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 MSR_IA32_ARCH_CAPABILITIES
,
1055 static unsigned int num_msr_based_features
;
1057 static int kvm_get_msr_feature(struct kvm_msr_entry
*msr
)
1059 switch (msr
->index
) {
1060 case MSR_IA32_UCODE_REV
:
1061 case MSR_IA32_ARCH_CAPABILITIES
:
1062 rdmsrl_safe(msr
->index
, &msr
->data
);
1065 if (kvm_x86_ops
->get_msr_feature(msr
))
1071 static int do_get_msr_feature(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1073 struct kvm_msr_entry msr
;
1077 r
= kvm_get_msr_feature(&msr
);
1086 bool kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1088 if (efer
& efer_reserved_bits
)
1091 if (efer
& EFER_FFXSR
&& !guest_cpuid_has(vcpu
, X86_FEATURE_FXSR_OPT
))
1094 if (efer
& EFER_SVME
&& !guest_cpuid_has(vcpu
, X86_FEATURE_SVM
))
1099 EXPORT_SYMBOL_GPL(kvm_valid_efer
);
1101 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1103 u64 old_efer
= vcpu
->arch
.efer
;
1105 if (!kvm_valid_efer(vcpu
, efer
))
1109 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
1113 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
1115 kvm_x86_ops
->set_efer(vcpu
, efer
);
1117 /* Update reserved bits */
1118 if ((efer
^ old_efer
) & EFER_NX
)
1119 kvm_mmu_reset_context(vcpu
);
1124 void kvm_enable_efer_bits(u64 mask
)
1126 efer_reserved_bits
&= ~mask
;
1128 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
1131 * Writes msr value into into the appropriate "register".
1132 * Returns 0 on success, non-0 otherwise.
1133 * Assumes vcpu_load() was already called.
1135 int kvm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1137 switch (msr
->index
) {
1140 case MSR_KERNEL_GS_BASE
:
1143 if (is_noncanonical_address(msr
->data
, vcpu
))
1146 case MSR_IA32_SYSENTER_EIP
:
1147 case MSR_IA32_SYSENTER_ESP
:
1149 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1150 * non-canonical address is written on Intel but not on
1151 * AMD (which ignores the top 32-bits, because it does
1152 * not implement 64-bit SYSENTER).
1154 * 64-bit code should hence be able to write a non-canonical
1155 * value on AMD. Making the address canonical ensures that
1156 * vmentry does not fail on Intel after writing a non-canonical
1157 * value, and that something deterministic happens if the guest
1158 * invokes 64-bit SYSENTER.
1160 msr
->data
= get_canonical(msr
->data
, vcpu_virt_addr_bits(vcpu
));
1162 return kvm_x86_ops
->set_msr(vcpu
, msr
);
1164 EXPORT_SYMBOL_GPL(kvm_set_msr
);
1167 * Adapt set_msr() to msr_io()'s calling convention
1169 static int do_get_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1171 struct msr_data msr
;
1175 msr
.host_initiated
= true;
1176 r
= kvm_get_msr(vcpu
, &msr
);
1184 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1186 struct msr_data msr
;
1190 msr
.host_initiated
= true;
1191 return kvm_set_msr(vcpu
, &msr
);
1194 #ifdef CONFIG_X86_64
1195 struct pvclock_gtod_data
{
1198 struct { /* extract of a clocksource struct */
1211 static struct pvclock_gtod_data pvclock_gtod_data
;
1213 static void update_pvclock_gtod(struct timekeeper
*tk
)
1215 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
1218 boot_ns
= ktime_to_ns(ktime_add(tk
->tkr_mono
.base
, tk
->offs_boot
));
1220 write_seqcount_begin(&vdata
->seq
);
1222 /* copy pvclock gtod data */
1223 vdata
->clock
.vclock_mode
= tk
->tkr_mono
.clock
->archdata
.vclock_mode
;
1224 vdata
->clock
.cycle_last
= tk
->tkr_mono
.cycle_last
;
1225 vdata
->clock
.mask
= tk
->tkr_mono
.mask
;
1226 vdata
->clock
.mult
= tk
->tkr_mono
.mult
;
1227 vdata
->clock
.shift
= tk
->tkr_mono
.shift
;
1229 vdata
->boot_ns
= boot_ns
;
1230 vdata
->nsec_base
= tk
->tkr_mono
.xtime_nsec
;
1232 vdata
->wall_time_sec
= tk
->xtime_sec
;
1234 write_seqcount_end(&vdata
->seq
);
1238 void kvm_set_pending_timer(struct kvm_vcpu
*vcpu
)
1241 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1242 * vcpu_enter_guest. This function is only called from
1243 * the physical CPU that is running vcpu.
1245 kvm_make_request(KVM_REQ_PENDING_TIMER
, vcpu
);
1248 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
1252 struct pvclock_wall_clock wc
;
1253 struct timespec64 boot
;
1258 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1263 ++version
; /* first time write, random junk */
1267 if (kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
)))
1271 * The guest calculates current wall clock time by adding
1272 * system time (updated by kvm_guest_time_update below) to the
1273 * wall clock specified here. guest system time equals host
1274 * system time for us, thus we must fill in host boot time here.
1276 getboottime64(&boot
);
1278 if (kvm
->arch
.kvmclock_offset
) {
1279 struct timespec64 ts
= ns_to_timespec64(kvm
->arch
.kvmclock_offset
);
1280 boot
= timespec64_sub(boot
, ts
);
1282 wc
.sec
= (u32
)boot
.tv_sec
; /* overflow in 2106 guest time */
1283 wc
.nsec
= boot
.tv_nsec
;
1284 wc
.version
= version
;
1286 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1289 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1292 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
1294 do_shl32_div32(dividend
, divisor
);
1298 static void kvm_get_time_scale(uint64_t scaled_hz
, uint64_t base_hz
,
1299 s8
*pshift
, u32
*pmultiplier
)
1307 scaled64
= scaled_hz
;
1308 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
1313 tps32
= (uint32_t)tps64
;
1314 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
1315 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
1323 *pmultiplier
= div_frac(scaled64
, tps32
);
1325 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1326 __func__
, base_hz
, scaled_hz
, shift
, *pmultiplier
);
1329 #ifdef CONFIG_X86_64
1330 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
1333 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
1334 static unsigned long max_tsc_khz
;
1336 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1338 u64 v
= (u64
)khz
* (1000000 + ppm
);
1343 static int set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
, bool scale
)
1347 /* Guest TSC same frequency as host TSC? */
1349 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1353 /* TSC scaling supported? */
1354 if (!kvm_has_tsc_control
) {
1355 if (user_tsc_khz
> tsc_khz
) {
1356 vcpu
->arch
.tsc_catchup
= 1;
1357 vcpu
->arch
.tsc_always_catchup
= 1;
1360 WARN(1, "user requested TSC rate below hardware speed\n");
1365 /* TSC scaling required - calculate ratio */
1366 ratio
= mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits
,
1367 user_tsc_khz
, tsc_khz
);
1369 if (ratio
== 0 || ratio
>= kvm_max_tsc_scaling_ratio
) {
1370 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1375 vcpu
->arch
.tsc_scaling_ratio
= ratio
;
1379 static int kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
)
1381 u32 thresh_lo
, thresh_hi
;
1382 int use_scaling
= 0;
1384 /* tsc_khz can be zero if TSC calibration fails */
1385 if (user_tsc_khz
== 0) {
1386 /* set tsc_scaling_ratio to a safe value */
1387 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1391 /* Compute a scale to convert nanoseconds in TSC cycles */
1392 kvm_get_time_scale(user_tsc_khz
* 1000LL, NSEC_PER_SEC
,
1393 &vcpu
->arch
.virtual_tsc_shift
,
1394 &vcpu
->arch
.virtual_tsc_mult
);
1395 vcpu
->arch
.virtual_tsc_khz
= user_tsc_khz
;
1398 * Compute the variation in TSC rate which is acceptable
1399 * within the range of tolerance and decide if the
1400 * rate being applied is within that bounds of the hardware
1401 * rate. If so, no scaling or compensation need be done.
1403 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1404 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1405 if (user_tsc_khz
< thresh_lo
|| user_tsc_khz
> thresh_hi
) {
1406 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz
, thresh_lo
, thresh_hi
);
1409 return set_tsc_khz(vcpu
, user_tsc_khz
, use_scaling
);
1412 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1414 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1415 vcpu
->arch
.virtual_tsc_mult
,
1416 vcpu
->arch
.virtual_tsc_shift
);
1417 tsc
+= vcpu
->arch
.this_tsc_write
;
1421 static void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
1423 #ifdef CONFIG_X86_64
1425 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1426 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1428 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1429 atomic_read(&vcpu
->kvm
->online_vcpus
));
1432 * Once the masterclock is enabled, always perform request in
1433 * order to update it.
1435 * In order to enable masterclock, the host clocksource must be TSC
1436 * and the vcpus need to have matched TSCs. When that happens,
1437 * perform request to enable masterclock.
1439 if (ka
->use_master_clock
||
1440 (gtod
->clock
.vclock_mode
== VCLOCK_TSC
&& vcpus_matched
))
1441 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1443 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
1444 atomic_read(&vcpu
->kvm
->online_vcpus
),
1445 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
1449 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu
*vcpu
, s64 offset
)
1451 u64 curr_offset
= vcpu
->arch
.tsc_offset
;
1452 vcpu
->arch
.ia32_tsc_adjust_msr
+= offset
- curr_offset
;
1456 * Multiply tsc by a fixed point number represented by ratio.
1458 * The most significant 64-N bits (mult) of ratio represent the
1459 * integral part of the fixed point number; the remaining N bits
1460 * (frac) represent the fractional part, ie. ratio represents a fixed
1461 * point number (mult + frac * 2^(-N)).
1463 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1465 static inline u64
__scale_tsc(u64 ratio
, u64 tsc
)
1467 return mul_u64_u64_shr(tsc
, ratio
, kvm_tsc_scaling_ratio_frac_bits
);
1470 u64
kvm_scale_tsc(struct kvm_vcpu
*vcpu
, u64 tsc
)
1473 u64 ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1475 if (ratio
!= kvm_default_tsc_scaling_ratio
)
1476 _tsc
= __scale_tsc(ratio
, tsc
);
1480 EXPORT_SYMBOL_GPL(kvm_scale_tsc
);
1482 static u64
kvm_compute_tsc_offset(struct kvm_vcpu
*vcpu
, u64 target_tsc
)
1486 tsc
= kvm_scale_tsc(vcpu
, rdtsc());
1488 return target_tsc
- tsc
;
1491 u64
kvm_read_l1_tsc(struct kvm_vcpu
*vcpu
, u64 host_tsc
)
1493 return vcpu
->arch
.tsc_offset
+ kvm_scale_tsc(vcpu
, host_tsc
);
1495 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc
);
1497 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1499 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1500 vcpu
->arch
.tsc_offset
= offset
;
1503 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1505 struct kvm
*kvm
= vcpu
->kvm
;
1506 u64 offset
, ns
, elapsed
;
1507 unsigned long flags
;
1509 bool already_matched
;
1510 u64 data
= msr
->data
;
1511 bool synchronizing
= false;
1513 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1514 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1515 ns
= ktime_get_boot_ns();
1516 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1518 if (vcpu
->arch
.virtual_tsc_khz
) {
1519 if (data
== 0 && msr
->host_initiated
) {
1521 * detection of vcpu initialization -- need to sync
1522 * with other vCPUs. This particularly helps to keep
1523 * kvm_clock stable after CPU hotplug
1525 synchronizing
= true;
1527 u64 tsc_exp
= kvm
->arch
.last_tsc_write
+
1528 nsec_to_cycles(vcpu
, elapsed
);
1529 u64 tsc_hz
= vcpu
->arch
.virtual_tsc_khz
* 1000LL;
1531 * Special case: TSC write with a small delta (1 second)
1532 * of virtual cycle time against real time is
1533 * interpreted as an attempt to synchronize the CPU.
1535 synchronizing
= data
< tsc_exp
+ tsc_hz
&&
1536 data
+ tsc_hz
> tsc_exp
;
1541 * For a reliable TSC, we can match TSC offsets, and for an unstable
1542 * TSC, we add elapsed time in this computation. We could let the
1543 * compensation code attempt to catch up if we fall behind, but
1544 * it's better to try to match offsets from the beginning.
1546 if (synchronizing
&&
1547 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1548 if (!check_tsc_unstable()) {
1549 offset
= kvm
->arch
.cur_tsc_offset
;
1550 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1552 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1554 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1555 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1558 already_matched
= (vcpu
->arch
.this_tsc_generation
== kvm
->arch
.cur_tsc_generation
);
1561 * We split periods of matched TSC writes into generations.
1562 * For each generation, we track the original measured
1563 * nanosecond time, offset, and write, so if TSCs are in
1564 * sync, we can match exact offset, and if not, we can match
1565 * exact software computation in compute_guest_tsc()
1567 * These values are tracked in kvm->arch.cur_xxx variables.
1569 kvm
->arch
.cur_tsc_generation
++;
1570 kvm
->arch
.cur_tsc_nsec
= ns
;
1571 kvm
->arch
.cur_tsc_write
= data
;
1572 kvm
->arch
.cur_tsc_offset
= offset
;
1574 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1575 kvm
->arch
.cur_tsc_generation
, data
);
1579 * We also track th most recent recorded KHZ, write and time to
1580 * allow the matching interval to be extended at each write.
1582 kvm
->arch
.last_tsc_nsec
= ns
;
1583 kvm
->arch
.last_tsc_write
= data
;
1584 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1586 vcpu
->arch
.last_guest_tsc
= data
;
1588 /* Keep track of which generation this VCPU has synchronized to */
1589 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1590 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1591 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1593 if (!msr
->host_initiated
&& guest_cpuid_has(vcpu
, X86_FEATURE_TSC_ADJUST
))
1594 update_ia32_tsc_adjust_msr(vcpu
, offset
);
1596 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
1597 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1599 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1601 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
1602 } else if (!already_matched
) {
1603 kvm
->arch
.nr_vcpus_matched_tsc
++;
1606 kvm_track_tsc_matching(vcpu
);
1607 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1610 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1612 static inline void adjust_tsc_offset_guest(struct kvm_vcpu
*vcpu
,
1615 kvm_vcpu_write_tsc_offset(vcpu
, vcpu
->arch
.tsc_offset
+ adjustment
);
1618 static inline void adjust_tsc_offset_host(struct kvm_vcpu
*vcpu
, s64 adjustment
)
1620 if (vcpu
->arch
.tsc_scaling_ratio
!= kvm_default_tsc_scaling_ratio
)
1621 WARN_ON(adjustment
< 0);
1622 adjustment
= kvm_scale_tsc(vcpu
, (u64
) adjustment
);
1623 adjust_tsc_offset_guest(vcpu
, adjustment
);
1626 #ifdef CONFIG_X86_64
1628 static u64
read_tsc(void)
1630 u64 ret
= (u64
)rdtsc_ordered();
1631 u64 last
= pvclock_gtod_data
.clock
.cycle_last
;
1633 if (likely(ret
>= last
))
1637 * GCC likes to generate cmov here, but this branch is extremely
1638 * predictable (it's just a function of time and the likely is
1639 * very likely) and there's a data dependence, so force GCC
1640 * to generate a branch instead. I don't barrier() because
1641 * we don't actually need a barrier, and if this function
1642 * ever gets inlined it will generate worse code.
1648 static inline u64
vgettsc(u64
*cycle_now
)
1651 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1653 *cycle_now
= read_tsc();
1655 v
= (*cycle_now
- gtod
->clock
.cycle_last
) & gtod
->clock
.mask
;
1656 return v
* gtod
->clock
.mult
;
1659 static int do_monotonic_boot(s64
*t
, u64
*cycle_now
)
1661 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1667 seq
= read_seqcount_begin(>od
->seq
);
1668 mode
= gtod
->clock
.vclock_mode
;
1669 ns
= gtod
->nsec_base
;
1670 ns
+= vgettsc(cycle_now
);
1671 ns
>>= gtod
->clock
.shift
;
1672 ns
+= gtod
->boot_ns
;
1673 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1679 static int do_realtime(struct timespec
*ts
, u64
*cycle_now
)
1681 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1687 seq
= read_seqcount_begin(>od
->seq
);
1688 mode
= gtod
->clock
.vclock_mode
;
1689 ts
->tv_sec
= gtod
->wall_time_sec
;
1690 ns
= gtod
->nsec_base
;
1691 ns
+= vgettsc(cycle_now
);
1692 ns
>>= gtod
->clock
.shift
;
1693 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1695 ts
->tv_sec
+= __iter_div_u64_rem(ns
, NSEC_PER_SEC
, &ns
);
1701 /* returns true if host is using tsc clocksource */
1702 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, u64
*cycle_now
)
1704 /* checked again under seqlock below */
1705 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1708 return do_monotonic_boot(kernel_ns
, cycle_now
) == VCLOCK_TSC
;
1711 /* returns true if host is using tsc clocksource */
1712 static bool kvm_get_walltime_and_clockread(struct timespec
*ts
,
1715 /* checked again under seqlock below */
1716 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1719 return do_realtime(ts
, cycle_now
) == VCLOCK_TSC
;
1725 * Assuming a stable TSC across physical CPUS, and a stable TSC
1726 * across virtual CPUs, the following condition is possible.
1727 * Each numbered line represents an event visible to both
1728 * CPUs at the next numbered event.
1730 * "timespecX" represents host monotonic time. "tscX" represents
1733 * VCPU0 on CPU0 | VCPU1 on CPU1
1735 * 1. read timespec0,tsc0
1736 * 2. | timespec1 = timespec0 + N
1738 * 3. transition to guest | transition to guest
1739 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1740 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1741 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1743 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1746 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1748 * - 0 < N - M => M < N
1750 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1751 * always the case (the difference between two distinct xtime instances
1752 * might be smaller then the difference between corresponding TSC reads,
1753 * when updating guest vcpus pvclock areas).
1755 * To avoid that problem, do not allow visibility of distinct
1756 * system_timestamp/tsc_timestamp values simultaneously: use a master
1757 * copy of host monotonic time values. Update that master copy
1760 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1764 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
1766 #ifdef CONFIG_X86_64
1767 struct kvm_arch
*ka
= &kvm
->arch
;
1769 bool host_tsc_clocksource
, vcpus_matched
;
1771 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1772 atomic_read(&kvm
->online_vcpus
));
1775 * If the host uses TSC clock, then passthrough TSC as stable
1778 host_tsc_clocksource
= kvm_get_time_and_clockread(
1779 &ka
->master_kernel_ns
,
1780 &ka
->master_cycle_now
);
1782 ka
->use_master_clock
= host_tsc_clocksource
&& vcpus_matched
1783 && !ka
->backwards_tsc_observed
1784 && !ka
->boot_vcpu_runs_old_kvmclock
;
1786 if (ka
->use_master_clock
)
1787 atomic_set(&kvm_guest_has_master_clock
, 1);
1789 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
1790 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
1795 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
1797 kvm_make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
1800 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
1802 #ifdef CONFIG_X86_64
1804 struct kvm_vcpu
*vcpu
;
1805 struct kvm_arch
*ka
= &kvm
->arch
;
1807 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1808 kvm_make_mclock_inprogress_request(kvm
);
1809 /* no guest entries from this point */
1810 pvclock_update_vm_gtod_copy(kvm
);
1812 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1813 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1815 /* guest entries allowed */
1816 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1817 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS
, vcpu
);
1819 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1823 u64
get_kvmclock_ns(struct kvm
*kvm
)
1825 struct kvm_arch
*ka
= &kvm
->arch
;
1826 struct pvclock_vcpu_time_info hv_clock
;
1829 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1830 if (!ka
->use_master_clock
) {
1831 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1832 return ktime_get_boot_ns() + ka
->kvmclock_offset
;
1835 hv_clock
.tsc_timestamp
= ka
->master_cycle_now
;
1836 hv_clock
.system_time
= ka
->master_kernel_ns
+ ka
->kvmclock_offset
;
1837 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1839 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1842 if (__this_cpu_read(cpu_tsc_khz
)) {
1843 kvm_get_time_scale(NSEC_PER_SEC
, __this_cpu_read(cpu_tsc_khz
) * 1000LL,
1844 &hv_clock
.tsc_shift
,
1845 &hv_clock
.tsc_to_system_mul
);
1846 ret
= __pvclock_read_cycles(&hv_clock
, rdtsc());
1848 ret
= ktime_get_boot_ns() + ka
->kvmclock_offset
;
1855 static void kvm_setup_pvclock_page(struct kvm_vcpu
*v
)
1857 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1858 struct pvclock_vcpu_time_info guest_hv_clock
;
1860 if (unlikely(kvm_read_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1861 &guest_hv_clock
, sizeof(guest_hv_clock
))))
1864 /* This VCPU is paused, but it's legal for a guest to read another
1865 * VCPU's kvmclock, so we really have to follow the specification where
1866 * it says that version is odd if data is being modified, and even after
1869 * Version field updates must be kept separate. This is because
1870 * kvm_write_guest_cached might use a "rep movs" instruction, and
1871 * writes within a string instruction are weakly ordered. So there
1872 * are three writes overall.
1874 * As a small optimization, only write the version field in the first
1875 * and third write. The vcpu->pv_time cache is still valid, because the
1876 * version field is the first in the struct.
1878 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info
, version
) != 0);
1880 if (guest_hv_clock
.version
& 1)
1881 ++guest_hv_clock
.version
; /* first time write, random junk */
1883 vcpu
->hv_clock
.version
= guest_hv_clock
.version
+ 1;
1884 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1886 sizeof(vcpu
->hv_clock
.version
));
1890 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1891 vcpu
->hv_clock
.flags
|= (guest_hv_clock
.flags
& PVCLOCK_GUEST_STOPPED
);
1893 if (vcpu
->pvclock_set_guest_stopped_request
) {
1894 vcpu
->hv_clock
.flags
|= PVCLOCK_GUEST_STOPPED
;
1895 vcpu
->pvclock_set_guest_stopped_request
= false;
1898 trace_kvm_pvclock_update(v
->vcpu_id
, &vcpu
->hv_clock
);
1900 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1902 sizeof(vcpu
->hv_clock
));
1906 vcpu
->hv_clock
.version
++;
1907 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1909 sizeof(vcpu
->hv_clock
.version
));
1912 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1914 unsigned long flags
, tgt_tsc_khz
;
1915 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1916 struct kvm_arch
*ka
= &v
->kvm
->arch
;
1918 u64 tsc_timestamp
, host_tsc
;
1920 bool use_master_clock
;
1926 * If the host uses TSC clock, then passthrough TSC as stable
1929 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1930 use_master_clock
= ka
->use_master_clock
;
1931 if (use_master_clock
) {
1932 host_tsc
= ka
->master_cycle_now
;
1933 kernel_ns
= ka
->master_kernel_ns
;
1935 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1937 /* Keep irq disabled to prevent changes to the clock */
1938 local_irq_save(flags
);
1939 tgt_tsc_khz
= __this_cpu_read(cpu_tsc_khz
);
1940 if (unlikely(tgt_tsc_khz
== 0)) {
1941 local_irq_restore(flags
);
1942 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1945 if (!use_master_clock
) {
1947 kernel_ns
= ktime_get_boot_ns();
1950 tsc_timestamp
= kvm_read_l1_tsc(v
, host_tsc
);
1953 * We may have to catch up the TSC to match elapsed wall clock
1954 * time for two reasons, even if kvmclock is used.
1955 * 1) CPU could have been running below the maximum TSC rate
1956 * 2) Broken TSC compensation resets the base at each VCPU
1957 * entry to avoid unknown leaps of TSC even when running
1958 * again on the same CPU. This may cause apparent elapsed
1959 * time to disappear, and the guest to stand still or run
1962 if (vcpu
->tsc_catchup
) {
1963 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1964 if (tsc
> tsc_timestamp
) {
1965 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1966 tsc_timestamp
= tsc
;
1970 local_irq_restore(flags
);
1972 /* With all the info we got, fill in the values */
1974 if (kvm_has_tsc_control
)
1975 tgt_tsc_khz
= kvm_scale_tsc(v
, tgt_tsc_khz
);
1977 if (unlikely(vcpu
->hw_tsc_khz
!= tgt_tsc_khz
)) {
1978 kvm_get_time_scale(NSEC_PER_SEC
, tgt_tsc_khz
* 1000LL,
1979 &vcpu
->hv_clock
.tsc_shift
,
1980 &vcpu
->hv_clock
.tsc_to_system_mul
);
1981 vcpu
->hw_tsc_khz
= tgt_tsc_khz
;
1984 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1985 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1986 vcpu
->last_guest_tsc
= tsc_timestamp
;
1988 /* If the host uses TSC clocksource, then it is stable */
1990 if (use_master_clock
)
1991 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
1993 vcpu
->hv_clock
.flags
= pvclock_flags
;
1995 if (vcpu
->pv_time_enabled
)
1996 kvm_setup_pvclock_page(v
);
1997 if (v
== kvm_get_vcpu(v
->kvm
, 0))
1998 kvm_hv_setup_tsc_page(v
->kvm
, &vcpu
->hv_clock
);
2003 * kvmclock updates which are isolated to a given vcpu, such as
2004 * vcpu->cpu migration, should not allow system_timestamp from
2005 * the rest of the vcpus to remain static. Otherwise ntp frequency
2006 * correction applies to one vcpu's system_timestamp but not
2009 * So in those cases, request a kvmclock update for all vcpus.
