXPORT_SYMBOL_GPL(kvm_get_msr
);
2409 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2412 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2413 unsigned bank_num
= mcg_cap
& 0xff;
2416 case MSR_IA32_P5_MC_ADDR
:
2417 case MSR_IA32_P5_MC_TYPE
:
2420 case MSR_IA32_MCG_CAP
:
2421 data
= vcpu
->arch
.mcg_cap
;
2423 case MSR_IA32_MCG_CTL
:
2424 if (!(mcg_cap
& MCG_CTL_P
))
2426 data
= vcpu
->arch
.mcg_ctl
;
2428 case MSR_IA32_MCG_STATUS
:
2429 data
= vcpu
->arch
.mcg_status
;
2432 if (msr
>= MSR_IA32_MC0_CTL
&&
2433 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2434 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2435 data
= vcpu
->arch
.mce_banks
[offset
];
2444 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2446 switch (msr_info
->index
) {
2447 case MSR_IA32_PLATFORM_ID
:
2448 case MSR_IA32_EBL_CR_POWERON
:
2449 case MSR_IA32_DEBUGCTLMSR
:
2450 case MSR_IA32_LASTBRANCHFROMIP
:
2451 case MSR_IA32_LASTBRANCHTOIP
:
2452 case MSR_IA32_LASTINTFROMIP
:
2453 case MSR_IA32_LASTINTTOIP
:
2455 case MSR_K8_TSEG_ADDR
:
2456 case MSR_K8_TSEG_MASK
:
2458 case MSR_VM_HSAVE_PA
:
2459 case MSR_K8_INT_PENDING_MSG
:
2460 case MSR_AMD64_NB_CFG
:
2461 case MSR_FAM10H_MMIO_CONF_BASE
:
2462 case MSR_AMD64_BU_CFG2
:
2463 case MSR_IA32_PERF_CTL
:
2464 case MSR_AMD64_DC_CFG
:
2467 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2468 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2469 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2470 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2471 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2472 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2475 case MSR_IA32_UCODE_REV
:
2476 msr_info
->data
= 0x100000000ULL
;
2479 case 0x200 ... 0x2ff:
2480 return kvm_mtrr_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2481 case 0xcd: /* fsb frequency */
2485 * MSR_EBC_FREQUENCY_ID
2486 * Conservative value valid for even the basic CPU models.
2487 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2488 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2489 * and 266MHz for model 3, or 4. Set Core Clock
2490 * Frequency to System Bus Frequency Ratio to 1 (bits
2491 * 31:24) even though these are only valid for CPU
2492 * models > 2, however guests may end up dividing or
2493 * multiplying by zero otherwise.
2495 case MSR_EBC_FREQUENCY_ID
:
2496 msr_info
->data
= 1 << 24;
2498 case MSR_IA32_APICBASE
:
2499 msr_info
->data
= kvm_get_apic_base(vcpu
);
2501 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2502 return kvm_x2apic_msr_read(vcpu
, msr_info
->index
, &msr_info
->data
);
2504 case MSR_IA32_TSCDEADLINE
:
2505 msr_info
->data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
2507 case MSR_IA32_TSC_ADJUST
:
2508 msr_info
->data
= (u64
)vcpu
->arch
.ia32_tsc_adjust_msr
;
2510 case MSR_IA32_MISC_ENABLE
:
2511 msr_info
->data
= vcpu
->arch
.ia32_misc_enable_msr
;
2513 case MSR_IA32_SMBASE
:
2514 if (!msr_info
->host_initiated
)
2516 msr_info
->data
= vcpu
->arch
.smbase
;
2518 case MSR_IA32_PERF_STATUS
:
2519 /* TSC increment by tick */
2520 msr_info
->data
= 1000ULL;
2521 /* CPU multiplier */
2522 msr_info
->data
|= (((uint64_t)4ULL) << 40);
2525 msr_info
->data
= vcpu
->arch
.efer
;
2527 case MSR_KVM_WALL_CLOCK
:
2528 case MSR_KVM_WALL_CLOCK_NEW
:
2529 msr_info
->data
= vcpu
->kvm
->arch
.wall_clock
;
2531 case MSR_KVM_SYSTEM_TIME
:
2532 case MSR_KVM_SYSTEM_TIME_NEW
:
2533 msr_info
->data
= vcpu
->arch
.time
;
2535 case MSR_KVM_ASYNC_PF_EN
:
2536 msr_info
->data
= vcpu
->arch
.apf
.msr_val
;
2538 case MSR_KVM_STEAL_TIME
:
2539 msr_info
->data
= vcpu
->arch
.st
.msr_val
;
2541 case MSR_KVM_PV_EOI_EN
:
2542 msr_info
->data
= vcpu
->arch
.pv_eoi
.msr_val
;
2544 case MSR_IA32_P5_MC_ADDR
:
2545 case MSR_IA32_P5_MC_TYPE
:
2546 case MSR_IA32_MCG_CAP
:
2547 case MSR_IA32_MCG_CTL
:
2548 case MSR_IA32_MCG_STATUS
:
2549 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2550 return get_msr_mce(vcpu
, msr_info
->index
, &msr_info
->data
);
2551 case MSR_K7_CLK_CTL
:
2553 * Provide expected ramp-up count for K7. All other
2554 * are set to zero, indicating minimum divisors for
2557 * This prevents guest kernels on AMD host with CPU
2558 * type 6, model 8 and higher from exploding due to
2559 * the rdmsr failing.
2561 msr_info
->data
= 0x20000000;
2563 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2564 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2565 case HV_X64_MSR_CRASH_CTL
:
2566 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2567 return kvm_hv_get_msr_common(vcpu
,
2568 msr_info
->index
, &msr_info
->data
);
2570 case MSR_IA32_BBL_CR_CTL3
:
2571 /* This legacy MSR exists but isn't fully documented in current
2572 * silicon. It is however accessed by winxp in very narrow
2573 * scenarios where it sets bit #19, itself documented as
2574 * a "reserved" bit. Best effort attempt to source coherent
2575 * read data here should the balance of the register be
2576 * interpreted by the guest:
2578 * L2 cache control register 3: 64GB range, 256KB size,
2579 * enabled, latency 0x1, configured
2581 msr_info
->data
= 0xbe702111;
2583 case MSR_AMD64_OSVW_ID_LENGTH
:
2584 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2586 msr_info
->data
= vcpu
->arch
.osvw
.length
;
2588 case MSR_AMD64_OSVW_STATUS
:
2589 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2591 msr_info
->data
= vcpu
->arch
.osvw
.status
;
2593 case MSR_PLATFORM_INFO
:
2594 msr_info
->data
= vcpu
->arch
.msr_platform_info
;
2596 case MSR_MISC_FEATURES_ENABLES
:
2597 msr_info
->data
= vcpu
->arch
.msr_misc_features_enables
;
2600 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2601 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2603 vcpu_debug_ratelimited(vcpu
, "unhandled rdmsr: 0x%x\n",
2607 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr_info
->index
);
2614 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2617 * Read or write a bunch of msrs. All parameters are kernel addresses.
2619 * @return number of msrs set successfully.
2621 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2622 struct kvm_msr_entry
*entries
,
2623 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2624 unsigned index
, u64
*data
))
2628 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2629 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2636 * Read or write a bunch of msrs. Parameters are user addresses.
2638 * @return number of msrs set successfully.
2640 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2641 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2642 unsigned index
, u64
*data
),
2645 struct kvm_msrs msrs
;
2646 struct kvm_msr_entry
*entries
;
2651 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2655 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2658 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2659 entries
= memdup_user(user_msrs
->entries
, size
);
2660 if (IS_ERR(entries
)) {
2661 r
= PTR_ERR(entries
);
2665 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2670 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2681 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
2686 case KVM_CAP_IRQCHIP
:
2688 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2689 case KVM_CAP_SET_TSS_ADDR
:
2690 case KVM_CAP_EXT_CPUID
:
2691 case KVM_CAP_EXT_EMUL_CPUID
:
2692 case KVM_CAP_CLOCKSOURCE
:
2694 case KVM_CAP_NOP_IO_DELAY
:
2695 case KVM_CAP_MP_STATE
:
2696 case KVM_CAP_SYNC_MMU
:
2697 case KVM_CAP_USER_NMI
:
2698 case KVM_CAP_REINJECT_CONTROL
:
2699 case KVM_CAP_IRQ_INJECT_STATUS
:
2700 case KVM_CAP_IOEVENTFD
:
2701 case KVM_CAP_IOEVENTFD_NO_LENGTH
:
2703 case KVM_CAP_PIT_STATE2
:
2704 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2705 case KVM_CAP_XEN_HVM
:
2706 case KVM_CAP_VCPU_EVENTS
:
2707 case KVM_CAP_HYPERV
:
2708 case KVM_CAP_HYPERV_VAPIC
:
2709 case KVM_CAP_HYPERV_SPIN
:
2710 case KVM_CAP_HYPERV_SYNIC
:
2711 case KVM_CAP_HYPERV_SYNIC2
:
2712 case KVM_CAP_HYPERV_VP_INDEX
:
2713 case KVM_CAP_PCI_SEGMENT
:
2714 case KVM_CAP_DEBUGREGS
:
2715 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2717 case KVM_CAP_ASYNC_PF
:
2718 case KVM_CAP_GET_TSC_KHZ
:
2719 case KVM_CAP_KVMCLOCK_CTRL
:
2720 case KVM_CAP_READONLY_MEM
:
2721 case KVM_CAP_HYPERV_TIME
:
2722 case KVM_CAP_IOAPIC_POLARITY_IGNORED
:
2723 case KVM_CAP_TSC_DEADLINE_TIMER
:
2724 case KVM_CAP_ENABLE_CAP_VM
:
2725 case KVM_CAP_DISABLE_QUIRKS
:
2726 case KVM_CAP_SET_BOOT_CPU_ID
:
2727 case KVM_CAP_SPLIT_IRQCHIP
:
2728 case KVM_CAP_IMMEDIATE_EXIT
:
2729 case KVM_CAP_GET_MSR_FEATURES
:
2732 case KVM_CAP_ADJUST_CLOCK
:
2733 r
= KVM_CLOCK_TSC_STABLE
;
2735 case KVM_CAP_X86_GUEST_MWAIT
:
2736 r
= kvm_mwait_in_guest();
2738 case KVM_CAP_X86_SMM
:
2739 /* SMBASE is usually relocated above 1M on modern chipsets,
2740 * and SMM handlers might indeed rely on 4G segment limits,
2741 * so do not report SMM to be available if real mode is
2742 * emulated via vm86 mode. Still, do not go to great lengths
2743 * to avoid userspace's usage of the feature, because it is a
2744 * fringe case that is not enabled except via specific settings
2745 * of the module parameters.
2747 r
= kvm_x86_ops
->has_emulated_msr(MSR_IA32_SMBASE
);
2750 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2752 case KVM_CAP_NR_VCPUS
:
2753 r
= KVM_SOFT_MAX_VCPUS
;
2755 case KVM_CAP_MAX_VCPUS
:
2758 case KVM_CAP_NR_MEMSLOTS
:
2759 r
= KVM_USER_MEM_SLOTS
;
2761 case KVM_CAP_PV_MMU
: /* obsolete */
2765 r
= KVM_MAX_MCE_BANKS
;
2768 r
= boot_cpu_has(X86_FEATURE_XSAVE
);
2770 case KVM_CAP_TSC_CONTROL
:
2771 r
= kvm_has_tsc_control
;
2773 case KVM_CAP_X2APIC_API
:
2774 r
= KVM_X2APIC_API_VALID_FLAGS
;
2784 long kvm_arch_dev_ioctl(struct file
*filp
,
2785 unsigned int ioctl
, unsigned long arg
)
2787 void __user
*argp
= (void __user
*)arg
;
2791 case KVM_GET_MSR_INDEX_LIST
: {
2792 struct kvm_msr_list __user
*user_msr_list
= argp
;
2793 struct kvm_msr_list msr_list
;
2797 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2800 msr_list
.nmsrs
= num_msrs_to_save
+ num_emulated_msrs
;
2801 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2804 if (n
< msr_list
.nmsrs
)
2807 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2808 num_msrs_to_save
* sizeof(u32
)))
2810 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2812 num_emulated_msrs
* sizeof(u32
)))
2817 case KVM_GET_SUPPORTED_CPUID
:
2818 case KVM_GET_EMULATED_CPUID
: {
2819 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2820 struct kvm_cpuid2 cpuid
;
2823 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2826 r
= kvm_dev_ioctl_get_cpuid(&cpuid
, cpuid_arg
->entries
,
2832 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2837 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2839 if (copy_to_user(argp
, &kvm_mce_cap_supported
,
2840 sizeof(kvm_mce_cap_supported
)))
2844 case KVM_GET_MSR_FEATURE_INDEX_LIST
: {
2845 struct kvm_msr_list __user
*user_msr_list
= argp
;
2846 struct kvm_msr_list msr_list
;
2850 if (copy_from_user(&msr_list
, user_msr_list
, sizeof(msr_list
)))
2853 msr_list
.nmsrs
= num_msr_based_features
;
2854 if (copy_to_user(user_msr_list
, &msr_list
, sizeof(msr_list
)))
2857 if (n
< msr_list
.nmsrs
)
2860 if (copy_to_user(user_msr_list
->indices
, &msr_based_features
,
2861 num_msr_based_features
* sizeof(u32
)))
2867 r
= msr_io(NULL
, argp
, do_get_msr_feature
, 1);
2877 static void wbinvd_ipi(void *garbage
)
2882 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2884 return kvm_arch_has_noncoherent_dma(vcpu
->kvm
);
2887 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2889 /* Address WBINVD may be executed by guest */
2890 if (need_emulate_wbinvd(vcpu
)) {
2891 if (kvm_x86_ops
->has_wbinvd_exit())
2892 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2893 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2894 smp_call_function_single(vcpu
->cpu
,
2895 wbinvd_ipi
, NULL
, 1);
2898 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2900 /* Apply any externally detected TSC adjustments (due to suspend) */
2901 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2902 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2903 vcpu
->arch
.tsc_offset_adjustment
= 0;
2904 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2907 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2908 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2909 rdtsc() - vcpu
->arch
.last_host_tsc
;
2911 mark_tsc_unstable("KVM discovered backwards TSC");
2913 if (check_tsc_unstable()) {
2914 u64 offset
= kvm_compute_tsc_offset(vcpu
,
2915 vcpu
->arch
.last_guest_tsc
);
2916 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
2917 vcpu
->arch
.tsc_catchup
= 1;
2920 if (kvm_lapic_hv_timer_in_use(vcpu
))
2921 kvm_lapic_restart_hv_timer(vcpu
);
2924 * On a host with synchronized TSC, there is no need to update
2925 * kvmclock on vcpu->cpu migration
2927 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
2928 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2929 if (vcpu
->cpu
!= cpu
)
2930 kvm_make_request(KVM_REQ_MIGRATE_TIMER
, vcpu
);
2934 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2937 static void kvm_steal_time_set_preempted(struct kvm_vcpu
*vcpu
)
2939 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
2942 vcpu
->arch
.st
.steal
.preempted
= 1;
2944 kvm_write_guest_offset_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2945 &vcpu
->arch
.st
.steal
.preempted
,
2946 offsetof(struct kvm_steal_time
, preempted
),
2947 sizeof(vcpu
->arch
.st
.steal
.preempted
));
2950 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2954 if (vcpu
->preempted
)
2955 vcpu
->arch
.preempted_in_kernel
= !kvm_x86_ops
->get_cpl(vcpu
);
2958 * Disable page faults because we're in atomic context here.
2959 * kvm_write_guest_offset_cached() would call might_fault()
2960 * that relies on pagefault_disable() to tell if there's a
2961 * bug. NOTE: the write to guest memory may not go through if
2962 * during postcopy live migration or if there's heavy guest
2965 pagefault_disable();
2967 * kvm_memslots() will be called by
2968 * kvm_write_guest_offset_cached() so take the srcu lock.
2970 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2971 kvm_steal_time_set_preempted(vcpu
);
2972 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2974 kvm_x86_ops
->vcpu_put(vcpu
);
2975 kvm_put_guest_fpu(vcpu
);
2976 vcpu
->arch
.last_host_tsc
= rdtsc();
2978 * If userspace has set any breakpoints or watchpoints, dr6 is restored
2979 * on every vmexit, but if not, we might have a stale dr6 from the
2980 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
2985 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
2986 struct kvm_lapic_state
*s
)
2988 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
2989 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
2991 return kvm_apic_get_state(vcpu
, s
);
2994 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
2995 struct kvm_lapic_state
*s
)
2999 r
= kvm_apic_set_state(vcpu
, s
);
3002 update_cr8_intercept(vcpu
);
3007 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu
*vcpu
)
3009 return (!lapic_in_kernel(vcpu
) ||
3010 kvm_apic_accept_pic_intr(vcpu
));
3014 * if userspace requested an interrupt window, check that the
3015 * interrupt window is open.
3017 * No need to exit to userspace if we already have an interrupt queued.
3019 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu
*vcpu
)
3021 return kvm_arch_interrupt_allowed(vcpu
) &&
3022 !kvm_cpu_has_interrupt(vcpu
) &&
3023 !kvm_event_needs_reinjection(vcpu
) &&
3024 kvm_cpu_accept_dm_intr(vcpu
);
3027 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
3028 struct kvm_interrupt
*irq
)
3030 if (irq
->irq
>= KVM_NR_INTERRUPTS
)
3033 if (!irqchip_in_kernel(vcpu
->kvm
)) {
3034 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
3035 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3040 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3041 * fail for in-kernel 8259.
3043 if (pic_in_kernel(vcpu
->kvm
))
3046 if (vcpu
->arch
.pending_external_vector
!= -1)
3049 vcpu
->arch
.pending_external_vector
= irq
->irq
;
3050 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3054 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
3056 kvm_inject_nmi(vcpu
);
3061 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu
*vcpu
)
3063 kvm_make_request(KVM_REQ_SMI
, vcpu
);
3068 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
3069 struct kvm_tpr_access_ctl
*tac
)
3073 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
3077 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
3081 unsigned bank_num
= mcg_cap
& 0xff, bank
;
3084 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
3086 if (mcg_cap
& ~(kvm_mce_cap_supported
| 0xff | 0xff0000))
3089 vcpu
->arch
.mcg_cap
= mcg_cap
;
3090 /* Init IA32_MCG_CTL to all 1s */
3091 if (mcg_cap
& MCG_CTL_P
)
3092 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
3093 /* Init IA32_MCi_CTL to all 1s */
3094 for (bank
= 0; bank
< bank_num
; bank
++)
3095 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
3097 if (kvm_x86_ops
->setup_mce
)
3098 kvm_x86_ops
->setup_mce(vcpu
);
3103 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
3104 struct kvm_x86_mce
*mce
)
3106 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
3107 unsigned bank_num
= mcg_cap
& 0xff;
3108 u64
*banks
= vcpu
->arch
.mce_banks
;
3110 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
3113 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3114 * reporting is disabled
3116 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
3117 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
3119 banks
+= 4 * mce
->bank
;
3121 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3122 * reporting is disabled for the bank
3124 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
3126 if (mce
->status
& MCI_STATUS_UC
) {
3127 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
3128 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
3129 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
3132 if (banks
[1] & MCI_STATUS_VAL
)
3133 mce
->status
|= MCI_STATUS_OVER
;
3134 banks
[2] = mce
->addr
;
3135 banks
[3] = mce
->misc
;
3136 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
3137 banks
[1] = mce
->status
;
3138 kvm_queue_exception(vcpu
, MC_VECTOR
);
3139 } else if (!(banks
[1] & MCI_STATUS_VAL
)
3140 || !(banks
[1] & MCI_STATUS_UC
)) {
3141 if (banks
[1] & MCI_STATUS_VAL
)
3142 mce
->status
|= MCI_STATUS_OVER
;
3143 banks
[2] = mce
->addr
;
3144 banks
[3] = mce
->misc
;
3145 banks
[1] = mce
->status
;
3147 banks
[1] |= MCI_STATUS_OVER
;
3151 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
3152 struct kvm_vcpu_events
*events
)
3156 * FIXME: pass injected and pending separately. This is only
3157 * needed for nested virtualization, whose state cannot be
3158 * migrated yet. For now we can combine them.
