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"
30 #include "assigned-dev.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/module.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 <trace/events/kvm.h>
58 #define CREATE_TRACE_POINTS
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 MAX_IO_MSRS 256
72 #define KVM_MAX_MCE_BANKS 32
73 u64 __read_mostly kvm_mce_cap_supported
= MCG_CTL_P
| MCG_SER_P
;
74 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported
);
76 #define emul_to_vcpu(ctxt) \
77 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
80 * - enable syscall per default because its emulated by KVM
81 * - enable LME and LMA per default on 64 bit KVM
85 u64 __read_mostly efer_reserved_bits
= ~((u64
)(EFER_SCE
| EFER_LME
| EFER_LMA
));
87 static u64 __read_mostly efer_reserved_bits
= ~((u64
)EFER_SCE
);
90 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
91 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
93 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
);
94 static void process_nmi(struct kvm_vcpu
*vcpu
);
95 static void enter_smm(struct kvm_vcpu
*vcpu
);
96 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
);
98 struct kvm_x86_ops
*kvm_x86_ops __read_mostly
;
99 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
101 static bool __read_mostly ignore_msrs
= 0;
102 module_param(ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
104 unsigned int min_timer_period_us
= 500;
105 module_param(min_timer_period_us
, uint
, S_IRUGO
| S_IWUSR
);
107 static bool __read_mostly kvmclock_periodic_sync
= true;
108 module_param(kvmclock_periodic_sync
, bool, S_IRUGO
);
110 bool __read_mostly kvm_has_tsc_control
;
111 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
112 u32 __read_mostly kvm_max_guest_tsc_khz
;
113 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
114 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits
;
115 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits
);
116 u64 __read_mostly kvm_max_tsc_scaling_ratio
;
117 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio
);
118 u64 __read_mostly kvm_default_tsc_scaling_ratio
;
119 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio
);
121 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
122 static u32 __read_mostly tsc_tolerance_ppm
= 250;
123 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
125 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
126 unsigned int __read_mostly lapic_timer_advance_ns
= 0;
127 module_param(lapic_timer_advance_ns
, uint
, S_IRUGO
| S_IWUSR
);
129 static bool __read_mostly vector_hashing
= true;
130 module_param(vector_hashing
, bool, S_IRUGO
);
132 static bool __read_mostly backwards_tsc_observed
= false;
134 #define KVM_NR_SHARED_MSRS 16
136 struct kvm_shared_msrs_global
{
138 u32 msrs
[KVM_NR_SHARED_MSRS
];
141 struct kvm_shared_msrs
{
142 struct user_return_notifier urn
;
144 struct kvm_shared_msr_values
{
147 } values
[KVM_NR_SHARED_MSRS
];
150 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
151 static struct kvm_shared_msrs __percpu
*shared_msrs
;
153 struct kvm_stats_debugfs_item debugfs_entries
[] = {
154 { "pf_fixed", VCPU_STAT(pf_fixed
) },
155 { "pf_guest", VCPU_STAT(pf_guest
) },
156 { "tlb_flush", VCPU_STAT(tlb_flush
) },
157 { "invlpg", VCPU_STAT(invlpg
) },
158 { "exits", VCPU_STAT(exits
) },
159 { "io_exits", VCPU_STAT(io_exits
) },
160 { "mmio_exits", VCPU_STAT(mmio_exits
) },
161 { "signal_exits", VCPU_STAT(signal_exits
) },
162 { "irq_window", VCPU_STAT(irq_window_exits
) },
163 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
164 { "halt_exits", VCPU_STAT(halt_exits
) },
165 { "halt_successful_poll", VCPU_STAT(halt_successful_poll
) },
166 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll
) },
167 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid
) },
168 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
169 { "hypercalls", VCPU_STAT(hypercalls
) },
170 { "request_irq", VCPU_STAT(request_irq_exits
) },
171 { "irq_exits", VCPU_STAT(irq_exits
) },
172 { "host_state_reload", VCPU_STAT(host_state_reload
) },
173 { "efer_reload", VCPU_STAT(efer_reload
) },
174 { "fpu_reload", VCPU_STAT(fpu_reload
) },
175 { "insn_emulation", VCPU_STAT(insn_emulation
) },
176 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
177 { "irq_injections", VCPU_STAT(irq_injections
) },
178 { "nmi_injections", VCPU_STAT(nmi_injections
) },
179 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
180 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
181 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
182 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
183 { "mmu_flooded", VM_STAT(mmu_flooded
) },
184 { "mmu_recycled", VM_STAT(mmu_recycled
) },