2010 * We need to rate-limit these requests though, as they can
2011 * considerably slow guests that have a large number of vcpus.
2012 * The time for a remote vcpu to update its kvmclock is bound
2013 * by the delay we use to rate-limit the updates.
2016 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2018 static void kvmclock_update_fn(struct work_struct
*work
)
2021 struct delayed_work
*dwork
= to_delayed_work(work
);
2022 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2023 kvmclock_update_work
);
2024 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2025 struct kvm_vcpu
*vcpu
;
2027 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2028 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2029 kvm_vcpu_kick(vcpu
);
2033 static void kvm_gen_kvmclock_update(struct kvm_vcpu
*v
)
2035 struct kvm
*kvm
= v
->kvm
;
2037 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
2038 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
,
2039 KVMCLOCK_UPDATE_DELAY
);
2042 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2044 static void kvmclock_sync_fn(struct work_struct
*work
)
2046 struct delayed_work
*dwork
= to_delayed_work(work
);
2047 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2048 kvmclock_sync_work
);
2049 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2051 if (!kvmclock_periodic_sync
)
2054 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
, 0);
2055 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
2056 KVMCLOCK_SYNC_PERIOD
);
2059 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
2061 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2062 unsigned bank_num
= mcg_cap
& 0xff;
2065 case MSR_IA32_MCG_STATUS
:
2066 vcpu
->arch
.mcg_status
= data
;
2068 case MSR_IA32_MCG_CTL
:
2069 if (!(mcg_cap
& MCG_CTL_P
))
2071 if (data
!= 0 && data
!= ~(u64
)0)
2073 vcpu
->arch
.mcg_ctl
= data
;
2076 if (msr
>= MSR_IA32_MC0_CTL
&&
2077 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2078 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2079 /* only 0 or all 1s can be written to IA32_MCi_CTL
2080 * some Linux kernels though clear bit 10 in bank 4 to
2081 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2082 * this to avoid an uncatched #GP in the guest
2084 if ((offset
& 0x3) == 0 &&
2085 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
2087 vcpu
->arch
.mce_banks
[offset
] = data
;
2095 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
2097 struct kvm
*kvm
= vcpu
->kvm
;
2098 int lm
= is_long_mode(vcpu
);
2099 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
2100 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
2101 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
2102 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
2103 u32 page_num
= data
& ~PAGE_MASK
;
2104 u64 page_addr
= data
& PAGE_MASK
;
2109 if (page_num
>= blob_size
)
2112 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
2117 if (kvm_vcpu_write_guest(vcpu
, page_addr
, page
, PAGE_SIZE
))
2126 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
2128 gpa_t gpa
= data
& ~0x3f;
2130 /* Bits 3:5 are reserved, Should be zero */
2134 vcpu
->arch
.apf
.msr_val
= data
;
2136 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
2137 kvm_clear_async_pf_completion_queue(vcpu
);
2138 kvm_async_pf_hash_reset(vcpu
);
2142 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
,
2146 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
2147 vcpu
->arch
.apf
.delivery_as_pf_vmexit
= data
& KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT
;
2148 kvm_async_pf_wakeup_all(vcpu
);
2152 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
2154 vcpu
->arch
.pv_time_enabled
= false;
2157 static void record_steal_time(struct kvm_vcpu
*vcpu
)
2159 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
2162 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2163 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
2166 vcpu
->arch
.st
.steal
.preempted
= 0;
2168 if (vcpu
->arch
.st
.steal
.version
& 1)
2169 vcpu
->arch
.st
.steal
.version
+= 1; /* first time write, random junk */
2171 vcpu
->arch
.st
.steal
.version
+= 1;
2173 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2174 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2178 vcpu
->arch
.st
.steal
.steal
+= current
->sched_info
.run_delay
-
2179 vcpu
->arch
.st
.last_steal
;
2180 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
2182 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2183 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2187 vcpu
->arch
.st
.steal
.version
+= 1;
2189 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2190 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2193 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2196 u32 msr
= msr_info
->index
;
2197 u64 data
= msr_info
->data
;
2200 case MSR_AMD64_NB_CFG
:
2201 case MSR_IA32_UCODE_WRITE
:
2202 case MSR_VM_HSAVE_PA
:
2203 case MSR_AMD64_PATCH_LOADER
:
2204 case MSR_AMD64_BU_CFG2
:
2205 case MSR_AMD64_DC_CFG
:
2208 case MSR_IA32_UCODE_REV
:
2209 if (msr_info
->host_initiated
)
2210 vcpu
->arch
.microcode_version
= data
;
2213 return set_efer(vcpu
, data
);
2215 data
&= ~(u64
)0x40; /* ignore flush filter disable */
2216 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
2217 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
2218 data
&= ~(u64
)0x40000; /* ignore Mc status write enable */
2220 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
2225 case MSR_FAM10H_MMIO_CONF_BASE
:
2227 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
2232 case MSR_IA32_DEBUGCTLMSR
:
2234 /* We support the non-activated case already */
2236 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
2237 /* Values other than LBR and BTF are vendor-specific,
2238 thus reserved and should throw a #GP */
2241 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2244 case 0x200 ... 0x2ff:
2245 return kvm_mtrr_set_msr(vcpu
, msr
, data
);
2246 case MSR_IA32_APICBASE
:
2247 return kvm_set_apic_base(vcpu
, msr_info
);
2248 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2249 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
2250 case MSR_IA32_TSCDEADLINE
:
2251 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
2253 case MSR_IA32_TSC_ADJUST
:
2254 if (guest_cpuid_has(vcpu
, X86_FEATURE_TSC_ADJUST
)) {
2255 if (!msr_info
->host_initiated
) {
2256 s64 adj
= data
- vcpu
->arch
.ia32_tsc_adjust_msr
;
2257 adjust_tsc_offset_guest(vcpu
, adj
);
2259 vcpu
->arch
.ia32_tsc_adjust_msr
= data
;
2262 case MSR_IA32_MISC_ENABLE
:
2263 vcpu
->arch
.ia32_misc_enable_msr
= data
;
2265 case MSR_IA32_SMBASE
:
2266 if (!msr_info
->host_initiated
)
2268 vcpu
->arch
.smbase
= data
;
2270 case MSR_KVM_WALL_CLOCK_NEW
:
2271 case MSR_KVM_WALL_CLOCK
:
2272 vcpu
->kvm
->arch
.wall_clock
= data
;
2273 kvm_write_wall_clock(vcpu
->kvm
, data
);
2275 case MSR_KVM_SYSTEM_TIME_NEW
:
2276 case MSR_KVM_SYSTEM_TIME
: {
2277 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
2279 kvmclock_reset(vcpu
);
2281 if (vcpu
->vcpu_id
== 0 && !msr_info
->host_initiated
) {
2282 bool tmp
= (msr
== MSR_KVM_SYSTEM_TIME
);
2284 if (ka
->boot_vcpu_runs_old_kvmclock
!= tmp
)
2285 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
2287 ka
->boot_vcpu_runs_old_kvmclock
= tmp
;
2290 vcpu
->arch
.time
= data
;
2291 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2293 /* we verify if the enable bit is set... */
2297 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2298 &vcpu
->arch
.pv_time
, data
& ~1ULL,
2299 sizeof(struct pvclock_vcpu_time_info
)))
2300 vcpu
->arch
.pv_time_enabled
= false;
2302 vcpu
->arch
.pv_time_enabled
= true;
2306 case MSR_KVM_ASYNC_PF_EN
:
2307 if (kvm_pv_enable_async_pf(vcpu
, data
))
2310 case MSR_KVM_STEAL_TIME
:
2312 if (unlikely(!sched_info_on()))
2315 if (data
& KVM_STEAL_RESERVED_MASK
)
2318 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2319 data
& KVM_STEAL_VALID_BITS
,
2320 sizeof(struct kvm_steal_time
)))
2323 vcpu
->arch
.st
.msr_val
= data
;
2325 if (!(data
& KVM_MSR_ENABLED
))
2328 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2331 case MSR_KVM_PV_EOI_EN
:
2332 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
2336 case MSR_IA32_MCG_CTL
:
2337 case MSR_IA32_MCG_STATUS
:
2338 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2339 return set_msr_mce(vcpu
, msr
, data
);
2341 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2342 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2343 pr
= true; /* fall through */
2344 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2345 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2346 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2347 return kvm_pmu_set_msr(vcpu
, msr_info
);
2349 if (pr
|| data
!= 0)
2350 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
2351 "0x%x data 0x%llx\n", msr
, data
);
2353 case MSR_K7_CLK_CTL
:
2355 * Ignore all writes to this no longer documented MSR.
2356 * Writes are only relevant for old K7 processors,
2357 * all pre-dating SVM, but a recommended workaround from
2358 * AMD for these chips. It is possible to specify the
2359 * affected processor models on the command line, hence
2360 * the need to ignore the workaround.
2363 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2364 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2365 case HV_X64_MSR_CRASH_CTL
:
2366 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2367 return kvm_hv_set_msr_common(vcpu
, msr
, data
,
2368 msr_info
->host_initiated
);
2369 case MSR_IA32_BBL_CR_CTL3
:
2370 /* Drop writes to this legacy MSR -- see rdmsr
2371 * counterpart for further detail.
2373 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data 0x%llx\n", msr
, data
);
2375 case MSR_AMD64_OSVW_ID_LENGTH
:
2376 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2378 vcpu
->arch
.osvw
.length
= data
;
2380 case MSR_AMD64_OSVW_STATUS
:
2381 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2383 vcpu
->arch
.osvw
.status
= data
;
2385 case MSR_PLATFORM_INFO
:
2386 if (!msr_info
->host_initiated
||
2387 data
& ~MSR_PLATFORM_INFO_CPUID_FAULT
||
2388 (!(data
& MSR_PLATFORM_INFO_CPUID_FAULT
) &&
2389 cpuid_fault_enabled(vcpu
)))
2391 vcpu
->arch
.msr_platform_info
= data
;
2393 case MSR_MISC_FEATURES_ENABLES
:
2394 if (data
& ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
||
2395 (data
& MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
&&
2396 !supports_cpuid_fault(vcpu
)))
2398 vcpu
->arch
.msr_misc_features_enables
= data
;
2401 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
2402 return xen_hvm_config(vcpu
, data
);
2403 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2404 return kvm_pmu_set_msr(vcpu
, msr_info
);
2406 vcpu_debug_ratelimited(vcpu
, "unhandled wrmsr: 0x%x data 0x%llx\n",
2410 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data 0x%llx\n",
2417 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
2421 * Reads an msr value (of 'msr_index') into 'pdata'.
2422 * Returns 0 on success, non-0 otherwise.
2423 * Assumes vcpu_load() was already called.
2425 int kvm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
2427 return kvm_x86_ops
->get_msr(vcpu
, msr
);
2429 EXPORT_SYMBOL_GPL(kvm_get_msr
);
2431 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2434 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2435 unsigned bank_num
= mcg_cap
& 0xff;
2438 case MSR_IA32_P5_MC_ADDR
:
2439 case MSR_IA32_P5_MC_TYPE
:
2442 case MSR_IA32_MCG_CAP
:
2443 data
= vcpu
->arch
.mcg_cap
;
2445 case MSR_IA32_MCG_CTL
:
2446 if (!(mcg_cap
& MCG_CTL_P
))
2448 data
= vcpu
->arch
.mcg_ctl
;
2450 case MSR_IA32_MCG_STATUS
:
2451 data
= vcpu
->arch
.mcg_status
;
2454 if (msr
>= MSR_IA32_MC0_CTL
&&
2455 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2456 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2457 data
= vcpu
->arch
.mce_banks
[offset
];
2466 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2468 switch (msr_info
->index
) {
2469 case MSR_IA32_PLATFORM_ID
:
2470 case MSR_IA32_EBL_CR_POWERON
:
2471 case MSR_IA32_DEBUGCTLMSR
:
2472 case MSR_IA32_LASTBRANCHFROMIP
:
2473 case MSR_IA32_LASTBRANCHTOIP
:
2474 case MSR_IA32_LASTINTFROMIP
:
2475 case MSR_IA32_LASTINTTOIP
:
2477 case MSR_K8_TSEG_ADDR
:
2478 case MSR_K8_TSEG_MASK
:
2480 case MSR_VM_HSAVE_PA
:
2481 case MSR_K8_INT_PENDING_MSG
:
2482 case MSR_AMD64_NB_CFG
:
2483 case MSR_FAM10H_MMIO_CONF_BASE
:
2484 case MSR_AMD64_BU_CFG2
:
2485 case MSR_IA32_PERF_CTL
:
2486 case MSR_AMD64_DC_CFG
:
2489 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2490 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2491 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2492 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2493 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2494 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2497 case MSR_IA32_UCODE_REV
:
2498 msr_info
->data
= vcpu
->arch
.microcode_version
;
2501 case 0x200 ... 0x2ff:
2502 return kvm_mtrr_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2503 case 0xcd: /* fsb frequency */
2507 * MSR_EBC_FREQUENCY_ID
2508 * Conservative value valid for even the basic CPU models.
2509 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2510 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2511 * and 266MHz for model 3, or 4. Set Core Clock
2512 * Frequency to System Bus Frequency Ratio to 1 (bits
2513 * 31:24) even though these are only valid for CPU
2514 * models > 2, however guests may end up dividing or
2515 * multiplying by zero otherwise.
2517 case MSR_EBC_FREQUENCY_ID
:
2518 msr_info
->data
= 1 << 24;
2520 case MSR_IA32_APICBASE
:
2521 msr_info
->data
= kvm_get_apic_base(vcpu
);
2523 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2524 return kvm_x2apic_msr_read(vcpu
, msr_info
->index
, &msr_info
->data
);
2526 case MSR_IA32_TSCDEADLINE
:
2527 msr_info
->data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
2529 case MSR_IA32_TSC_ADJUST
:
2530 msr_info
->data
= (u64
)vcpu
->arch
.ia32_tsc_adjust_msr
;
2532 case MSR_IA32_MISC_ENABLE
:
2533 msr_info
->data
= vcpu
->arch
.ia32_misc_enable_msr
;
2535 case MSR_IA32_SMBASE
:
2536 if (!msr_info
->host_initiated
)
2538 msr_info
->data
= vcpu
->arch
.smbase
;
2540 case MSR_IA32_PERF_STATUS
:
2541 /* TSC increment by tick */
2542 msr_info
->data
= 1000ULL;
2543 /* CPU multiplier */
2544 msr_info
->data
|= (((uint64_t)4ULL) << 40);
2547 msr_info
->data
= vcpu
->arch
.efer
;
2549 case MSR_KVM_WALL_CLOCK
:
2550 case MSR_KVM_WALL_CLOCK_NEW
:
2551 msr_info
->data
= vcpu
->kvm
->arch
.wall_clock
;
2553 case MSR_KVM_SYSTEM_TIME
:
2554 case MSR_KVM_SYSTEM_TIME_NEW
:
2555 msr_info
->data
= vcpu
->arch
.time
;
2557 case MSR_KVM_ASYNC_PF_EN
:
2558 msr_info
->data
= vcpu
->arch
.apf
.msr_val
;
2560 case MSR_KVM_STEAL_TIME
:
2561 msr_info
->data
= vcpu
->arch
.st
.msr_val
;
2563 case MSR_KVM_PV_EOI_EN
:
2564 msr_info
->data
= vcpu
->arch
.pv_eoi
.msr_val
;
2566 case MSR_IA32_P5_MC_ADDR
:
2567 case MSR_IA32_P5_MC_TYPE
:
2568 case MSR_IA32_MCG_CAP
:
2569 case MSR_IA32_MCG_CTL
:
2570 case MSR_IA32_MCG_STATUS
:
2571 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2572 return get_msr_mce(vcpu
, msr_info
->index
, &msr_info
->data
);
2573 case MSR_K7_CLK_CTL
:
2575 * Provide expected ramp-up count for K7. All other
2576 * are set to zero, indicating minimum divisors for
2579 * This prevents guest kernels on AMD host with CPU
2580 * type 6, model 8 and higher from exploding due to
2581 * the rdmsr failing.
2583 msr_info
->data
= 0x20000000;
2585 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2586 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2587 case HV_X64_MSR_CRASH_CTL
:
2588 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2589 return kvm_hv_get_msr_common(vcpu
,
2590 msr_info
->index
, &msr_info
->data
);
2592 case MSR_IA32_BBL_CR_CTL3
:
2593 /* This legacy MSR exists but isn't fully documented in current
2594 * silicon. It is however accessed by winxp in very narrow
2595 * scenarios where it sets bit #19, itself documented as
2596 * a "reserved" bit. Best effort attempt to source coherent
2597 * read data here should the balance of the register be
2598 * interpreted by the guest:
2600 * L2 cache control register 3: 64GB range, 256KB size,
2601 * enabled, latency 0x1, configured
2603 msr_info
->data
= 0xbe702111;
2605 case MSR_AMD64_OSVW_ID_LENGTH
:
2606 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2608 msr_info
->data
= vcpu
->arch
.osvw
.length
;
2610 case MSR_AMD64_OSVW_STATUS
:
2611 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2613 msr_info
->data
= vcpu
->arch
.osvw
.status
;
2615 case MSR_PLATFORM_INFO
:
2616 msr_info
->data
= vcpu
->arch
.msr_platform_info
;
2618 case MSR_MISC_FEATURES_ENABLES
:
2619 msr_info
->data
= vcpu
->arch
.msr_misc_features_enables
;
2622 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2623 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2625 vcpu_debug_ratelimited(vcpu
, "unhandled rdmsr: 0x%x\n",
2629 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr_info
->index
);
2636 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2639 * Read or write a bunch of msrs. All parameters are kernel addresses.
2641 * @return number of msrs set successfully.
2643 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2644 struct kvm_msr_entry
*entries
,
2645 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2646 unsigned index
, u64
*data
))
2650 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2651 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2658 * Read or write a bunch of msrs. Parameters are user addresses.
2660 * @return number of msrs set successfully.
2662 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2663 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2664 unsigned index
, u64
*data
),
2667 struct kvm_msrs msrs
;
2668 struct kvm_msr_entry
*entries
;
2673 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2677 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2680 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2681 entries
= memdup_user(user_msrs
->entries
, size
);
2682 if (IS_ERR(entries
)) {
2683 r
= PTR_ERR(entries
);
2687 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2692 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2703 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
2708 case KVM_CAP_IRQCHIP
:
2710 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2711 case KVM_CAP_SET_TSS_ADDR
:
2712 case KVM_CAP_EXT_CPUID
:
2713 case KVM_CAP_EXT_EMUL_CPUID
:
2714 case KVM_CAP_CLOCKSOURCE
:
2716 case KVM_CAP_NOP_IO_DELAY
:
2717 case KVM_CAP_MP_STATE
:
2718 case KVM_CAP_SYNC_MMU
:
2719 case KVM_CAP_USER_NMI
:
2720 case KVM_CAP_REINJECT_CONTROL
:
2721 case KVM_CAP_IRQ_INJECT_STATUS
:
2722 case KVM_CAP_IOEVENTFD
:
2723 case KVM_CAP_IOEVENTFD_NO_LENGTH
:
2725 case KVM_CAP_PIT_STATE2
:
2726 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2727 case KVM_CAP_XEN_HVM
:
2728 case KVM_CAP_VCPU_EVENTS
:
2729 case KVM_CAP_HYPERV
:
2730 case KVM_CAP_HYPERV_VAPIC
:
2731 case KVM_CAP_HYPERV_SPIN
:
2732 case KVM_CAP_HYPERV_SYNIC
:
2733 case KVM_CAP_HYPERV_SYNIC2
:
2734 case KVM_CAP_HYPERV_VP_INDEX
:
2735 case KVM_CAP_PCI_SEGMENT
:
2736 case KVM_CAP_DEBUGREGS
:
2737 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2739 case KVM_CAP_ASYNC_PF
:
2740 case KVM_CAP_GET_TSC_KHZ
:
2741 case KVM_CAP_KVMCLOCK_CTRL
:
2742 case KVM_CAP_READONLY_MEM
:
2743 case KVM_CAP_HYPERV_TIME
:
2744 case KVM_CAP_IOAPIC_POLARITY_IGNORED
:
2745 case KVM_CAP_TSC_DEADLINE_TIMER
:
2746 case KVM_CAP_ENABLE_CAP_VM
:
2747 case KVM_CAP_DISABLE_QUIRKS
:
2748 case KVM_CAP_SET_BOOT_CPU_ID
:
2749 case KVM_CAP_SPLIT_IRQCHIP
:
2750 case KVM_CAP_IMMEDIATE_EXIT
:
2751 case KVM_CAP_GET_MSR_FEATURES
:
2754 case KVM_CAP_ADJUST_CLOCK
:
2755 r
= KVM_CLOCK_TSC_STABLE
;
2757 case KVM_CAP_X86_GUEST_MWAIT
:
2758 r
= kvm_mwait_in_guest();
2760 case KVM_CAP_X86_SMM
:
2761 /* SMBASE is usually relocated above 1M on modern chipsets,
2762 * and SMM handlers might indeed rely on 4G segment limits,
2763 * so do not report SMM to be available if real mode is
2764 * emulated via vm86 mode. Still, do not go to great lengths
2765 * to avoid userspace's usage of the feature, because it is a
2766 * fringe case that is not enabled except via specific settings
2767 * of the module parameters.
2769 r
= kvm_x86_ops
->has_emulated_msr(MSR_IA32_SMBASE
);
2772 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2774 case KVM_CAP_NR_VCPUS
:
2775 r
= KVM_SOFT_MAX_VCPUS
;
2777 case KVM_CAP_MAX_VCPUS
:
2780 case KVM_CAP_NR_MEMSLOTS
:
2781 r
= KVM_USER_MEM_SLOTS
;
2783 case KVM_CAP_PV_MMU
: /* obsolete */
2787 r
= KVM_MAX_MCE_BANKS
;
2790 r
= boot_cpu_has(X86_FEATURE_XSAVE
);
2792 case KVM_CAP_TSC_CONTROL
:
2793 r
= kvm_has_tsc_control
;
2795 case KVM_CAP_X2APIC_API
:
2796 r
= KVM_X2APIC_API_VALID_FLAGS
;
2806 long kvm_arch_dev_ioctl(struct file
*filp
,
2807 unsigned int ioctl
, unsigned long arg
)
2809 void __user
*argp
= (void __user
*)arg
;
2813 case KVM_GET_MSR_INDEX_LIST
: {
2814 struct kvm_msr_list __user
*user_msr_list
= argp
;
2815 struct kvm_msr_list msr_list
;
2819 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2822 msr_list
.nmsrs
= num_msrs_to_save
+ num_emulated_msrs
;
2823 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2826 if (n
< msr_list
.nmsrs
)
2829 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2830 num_msrs_to_save
* sizeof(u32
)))
2832 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2834 num_emulated_msrs
* sizeof(u32
)))
2839 case KVM_GET_SUPPORTED_CPUID
:
2840 case KVM_GET_EMULATED_CPUID
: {
2841 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2842 struct kvm_cpuid2 cpuid
;
2845 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2848 r
= kvm_dev_ioctl_get_cpuid(&cpuid
, cpuid_arg
->entries
,
2854 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2859 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2861 if (copy_to_user(argp
, &kvm_mce_cap_supported
,
2862 sizeof(kvm_mce_cap_supported
)))
2866 case KVM_GET_MSR_FEATURE_INDEX_LIST
: {
2867 struct kvm_msr_list __user
*user_msr_list
= argp
;
2868 struct kvm_msr_list msr_list
;
2872 if (copy_from_user(&msr_list
, user_msr_list
, sizeof(msr_list
)))
2875 msr_list
.nmsrs
= num_msr_based_features
;
2876 if (copy_to_user(user_msr_list
, &msr_list
, sizeof(msr_list
)))
2879 if (n
< msr_list
.nmsrs
)
2882 if (copy_to_user(user_msr_list
->indices
, &msr_based_features
,
2883 num_msr_based_features
* sizeof(u32
)))
2889 r
= msr_io(NULL
, argp
, do_get_msr_feature
, 1);
2899 static void wbinvd_ipi(void *garbage
)
2904 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2906 return kvm_arch_has_noncoherent_dma(vcpu
->kvm
);
2909 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2911 /* Address WBINVD may be executed by guest */
2912 if (need_emulate_wbinvd(vcpu
)) {
2913 if (kvm_x86_ops
->has_wbinvd_exit())
2914 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2915 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2916 smp_call_function_single(vcpu
->cpu
,
2917 wbinvd_ipi
, NULL
, 1);
2920 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2922 /* Apply any externally detected TSC adjustments (due to suspend) */
2923 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2924 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2925 vcpu
->arch
.tsc_offset_adjustment
= 0;
2926 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2929 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2930 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2931 rdtsc() - vcpu
->arch
.last_host_tsc
;
2933 mark_tsc_unstable("KVM discovered backwards TSC");
2935 if (check_tsc_unstable()) {
2936 u64 offset
= kvm_compute_tsc_offset(vcpu
,
2937 vcpu
->arch
.last_guest_tsc
);
2938 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
2939 vcpu
->arch
.tsc_catchup
= 1;
2942 if (kvm_lapic_hv_timer_in_use(vcpu
))
2943 kvm_lapic_restart_hv_timer(vcpu
);
2946 * On a host with synchronized TSC, there is no need to update
2947 * kvmclock on vcpu->cpu migration
2949 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
2950 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2951 if (vcpu
->cpu
!= cpu
)
2952 kvm_make_request(KVM_REQ_MIGRATE_TIMER
, vcpu
);
2956 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2959 static void kvm_steal_time_set_preempted(struct kvm_vcpu
*vcpu
)
2961 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
2964 vcpu
->arch
.st
.steal
.preempted
= 1;
2966 kvm_write_guest_offset_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2967 &vcpu
->arch
.st
.steal
.preempted
,
2968 offsetof(struct kvm_steal_time
, preempted
),
2969 sizeof(vcpu
->arch
.st
.steal
.preempted
));
2972 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2976 if (vcpu
->preempted
)
2977 vcpu
->arch
.preempted_in_kernel
= !kvm_x86_ops
->get_cpl(vcpu
);
2980 * Disable page faults because we're in atomic context here.