3160 events
->exception
.injected
=
3161 (vcpu
->arch
.exception
.pending
||
3162 vcpu
->arch
.exception
.injected
) &&
3163 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
3164 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
3165 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
3166 events
->exception
.pad
= 0;
3167 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
3169 events
->interrupt
.injected
=
3170 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
3171 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
3172 events
->interrupt
.soft
= 0;
3173 events
->interrupt
.shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
3175 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
3176 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
3177 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
3178 events
->nmi
.pad
= 0;
3180 events
->sipi_vector
= 0; /* never valid when reporting to user space */
3182 events
->smi
.smm
= is_smm(vcpu
);
3183 events
->smi
.pending
= vcpu
->arch
.smi_pending
;
3184 events
->smi
.smm_inside_nmi
=
3185 !!(vcpu
->arch
.hflags
& HF_SMM_INSIDE_NMI_MASK
);
3186 events
->smi
.latched_init
= kvm_lapic_latched_init(vcpu
);
3188 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
3189 | KVM_VCPUEVENT_VALID_SHADOW
3190 | KVM_VCPUEVENT_VALID_SMM
);
3191 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
3194 static void kvm_set_hflags(struct kvm_vcpu
*vcpu
, unsigned emul_flags
);
3196 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
3197 struct kvm_vcpu_events
*events
)
3199 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3200 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3201 | KVM_VCPUEVENT_VALID_SHADOW
3202 | KVM_VCPUEVENT_VALID_SMM
))
3205 if (events
->exception
.injected
&&
3206 (events
->exception
.nr
> 31 || events
->exception
.nr
== NMI_VECTOR
||
3207 is_guest_mode(vcpu
)))
3210 /* INITs are latched while in SMM */
3211 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
&&
3212 (events
->smi
.smm
|| events
->smi
.pending
) &&
3213 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
)
3217 vcpu
->arch
.exception
.injected
= false;
3218 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
3219 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
3220 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
3221 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
3223 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
3224 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
3225 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
3226 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
3227 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
3228 events
->interrupt
.shadow
);
3230 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
3231 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
3232 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
3233 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
3235 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
&&
3236 lapic_in_kernel(vcpu
))
3237 vcpu
->arch
.apic
->sipi_vector
= events
->sipi_vector
;
3239 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
) {
3240 u32 hflags
= vcpu
->arch
.hflags
;
3241 if (events
->smi
.smm
)
3242 hflags
|= HF_SMM_MASK
;
3244 hflags
&= ~HF_SMM_MASK
;
3245 kvm_set_hflags(vcpu
, hflags
);
3247 vcpu
->arch
.smi_pending
= events
->smi
.pending
;
3249 if (events
->smi
.smm
) {
3250 if (events
->smi
.smm_inside_nmi
)
3251 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
3253 vcpu
->arch
.hflags
&= ~HF_SMM_INSIDE_NMI_MASK
;
3254 if (lapic_in_kernel(vcpu
)) {
3255 if (events
->smi
.latched_init
)
3256 set_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3258 clear_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3263 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3268 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
3269 struct kvm_debugregs
*dbgregs
)
3273 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
3274 kvm_get_dr(vcpu
, 6, &val
);
3276 dbgregs
->dr7
= vcpu
->arch
.dr7
;
3278 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
3281 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
3282 struct kvm_debugregs
*dbgregs
)
3287 if (dbgregs
->dr6
& ~0xffffffffull
)
3289 if (dbgregs
->dr7
& ~0xffffffffull
)
3292 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
3293 kvm_update_dr0123(vcpu
);
3294 vcpu
->arch
.dr6
= dbgregs
->dr6
;
3295 kvm_update_dr6(vcpu
);
3296 vcpu
->arch
.dr7
= dbgregs
->dr7
;
3297 kvm_update_dr7(vcpu
);
3302 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3304 static void fill_xsave(u8
*dest
, struct kvm_vcpu
*vcpu
)
3306 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3307 u64 xstate_bv
= xsave
->header
.xfeatures
;
3311 * Copy legacy XSAVE area, to avoid complications with CPUID
3312 * leaves 0 and 1 in the loop below.
3314 memcpy(dest
, xsave
, XSAVE_HDR_OFFSET
);
3317 xstate_bv
&= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FPSSE
;
3318 *(u64
*)(dest
+ XSAVE_HDR_OFFSET
) = xstate_bv
;
3321 * Copy each region from the possibly compacted offset to the
3322 * non-compacted offset.
3324 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3326 u64 feature
= valid
& -valid
;
3327 int index
= fls64(feature
) - 1;
3328 void *src
= get_xsave_addr(xsave
, feature
);
3331 u32 size
, offset
, ecx
, edx
;
3332 cpuid_count(XSTATE_CPUID
, index
,
3333 &size
, &offset
, &ecx
, &edx
);
3334 if (feature
== XFEATURE_MASK_PKRU
)
3335 memcpy(dest
+ offset
, &vcpu
->arch
.pkru
,
3336 sizeof(vcpu
->arch
.pkru
));
3338 memcpy(dest
+ offset
, src
, size
);
3346 static void load_xsave(struct kvm_vcpu
*vcpu
, u8
*src
)
3348 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3349 u64 xstate_bv
= *(u64
*)(src
+ XSAVE_HDR_OFFSET
);
3353 * Copy legacy XSAVE area, to avoid complications with CPUID
3354 * leaves 0 and 1 in the loop below.
3356 memcpy(xsave
, src
, XSAVE_HDR_OFFSET
);
3358 /* Set XSTATE_BV and possibly XCOMP_BV. */
3359 xsave
->header
.xfeatures
= xstate_bv
;
3360 if (boot_cpu_has(X86_FEATURE_XSAVES
))
3361 xsave
->header
.xcomp_bv
= host_xcr0
| XSTATE_COMPACTION_ENABLED
;
3364 * Copy each region from the non-compacted offset to the
3365 * possibly compacted offset.
3367 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3369 u64 feature
= valid
& -valid
;
3370 int index
= fls64(feature
) - 1;
3371 void *dest
= get_xsave_addr(xsave
, feature
);
3374 u32 size
, offset
, ecx
, edx
;
3375 cpuid_count(XSTATE_CPUID
, index
,
3376 &size
, &offset
, &ecx
, &edx
);
3377 if (feature
== XFEATURE_MASK_PKRU
)
3378 memcpy(&vcpu
->arch
.pkru
, src
+ offset
,
3379 sizeof(vcpu
->arch
.pkru
));
3381 memcpy(dest
, src
+ offset
, size
);
3388 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
3389 struct kvm_xsave
*guest_xsave
)
3391 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3392 memset(guest_xsave
, 0, sizeof(struct kvm_xsave
));
3393 fill_xsave((u8
*) guest_xsave
->region
, vcpu
);
3395 memcpy(guest_xsave
->region
,
3396 &vcpu
->arch
.guest_fpu
.state
.fxsave
,
3397 sizeof(struct fxregs_state
));
3398 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
3399 XFEATURE_MASK_FPSSE
;
3403 #define XSAVE_MXCSR_OFFSET 24
3405 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
3406 struct kvm_xsave
*guest_xsave
)
3409 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
3410 u32 mxcsr
= *(u32
*)&guest_xsave
->region
[XSAVE_MXCSR_OFFSET
/ sizeof(u32
)];
3412 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3414 * Here we allow setting states that are not present in
3415 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3416 * with old userspace.
3418 if (xstate_bv
& ~kvm_supported_xcr0() ||
3419 mxcsr
& ~mxcsr_feature_mask
)
3421 load_xsave(vcpu
, (u8
*)guest_xsave
->region
);
3423 if (xstate_bv
& ~XFEATURE_MASK_FPSSE
||
3424 mxcsr
& ~mxcsr_feature_mask
)
3426 memcpy(&vcpu
->arch
.guest_fpu
.state
.fxsave
,
3427 guest_xsave
->region
, sizeof(struct fxregs_state
));
3432 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
3433 struct kvm_xcrs
*guest_xcrs
)
3435 if (!boot_cpu_has(X86_FEATURE_XSAVE
)) {
3436 guest_xcrs
->nr_xcrs
= 0;
3440 guest_xcrs
->nr_xcrs
= 1;
3441 guest_xcrs
->flags
= 0;
3442 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
3443 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
3446 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
3447 struct kvm_xcrs
*guest_xcrs
)
3451 if (!boot_cpu_has(X86_FEATURE_XSAVE
))
3454 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
3457 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
3458 /* Only support XCR0 currently */
3459 if (guest_xcrs
->xcrs
[i
].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
3460 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
3461 guest_xcrs
->xcrs
[i
].value
);
3470 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3471 * stopped by the hypervisor. This function will be called from the host only.
3472 * EINVAL is returned when the host attempts to set the flag for a guest that
3473 * does not support pv clocks.
3475 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
3477 if (!vcpu
->arch
.pv_time_enabled
)
3479 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
3480 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
3484 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
3485 struct kvm_enable_cap
*cap
)
3491 case KVM_CAP_HYPERV_SYNIC2
:
3494 case KVM_CAP_HYPERV_SYNIC
:
3495 if (!irqchip_in_kernel(vcpu
->kvm
))
3497 return kvm_hv_activate_synic(vcpu
, cap
->cap
==
3498 KVM_CAP_HYPERV_SYNIC2
);
3504 long kvm_arch_vcpu_ioctl(struct file
*filp
,
3505 unsigned int ioctl
, unsigned long arg
)
3507 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3508 void __user
*argp
= (void __user
*)arg
;
3511 struct kvm_lapic_state
*lapic
;
3512 struct kvm_xsave
*xsave
;
3513 struct kvm_xcrs
*xcrs
;
3519 case KVM_GET_LAPIC
: {
3521 if (!lapic_in_kernel(vcpu
))
3523 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
3528 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
3532 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
3537 case KVM_SET_LAPIC
: {
3539 if (!lapic_in_kernel(vcpu
))
3541 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
3542 if (IS_ERR(u
.lapic
))
3543 return PTR_ERR(u
.lapic
);
3545 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
3548 case KVM_INTERRUPT
: {
3549 struct kvm_interrupt irq
;
3552 if (copy_from_user(&irq
, argp
, sizeof irq
))
3554 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
3558 r
= kvm_vcpu_ioctl_nmi(vcpu
);
3562 r
= kvm_vcpu_ioctl_smi(vcpu
);
3565 case KVM_SET_CPUID
: {
3566 struct kvm_cpuid __user
*cpuid_arg
= argp
;
3567 struct kvm_cpuid cpuid
;
3570 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3572 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3575 case KVM_SET_CPUID2
: {
3576 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3577 struct kvm_cpuid2 cpuid
;
3580 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3582 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
3583 cpuid_arg
->entries
);
3586 case KVM_GET_CPUID2
: {
3587 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3588 struct kvm_cpuid2 cpuid
;
3591 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3593 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
3594 cpuid_arg
->entries
);
3598 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
3603 case KVM_GET_MSRS
: {
3604 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3605 r
= msr_io(vcpu
, argp
, do_get_msr
, 1);
3606 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3609 case KVM_SET_MSRS
: {
3610 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3611 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
3612 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3615 case KVM_TPR_ACCESS_REPORTING
: {
3616 struct kvm_tpr_access_ctl tac
;
3619 if (copy_from_user(&tac
, argp
, sizeof tac
))
3621 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
3625 if (copy_to_user(argp
, &tac
, sizeof tac
))
3630 case KVM_SET_VAPIC_ADDR
: {
3631 struct kvm_vapic_addr va
;
3635 if (!lapic_in_kernel(vcpu
))
3638 if (copy_from_user(&va
, argp
, sizeof va
))
3640 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3641 r
= kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
3642 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3645 case KVM_X86_SETUP_MCE
: {
3649 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
3651 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
3654 case KVM_X86_SET_MCE
: {
3655 struct kvm_x86_mce mce
;
3658 if (copy_from_user(&mce
, argp
, sizeof mce
))
3660 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
3663 case KVM_GET_VCPU_EVENTS
: {
3664 struct kvm_vcpu_events events
;
3666 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
3669 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
3674 case KVM_SET_VCPU_EVENTS
: {
3675 struct kvm_vcpu_events events
;
3678 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
3681 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
3684 case KVM_GET_DEBUGREGS
: {
3685 struct kvm_debugregs dbgregs
;
3687 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
3690 if (copy_to_user(argp
, &dbgregs
,
3691 sizeof(struct kvm_debugregs
)))
3696 case KVM_SET_DEBUGREGS
: {
3697 struct kvm_debugregs dbgregs
;
3700 if (copy_from_user(&dbgregs
, argp
,
3701 sizeof(struct kvm_debugregs
)))
3704 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3707 case KVM_GET_XSAVE
: {
3708 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3713 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3716 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3721 case KVM_SET_XSAVE
: {
3722 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
3723 if (IS_ERR(u
.xsave
))
3724 return PTR_ERR(u
.xsave
);
3726 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3729 case KVM_GET_XCRS
: {
3730 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3735 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3738 if (copy_to_user(argp
, u
.xcrs
,
3739 sizeof(struct kvm_xcrs
)))
3744 case KVM_SET_XCRS
: {
3745 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
3747 return PTR_ERR(u
.xcrs
);
3749 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3752 case KVM_SET_TSC_KHZ
: {
3756 user_tsc_khz
= (u32
)arg
;
3758 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3761 if (user_tsc_khz
== 0)
3762 user_tsc_khz
= tsc_khz
;
3764 if (!kvm_set_tsc_khz(vcpu
, user_tsc_khz
))
3769 case KVM_GET_TSC_KHZ
: {
3770 r
= vcpu
->arch
.virtual_tsc_khz
;
3773 case KVM_KVMCLOCK_CTRL
: {
3774 r
= kvm_set_guest_paused(vcpu
);
3777 case KVM_ENABLE_CAP
: {
3778 struct kvm_enable_cap cap
;
3781 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3783 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
3794 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
3796 return VM_FAULT_SIGBUS
;
3799 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3803 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3805 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3809 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3812 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3816 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3817 u32 kvm_nr_mmu_pages
)
3819 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3822 mutex_lock(&kvm
->slots_lock
);
3824 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3825 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3827 mutex_unlock(&kvm
->slots_lock
);
3831 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3833 return kvm
->arch
.n_max_mmu_pages
;
3836 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3838 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
3842 switch (chip
->chip_id
) {
3843 case KVM_IRQCHIP_PIC_MASTER
:
3844 memcpy(&chip
->chip
.pic
, &pic
->pics
[0],
3845 sizeof(struct kvm_pic_state
));
3847 case KVM_IRQCHIP_PIC_SLAVE
:
3848 memcpy(&chip
->chip
.pic
, &pic
->pics
[1],
3849 sizeof(struct kvm_pic_state
));
3851 case KVM_IRQCHIP_IOAPIC
:
3852 kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3861 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3863 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
3867 switch (chip
->chip_id
) {
3868 case KVM_IRQCHIP_PIC_MASTER
:
3869 spin_lock(&pic
->lock
);
3870 memcpy(&pic
->pics
[0], &chip
->chip
.pic
,
3871 sizeof(struct kvm_pic_state
));
3872 spin_unlock(&pic
->lock
);
3874 case KVM_IRQCHIP_PIC_SLAVE
:
3875 spin_lock(&pic
->lock
);
3876 memcpy(&pic
->pics
[1], &chip
->chip
.pic
,
3877 sizeof(struct kvm_pic_state
));
3878 spin_unlock(&pic
->lock
);
3880 case KVM_IRQCHIP_IOAPIC
:
3881 kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3887 kvm_pic_update_irq(pic
);
3891 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3893 struct kvm_kpit_state
*kps
= &kvm
->arch
.vpit
->pit_state
;
3895 BUILD_BUG_ON(sizeof(*ps
) != sizeof(kps
->channels
));
3897 mutex_lock(&kps
->lock
);
3898 memcpy(ps
, &kps
->channels
, sizeof(*ps
));
3899 mutex_unlock(&kps
->lock
);
3903 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3906 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3908 mutex_lock(&pit
->pit_state
.lock
);
3909 memcpy(&pit
->pit_state
.channels
, ps
, sizeof(*ps
));
3910 for (i
= 0; i
< 3; i
++)
3911 kvm_pit_load_count(pit
, i
, ps
->channels
[i
].count
, 0);
3912 mutex_unlock(&pit
->pit_state
.lock
);
3916 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3918 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3919 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3920 sizeof(ps
->channels
));
3921 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3922 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3923 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3927 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3931 u32 prev_legacy
, cur_legacy
;
3932 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3934 mutex_lock(&pit
->pit_state
.lock
);
3935 prev_legacy
= pit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3936 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3937 if (!prev_legacy
&& cur_legacy
)
3939 memcpy(&pit
->pit_state
.channels
, &ps
->channels
,
3940 sizeof(pit
->pit_state
.channels
));
3941 pit
->pit_state
.flags
= ps
->flags
;
3942 for (i
= 0; i
< 3; i
++)
3943 kvm_pit_load_count(pit
, i
, pit
->pit_state
.channels
[i
].count
,
3945 mutex_unlock(&pit
->pit_state
.lock
);
3949 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3950 struct kvm_reinject_control
*control
)
3952 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3957 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3958 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3959 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3961 mutex_lock(&pit
->pit_state
.lock
);
3962 kvm_pit_set_reinject(pit
, control
->pit_reinject
);
3963 mutex_unlock(&pit
->pit_state
.lock
);
3969 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3970 * @kvm: kvm instance
3971 * @log: slot id and address to which we copy the log
3973 * Steps 1-4 below provide general overview of dirty page logging. See
3974 * kvm_get_dirty_log_protect() function description for additional details.
3976 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3977 * always flush the TLB (step 4) even if previous step failed and the dirty
3978 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3979 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3980 * writes will be marked dirty for next log read.
3982 * 1. Take a snapshot of the bit and clear it if needed.
3983 * 2. Write protect the corresponding page.
3984 * 3. Copy the snapshot to the userspace.
3985 * 4. Flush TLB's if needed.
3987 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
3989 bool is_dirty
= false;
3992 mutex_lock(&kvm
->slots_lock
);
3995 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3997 if (kvm_x86_ops
->flush_log_dirty
)
3998 kvm_x86_ops
->flush_log_dirty(kvm
);
4000 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
4003 * All the TLBs can be flushed out of mmu lock, see the comments in
4004 * kvm_mmu_slot_remove_write_access().
4006 lockdep_assert_held(&kvm
->slots_lock
);
4008 kvm_flush_remote_tlbs(kvm
);
4010 mutex_unlock(&kvm
->slots_lock
);
4014 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
4017 if (!irqchip_in_kernel(kvm
))
4020 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
4021 irq_event
->irq
, irq_event
->level
,
4026 static int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4027 struct kvm_enable_cap
*cap
)
4035 case KVM_CAP_DISABLE_QUIRKS
:
4036 kvm
->arch
.disabled_quirks
= cap
->args
[0];
4039 case KVM_CAP_SPLIT_IRQCHIP
: {
4040 mutex_lock(&kvm
->lock
);
4042 if (cap
->args
[0] > MAX_NR_RESERVED_IOAPIC_PINS
)
4043 goto split_irqchip_unlock
;
4045 if (irqchip_in_kernel(kvm
))
4046 goto split_irqchip_unlock
;
4047 if (kvm
->created_vcpus
)
4048 goto split_irqchip_unlock
;
4049 r
= kvm_setup_empty_irq_routing(kvm
);
4051 goto split_irqchip_unlock
;
4052 /* Pairs with irqchip_in_kernel. */
4054 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_SPLIT
;
4055 kvm
->arch
.nr_reserved_ioapic_pins
= cap
->args
[0];
4057 split_irqchip_unlock
:
4058 mutex_unlock(&kvm
->lock
);
4061 case KVM_CAP_X2APIC_API
:
4063 if (cap
->args
[0] & ~KVM_X2APIC_API_VALID_FLAGS
)
4066 if (cap
->args
[0] & KVM_X2APIC_API_USE_32BIT_IDS
)
4067 kvm
->arch
.x2apic_format
= true;
4068 if (cap
->args
[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK
)
4069 kvm
->arch
.x2apic_broadcast_quirk_disabled
= true;
4080 long kvm_arch_vm_ioctl(struct file
*filp
,
4081 unsigned int ioctl
, unsigned long arg
)
4083 struct kvm
*kvm
= filp
->private_data
;
4084 void __user
*argp
= (void __user
*)arg
;
4087 * This union makes it completely explicit to gcc-3.x
4088 * that these two variables' stack usage should be
4089 * combined, not added together.