2981 * kvm_write_guest_offset_cached() would call might_fault()
2982 * that relies on pagefault_disable() to tell if there's a
2983 * bug. NOTE: the write to guest memory may not go through if
2984 * during postcopy live migration or if there's heavy guest
2987 pagefault_disable();
2989 * kvm_memslots() will be called by
2990 * kvm_write_guest_offset_cached() so take the srcu lock.
2992 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2993 kvm_steal_time_set_preempted(vcpu
);
2994 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2996 kvm_x86_ops
->vcpu_put(vcpu
);
2997 kvm_put_guest_fpu(vcpu
);
2998 vcpu
->arch
.last_host_tsc
= rdtsc();
3000 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3001 * on every vmexit, but if not, we might have a stale dr6 from the
3002 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3007 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
3008 struct kvm_lapic_state
*s
)
3010 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
3011 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
3013 return kvm_apic_get_state(vcpu
, s
);
3016 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
3017 struct kvm_lapic_state
*s
)
3021 r
= kvm_apic_set_state(vcpu
, s
);
3024 update_cr8_intercept(vcpu
);
3029 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu
*vcpu
)
3031 return (!lapic_in_kernel(vcpu
) ||
3032 kvm_apic_accept_pic_intr(vcpu
));
3036 * if userspace requested an interrupt window, check that the
3037 * interrupt window is open.
3039 * No need to exit to userspace if we already have an interrupt queued.
3041 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu
*vcpu
)
3043 return kvm_arch_interrupt_allowed(vcpu
) &&
3044 !kvm_cpu_has_interrupt(vcpu
) &&
3045 !kvm_event_needs_reinjection(vcpu
) &&
3046 kvm_cpu_accept_dm_intr(vcpu
);
3049 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
3050 struct kvm_interrupt
*irq
)
3052 if (irq
->irq
>= KVM_NR_INTERRUPTS
)
3055 if (!irqchip_in_kernel(vcpu
->kvm
)) {
3056 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
3057 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3062 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3063 * fail for in-kernel 8259.
3065 if (pic_in_kernel(vcpu
->kvm
))
3068 if (vcpu
->arch
.pending_external_vector
!= -1)
3071 vcpu
->arch
.pending_external_vector
= irq
->irq
;
3072 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3076 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
3078 kvm_inject_nmi(vcpu
);
3083 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu
*vcpu
)
3085 kvm_make_request(KVM_REQ_SMI
, vcpu
);
3090 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
3091 struct kvm_tpr_access_ctl
*tac
)
3095 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
3099 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
3103 unsigned bank_num
= mcg_cap
& 0xff, bank
;
3106 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
3108 if (mcg_cap
& ~(kvm_mce_cap_supported
| 0xff | 0xff0000))
3111 vcpu
->arch
.mcg_cap
= mcg_cap
;
3112 /* Init IA32_MCG_CTL to all 1s */
3113 if (mcg_cap
& MCG_CTL_P
)
3114 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
3115 /* Init IA32_MCi_CTL to all 1s */
3116 for (bank
= 0; bank
< bank_num
; bank
++)
3117 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
3119 if (kvm_x86_ops
->setup_mce
)
3120 kvm_x86_ops
->setup_mce(vcpu
);
3125 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
3126 struct kvm_x86_mce
*mce
)
3128 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
3129 unsigned bank_num
= mcg_cap
& 0xff;
3130 u64
*banks
= vcpu
->arch
.mce_banks
;
3132 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
3135 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3136 * reporting is disabled
3138 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
3139 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
3141 banks
+= 4 * mce
->bank
;
3143 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3144 * reporting is disabled for the bank
3146 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
3148 if (mce
->status
& MCI_STATUS_UC
) {
3149 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
3150 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
3151 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
3154 if (banks
[1] & MCI_STATUS_VAL
)
3155 mce
->status
|= MCI_STATUS_OVER
;
3156 banks
[2] = mce
->addr
;
3157 banks
[3] = mce
->misc
;
3158 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
3159 banks
[1] = mce
->status
;
3160 kvm_queue_exception(vcpu
, MC_VECTOR
);
3161 } else if (!(banks
[1] & MCI_STATUS_VAL
)
3162 || !(banks
[1] & MCI_STATUS_UC
)) {
3163 if (banks
[1] & MCI_STATUS_VAL
)
3164 mce
->status
|= MCI_STATUS_OVER
;
3165 banks
[2] = mce
->addr
;
3166 banks
[3] = mce
->misc
;
3167 banks
[1] = mce
->status
;
3169 banks
[1] |= MCI_STATUS_OVER
;
3173 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
3174 struct kvm_vcpu_events
*events
)
3178 * FIXME: pass injected and pending separately. This is only
3179 * needed for nested virtualization, whose state cannot be
3180 * migrated yet. For now we can combine them.
3182 events
->exception
.injected
=
3183 (vcpu
->arch
.exception
.pending
||
3184 vcpu
->arch
.exception
.injected
) &&
3185 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
3186 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
3187 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
3188 events
->exception
.pad
= 0;
3189 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
3191 events
->interrupt
.injected
=
3192 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
3193 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
3194 events
->interrupt
.soft
= 0;
3195 events
->interrupt
.shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
3197 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
3198 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
3199 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
3200 events
->nmi
.pad
= 0;
3202 events
->sipi_vector
= 0; /* never valid when reporting to user space */
3204 events
->smi
.smm
= is_smm(vcpu
);
3205 events
->smi
.pending
= vcpu
->arch
.smi_pending
;
3206 events
->smi
.smm_inside_nmi
=
3207 !!(vcpu
->arch
.hflags
& HF_SMM_INSIDE_NMI_MASK
);
3208 events
->smi
.latched_init
= kvm_lapic_latched_init(vcpu
);
3210 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
3211 | KVM_VCPUEVENT_VALID_SHADOW
3212 | KVM_VCPUEVENT_VALID_SMM
);
3213 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
3216 static void kvm_set_hflags(struct kvm_vcpu
*vcpu
, unsigned emul_flags
);
3218 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
3219 struct kvm_vcpu_events
*events
)
3221 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3222 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3223 | KVM_VCPUEVENT_VALID_SHADOW
3224 | KVM_VCPUEVENT_VALID_SMM
))
3227 if (events
->exception
.injected
&&
3228 (events
->exception
.nr
> 31 || events
->exception
.nr
== NMI_VECTOR
||
3229 is_guest_mode(vcpu
)))
3232 /* INITs are latched while in SMM */
3233 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
&&
3234 (events
->smi
.smm
|| events
->smi
.pending
) &&
3235 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
)
3239 vcpu
->arch
.exception
.injected
= false;
3240 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
3241 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
3242 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
3243 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
3245 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
3246 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
3247 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
3248 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
3249 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
3250 events
->interrupt
.shadow
);
3252 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
3253 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
3254 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
3255 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
3257 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
&&
3258 lapic_in_kernel(vcpu
))
3259 vcpu
->arch
.apic
->sipi_vector
= events
->sipi_vector
;
3261 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
) {
3262 u32 hflags
= vcpu
->arch
.hflags
;
3263 if (events
->smi
.smm
)
3264 hflags
|= HF_SMM_MASK
;
3266 hflags
&= ~HF_SMM_MASK
;
3267 kvm_set_hflags(vcpu
, hflags
);
3269 vcpu
->arch
.smi_pending
= events
->smi
.pending
;
3271 if (events
->smi
.smm
) {
3272 if (events
->smi
.smm_inside_nmi
)
3273 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
3275 vcpu
->arch
.hflags
&= ~HF_SMM_INSIDE_NMI_MASK
;
3276 if (lapic_in_kernel(vcpu
)) {
3277 if (events
->smi
.latched_init
)
3278 set_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3280 clear_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3285 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3290 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
3291 struct kvm_debugregs
*dbgregs
)
3295 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
3296 kvm_get_dr(vcpu
, 6, &val
);
3298 dbgregs
->dr7
= vcpu
->arch
.dr7
;
3300 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
3303 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
3304 struct kvm_debugregs
*dbgregs
)
3309 if (dbgregs
->dr6
& ~0xffffffffull
)
3311 if (dbgregs
->dr7
& ~0xffffffffull
)
3314 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
3315 kvm_update_dr0123(vcpu
);
3316 vcpu
->arch
.dr6
= dbgregs
->dr6
;
3317 kvm_update_dr6(vcpu
);
3318 vcpu
->arch
.dr7
= dbgregs
->dr7
;
3319 kvm_update_dr7(vcpu
);
3324 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3326 static void fill_xsave(u8
*dest
, struct kvm_vcpu
*vcpu
)
3328 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3329 u64 xstate_bv
= xsave
->header
.xfeatures
;
3333 * Copy legacy XSAVE area, to avoid complications with CPUID
3334 * leaves 0 and 1 in the loop below.
3336 memcpy(dest
, xsave
, XSAVE_HDR_OFFSET
);
3339 xstate_bv
&= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FPSSE
;
3340 *(u64
*)(dest
+ XSAVE_HDR_OFFSET
) = xstate_bv
;
3343 * Copy each region from the possibly compacted offset to the
3344 * non-compacted offset.
3346 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3348 u64 feature
= valid
& -valid
;
3349 int index
= fls64(feature
) - 1;
3350 void *src
= get_xsave_addr(xsave
, feature
);
3353 u32 size
, offset
, ecx
, edx
;
3354 cpuid_count(XSTATE_CPUID
, index
,
3355 &size
, &offset
, &ecx
, &edx
);
3356 if (feature
== XFEATURE_MASK_PKRU
)
3357 memcpy(dest
+ offset
, &vcpu
->arch
.pkru
,
3358 sizeof(vcpu
->arch
.pkru
));
3360 memcpy(dest
+ offset
, src
, size
);
3368 static void load_xsave(struct kvm_vcpu
*vcpu
, u8
*src
)
3370 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3371 u64 xstate_bv
= *(u64
*)(src
+ XSAVE_HDR_OFFSET
);
3375 * Copy legacy XSAVE area, to avoid complications with CPUID
3376 * leaves 0 and 1 in the loop below.
3378 memcpy(xsave
, src
, XSAVE_HDR_OFFSET
);
3380 /* Set XSTATE_BV and possibly XCOMP_BV. */
3381 xsave
->header
.xfeatures
= xstate_bv
;
3382 if (boot_cpu_has(X86_FEATURE_XSAVES
))
3383 xsave
->header
.xcomp_bv
= host_xcr0
| XSTATE_COMPACTION_ENABLED
;
3386 * Copy each region from the non-compacted offset to the
3387 * possibly compacted offset.
3389 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3391 u64 feature
= valid
& -valid
;
3392 int index
= fls64(feature
) - 1;
3393 void *dest
= get_xsave_addr(xsave
, feature
);
3396 u32 size
, offset
, ecx
, edx
;
3397 cpuid_count(XSTATE_CPUID
, index
,
3398 &size
, &offset
, &ecx
, &edx
);
3399 if (feature
== XFEATURE_MASK_PKRU
)
3400 memcpy(&vcpu
->arch
.pkru
, src
+ offset
,
3401 sizeof(vcpu
->arch
.pkru
));
3403 memcpy(dest
, src
+ offset
, size
);
3410 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
3411 struct kvm_xsave
*guest_xsave
)
3413 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3414 memset(guest_xsave
, 0, sizeof(struct kvm_xsave
));
3415 fill_xsave((u8
*) guest_xsave
->region
, vcpu
);
3417 memcpy(guest_xsave
->region
,
3418 &vcpu
->arch
.guest_fpu
.state
.fxsave
,
3419 sizeof(struct fxregs_state
));
3420 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
3421 XFEATURE_MASK_FPSSE
;
3425 #define XSAVE_MXCSR_OFFSET 24
3427 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
3428 struct kvm_xsave
*guest_xsave
)
3431 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
3432 u32 mxcsr
= *(u32
*)&guest_xsave
->region
[XSAVE_MXCSR_OFFSET
/ sizeof(u32
)];
3434 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3436 * Here we allow setting states that are not present in
3437 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3438 * with old userspace.
3440 if (xstate_bv
& ~kvm_supported_xcr0() ||
3441 mxcsr
& ~mxcsr_feature_mask
)
3443 load_xsave(vcpu
, (u8
*)guest_xsave
->region
);
3445 if (xstate_bv
& ~XFEATURE_MASK_FPSSE
||
3446 mxcsr
& ~mxcsr_feature_mask
)
3448 memcpy(&vcpu
->arch
.guest_fpu
.state
.fxsave
,
3449 guest_xsave
->region
, sizeof(struct fxregs_state
));
3454 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
3455 struct kvm_xcrs
*guest_xcrs
)
3457 if (!boot_cpu_has(X86_FEATURE_XSAVE
)) {
3458 guest_xcrs
->nr_xcrs
= 0;
3462 guest_xcrs
->nr_xcrs
= 1;
3463 guest_xcrs
->flags
= 0;
3464 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
3465 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
3468 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
3469 struct kvm_xcrs
*guest_xcrs
)
3473 if (!boot_cpu_has(X86_FEATURE_XSAVE
))
3476 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
3479 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
3480 /* Only support XCR0 currently */
3481 if (guest_xcrs
->xcrs
[i
].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
3482 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
3483 guest_xcrs
->xcrs
[i
].value
);
3492 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3493 * stopped by the hypervisor. This function will be called from the host only.
3494 * EINVAL is returned when the host attempts to set the flag for a guest that
3495 * does not support pv clocks.
3497 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
3499 if (!vcpu
->arch
.pv_time_enabled
)
3501 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
3502 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
3506 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
3507 struct kvm_enable_cap
*cap
)
3513 case KVM_CAP_HYPERV_SYNIC2
:
3516 case KVM_CAP_HYPERV_SYNIC
:
3517 if (!irqchip_in_kernel(vcpu
->kvm
))
3519 return kvm_hv_activate_synic(vcpu
, cap
->cap
==
3520 KVM_CAP_HYPERV_SYNIC2
);
3526 long kvm_arch_vcpu_ioctl(struct file
*filp
,
3527 unsigned int ioctl
, unsigned long arg
)
3529 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3530 void __user
*argp
= (void __user
*)arg
;
3533 struct kvm_lapic_state
*lapic
;
3534 struct kvm_xsave
*xsave
;
3535 struct kvm_xcrs
*xcrs
;
3541 case KVM_GET_LAPIC
: {
3543 if (!lapic_in_kernel(vcpu
))
3545 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
3550 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
3554 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
3559 case KVM_SET_LAPIC
: {
3561 if (!lapic_in_kernel(vcpu
))
3563 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
3564 if (IS_ERR(u
.lapic
))
3565 return PTR_ERR(u
.lapic
);
3567 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
3570 case KVM_INTERRUPT
: {
3571 struct kvm_interrupt irq
;
3574 if (copy_from_user(&irq
, argp
, sizeof irq
))
3576 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
3580 r
= kvm_vcpu_ioctl_nmi(vcpu
);
3584 r
= kvm_vcpu_ioctl_smi(vcpu
);
3587 case KVM_SET_CPUID
: {
3588 struct kvm_cpuid __user
*cpuid_arg
= argp
;
3589 struct kvm_cpuid cpuid
;
3592 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3594 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3597 case KVM_SET_CPUID2
: {
3598 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3599 struct kvm_cpuid2 cpuid
;
3602 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3604 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
3605 cpuid_arg
->entries
);
3608 case KVM_GET_CPUID2
: {
3609 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3610 struct kvm_cpuid2 cpuid
;
3613 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3615 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
3616 cpuid_arg
->entries
);
3620 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
3625 case KVM_GET_MSRS
: {
3626 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3627 r
= msr_io(vcpu
, argp
, do_get_msr
, 1);
3628 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3631 case KVM_SET_MSRS
: {
3632 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3633 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
3634 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3637 case KVM_TPR_ACCESS_REPORTING
: {
3638 struct kvm_tpr_access_ctl tac
;
3641 if (copy_from_user(&tac
, argp
, sizeof tac
))
3643 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
3647 if (copy_to_user(argp
, &tac
, sizeof tac
))
3652 case KVM_SET_VAPIC_ADDR
: {
3653 struct kvm_vapic_addr va
;
3657 if (!lapic_in_kernel(vcpu
))
3660 if (copy_from_user(&va
, argp
, sizeof va
))
3662 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3663 r
= kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
3664 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3667 case KVM_X86_SETUP_MCE
: {
3671 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
3673 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
3676 case KVM_X86_SET_MCE
: {
3677 struct kvm_x86_mce mce
;
3680 if (copy_from_user(&mce
, argp
, sizeof mce
))
3682 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
3685 case KVM_GET_VCPU_EVENTS
: {
3686 struct kvm_vcpu_events events
;
3688 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
3691 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
3696 case KVM_SET_VCPU_EVENTS
: {
3697 struct kvm_vcpu_events events
;
3700 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
3703 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
3706 case KVM_GET_DEBUGREGS
: {
3707 struct kvm_debugregs dbgregs
;
3709 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
3712 if (copy_to_user(argp
, &dbgregs
,
3713 sizeof(struct kvm_debugregs
)))
3718 case KVM_SET_DEBUGREGS
: {
3719 struct kvm_debugregs dbgregs
;
3722 if (copy_from_user(&dbgregs
, argp
,
3723 sizeof(struct kvm_debugregs
)))
3726 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3729 case KVM_GET_XSAVE
: {
3730 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3735 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3738 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3743 case KVM_SET_XSAVE
: {
3744 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
3745 if (IS_ERR(u
.xsave
))
3746 return PTR_ERR(u
.xsave
);
3748 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3751 case KVM_GET_XCRS
: {
3752 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3757 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3760 if (copy_to_user(argp
, u
.xcrs
,
3761 sizeof(struct kvm_xcrs
)))
3766 case KVM_SET_XCRS
: {
3767 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
3769 return PTR_ERR(u
.xcrs
);
3771 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3774 case KVM_SET_TSC_KHZ
: {
3778 user_tsc_khz
= (u32
)arg
;
3780 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3783 if (user_tsc_khz
== 0)
3784 user_tsc_khz
= tsc_khz
;
3786 if (!kvm_set_tsc_khz(vcpu
, user_tsc_khz
))
3791 case KVM_GET_TSC_KHZ
: {
3792 r
= vcpu
->arch
.virtual_tsc_khz
;
3795 case KVM_KVMCLOCK_CTRL
: {
3796 r
= kvm_set_guest_paused(vcpu
);
3799 case KVM_ENABLE_CAP
: {
3800 struct kvm_enable_cap cap
;
3803 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3805 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
3816 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
3818 return VM_FAULT_SIGBUS
;
3821 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3825 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3827 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3831 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3834 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3838 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3839 u32 kvm_nr_mmu_pages
)
3841 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3844 mutex_lock(&kvm
->slots_lock
);
3846 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3847 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3849 mutex_unlock(&kvm
->slots_lock
);
3853 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3855 return kvm
->arch
.n_max_mmu_pages
;
3858 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3860 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
3864 switch (chip
->chip_id
) {
3865 case KVM_IRQCHIP_PIC_MASTER
:
3866 memcpy(&chip
->chip
.pic
, &pic
->pics
[0],
3867 sizeof(struct kvm_pic_state
));
3869 case KVM_IRQCHIP_PIC_SLAVE
:
3870 memcpy(&chip
->chip
.pic
, &pic
->pics
[1],
3871 sizeof(struct kvm_pic_state
));
3873 case KVM_IRQCHIP_IOAPIC
:
3874 kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3883 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3885 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
3889 switch (chip
->chip_id
) {
3890 case KVM_IRQCHIP_PIC_MASTER
:
3891 spin_lock(&pic
->lock
);
3892 memcpy(&pic
->pics
[0], &chip
->chip
.pic
,
3893 sizeof(struct kvm_pic_state
));
3894 spin_unlock(&pic
->lock
);
3896 case KVM_IRQCHIP_PIC_SLAVE
:
3897 spin_lock(&pic
->lock
);
3898 memcpy(&pic
->pics
[1], &chip
->chip
.pic
,
3899 sizeof(struct kvm_pic_state
));
3900 spin_unlock(&pic
->lock
);
3902 case KVM_IRQCHIP_IOAPIC
:
3903 kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3909 kvm_pic_update_irq(pic
);
3913 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3915 struct kvm_kpit_state
*kps
= &kvm
->arch
.vpit
->pit_state
;
3917 BUILD_BUG_ON(sizeof(*ps
) != sizeof(kps
->channels
));
3919 mutex_lock(&kps
->lock
);
3920 memcpy(ps
, &kps
->channels
, sizeof(*ps
));
3921 mutex_unlock(&kps
->lock
);
3925 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3928 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3930 mutex_lock(&pit
->pit_state
.lock
);
3931 memcpy(&pit
->pit_state
.channels
, ps
, sizeof(*ps
));
3932 for (i
= 0; i
< 3; i
++)
3933 kvm_pit_load_count(pit
, i
, ps
->channels
[i
].count
, 0);
3934 mutex_unlock(&pit
->pit_state
.lock
);
3938 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3940 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3941 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3942 sizeof(ps
->channels
));
3943 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3944 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3945 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3949 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3953 u32 prev_legacy
, cur_legacy
;
3954 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3956 mutex_lock(&pit
->pit_state
.lock
);
3957 prev_legacy
= pit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3958 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3959 if (!prev_legacy
&& cur_legacy
)
3961 memcpy(&pit
->pit_state
.channels
, &ps
->channels
,
3962 sizeof(pit
->pit_state
.channels
));
3963 pit
->pit_state
.flags
= ps
->flags
;
3964 for (i
= 0; i
< 3; i
++)
3965 kvm_pit_load_count(pit
, i
, pit
->pit_state
.channels
[i
].count
,
3967 mutex_unlock(&pit
->pit_state
.lock
);
3971 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3972 struct kvm_reinject_control
*control
)
3974 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3979 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3980 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3981 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3983 mutex_lock(&pit
->pit_state
.lock
);
3984 kvm_pit_set_reinject(pit
, control
->pit_reinject
);
3985 mutex_unlock(&pit
->pit_state
.lock
);
3991 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3992 * @kvm: kvm instance
3993 * @log: slot id and address to which we copy the log
3995 * Steps 1-4 below provide general overview of dirty page logging. See
3996 * kvm_get_dirty_log_protect() function description for additional details.
3998 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3999 * always flush the TLB (step 4) even if previous step failed and the dirty
4000 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4001 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4002 * writes will be marked dirty for next log read.
4004 * 1. Take a snapshot of the bit and clear it if needed.
4005 * 2. Write protect the corresponding page.
4006 * 3. Copy the snapshot to the userspace.
4007 * 4. Flush TLB's if needed.
4009 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
4011 bool is_dirty
= false;
4014 mutex_lock(&kvm
->slots_lock
);
4017 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4019 if (kvm_x86_ops
->flush_log_dirty
)
4020 kvm_x86_ops
->flush_log_dirty(kvm
);
4022 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
4025 * All the TLBs can be flushed out of mmu lock, see the comments in
4026 * kvm_mmu_slot_remove_write_access().
4028 lockdep_assert_held(&kvm
->slots_lock
);
4030 kvm_flush_remote_tlbs(kvm
);
4032 mutex_unlock(&kvm
->slots_lock
);
4036 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
4039 if (!irqchip_in_kernel(kvm
))
4042 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
4043 irq_event
->irq
, irq_event
->level
,
4048 static int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4049 struct kvm_enable_cap
*cap
)
4057 case KVM_CAP_DISABLE_QUIRKS
:
4058 kvm
->arch
.disabled_quirks
= cap
->args
[0];
4061 case KVM_CAP_SPLIT_IRQCHIP
: {
4062 mutex_lock(&kvm
->lock
);
4064 if (cap
->args
[0] > MAX_NR_RESERVED_IOAPIC_PINS
)
4065 goto split_irqchip_unlock
;
4067 if (irqchip_in_kernel(kvm
))
4068 goto split_irqchip_unlock
;
4069 if (kvm
->created_vcpus
)
4070 goto split_irqchip_unlock
;
4071 r
= kvm_setup_empty_irq_routing(kvm
);
4073 goto split_irqchip_unlock
;
4074 /* Pairs with irqchip_in_kernel. */
4076 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_SPLIT
;
4077 kvm
->arch
.nr_reserved_ioapic_pins
= cap
->args
[0];
4079 split_irqchip_unlock
:
4080 mutex_unlock(&kvm
->lock
);
4083 case KVM_CAP_X2APIC_API
:
4085 if (cap
->args
[0] & ~KVM_X2APIC_API_VALID_FLAGS
)
4088 if (cap
->args
[0] & KVM_X2APIC_API_USE_32BIT_IDS
)
4089 kvm
->arch
.x2apic_format
= true;
4090 if (cap
->args
[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK
)
4091 kvm
->arch
.x2apic_broadcast_quirk_disabled
= true;
4102 long kvm_arch_vm_ioctl(struct file
*filp
,
4103 unsigned int ioctl
, unsigned long arg
)
4105 struct kvm
*kvm
= filp
->private_data
;
4106 void __user
*argp
= (void __user
*)arg
;
4109 * This union makes it completely explicit to gcc-3.x
4110 * that these two variables' stack usage should be
4111 * combined, not added together.
4114 struct kvm_pit_state ps
;
4115 struct kvm_pit_state2 ps2
;
4116 struct kvm_pit_config pit_config
;
4120 case KVM_SET_TSS_ADDR
:
4121 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
4123 case KVM_SET_IDENTITY_MAP_ADDR
: {
4127 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
4129 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
4132 case KVM_SET_NR_MMU_PAGES
:
4133 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
4135 case KVM_GET_NR_MMU_PAGES
:
4136 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
4138 case KVM_CREATE_IRQCHIP
: {
4139 mutex_lock(&kvm
->lock
);
4142 if (irqchip_in_kernel(kvm
))
4143 goto create_irqchip_unlock
;
4146 if (kvm
->created_vcpus
)
4147 goto create_irqchip_unlock
;
4149 r
= kvm_pic_init(kvm
);
4151 goto create_irqchip_unlock
;
4153 r
= kvm_ioapic_init(kvm
);
4155 kvm_pic_destroy(kvm
);
4156 goto create_irqchip_unlock
;
4159 r
= kvm_setup_default_irq_routing(kvm
);
4161 kvm_ioapic_destroy(kvm
);
4162 kvm_pic_destroy(kvm
);
4163 goto create_irqchip_unlock
;
4165 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4167 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_KERNEL
;
4168 create_irqchip_unlock
:
4169 mutex_unlock(&kvm
->lock
);
4172 case KVM_CREATE_PIT
:
4173 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
4175 case KVM_CREATE_PIT2
:
4177 if (copy_from_user(&u
.pit_config
, argp
,
4178 sizeof(struct kvm_pit_config
)))
4181 mutex_lock(&kvm
->lock
);
4184 goto create_pit_unlock
;
4186 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
4190 mutex_unlock(&kvm
->lock
);
4192 case KVM_GET_IRQCHIP
: {
4193 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4194 struct kvm_irqchip
*chip
;
4196 chip
= memdup_user(argp
, sizeof(*chip
));
4203 if (!irqchip_kernel(kvm
))
4204 goto get_irqchip_out
;
4205 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
4207 goto get_irqchip_out
;
4209 if (copy_to_user(argp
, chip
, sizeof *chip
))
4210 goto get_irqchip_out
;
4216 case KVM_SET_IRQCHIP
: {
4217 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4218 struct kvm_irqchip
*chip
;
4220 chip
= memdup_user(argp
, sizeof(*chip
));
4227 if (!irqchip_kernel(kvm
))
4228 goto set_irqchip_out
;
4229 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
4231 goto set_irqchip_out
;
4239 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
4242 if (!kvm
->arch
.vpit
)
4244 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
4248 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
4255 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
4258 if (!kvm
->arch
.vpit
)
4260 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
4263 case KVM_GET_PIT2
: {
4265 if (!kvm
->arch
.vpit
)
4267 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
4271 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
4276 case KVM_SET_PIT2
: {
4278 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
4281 if (!kvm
->arch
.vpit
)
4283 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
4286 case KVM_REINJECT_CONTROL
: {
4287 struct kvm_reinject_control control
;
4289 if (copy_from_user(&control
, argp
, sizeof(control
)))
4291 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
4294 case KVM_SET_BOOT_CPU_ID
:
4296 mutex_lock(&kvm
->lock
);
4297 if (kvm
->created_vcpus
)
4300 kvm
->arch
.bsp_vcpu_id
= arg
;
4301 mutex_unlock(&kvm
->lock
);
4303 case KVM_XEN_HVM_CONFIG
: {
4304 struct kvm_xen_hvm_config xhc
;
4306 if (copy_from_user(&xhc
, argp
, sizeof(xhc
)))
4311 memcpy(&kvm
->arch
.xen_hvm_config
, &xhc
, sizeof(xhc
));
4315 case KVM_SET_CLOCK
: {
4316 struct kvm_clock_data user_ns
;
4320 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
4329 * TODO: userspace has to take care of races with VCPU_RUN, so
4330 * kvm_gen_update_masterclock() can be cut down to locked
4331 * pvclock_update_vm_gtod_copy().
4333 kvm_gen_update_masterclock(kvm
);
4334 now_ns
= get_kvmclock_ns(kvm
);
4335 kvm
->arch
.kvmclock_offset
+= user_ns
.clock
- now_ns
;
4336 kvm_make_all_cpus_request(kvm
, KVM_REQ_CLOCK_UPDATE
);
4339 case KVM_GET_CLOCK
: {
4340 struct kvm_clock_data user_ns
;
4343 now_ns
= get_kvmclock_ns(kvm
);
4344 user_ns
.clock
= now_ns
;
4345 user_ns
.flags
= kvm
->arch
.use_master_clock
? KVM_CLOCK_TSC_STABLE
: 0;
4346 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
4349 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
4354 case KVM_ENABLE_CAP
: {
4355 struct kvm_enable_cap cap
;
4358 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4360 r
= kvm_vm_ioctl_enable_cap(kvm
, &cap
);
4370 static void kvm_init_msr_list(void)
4375 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
4376 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
4380 * Even MSRs that are valid in the host may not be exposed
4381 * to the guests in some cases.
4383 switch (msrs_to_save
[i
]) {
4384 case MSR_IA32_BNDCFGS
:
4385 if (!kvm_x86_ops
->mpx_supported())
4389 if (!kvm_x86_ops
->rdtscp_supported())
4397 msrs_to_save
[j
] = msrs_to_save
[i
];
4400 num_msrs_to_save
= j
;
4402 for (i
= j
= 0; i
< ARRAY_SIZE(emulated_msrs
); i
++) {
4403 if (!kvm_x86_ops
->has_emulated_msr(emulated_msrs
[i
]))
4407 emulated_msrs
[j
] = emulated_msrs
[i
];
4410 num_emulated_msrs
= j
;
4412 for (i
= j
= 0; i
< ARRAY_SIZE(msr_based_features
); i
++) {
4413 struct kvm_msr_entry msr
;
4415 msr
.index
= msr_based_features
[i
];
4416 if (kvm_get_msr_feature(&msr
))
4420 msr_based_features
[j
] = msr_based_features
[i
];
4423 num_msr_based_features
= j
;
4426 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
4434 if (!(lapic_in_kernel(vcpu
) &&
4435 !kvm_iodevice_write(vcpu
, &vcpu
->arch
.apic
->dev
, addr
, n
, v
))
4436 && kvm_io_bus_write(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4447 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
4454 if (!(lapic_in_kernel(vcpu
) &&
4455 !kvm_iodevice_read(vcpu
, &vcpu
->arch
.apic
->dev
,
4457 && kvm_io_bus_read(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4459 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, v
);
4469 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
4470 struct kvm_segment
*var
, int seg
)
4472 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
4475 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
4476 struct kvm_segment
*var
, int seg
)
4478 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
4481 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
,
4482 struct x86_exception
*exception
)
4486 BUG_ON(!mmu_is_nested(vcpu
));
4488 /* NPT walks are always user-walks */
4489 access
|= PFERR_USER_MASK
;
4490 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, exception
);
4495 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
4496 struct x86_exception
*exception
)
4498 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4499 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4502 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
4503 struct x86_exception
*exception
)
4505 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4506 access
|= PFERR_FETCH_MASK
;
4507 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4510 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
4511 struct x86_exception
*exception
)
4513 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4514 access
|= PFERR_WRITE_MASK
;
4515 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4518 /* uses this to access any guest's mapped memory without checking CPL */
4519 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
4520 struct x86_exception
*exception
)
4522 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
4525 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
4526 struct kvm_vcpu
*vcpu
, u32 access
,
4527 struct x86_exception
*exception
)
4530 int r
= X86EMUL_CONTINUE
;
4533 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
4535 unsigned offset
= addr
& (PAGE_SIZE
-1);
4536 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4539 if (gpa
== UNMAPPED_GVA
)
4540 return X86EMUL_PROPAGATE_FAULT
;
4541 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, data
,
4544 r
= X86EMUL_IO_NEEDED
;
4556 /* used for instruction fetching */
4557 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
4558 gva_t addr
, void *val
, unsigned int bytes
,
4559 struct x86_exception
*exception
)
4561 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4562 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4566 /* Inline kvm_read_guest_virt_helper for speed. */
4567 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
|PFERR_FETCH_MASK
,
4569 if (unlikely(gpa
== UNMAPPED_GVA
))
4570 return X86EMUL_PROPAGATE_FAULT
;
4572 offset
= addr
& (PAGE_SIZE
-1);
4573 if (WARN_ON(offset
+ bytes
> PAGE_SIZE
))
4574 bytes
= (unsigned)PAGE_SIZE
- offset
;
4575 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, val
,
4577 if (unlikely(ret
< 0))
4578 return X86EMUL_IO_NEEDED
;
4580 return X86EMUL_CONTINUE
;
4583 int kvm_read_guest_virt(struct kvm_vcpu
*vcpu
,
4584 gva_t addr
, void *val
, unsigned int bytes
,
4585 struct x86_exception
*exception
)
4587 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4589 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
4592 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
4594 static int emulator_read_std(struct x86_emulate_ctxt
*ctxt
,
4595 gva_t addr
, void *val
, unsigned int bytes
,
4596 struct x86_exception
*exception
, bool system
)
4598 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4601 if (!system
&& kvm_x86_ops
->get_cpl(vcpu
) == 3)
4602 access
|= PFERR_USER_MASK
;
4604 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
, exception
);
4607 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt
*ctxt
,
4608 unsigned long addr
, void *val
, unsigned int bytes
)
4610 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4611 int r
= kvm_vcpu_read_guest(vcpu
, addr
, val
, bytes
);
4613 return r
< 0 ? X86EMUL_IO_NEEDED
: X86EMUL_CONTINUE
;
4616 static int kvm_write_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
4617 struct kvm_vcpu
*vcpu
, u32 access
,
4618 struct x86_exception
*exception
)
4621 int r
= X86EMUL_CONTINUE
;
4624 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
4627 unsigned offset
= addr
& (PAGE_SIZE
-1);
4628 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4631 if (gpa
== UNMAPPED_GVA
)
4632 return X86EMUL_PROPAGATE_FAULT
;
4633 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, data
, towrite
);
4635 r
= X86EMUL_IO_NEEDED
;
4647 static int emulator_write_std(struct x86_emulate_ctxt
*ctxt
, gva_t addr
, void *val
,
4648 unsigned int bytes
, struct x86_exception
*exception
,
4651 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4652 u32 access
= PFERR_WRITE_MASK
;
4654 if (!system
&& kvm_x86_ops
->get_cpl(vcpu
) == 3)
4655 access
|= PFERR_USER_MASK
;
4657 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
4661 int kvm_write_guest_virt_system(struct kvm_vcpu
*vcpu
, gva_t addr
, void *val
,
4662 unsigned int bytes
, struct x86_exception
*exception
)
4664 /* kvm_write_guest_virt_system can pull in tons of pages. */
4665 vcpu
->arch
.l1tf_flush_l1d
= true;
4667 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
4668 PFERR_WRITE_MASK
, exception
);
4670 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
4672 static int vcpu_is_mmio_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4673 gpa_t gpa
, bool write
)
4675 /* For APIC access vmexit */
4676 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4679 if (vcpu_match_mmio_gpa(vcpu
, gpa
)) {
4680 trace_vcpu_match_mmio(gva
, gpa
, write
, true);
4687 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4688 gpa_t
*gpa
, struct x86_exception
*exception
,
4691 u32 access
= ((kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
4692 | (write
? PFERR_WRITE_MASK
: 0);
4695 * currently PKRU is only applied to ept enabled guest so
4696 * there is no pkey in EPT page table for L1 guest or EPT
4697 * shadow page table for L2 guest.
4699 if (vcpu_match_mmio_gva(vcpu
, gva
)
4700 && !permission_fault(vcpu
, vcpu
->arch
.walk_mmu
,
4701 vcpu
->arch
.access
, 0, access
)) {
4702 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
4703 (gva
& (PAGE_SIZE
- 1));
4704 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
4708 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4710 if (*gpa
== UNMAPPED_GVA
)
4713 return vcpu_is_mmio_gpa(vcpu
, gva
, *gpa
, write
);
4716 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4717 const void *val
, int bytes
)
4721 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, val
, bytes
);
4724 kvm_page_track_write(vcpu
, gpa
, val
, bytes
);
4728 struct read_write_emulator_ops
{
4729 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
4731 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4732 void *val
, int bytes
);
4733 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4734 int bytes
, void *val
);
4735 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4736 void *val
, int bytes
);
4740 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
4742 if (vcpu
->mmio_read_completed
) {
4743 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
4744 vcpu
->mmio_fragments
[0].gpa
, val
);
4745 vcpu
->mmio_read_completed
= 0;
4752 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4753 void *val
, int bytes
)
4755 return !kvm_vcpu_read_guest(vcpu
, gpa
, val
, bytes
);
4758 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4759 void *val
, int bytes
)
4761 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
4764 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
4766 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, val
);
4767 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
4770 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4771 void *val
, int bytes
)
4773 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, NULL
);
4774 return X86EMUL_IO_NEEDED
;
4777 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4778 void *val
, int bytes
)
4780 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
4782 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
4783 return X86EMUL_CONTINUE
;
4786 static const struct read_write_emulator_ops read_emultor
= {
4787 .read_write_prepare
= read_prepare
,
4788 .read_write_emulate
= read_emulate
,
4789 .read_write_mmio
= vcpu_mmio_read
,
4790 .read_write_exit_mmio
= read_exit_mmio
,
4793 static const struct read_write_emulator_ops write_emultor
= {
4794 .read_write_emulate
= write_emulate
,
4795 .read_write_mmio
= write_mmio
,
4796 .read_write_exit_mmio
= write_exit_mmio
,
4800 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
4802 struct x86_exception
*exception
,
4803 struct kvm_vcpu
*vcpu
,
4804 const struct read_write_emulator_ops
*ops
)
4808 bool write
= ops
->write
;
4809 struct kvm_mmio_fragment
*frag
;
4810 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4813 * If the exit was due to a NPF we may already have a GPA.
4814 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4815 * Note, this cannot be used on string operations since string
4816 * operation using rep will only have the initial GPA from the NPF
4819 if (vcpu
->arch
.gpa_available
&&
4820 emulator_can_use_gpa(ctxt
) &&
4821 (addr
& ~PAGE_MASK
) == (vcpu
->arch
.gpa_val
& ~PAGE_MASK
)) {
4822 gpa
= vcpu
->arch
.gpa_val
;
4823 ret
= vcpu_is_mmio_gpa(vcpu
, addr
, gpa
, write
);
4825 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
4827 return X86EMUL_PROPAGATE_FAULT
;
4830 if (!ret
&& ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
4831 return X86EMUL_CONTINUE
;
4834 * Is this MMIO handled locally?