4092 struct kvm_pit_state ps
;
4093 struct kvm_pit_state2 ps2
;
4094 struct kvm_pit_config pit_config
;
4098 case KVM_SET_TSS_ADDR
:
4099 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
4101 case KVM_SET_IDENTITY_MAP_ADDR
: {
4105 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
4107 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
4110 case KVM_SET_NR_MMU_PAGES
:
4111 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
4113 case KVM_GET_NR_MMU_PAGES
:
4114 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
4116 case KVM_CREATE_IRQCHIP
: {
4117 mutex_lock(&kvm
->lock
);
4120 if (irqchip_in_kernel(kvm
))
4121 goto create_irqchip_unlock
;
4124 if (kvm
->created_vcpus
)
4125 goto create_irqchip_unlock
;
4127 r
= kvm_pic_init(kvm
);
4129 goto create_irqchip_unlock
;
4131 r
= kvm_ioapic_init(kvm
);
4133 kvm_pic_destroy(kvm
);
4134 goto create_irqchip_unlock
;
4137 r
= kvm_setup_default_irq_routing(kvm
);
4139 kvm_ioapic_destroy(kvm
);
4140 kvm_pic_destroy(kvm
);
4141 goto create_irqchip_unlock
;
4143 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4145 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_KERNEL
;
4146 create_irqchip_unlock
:
4147 mutex_unlock(&kvm
->lock
);
4150 case KVM_CREATE_PIT
:
4151 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
4153 case KVM_CREATE_PIT2
:
4155 if (copy_from_user(&u
.pit_config
, argp
,
4156 sizeof(struct kvm_pit_config
)))
4159 mutex_lock(&kvm
->lock
);
4162 goto create_pit_unlock
;
4164 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
4168 mutex_unlock(&kvm
->lock
);
4170 case KVM_GET_IRQCHIP
: {
4171 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4172 struct kvm_irqchip
*chip
;
4174 chip
= memdup_user(argp
, sizeof(*chip
));
4181 if (!irqchip_kernel(kvm
))
4182 goto get_irqchip_out
;
4183 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
4185 goto get_irqchip_out
;
4187 if (copy_to_user(argp
, chip
, sizeof *chip
))
4188 goto get_irqchip_out
;
4194 case KVM_SET_IRQCHIP
: {
4195 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4196 struct kvm_irqchip
*chip
;
4198 chip
= memdup_user(argp
, sizeof(*chip
));
4205 if (!irqchip_kernel(kvm
))
4206 goto set_irqchip_out
;
4207 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
4209 goto set_irqchip_out
;
4217 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
4220 if (!kvm
->arch
.vpit
)
4222 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
4226 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
4233 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
4236 if (!kvm
->arch
.vpit
)
4238 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
4241 case KVM_GET_PIT2
: {
4243 if (!kvm
->arch
.vpit
)
4245 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
4249 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
4254 case KVM_SET_PIT2
: {
4256 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
4259 if (!kvm
->arch
.vpit
)
4261 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
4264 case KVM_REINJECT_CONTROL
: {
4265 struct kvm_reinject_control control
;
4267 if (copy_from_user(&control
, argp
, sizeof(control
)))
4269 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
4272 case KVM_SET_BOOT_CPU_ID
:
4274 mutex_lock(&kvm
->lock
);
4275 if (kvm
->created_vcpus
)
4278 kvm
->arch
.bsp_vcpu_id
= arg
;
4279 mutex_unlock(&kvm
->lock
);
4281 case KVM_XEN_HVM_CONFIG
: {
4282 struct kvm_xen_hvm_config xhc
;
4284 if (copy_from_user(&xhc
, argp
, sizeof(xhc
)))
4289 memcpy(&kvm
->arch
.xen_hvm_config
, &xhc
, sizeof(xhc
));
4293 case KVM_SET_CLOCK
: {
4294 struct kvm_clock_data user_ns
;
4298 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
4307 * TODO: userspace has to take care of races with VCPU_RUN, so
4308 * kvm_gen_update_masterclock() can be cut down to locked
4309 * pvclock_update_vm_gtod_copy().
4311 kvm_gen_update_masterclock(kvm
);
4312 now_ns
= get_kvmclock_ns(kvm
);
4313 kvm
->arch
.kvmclock_offset
+= user_ns
.clock
- now_ns
;
4314 kvm_make_all_cpus_request(kvm
, KVM_REQ_CLOCK_UPDATE
);
4317 case KVM_GET_CLOCK
: {
4318 struct kvm_clock_data user_ns
;
4321 now_ns
= get_kvmclock_ns(kvm
);
4322 user_ns
.clock
= now_ns
;
4323 user_ns
.flags
= kvm
->arch
.use_master_clock
? KVM_CLOCK_TSC_STABLE
: 0;
4324 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
4327 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
4332 case KVM_ENABLE_CAP
: {
4333 struct kvm_enable_cap cap
;
4336 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4338 r
= kvm_vm_ioctl_enable_cap(kvm
, &cap
);
4348 static void kvm_init_msr_list(void)
4353 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
4354 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
4358 * Even MSRs that are valid in the host may not be exposed
4359 * to the guests in some cases.
4361 switch (msrs_to_save
[i
]) {
4362 case MSR_IA32_BNDCFGS
:
4363 if (!kvm_x86_ops
->mpx_supported())
4367 if (!kvm_x86_ops
->rdtscp_supported())
4375 msrs_to_save
[j
] = msrs_to_save
[i
];
4378 num_msrs_to_save
= j
;
4380 for (i
= j
= 0; i
< ARRAY_SIZE(emulated_msrs
); i
++) {
4381 if (!kvm_x86_ops
->has_emulated_msr(emulated_msrs
[i
]))
4385 emulated_msrs
[j
] = emulated_msrs
[i
];
4388 num_emulated_msrs
= j
;
4390 for (i
= j
= 0; i
< ARRAY_SIZE(msr_based_features
); i
++) {
4391 struct kvm_msr_entry msr
;
4393 msr
.index
= msr_based_features
[i
];
4394 if (kvm_x86_ops
->get_msr_feature(&msr
))
4398 msr_based_features
[j
] = msr_based_features
[i
];
4401 num_msr_based_features
= j
;
4404 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
4412 if (!(lapic_in_kernel(vcpu
) &&
4413 !kvm_iodevice_write(vcpu
, &vcpu
->arch
.apic
->dev
, addr
, n
, v
))
4414 && kvm_io_bus_write(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4425 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
4432 if (!(lapic_in_kernel(vcpu
) &&
4433 !kvm_iodevice_read(vcpu
, &vcpu
->arch
.apic
->dev
,
4435 && kvm_io_bus_read(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4437 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, v
);
4447 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
4448 struct kvm_segment
*var
, int seg
)
4450 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
4453 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
4454 struct kvm_segment
*var
, int seg
)
4456 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
4459 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
,
4460 struct x86_exception
*exception
)
4464 BUG_ON(!mmu_is_nested(vcpu
));
4466 /* NPT walks are always user-walks */
4467 access
|= PFERR_USER_MASK
;
4468 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, exception
);
4473 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
4474 struct x86_exception
*exception
)
4476 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4477 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4480 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
4481 struct x86_exception
*exception
)
4483 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4484 access
|= PFERR_FETCH_MASK
;
4485 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4488 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
4489 struct x86_exception
*exception
)
4491 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4492 access
|= PFERR_WRITE_MASK
;
4493 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4496 /* uses this to access any guest's mapped memory without checking CPL */
4497 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
4498 struct x86_exception
*exception
)
4500 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
4503 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
4504 struct kvm_vcpu
*vcpu
, u32 access
,
4505 struct x86_exception
*exception
)
4508 int r
= X86EMUL_CONTINUE
;
4511 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
4513 unsigned offset
= addr
& (PAGE_SIZE
-1);
4514 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4517 if (gpa
== UNMAPPED_GVA
)
4518 return X86EMUL_PROPAGATE_FAULT
;
4519 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, data
,
4522 r
= X86EMUL_IO_NEEDED
;
4534 /* used for instruction fetching */
4535 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
4536 gva_t addr
, void *val
, unsigned int bytes
,
4537 struct x86_exception
*exception
)
4539 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4540 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4544 /* Inline kvm_read_guest_virt_helper for speed. */
4545 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
|PFERR_FETCH_MASK
,
4547 if (unlikely(gpa
== UNMAPPED_GVA
))
4548 return X86EMUL_PROPAGATE_FAULT
;
4550 offset
= addr
& (PAGE_SIZE
-1);
4551 if (WARN_ON(offset
+ bytes
> PAGE_SIZE
))
4552 bytes
= (unsigned)PAGE_SIZE
- offset
;
4553 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, val
,
4555 if (unlikely(ret
< 0))
4556 return X86EMUL_IO_NEEDED
;
4558 return X86EMUL_CONTINUE
;
4561 int kvm_read_guest_virt(struct kvm_vcpu
*vcpu
,
4562 gva_t addr
, void *val
, unsigned int bytes
,
4563 struct x86_exception
*exception
)
4565 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4567 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
4570 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
4572 static int emulator_read_std(struct x86_emulate_ctxt
*ctxt
,
4573 gva_t addr
, void *val
, unsigned int bytes
,
4574 struct x86_exception
*exception
, bool system
)
4576 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4579 if (!system
&& kvm_x86_ops
->get_cpl(vcpu
) == 3)
4580 access
|= PFERR_USER_MASK
;
4582 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
, exception
);
4585 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt
*ctxt
,
4586 unsigned long addr
, void *val
, unsigned int bytes
)
4588 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4589 int r
= kvm_vcpu_read_guest(vcpu
, addr
, val
, bytes
);
4591 return r
< 0 ? X86EMUL_IO_NEEDED
: X86EMUL_CONTINUE
;
4594 static int kvm_write_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
4595 struct kvm_vcpu
*vcpu
, u32 access
,
4596 struct x86_exception
*exception
)
4599 int r
= X86EMUL_CONTINUE
;
4602 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
4605 unsigned offset
= addr
& (PAGE_SIZE
-1);
4606 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4609 if (gpa
== UNMAPPED_GVA
)
4610 return X86EMUL_PROPAGATE_FAULT
;
4611 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, data
, towrite
);
4613 r
= X86EMUL_IO_NEEDED
;
4625 static int emulator_write_std(struct x86_emulate_ctxt
*ctxt
, gva_t addr
, void *val
,
4626 unsigned int bytes
, struct x86_exception
*exception
,
4629 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4630 u32 access
= PFERR_WRITE_MASK
;
4632 if (!system
&& kvm_x86_ops
->get_cpl(vcpu
) == 3)
4633 access
|= PFERR_USER_MASK
;
4635 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
4639 int kvm_write_guest_virt_system(struct kvm_vcpu
*vcpu
, gva_t addr
, void *val
,
4640 unsigned int bytes
, struct x86_exception
*exception
)
4642 /* kvm_write_guest_virt_system can pull in tons of pages. */
4643 vcpu
->arch
.l1tf_flush_l1d
= true;
4645 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
4646 PFERR_WRITE_MASK
, exception
);
4648 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
4650 static int vcpu_is_mmio_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4651 gpa_t gpa
, bool write
)
4653 /* For APIC access vmexit */
4654 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4657 if (vcpu_match_mmio_gpa(vcpu
, gpa
)) {
4658 trace_vcpu_match_mmio(gva
, gpa
, write
, true);
4665 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4666 gpa_t
*gpa
, struct x86_exception
*exception
,
4669 u32 access
= ((kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
4670 | (write
? PFERR_WRITE_MASK
: 0);
4673 * currently PKRU is only applied to ept enabled guest so
4674 * there is no pkey in EPT page table for L1 guest or EPT
4675 * shadow page table for L2 guest.
4677 if (vcpu_match_mmio_gva(vcpu
, gva
)
4678 && !permission_fault(vcpu
, vcpu
->arch
.walk_mmu
,
4679 vcpu
->arch
.access
, 0, access
)) {
4680 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
4681 (gva
& (PAGE_SIZE
- 1));
4682 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
4686 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4688 if (*gpa
== UNMAPPED_GVA
)
4691 return vcpu_is_mmio_gpa(vcpu
, gva
, *gpa
, write
);
4694 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4695 const void *val
, int bytes
)
4699 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, val
, bytes
);
4702 kvm_page_track_write(vcpu
, gpa
, val
, bytes
);
4706 struct read_write_emulator_ops
{
4707 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
4709 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4710 void *val
, int bytes
);
4711 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4712 int bytes
, void *val
);
4713 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4714 void *val
, int bytes
);
4718 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
4720 if (vcpu
->mmio_read_completed
) {
4721 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
4722 vcpu
->mmio_fragments
[0].gpa
, val
);
4723 vcpu
->mmio_read_completed
= 0;
4730 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4731 void *val
, int bytes
)
4733 return !kvm_vcpu_read_guest(vcpu
, gpa
, val
, bytes
);
4736 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4737 void *val
, int bytes
)
4739 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
4742 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
4744 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, val
);
4745 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
4748 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4749 void *val
, int bytes
)
4751 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, NULL
);
4752 return X86EMUL_IO_NEEDED
;
4755 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4756 void *val
, int bytes
)
4758 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
4760 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
4761 return X86EMUL_CONTINUE
;
4764 static const struct read_write_emulator_ops read_emultor
= {
4765 .read_write_prepare
= read_prepare
,
4766 .read_write_emulate
= read_emulate
,
4767 .read_write_mmio
= vcpu_mmio_read
,
4768 .read_write_exit_mmio
= read_exit_mmio
,
4771 static const struct read_write_emulator_ops write_emultor
= {
4772 .read_write_emulate
= write_emulate
,
4773 .read_write_mmio
= write_mmio
,
4774 .read_write_exit_mmio
= write_exit_mmio
,
4778 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
4780 struct x86_exception
*exception
,
4781 struct kvm_vcpu
*vcpu
,
4782 const struct read_write_emulator_ops
*ops
)
4786 bool write
= ops
->write
;
4787 struct kvm_mmio_fragment
*frag
;
4788 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4791 * If the exit was due to a NPF we may already have a GPA.
4792 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4793 * Note, this cannot be used on string operations since string
4794 * operation using rep will only have the initial GPA from the NPF
4797 if (vcpu
->arch
.gpa_available
&&
4798 emulator_can_use_gpa(ctxt
) &&
4799 (addr
& ~PAGE_MASK
) == (vcpu
->arch
.gpa_val
& ~PAGE_MASK
)) {
4800 gpa
= vcpu
->arch
.gpa_val
;
4801 ret
= vcpu_is_mmio_gpa(vcpu
, addr
, gpa
, write
);
4803 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
4805 return X86EMUL_PROPAGATE_FAULT
;
4808 if (!ret
&& ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
4809 return X86EMUL_CONTINUE
;
4812 * Is this MMIO handled locally?
4814 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
4815 if (handled
== bytes
)
4816 return X86EMUL_CONTINUE
;
4822 WARN_ON(vcpu
->mmio_nr_fragments
>= KVM_MAX_MMIO_FRAGMENTS
);
4823 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
4827 return X86EMUL_CONTINUE
;
4830 static int emulator_read_write(struct x86_emulate_ctxt
*ctxt
,
4832 void *val
, unsigned int bytes
,
4833 struct x86_exception
*exception
,
4834 const struct read_write_emulator_ops
*ops
)
4836 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4840 if (ops
->read_write_prepare
&&
4841 ops
->read_write_prepare(vcpu
, val
, bytes
))
4842 return X86EMUL_CONTINUE
;
4844 vcpu
->mmio_nr_fragments
= 0;
4846 /* Crossing a page boundary? */
4847 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
4850 now
= -addr
& ~PAGE_MASK
;
4851 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
4854 if (rc
!= X86EMUL_CONTINUE
)
4857 if (ctxt
->mode
!= X86EMUL_MODE_PROT64
)
4863 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
4865 if (rc
!= X86EMUL_CONTINUE
)
4868 if (!vcpu
->mmio_nr_fragments
)
4871 gpa
= vcpu
->mmio_fragments
[0].gpa
;
4873 vcpu
->mmio_needed
= 1;
4874 vcpu
->mmio_cur_fragment
= 0;
4876 vcpu
->run
->mmio
.len
= min(8u, vcpu
->mmio_fragments
[0].len
);
4877 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
4878 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
4879 vcpu
->run
->mmio
.phys_addr
= gpa
;
4881 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
4884 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
4888 struct x86_exception
*exception
)
4890 return emulator_read_write(ctxt
, addr
, val
, bytes
,
4891 exception
, &read_emultor
);
4894 static int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
4898 struct x86_exception
*exception
)
4900 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
4901 exception
, &write_emultor
);
4904 #define CMPXCHG_TYPE(t, ptr, old, new) \
4905 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4907 #ifdef CONFIG_X86_64
4908 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4910 # define CMPXCHG64(ptr, old, new) \
4911 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4914 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
4919 struct x86_exception
*exception
)
4921 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4927 /* guests cmpxchg8b have to be emulated atomically */
4928 if (bytes
> 8 || (bytes
& (bytes
- 1)))
4931 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
4933 if (gpa
== UNMAPPED_GVA
||
4934 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4937 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
4940 page
= kvm_vcpu_gfn_to_page(vcpu
, gpa
>> PAGE_SHIFT
);
4941 if (is_error_page(page
))
4944 kaddr
= kmap_atomic(page
);
4945 kaddr
+= offset_in_page(gpa
);
4948 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
4951 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
4954 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
4957 exchanged
= CMPXCHG64(kaddr
, old
, new);
4962 kunmap_atomic(kaddr
);
4963 kvm_release_page_dirty(page
);
4966 return X86EMUL_CMPXCHG_FAILED
;
4968 kvm_vcpu_mark_page_dirty(vcpu
, gpa
>> PAGE_SHIFT
);
4969 kvm_page_track_write(vcpu
, gpa
, new, bytes
);
4971 return X86EMUL_CONTINUE
;
4974 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
4976 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
4979 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
4983 for (i
= 0; i
< vcpu
->arch
.pio
.count
; i
++) {
4984 if (vcpu
->arch
.pio
.in
)
4985 r
= kvm_io_bus_read(vcpu
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
4986 vcpu
->arch
.pio
.size
, pd
);
4988 r
= kvm_io_bus_write(vcpu
, KVM_PIO_BUS
,
4989 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
4993 pd
+= vcpu
->arch
.pio
.size
;
4998 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
4999 unsigned short port
, void *val
,
5000 unsigned int count
, bool in
)
5002 vcpu
->arch
.pio
.port
= port
;
5003 vcpu
->arch
.pio
.in
= in
;
5004 vcpu
->arch
.pio
.count
= count
;
5005 vcpu
->arch
.pio
.size
= size
;
5007 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
5008 vcpu
->arch
.pio
.count
= 0;
5012 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
5013 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
5014 vcpu
->run
->io
.size
= size
;
5015 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
5016 vcpu
->run
->io
.count
= count
;
5017 vcpu
->run
->io
.port
= port
;
5022 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
5023 int size
, unsigned short port
, void *val
,
5026 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5029 if (vcpu
->arch
.pio
.count
)
5032 memset(vcpu
->arch
.pio_data
, 0, size
* count
);
5034 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
5037 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
5038 trace_kvm_pio(KVM_PIO_IN
, port
, size
, count
, vcpu
->arch
.pio_data
);
5039 vcpu
->arch
.pio
.count
= 0;
5046 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
5047 int size
, unsigned short port
,
5048 const void *val
, unsigned int count
)
5050 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5052 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
5053 trace_kvm_pio(KVM_PIO_OUT
, port
, size
, count
, vcpu
->arch
.pio_data
);
5054 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
5057 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
5059 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
5062 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
5064 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
5067 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu
*vcpu
)
5069 if (!need_emulate_wbinvd(vcpu
))
5070 return X86EMUL_CONTINUE
;
5072 if (kvm_x86_ops
->has_wbinvd_exit()) {
5073 int cpu
= get_cpu();
5075 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
5076 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
5077 wbinvd_ipi
, NULL
, 1);
5079 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
5082 return X86EMUL_CONTINUE
;
5085 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
5087 kvm_emulate_wbinvd_noskip(vcpu
);
5088 return kvm_skip_emulated_instruction(vcpu
);
5090 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
5094 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
5096 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt
));
5099 static int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
5100 unsigned long *dest
)
5102 return kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
5105 static int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
5106 unsigned long value
)
5109 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
5112 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
5114 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
5117 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
5119 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5120 unsigned long value
;
5124 value
= kvm_read_cr0(vcpu
);
5127 value
= vcpu
->arch
.cr2
;
5130 value
= kvm_read_cr3(vcpu
);
5133 value
= kvm_read_cr4(vcpu
);
5136 value
= kvm_get_cr8(vcpu
);
5139 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
5146 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
5148 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5153 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
5156 vcpu
->arch
.cr2
= val
;
5159 res
= kvm_set_cr3(vcpu
, val
);
5162 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
5165 res
= kvm_set_cr8(vcpu
, val
);
5168 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
5175 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
5177 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
5180 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5182 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
5185 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5187 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
5190 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5192 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
5195 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5197 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
5200 static unsigned long emulator_get_cached_segment_base(
5201 struct x86_emulate_ctxt
*ctxt
, int seg
)
5203 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
5206 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
5207 struct desc_struct
*desc
, u32
*base3
,
5210 struct kvm_segment var
;
5212 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
5213 *selector
= var
.selector
;
5216 memset(desc
, 0, sizeof(*desc
));
5224 set_desc_limit(desc
, var
.limit
);
5225 set_desc_base(desc
, (unsigned long)var
.base
);
5226 #ifdef CONFIG_X86_64
5228 *base3
= var
.base
>> 32;
5230 desc
->type
= var
.type
;
5232 desc
->dpl
= var
.dpl
;
5233 desc
->p
= var
.present
;
5234 desc
->avl
= var
.avl
;
5242 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
5243 struct desc_struct
*desc
, u32 base3
,
5246 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5247 struct kvm_segment var
;
5249 var
.selector
= selector
;
5250 var
.base
= get_desc_base(desc
);
5251 #ifdef CONFIG_X86_64
5252 var
.base
|= ((u64
)base3
) << 32;
5254 var
.limit
= get_desc_limit(desc
);
5256 var
.limit
= (var
.limit
<< 12) | 0xfff;
5257 var
.type
= desc
->type
;
5258 var
.dpl
= desc
->dpl
;
5263 var
.avl
= desc
->avl
;
5264 var
.present
= desc
->p
;
5265 var
.unusable
= !var
.present
;
5268 kvm_set_segment(vcpu
, &var
, seg
);
5272 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
5273 u32 msr_index
, u64
*pdata
)
5275 struct msr_data msr
;
5278 msr
.index
= msr_index
;
5279 msr
.host_initiated
= false;
5280 r
= kvm_get_msr(emul_to_vcpu(ctxt
), &msr
);
5288 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
5289 u32 msr_index
, u64 data
)
5291 struct msr_data msr
;
5294 msr
.index
= msr_index
;
5295 msr
.host_initiated
= false;
5296 return kvm_set_msr(emul_to_vcpu(ctxt
), &msr
);
5299 static u64
emulator_get_smbase(struct x86_emulate_ctxt
*ctxt
)
5301 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5303 return vcpu
->arch
.smbase
;
5306 static void emulator_set_smbase(struct x86_emulate_ctxt
*ctxt
, u64 smbase
)
5308 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5310 vcpu
->arch
.smbase
= smbase
;
5313 static int emulator_check_pmc(struct x86_emulate_ctxt
*ctxt
,
5316 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt
), pmc
);
5319 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
5320 u32 pmc
, u64
*pdata
)
5322 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
5325 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
5327 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
5330 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
5333 kvm_load_guest_fpu(emul_to_vcpu(ctxt
));
5336 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
5341 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
5342 struct x86_instruction_info
*info
,
5343 enum x86_intercept_stage stage
)
5345 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
5348 static bool emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
5349 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
, bool check_limit
)
5351 return kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
, check_limit
);
5354 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
5356 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
5359 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
5361 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
5364 static void emulator_set_nmi_mask(struct x86_emulate_ctxt
*ctxt
, bool masked
)
5366 kvm_x86_ops
->set_nmi_mask(emul_to_vcpu(ctxt
), masked
);
5369 static unsigned emulator_get_hflags(struct x86_emulate_ctxt
*ctxt
)
5371 return emul_to_vcpu(ctxt
)->arch
.hflags
;
5374 static void emulator_set_hflags(struct x86_emulate_ctxt
*ctxt
, unsigned emul_flags
)
5376 kvm_set_hflags(emul_to_vcpu(ctxt
), emul_flags
);
5379 static const struct x86_emulate_ops emulate_ops
= {
5380 .read_gpr
= emulator_read_gpr
,
5381 .write_gpr
= emulator_write_gpr
,
5382 .read_std
= emulator_read_std
,
5383 .write_std
= emulator_write_std
,
5384 .read_phys
= kvm_read_guest_phys_system
,
5385 .fetch
= kvm_fetch_guest_virt
,
5386 .read_emulated
= emulator_read_emulated
,
5387 .write_emulated
= emulator_write_emulated
,
5388 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
5389 .invlpg
= emulator_invlpg
,
5390 .pio_in_emulated
= emulator_pio_in_emulated
,
5391 .pio_out_emulated
= emulator_pio_out_emulated
,
5392 .get_segment
= emulator_get_segment
,
5393 .set_segment
= emulator_set_segment
,
5394 .get_cached_segment_base
= emulator_get_cached_segment_base
,
5395 .get_gdt
= emulator_get_gdt
,
5396 .get_idt
= emulator_get_idt
,
5397 .set_gdt
= emulator_set_gdt
,
5398 .set_idt
= emulator_set_idt
,
5399 .get_cr
= emulator_get_cr
,
5400 .set_cr
= emulator_set_cr
,
5401 .cpl
= emulator_get_cpl
,
5402 .get_dr
= emulator_get_dr
,
5403 .set_dr
= emulator_set_dr
,
5404 .get_smbase
= emulator_get_smbase
,
5405 .set_smbase
= emulator_set_smbase
,
5406 .set_msr
= emulator_set_msr
,
5407 .get_msr
= emulator_get_msr
,
5408 .check_pmc
= emulator_check_pmc
,
5409 .read_pmc
= emulator_read_pmc
,
5410 .halt
= emulator_halt
,
5411 .wbinvd
= emulator_wbinvd
,
5412 .fix_hypercall
= emulator_fix_hypercall
,
5413 .get_fpu
= emulator_get_fpu
,
5414 .put_fpu
= emulator_put_fpu
,
5415 .intercept
= emulator_intercept
,
5416 .get_cpuid
= emulator_get_cpuid
,
5417 .set_nmi_mask
= emulator_set_nmi_mask
,
5418 .get_hflags
= emulator_get_hflags
,
5419 .set_hflags
= emulator_set_hflags
,
5422 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
5424 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
5426 * an sti; sti; sequence only disable interrupts for the first
5427 * instruction. So, if the last instruction, be it emulated or
5428 * not, left the system with the INT_STI flag enabled, it
5429 * means that the last instruction is an sti. We should not
5430 * leave the flag on in this case. The same goes for mov ss
5432 if (int_shadow
& mask
)
5434 if (unlikely(int_shadow
|| mask
)) {
5435 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
5437 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5441 static bool inject_emulated_exception(struct kvm_vcpu
*vcpu
)
5443 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5444 if (ctxt
->exception
.vector
== PF_VECTOR
)
5445 return kvm_propagate_fault(vcpu
, &ctxt
->exception
);
5447 if (ctxt
->exception
.error_code_valid
)
5448 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
5449 ctxt
->exception
.error_code
);
5451 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
5455 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
5457 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5460 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
5462 ctxt
->eflags
= kvm_get_rflags(vcpu
);
5463 ctxt
->tf
= (ctxt
->eflags
& X86_EFLAGS_TF
) != 0;
5465 ctxt
->eip
= kvm_rip_read(vcpu
);
5466 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
5467 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
5468 (cs_l
&& is_long_mode(vcpu
)) ? X86EMUL_MODE_PROT64
:
5469 cs_db
? X86EMUL_MODE_PROT32
:
5470 X86EMUL_MODE_PROT16
;
5471 BUILD_BUG_ON(HF_GUEST_MASK
!= X86EMUL_GUEST_MASK
);
5472 BUILD_BUG_ON(HF_SMM_MASK
!= X86EMUL_SMM_MASK
);
5473 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK
!= X86EMUL_SMM_INSIDE_NMI_MASK
);
5475 init_decode_cache(ctxt
);
5476 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5479 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
5481 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5484 init_emulate_ctxt(vcpu
);
5488 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
5489 ret
= emulate_int_real(ctxt
, irq
);
5491 if (ret
!= X86EMUL_CONTINUE
)
5492 return EMULATE_FAIL
;
5494 ctxt
->eip
= ctxt
->_eip
;
5495 kvm_rip_write(vcpu
, ctxt
->eip
);
5496 kvm_set_rflags(vcpu
, ctxt
->eflags
);
5498 if (irq
== NMI_VECTOR
)
5499 vcpu
->arch
.nmi_pending
= 0;
5501 vcpu
->arch
.interrupt
.pending
= false;
5503 return EMULATE_DONE
;
5505 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
5507 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
5509 int r
= EMULATE_DONE
;
5511 ++vcpu
->stat
.insn_emulation_fail
;
5512 trace_kvm_emulate_insn_failed(vcpu
);
5513 if (!is_guest_mode(vcpu
) && kvm_x86_ops
->get_cpl(vcpu
) == 0) {
5514 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
5515 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
5516 vcpu
->run
->internal
.ndata
= 0;
5517 r
= EMULATE_USER_EXIT
;
5519 kvm_queue_exception(vcpu
, UD_VECTOR
);
5524 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t cr2
,
5525 bool write_fault_to_shadow_pgtable
,
5531 if (emulation_type
& EMULTYPE_NO_REEXECUTE
)
5534 if (!vcpu
->arch
.mmu
.direct_map
) {
5536 * Write permission should be allowed since only
5537 * write access need to be emulated.
5539 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5542 * If the mapping is invalid in guest, let cpu retry
5543 * it to generate fault.
5545 if (gpa
== UNMAPPED_GVA
)
5550 * Do not retry the unhandleable instruction if it faults on the
5551 * readonly host memory, otherwise it will goto a infinite loop:
5552 * retry instruction -> write #PF -> emulation fail -> retry
5553 * instruction -> ...
5555 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
5558 * If the instruction failed on the error pfn, it can not be fixed,
5559 * report the error to userspace.
5561 if (is_error_noslot_pfn(pfn
))
5564 kvm_release_pfn_clean(pfn
);
5566 /* The instructions are well-emulated on direct mmu. */
5567 if (vcpu
->arch
.mmu
.direct_map
) {
5568 unsigned int indirect_shadow_pages
;
5570 spin_lock(&vcpu
->kvm
->mmu_lock
);
5571 indirect_shadow_pages
= vcpu
->kvm
->arch
.indirect_shadow_pages
;
5572 spin_unlock(&vcpu
->kvm
->mmu_lock
);
5574 if (indirect_shadow_pages
)
5575 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5581 * if emulation was due to access to shadowed page table
5582 * and it failed try to unshadow page and re-enter the
5583 * guest to let CPU execute the instruction.
5585 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5588 * If the access faults on its page table, it can not
5589 * be fixed by unprotecting shadow page and it should
5590 * be reported to userspace.
5592 return !write_fault_to_shadow_pgtable
;
5595 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
5596 unsigned long cr2
, int emulation_type
)
5598 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5599 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
5601 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
5602 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
5605 * If the emulation is caused by #PF and it is non-page_table
5606 * writing instruction, it means the VM-EXIT is caused by shadow
5607 * page protected, we can zap the shadow page and retry this
5608 * instruction directly.
5610 * Note: if the guest uses a non-page-table modifying instruction
5611 * on the PDE that points to the instruction, then we will unmap
5612 * the instruction and go to an infinite loop. So, we cache the
5613 * last retried eip and the last fault address, if we meet the eip
5614 * and the address again, we can break out of the potential infinite
5617 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
5619 if (!(emulation_type
& EMULTYPE_RETRY
))
5622 if (x86_page_table_writing_insn(ctxt
))
5625 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
5628 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
5629 vcpu
->arch
.last_retry_addr
= cr2
;
5631 if (!vcpu
->arch
.mmu
.direct_map
)
5632 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5634 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5639 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
5640 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
5642 static void kvm_smm_changed(struct kvm_vcpu
*vcpu
)
5644 if (!(vcpu
->arch
.hflags
& HF_SMM_MASK
)) {
5645 /* This is a good place to trace that we are exiting SMM. */
5646 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, false);
5648 /* Process a latched INIT or SMI, if any. */
5649 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5652 kvm_mmu_reset_context(vcpu
);
5655 static void kvm_set_hflags(struct kvm_vcpu
*vcpu
, unsigned emul_flags
)
5657 unsigned changed
= vcpu
->arch
.hflags
^ emul_flags
;
5659 vcpu
->arch
.hflags
= emul_flags
;
5661 if (changed
& HF_SMM_MASK
)
5662 kvm_smm_changed(vcpu
);
5665 static int kvm_vcpu_check_hw_bp(unsigned long addr
, u32 type
, u32 dr7
,
5674 for (i
= 0; i
< 4; i
++, enable
>>= 2, rwlen
>>= 4)
5675 if ((enable
& 3) && (rwlen
& 15) == type
&& db
[i
] == addr
)
5680 static void kvm_vcpu_do_singlestep(struct kvm_vcpu
*vcpu
, int *r
)
5682 struct kvm_run
*kvm_run
= vcpu
->run
;
5684 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
) {
5685 kvm_run
->debug
.arch
.dr6
= DR6_BS
| DR6_FIXED_1
| DR6_RTM
;
5686 kvm_run
->debug
.arch
.pc
= vcpu
->arch
.singlestep_rip
;
5687 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5688 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5689 *r
= EMULATE_USER_EXIT
;
5692 * "Certain debug exceptions may clear bit 0-3. The
5693 * remaining contents of the DR6 register are never
5694 * cleared by the processor".
5696 vcpu
->arch
.dr6
&= ~15;
5697 vcpu
->arch
.dr6
|= DR6_BS
| DR6_RTM
;
5698 kvm_queue_exception(vcpu
, DB_VECTOR
);
5702 int kvm_skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
5704 unsigned long rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5705 int r
= EMULATE_DONE
;
5707 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
5710 * rflags is the old, "raw" value of the flags. The new value has
5711 * not been saved yet.
5713 * This is correct even for TF set by the guest, because "the
5714 * processor will not generate this exception after the instruction
5715 * that sets the TF flag".
5717 if (unlikely(rflags
& X86_EFLAGS_TF
))
5718 kvm_vcpu_do_singlestep(vcpu
, &r
);
5719 return r
== EMULATE_DONE
;
5721 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction
);
5723 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu
*vcpu
, int *r
)
5725 if (unlikely(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) &&
5726 (vcpu
->arch
.guest_debug_dr7
& DR7_BP_EN_MASK
)) {
5727 struct kvm_run
*kvm_run
= vcpu
->run
;
5728 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5729 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5730 vcpu
->arch
.guest_debug_dr7
,
5734 kvm_run
->debug
.arch
.dr6
= dr6
| DR6_FIXED_1
| DR6_RTM
;
5735 kvm_run
->debug
.arch
.pc
= eip
;
5736 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5737 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5738 *r
= EMULATE_USER_EXIT
;
5743 if (unlikely(vcpu
->arch
.dr7
& DR7_BP_EN_MASK
) &&
5744 !(kvm_get_rflags(vcpu
) & X86_EFLAGS_RF
)) {
5745 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5746 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5751 vcpu
->arch
.dr6
&= ~15;
5752 vcpu
->arch
.dr6
|= dr6
| DR6_RTM
;
5753 kvm_queue_exception(vcpu
, DB_VECTOR
);
5762 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
5769 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5770 bool writeback
= true;
5771 bool write_fault_to_spt
= vcpu
->arch
.write_fault_to_shadow_pgtable
;
5773 vcpu
->arch
.l1tf_flush_l1d
= true;
5776 * Clear write_fault_to_shadow_pgtable here to ensure it is
5779 vcpu
->arch
.write_fault_to_shadow_pgtable
= false;
5780 kvm_clear_exception_queue(vcpu
);
5782 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
5783 init_emulate_ctxt(vcpu
);
5786 * We will reenter on the same instruction since
5787 * we do not set complete_userspace_io. This does not
5788 * handle watchpoints yet, those would be handled in
5791 if (!(emulation_type
& EMULTYPE_SKIP
) &&
5792 kvm_vcpu_check_breakpoint(vcpu
, &r
))
5795 ctxt
->interruptibility
= 0;
5796 ctxt
->have_exception
= false;
5797 ctxt
->exception
.vector
= -1;
5798 ctxt
->perm_ok
= false;
5800 ctxt
->ud
= emulation_type
& EMULTYPE_TRAP_UD
;
5802 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
5804 trace_kvm_emulate_insn_start(vcpu
);
5805 ++vcpu
->stat
.insn_emulation
;
5806 if (r
!= EMULATION_OK
) {
5807 if (emulation_type
& EMULTYPE_TRAP_UD
)
5808 return EMULATE_FAIL
;
5809 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5811 return EMULATE_DONE
;
5812 if (ctxt
->have_exception
&& inject_emulated_exception(vcpu
))
5813 return EMULATE_DONE
;
5814 if (emulation_type
& EMULTYPE_SKIP
)
5815 return EMULATE_FAIL
;
5816 return handle_emulation_failure(vcpu
);
5820 if (emulation_type
& EMULTYPE_SKIP
) {
5821 kvm_rip_write(vcpu
, ctxt
->_eip
);
5822 if (ctxt
->eflags
& X86_EFLAGS_RF
)
5823 kvm_set_rflags(vcpu
, ctxt
->eflags
& ~X86_EFLAGS_RF
);
5824 return EMULATE_DONE
;
5827 if (retry_instruction(ctxt
, cr2
, emulation_type
))
5828 return EMULATE_DONE
;
5830 /* this is needed for vmware backdoor interface to work since it
5831 changes registers values during IO operation */
5832 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
5833 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5834 emulator_invalidate_register_cache(ctxt
);
5838 /* Save the faulting GPA (cr2) in the address field */
5839 ctxt
->exception
.address
= cr2
;
5841 r
= x86_emulate_insn(ctxt
);
5843 if (r
== EMULATION_INTERCEPTED
)
5844 return EMULATE_DONE
;
5846 if (r
== EMULATION_FAILED
) {
5847 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5849 return EMULATE_DONE
;
5851 return handle_emulation_failure(vcpu
);
5854 if (ctxt
->have_exception
) {
5856 if (inject_emulated_exception(vcpu
))
5858 } else if (vcpu
->arch
.pio
.count
) {
5859 if (!vcpu
->arch
.pio
.in
) {
5860 /* FIXME: return into emulator if single-stepping. */
5861 vcpu
->arch
.pio
.count
= 0;
5864 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
5866 r
= EMULATE_USER_EXIT
;
5867 } else if (vcpu
->mmio_needed
) {
5868 if (!vcpu
->mmio_is_write
)
5870 r
= EMULATE_USER_EXIT
;
5871 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
5872 } else if (r
== EMULATION_RESTART
)
5878 unsigned long rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5879 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
5880 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5881 kvm_rip_write(vcpu
, ctxt
->eip
);
5882 if (r
== EMULATE_DONE
&&
5883 (ctxt
->tf
|| (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)))
5884 kvm_vcpu_do_singlestep(vcpu
, &r
);
5885 if (!ctxt
->have_exception
||
5886 exception_type(ctxt
->exception
.vector
) == EXCPT_TRAP
)
5887 __kvm_set_rflags(vcpu
, ctxt
->eflags
);
5890 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5891 * do nothing, and it will be requested again as soon as
5892 * the shadow expires. But we still need to check here,
5893 * because POPF has no interrupt shadow.