4836 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
4837 if (handled
== bytes
)
4838 return X86EMUL_CONTINUE
;
4844 WARN_ON(vcpu
->mmio_nr_fragments
>= KVM_MAX_MMIO_FRAGMENTS
);
4845 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
4849 return X86EMUL_CONTINUE
;
4852 static int emulator_read_write(struct x86_emulate_ctxt
*ctxt
,
4854 void *val
, unsigned int bytes
,
4855 struct x86_exception
*exception
,
4856 const struct read_write_emulator_ops
*ops
)
4858 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4862 if (ops
->read_write_prepare
&&
4863 ops
->read_write_prepare(vcpu
, val
, bytes
))
4864 return X86EMUL_CONTINUE
;
4866 vcpu
->mmio_nr_fragments
= 0;
4868 /* Crossing a page boundary? */
4869 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
4872 now
= -addr
& ~PAGE_MASK
;
4873 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
4876 if (rc
!= X86EMUL_CONTINUE
)
4879 if (ctxt
->mode
!= X86EMUL_MODE_PROT64
)
4885 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
4887 if (rc
!= X86EMUL_CONTINUE
)
4890 if (!vcpu
->mmio_nr_fragments
)
4893 gpa
= vcpu
->mmio_fragments
[0].gpa
;
4895 vcpu
->mmio_needed
= 1;
4896 vcpu
->mmio_cur_fragment
= 0;
4898 vcpu
->run
->mmio
.len
= min(8u, vcpu
->mmio_fragments
[0].len
);
4899 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
4900 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
4901 vcpu
->run
->mmio
.phys_addr
= gpa
;
4903 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
4906 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
4910 struct x86_exception
*exception
)
4912 return emulator_read_write(ctxt
, addr
, val
, bytes
,
4913 exception
, &read_emultor
);
4916 static int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
4920 struct x86_exception
*exception
)
4922 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
4923 exception
, &write_emultor
);
4926 #define CMPXCHG_TYPE(t, ptr, old, new) \
4927 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4929 #ifdef CONFIG_X86_64
4930 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4932 # define CMPXCHG64(ptr, old, new) \
4933 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4936 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
4941 struct x86_exception
*exception
)
4943 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4949 /* guests cmpxchg8b have to be emulated atomically */
4950 if (bytes
> 8 || (bytes
& (bytes
- 1)))
4953 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
4955 if (gpa
== UNMAPPED_GVA
||
4956 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4959 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
4962 page
= kvm_vcpu_gfn_to_page(vcpu
, gpa
>> PAGE_SHIFT
);
4963 if (is_error_page(page
))
4966 kaddr
= kmap_atomic(page
);
4967 kaddr
+= offset_in_page(gpa
);
4970 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
4973 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
4976 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
4979 exchanged
= CMPXCHG64(kaddr
, old
, new);
4984 kunmap_atomic(kaddr
);
4985 kvm_release_page_dirty(page
);
4988 return X86EMUL_CMPXCHG_FAILED
;
4990 kvm_vcpu_mark_page_dirty(vcpu
, gpa
>> PAGE_SHIFT
);
4991 kvm_page_track_write(vcpu
, gpa
, new, bytes
);
4993 return X86EMUL_CONTINUE
;
4996 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
4998 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
5001 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
5005 for (i
= 0; i
< vcpu
->arch
.pio
.count
; i
++) {
5006 if (vcpu
->arch
.pio
.in
)
5007 r
= kvm_io_bus_read(vcpu
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
5008 vcpu
->arch
.pio
.size
, pd
);
5010 r
= kvm_io_bus_write(vcpu
, KVM_PIO_BUS
,
5011 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
5015 pd
+= vcpu
->arch
.pio
.size
;
5020 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
5021 unsigned short port
, void *val
,
5022 unsigned int count
, bool in
)
5024 vcpu
->arch
.pio
.port
= port
;
5025 vcpu
->arch
.pio
.in
= in
;
5026 vcpu
->arch
.pio
.count
= count
;
5027 vcpu
->arch
.pio
.size
= size
;
5029 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
5030 vcpu
->arch
.pio
.count
= 0;
5034 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
5035 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
5036 vcpu
->run
->io
.size
= size
;
5037 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
5038 vcpu
->run
->io
.count
= count
;
5039 vcpu
->run
->io
.port
= port
;
5044 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
5045 int size
, unsigned short port
, void *val
,
5048 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5051 if (vcpu
->arch
.pio
.count
)
5054 memset(vcpu
->arch
.pio_data
, 0, size
* count
);
5056 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
5059 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
5060 trace_kvm_pio(KVM_PIO_IN
, port
, size
, count
, vcpu
->arch
.pio_data
);
5061 vcpu
->arch
.pio
.count
= 0;
5068 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
5069 int size
, unsigned short port
,
5070 const void *val
, unsigned int count
)
5072 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5074 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
5075 trace_kvm_pio(KVM_PIO_OUT
, port
, size
, count
, vcpu
->arch
.pio_data
);
5076 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
5079 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
5081 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
5084 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
5086 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
5089 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu
*vcpu
)
5091 if (!need_emulate_wbinvd(vcpu
))
5092 return X86EMUL_CONTINUE
;
5094 if (kvm_x86_ops
->has_wbinvd_exit()) {
5095 int cpu
= get_cpu();
5097 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
5098 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
5099 wbinvd_ipi
, NULL
, 1);
5101 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
5104 return X86EMUL_CONTINUE
;
5107 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
5109 kvm_emulate_wbinvd_noskip(vcpu
);
5110 return kvm_skip_emulated_instruction(vcpu
);
5112 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
5116 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
5118 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt
));
5121 static int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
5122 unsigned long *dest
)
5124 return kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
5127 static int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
5128 unsigned long value
)
5131 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
5134 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
5136 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
5139 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
5141 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5142 unsigned long value
;
5146 value
= kvm_read_cr0(vcpu
);
5149 value
= vcpu
->arch
.cr2
;
5152 value
= kvm_read_cr3(vcpu
);
5155 value
= kvm_read_cr4(vcpu
);
5158 value
= kvm_get_cr8(vcpu
);
5161 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
5168 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
5170 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5175 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
5178 vcpu
->arch
.cr2
= val
;
5181 res
= kvm_set_cr3(vcpu
, val
);
5184 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
5187 res
= kvm_set_cr8(vcpu
, val
);
5190 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
5197 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
5199 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
5202 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5204 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
5207 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5209 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
5212 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5214 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
5217 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5219 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
5222 static unsigned long emulator_get_cached_segment_base(
5223 struct x86_emulate_ctxt
*ctxt
, int seg
)
5225 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
5228 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
5229 struct desc_struct
*desc
, u32
*base3
,
5232 struct kvm_segment var
;
5234 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
5235 *selector
= var
.selector
;
5238 memset(desc
, 0, sizeof(*desc
));
5246 set_desc_limit(desc
, var
.limit
);
5247 set_desc_base(desc
, (unsigned long)var
.base
);
5248 #ifdef CONFIG_X86_64
5250 *base3
= var
.base
>> 32;
5252 desc
->type
= var
.type
;
5254 desc
->dpl
= var
.dpl
;
5255 desc
->p
= var
.present
;
5256 desc
->avl
= var
.avl
;
5264 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
5265 struct desc_struct
*desc
, u32 base3
,
5268 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5269 struct kvm_segment var
;
5271 var
.selector
= selector
;
5272 var
.base
= get_desc_base(desc
);
5273 #ifdef CONFIG_X86_64
5274 var
.base
|= ((u64
)base3
) << 32;
5276 var
.limit
= get_desc_limit(desc
);
5278 var
.limit
= (var
.limit
<< 12) | 0xfff;
5279 var
.type
= desc
->type
;
5280 var
.dpl
= desc
->dpl
;
5285 var
.avl
= desc
->avl
;
5286 var
.present
= desc
->p
;
5287 var
.unusable
= !var
.present
;
5290 kvm_set_segment(vcpu
, &var
, seg
);
5294 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
5295 u32 msr_index
, u64
*pdata
)
5297 struct msr_data msr
;
5300 msr
.index
= msr_index
;
5301 msr
.host_initiated
= false;
5302 r
= kvm_get_msr(emul_to_vcpu(ctxt
), &msr
);
5310 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
5311 u32 msr_index
, u64 data
)
5313 struct msr_data msr
;
5316 msr
.index
= msr_index
;
5317 msr
.host_initiated
= false;
5318 return kvm_set_msr(emul_to_vcpu(ctxt
), &msr
);
5321 static u64
emulator_get_smbase(struct x86_emulate_ctxt
*ctxt
)
5323 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5325 return vcpu
->arch
.smbase
;
5328 static void emulator_set_smbase(struct x86_emulate_ctxt
*ctxt
, u64 smbase
)
5330 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5332 vcpu
->arch
.smbase
= smbase
;
5335 static int emulator_check_pmc(struct x86_emulate_ctxt
*ctxt
,
5338 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt
), pmc
);
5341 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
5342 u32 pmc
, u64
*pdata
)
5344 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
5347 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
5349 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
5352 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
5355 kvm_load_guest_fpu(emul_to_vcpu(ctxt
));
5358 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
5363 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
5364 struct x86_instruction_info
*info
,
5365 enum x86_intercept_stage stage
)
5367 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
5370 static bool emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
5371 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
, bool check_limit
)
5373 return kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
, check_limit
);
5376 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
5378 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
5381 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
5383 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
5386 static void emulator_set_nmi_mask(struct x86_emulate_ctxt
*ctxt
, bool masked
)
5388 kvm_x86_ops
->set_nmi_mask(emul_to_vcpu(ctxt
), masked
);
5391 static unsigned emulator_get_hflags(struct x86_emulate_ctxt
*ctxt
)
5393 return emul_to_vcpu(ctxt
)->arch
.hflags
;
5396 static void emulator_set_hflags(struct x86_emulate_ctxt
*ctxt
, unsigned emul_flags
)
5398 kvm_set_hflags(emul_to_vcpu(ctxt
), emul_flags
);
5401 static const struct x86_emulate_ops emulate_ops
= {
5402 .read_gpr
= emulator_read_gpr
,
5403 .write_gpr
= emulator_write_gpr
,
5404 .read_std
= emulator_read_std
,
5405 .write_std
= emulator_write_std
,
5406 .read_phys
= kvm_read_guest_phys_system
,
5407 .fetch
= kvm_fetch_guest_virt
,
5408 .read_emulated
= emulator_read_emulated
,
5409 .write_emulated
= emulator_write_emulated
,
5410 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
5411 .invlpg
= emulator_invlpg
,
5412 .pio_in_emulated
= emulator_pio_in_emulated
,
5413 .pio_out_emulated
= emulator_pio_out_emulated
,
5414 .get_segment
= emulator_get_segment
,
5415 .set_segment
= emulator_set_segment
,
5416 .get_cached_segment_base
= emulator_get_cached_segment_base
,
5417 .get_gdt
= emulator_get_gdt
,
5418 .get_idt
= emulator_get_idt
,
5419 .set_gdt
= emulator_set_gdt
,
5420 .set_idt
= emulator_set_idt
,
5421 .get_cr
= emulator_get_cr
,
5422 .set_cr
= emulator_set_cr
,
5423 .cpl
= emulator_get_cpl
,
5424 .get_dr
= emulator_get_dr
,
5425 .set_dr
= emulator_set_dr
,
5426 .get_smbase
= emulator_get_smbase
,
5427 .set_smbase
= emulator_set_smbase
,
5428 .set_msr
= emulator_set_msr
,
5429 .get_msr
= emulator_get_msr
,
5430 .check_pmc
= emulator_check_pmc
,
5431 .read_pmc
= emulator_read_pmc
,
5432 .halt
= emulator_halt
,
5433 .wbinvd
= emulator_wbinvd
,
5434 .fix_hypercall
= emulator_fix_hypercall
,
5435 .get_fpu
= emulator_get_fpu
,
5436 .put_fpu
= emulator_put_fpu
,
5437 .intercept
= emulator_intercept
,
5438 .get_cpuid
= emulator_get_cpuid
,
5439 .set_nmi_mask
= emulator_set_nmi_mask
,
5440 .get_hflags
= emulator_get_hflags
,
5441 .set_hflags
= emulator_set_hflags
,
5444 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
5446 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
5448 * an sti; sti; sequence only disable interrupts for the first
5449 * instruction. So, if the last instruction, be it emulated or
5450 * not, left the system with the INT_STI flag enabled, it
5451 * means that the last instruction is an sti. We should not
5452 * leave the flag on in this case. The same goes for mov ss
5454 if (int_shadow
& mask
)
5456 if (unlikely(int_shadow
|| mask
)) {
5457 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
5459 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5463 static bool inject_emulated_exception(struct kvm_vcpu
*vcpu
)
5465 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5466 if (ctxt
->exception
.vector
== PF_VECTOR
)
5467 return kvm_propagate_fault(vcpu
, &ctxt
->exception
);
5469 if (ctxt
->exception
.error_code_valid
)
5470 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
5471 ctxt
->exception
.error_code
);
5473 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
5477 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
5479 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5482 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
5484 ctxt
->eflags
= kvm_get_rflags(vcpu
);
5485 ctxt
->tf
= (ctxt
->eflags
& X86_EFLAGS_TF
) != 0;
5487 ctxt
->eip
= kvm_rip_read(vcpu
);
5488 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
5489 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
5490 (cs_l
&& is_long_mode(vcpu
)) ? X86EMUL_MODE_PROT64
:
5491 cs_db
? X86EMUL_MODE_PROT32
:
5492 X86EMUL_MODE_PROT16
;
5493 BUILD_BUG_ON(HF_GUEST_MASK
!= X86EMUL_GUEST_MASK
);
5494 BUILD_BUG_ON(HF_SMM_MASK
!= X86EMUL_SMM_MASK
);
5495 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK
!= X86EMUL_SMM_INSIDE_NMI_MASK
);
5497 init_decode_cache(ctxt
);
5498 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5501 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
5503 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5506 init_emulate_ctxt(vcpu
);
5510 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
5511 ret
= emulate_int_real(ctxt
, irq
);
5513 if (ret
!= X86EMUL_CONTINUE
)
5514 return EMULATE_FAIL
;
5516 ctxt
->eip
= ctxt
->_eip
;
5517 kvm_rip_write(vcpu
, ctxt
->eip
);
5518 kvm_set_rflags(vcpu
, ctxt
->eflags
);
5520 if (irq
== NMI_VECTOR
)
5521 vcpu
->arch
.nmi_pending
= 0;
5523 vcpu
->arch
.interrupt
.pending
= false;
5525 return EMULATE_DONE
;
5527 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
5529 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
5531 int r
= EMULATE_DONE
;
5533 ++vcpu
->stat
.insn_emulation_fail
;
5534 trace_kvm_emulate_insn_failed(vcpu
);
5535 if (!is_guest_mode(vcpu
) && kvm_x86_ops
->get_cpl(vcpu
) == 0) {
5536 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
5537 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
5538 vcpu
->run
->internal
.ndata
= 0;
5539 r
= EMULATE_USER_EXIT
;
5541 kvm_queue_exception(vcpu
, UD_VECTOR
);
5546 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t cr2
,
5547 bool write_fault_to_shadow_pgtable
,
5553 if (emulation_type
& EMULTYPE_NO_REEXECUTE
)
5556 if (!vcpu
->arch
.mmu
.direct_map
) {
5558 * Write permission should be allowed since only
5559 * write access need to be emulated.
5561 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5564 * If the mapping is invalid in guest, let cpu retry
5565 * it to generate fault.
5567 if (gpa
== UNMAPPED_GVA
)
5572 * Do not retry the unhandleable instruction if it faults on the
5573 * readonly host memory, otherwise it will goto a infinite loop:
5574 * retry instruction -> write #PF -> emulation fail -> retry
5575 * instruction -> ...
5577 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
5580 * If the instruction failed on the error pfn, it can not be fixed,
5581 * report the error to userspace.
5583 if (is_error_noslot_pfn(pfn
))
5586 kvm_release_pfn_clean(pfn
);
5588 /* The instructions are well-emulated on direct mmu. */
5589 if (vcpu
->arch
.mmu
.direct_map
) {
5590 unsigned int indirect_shadow_pages
;
5592 spin_lock(&vcpu
->kvm
->mmu_lock
);
5593 indirect_shadow_pages
= vcpu
->kvm
->arch
.indirect_shadow_pages
;
5594 spin_unlock(&vcpu
->kvm
->mmu_lock
);
5596 if (indirect_shadow_pages
)
5597 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5603 * if emulation was due to access to shadowed page table
5604 * and it failed try to unshadow page and re-enter the
5605 * guest to let CPU execute the instruction.
5607 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5610 * If the access faults on its page table, it can not
5611 * be fixed by unprotecting shadow page and it should
5612 * be reported to userspace.
5614 return !write_fault_to_shadow_pgtable
;
5617 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
5618 unsigned long cr2
, int emulation_type
)
5620 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5621 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
5623 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
5624 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
5627 * If the emulation is caused by #PF and it is non-page_table
5628 * writing instruction, it means the VM-EXIT is caused by shadow
5629 * page protected, we can zap the shadow page and retry this
5630 * instruction directly.
5632 * Note: if the guest uses a non-page-table modifying instruction
5633 * on the PDE that points to the instruction, then we will unmap
5634 * the instruction and go to an infinite loop. So, we cache the
5635 * last retried eip and the last fault address, if we meet the eip
5636 * and the address again, we can break out of the potential infinite
5639 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
5641 if (!(emulation_type
& EMULTYPE_RETRY
))
5644 if (x86_page_table_writing_insn(ctxt
))
5647 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
5650 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
5651 vcpu
->arch
.last_retry_addr
= cr2
;
5653 if (!vcpu
->arch
.mmu
.direct_map
)
5654 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5656 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5661 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
5662 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
5664 static void kvm_smm_changed(struct kvm_vcpu
*vcpu
)
5666 if (!(vcpu
->arch
.hflags
& HF_SMM_MASK
)) {
5667 /* This is a good place to trace that we are exiting SMM. */
5668 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, false);
5670 /* Process a latched INIT or SMI, if any. */
5671 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5674 kvm_mmu_reset_context(vcpu
);
5677 static void kvm_set_hflags(struct kvm_vcpu
*vcpu
, unsigned emul_flags
)
5679 unsigned changed
= vcpu
->arch
.hflags
^ emul_flags
;
5681 vcpu
->arch
.hflags
= emul_flags
;
5683 if (changed
& HF_SMM_MASK
)
5684 kvm_smm_changed(vcpu
);
5687 static int kvm_vcpu_check_hw_bp(unsigned long addr
, u32 type
, u32 dr7
,
5696 for (i
= 0; i
< 4; i
++, enable
>>= 2, rwlen
>>= 4)
5697 if ((enable
& 3) && (rwlen
& 15) == type
&& db
[i
] == addr
)
5702 static void kvm_vcpu_do_singlestep(struct kvm_vcpu
*vcpu
, int *r
)
5704 struct kvm_run
*kvm_run
= vcpu
->run
;
5706 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
) {
5707 kvm_run
->debug
.arch
.dr6
= DR6_BS
| DR6_FIXED_1
| DR6_RTM
;
5708 kvm_run
->debug
.arch
.pc
= vcpu
->arch
.singlestep_rip
;
5709 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5710 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5711 *r
= EMULATE_USER_EXIT
;
5714 * "Certain debug exceptions may clear bit 0-3. The
5715 * remaining contents of the DR6 register are never
5716 * cleared by the processor".
5718 vcpu
->arch
.dr6
&= ~15;
5719 vcpu
->arch
.dr6
|= DR6_BS
| DR6_RTM
;
5720 kvm_queue_exception(vcpu
, DB_VECTOR
);
5724 int kvm_skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
5726 unsigned long rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5727 int r
= EMULATE_DONE
;
5729 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
5732 * rflags is the old, "raw" value of the flags. The new value has
5733 * not been saved yet.
5735 * This is correct even for TF set by the guest, because "the
5736 * processor will not generate this exception after the instruction
5737 * that sets the TF flag".
5739 if (unlikely(rflags
& X86_EFLAGS_TF
))
5740 kvm_vcpu_do_singlestep(vcpu
, &r
);
5741 return r
== EMULATE_DONE
;
5743 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction
);
5745 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu
*vcpu
, int *r
)
5747 if (unlikely(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) &&
5748 (vcpu
->arch
.guest_debug_dr7
& DR7_BP_EN_MASK
)) {
5749 struct kvm_run
*kvm_run
= vcpu
->run
;
5750 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5751 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5752 vcpu
->arch
.guest_debug_dr7
,
5756 kvm_run
->debug
.arch
.dr6
= dr6
| DR6_FIXED_1
| DR6_RTM
;
5757 kvm_run
->debug
.arch
.pc
= eip
;
5758 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5759 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5760 *r
= EMULATE_USER_EXIT
;
5765 if (unlikely(vcpu
->arch
.dr7
& DR7_BP_EN_MASK
) &&
5766 !(kvm_get_rflags(vcpu
) & X86_EFLAGS_RF
)) {
5767 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5768 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5773 vcpu
->arch
.dr6
&= ~15;
5774 vcpu
->arch
.dr6
|= dr6
| DR6_RTM
;
5775 kvm_queue_exception(vcpu
, DB_VECTOR
);
5784 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
5791 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5792 bool writeback
= true;
5793 bool write_fault_to_spt
= vcpu
->arch
.write_fault_to_shadow_pgtable
;
5795 vcpu
->arch
.l1tf_flush_l1d
= true;
5798 * Clear write_fault_to_shadow_pgtable here to ensure it is
5801 vcpu
->arch
.write_fault_to_shadow_pgtable
= false;
5802 kvm_clear_exception_queue(vcpu
);
5804 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
5805 init_emulate_ctxt(vcpu
);
5808 * We will reenter on the same instruction since
5809 * we do not set complete_userspace_io. This does not
5810 * handle watchpoints yet, those would be handled in
5813 if (!(emulation_type
& EMULTYPE_SKIP
) &&
5814 kvm_vcpu_check_breakpoint(vcpu
, &r
))
5817 ctxt
->interruptibility
= 0;
5818 ctxt
->have_exception
= false;
5819 ctxt
->exception
.vector
= -1;
5820 ctxt
->perm_ok
= false;
5822 ctxt
->ud
= emulation_type
& EMULTYPE_TRAP_UD
;
5824 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
5826 trace_kvm_emulate_insn_start(vcpu
);
5827 ++vcpu
->stat
.insn_emulation
;
5828 if (r
!= EMULATION_OK
) {
5829 if (emulation_type
& EMULTYPE_TRAP_UD
)
5830 return EMULATE_FAIL
;
5831 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5833 return EMULATE_DONE
;
5834 if (ctxt
->have_exception
&& inject_emulated_exception(vcpu
))
5835 return EMULATE_DONE
;
5836 if (emulation_type
& EMULTYPE_SKIP
)
5837 return EMULATE_FAIL
;
5838 return handle_emulation_failure(vcpu
);
5842 if (emulation_type
& EMULTYPE_SKIP
) {
5843 kvm_rip_write(vcpu
, ctxt
->_eip
);
5844 if (ctxt
->eflags
& X86_EFLAGS_RF
)
5845 kvm_set_rflags(vcpu
, ctxt
->eflags
& ~X86_EFLAGS_RF
);
5846 return EMULATE_DONE
;
5849 if (retry_instruction(ctxt
, cr2
, emulation_type
))
5850 return EMULATE_DONE
;
5852 /* this is needed for vmware backdoor interface to work since it
5853 changes registers values during IO operation */
5854 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
5855 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5856 emulator_invalidate_register_cache(ctxt
);
5860 /* Save the faulting GPA (cr2) in the address field */
5861 ctxt
->exception
.address
= cr2
;
5863 r
= x86_emulate_insn(ctxt
);
5865 if (r
== EMULATION_INTERCEPTED
)
5866 return EMULATE_DONE
;
5868 if (r
== EMULATION_FAILED
) {
5869 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5871 return EMULATE_DONE
;
5873 return handle_emulation_failure(vcpu
);
5876 if (ctxt
->have_exception
) {
5878 if (inject_emulated_exception(vcpu
))
5880 } else if (vcpu
->arch
.pio
.count
) {
5881 if (!vcpu
->arch
.pio
.in
) {
5882 /* FIXME: return into emulator if single-stepping. */
5883 vcpu
->arch
.pio
.count
= 0;
5886 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
5888 r
= EMULATE_USER_EXIT
;
5889 } else if (vcpu
->mmio_needed
) {
5890 if (!vcpu
->mmio_is_write
)
5892 r
= EMULATE_USER_EXIT
;
5893 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
5894 } else if (r
== EMULATION_RESTART
)
5900 unsigned long rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5901 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
5902 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5903 kvm_rip_write(vcpu
, ctxt
->eip
);
5904 if (r
== EMULATE_DONE
&&
5905 (ctxt
->tf
|| (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)))
5906 kvm_vcpu_do_singlestep(vcpu
, &r
);
5907 if (!ctxt
->have_exception
||
5908 exception_type(ctxt
->exception
.vector
) == EXCPT_TRAP
)
5909 __kvm_set_rflags(vcpu
, ctxt
->eflags
);
5912 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5913 * do nothing, and it will be requested again as soon as
5914 * the shadow expires. But we still need to check here,
5915 * because POPF has no interrupt shadow.
5917 if (unlikely((ctxt
->eflags
& ~rflags
) & X86_EFLAGS_IF
))
5918 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5920 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
5924 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
5926 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5928 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5929 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
5930 size
, port
, &val
, 1);
5931 /* do not return to emulator after return from userspace */
5932 vcpu
->arch
.pio
.count
= 0;
5935 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
5937 static int complete_fast_pio_in(struct kvm_vcpu
*vcpu
)
5941 /* We should only ever be called with arch.pio.count equal to 1 */
5942 BUG_ON(vcpu
->arch
.pio
.count
!= 1);
5944 /* For size less than 4 we merge, else we zero extend */
5945 val
= (vcpu
->arch
.pio
.size
< 4) ? kvm_register_read(vcpu
, VCPU_REGS_RAX
)
5949 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5950 * the copy and tracing
5952 emulator_pio_in_emulated(&vcpu
->arch
.emulate_ctxt
, vcpu
->arch
.pio
.size
,
5953 vcpu
->arch
.pio
.port
, &val
, 1);
5954 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
5959 int kvm_fast_pio_in(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5964 /* For size less than 4 we merge, else we zero extend */
5965 val
= (size
< 4) ? kvm_register_read(vcpu
, VCPU_REGS_RAX
) : 0;
5967 ret
= emulator_pio_in_emulated(&vcpu
->arch
.emulate_ctxt
, size
, port
,
5970 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
5974 vcpu
->arch
.complete_userspace_io
= complete_fast_pio_in
;
5978 EXPORT_SYMBOL_GPL(kvm_fast_pio_in
);
5980 static int kvmclock_cpu_down_prep(unsigned int cpu
)
5982 __this_cpu_write(cpu_tsc_khz
, 0);
5986 static void tsc_khz_changed(void *data
)
5988 struct cpufreq_freqs
*freq
= data
;
5989 unsigned long khz
= 0;
5993 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5994 khz
= cpufreq_quick_get(raw_smp_processor_id());
5997 __this_cpu_write(cpu_tsc_khz
, khz
);
6000 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
6003 struct cpufreq_freqs
*freq
= data
;
6005 struct kvm_vcpu
*vcpu
;
6006 int i
, send_ipi
= 0;
6009 * We allow guests to temporarily run on slowing clocks,
6010 * provided we notify them after, or to run on accelerating
6011 * clocks, provided we notify them before. Thus time never
6014 * However, we have a problem. We can't atomically update
6015 * the frequency of a given CPU from this function; it is
6016 * merely a notifier, which can be called from any CPU.
6017 * Changing the TSC frequency at arbitrary points in time
6018 * requires a recomputation of local variables related to
6019 * the TSC for each VCPU. We must flag these local variables
6020 * to be updated and be sure the update takes place with the
6021 * new frequency before any guests proceed.
6023 * Unfortunately, the combination of hotplug CPU and frequency
6024 * change creates an intractable locking scenario; the order
6025 * of when these callouts happen is undefined with respect to
6026 * CPU hotplug, and they can race with each other. As such,
6027 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6028 * undefined; you can actually have a CPU frequency change take
6029 * place in between the computation of X and the setting of the
6030 * variable. To protect against this problem, all updates of
6031 * the per_cpu tsc_khz variable are done in an interrupt
6032 * protected IPI, and all callers wishing to update the value
6033 * must wait for a synchronous IPI to complete (which is trivial
6034 * if the caller is on the CPU already). This establishes the
6035 * necessary total order on variable updates.
6037 * Note that because a guest time update may take place
6038 * anytime after the setting of the VCPU's request bit, the
6039 * correct TSC value must be set before the request. However,
6040 * to ensure the update actually makes it to any guest which
6041 * starts running in hardware virtualization between the set
6042 * and the acquisition of the spinlock, we must also ping the
6043 * CPU after setting the request bit.
6047 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
6049 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
6052 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
6054 spin_lock(&kvm_lock
);
6055 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6056 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6057 if (vcpu
->cpu
!= freq
->cpu
)
6059 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
6060 if (vcpu
->cpu
!= smp_processor_id())
6064 spin_unlock(&kvm_lock
);
6066 if (freq
->old
< freq
->new && send_ipi
) {
6068 * We upscale the frequency. Must make the guest
6069 * doesn't see old kvmclock values while running with
6070 * the new frequency, otherwise we risk the guest sees
6071 * time go backwards.
6073 * In case we update the frequency for another cpu
6074 * (which might be in guest context) send an interrupt
6075 * to kick the cpu out of guest context. Next time
6076 * guest context is entered kvmclock will be updated,
6077 * so the guest will not see stale values.
6079 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
6084 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
6085 .notifier_call
= kvmclock_cpufreq_notifier
6088 static int kvmclock_cpu_online(unsigned int cpu
)
6090 tsc_khz_changed(NULL
);
6094 static void kvm_timer_init(void)
6096 max_tsc_khz
= tsc_khz
;
6098 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
6099 #ifdef CONFIG_CPU_FREQ
6100 struct cpufreq_policy policy
;
6103 memset(&policy
, 0, sizeof(policy
));
6105 cpufreq_get_policy(&policy
, cpu
);
6106 if (policy
.cpuinfo
.max_freq
)
6107 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
6110 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
6111 CPUFREQ_TRANSITION_NOTIFIER
);
6113 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
6115 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE
, "x86/kvm/clk:online",
6116 kvmclock_cpu_online
, kvmclock_cpu_down_prep
);
6119 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
6121 int kvm_is_in_guest(void)
6123 return __this_cpu_read(current_vcpu
) != NULL
;
6126 static int kvm_is_user_mode(void)
6130 if (__this_cpu_read(current_vcpu
))
6131 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
6133 return user_mode
!= 0;
6136 static unsigned long kvm_get_guest_ip(void)
6138 unsigned long ip
= 0;
6140 if (__this_cpu_read(current_vcpu
))
6141 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
6146 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6147 .is_in_guest
= kvm_is_in_guest
,
6148 .is_user_mode
= kvm_is_user_mode
,
6149 .get_guest_ip
= kvm_get_guest_ip
,
6152 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
6154 __this_cpu_write(current_vcpu
, vcpu
);
6156 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
6158 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
6160 __this_cpu_write(current_vcpu
, NULL
);
6162 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
6164 static void kvm_set_mmio_spte_mask(void)
6167 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
6170 * Set the reserved bits and the present bit of an paging-structure
6171 * entry to generate page fault with PFER.RSV = 1.