5895 if (unlikely((ctxt
->eflags
& ~rflags
) & X86_EFLAGS_IF
))
5896 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5898 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
5902 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
5904 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5906 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5907 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
5908 size
, port
, &val
, 1);
5909 /* do not return to emulator after return from userspace */
5910 vcpu
->arch
.pio
.count
= 0;
5913 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
5915 static int complete_fast_pio_in(struct kvm_vcpu
*vcpu
)
5919 /* We should only ever be called with arch.pio.count equal to 1 */
5920 BUG_ON(vcpu
->arch
.pio
.count
!= 1);
5922 /* For size less than 4 we merge, else we zero extend */
5923 val
= (vcpu
->arch
.pio
.size
< 4) ? kvm_register_read(vcpu
, VCPU_REGS_RAX
)
5927 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5928 * the copy and tracing
5930 emulator_pio_in_emulated(&vcpu
->arch
.emulate_ctxt
, vcpu
->arch
.pio
.size
,
5931 vcpu
->arch
.pio
.port
, &val
, 1);
5932 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
5937 int kvm_fast_pio_in(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5942 /* For size less than 4 we merge, else we zero extend */
5943 val
= (size
< 4) ? kvm_register_read(vcpu
, VCPU_REGS_RAX
) : 0;
5945 ret
= emulator_pio_in_emulated(&vcpu
->arch
.emulate_ctxt
, size
, port
,
5948 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
5952 vcpu
->arch
.complete_userspace_io
= complete_fast_pio_in
;
5956 EXPORT_SYMBOL_GPL(kvm_fast_pio_in
);
5958 static int kvmclock_cpu_down_prep(unsigned int cpu
)
5960 __this_cpu_write(cpu_tsc_khz
, 0);
5964 static void tsc_khz_changed(void *data
)
5966 struct cpufreq_freqs
*freq
= data
;
5967 unsigned long khz
= 0;
5971 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5972 khz
= cpufreq_quick_get(raw_smp_processor_id());
5975 __this_cpu_write(cpu_tsc_khz
, khz
);
5978 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
5981 struct cpufreq_freqs
*freq
= data
;
5983 struct kvm_vcpu
*vcpu
;
5984 int i
, send_ipi
= 0;
5987 * We allow guests to temporarily run on slowing clocks,
5988 * provided we notify them after, or to run on accelerating
5989 * clocks, provided we notify them before. Thus time never
5992 * However, we have a problem. We can't atomically update
5993 * the frequency of a given CPU from this function; it is
5994 * merely a notifier, which can be called from any CPU.
5995 * Changing the TSC frequency at arbitrary points in time
5996 * requires a recomputation of local variables related to
5997 * the TSC for each VCPU. We must flag these local variables
5998 * to be updated and be sure the update takes place with the
5999 * new frequency before any guests proceed.
6001 * Unfortunately, the combination of hotplug CPU and frequency
6002 * change creates an intractable locking scenario; the order
6003 * of when these callouts happen is undefined with respect to
6004 * CPU hotplug, and they can race with each other. As such,
6005 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6006 * undefined; you can actually have a CPU frequency change take
6007 * place in between the computation of X and the setting of the
6008 * variable. To protect against this problem, all updates of
6009 * the per_cpu tsc_khz variable are done in an interrupt
6010 * protected IPI, and all callers wishing to update the value
6011 * must wait for a synchronous IPI to complete (which is trivial
6012 * if the caller is on the CPU already). This establishes the
6013 * necessary total order on variable updates.
6015 * Note that because a guest time update may take place
6016 * anytime after the setting of the VCPU's request bit, the
6017 * correct TSC value must be set before the request. However,
6018 * to ensure the update actually makes it to any guest which
6019 * starts running in hardware virtualization between the set
6020 * and the acquisition of the spinlock, we must also ping the
6021 * CPU after setting the request bit.
6025 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
6027 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
6030 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
6032 spin_lock(&kvm_lock
);
6033 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6034 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6035 if (vcpu
->cpu
!= freq
->cpu
)
6037 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
6038 if (vcpu
->cpu
!= smp_processor_id())
6042 spin_unlock(&kvm_lock
);
6044 if (freq
->old
< freq
->new && send_ipi
) {
6046 * We upscale the frequency. Must make the guest
6047 * doesn't see old kvmclock values while running with
6048 * the new frequency, otherwise we risk the guest sees
6049 * time go backwards.
6051 * In case we update the frequency for another cpu
6052 * (which might be in guest context) send an interrupt
6053 * to kick the cpu out of guest context. Next time
6054 * guest context is entered kvmclock will be updated,
6055 * so the guest will not see stale values.
6057 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
6062 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
6063 .notifier_call
= kvmclock_cpufreq_notifier
6066 static int kvmclock_cpu_online(unsigned int cpu
)
6068 tsc_khz_changed(NULL
);
6072 static void kvm_timer_init(void)
6074 max_tsc_khz
= tsc_khz
;
6076 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
6077 #ifdef CONFIG_CPU_FREQ
6078 struct cpufreq_policy policy
;
6081 memset(&policy
, 0, sizeof(policy
));
6083 cpufreq_get_policy(&policy
, cpu
);
6084 if (policy
.cpuinfo
.max_freq
)
6085 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
6088 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
6089 CPUFREQ_TRANSITION_NOTIFIER
);
6091 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
6093 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE
, "x86/kvm/clk:online",
6094 kvmclock_cpu_online
, kvmclock_cpu_down_prep
);
6097 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
6099 int kvm_is_in_guest(void)
6101 return __this_cpu_read(current_vcpu
) != NULL
;
6104 static int kvm_is_user_mode(void)
6108 if (__this_cpu_read(current_vcpu
))
6109 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
6111 return user_mode
!= 0;
6114 static unsigned long kvm_get_guest_ip(void)
6116 unsigned long ip
= 0;
6118 if (__this_cpu_read(current_vcpu
))
6119 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
6124 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6125 .is_in_guest
= kvm_is_in_guest
,
6126 .is_user_mode
= kvm_is_user_mode
,
6127 .get_guest_ip
= kvm_get_guest_ip
,
6130 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
6132 __this_cpu_write(current_vcpu
, vcpu
);
6134 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
6136 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
6138 __this_cpu_write(current_vcpu
, NULL
);
6140 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
6142 static void kvm_set_mmio_spte_mask(void)
6145 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
6148 * Set the reserved bits and the present bit of an paging-structure
6149 * entry to generate page fault with PFER.RSV = 1.
6151 /* Mask the reserved physical address bits. */
6152 mask
= rsvd_bits(maxphyaddr
, 51);
6154 /* Set the present bit. */
6157 #ifdef CONFIG_X86_64
6159 * If reserved bit is not supported, clear the present bit to disable
6162 if (maxphyaddr
== 52)
6166 kvm_mmu_set_mmio_spte_mask(mask
, mask
);
6169 #ifdef CONFIG_X86_64
6170 static void pvclock_gtod_update_fn(struct work_struct
*work
)
6174 struct kvm_vcpu
*vcpu
;
6177 spin_lock(&kvm_lock
);
6178 list_for_each_entry(kvm
, &vm_list
, vm_list
)
6179 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6180 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
6181 atomic_set(&kvm_guest_has_master_clock
, 0);
6182 spin_unlock(&kvm_lock
);
6185 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
6188 * Notification about pvclock gtod data update.
6190 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
6193 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
6194 struct timekeeper
*tk
= priv
;
6196 update_pvclock_gtod(tk
);
6198 /* disable master clock if host does not trust, or does not
6199 * use, TSC clocksource
6201 if (gtod
->clock
.vclock_mode
!= VCLOCK_TSC
&&
6202 atomic_read(&kvm_guest_has_master_clock
) != 0)
6203 queue_work(system_long_wq
, &pvclock_gtod_work
);
6208 static struct notifier_block pvclock_gtod_notifier
= {
6209 .notifier_call
= pvclock_gtod_notify
,
6213 int kvm_arch_init(void *opaque
)
6216 struct kvm_x86_ops
*ops
= opaque
;
6219 printk(KERN_ERR
"kvm: already loaded the other module\n");
6224 if (!ops
->cpu_has_kvm_support()) {
6225 printk(KERN_ERR
"kvm: no hardware support\n");
6229 if (ops
->disabled_by_bios()) {
6230 printk(KERN_ERR
"kvm: disabled by bios\n");
6236 shared_msrs
= alloc_percpu(struct kvm_shared_msrs
);
6238 printk(KERN_ERR
"kvm: failed to allocate percpu kvm_shared_msrs\n");
6242 r
= kvm_mmu_module_init();
6244 goto out_free_percpu
;
6246 kvm_set_mmio_spte_mask();
6250 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
6251 PT_DIRTY_MASK
, PT64_NX_MASK
, 0,
6252 PT_PRESENT_MASK
, 0, sme_me_mask
);
6255 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6257 if (boot_cpu_has(X86_FEATURE_XSAVE
))
6258 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
6261 #ifdef CONFIG_X86_64
6262 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
6268 free_percpu(shared_msrs
);
6273 void kvm_arch_exit(void)
6276 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6278 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
6279 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
6280 CPUFREQ_TRANSITION_NOTIFIER
);
6281 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE
);
6282 #ifdef CONFIG_X86_64
6283 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
6286 kvm_mmu_module_exit();
6287 free_percpu(shared_msrs
);
6290 int kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
6292 ++vcpu
->stat
.halt_exits
;
6293 if (lapic_in_kernel(vcpu
)) {
6294 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
6297 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
6301 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
6303 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
6305 int ret
= kvm_skip_emulated_instruction(vcpu
);
6307 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6308 * KVM_EXIT_DEBUG here.
6310 return kvm_vcpu_halt(vcpu
) && ret
;
6312 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
6314 #ifdef CONFIG_X86_64
6315 static int kvm_pv_clock_pairing(struct kvm_vcpu
*vcpu
, gpa_t paddr
,
6316 unsigned long clock_type
)
6318 struct kvm_clock_pairing clock_pairing
;
6323 if (clock_type
!= KVM_CLOCK_PAIRING_WALLCLOCK
)
6324 return -KVM_EOPNOTSUPP
;
6326 if (kvm_get_walltime_and_clockread(&ts
, &cycle
) == false)
6327 return -KVM_EOPNOTSUPP
;
6329 clock_pairing
.sec
= ts
.tv_sec
;
6330 clock_pairing
.nsec
= ts
.tv_nsec
;
6331 clock_pairing
.tsc
= kvm_read_l1_tsc(vcpu
, cycle
);
6332 clock_pairing
.flags
= 0;
6335 if (kvm_write_guest(vcpu
->kvm
, paddr
, &clock_pairing
,
6336 sizeof(struct kvm_clock_pairing
)))
6344 * kvm_pv_kick_cpu_op: Kick a vcpu.
6346 * @apicid - apicid of vcpu to be kicked.
6348 static void kvm_pv_kick_cpu_op(struct kvm
*kvm
, unsigned long flags
, int apicid
)
6350 struct kvm_lapic_irq lapic_irq
;
6352 lapic_irq
.shorthand
= 0;
6353 lapic_irq
.dest_mode
= 0;
6354 lapic_irq
.level
= 0;
6355 lapic_irq
.dest_id
= apicid
;
6356 lapic_irq
.msi_redir_hint
= false;
6358 lapic_irq
.delivery_mode
= APIC_DM_REMRD
;
6359 kvm_irq_delivery_to_apic(kvm
, NULL
, &lapic_irq
, NULL
);
6362 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu
*vcpu
)
6364 vcpu
->arch
.apicv_active
= false;
6365 kvm_x86_ops
->refresh_apicv_exec_ctrl(vcpu
);
6368 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
6370 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
6373 if (kvm_hv_hypercall_enabled(vcpu
->kvm
)) {
6374 if (!kvm_hv_hypercall(vcpu
))
6379 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
6380 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
6381 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
6382 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
6383 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
6385 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
6387 op_64_bit
= is_64_bit_mode(vcpu
);
6396 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
6402 case KVM_HC_VAPIC_POLL_IRQ
:
6405 case KVM_HC_KICK_CPU
:
6406 kvm_pv_kick_cpu_op(vcpu
->kvm
, a0
, a1
);
6409 #ifdef CONFIG_X86_64
6410 case KVM_HC_CLOCK_PAIRING
:
6411 ret
= kvm_pv_clock_pairing(vcpu
, a0
, a1
);
6421 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
6424 ++vcpu
->stat
.hypercalls
;
6425 return kvm_skip_emulated_instruction(vcpu
);
6427 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
6429 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
6431 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6432 char instruction
[3];
6433 unsigned long rip
= kvm_rip_read(vcpu
);
6435 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
6437 return emulator_write_emulated(ctxt
, rip
, instruction
, 3,
6441 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
6443 return vcpu
->run
->request_interrupt_window
&&
6444 likely(!pic_in_kernel(vcpu
->kvm
));
6447 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
6449 struct kvm_run
*kvm_run
= vcpu
->run
;
6451 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
6452 kvm_run
->flags
= is_smm(vcpu
) ? KVM_RUN_X86_SMM
: 0;
6453 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
6454 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
6455 kvm_run
->ready_for_interrupt_injection
=
6456 pic_in_kernel(vcpu
->kvm
) ||
6457 kvm_vcpu_ready_for_interrupt_injection(vcpu
);
6460 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
6464 if (!kvm_x86_ops
->update_cr8_intercept
)
6467 if (!lapic_in_kernel(vcpu
))
6470 if (vcpu
->arch
.apicv_active
)
6473 if (!vcpu
->arch
.apic
->vapic_addr
)
6474 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
6481 tpr
= kvm_lapic_get_cr8(vcpu
);
6483 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
6486 static int inject_pending_event(struct kvm_vcpu
*vcpu
, bool req_int_win
)
6490 /* try to reinject previous events if any */
6491 if (vcpu
->arch
.exception
.injected
) {
6492 kvm_x86_ops
->queue_exception(vcpu
);
6497 * Exceptions must be injected immediately, or the exception
6498 * frame will have the address of the NMI or interrupt handler.
6500 if (!vcpu
->arch
.exception
.pending
) {
6501 if (vcpu
->arch
.nmi_injected
) {
6502 kvm_x86_ops
->set_nmi(vcpu
);
6506 if (vcpu
->arch
.interrupt
.pending
) {
6507 kvm_x86_ops
->set_irq(vcpu
);
6512 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6513 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6518 /* try to inject new event if pending */
6519 if (vcpu
->arch
.exception
.pending
) {
6520 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
6521 vcpu
->arch
.exception
.has_error_code
,
6522 vcpu
->arch
.exception
.error_code
);
6524 vcpu
->arch
.exception
.pending
= false;
6525 vcpu
->arch
.exception
.injected
= true;
6527 if (exception_type(vcpu
->arch
.exception
.nr
) == EXCPT_FAULT
)
6528 __kvm_set_rflags(vcpu
, kvm_get_rflags(vcpu
) |
6531 if (vcpu
->arch
.exception
.nr
== DB_VECTOR
&&
6532 (vcpu
->arch
.dr7
& DR7_GD
)) {
6533 vcpu
->arch
.dr7
&= ~DR7_GD
;
6534 kvm_update_dr7(vcpu
);
6537 kvm_x86_ops
->queue_exception(vcpu
);
6538 } else if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
)) {
6539 vcpu
->arch
.smi_pending
= false;
6541 } else if (vcpu
->arch
.nmi_pending
&& kvm_x86_ops
->nmi_allowed(vcpu
)) {
6542 --vcpu
->arch
.nmi_pending
;
6543 vcpu
->arch
.nmi_injected
= true;
6544 kvm_x86_ops
->set_nmi(vcpu
);
6545 } else if (kvm_cpu_has_injectable_intr(vcpu
)) {
6547 * Because interrupts can be injected asynchronously, we are
6548 * calling check_nested_events again here to avoid a race condition.
6549 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6550 * proposal and current concerns. Perhaps we should be setting
6551 * KVM_REQ_EVENT only on certain events and not unconditionally?
6553 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6554 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6558 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
6559 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
6561 kvm_x86_ops
->set_irq(vcpu
);
6568 static void process_nmi(struct kvm_vcpu
*vcpu
)
6573 * x86 is limited to one NMI running, and one NMI pending after it.
6574 * If an NMI is already in progress, limit further NMIs to just one.
6575 * Otherwise, allow two (and we'll inject the first one immediately).
6577 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
6580 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
6581 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
6582 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6585 #define put_smstate(type, buf, offset, val) \
6586 *(type *)((buf) + (offset) - 0x7e00) = val
6588 static u32
enter_smm_get_segment_flags(struct kvm_segment
*seg
)
6591 flags
|= seg
->g
<< 23;
6592 flags
|= seg
->db
<< 22;
6593 flags
|= seg
->l
<< 21;
6594 flags
|= seg
->avl
<< 20;
6595 flags
|= seg
->present
<< 15;
6596 flags
|= seg
->dpl
<< 13;
6597 flags
|= seg
->s
<< 12;
6598 flags
|= seg
->type
<< 8;
6602 static void enter_smm_save_seg_32(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6604 struct kvm_segment seg
;
6607 kvm_get_segment(vcpu
, &seg
, n
);
6608 put_smstate(u32
, buf
, 0x7fa8 + n
* 4, seg
.selector
);
6611 offset
= 0x7f84 + n
* 12;
6613 offset
= 0x7f2c + (n
- 3) * 12;
6615 put_smstate(u32
, buf
, offset
+ 8, seg
.base
);
6616 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6617 put_smstate(u32
, buf
, offset
, enter_smm_get_segment_flags(&seg
));
6620 #ifdef CONFIG_X86_64
6621 static void enter_smm_save_seg_64(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6623 struct kvm_segment seg
;
6627 kvm_get_segment(vcpu
, &seg
, n
);
6628 offset
= 0x7e00 + n
* 16;
6630 flags
= enter_smm_get_segment_flags(&seg
) >> 8;
6631 put_smstate(u16
, buf
, offset
, seg
.selector
);
6632 put_smstate(u16
, buf
, offset
+ 2, flags
);
6633 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6634 put_smstate(u64
, buf
, offset
+ 8, seg
.base
);
6638 static void enter_smm_save_state_32(struct kvm_vcpu
*vcpu
, char *buf
)
6641 struct kvm_segment seg
;
6645 put_smstate(u32
, buf
, 0x7ffc, kvm_read_cr0(vcpu
));
6646 put_smstate(u32
, buf
, 0x7ff8, kvm_read_cr3(vcpu
));
6647 put_smstate(u32
, buf
, 0x7ff4, kvm_get_rflags(vcpu
));
6648 put_smstate(u32
, buf
, 0x7ff0, kvm_rip_read(vcpu
));
6650 for (i
= 0; i
< 8; i
++)
6651 put_smstate(u32
, buf
, 0x7fd0 + i
* 4, kvm_register_read(vcpu
, i
));
6653 kvm_get_dr(vcpu
, 6, &val
);
6654 put_smstate(u32
, buf
, 0x7fcc, (u32
)val
);
6655 kvm_get_dr(vcpu
, 7, &val
);
6656 put_smstate(u32
, buf
, 0x7fc8, (u32
)val
);
6658 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6659 put_smstate(u32
, buf
, 0x7fc4, seg
.selector
);
6660 put_smstate(u32
, buf
, 0x7f64, seg
.base
);
6661 put_smstate(u32
, buf
, 0x7f60, seg
.limit
);
6662 put_smstate(u32
, buf
, 0x7f5c, enter_smm_get_segment_flags(&seg
));
6664 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6665 put_smstate(u32
, buf
, 0x7fc0, seg
.selector
);
6666 put_smstate(u32
, buf
, 0x7f80, seg
.base
);
6667 put_smstate(u32
, buf
, 0x7f7c, seg
.limit
);
6668 put_smstate(u32
, buf
, 0x7f78, enter_smm_get_segment_flags(&seg
));
6670 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6671 put_smstate(u32
, buf
, 0x7f74, dt
.address
);
6672 put_smstate(u32
, buf
, 0x7f70, dt
.size
);
6674 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6675 put_smstate(u32
, buf
, 0x7f58, dt
.address
);
6676 put_smstate(u32
, buf
, 0x7f54, dt
.size
);
6678 for (i
= 0; i
< 6; i
++)
6679 enter_smm_save_seg_32(vcpu
, buf
, i
);
6681 put_smstate(u32
, buf
, 0x7f14, kvm_read_cr4(vcpu
));
6684 put_smstate(u32
, buf
, 0x7efc, 0x00020000);
6685 put_smstate(u32
, buf
, 0x7ef8, vcpu
->arch
.smbase
);
6688 static void enter_smm_save_state_64(struct kvm_vcpu
*vcpu
, char *buf
)
6690 #ifdef CONFIG_X86_64
6692 struct kvm_segment seg
;
6696 for (i
= 0; i
< 16; i
++)
6697 put_smstate(u64
, buf
, 0x7ff8 - i
* 8, kvm_register_read(vcpu
, i
));
6699 put_smstate(u64
, buf
, 0x7f78, kvm_rip_read(vcpu
));
6700 put_smstate(u32
, buf
, 0x7f70, kvm_get_rflags(vcpu
));
6702 kvm_get_dr(vcpu
, 6, &val
);
6703 put_smstate(u64
, buf
, 0x7f68, val
);
6704 kvm_get_dr(vcpu
, 7, &val
);
6705 put_smstate(u64
, buf
, 0x7f60, val
);
6707 put_smstate(u64
, buf
, 0x7f58, kvm_read_cr0(vcpu
));
6708 put_smstate(u64
, buf
, 0x7f50, kvm_read_cr3(vcpu
));
6709 put_smstate(u64
, buf
, 0x7f48, kvm_read_cr4(vcpu
));
6711 put_smstate(u32
, buf
, 0x7f00, vcpu
->arch
.smbase
);
6714 put_smstate(u32
, buf
, 0x7efc, 0x00020064);
6716 put_smstate(u64
, buf
, 0x7ed0, vcpu
->arch
.efer
);
6718 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6719 put_smstate(u16
, buf
, 0x7e90, seg
.selector
);
6720 put_smstate(u16
, buf
, 0x7e92, enter_smm_get_segment_flags(&seg
) >> 8);
6721 put_smstate(u32
, buf
, 0x7e94, seg
.limit
);
6722 put_smstate(u64
, buf
, 0x7e98, seg
.base
);
6724 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6725 put_smstate(u32
, buf
, 0x7e84, dt
.size
);
6726 put_smstate(u64
, buf
, 0x7e88, dt
.address
);
6728 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6729 put_smstate(u16
, buf
, 0x7e70, seg
.selector
);
6730 put_smstate(u16
, buf
, 0x7e72, enter_smm_get_segment_flags(&seg
) >> 8);
6731 put_smstate(u32
, buf
, 0x7e74, seg
.limit
);
6732 put_smstate(u64
, buf
, 0x7e78, seg
.base
);
6734 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6735 put_smstate(u32
, buf
, 0x7e64, dt
.size
);
6736 put_smstate(u64
, buf
, 0x7e68, dt
.address
);
6738 for (i
= 0; i
< 6; i
++)
6739 enter_smm_save_seg_64(vcpu
, buf
, i
);
6745 static void enter_smm(struct kvm_vcpu
*vcpu
)
6747 struct kvm_segment cs
, ds
;
6752 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, true);
6753 vcpu
->arch
.hflags
|= HF_SMM_MASK
;
6754 memset(buf
, 0, 512);
6755 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
6756 enter_smm_save_state_64(vcpu
, buf
);
6758 enter_smm_save_state_32(vcpu
, buf
);
6760 kvm_vcpu_write_guest(vcpu
, vcpu
->arch
.smbase
+ 0xfe00, buf
, sizeof(buf
));
6762 if (kvm_x86_ops
->get_nmi_mask(vcpu
))
6763 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
6765 kvm_x86_ops
->set_nmi_mask(vcpu
, true);
6767 kvm_set_rflags(vcpu
, X86_EFLAGS_FIXED
);
6768 kvm_rip_write(vcpu
, 0x8000);
6770 cr0
= vcpu
->arch
.cr0
& ~(X86_CR0_PE
| X86_CR0_EM
| X86_CR0_TS
| X86_CR0_PG
);
6771 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
6772 vcpu
->arch
.cr0
= cr0
;
6774 kvm_x86_ops
->set_cr4(vcpu
, 0);
6776 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6777 dt
.address
= dt
.size
= 0;
6778 kvm_x86_ops
->set_idt(vcpu
, &dt
);
6780 __kvm_set_dr(vcpu
, 7, DR7_FIXED_1
);
6782 cs
.selector
= (vcpu
->arch
.smbase
>> 4) & 0xffff;
6783 cs
.base
= vcpu
->arch
.smbase
;
6788 cs
.limit
= ds
.limit
= 0xffffffff;
6789 cs
.type
= ds
.type
= 0x3;
6790 cs
.dpl
= ds
.dpl
= 0;
6795 cs
.avl
= ds
.avl
= 0;
6796 cs
.present
= ds
.present
= 1;
6797 cs
.unusable
= ds
.unusable
= 0;
6798 cs
.padding
= ds
.padding
= 0;
6800 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6801 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_DS
);
6802 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_ES
);
6803 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_FS
);
6804 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_GS
);
6805 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_SS
);
6807 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
6808 kvm_x86_ops
->set_efer(vcpu
, 0);
6810 kvm_update_cpuid(vcpu
);
6811 kvm_mmu_reset_context(vcpu
);
6814 static void process_smi(struct kvm_vcpu
*vcpu
)
6816 vcpu
->arch
.smi_pending
= true;
6817 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6820 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
6822 kvm_make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
6825 static void vcpu_scan_ioapic(struct kvm_vcpu
*vcpu
)
6827 u64 eoi_exit_bitmap
[4];
6829 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
6832 bitmap_zero(vcpu
->arch
.ioapic_handled_vectors
, 256);
6834 if (irqchip_split(vcpu
->kvm
))
6835 kvm_scan_ioapic_routes(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6837 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
6838 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
6839 kvm_ioapic_scan_entry(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6841 bitmap_or((ulong
*)eoi_exit_bitmap
, vcpu
->arch
.ioapic_handled_vectors
,
6842 vcpu_to_synic(vcpu
)->vec_bitmap
, 256);
6843 kvm_x86_ops
->load_eoi_exitmap(vcpu
, eoi_exit_bitmap
);
6846 static void kvm_vcpu_flush_tlb(struct kvm_vcpu
*vcpu
)
6848 ++vcpu
->stat
.tlb_flush
;
6849 kvm_x86_ops
->tlb_flush(vcpu
);
6852 void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
6853 unsigned long start
, unsigned long end
)
6855 unsigned long apic_address
;
6858 * The physical address of apic access page is stored in the VMCS.
6859 * Update it when it becomes invalid.
6861 apic_address
= gfn_to_hva(kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
6862 if (start
<= apic_address
&& apic_address
< end
)
6863 kvm_make_all_cpus_request(kvm
, KVM_REQ_APIC_PAGE_RELOAD
);
6866 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu
*vcpu
)
6868 struct page
*page
= NULL
;
6870 if (!lapic_in_kernel(vcpu
))
6873 if (!kvm_x86_ops
->set_apic_access_page_addr
)
6876 page
= gfn_to_page(vcpu
->kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
6877 if (is_error_page(page
))
6879 kvm_x86_ops
->set_apic_access_page_addr(vcpu
, page_to_phys(page
));
6882 * Do not pin apic access page in memory, the MMU notifier
6883 * will call us again if it is migrated or swapped out.
6887 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page
);
6890 * Returns 1 to let vcpu_run() continue the guest execution loop without
6891 * exiting to the userspace. Otherwise, the value will be returned to the
6894 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
6898 dm_request_for_irq_injection(vcpu
) &&
6899 kvm_cpu_accept_dm_intr(vcpu
);
6901 bool req_immediate_exit
= false;
6903 if (kvm_request_pending(vcpu
)) {
6904 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
6905 kvm_mmu_unload(vcpu
);
6906 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
6907 __kvm_migrate_timers(vcpu
);
6908 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
6909 kvm_gen_update_masterclock(vcpu
->kvm
);
6910 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
))
6911 kvm_gen_kvmclock_update(vcpu
);
6912 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
6913 r
= kvm_guest_time_update(vcpu
);
6917 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
6918 kvm_mmu_sync_roots(vcpu
);
6919 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
6920 kvm_vcpu_flush_tlb(vcpu
);
6921 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
6922 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
6926 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
6927 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
6928 vcpu
->mmio_needed
= 0;
6932 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
6933 /* Page is swapped out. Do synthetic halt */
6934 vcpu
->arch
.apf
.halted
= true;
6938 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
6939 record_steal_time(vcpu
);
6940 if (kvm_check_request(KVM_REQ_SMI
, vcpu
))
6942 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
6944 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
6945 kvm_pmu_handle_event(vcpu
);
6946 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
6947 kvm_pmu_deliver_pmi(vcpu
);
6948 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT
, vcpu
)) {
6949 BUG_ON(vcpu
->arch
.pending_ioapic_eoi
> 255);
6950 if (test_bit(vcpu
->arch
.pending_ioapic_eoi
,
6951 vcpu
->arch
.ioapic_handled_vectors
)) {
6952 vcpu
->run
->exit_reason
= KVM_EXIT_IOAPIC_EOI
;
6953 vcpu
->run
->eoi
.vector
=
6954 vcpu
->arch
.pending_ioapic_eoi
;
6959 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC
, vcpu
))
6960 vcpu_scan_ioapic(vcpu
);
6961 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD
, vcpu
))
6962 kvm_vcpu_reload_apic_access_page(vcpu
);
6963 if (kvm_check_request(KVM_REQ_HV_CRASH
, vcpu
)) {
6964 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
6965 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_CRASH
;
6969 if (kvm_check_request(KVM_REQ_HV_RESET
, vcpu
)) {
6970 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
6971 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_RESET
;
6975 if (kvm_check_request(KVM_REQ_HV_EXIT
, vcpu
)) {
6976 vcpu
->run
->exit_reason
= KVM_EXIT_HYPERV
;
6977 vcpu
->run
->hyperv
= vcpu
->arch
.hyperv
.exit
;
6983 * KVM_REQ_HV_STIMER has to be processed after
6984 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6985 * depend on the guest clock being up-to-date
6987 if (kvm_check_request(KVM_REQ_HV_STIMER
, vcpu
))
6988 kvm_hv_process_stimers(vcpu
);
6991 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
6992 ++vcpu
->stat
.req_event
;
6993 kvm_apic_accept_events(vcpu
);
6994 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
6999 if (inject_pending_event(vcpu
, req_int_win
) != 0)
7000 req_immediate_exit
= true;
7002 /* Enable NMI/IRQ window open exits if needed.
7004 * SMIs have two cases: 1) they can be nested, and
7005 * then there is nothing to do here because RSM will
7006 * cause a vmexit anyway; 2) or the SMI can be pending
7007 * because inject_pending_event has completed the
7008 * injection of an IRQ or NMI from the previous vmexit,
7009 * and then we request an immediate exit to inject the SMI.
7011 if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
))
7012 req_immediate_exit
= true;
7013 if (vcpu
->arch
.nmi_pending
)
7014 kvm_x86_ops
->enable_nmi_window(vcpu
);
7015 if (kvm_cpu_has_injectable_intr(vcpu
) || req_int_win
)
7016 kvm_x86_ops
->enable_irq_window(vcpu
);
7017 WARN_ON(vcpu
->arch
.exception
.pending
);
7020 if (kvm_lapic_enabled(vcpu
)) {
7021 update_cr8_intercept(vcpu
);
7022 kvm_lapic_sync_to_vapic(vcpu
);
7026 r
= kvm_mmu_reload(vcpu
);
7028 goto cancel_injection
;
7033 kvm_x86_ops
->prepare_guest_switch(vcpu
);
7034 kvm_load_guest_fpu(vcpu
);
7037 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7038 * IPI are then delayed after guest entry, which ensures that they
7039 * result in virtual interrupt delivery.
7041 local_irq_disable();
7042 vcpu
->mode
= IN_GUEST_MODE
;
7044 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
7047 * 1) We should set ->mode before checking ->requests. Please see
7048 * the comment in kvm_vcpu_exiting_guest_mode().
7050 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7051 * pairs with the memory barrier implicit in pi_test_and_set_on
7052 * (see vmx_deliver_posted_interrupt).
7054 * 3) This also orders the write to mode from any reads to the page
7055 * tables done while the VCPU is running. Please see the comment
7056 * in kvm_flush_remote_tlbs.
7058 smp_mb__after_srcu_read_unlock();
7061 * This handles the case where a posted interrupt was
7062 * notified with kvm_vcpu_kick.
7064 if (kvm_lapic_enabled(vcpu
)) {
7065 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
7066 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
7069 if (vcpu
->mode
== EXITING_GUEST_MODE
|| kvm_request_pending(vcpu
)
7070 || need_resched() || signal_pending(current
)) {
7071 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
7075 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7077 goto cancel_injection
;
7080 kvm_load_guest_xcr0(vcpu
);
7082 if (req_immediate_exit
) {
7083 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7084 smp_send_reschedule(vcpu
->cpu
);
7087 trace_kvm_entry(vcpu
->vcpu_id
);
7088 wait_lapic_expire(vcpu
);
7089 guest_enter_irqoff();
7091 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
7093 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
7094 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
7095 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
7096 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
7097 set_debugreg(vcpu
->arch
.dr6
, 6);
7098 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
7101 kvm_x86_ops
->run(vcpu
);
7104 * Do this here before restoring debug registers on the host. And
7105 * since we do this before handling the vmexit, a DR access vmexit
7106 * can (a) read the correct value of the debug registers, (b) set
7107 * KVM_DEBUGREG_WONT_EXIT again.
7109 if (unlikely(vcpu
->arch
.switch_db_regs
& KVM_DEBUGREG_WONT_EXIT
)) {
7110 WARN_ON(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
);
7111 kvm_x86_ops
->sync_dirty_debug_regs(vcpu
);
7112 kvm_update_dr0123(vcpu
);
7113 kvm_update_dr6(vcpu
);
7114 kvm_update_dr7(vcpu
);
7115 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
7119 * If the guest has used debug registers, at least dr7
7120 * will be disabled while returning to the host.
7121 * If we don't have active breakpoints in the host, we don't
7122 * care about the messed up debug address registers. But if
7123 * we have some of them active, restore the old state.
7125 if (hw_breakpoint_active())
7126 hw_breakpoint_restore();
7128 vcpu
->arch
.last_guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
7130 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
7133 kvm_put_guest_xcr0(vcpu
);
7135 kvm_x86_ops
->handle_external_intr(vcpu
);
7139 guest_exit_irqoff();
7144 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7147 * Profile KVM exit RIPs:
7149 if (unlikely(prof_on
== KVM_PROFILING
)) {
7150 unsigned long rip
= kvm_rip_read(vcpu
);
7151 profile_hit(KVM_PROFILING
, (void *)rip
);
7154 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
7155 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
7157 if (vcpu
->arch
.apic_attention
)
7158 kvm_lapic_sync_from_vapic(vcpu
);
7160 vcpu
->arch
.gpa_available
= false;
7161 r
= kvm_x86_ops
->handle_exit(vcpu
);
7165 kvm_x86_ops
->cancel_injection(vcpu
);
7166 if (unlikely(vcpu
->arch
.apic_attention
))
7167 kvm_lapic_sync_from_vapic(vcpu
);
7172 static inline int vcpu_block(struct kvm
*kvm
, struct kvm_vcpu
*vcpu
)
7174 if (!kvm_arch_vcpu_runnable(vcpu
) &&
7175 (!kvm_x86_ops
->pre_block
|| kvm_x86_ops
->pre_block(vcpu
) == 0)) {
7176 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7177 kvm_vcpu_block(vcpu
);
7178 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7180 if (kvm_x86_ops
->post_block
)
7181 kvm_x86_ops
->post_block(vcpu
);
7183 if (!kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
7187 kvm_apic_accept_events(vcpu
);
7188 switch(vcpu
->arch
.mp_state
) {
7189 case KVM_MP_STATE_HALTED
:
7190 vcpu
->arch
.pv
.pv_unhalted
= false;
7191 vcpu
->arch
.mp_state
=
7192 KVM_MP_STATE_RUNNABLE
;
7193 case KVM_MP_STATE_RUNNABLE
:
7194 vcpu
->arch
.apf
.halted
= false;
7196 case KVM_MP_STATE_INIT_RECEIVED
:
7205 static inline bool kvm_vcpu_running(struct kvm_vcpu
*vcpu
)
7207 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
)
7208 kvm_x86_ops
->check_nested_events(vcpu
, false);
7210 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
7211 !vcpu
->arch
.apf
.halted
);
7214 static int vcpu_run(struct kvm_vcpu
*vcpu
)
7217 struct kvm
*kvm
= vcpu
->kvm
;
7219 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7220 vcpu
->arch
.l1tf_flush_l1d
= true;
7223 if (kvm_vcpu_running(vcpu
)) {
7224 r
= vcpu_enter_guest(vcpu
);
7226 r
= vcpu_block(kvm
, vcpu
);
7232 kvm_clear_request(KVM_REQ_PENDING_TIMER
, vcpu
);
7233 if (kvm_cpu_has_pending_timer(vcpu
))
7234 kvm_inject_pending_timer_irqs(vcpu
);
7236 if (dm_request_for_irq_injection(vcpu
) &&
7237 kvm_vcpu_ready_for_interrupt_injection(vcpu
)) {
7239 vcpu
->run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
7240 ++vcpu
->stat
.request_irq_exits
;
7244 kvm_check_async_pf_completion(vcpu
);
7246 if (signal_pending(current
)) {
7248 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
7249 ++vcpu
->stat
.signal_exits
;
7252 if (need_resched()) {
7253 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7255 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7259 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7264 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
7267 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7268 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
7269 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
7270 if (r
!= EMULATE_DONE
)
7275 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
7277 BUG_ON(!vcpu
->arch
.pio
.count
);
7279 return complete_emulated_io(vcpu
);
7283 * Implements the following, as a state machine:
7287 * for each mmio piece in the fragment
7295 * for each mmio piece in the fragment
7300 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
7302 struct kvm_run
*run
= vcpu
->run
;
7303 struct kvm_mmio_fragment
*frag
;
7306 BUG_ON(!vcpu
->mmio_needed
);
7308 /* Complete previous fragment */
7309 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
7310 len
= min(8u, frag
->len
);
7311 if (!vcpu
->mmio_is_write
)
7312 memcpy(frag
->data
, run
->mmio
.data
, len
);
7314 if (frag
->len
<= 8) {
7315 /* Switch to the next fragment. */
7317 vcpu
->mmio_cur_fragment
++;
7319 /* Go forward to the next mmio piece. */
7325 if (vcpu
->mmio_cur_fragment
>= vcpu
->mmio_nr_fragments
) {
7326 vcpu
->mmio_needed
= 0;
7328 /* FIXME: return into emulator if single-stepping. */
7329 if (vcpu
->mmio_is_write
)
7331 vcpu
->mmio_read_completed
= 1;
7332 return complete_emulated_io(vcpu
);
7335 run
->exit_reason
= KVM_EXIT_MMIO
;
7336 run
->mmio
.phys_addr
= frag
->gpa
;
7337 if (vcpu
->mmio_is_write
)
7338 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
7339 run
->mmio
.len
= min(8u, frag
->len
);
7340 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
7341 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
7346 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
7348 struct fpu
*fpu
= ¤t
->thread
.fpu
;
7351 fpu__initialize(fpu
);
7353 kvm_sigset_activate(vcpu
);
7355 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
7356 if (kvm_run
->immediate_exit
) {
7360 kvm_vcpu_block(vcpu
);
7361 kvm_apic_accept_events(vcpu
);
7362 kvm_clear_request(KVM_REQ_UNHALT
, vcpu
);
7364 if (signal_pending(current
)) {
7366 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
7367 ++vcpu
->stat
.signal_exits
;
7372 /* re-sync apic's tpr */
7373 if (!lapic_in_kernel(vcpu
)) {
7374 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
7380 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
7381 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
7382 vcpu
->arch
.complete_userspace_io
= NULL
;
7387 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
7389 if (kvm_run
->immediate_exit
)
7395 post_kvm_run_save(vcpu
);
7396 kvm_sigset_deactivate(vcpu
);
7401 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
7403 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
7405 * We are here if userspace calls get_regs() in the middle of
7406 * instruction emulation. Registers state needs to be copied
7407 * back from emulation context to vcpu. Userspace shouldn't do
7408 * that usually, but some bad designed PV devices (vmware
7409 * backdoor interface) need this to work
7411 emulator_writeback_register_cache(&vcpu
->arch
.emulate_ctxt
);
7412 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7414 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
7415 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
7416 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
7417 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
7418 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
7419 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
7420 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
7421 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
7422 #ifdef CONFIG_X86_64
7423 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
7424 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
7425 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
7426 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
7427 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
7428 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
7429 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
7430 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
7433 regs
->rip
= kvm_rip_read(vcpu
);
7434 regs
->rflags
= kvm_get_rflags(vcpu
);
7439 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
7441 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
7442 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7444 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
7445 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
7446 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
7447 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
7448 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
7449 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
7450 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
7451 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
7452 #ifdef CONFIG_X86_64
7453 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
7454 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
7455 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
7456 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
7457 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
7458 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
7459 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
7460 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
7463 kvm_rip_write(vcpu
, regs
->rip
);
7464 kvm_set_rflags(vcpu
, regs
->rflags
| X86_EFLAGS_FIXED
);
7466 vcpu
->arch
.exception
.pending
= false;
7468 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7473 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
7475 struct kvm_segment cs
;
7477 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7481 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
7483 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
7484 struct kvm_sregs
*sregs
)
7488 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7489 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7490 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7491 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7492 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7493 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7495 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7496 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7498 kvm_x86_ops
->get_idt(vcpu
, &dt
);
7499 sregs
->idt
.limit
= dt
.size
;
7500 sregs
->idt
.base
= dt
.address
;
7501 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
7502 sregs
->gdt
.limit
= dt
.size
;
7503 sregs
->gdt
.base
= dt
.address
;
7505 sregs
->cr0
= kvm_read_cr0(vcpu
);
7506 sregs
->cr2
= vcpu
->arch
.cr2
;
7507 sregs
->cr3
= kvm_read_cr3(vcpu
);
7508 sregs
->cr4
= kvm_read_cr4(vcpu
);
7509 sregs
->cr8
= kvm_get_cr8(vcpu
);
7510 sregs
->efer
= vcpu
->arch
.efer
;
7511 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
7513 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
7515 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
7516 set_bit(vcpu
->arch
.interrupt
.nr
,
7517 (unsigned long *)sregs
->interrupt_bitmap
);
7522 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
7523 struct kvm_mp_state
*mp_state
)
7525 kvm_apic_accept_events(vcpu
);
7526 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_HALTED
&&
7527 vcpu
->arch
.pv
.pv_unhalted
)
7528 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
7530 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
7535 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
7536 struct kvm_mp_state
*mp_state
)
7538 if (!lapic_in_kernel(vcpu
) &&
7539 mp_state
->mp_state
!= KVM_MP_STATE_RUNNABLE
)
7542 /* INITs are latched while in SMM */
7543 if ((is_smm(vcpu
) || vcpu
->arch
.smi_pending
) &&
7544 (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
||
7545 mp_state
->mp_state
== KVM_MP_STATE_INIT_RECEIVED
))
7548 if (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
) {
7549 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
7550 set_bit(KVM_APIC_SIPI
, &vcpu
->arch
.apic
->pending_events
);
7552 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
7553 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7557 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
7558 int reason
, bool has_error_code
, u32 error_code
)
7560 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
7563 init_emulate_ctxt(vcpu
);
7565 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
7566 has_error_code
, error_code
);
7569 return EMULATE_FAIL
;
7571 kvm_rip_write(vcpu
, ctxt
->eip
);
7572 kvm_set_rflags(vcpu
, ctxt
->eflags
);
7573 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7574 return EMULATE_DONE
;
7576 EXPORT_SYMBOL_GPL(kvm_task_switch
);
7578 int kvm_valid_sregs(struct kvm_vcpu
*vcpu
, struct kvm_sregs
*sregs
)
7580 if ((sregs
->efer
& EFER_LME
) && (sregs
->cr0
& X86_CR0_PG
)) {
7582 * When EFER.LME and CR0.PG are set, the processor is in
7583 * 64-bit mode (though maybe in a 32-bit code segment).