6173 /* Mask the reserved physical address bits. */
6174 mask
= rsvd_bits(maxphyaddr
, 51);
6176 /* Set the present bit. */
6179 #ifdef CONFIG_X86_64
6181 * If reserved bit is not supported, clear the present bit to disable
6184 if (maxphyaddr
== 52)
6188 kvm_mmu_set_mmio_spte_mask(mask
, mask
);
6191 #ifdef CONFIG_X86_64
6192 static void pvclock_gtod_update_fn(struct work_struct
*work
)
6196 struct kvm_vcpu
*vcpu
;
6199 spin_lock(&kvm_lock
);
6200 list_for_each_entry(kvm
, &vm_list
, vm_list
)
6201 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6202 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
6203 atomic_set(&kvm_guest_has_master_clock
, 0);
6204 spin_unlock(&kvm_lock
);
6207 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
6210 * Notification about pvclock gtod data update.
6212 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
6215 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
6216 struct timekeeper
*tk
= priv
;
6218 update_pvclock_gtod(tk
);
6220 /* disable master clock if host does not trust, or does not
6221 * use, TSC clocksource
6223 if (gtod
->clock
.vclock_mode
!= VCLOCK_TSC
&&
6224 atomic_read(&kvm_guest_has_master_clock
) != 0)
6225 queue_work(system_long_wq
, &pvclock_gtod_work
);
6230 static struct notifier_block pvclock_gtod_notifier
= {
6231 .notifier_call
= pvclock_gtod_notify
,
6235 int kvm_arch_init(void *opaque
)
6238 struct kvm_x86_ops
*ops
= opaque
;
6241 printk(KERN_ERR
"kvm: already loaded the other module\n");
6246 if (!ops
->cpu_has_kvm_support()) {
6247 printk(KERN_ERR
"kvm: no hardware support\n");
6251 if (ops
->disabled_by_bios()) {
6252 printk(KERN_ERR
"kvm: disabled by bios\n");
6258 shared_msrs
= alloc_percpu(struct kvm_shared_msrs
);
6260 printk(KERN_ERR
"kvm: failed to allocate percpu kvm_shared_msrs\n");
6264 r
= kvm_mmu_module_init();
6266 goto out_free_percpu
;
6268 kvm_set_mmio_spte_mask();
6272 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
6273 PT_DIRTY_MASK
, PT64_NX_MASK
, 0,
6274 PT_PRESENT_MASK
, 0, sme_me_mask
);
6277 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6279 if (boot_cpu_has(X86_FEATURE_XSAVE
))
6280 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
6283 #ifdef CONFIG_X86_64
6284 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
6290 free_percpu(shared_msrs
);
6295 void kvm_arch_exit(void)
6298 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6300 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
6301 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
6302 CPUFREQ_TRANSITION_NOTIFIER
);
6303 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE
);
6304 #ifdef CONFIG_X86_64
6305 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
6308 kvm_mmu_module_exit();
6309 free_percpu(shared_msrs
);
6312 int kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
6314 ++vcpu
->stat
.halt_exits
;
6315 if (lapic_in_kernel(vcpu
)) {
6316 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
6319 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
6323 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
6325 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
6327 int ret
= kvm_skip_emulated_instruction(vcpu
);
6329 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6330 * KVM_EXIT_DEBUG here.
6332 return kvm_vcpu_halt(vcpu
) && ret
;
6334 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
6336 #ifdef CONFIG_X86_64
6337 static int kvm_pv_clock_pairing(struct kvm_vcpu
*vcpu
, gpa_t paddr
,
6338 unsigned long clock_type
)
6340 struct kvm_clock_pairing clock_pairing
;
6345 if (clock_type
!= KVM_CLOCK_PAIRING_WALLCLOCK
)
6346 return -KVM_EOPNOTSUPP
;
6348 if (kvm_get_walltime_and_clockread(&ts
, &cycle
) == false)
6349 return -KVM_EOPNOTSUPP
;
6351 clock_pairing
.sec
= ts
.tv_sec
;
6352 clock_pairing
.nsec
= ts
.tv_nsec
;
6353 clock_pairing
.tsc
= kvm_read_l1_tsc(vcpu
, cycle
);
6354 clock_pairing
.flags
= 0;
6357 if (kvm_write_guest(vcpu
->kvm
, paddr
, &clock_pairing
,
6358 sizeof(struct kvm_clock_pairing
)))
6366 * kvm_pv_kick_cpu_op: Kick a vcpu.
6368 * @apicid - apicid of vcpu to be kicked.
6370 static void kvm_pv_kick_cpu_op(struct kvm
*kvm
, unsigned long flags
, int apicid
)
6372 struct kvm_lapic_irq lapic_irq
;
6374 lapic_irq
.shorthand
= 0;
6375 lapic_irq
.dest_mode
= 0;
6376 lapic_irq
.level
= 0;
6377 lapic_irq
.dest_id
= apicid
;
6378 lapic_irq
.msi_redir_hint
= false;
6380 lapic_irq
.delivery_mode
= APIC_DM_REMRD
;
6381 kvm_irq_delivery_to_apic(kvm
, NULL
, &lapic_irq
, NULL
);
6384 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu
*vcpu
)
6386 vcpu
->arch
.apicv_active
= false;
6387 kvm_x86_ops
->refresh_apicv_exec_ctrl(vcpu
);
6390 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
6392 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
6395 if (kvm_hv_hypercall_enabled(vcpu
->kvm
)) {
6396 if (!kvm_hv_hypercall(vcpu
))
6401 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
6402 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
6403 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
6404 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
6405 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
6407 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
6409 op_64_bit
= is_64_bit_mode(vcpu
);
6418 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
6424 case KVM_HC_VAPIC_POLL_IRQ
:
6427 case KVM_HC_KICK_CPU
:
6428 kvm_pv_kick_cpu_op(vcpu
->kvm
, a0
, a1
);
6431 #ifdef CONFIG_X86_64
6432 case KVM_HC_CLOCK_PAIRING
:
6433 ret
= kvm_pv_clock_pairing(vcpu
, a0
, a1
);
6443 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
6446 ++vcpu
->stat
.hypercalls
;
6447 return kvm_skip_emulated_instruction(vcpu
);
6449 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
6451 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
6453 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6454 char instruction
[3];
6455 unsigned long rip
= kvm_rip_read(vcpu
);
6457 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
6459 return emulator_write_emulated(ctxt
, rip
, instruction
, 3,
6463 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
6465 return vcpu
->run
->request_interrupt_window
&&
6466 likely(!pic_in_kernel(vcpu
->kvm
));
6469 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
6471 struct kvm_run
*kvm_run
= vcpu
->run
;
6473 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
6474 kvm_run
->flags
= is_smm(vcpu
) ? KVM_RUN_X86_SMM
: 0;
6475 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
6476 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
6477 kvm_run
->ready_for_interrupt_injection
=
6478 pic_in_kernel(vcpu
->kvm
) ||
6479 kvm_vcpu_ready_for_interrupt_injection(vcpu
);
6482 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
6486 if (!kvm_x86_ops
->update_cr8_intercept
)
6489 if (!lapic_in_kernel(vcpu
))
6492 if (vcpu
->arch
.apicv_active
)
6495 if (!vcpu
->arch
.apic
->vapic_addr
)
6496 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
6503 tpr
= kvm_lapic_get_cr8(vcpu
);
6505 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
6508 static int inject_pending_event(struct kvm_vcpu
*vcpu
, bool req_int_win
)
6512 /* try to reinject previous events if any */
6513 if (vcpu
->arch
.exception
.injected
) {
6514 kvm_x86_ops
->queue_exception(vcpu
);
6519 * Exceptions must be injected immediately, or the exception
6520 * frame will have the address of the NMI or interrupt handler.
6522 if (!vcpu
->arch
.exception
.pending
) {
6523 if (vcpu
->arch
.nmi_injected
) {
6524 kvm_x86_ops
->set_nmi(vcpu
);
6528 if (vcpu
->arch
.interrupt
.pending
) {
6529 kvm_x86_ops
->set_irq(vcpu
);
6534 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6535 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6540 /* try to inject new event if pending */
6541 if (vcpu
->arch
.exception
.pending
) {
6542 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
6543 vcpu
->arch
.exception
.has_error_code
,
6544 vcpu
->arch
.exception
.error_code
);
6546 vcpu
->arch
.exception
.pending
= false;
6547 vcpu
->arch
.exception
.injected
= true;
6549 if (exception_type(vcpu
->arch
.exception
.nr
) == EXCPT_FAULT
)
6550 __kvm_set_rflags(vcpu
, kvm_get_rflags(vcpu
) |
6553 if (vcpu
->arch
.exception
.nr
== DB_VECTOR
&&
6554 (vcpu
->arch
.dr7
& DR7_GD
)) {
6555 vcpu
->arch
.dr7
&= ~DR7_GD
;
6556 kvm_update_dr7(vcpu
);
6559 kvm_x86_ops
->queue_exception(vcpu
);
6560 } else if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
)) {
6561 vcpu
->arch
.smi_pending
= false;
6563 } else if (vcpu
->arch
.nmi_pending
&& kvm_x86_ops
->nmi_allowed(vcpu
)) {
6564 --vcpu
->arch
.nmi_pending
;
6565 vcpu
->arch
.nmi_injected
= true;
6566 kvm_x86_ops
->set_nmi(vcpu
);
6567 } else if (kvm_cpu_has_injectable_intr(vcpu
)) {
6569 * Because interrupts can be injected asynchronously, we are
6570 * calling check_nested_events again here to avoid a race condition.
6571 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6572 * proposal and current concerns. Perhaps we should be setting
6573 * KVM_REQ_EVENT only on certain events and not unconditionally?
6575 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6576 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6580 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
6581 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
6583 kvm_x86_ops
->set_irq(vcpu
);
6590 static void process_nmi(struct kvm_vcpu
*vcpu
)
6595 * x86 is limited to one NMI running, and one NMI pending after it.
6596 * If an NMI is already in progress, limit further NMIs to just one.
6597 * Otherwise, allow two (and we'll inject the first one immediately).
6599 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
6602 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
6603 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
6604 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6607 #define put_smstate(type, buf, offset, val) \
6608 *(type *)((buf) + (offset) - 0x7e00) = val
6610 static u32
enter_smm_get_segment_flags(struct kvm_segment
*seg
)
6613 flags
|= seg
->g
<< 23;
6614 flags
|= seg
->db
<< 22;
6615 flags
|= seg
->l
<< 21;
6616 flags
|= seg
->avl
<< 20;
6617 flags
|= seg
->present
<< 15;
6618 flags
|= seg
->dpl
<< 13;
6619 flags
|= seg
->s
<< 12;
6620 flags
|= seg
->type
<< 8;
6624 static void enter_smm_save_seg_32(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6626 struct kvm_segment seg
;
6629 kvm_get_segment(vcpu
, &seg
, n
);
6630 put_smstate(u32
, buf
, 0x7fa8 + n
* 4, seg
.selector
);
6633 offset
= 0x7f84 + n
* 12;
6635 offset
= 0x7f2c + (n
- 3) * 12;
6637 put_smstate(u32
, buf
, offset
+ 8, seg
.base
);
6638 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6639 put_smstate(u32
, buf
, offset
, enter_smm_get_segment_flags(&seg
));
6642 #ifdef CONFIG_X86_64
6643 static void enter_smm_save_seg_64(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6645 struct kvm_segment seg
;
6649 kvm_get_segment(vcpu
, &seg
, n
);
6650 offset
= 0x7e00 + n
* 16;
6652 flags
= enter_smm_get_segment_flags(&seg
) >> 8;
6653 put_smstate(u16
, buf
, offset
, seg
.selector
);
6654 put_smstate(u16
, buf
, offset
+ 2, flags
);
6655 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6656 put_smstate(u64
, buf
, offset
+ 8, seg
.base
);
6660 static void enter_smm_save_state_32(struct kvm_vcpu
*vcpu
, char *buf
)
6663 struct kvm_segment seg
;
6667 put_smstate(u32
, buf
, 0x7ffc, kvm_read_cr0(vcpu
));
6668 put_smstate(u32
, buf
, 0x7ff8, kvm_read_cr3(vcpu
));
6669 put_smstate(u32
, buf
, 0x7ff4, kvm_get_rflags(vcpu
));
6670 put_smstate(u32
, buf
, 0x7ff0, kvm_rip_read(vcpu
));
6672 for (i
= 0; i
< 8; i
++)
6673 put_smstate(u32
, buf
, 0x7fd0 + i
* 4, kvm_register_read(vcpu
, i
));
6675 kvm_get_dr(vcpu
, 6, &val
);
6676 put_smstate(u32
, buf
, 0x7fcc, (u32
)val
);
6677 kvm_get_dr(vcpu
, 7, &val
);
6678 put_smstate(u32
, buf
, 0x7fc8, (u32
)val
);
6680 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6681 put_smstate(u32
, buf
, 0x7fc4, seg
.selector
);
6682 put_smstate(u32
, buf
, 0x7f64, seg
.base
);
6683 put_smstate(u32
, buf
, 0x7f60, seg
.limit
);
6684 put_smstate(u32
, buf
, 0x7f5c, enter_smm_get_segment_flags(&seg
));
6686 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6687 put_smstate(u32
, buf
, 0x7fc0, seg
.selector
);
6688 put_smstate(u32
, buf
, 0x7f80, seg
.base
);
6689 put_smstate(u32
, buf
, 0x7f7c, seg
.limit
);
6690 put_smstate(u32
, buf
, 0x7f78, enter_smm_get_segment_flags(&seg
));
6692 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6693 put_smstate(u32
, buf
, 0x7f74, dt
.address
);
6694 put_smstate(u32
, buf
, 0x7f70, dt
.size
);
6696 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6697 put_smstate(u32
, buf
, 0x7f58, dt
.address
);
6698 put_smstate(u32
, buf
, 0x7f54, dt
.size
);
6700 for (i
= 0; i
< 6; i
++)
6701 enter_smm_save_seg_32(vcpu
, buf
, i
);
6703 put_smstate(u32
, buf
, 0x7f14, kvm_read_cr4(vcpu
));
6706 put_smstate(u32
, buf
, 0x7efc, 0x00020000);
6707 put_smstate(u32
, buf
, 0x7ef8, vcpu
->arch
.smbase
);
6710 static void enter_smm_save_state_64(struct kvm_vcpu
*vcpu
, char *buf
)
6712 #ifdef CONFIG_X86_64
6714 struct kvm_segment seg
;
6718 for (i
= 0; i
< 16; i
++)
6719 put_smstate(u64
, buf
, 0x7ff8 - i
* 8, kvm_register_read(vcpu
, i
));
6721 put_smstate(u64
, buf
, 0x7f78, kvm_rip_read(vcpu
));
6722 put_smstate(u32
, buf
, 0x7f70, kvm_get_rflags(vcpu
));
6724 kvm_get_dr(vcpu
, 6, &val
);
6725 put_smstate(u64
, buf
, 0x7f68, val
);
6726 kvm_get_dr(vcpu
, 7, &val
);
6727 put_smstate(u64
, buf
, 0x7f60, val
);
6729 put_smstate(u64
, buf
, 0x7f58, kvm_read_cr0(vcpu
));
6730 put_smstate(u64
, buf
, 0x7f50, kvm_read_cr3(vcpu
));
6731 put_smstate(u64
, buf
, 0x7f48, kvm_read_cr4(vcpu
));
6733 put_smstate(u32
, buf
, 0x7f00, vcpu
->arch
.smbase
);
6736 put_smstate(u32
, buf
, 0x7efc, 0x00020064);
6738 put_smstate(u64
, buf
, 0x7ed0, vcpu
->arch
.efer
);
6740 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6741 put_smstate(u16
, buf
, 0x7e90, seg
.selector
);
6742 put_smstate(u16
, buf
, 0x7e92, enter_smm_get_segment_flags(&seg
) >> 8);
6743 put_smstate(u32
, buf
, 0x7e94, seg
.limit
);
6744 put_smstate(u64
, buf
, 0x7e98, seg
.base
);
6746 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6747 put_smstate(u32
, buf
, 0x7e84, dt
.size
);
6748 put_smstate(u64
, buf
, 0x7e88, dt
.address
);
6750 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6751 put_smstate(u16
, buf
, 0x7e70, seg
.selector
);
6752 put_smstate(u16
, buf
, 0x7e72, enter_smm_get_segment_flags(&seg
) >> 8);
6753 put_smstate(u32
, buf
, 0x7e74, seg
.limit
);
6754 put_smstate(u64
, buf
, 0x7e78, seg
.base
);
6756 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6757 put_smstate(u32
, buf
, 0x7e64, dt
.size
);
6758 put_smstate(u64
, buf
, 0x7e68, dt
.address
);
6760 for (i
= 0; i
< 6; i
++)
6761 enter_smm_save_seg_64(vcpu
, buf
, i
);
6767 static void enter_smm(struct kvm_vcpu
*vcpu
)
6769 struct kvm_segment cs
, ds
;
6774 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, true);
6775 vcpu
->arch
.hflags
|= HF_SMM_MASK
;
6776 memset(buf
, 0, 512);
6777 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
6778 enter_smm_save_state_64(vcpu
, buf
);
6780 enter_smm_save_state_32(vcpu
, buf
);
6782 kvm_vcpu_write_guest(vcpu
, vcpu
->arch
.smbase
+ 0xfe00, buf
, sizeof(buf
));
6784 if (kvm_x86_ops
->get_nmi_mask(vcpu
))
6785 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
6787 kvm_x86_ops
->set_nmi_mask(vcpu
, true);
6789 kvm_set_rflags(vcpu
, X86_EFLAGS_FIXED
);
6790 kvm_rip_write(vcpu
, 0x8000);
6792 cr0
= vcpu
->arch
.cr0
& ~(X86_CR0_PE
| X86_CR0_EM
| X86_CR0_TS
| X86_CR0_PG
);
6793 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
6794 vcpu
->arch
.cr0
= cr0
;
6796 kvm_x86_ops
->set_cr4(vcpu
, 0);
6798 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6799 dt
.address
= dt
.size
= 0;
6800 kvm_x86_ops
->set_idt(vcpu
, &dt
);
6802 __kvm_set_dr(vcpu
, 7, DR7_FIXED_1
);
6804 cs
.selector
= (vcpu
->arch
.smbase
>> 4) & 0xffff;
6805 cs
.base
= vcpu
->arch
.smbase
;
6810 cs
.limit
= ds
.limit
= 0xffffffff;
6811 cs
.type
= ds
.type
= 0x3;
6812 cs
.dpl
= ds
.dpl
= 0;
6817 cs
.avl
= ds
.avl
= 0;
6818 cs
.present
= ds
.present
= 1;
6819 cs
.unusable
= ds
.unusable
= 0;
6820 cs
.padding
= ds
.padding
= 0;
6822 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6823 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_DS
);
6824 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_ES
);
6825 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_FS
);
6826 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_GS
);
6827 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_SS
);
6829 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
6830 kvm_x86_ops
->set_efer(vcpu
, 0);
6832 kvm_update_cpuid(vcpu
);
6833 kvm_mmu_reset_context(vcpu
);
6836 static void process_smi(struct kvm_vcpu
*vcpu
)
6838 vcpu
->arch
.smi_pending
= true;
6839 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6842 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
6844 kvm_make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
6847 static void vcpu_scan_ioapic(struct kvm_vcpu
*vcpu
)
6849 u64 eoi_exit_bitmap
[4];
6851 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
6854 bitmap_zero(vcpu
->arch
.ioapic_handled_vectors
, 256);
6856 if (irqchip_split(vcpu
->kvm
))
6857 kvm_scan_ioapic_routes(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6859 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
6860 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
6861 kvm_ioapic_scan_entry(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6863 bitmap_or((ulong
*)eoi_exit_bitmap
, vcpu
->arch
.ioapic_handled_vectors
,
6864 vcpu_to_synic(vcpu
)->vec_bitmap
, 256);
6865 kvm_x86_ops
->load_eoi_exitmap(vcpu
, eoi_exit_bitmap
);
6868 static void kvm_vcpu_flush_tlb(struct kvm_vcpu
*vcpu
)
6870 ++vcpu
->stat
.tlb_flush
;
6871 kvm_x86_ops
->tlb_flush(vcpu
);
6874 void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
6875 unsigned long start
, unsigned long end
)
6877 unsigned long apic_address
;
6880 * The physical address of apic access page is stored in the VMCS.
6881 * Update it when it becomes invalid.
6883 apic_address
= gfn_to_hva(kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
6884 if (start
<= apic_address
&& apic_address
< end
)
6885 kvm_make_all_cpus_request(kvm
, KVM_REQ_APIC_PAGE_RELOAD
);
6888 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu
*vcpu
)
6890 struct page
*page
= NULL
;
6892 if (!lapic_in_kernel(vcpu
))
6895 if (!kvm_x86_ops
->set_apic_access_page_addr
)
6898 page
= gfn_to_page(vcpu
->kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
6899 if (is_error_page(page
))
6901 kvm_x86_ops
->set_apic_access_page_addr(vcpu
, page_to_phys(page
));
6904 * Do not pin apic access page in memory, the MMU notifier
6905 * will call us again if it is migrated or swapped out.
6909 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page
);
6912 * Returns 1 to let vcpu_run() continue the guest execution loop without
6913 * exiting to the userspace. Otherwise, the value will be returned to the
6916 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
6920 dm_request_for_irq_injection(vcpu
) &&
6921 kvm_cpu_accept_dm_intr(vcpu
);
6923 bool req_immediate_exit
= false;
6925 if (kvm_request_pending(vcpu
)) {
6926 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
6927 kvm_mmu_unload(vcpu
);
6928 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
6929 __kvm_migrate_timers(vcpu
);
6930 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
6931 kvm_gen_update_masterclock(vcpu
->kvm
);
6932 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
))
6933 kvm_gen_kvmclock_update(vcpu
);
6934 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
6935 r
= kvm_guest_time_update(vcpu
);
6939 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
6940 kvm_mmu_sync_roots(vcpu
);
6941 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
6942 kvm_vcpu_flush_tlb(vcpu
);
6943 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
6944 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
6948 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
6949 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
6950 vcpu
->mmio_needed
= 0;
6954 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
6955 /* Page is swapped out. Do synthetic halt */
6956 vcpu
->arch
.apf
.halted
= true;
6960 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
6961 record_steal_time(vcpu
);
6962 if (kvm_check_request(KVM_REQ_SMI
, vcpu
))
6964 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
6966 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
6967 kvm_pmu_handle_event(vcpu
);
6968 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
6969 kvm_pmu_deliver_pmi(vcpu
);
6970 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT
, vcpu
)) {
6971 BUG_ON(vcpu
->arch
.pending_ioapic_eoi
> 255);
6972 if (test_bit(vcpu
->arch
.pending_ioapic_eoi
,
6973 vcpu
->arch
.ioapic_handled_vectors
)) {
6974 vcpu
->run
->exit_reason
= KVM_EXIT_IOAPIC_EOI
;
6975 vcpu
->run
->eoi
.vector
=
6976 vcpu
->arch
.pending_ioapic_eoi
;
6981 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC
, vcpu
))
6982 vcpu_scan_ioapic(vcpu
);
6983 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD
, vcpu
))
6984 kvm_vcpu_reload_apic_access_page(vcpu
);
6985 if (kvm_check_request(KVM_REQ_HV_CRASH
, vcpu
)) {
6986 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
6987 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_CRASH
;
6991 if (kvm_check_request(KVM_REQ_HV_RESET
, vcpu
)) {
6992 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
6993 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_RESET
;
6997 if (kvm_check_request(KVM_REQ_HV_EXIT
, vcpu
)) {
6998 vcpu
->run
->exit_reason
= KVM_EXIT_HYPERV
;
6999 vcpu
->run
->hyperv
= vcpu
->arch
.hyperv
.exit
;
7005 * KVM_REQ_HV_STIMER has to be processed after
7006 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7007 * depend on the guest clock being up-to-date
7009 if (kvm_check_request(KVM_REQ_HV_STIMER
, vcpu
))
7010 kvm_hv_process_stimers(vcpu
);
7013 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
7014 ++vcpu
->stat
.req_event
;
7015 kvm_apic_accept_events(vcpu
);
7016 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
7021 if (inject_pending_event(vcpu
, req_int_win
) != 0)
7022 req_immediate_exit
= true;
7024 /* Enable NMI/IRQ window open exits if needed.
7026 * SMIs have two cases: 1) they can be nested, and
7027 * then there is nothing to do here because RSM will
7028 * cause a vmexit anyway; 2) or the SMI can be pending
7029 * because inject_pending_event has completed the
7030 * injection of an IRQ or NMI from the previous vmexit,
7031 * and then we request an immediate exit to inject the SMI.
7033 if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
))
7034 req_immediate_exit
= true;
7035 if (vcpu
->arch
.nmi_pending
)
7036 kvm_x86_ops
->enable_nmi_window(vcpu
);
7037 if (kvm_cpu_has_injectable_intr(vcpu
) || req_int_win
)
7038 kvm_x86_ops
->enable_irq_window(vcpu
);
7039 WARN_ON(vcpu
->arch
.exception
.pending
);
7042 if (kvm_lapic_enabled(vcpu
)) {
7043 update_cr8_intercept(vcpu
);
7044 kvm_lapic_sync_to_vapic(vcpu
);
7048 r
= kvm_mmu_reload(vcpu
);
7050 goto cancel_injection
;
7055 kvm_x86_ops
->prepare_guest_switch(vcpu
);
7056 kvm_load_guest_fpu(vcpu
);
7059 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7060 * IPI are then delayed after guest entry, which ensures that they
7061 * result in virtual interrupt delivery.
7063 local_irq_disable();
7064 vcpu
->mode
= IN_GUEST_MODE
;
7066 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
7069 * 1) We should set ->mode before checking ->requests. Please see
7070 * the comment in kvm_vcpu_exiting_guest_mode().
7072 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7073 * pairs with the memory barrier implicit in pi_test_and_set_on
7074 * (see vmx_deliver_posted_interrupt).
7076 * 3) This also orders the write to mode from any reads to the page
7077 * tables done while the VCPU is running. Please see the comment
7078 * in kvm_flush_remote_tlbs.
7080 smp_mb__after_srcu_read_unlock();
7083 * This handles the case where a posted interrupt was
7084 * notified with kvm_vcpu_kick.
7086 if (kvm_lapic_enabled(vcpu
)) {
7087 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
7088 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
7091 if (vcpu
->mode
== EXITING_GUEST_MODE
|| kvm_request_pending(vcpu
)
7092 || need_resched() || signal_pending(current
)) {
7093 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
7097 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7099 goto cancel_injection
;
7102 kvm_load_guest_xcr0(vcpu
);
7104 if (req_immediate_exit
) {
7105 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7106 smp_send_reschedule(vcpu
->cpu
);
7109 trace_kvm_entry(vcpu
->vcpu_id
);
7110 wait_lapic_expire(vcpu
);
7111 guest_enter_irqoff();
7113 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
7115 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
7116 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
7117 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
7118 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
7119 set_debugreg(vcpu
->arch
.dr6
, 6);
7120 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
7123 kvm_x86_ops
->run(vcpu
);
7126 * Do this here before restoring debug registers on the host. And
7127 * since we do this before handling the vmexit, a DR access vmexit
7128 * can (a) read the correct value of the debug registers, (b) set
7129 * KVM_DEBUGREG_WONT_EXIT again.
7131 if (unlikely(vcpu
->arch
.switch_db_regs
& KVM_DEBUGREG_WONT_EXIT
)) {
7132 WARN_ON(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
);
7133 kvm_x86_ops
->sync_dirty_debug_regs(vcpu
);
7134 kvm_update_dr0123(vcpu
);
7135 kvm_update_dr6(vcpu
);
7136 kvm_update_dr7(vcpu
);
7137 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
7141 * If the guest has used debug registers, at least dr7
7142 * will be disabled while returning to the host.
7143 * If we don't have active breakpoints in the host, we don't
7144 * care about the messed up debug address registers. But if
7145 * we have some of them active, restore the old state.
7147 if (hw_breakpoint_active())
7148 hw_breakpoint_restore();
7150 vcpu
->arch
.last_guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
7152 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
7155 kvm_put_guest_xcr0(vcpu
);
7157 kvm_x86_ops
->handle_external_intr(vcpu
);
7161 guest_exit_irqoff();
7166 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7169 * Profile KVM exit RIPs:
7171 if (unlikely(prof_on
== KVM_PROFILING
)) {
7172 unsigned long rip
= kvm_rip_read(vcpu
);
7173 profile_hit(KVM_PROFILING
, (void *)rip
);
7176 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
7177 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
7179 if (vcpu
->arch
.apic_attention
)
7180 kvm_lapic_sync_from_vapic(vcpu
);
7182 vcpu
->arch
.gpa_available
= false;
7183 r
= kvm_x86_ops
->handle_exit(vcpu
);
7187 kvm_x86_ops
->cancel_injection(vcpu
);
7188 if (unlikely(vcpu
->arch
.apic_attention
))
7189 kvm_lapic_sync_from_vapic(vcpu
);
7194 static inline int vcpu_block(struct kvm
*kvm
, struct kvm_vcpu
*vcpu
)
7196 if (!kvm_arch_vcpu_runnable(vcpu
) &&
7197 (!kvm_x86_ops
->pre_block
|| kvm_x86_ops
->pre_block(vcpu
) == 0)) {
7198 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7199 kvm_vcpu_block(vcpu
);
7200 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7202 if (kvm_x86_ops
->post_block
)
7203 kvm_x86_ops
->post_block(vcpu
);
7205 if (!kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
7209 kvm_apic_accept_events(vcpu
);
7210 switch(vcpu
->arch
.mp_state
) {
7211 case KVM_MP_STATE_HALTED
:
7212 vcpu
->arch
.pv
.pv_unhalted
= false;
7213 vcpu
->arch
.mp_state
=
7214 KVM_MP_STATE_RUNNABLE
;
7215 case KVM_MP_STATE_RUNNABLE
:
7216 vcpu
->arch
.apf
.halted
= false;
7218 case KVM_MP_STATE_INIT_RECEIVED
:
7227 static inline bool kvm_vcpu_running(struct kvm_vcpu
*vcpu
)
7229 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
)
7230 kvm_x86_ops
->check_nested_events(vcpu
, false);
7232 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
7233 !vcpu
->arch
.apf
.halted
);
7236 static int vcpu_run(struct kvm_vcpu
*vcpu
)
7239 struct kvm
*kvm
= vcpu
->kvm
;
7241 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7242 vcpu
->arch
.l1tf_flush_l1d
= true;
7245 if (kvm_vcpu_running(vcpu
)) {
7246 r
= vcpu_enter_guest(vcpu
);
7248 r
= vcpu_block(kvm
, vcpu
);
7254 kvm_clear_request(KVM_REQ_PENDING_TIMER
, vcpu
);
7255 if (kvm_cpu_has_pending_timer(vcpu
))
7256 kvm_inject_pending_timer_irqs(vcpu
);
7258 if (dm_request_for_irq_injection(vcpu
) &&
7259 kvm_vcpu_ready_for_interrupt_injection(vcpu
)) {
7261 vcpu
->run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
7262 ++vcpu
->stat
.request_irq_exits
;
7266 kvm_check_async_pf_completion(vcpu
);
7268 if (signal_pending(current
)) {
7270 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
7271 ++vcpu
->stat
.signal_exits
;
7274 if (need_resched()) {
7275 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7277 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7281 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7286 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
7289 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7290 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
7291 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
7292 if (r
!= EMULATE_DONE
)
7297 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
7299 BUG_ON(!vcpu
->arch
.pio
.count
);
7301 return complete_emulated_io(vcpu
);
7305 * Implements the following, as a state machine:
7309 * for each mmio piece in the fragment
7317 * for each mmio piece in the fragment
7322 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
7324 struct kvm_run
*run
= vcpu
->run
;
7325 struct kvm_mmio_fragment
*frag
;
7328 BUG_ON(!vcpu
->mmio_needed
);
7330 /* Complete previous fragment */
7331 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
7332 len
= min(8u, frag
->len
);
7333 if (!vcpu
->mmio_is_write
)
7334 memcpy(frag
->data
, run
->mmio
.data
, len
);
7336 if (frag
->len
<= 8) {
7337 /* Switch to the next fragment. */
7339 vcpu
->mmio_cur_fragment
++;
7341 /* Go forward to the next mmio piece. */
7347 if (vcpu
->mmio_cur_fragment
>= vcpu
->mmio_nr_fragments
) {
7348 vcpu
->mmio_needed
= 0;
7350 /* FIXME: return into emulator if single-stepping. */
7351 if (vcpu
->mmio_is_write
)
7353 vcpu
->mmio_read_completed
= 1;
7354 return complete_emulated_io(vcpu
);
7357 run
->exit_reason
= KVM_EXIT_MMIO
;
7358 run
->mmio
.phys_addr
= frag
->gpa
;
7359 if (vcpu
->mmio_is_write
)
7360 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
7361 run
->mmio
.len
= min(8u, frag
->len
);
7362 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
7363 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
7368 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
7370 struct fpu
*fpu
= ¤t
->thread
.fpu
;
7373 fpu__initialize(fpu
);
7375 kvm_sigset_activate(vcpu
);
7377 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
7378 if (kvm_run
->immediate_exit
) {
7382 kvm_vcpu_block(vcpu
);
7383 kvm_apic_accept_events(vcpu
);
7384 kvm_clear_request(KVM_REQ_UNHALT
, vcpu
);
7386 if (signal_pending(current
)) {
7388 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
7389 ++vcpu
->stat
.signal_exits
;
7394 /* re-sync apic's tpr */
7395 if (!lapic_in_kernel(vcpu
)) {
7396 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
7402 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
7403 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
7404 vcpu
->arch
.complete_userspace_io
= NULL
;
7409 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
7411 if (kvm_run
->immediate_exit
)
7417 post_kvm_run_save(vcpu
);
7418 kvm_sigset_deactivate(vcpu
);
7423 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
7425 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
7427 * We are here if userspace calls get_regs() in the middle of
7428 * instruction emulation. Registers state needs to be copied
7429 * back from emulation context to vcpu. Userspace shouldn't do
7430 * that usually, but some bad designed PV devices (vmware
7431 * backdoor interface) need this to work
7433 emulator_writeback_register_cache(&vcpu
->arch
.emulate_ctxt
);
7434 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7436 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
7437 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
7438 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
7439 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
7440 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
7441 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
7442 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
7443 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
7444 #ifdef CONFIG_X86_64
7445 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
7446 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
7447 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
7448 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
7449 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
7450 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
7451 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
7452 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
7455 regs
->rip
= kvm_rip_read(vcpu
);
7456 regs
->rflags
= kvm_get_rflags(vcpu
);
7461 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
7463 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
7464 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7466 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
7467 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
7468 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
7469 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
7470 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
7471 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
7472 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
7473 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
7474 #ifdef CONFIG_X86_64
7475 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
7476 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
7477 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
7478 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
7479 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
7480 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
7481 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
7482 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
7485 kvm_rip_write(vcpu
, regs
->rip
);
7486 kvm_set_rflags(vcpu
, regs
->rflags
| X86_EFLAGS_FIXED
);
7488 vcpu
->arch
.exception
.pending
= false;
7490 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7495 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
7497 struct kvm_segment cs
;
7499 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7503 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
7505 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
7506 struct kvm_sregs
*sregs
)
7510 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7511 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7512 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7513 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7514 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7515 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7517 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7518 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7520 kvm_x86_ops
->get_idt(vcpu
, &dt
);
7521 sregs
->idt
.limit
= dt
.size
;
7522 sregs
->idt
.base
= dt
.address
;
7523 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
7524 sregs
->gdt
.limit
= dt
.size
;
7525 sregs
->gdt
.base
= dt
.address
;
7527 sregs
->cr0
= kvm_read_cr0(vcpu
);
7528 sregs
->cr2
= vcpu
->arch
.cr2
;
7529 sregs
->cr3
= kvm_read_cr3(vcpu
);
7530 sregs
->cr4
= kvm_read_cr4(vcpu
);
7531 sregs
->cr8
= kvm_get_cr8(vcpu
);
7532 sregs
->efer
= vcpu
->arch
.efer
;
7533 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
7535 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
7537 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
7538 set_bit(vcpu
->arch
.interrupt
.nr
,
7539 (unsigned long *)sregs
->interrupt_bitmap
);
7544 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
7545 struct kvm_mp_state
*mp_state
)
7547 kvm_apic_accept_events(vcpu
);
7548 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_HALTED
&&
7549 vcpu
->arch
.pv
.pv_unhalted
)
7550 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
7552 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
7557 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
7558 struct kvm_mp_state
*mp_state
)
7560 if (!lapic_in_kernel(vcpu
) &&
7561 mp_state
->mp_state
!= KVM_MP_STATE_RUNNABLE
)
7564 /* INITs are latched while in SMM */
7565 if ((is_smm(vcpu
) || vcpu
->arch
.smi_pending
) &&
7566 (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
||
7567 mp_state
->mp_state
== KVM_MP_STATE_INIT_RECEIVED
))
7570 if (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
) {
7571 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
7572 set_bit(KVM_APIC_SIPI
, &vcpu
->arch
.apic
->pending_events
);
7574 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
7575 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7579 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
7580 int reason
, bool has_error_code
, u32 error_code
)
7582 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
7585 init_emulate_ctxt(vcpu
);
7587 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
7588 has_error_code
, error_code
);
7591 return EMULATE_FAIL
;
7593 kvm_rip_write(vcpu
, ctxt
->eip
);
7594 kvm_set_rflags(vcpu
, ctxt
->eflags
);
7595 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7596 return EMULATE_DONE
;
7598 EXPORT_SYMBOL_GPL(kvm_task_switch
);
7600 int kvm_valid_sregs(struct kvm_vcpu
*vcpu
, struct kvm_sregs
*sregs
)
7602 if ((sregs
->efer
& EFER_LME
) && (sregs
->cr0
& X86_CR0_PG
)) {
7604 * When EFER.LME and CR0.PG are set, the processor is in
7605 * 64-bit mode (though maybe in a 32-bit code segment).
7606 * CR4.PAE and EFER.LMA must be set.
7608 if (!(sregs
->cr4
& X86_CR4_PAE
)
7609 || !(sregs
->efer
& EFER_LMA
))
7613 * Not in 64-bit mode: EFER.LMA is clear and the code
7614 * segment cannot be 64-bit.
7616 if (sregs
->efer
& EFER_LMA
|| sregs
->cs
.l
)
7623 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
7624 struct kvm_sregs
*sregs
)
7626 struct msr_data apic_base_msr
;
7627 int mmu_reset_needed
= 0;
7628 int cpuid_update_needed
= 0;
7629 int pending_vec
, max_bits
, idx
;
7632 if (!guest_cpuid_has(vcpu
, X86_FEATURE_XSAVE
) &&
7633 (sregs
->cr4
& X86_CR4_OSXSAVE
))
7636 if (kvm_valid_sregs(vcpu
, sregs
))
7639 apic_base_msr
.data
= sregs
->apic_base
;
7640 apic_base_msr
.host_initiated
= true;
7641 if (kvm_set_apic_base(vcpu
, &apic_base_msr
))
7644 dt
.size
= sregs
->idt
.limit
;
7645 dt
.address
= sregs
->idt
.base
;
7646 kvm_x86_ops
->set_idt(vcpu
, &dt
);
7647 dt
.size
= sregs
->gdt
.limit
;
7648 dt
.address
= sregs
->gdt
.base
;
7649 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
7651 vcpu
->arch
.cr2
= sregs
->cr2
;
7652 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
7653 vcpu
->arch
.cr3
= sregs
->cr3
;
7654 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
7656 kvm_set_cr8(vcpu
, sregs
->cr8
);
7658 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
7659 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
7661 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
7662 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
7663 vcpu
->arch
.cr0
= sregs
->cr0
;
7665 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
7666 cpuid_update_needed
|= ((kvm_read_cr4(vcpu
) ^ sregs
->cr4
) &
7667 (X86_CR4_OSXSAVE
| X86_CR4_PKE
));
7668 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
7669 if (cpuid_update_needed
)
7670 kvm_update_cpuid(vcpu
);
7672 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7673 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
7674 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
7675 mmu_reset_needed
= 1;
7677 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7679 if (mmu_reset_needed
)
7680 kvm_mmu_reset_context(vcpu
);
7682 max_bits
= KVM_NR_INTERRUPTS
;
7683 pending_vec
= find_first_bit(
7684 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
7685 if (pending_vec
< max_bits
) {
7686 kvm_queue_interrupt(vcpu
, pending_vec
, false);
7687 pr_debug("Set back pending irq %d\n", pending_vec
);
7690 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7691 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7692 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7693 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7694 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7695 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7697 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7698 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7700 update_cr8_intercept(vcpu
);
7702 /* Older userspace won't unhalt the vcpu on reset. */
7703 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
7704 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
7706 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7708 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7713 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
7714 struct kvm_guest_debug
*dbg
)
7716 unsigned long rflags
;
7719 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
7721 if (vcpu
->arch
.exception
.pending
)
7723 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
7724 kvm_queue_exception(vcpu
, DB_VECTOR
);
7726 kvm_queue_exception(vcpu
, BP_VECTOR
);
7730 * Read rflags as long as potentially injected trace flags are still
7733 rflags
= kvm_get_rflags(vcpu
);
7735 vcpu
->guest_debug
= dbg
->control
;
7736 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
7737 vcpu
->guest_debug
= 0;
7739 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
7740 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
7741 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
7742 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
7744 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
7745 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
7747 kvm_update_dr7(vcpu
);
7749 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
7750 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
7751 get_segment_base(vcpu
, VCPU_SREG_CS
);
7754 * Trigger an rflags update that will inject or remove the trace
7757 kvm_set_rflags(vcpu
, rflags
);
7759 kvm_x86_ops
->update_bp_intercept(vcpu
);
7769 * Translate a guest virtual address to a guest physical address.
7771 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
7772 struct kvm_translation
*tr
)
7774 unsigned long vaddr
= tr
->linear_address
;
7778 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7779 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
7780 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7781 tr
->physical_address
= gpa
;
7782 tr
->valid
= gpa
!= UNMAPPED_GVA
;
7789 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7791 struct fxregs_state
*fxsave
=
7792 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7794 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
7795 fpu
->fcw
= fxsave
->cwd
;
7796 fpu
->fsw
= fxsave
->swd
;
7797 fpu
->ftwx
= fxsave
->twd
;
7798 fpu
->last_opcode
= fxsave
->fop
;
7799 fpu
->last_ip
= fxsave
->rip
;
7800 fpu
->last_dp
= fxsave
->rdp
;
7801 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
7806 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7808 struct fxregs_state
*fxsave
=
7809 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7811 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
7812 fxsave
->cwd
= fpu
->fcw
;
7813 fxsave
->swd
= fpu
->fsw
;
7814 fxsave
->twd
= fpu
->ftwx
;
7815 fxsave
->fop
= fpu
->last_opcode
;
7816 fxsave
->rip
= fpu
->last_ip
;
7817 fxsave
->rdp
= fpu
->last_dp
;
7818 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
7823 static void fx_init(struct kvm_vcpu
*vcpu
)
7825 fpstate_init(&vcpu
->arch
.guest_fpu
.state
);
7826 if (boot_cpu_has(X86_FEATURE_XSAVES
))
7827 vcpu
->arch
.guest_fpu
.state
.xsave
.header
.xcomp_bv
=
7828 host_xcr0
| XSTATE_COMPACTION_ENABLED
;
7831 * Ensure guest xcr0 is valid for loading
7833 vcpu
->arch
.xcr0
= XFEATURE_MASK_FP
;
7835 vcpu
->arch
.cr0
|= X86_CR0_ET
;
7838 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
7840 if (vcpu
->guest_fpu_loaded
)
7844 * Restore all possible states in the guest,
7845 * and assume host would use all available bits.
7846 * Guest xcr0 would be loaded later.
7848 vcpu
->guest_fpu_loaded
= 1;
7849 __kernel_fpu_begin();
7850 /* PKRU is separately restored in kvm_x86_ops->run. */
7851 __copy_kernel_to_fpregs(&vcpu
->arch
.guest_fpu
.state
,
7852 ~XFEATURE_MASK_PKRU
);
7856 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
7858 if (!vcpu
->guest_fpu_loaded
)
7861 vcpu
->guest_fpu_loaded
= 0;
7862 copy_fpregs_to_fpstate(&vcpu
->arch
.guest_fpu
);
7864 ++vcpu
->stat
.fpu_reload
;
7868 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
7870 void *wbinvd_dirty_mask
= vcpu
->arch
.wbinvd_dirty_mask
;
7872 kvmclock_reset(vcpu
);
7874 kvm_x86_ops
->vcpu_free(vcpu
);
7875 free_cpumask_var(wbinvd_dirty_mask
);
7878 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
7881 struct kvm_vcpu
*vcpu
;
7883 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
7884 printk_once(KERN_WARNING
7885 "kvm: SMP vm created on host with unstable TSC; "
7886 "guest TSC will not be reliable\n");
7888 vcpu
= kvm_x86_ops
->vcpu_create(kvm
, id
);
7893 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
7897 kvm_vcpu_mtrr_init(vcpu
);
7898 r
= vcpu_load(vcpu
);
7901 kvm_vcpu_reset(vcpu
, false);
7902 kvm_mmu_setup(vcpu
);
7907 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
7909 struct msr_data msr
;
7910 struct kvm
*kvm
= vcpu
->kvm
;
7912 kvm_hv_vcpu_postcreate(vcpu
);
7914 if (vcpu_load(vcpu
))
7917 msr
.index
= MSR_IA32_TSC
;
7918 msr
.host_initiated
= true;
7919 kvm_write_tsc(vcpu
, &msr
);
7922 if (!kvmclock_periodic_sync
)
7925 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
7926 KVMCLOCK_SYNC_PERIOD
);
7929 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
7932 vcpu
->arch
.apf
.msr_val
= 0;
7934 r
= vcpu_load(vcpu
);
7936 kvm_mmu_unload(vcpu
);
7939 kvm_x86_ops
->vcpu_free(vcpu
);
7942 void kvm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
7944 kvm_lapic_reset(vcpu
, init_event
);
7946 vcpu
->arch
.hflags
= 0;
7948 vcpu
->arch
.smi_pending
= 0;
7949 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
7950 vcpu
->arch
.nmi_pending
= 0;
7951 vcpu
->arch
.nmi_injected
= false;
7952 kvm_clear_interrupt_queue(vcpu
);
7953 kvm_clear_exception_queue(vcpu
);
7954 vcpu
->arch
.exception
.pending
= false;
7956 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
7957 kvm_update_dr0123(vcpu
);
7958 vcpu
->arch
.dr6
= DR6_INIT
;
7959 kvm_update_dr6(vcpu
);
7960 vcpu
->arch
.dr7
= DR7_FIXED_1
;
7961 kvm_update_dr7(vcpu
);
7965 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7966 vcpu
->arch
.apf
.msr_val
= 0;
7967 vcpu
->arch
.st
.msr_val
= 0;
7969 kvmclock_reset(vcpu
);
7971 kvm_clear_async_pf_completion_queue(vcpu
);
7972 kvm_async_pf_hash_reset(vcpu
);
7973 vcpu
->arch
.apf
.halted
= false;
7976 kvm_pmu_reset(vcpu
);
7977 vcpu
->arch
.smbase
= 0x30000;
7979 vcpu
->arch
.msr_platform_info
= MSR_PLATFORM_INFO_CPUID_FAULT
;
7980 vcpu
->arch
.msr_misc_features_enables
= 0;
7983 memset(vcpu
->arch
.regs
, 0, sizeof(vcpu
->arch
.regs
));
7984 vcpu
->arch
.regs_avail
= ~0;
7985 vcpu
->arch
.regs_dirty
= ~0;
7987 kvm_x86_ops
->vcpu_reset(vcpu
, init_event
);
7990 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu
*vcpu
, u8 vector
)
7992 struct kvm_segment cs
;
7994 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7995 cs
.selector
= vector
<< 8;
7996 cs
.base
= vector
<< 12;
7997 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7998 kvm_rip_write(vcpu
, 0);
8001 int kvm_arch_hardware_enable(void)
8004 struct kvm_vcpu
*vcpu
;
8009 bool stable
, backwards_tsc
= false;
8011 kvm_shared_msr_cpu_online();
8012 ret
= kvm_x86_ops
->hardware_enable();
8016 local_tsc
= rdtsc();
8017 stable
= !check_tsc_unstable();
8018 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
8019 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8020 if (!stable
&& vcpu
->cpu
== smp_processor_id())
8021 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
8022 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
8023 backwards_tsc
= true;
8024 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
8025 max_tsc
= vcpu
->arch
.last_host_tsc
;
8031 * Sometimes, even reliable TSCs go backwards. This happens on
8032 * platforms that reset TSC during suspend or hibernate actions, but
8033 * maintain synchronization. We must compensate. Fortunately, we can
8034 * detect that condition here, which happens early in CPU bringup,
8035 * before any KVM threads can be running. Unfortunately, we can't
8036 * bring the TSCs fully up to date with real time, as we aren't yet far
8037 * enough into CPU bringup that we know how much real time has actually
8038 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8039 * variables that haven't been updated yet.