7584 * CR4.PAE and EFER.LMA must be set.
7586 if (!(sregs
->cr4
& X86_CR4_PAE
)
7587 || !(sregs
->efer
& EFER_LMA
))
7591 * Not in 64-bit mode: EFER.LMA is clear and the code
7592 * segment cannot be 64-bit.
7594 if (sregs
->efer
& EFER_LMA
|| sregs
->cs
.l
)
7601 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
7602 struct kvm_sregs
*sregs
)
7604 struct msr_data apic_base_msr
;
7605 int mmu_reset_needed
= 0;
7606 int cpuid_update_needed
= 0;
7607 int pending_vec
, max_bits
, idx
;
7610 if (!guest_cpuid_has(vcpu
, X86_FEATURE_XSAVE
) &&
7611 (sregs
->cr4
& X86_CR4_OSXSAVE
))
7614 if (kvm_valid_sregs(vcpu
, sregs
))
7617 apic_base_msr
.data
= sregs
->apic_base
;
7618 apic_base_msr
.host_initiated
= true;
7619 if (kvm_set_apic_base(vcpu
, &apic_base_msr
))
7622 dt
.size
= sregs
->idt
.limit
;
7623 dt
.address
= sregs
->idt
.base
;
7624 kvm_x86_ops
->set_idt(vcpu
, &dt
);
7625 dt
.size
= sregs
->gdt
.limit
;
7626 dt
.address
= sregs
->gdt
.base
;
7627 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
7629 vcpu
->arch
.cr2
= sregs
->cr2
;
7630 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
7631 vcpu
->arch
.cr3
= sregs
->cr3
;
7632 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
7634 kvm_set_cr8(vcpu
, sregs
->cr8
);
7636 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
7637 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
7639 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
7640 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
7641 vcpu
->arch
.cr0
= sregs
->cr0
;
7643 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
7644 cpuid_update_needed
|= ((kvm_read_cr4(vcpu
) ^ sregs
->cr4
) &
7645 (X86_CR4_OSXSAVE
| X86_CR4_PKE
));
7646 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
7647 if (cpuid_update_needed
)
7648 kvm_update_cpuid(vcpu
);
7650 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7651 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
7652 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
7653 mmu_reset_needed
= 1;
7655 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7657 if (mmu_reset_needed
)
7658 kvm_mmu_reset_context(vcpu
);
7660 max_bits
= KVM_NR_INTERRUPTS
;
7661 pending_vec
= find_first_bit(
7662 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
7663 if (pending_vec
< max_bits
) {
7664 kvm_queue_interrupt(vcpu
, pending_vec
, false);
7665 pr_debug("Set back pending irq %d\n", pending_vec
);
7668 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7669 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7670 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7671 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7672 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7673 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7675 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7676 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7678 update_cr8_intercept(vcpu
);
7680 /* Older userspace won't unhalt the vcpu on reset. */
7681 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
7682 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
7684 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7686 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7691 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
7692 struct kvm_guest_debug
*dbg
)
7694 unsigned long rflags
;
7697 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
7699 if (vcpu
->arch
.exception
.pending
)
7701 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
7702 kvm_queue_exception(vcpu
, DB_VECTOR
);
7704 kvm_queue_exception(vcpu
, BP_VECTOR
);
7708 * Read rflags as long as potentially injected trace flags are still
7711 rflags
= kvm_get_rflags(vcpu
);
7713 vcpu
->guest_debug
= dbg
->control
;
7714 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
7715 vcpu
->guest_debug
= 0;
7717 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
7718 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
7719 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
7720 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
7722 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
7723 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
7725 kvm_update_dr7(vcpu
);
7727 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
7728 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
7729 get_segment_base(vcpu
, VCPU_SREG_CS
);
7732 * Trigger an rflags update that will inject or remove the trace
7735 kvm_set_rflags(vcpu
, rflags
);
7737 kvm_x86_ops
->update_bp_intercept(vcpu
);
7747 * Translate a guest virtual address to a guest physical address.
7749 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
7750 struct kvm_translation
*tr
)
7752 unsigned long vaddr
= tr
->linear_address
;
7756 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7757 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
7758 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7759 tr
->physical_address
= gpa
;
7760 tr
->valid
= gpa
!= UNMAPPED_GVA
;
7767 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7769 struct fxregs_state
*fxsave
=
7770 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7772 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
7773 fpu
->fcw
= fxsave
->cwd
;
7774 fpu
->fsw
= fxsave
->swd
;
7775 fpu
->ftwx
= fxsave
->twd
;
7776 fpu
->last_opcode
= fxsave
->fop
;
7777 fpu
->last_ip
= fxsave
->rip
;
7778 fpu
->last_dp
= fxsave
->rdp
;
7779 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
7784 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7786 struct fxregs_state
*fxsave
=
7787 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7789 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
7790 fxsave
->cwd
= fpu
->fcw
;
7791 fxsave
->swd
= fpu
->fsw
;
7792 fxsave
->twd
= fpu
->ftwx
;
7793 fxsave
->fop
= fpu
->last_opcode
;
7794 fxsave
->rip
= fpu
->last_ip
;
7795 fxsave
->rdp
= fpu
->last_dp
;
7796 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
7801 static void fx_init(struct kvm_vcpu
*vcpu
)
7803 fpstate_init(&vcpu
->arch
.guest_fpu
.state
);
7804 if (boot_cpu_has(X86_FEATURE_XSAVES
))
7805 vcpu
->arch
.guest_fpu
.state
.xsave
.header
.xcomp_bv
=
7806 host_xcr0
| XSTATE_COMPACTION_ENABLED
;
7809 * Ensure guest xcr0 is valid for loading
7811 vcpu
->arch
.xcr0
= XFEATURE_MASK_FP
;
7813 vcpu
->arch
.cr0
|= X86_CR0_ET
;
7816 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
7818 if (vcpu
->guest_fpu_loaded
)
7822 * Restore all possible states in the guest,
7823 * and assume host would use all available bits.
7824 * Guest xcr0 would be loaded later.
7826 vcpu
->guest_fpu_loaded
= 1;
7827 __kernel_fpu_begin();
7828 /* PKRU is separately restored in kvm_x86_ops->run. */
7829 __copy_kernel_to_fpregs(&vcpu
->arch
.guest_fpu
.state
,
7830 ~XFEATURE_MASK_PKRU
);
7834 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
7836 if (!vcpu
->guest_fpu_loaded
)
7839 vcpu
->guest_fpu_loaded
= 0;
7840 copy_fpregs_to_fpstate(&vcpu
->arch
.guest_fpu
);
7842 ++vcpu
->stat
.fpu_reload
;
7846 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
7848 void *wbinvd_dirty_mask
= vcpu
->arch
.wbinvd_dirty_mask
;
7850 kvmclock_reset(vcpu
);
7852 kvm_x86_ops
->vcpu_free(vcpu
);
7853 free_cpumask_var(wbinvd_dirty_mask
);
7856 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
7859 struct kvm_vcpu
*vcpu
;
7861 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
7862 printk_once(KERN_WARNING
7863 "kvm: SMP vm created on host with unstable TSC; "
7864 "guest TSC will not be reliable\n");
7866 vcpu
= kvm_x86_ops
->vcpu_create(kvm
, id
);
7871 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
7875 kvm_vcpu_mtrr_init(vcpu
);
7876 r
= vcpu_load(vcpu
);
7879 kvm_vcpu_reset(vcpu
, false);
7880 kvm_mmu_setup(vcpu
);
7885 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
7887 struct msr_data msr
;
7888 struct kvm
*kvm
= vcpu
->kvm
;
7890 kvm_hv_vcpu_postcreate(vcpu
);
7892 if (vcpu_load(vcpu
))
7895 msr
.index
= MSR_IA32_TSC
;
7896 msr
.host_initiated
= true;
7897 kvm_write_tsc(vcpu
, &msr
);
7900 if (!kvmclock_periodic_sync
)
7903 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
7904 KVMCLOCK_SYNC_PERIOD
);
7907 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
7910 vcpu
->arch
.apf
.msr_val
= 0;
7912 r
= vcpu_load(vcpu
);
7914 kvm_mmu_unload(vcpu
);
7917 kvm_x86_ops
->vcpu_free(vcpu
);
7920 void kvm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
7922 kvm_lapic_reset(vcpu
, init_event
);
7924 vcpu
->arch
.hflags
= 0;
7926 vcpu
->arch
.smi_pending
= 0;
7927 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
7928 vcpu
->arch
.nmi_pending
= 0;
7929 vcpu
->arch
.nmi_injected
= false;
7930 kvm_clear_interrupt_queue(vcpu
);
7931 kvm_clear_exception_queue(vcpu
);
7932 vcpu
->arch
.exception
.pending
= false;
7934 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
7935 kvm_update_dr0123(vcpu
);
7936 vcpu
->arch
.dr6
= DR6_INIT
;
7937 kvm_update_dr6(vcpu
);
7938 vcpu
->arch
.dr7
= DR7_FIXED_1
;
7939 kvm_update_dr7(vcpu
);
7943 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7944 vcpu
->arch
.apf
.msr_val
= 0;
7945 vcpu
->arch
.st
.msr_val
= 0;
7947 kvmclock_reset(vcpu
);
7949 kvm_clear_async_pf_completion_queue(vcpu
);
7950 kvm_async_pf_hash_reset(vcpu
);
7951 vcpu
->arch
.apf
.halted
= false;
7954 kvm_pmu_reset(vcpu
);
7955 vcpu
->arch
.smbase
= 0x30000;
7957 vcpu
->arch
.msr_platform_info
= MSR_PLATFORM_INFO_CPUID_FAULT
;
7958 vcpu
->arch
.msr_misc_features_enables
= 0;
7961 memset(vcpu
->arch
.regs
, 0, sizeof(vcpu
->arch
.regs
));
7962 vcpu
->arch
.regs_avail
= ~0;
7963 vcpu
->arch
.regs_dirty
= ~0;
7965 kvm_x86_ops
->vcpu_reset(vcpu
, init_event
);
7968 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu
*vcpu
, u8 vector
)
7970 struct kvm_segment cs
;
7972 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7973 cs
.selector
= vector
<< 8;
7974 cs
.base
= vector
<< 12;
7975 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7976 kvm_rip_write(vcpu
, 0);
7979 int kvm_arch_hardware_enable(void)
7982 struct kvm_vcpu
*vcpu
;
7987 bool stable
, backwards_tsc
= false;
7989 kvm_shared_msr_cpu_online();
7990 ret
= kvm_x86_ops
->hardware_enable();
7994 local_tsc
= rdtsc();
7995 stable
= !check_tsc_unstable();
7996 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
7997 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
7998 if (!stable
&& vcpu
->cpu
== smp_processor_id())
7999 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
8000 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
8001 backwards_tsc
= true;
8002 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
8003 max_tsc
= vcpu
->arch
.last_host_tsc
;
8009 * Sometimes, even reliable TSCs go backwards. This happens on
8010 * platforms that reset TSC during suspend or hibernate actions, but
8011 * maintain synchronization. We must compensate. Fortunately, we can
8012 * detect that condition here, which happens early in CPU bringup,
8013 * before any KVM threads can be running. Unfortunately, we can't
8014 * bring the TSCs fully up to date with real time, as we aren't yet far
8015 * enough into CPU bringup that we know how much real time has actually
8016 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8017 * variables that haven't been updated yet.
8019 * So we simply find the maximum observed TSC above, then record the
8020 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8021 * the adjustment will be applied. Note that we accumulate
8022 * adjustments, in case multiple suspend cycles happen before some VCPU
8023 * gets a chance to run again. In the event that no KVM threads get a
8024 * chance to run, we will miss the entire elapsed period, as we'll have
8025 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8026 * loose cycle time. This isn't too big a deal, since the loss will be
8027 * uniform across all VCPUs (not to mention the scenario is extremely
8028 * unlikely). It is possible that a second hibernate recovery happens
8029 * much faster than a first, causing the observed TSC here to be
8030 * smaller; this would require additional padding adjustment, which is
8031 * why we set last_host_tsc to the local tsc observed here.
8033 * N.B. - this code below runs only on platforms with reliable TSC,
8034 * as that is the only way backwards_tsc is set above. Also note
8035 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8036 * have the same delta_cyc adjustment applied if backwards_tsc
8037 * is detected. Note further, this adjustment is only done once,
8038 * as we reset last_host_tsc on all VCPUs to stop this from being
8039 * called multiple times (one for each physical CPU bringup).
8041 * Platforms with unreliable TSCs don't have to deal with this, they
8042 * will be compensated by the logic in vcpu_load, which sets the TSC to
8043 * catchup mode. This will catchup all VCPUs to real time, but cannot
8044 * guarantee that they stay in perfect synchronization.
8046 if (backwards_tsc
) {
8047 u64 delta_cyc
= max_tsc
- local_tsc
;
8048 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
8049 kvm
->arch
.backwards_tsc_observed
= true;
8050 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8051 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
8052 vcpu
->arch
.last_host_tsc
= local_tsc
;
8053 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
8057 * We have to disable TSC offset matching.. if you were
8058 * booting a VM while issuing an S4 host suspend....
8059 * you may have some problem. Solving this issue is
8060 * left as an exercise to the reader.
8062 kvm
->arch
.last_tsc_nsec
= 0;
8063 kvm
->arch
.last_tsc_write
= 0;
8070 void kvm_arch_hardware_disable(void)
8072 kvm_x86_ops
->hardware_disable();
8073 drop_user_return_notifiers();
8076 int kvm_arch_hardware_setup(void)
8080 r
= kvm_x86_ops
->hardware_setup();
8084 if (kvm_has_tsc_control
) {
8086 * Make sure the user can only configure tsc_khz values that
8087 * fit into a signed integer.
8088 * A min value is not calculated needed because it will always
8089 * be 1 on all machines.
8091 u64 max
= min(0x7fffffffULL
,
8092 __scale_tsc(kvm_max_tsc_scaling_ratio
, tsc_khz
));
8093 kvm_max_guest_tsc_khz
= max
;
8095 kvm_default_tsc_scaling_ratio
= 1ULL << kvm_tsc_scaling_ratio_frac_bits
;
8098 kvm_init_msr_list();
8102 void kvm_arch_hardware_unsetup(void)
8104 kvm_x86_ops
->hardware_unsetup();
8107 void kvm_arch_check_processor_compat(void *rtn
)
8109 kvm_x86_ops
->check_processor_compatibility(rtn
);
8112 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu
*vcpu
)
8114 return vcpu
->kvm
->arch
.bsp_vcpu_id
== vcpu
->vcpu_id
;
8116 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp
);
8118 bool kvm_vcpu_is_bsp(struct kvm_vcpu
*vcpu
)
8120 return (vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_BSP
) != 0;
8123 struct static_key kvm_no_apic_vcpu __read_mostly
;
8124 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu
);
8126 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
8132 BUG_ON(vcpu
->kvm
== NULL
);
8135 vcpu
->arch
.apicv_active
= kvm_x86_ops
->get_enable_apicv(vcpu
);
8136 vcpu
->arch
.pv
.pv_unhalted
= false;
8137 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
8138 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_reset_bsp(vcpu
))
8139 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
8141 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
8143 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
8148 vcpu
->arch
.pio_data
= page_address(page
);
8150 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
8152 r
= kvm_mmu_create(vcpu
);
8154 goto fail_free_pio_data
;
8156 if (irqchip_in_kernel(kvm
)) {
8157 r
= kvm_create_lapic(vcpu
);
8159 goto fail_mmu_destroy
;
8161 static_key_slow_inc(&kvm_no_apic_vcpu
);
8163 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
8165 if (!vcpu
->arch
.mce_banks
) {
8167 goto fail_free_lapic
;
8169 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
8171 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
)) {
8173 goto fail_free_mce_banks
;
8178 vcpu
->arch
.ia32_tsc_adjust_msr
= 0x0;
8179 vcpu
->arch
.pv_time_enabled
= false;
8181 vcpu
->arch
.guest_supported_xcr0
= 0;
8182 vcpu
->arch
.guest_xstate_size
= XSAVE_HDR_SIZE
+ XSAVE_HDR_OFFSET
;
8184 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
8186 vcpu
->arch
.pat
= MSR_IA32_CR_PAT_DEFAULT
;
8188 kvm_async_pf_hash_reset(vcpu
);
8191 vcpu
->arch
.pending_external_vector
= -1;
8192 vcpu
->arch
.preempted_in_kernel
= false;
8194 kvm_hv_vcpu_init(vcpu
);
8198 fail_free_mce_banks
:
8199 kfree(vcpu
->arch
.mce_banks
);
8201 kvm_free_lapic(vcpu
);
8203 kvm_mmu_destroy(vcpu
);
8205 free_page((unsigned long)vcpu
->arch
.pio_data
);
8210 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
8214 kvm_hv_vcpu_uninit(vcpu
);
8215 kvm_pmu_destroy(vcpu
);
8216 kfree(vcpu
->arch
.mce_banks
);
8217 kvm_free_lapic(vcpu
);
8218 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
8219 kvm_mmu_destroy(vcpu
);
8220 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
8221 free_page((unsigned long)vcpu
->arch
.pio_data
);
8222 if (!lapic_in_kernel(vcpu
))
8223 static_key_slow_dec(&kvm_no_apic_vcpu
);
8226 void kvm_arch_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
8228 vcpu
->arch
.l1tf_flush_l1d
= true;
8229 kvm_x86_ops
->sched_in(vcpu
, cpu
);
8232 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
8237 INIT_HLIST_HEAD(&kvm
->arch
.mask_notifier_list
);
8238 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
8239 INIT_LIST_HEAD(&kvm
->arch
.zapped_obsolete_pages
);
8240 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
8241 atomic_set(&kvm
->arch
.noncoherent_dma_count
, 0);
8243 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8244 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
8245 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8246 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
8247 &kvm
->arch
.irq_sources_bitmap
);
8249 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
8250 mutex_init(&kvm
->arch
.apic_map_lock
);
8251 mutex_init(&kvm
->arch
.hyperv
.hv_lock
);
8252 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
8254 kvm
->arch
.kvmclock_offset
= -ktime_get_boot_ns();
8255 pvclock_update_vm_gtod_copy(kvm
);
8257 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_update_work
, kvmclock_update_fn
);
8258 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_sync_work
, kvmclock_sync_fn
);
8260 kvm_page_track_init(kvm
);
8261 kvm_mmu_init_vm(kvm
);
8263 if (kvm_x86_ops
->vm_init
)
8264 return kvm_x86_ops
->vm_init(kvm
);
8269 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
8272 r
= vcpu_load(vcpu
);
8274 kvm_mmu_unload(vcpu
);
8278 static void kvm_free_vcpus(struct kvm
*kvm
)
8281 struct kvm_vcpu
*vcpu
;
8284 * Unpin any mmu pages first.
8286 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8287 kvm_clear_async_pf_completion_queue(vcpu
);
8288 kvm_unload_vcpu_mmu(vcpu
);
8290 kvm_for_each_vcpu(i
, vcpu
, kvm
)
8291 kvm_arch_vcpu_free(vcpu
);
8293 mutex_lock(&kvm
->lock
);
8294 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
8295 kvm
->vcpus
[i
] = NULL
;
8297 atomic_set(&kvm
->online_vcpus
, 0);
8298 mutex_unlock(&kvm
->lock
);
8301 void kvm_arch_sync_events(struct kvm
*kvm
)
8303 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_sync_work
);
8304 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_update_work
);
8308 int __x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
8312 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
8313 struct kvm_memory_slot
*slot
, old
;
8315 /* Called with kvm->slots_lock held. */
8316 if (WARN_ON(id
>= KVM_MEM_SLOTS_NUM
))
8319 slot
= id_to_memslot(slots
, id
);
8325 * MAP_SHARED to prevent internal slot pages from being moved
8328 hva
= vm_mmap(NULL
, 0, size
, PROT_READ
| PROT_WRITE
,
8329 MAP_SHARED
| MAP_ANONYMOUS
, 0);
8330 if (IS_ERR((void *)hva
))
8331 return PTR_ERR((void *)hva
);
8340 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
8341 struct kvm_userspace_memory_region m
;
8343 m
.slot
= id
| (i
<< 16);
8345 m
.guest_phys_addr
= gpa
;
8346 m
.userspace_addr
= hva
;
8347 m
.memory_size
= size
;
8348 r
= __kvm_set_memory_region(kvm
, &m
);
8354 vm_munmap(old
.userspace_addr
, old
.npages
* PAGE_SIZE
);
8358 EXPORT_SYMBOL_GPL(__x86_set_memory_region
);
8360 int x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
8364 mutex_lock(&kvm
->slots_lock
);
8365 r
= __x86_set_memory_region(kvm
, id
, gpa
, size
);
8366 mutex_unlock(&kvm
->slots_lock
);
8370 EXPORT_SYMBOL_GPL(x86_set_memory_region
);
8372 void kvm_arch_destroy_vm(struct kvm
*kvm
)
8374 if (current
->mm
== kvm
->mm
) {
8376 * Free memory regions allocated on behalf of userspace,
8377 * unless the the memory map has changed due to process exit
8380 x86_set_memory_region(kvm
, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
, 0, 0);
8381 x86_set_memory_region(kvm
, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
, 0, 0);
8382 x86_set_memory_region(kvm
, TSS_PRIVATE_MEMSLOT
, 0, 0);
8384 if (kvm_x86_ops
->vm_destroy
)
8385 kvm_x86_ops
->vm_destroy(kvm
);
8386 kvm_pic_destroy(kvm
);
8387 kvm_ioapic_destroy(kvm
);
8388 kvm_free_vcpus(kvm
);
8389 kvfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
8390 kvm_mmu_uninit_vm(kvm
);
8391 kvm_page_track_cleanup(kvm
);
8394 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
8395 struct kvm_memory_slot
*dont
)
8399 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8400 if (!dont
|| free
->arch
.rmap
[i
] != dont
->arch
.rmap
[i
]) {
8401 kvfree(free
->arch
.rmap
[i
]);
8402 free
->arch
.rmap
[i
] = NULL
;
8407 if (!dont
|| free
->arch
.lpage_info
[i
- 1] !=
8408 dont
->arch
.lpage_info
[i
- 1]) {
8409 kvfree(free
->arch
.lpage_info
[i
- 1]);
8410 free
->arch
.lpage_info
[i
- 1] = NULL
;
8414 kvm_page_track_free_memslot(free
, dont
);
8417 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
8418 unsigned long npages
)
8422 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8423 struct kvm_lpage_info
*linfo
;
8428 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
8429 slot
->base_gfn
, level
) + 1;
8431 slot
->arch
.rmap
[i
] =
8432 kvzalloc(lpages
* sizeof(*slot
->arch
.rmap
[i
]), GFP_KERNEL
);
8433 if (!slot
->arch
.rmap
[i
])
8438 linfo
= kvzalloc(lpages
* sizeof(*linfo
), GFP_KERNEL
);
8442 slot
->arch
.lpage_info
[i
- 1] = linfo
;
8444 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
8445 linfo
[0].disallow_lpage
= 1;
8446 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
8447 linfo
[lpages
- 1].disallow_lpage
= 1;
8448 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
8450 * If the gfn and userspace address are not aligned wrt each
8451 * other, or if explicitly asked to, disable large page
8452 * support for this slot
8454 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
8455 !kvm_largepages_enabled()) {
8458 for (j
= 0; j
< lpages
; ++j
)
8459 linfo
[j
].disallow_lpage
= 1;
8463 if (kvm_page_track_create_memslot(slot
, npages
))
8469 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8470 kvfree(slot
->arch
.rmap
[i
]);
8471 slot
->arch
.rmap
[i
] = NULL
;
8475 kvfree(slot
->arch
.lpage_info
[i
- 1]);
8476 slot
->arch
.lpage_info
[i
- 1] = NULL
;
8481 void kvm_arch_memslots_updated(struct kvm
*kvm
, struct kvm_memslots
*slots
)
8484 * memslots->generation has been incremented.
8485 * mmio generation may have reached its maximum value.
8487 kvm_mmu_invalidate_mmio_sptes(kvm
, slots
);
8490 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
8491 struct kvm_memory_slot
*memslot
,
8492 const struct kvm_userspace_memory_region
*mem
,
8493 enum kvm_mr_change change
)
8498 static void kvm_mmu_slot_apply_flags(struct kvm
*kvm
,
8499 struct kvm_memory_slot
*new)
8501 /* Still write protect RO slot */
8502 if (new->flags
& KVM_MEM_READONLY
) {
8503 kvm_mmu_slot_remove_write_access(kvm
, new);
8508 * Call kvm_x86_ops dirty logging hooks when they are valid.
8510 * kvm_x86_ops->slot_disable_log_dirty is called when:
8512 * - KVM_MR_CREATE with dirty logging is disabled
8513 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8515 * The reason is, in case of PML, we need to set D-bit for any slots
8516 * with dirty logging disabled in order to eliminate unnecessary GPA
8517 * logging in PML buffer (and potential PML buffer full VMEXT). This
8518 * guarantees leaving PML enabled during guest's lifetime won't have
8519 * any additonal overhead from PML when guest is running with dirty
8520 * logging disabled for memory slots.
8522 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8523 * to dirty logging mode.
8525 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8527 * In case of write protect:
8529 * Write protect all pages for dirty logging.
8531 * All the sptes including the large sptes which point to this
8532 * slot are set to readonly. We can not create any new large
8533 * spte on this slot until the end of the logging.
8535 * See the comments in fast_page_fault().
8537 if (new->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
8538 if (kvm_x86_ops
->slot_enable_log_dirty
)
8539 kvm_x86_ops
->slot_enable_log_dirty(kvm
, new);
8541 kvm_mmu_slot_remove_write_access(kvm
, new);
8543 if (kvm_x86_ops
->slot_disable_log_dirty
)
8544 kvm_x86_ops
->slot_disable_log_dirty(kvm
, new);
8548 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
8549 const struct kvm_userspace_memory_region
*mem
,
8550 const struct kvm_memory_slot
*old
,
8551 const struct kvm_memory_slot
*new,
8552 enum kvm_mr_change change
)
8554 int nr_mmu_pages
= 0;
8556 if (!kvm
->arch
.n_requested_mmu_pages
)
8557 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
8560 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
8563 * Dirty logging tracks sptes in 4k granularity, meaning that large
8564 * sptes have to be split. If live migration is successful, the guest
8565 * in the source machine will be destroyed and large sptes will be
8566 * created in the destination. However, if the guest continues to run
8567 * in the source machine (for example if live migration fails), small
8568 * sptes will remain around and cause bad performance.
8570 * Scan sptes if dirty logging has been stopped, dropping those
8571 * which can be collapsed into a single large-page spte. Later
8572 * page faults will create the large-page sptes.
8574 if ((change
!= KVM_MR_DELETE
) &&
8575 (old
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) &&
8576 !(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
8577 kvm_mmu_zap_collapsible_sptes(kvm
, new);
8580 * Set up write protection and/or dirty logging for the new slot.
8582 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8583 * been zapped so no dirty logging staff is needed for old slot. For
8584 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8585 * new and it's also covered when dealing with the new slot.
8587 * FIXME: const-ify all uses of struct kvm_memory_slot.
8589 if (change
!= KVM_MR_DELETE
)
8590 kvm_mmu_slot_apply_flags(kvm
, (struct kvm_memory_slot
*) new);
8593 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
8595 kvm_mmu_invalidate_zap_all_pages(kvm
);
8598 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
8599 struct kvm_memory_slot
*slot
)
8601 kvm_page_track_flush_slot(kvm
, slot
);
8604 static inline bool kvm_vcpu_has_events(struct kvm_vcpu
*vcpu
)
8606 if (!list_empty_careful(&vcpu
->async_pf
.done
))
8609 if (kvm_apic_has_events(vcpu
))
8612 if (vcpu
->arch
.pv
.pv_unhalted
)
8615 if (vcpu
->arch
.exception
.pending
)
8618 if (kvm_test_request(KVM_REQ_NMI
, vcpu
) ||
8619 (vcpu
->arch
.nmi_pending
&&
8620 kvm_x86_ops
->nmi_allowed(vcpu
)))
8623 if (kvm_test_request(KVM_REQ_SMI
, vcpu
) ||
8624 (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
)))
8627 if (kvm_arch_interrupt_allowed(vcpu
) &&
8628 kvm_cpu_has_interrupt(vcpu
))
8631 if (kvm_hv_has_stimer_pending(vcpu
))
8637 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
8639 return kvm_vcpu_running(vcpu
) || kvm_vcpu_has_events(vcpu
);
8642 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
8644 return vcpu
->arch
.preempted_in_kernel
;
8647 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
8649 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
8652 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
8654 return kvm_x86_ops
->interrupt_allowed(vcpu
);
8657 unsigned long kvm_get_linear_rip(struct kvm_vcpu
*vcpu
)
8659 if (is_64_bit_mode(vcpu
))
8660 return kvm_rip_read(vcpu
);
8661 return (u32
)(get_segment_base(vcpu
, VCPU_SREG_CS
) +
8662 kvm_rip_read(vcpu
));
8664 EXPORT_SYMBOL_GPL(kvm_get_linear_rip
);
8666 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
8668 return kvm_get_linear_rip(vcpu
) == linear_rip
;
8670 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
8672 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
8674 unsigned long rflags
;
8676 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
8677 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
8678 rflags
&= ~X86_EFLAGS_TF
;
8681 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
8683 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8685 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
8686 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
8687 rflags
|= X86_EFLAGS_TF
;
8688 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
8691 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8693 __kvm_set_rflags(vcpu
, rflags
);
8694 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8696 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
8698 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
8702 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
8706 r
= kvm_mmu_reload(vcpu
);
8710 if (!vcpu
->arch
.mmu
.direct_map
&&
8711 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
8714 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
8717 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
8719 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
8722 static inline u32
kvm_async_pf_next_probe(u32 key
)
8724 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
8727 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8729 u32 key
= kvm_async_pf_hash_fn(gfn
);
8731 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
8732 key
= kvm_async_pf_next_probe(key
);
8734 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
8737 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8740 u32 key
= kvm_async_pf_hash_fn(gfn
);
8742 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
8743 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
8744 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
8745 key
= kvm_async_pf_next_probe(key
);
8750 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8752 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
8755 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8759 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
8761 vcpu
->arch
.apf
.gfns
[i
] = ~0;
8763 j
= kvm_async_pf_next_probe(j
);
8764 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
8766 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
8768 * k lies cyclically in ]i,j]
8770 * |....j i.k.| or |.k..j i...|
8772 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
8773 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
8778 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
8781 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
8785 static int apf_get_user(struct kvm_vcpu
*vcpu
, u32
*val
)
8788 return kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, val
,
8792 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
8793 struct kvm_async_pf
*work
)
8795 struct x86_exception fault
;
8797 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
8798 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8800 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
8801 (vcpu
->arch
.apf
.send_user_only
&&
8802 kvm_x86_ops
->get_cpl(vcpu
) == 0))
8803 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
8804 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
8805 fault
.vector
= PF_VECTOR
;
8806 fault
.error_code_valid
= true;
8807 fault
.error_code
= 0;
8808 fault
.nested_page_fault
= false;
8809 fault
.address
= work
->arch
.token
;
8810 fault
.async_page_fault
= true;
8811 kvm_inject_page_fault(vcpu
, &fault
);
8815 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
8816 struct kvm_async_pf
*work
)
8818 struct x86_exception fault
;
8821 if (work
->wakeup_all
)
8822 work
->arch
.token
= ~0; /* broadcast wakeup */
8824 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8825 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
8827 if (vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
&&
8828 !apf_get_user(vcpu
, &val
)) {
8829 if (val
== KVM_PV_REASON_PAGE_NOT_PRESENT
&&
8830 vcpu
->arch
.exception
.pending
&&
8831 vcpu
->arch
.exception
.nr
== PF_VECTOR
&&
8832 !apf_put_user(vcpu
, 0)) {
8833 vcpu
->arch
.exception
.injected
= false;
8834 vcpu
->arch
.exception
.pending
= false;
8835 vcpu
->arch
.exception
.nr
= 0;
8836 vcpu
->arch
.exception
.has_error_code
= false;
8837 vcpu
->arch
.exception
.error_code
= 0;
8838 } else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
8839 fault
.vector
= PF_VECTOR
;
8840 fault
.error_code_valid
= true;
8841 fault
.error_code
= 0;
8842 fault
.nested_page_fault
= false;
8843 fault
.address
= work
->arch
.token
;
8844 fault
.async_page_fault
= true;
8845 kvm_inject_page_fault(vcpu
, &fault
);
8848 vcpu
->arch
.apf
.halted
= false;
8849 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
8852 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
8854 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
8857 return kvm_can_do_async_pf(vcpu
);
8860 void kvm_arch_start_assignment(struct kvm
*kvm
)
8862 atomic_inc(&kvm
->arch
.assigned_device_count
);
8864 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment
);
8866 void kvm_arch_end_assignment(struct kvm
*kvm
)
8868 atomic_dec(&kvm
->arch
.assigned_device_count
);
8870 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment
);
8872 bool kvm_arch_has_assigned_device(struct kvm
*kvm
)
8874 return atomic_read(&kvm
->arch
.assigned_device_count
);
8876 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device
);
8878 void kvm_arch_register_noncoherent_dma(struct kvm
*kvm
)
8880 atomic_inc(&kvm
->arch
.noncoherent_dma_count
);
8882 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma
);
8884 void kvm_arch_unregister_noncoherent_dma(struct kvm
*kvm
)
8886 atomic_dec(&kvm
->arch
.noncoherent_dma_count
);
8888 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma
);
8890 bool kvm_arch_has_noncoherent_dma(struct kvm
*kvm
)
8892 return atomic_read(&kvm
->arch
.noncoherent_dma_count
);
8894 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma
);
8896 bool kvm_arch_has_irq_bypass(void)
8898 return kvm_x86_ops
->update_pi_irte
!= NULL
;
8901 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
8902 struct irq_bypass_producer
*prod
)
8904 struct kvm_kernel_irqfd
*irqfd
=
8905 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
8907 irqfd
->producer
= prod
;
8909 return kvm_x86_ops
->update_pi_irte(irqfd
->kvm
,
8910 prod
->irq
, irqfd
->gsi
, 1);
8913 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
8914 struct irq_bypass_producer
*prod
)
8917 struct kvm_kernel_irqfd
*irqfd
=
8918 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
8920 WARN_ON(irqfd
->producer
!= prod
);
8921 irqfd
->producer
= NULL
;
8924 * When producer of consumer is unregistered, we change back to
8925 * remapped mode, so we can re-use the current implementation
8926 * when the irq is masked/disabled or the consumer side (KVM
8927 * int this case doesn't want to receive the interrupts.
8929 ret
= kvm_x86_ops
->update_pi_irte(irqfd
->kvm
, prod
->irq
, irqfd
->gsi
, 0);
8931 printk(KERN_INFO
"irq bypass consumer (token %p) unregistration"
8932 " fails: %d\n", irqfd
->consumer
.token
, ret
);
8935 int kvm_arch_update_irqfd_routing(struct kvm
*kvm
, unsigned int host_irq
,
8936 uint32_t guest_irq
, bool set
)
8938 if (!kvm_x86_ops
->update_pi_irte
)
8941 return kvm_x86_ops
->update_pi_irte(kvm
, host_irq
, guest_irq
, set
);
8944 bool kvm_vector_hashing_enabled(void)
8946 return vector_hashing
;
8948 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled
);
8950 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
8951 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio
);
8952 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
8953 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
8954 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
8955 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
8956 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
8957 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
8958 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
8959 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
8960 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
8961 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
8962 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);
8963 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset
);
8964 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window
);
8965 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full
);
8966 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update
);
8967 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access
);
8968 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi
);