8041 * So we simply find the maximum observed TSC above, then record the
8042 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8043 * the adjustment will be applied. Note that we accumulate
8044 * adjustments, in case multiple suspend cycles happen before some VCPU
8045 * gets a chance to run again. In the event that no KVM threads get a
8046 * chance to run, we will miss the entire elapsed period, as we'll have
8047 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8048 * loose cycle time. This isn't too big a deal, since the loss will be
8049 * uniform across all VCPUs (not to mention the scenario is extremely
8050 * unlikely). It is possible that a second hibernate recovery happens
8051 * much faster than a first, causing the observed TSC here to be
8052 * smaller; this would require additional padding adjustment, which is
8053 * why we set last_host_tsc to the local tsc observed here.
8055 * N.B. - this code below runs only on platforms with reliable TSC,
8056 * as that is the only way backwards_tsc is set above. Also note
8057 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8058 * have the same delta_cyc adjustment applied if backwards_tsc
8059 * is detected. Note further, this adjustment is only done once,
8060 * as we reset last_host_tsc on all VCPUs to stop this from being
8061 * called multiple times (one for each physical CPU bringup).
8063 * Platforms with unreliable TSCs don't have to deal with this, they
8064 * will be compensated by the logic in vcpu_load, which sets the TSC to
8065 * catchup mode. This will catchup all VCPUs to real time, but cannot
8066 * guarantee that they stay in perfect synchronization.
8068 if (backwards_tsc
) {
8069 u64 delta_cyc
= max_tsc
- local_tsc
;
8070 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
8071 kvm
->arch
.backwards_tsc_observed
= true;
8072 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8073 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
8074 vcpu
->arch
.last_host_tsc
= local_tsc
;
8075 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
8079 * We have to disable TSC offset matching.. if you were
8080 * booting a VM while issuing an S4 host suspend....
8081 * you may have some problem. Solving this issue is
8082 * left as an exercise to the reader.
8084 kvm
->arch
.last_tsc_nsec
= 0;
8085 kvm
->arch
.last_tsc_write
= 0;
8092 void kvm_arch_hardware_disable(void)
8094 kvm_x86_ops
->hardware_disable();
8095 drop_user_return_notifiers();
8098 int kvm_arch_hardware_setup(void)
8102 r
= kvm_x86_ops
->hardware_setup();
8106 if (kvm_has_tsc_control
) {
8108 * Make sure the user can only configure tsc_khz values that
8109 * fit into a signed integer.
8110 * A min value is not calculated needed because it will always
8111 * be 1 on all machines.
8113 u64 max
= min(0x7fffffffULL
,
8114 __scale_tsc(kvm_max_tsc_scaling_ratio
, tsc_khz
));
8115 kvm_max_guest_tsc_khz
= max
;
8117 kvm_default_tsc_scaling_ratio
= 1ULL << kvm_tsc_scaling_ratio_frac_bits
;
8120 kvm_init_msr_list();
8124 void kvm_arch_hardware_unsetup(void)
8126 kvm_x86_ops
->hardware_unsetup();
8129 void kvm_arch_check_processor_compat(void *rtn
)
8131 kvm_x86_ops
->check_processor_compatibility(rtn
);
8134 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu
*vcpu
)
8136 return vcpu
->kvm
->arch
.bsp_vcpu_id
== vcpu
->vcpu_id
;
8138 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp
);
8140 bool kvm_vcpu_is_bsp(struct kvm_vcpu
*vcpu
)
8142 return (vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_BSP
) != 0;
8145 struct static_key kvm_no_apic_vcpu __read_mostly
;
8146 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu
);
8148 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
8154 BUG_ON(vcpu
->kvm
== NULL
);
8157 vcpu
->arch
.apicv_active
= kvm_x86_ops
->get_enable_apicv(vcpu
);
8158 vcpu
->arch
.pv
.pv_unhalted
= false;
8159 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
8160 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_reset_bsp(vcpu
))
8161 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
8163 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
8165 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
8170 vcpu
->arch
.pio_data
= page_address(page
);
8172 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
8174 r
= kvm_mmu_create(vcpu
);
8176 goto fail_free_pio_data
;
8178 if (irqchip_in_kernel(kvm
)) {
8179 r
= kvm_create_lapic(vcpu
);
8181 goto fail_mmu_destroy
;
8183 static_key_slow_inc(&kvm_no_apic_vcpu
);
8185 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
8187 if (!vcpu
->arch
.mce_banks
) {
8189 goto fail_free_lapic
;
8191 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
8193 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
)) {
8195 goto fail_free_mce_banks
;
8200 vcpu
->arch
.ia32_tsc_adjust_msr
= 0x0;
8201 vcpu
->arch
.pv_time_enabled
= false;
8203 vcpu
->arch
.guest_supported_xcr0
= 0;
8204 vcpu
->arch
.guest_xstate_size
= XSAVE_HDR_SIZE
+ XSAVE_HDR_OFFSET
;
8206 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
8208 vcpu
->arch
.pat
= MSR_IA32_CR_PAT_DEFAULT
;
8210 kvm_async_pf_hash_reset(vcpu
);
8213 vcpu
->arch
.pending_external_vector
= -1;
8214 vcpu
->arch
.preempted_in_kernel
= false;
8216 kvm_hv_vcpu_init(vcpu
);
8220 fail_free_mce_banks
:
8221 kfree(vcpu
->arch
.mce_banks
);
8223 kvm_free_lapic(vcpu
);
8225 kvm_mmu_destroy(vcpu
);
8227 free_page((unsigned long)vcpu
->arch
.pio_data
);
8232 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
8236 kvm_hv_vcpu_uninit(vcpu
);
8237 kvm_pmu_destroy(vcpu
);
8238 kfree(vcpu
->arch
.mce_banks
);
8239 kvm_free_lapic(vcpu
);
8240 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
8241 kvm_mmu_destroy(vcpu
);
8242 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
8243 free_page((unsigned long)vcpu
->arch
.pio_data
);
8244 if (!lapic_in_kernel(vcpu
))
8245 static_key_slow_dec(&kvm_no_apic_vcpu
);
8248 void kvm_arch_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
8250 vcpu
->arch
.l1tf_flush_l1d
= true;
8251 kvm_x86_ops
->sched_in(vcpu
, cpu
);
8254 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
8259 INIT_HLIST_HEAD(&kvm
->arch
.mask_notifier_list
);
8260 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
8261 INIT_LIST_HEAD(&kvm
->arch
.zapped_obsolete_pages
);
8262 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
8263 atomic_set(&kvm
->arch
.noncoherent_dma_count
, 0);
8265 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8266 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
8267 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8268 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
8269 &kvm
->arch
.irq_sources_bitmap
);
8271 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
8272 mutex_init(&kvm
->arch
.apic_map_lock
);
8273 mutex_init(&kvm
->arch
.hyperv
.hv_lock
);
8274 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
8276 kvm
->arch
.kvmclock_offset
= -ktime_get_boot_ns();
8277 pvclock_update_vm_gtod_copy(kvm
);
8279 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_update_work
, kvmclock_update_fn
);
8280 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_sync_work
, kvmclock_sync_fn
);
8282 kvm_page_track_init(kvm
);
8283 kvm_mmu_init_vm(kvm
);
8285 if (kvm_x86_ops
->vm_init
)
8286 return kvm_x86_ops
->vm_init(kvm
);
8291 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
8294 r
= vcpu_load(vcpu
);
8296 kvm_mmu_unload(vcpu
);
8300 static void kvm_free_vcpus(struct kvm
*kvm
)
8303 struct kvm_vcpu
*vcpu
;
8306 * Unpin any mmu pages first.
8308 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8309 kvm_clear_async_pf_completion_queue(vcpu
);
8310 kvm_unload_vcpu_mmu(vcpu
);
8312 kvm_for_each_vcpu(i
, vcpu
, kvm
)
8313 kvm_arch_vcpu_free(vcpu
);
8315 mutex_lock(&kvm
->lock
);
8316 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
8317 kvm
->vcpus
[i
] = NULL
;
8319 atomic_set(&kvm
->online_vcpus
, 0);
8320 mutex_unlock(&kvm
->lock
);
8323 void kvm_arch_sync_events(struct kvm
*kvm
)
8325 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_sync_work
);
8326 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_update_work
);
8330 int __x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
8334 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
8335 struct kvm_memory_slot
*slot
, old
;
8337 /* Called with kvm->slots_lock held. */
8338 if (WARN_ON(id
>= KVM_MEM_SLOTS_NUM
))
8341 slot
= id_to_memslot(slots
, id
);
8347 * MAP_SHARED to prevent internal slot pages from being moved
8350 hva
= vm_mmap(NULL
, 0, size
, PROT_READ
| PROT_WRITE
,
8351 MAP_SHARED
| MAP_ANONYMOUS
, 0);
8352 if (IS_ERR((void *)hva
))
8353 return PTR_ERR((void *)hva
);
8362 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
8363 struct kvm_userspace_memory_region m
;
8365 m
.slot
= id
| (i
<< 16);
8367 m
.guest_phys_addr
= gpa
;
8368 m
.userspace_addr
= hva
;
8369 m
.memory_size
= size
;
8370 r
= __kvm_set_memory_region(kvm
, &m
);
8376 vm_munmap(old
.userspace_addr
, old
.npages
* PAGE_SIZE
);
8380 EXPORT_SYMBOL_GPL(__x86_set_memory_region
);
8382 int x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
8386 mutex_lock(&kvm
->slots_lock
);
8387 r
= __x86_set_memory_region(kvm
, id
, gpa
, size
);
8388 mutex_unlock(&kvm
->slots_lock
);
8392 EXPORT_SYMBOL_GPL(x86_set_memory_region
);
8394 void kvm_arch_destroy_vm(struct kvm
*kvm
)
8396 if (current
->mm
== kvm
->mm
) {
8398 * Free memory regions allocated on behalf of userspace,
8399 * unless the the memory map has changed due to process exit
8402 x86_set_memory_region(kvm
, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
, 0, 0);
8403 x86_set_memory_region(kvm
, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
, 0, 0);
8404 x86_set_memory_region(kvm
, TSS_PRIVATE_MEMSLOT
, 0, 0);
8406 if (kvm_x86_ops
->vm_destroy
)
8407 kvm_x86_ops
->vm_destroy(kvm
);
8408 kvm_pic_destroy(kvm
);
8409 kvm_ioapic_destroy(kvm
);
8410 kvm_free_vcpus(kvm
);
8411 kvfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
8412 kvm_mmu_uninit_vm(kvm
);
8413 kvm_page_track_cleanup(kvm
);
8416 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
8417 struct kvm_memory_slot
*dont
)
8421 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8422 if (!dont
|| free
->arch
.rmap
[i
] != dont
->arch
.rmap
[i
]) {
8423 kvfree(free
->arch
.rmap
[i
]);
8424 free
->arch
.rmap
[i
] = NULL
;
8429 if (!dont
|| free
->arch
.lpage_info
[i
- 1] !=
8430 dont
->arch
.lpage_info
[i
- 1]) {
8431 kvfree(free
->arch
.lpage_info
[i
- 1]);
8432 free
->arch
.lpage_info
[i
- 1] = NULL
;
8436 kvm_page_track_free_memslot(free
, dont
);
8439 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
8440 unsigned long npages
)
8444 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8445 struct kvm_lpage_info
*linfo
;
8450 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
8451 slot
->base_gfn
, level
) + 1;
8453 slot
->arch
.rmap
[i
] =
8454 kvzalloc(lpages
* sizeof(*slot
->arch
.rmap
[i
]), GFP_KERNEL
);
8455 if (!slot
->arch
.rmap
[i
])
8460 linfo
= kvzalloc(lpages
* sizeof(*linfo
), GFP_KERNEL
);
8464 slot
->arch
.lpage_info
[i
- 1] = linfo
;
8466 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
8467 linfo
[0].disallow_lpage
= 1;
8468 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
8469 linfo
[lpages
- 1].disallow_lpage
= 1;
8470 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
8472 * If the gfn and userspace address are not aligned wrt each
8473 * other, or if explicitly asked to, disable large page
8474 * support for this slot
8476 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
8477 !kvm_largepages_enabled()) {
8480 for (j
= 0; j
< lpages
; ++j
)
8481 linfo
[j
].disallow_lpage
= 1;
8485 if (kvm_page_track_create_memslot(slot
, npages
))
8491 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8492 kvfree(slot
->arch
.rmap
[i
]);
8493 slot
->arch
.rmap
[i
] = NULL
;
8497 kvfree(slot
->arch
.lpage_info
[i
- 1]);
8498 slot
->arch
.lpage_info
[i
- 1] = NULL
;
8503 void kvm_arch_memslots_updated(struct kvm
*kvm
, struct kvm_memslots
*slots
)
8506 * memslots->generation has been incremented.
8507 * mmio generation may have reached its maximum value.
8509 kvm_mmu_invalidate_mmio_sptes(kvm
, slots
);
8512 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
8513 struct kvm_memory_slot
*memslot
,
8514 const struct kvm_userspace_memory_region
*mem
,
8515 enum kvm_mr_change change
)
8520 static void kvm_mmu_slot_apply_flags(struct kvm
*kvm
,
8521 struct kvm_memory_slot
*new)
8523 /* Still write protect RO slot */
8524 if (new->flags
& KVM_MEM_READONLY
) {
8525 kvm_mmu_slot_remove_write_access(kvm
, new);
8530 * Call kvm_x86_ops dirty logging hooks when they are valid.
8532 * kvm_x86_ops->slot_disable_log_dirty is called when:
8534 * - KVM_MR_CREATE with dirty logging is disabled
8535 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8537 * The reason is, in case of PML, we need to set D-bit for any slots
8538 * with dirty logging disabled in order to eliminate unnecessary GPA
8539 * logging in PML buffer (and potential PML buffer full VMEXT). This
8540 * guarantees leaving PML enabled during guest's lifetime won't have
8541 * any additonal overhead from PML when guest is running with dirty
8542 * logging disabled for memory slots.
8544 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8545 * to dirty logging mode.
8547 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8549 * In case of write protect:
8551 * Write protect all pages for dirty logging.
8553 * All the sptes including the large sptes which point to this
8554 * slot are set to readonly. We can not create any new large
8555 * spte on this slot until the end of the logging.
8557 * See the comments in fast_page_fault().
8559 if (new->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
8560 if (kvm_x86_ops
->slot_enable_log_dirty
)
8561 kvm_x86_ops
->slot_enable_log_dirty(kvm
, new);
8563 kvm_mmu_slot_remove_write_access(kvm
, new);
8565 if (kvm_x86_ops
->slot_disable_log_dirty
)
8566 kvm_x86_ops
->slot_disable_log_dirty(kvm
, new);
8570 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
8571 const struct kvm_userspace_memory_region
*mem
,
8572 const struct kvm_memory_slot
*old
,
8573 const struct kvm_memory_slot
*new,
8574 enum kvm_mr_change change
)
8576 int nr_mmu_pages
= 0;
8578 if (!kvm
->arch
.n_requested_mmu_pages
)
8579 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
8582 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
8585 * Dirty logging tracks sptes in 4k granularity, meaning that large
8586 * sptes have to be split. If live migration is successful, the guest
8587 * in the source machine will be destroyed and large sptes will be
8588 * created in the destination. However, if the guest continues to run
8589 * in the source machine (for example if live migration fails), small
8590 * sptes will remain around and cause bad performance.
8592 * Scan sptes if dirty logging has been stopped, dropping those
8593 * which can be collapsed into a single large-page spte. Later
8594 * page faults will create the large-page sptes.
8596 if ((change
!= KVM_MR_DELETE
) &&
8597 (old
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) &&
8598 !(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
8599 kvm_mmu_zap_collapsible_sptes(kvm
, new);
8602 * Set up write protection and/or dirty logging for the new slot.
8604 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8605 * been zapped so no dirty logging staff is needed for old slot. For
8606 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8607 * new and it's also covered when dealing with the new slot.
8609 * FIXME: const-ify all uses of struct kvm_memory_slot.
8611 if (change
!= KVM_MR_DELETE
)
8612 kvm_mmu_slot_apply_flags(kvm
, (struct kvm_memory_slot
*) new);
8615 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
8617 kvm_mmu_invalidate_zap_all_pages(kvm
);
8620 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
8621 struct kvm_memory_slot
*slot
)
8623 kvm_page_track_flush_slot(kvm
, slot
);
8626 static inline bool kvm_vcpu_has_events(struct kvm_vcpu
*vcpu
)
8628 if (!list_empty_careful(&vcpu
->async_pf
.done
))
8631 if (kvm_apic_has_events(vcpu
))
8634 if (vcpu
->arch
.pv
.pv_unhalted
)
8637 if (vcpu
->arch
.exception
.pending
)
8640 if (kvm_test_request(KVM_REQ_NMI
, vcpu
) ||
8641 (vcpu
->arch
.nmi_pending
&&
8642 kvm_x86_ops
->nmi_allowed(vcpu
)))
8645 if (kvm_test_request(KVM_REQ_SMI
, vcpu
) ||
8646 (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
)))
8649 if (kvm_arch_interrupt_allowed(vcpu
) &&
8650 kvm_cpu_has_interrupt(vcpu
))
8653 if (kvm_hv_has_stimer_pending(vcpu
))
8659 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
8661 return kvm_vcpu_running(vcpu
) || kvm_vcpu_has_events(vcpu
);
8664 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
8666 return vcpu
->arch
.preempted_in_kernel
;
8669 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
8671 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
8674 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
8676 return kvm_x86_ops
->interrupt_allowed(vcpu
);
8679 unsigned long kvm_get_linear_rip(struct kvm_vcpu
*vcpu
)
8681 if (is_64_bit_mode(vcpu
))
8682 return kvm_rip_read(vcpu
);
8683 return (u32
)(get_segment_base(vcpu
, VCPU_SREG_CS
) +
8684 kvm_rip_read(vcpu
));
8686 EXPORT_SYMBOL_GPL(kvm_get_linear_rip
);
8688 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
8690 return kvm_get_linear_rip(vcpu
) == linear_rip
;
8692 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
8694 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
8696 unsigned long rflags
;
8698 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
8699 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
8700 rflags
&= ~X86_EFLAGS_TF
;
8703 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
8705 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8707 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
8708 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
8709 rflags
|= X86_EFLAGS_TF
;
8710 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
8713 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8715 __kvm_set_rflags(vcpu
, rflags
);
8716 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8718 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
8720 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
8724 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
8728 r
= kvm_mmu_reload(vcpu
);
8732 if (!vcpu
->arch
.mmu
.direct_map
&&
8733 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
8736 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
8739 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
8741 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
8744 static inline u32
kvm_async_pf_next_probe(u32 key
)
8746 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
8749 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8751 u32 key
= kvm_async_pf_hash_fn(gfn
);
8753 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
8754 key
= kvm_async_pf_next_probe(key
);
8756 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
8759 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8762 u32 key
= kvm_async_pf_hash_fn(gfn
);
8764 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
8765 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
8766 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
8767 key
= kvm_async_pf_next_probe(key
);
8772 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8774 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
8777 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8781 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
8783 vcpu
->arch
.apf
.gfns
[i
] = ~0;
8785 j
= kvm_async_pf_next_probe(j
);
8786 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
8788 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
8790 * k lies cyclically in ]i,j]
8792 * |....j i.k.| or |.k..j i...|
8794 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
8795 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
8800 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
8803 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
8807 static int apf_get_user(struct kvm_vcpu
*vcpu
, u32
*val
)
8810 return kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, val
,
8814 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
8815 struct kvm_async_pf
*work
)
8817 struct x86_exception fault
;
8819 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
8820 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8822 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
8823 (vcpu
->arch
.apf
.send_user_only
&&
8824 kvm_x86_ops
->get_cpl(vcpu
) == 0))
8825 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
8826 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
8827 fault
.vector
= PF_VECTOR
;
8828 fault
.error_code_valid
= true;
8829 fault
.error_code
= 0;
8830 fault
.nested_page_fault
= false;
8831 fault
.address
= work
->arch
.token
;
8832 fault
.async_page_fault
= true;
8833 kvm_inject_page_fault(vcpu
, &fault
);
8837 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
8838 struct kvm_async_pf
*work
)
8840 struct x86_exception fault
;
8843 if (work
->wakeup_all
)
8844 work
->arch
.token
= ~0; /* broadcast wakeup */
8846 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8847 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
8849 if (vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
&&
8850 !apf_get_user(vcpu
, &val
)) {
8851 if (val
== KVM_PV_REASON_PAGE_NOT_PRESENT
&&
8852 vcpu
->arch
.exception
.pending
&&
8853 vcpu
->arch
.exception
.nr
== PF_VECTOR
&&
8854 !apf_put_user(vcpu
, 0)) {
8855 vcpu
->arch
.exception
.injected
= false;
8856 vcpu
->arch
.exception
.pending
= false;
8857 vcpu
->arch
.exception
.nr
= 0;
8858 vcpu
->arch
.exception
.has_error_code
= false;
8859 vcpu
->arch
.exception
.error_code
= 0;
8860 } else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
8861 fault
.vector
= PF_VECTOR
;
8862 fault
.error_code_valid
= true;
8863 fault
.error_code
= 0;
8864 fault
.nested_page_fault
= false;
8865 fault
.address
= work
->arch
.token
;
8866 fault
.async_page_fault
= true;
8867 kvm_inject_page_fault(vcpu
, &fault
);
8870 vcpu
->arch
.apf
.halted
= false;
8871 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
8874 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
8876 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
8879 return kvm_can_do_async_pf(vcpu
);
8882 void kvm_arch_start_assignment(struct kvm
*kvm
)
8884 atomic_inc(&kvm
->arch
.assigned_device_count
);
8886 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment
);
8888 void kvm_arch_end_assignment(struct kvm
*kvm
)
8890 atomic_dec(&kvm
->arch
.assigned_device_count
);
8892 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment
);
8894 bool kvm_arch_has_assigned_device(struct kvm
*kvm
)
8896 return atomic_read(&kvm
->arch
.assigned_device_count
);
8898 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device
);
8900 void kvm_arch_register_noncoherent_dma(struct kvm
*kvm
)
8902 atomic_inc(&kvm
->arch
.noncoherent_dma_count
);
8904 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma
);
8906 void kvm_arch_unregister_noncoherent_dma(struct kvm
*kvm
)
8908 atomic_dec(&kvm
->arch
.noncoherent_dma_count
);
8910 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma
);
8912 bool kvm_arch_has_noncoherent_dma(struct kvm
*kvm
)
8914 return atomic_read(&kvm
->arch
.noncoherent_dma_count
);
8916 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma
);
8918 bool kvm_arch_has_irq_bypass(void)
8920 return kvm_x86_ops
->update_pi_irte
!= NULL
;
8923 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
8924 struct irq_bypass_producer
*prod
)
8926 struct kvm_kernel_irqfd
*irqfd
=
8927 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
8929 irqfd
->producer
= prod
;
8931 return kvm_x86_ops
->update_pi_irte(irqfd
->kvm
,
8932 prod
->irq
, irqfd
->gsi
, 1);
8935 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
8936 struct irq_bypass_producer
*prod
)
8939 struct kvm_kernel_irqfd
*irqfd
=
8940 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
8942 WARN_ON(irqfd
->producer
!= prod
);
8943 irqfd
->producer
= NULL
;
8946 * When producer of consumer is unregistered, we change back to
8947 * remapped mode, so we can re-use the current implementation
8948 * when the irq is masked/disabled or the consumer side (KVM
8949 * int this case doesn't want to receive the interrupts.
8951 ret
= kvm_x86_ops
->update_pi_irte(irqfd
->kvm
, prod
->irq
, irqfd
->gsi
, 0);
8953 printk(KERN_INFO
"irq bypass consumer (token %p) unregistration"
8954 " fails: %d\n", irqfd
->consumer
.token
, ret
);
8957 int kvm_arch_update_irqfd_routing(struct kvm
*kvm
, unsigned int host_irq
,
8958 uint32_t guest_irq
, bool set
)
8960 if (!kvm_x86_ops
->update_pi_irte
)
8963 return kvm_x86_ops
->update_pi_irte(kvm
, host_irq
, guest_irq
, set
);
8966 bool kvm_vector_hashing_enabled(void)
8968 return vector_hashing
;
8970 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled
);
8972 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
8973 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio
);
8974 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
8975 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
8976 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
8977 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
8978 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
8979 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
8980 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
8981 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
8982 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
8983 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
8984 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);
8985 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset
);
8986 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window
);
8987 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full
);
8988 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update
);
8989 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access
);
8990 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi
);