185 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
186 { "mmu_unsync", VM_STAT(mmu_unsync
) },
187 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
188 { "largepages", VM_STAT(lpages
) },
192 u64 __read_mostly host_xcr0
;
194 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
196 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
199 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
200 vcpu
->arch
.apf
.gfns
[i
] = ~0;
203 static void kvm_on_user_return(struct user_return_notifier
*urn
)
206 struct kvm_shared_msrs
*locals
207 = container_of(urn
, struct kvm_shared_msrs
, urn
);
208 struct kvm_shared_msr_values
*values
;
210 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
211 values
= &locals
->values
[slot
];
212 if (values
->host
!= values
->curr
) {
213 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
214 values
->curr
= values
->host
;
217 locals
->registered
= false;
218 user_return_notifier_unregister(urn
);
221 static void shared_msr_update(unsigned slot
, u32 msr
)
224 unsigned int cpu
= smp_processor_id();
225 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
227 /* only read, and nobody should modify it at this time,
228 * so don't need lock */
229 if (slot
>= shared_msrs_global
.nr
) {
230 printk(KERN_ERR
"kvm: invalid MSR slot!");
233 rdmsrl_safe(msr
, &value
);
234 smsr
->values
[slot
].host
= value
;
235 smsr
->values
[slot
].curr
= value
;
238 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
240 BUG_ON(slot
>= KVM_NR_SHARED_MSRS
);
241 shared_msrs_global
.msrs
[slot
] = msr
;
242 if (slot
>= shared_msrs_global
.nr
)
243 shared_msrs_global
.nr
= slot
+ 1;
245 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
247 static void kvm_shared_msr_cpu_online(void)
251 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
252 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
255 int kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
257 unsigned int cpu
= smp_processor_id();
258 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
261 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
263 smsr
->values
[slot
].curr
= value
;
264 err
= wrmsrl_safe(shared_msrs_global
.msrs
[slot
], value
);
268 if (!smsr
->registered
) {
269 smsr
->urn
.on_user_return
= kvm_on_user_return
;
270 user_return_notifier_register(&smsr
->urn
);
271 smsr
->registered
= true;
275 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
277 static void drop_user_return_notifiers(void)
279 unsigned int cpu
= smp_processor_id();
280 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
282 if (smsr
->registered
)
283 kvm_on_user_return(&smsr
->urn
);
286 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
288 return vcpu
->arch
.apic_base
;
290 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
292 int kvm_set_apic_base(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
294 u64 old_state
= vcpu
->arch
.apic_base
&
295 (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
);
296 u64 new_state
= msr_info
->data
&
297 (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
);
298 u64 reserved_bits
= ((~0ULL) << cpuid_maxphyaddr(vcpu
)) |
299 0x2ff | (guest_cpuid_has_x2apic(vcpu
) ? 0 : X2APIC_ENABLE
);
301 if (!msr_info
->host_initiated
&&
302 ((msr_info
->data
& reserved_bits
) != 0 ||
303 new_state
== X2APIC_ENABLE
||
304 (new_state
== MSR_IA32_APICBASE_ENABLE
&&
305 old_state
== (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
)) ||
306 (new_state
== (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
) &&
310 kvm_lapic_set_base(vcpu
, msr_info
->data
);
313 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
315 asmlinkage __visible
void kvm_spurious_fault(void)
317 /* Fault while not rebooting. We want the trace. */
320 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
322 #define EXCPT_BENIGN 0
323 #define EXCPT_CONTRIBUTORY 1
326 static int exception_class(int vector
)
336 return EXCPT_CONTRIBUTORY
;
343 #define EXCPT_FAULT 0
345 #define EXCPT_ABORT 2
346 #define EXCPT_INTERRUPT 3
348 static int exception_type(int vector
)
352 if (WARN_ON(vector
> 31 || vector
== NMI_VECTOR
))
353 return EXCPT_INTERRUPT
;
357 /* #DB is trap, as instruction watchpoints are handled elsewhere */
358 if (mask
& ((1 << DB_VECTOR
) | (1 << BP_VECTOR
) | (1 << OF_VECTOR
)))
361 if (mask
& ((1 << DF_VECTOR
) | (1 << MC_VECTOR
)))
364 /* Reserved exceptions will result in fault */
368 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
369 unsigned nr
, bool has_error
, u32 error_code
,
375 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
377 if (!vcpu
->arch
.exception
.pending
) {
379 if (has_error
&& !is_protmode(vcpu
))
381 vcpu
->arch
.exception
.pending
= true;
382 vcpu
->arch
.exception
.has_error_code
= has_error
;
383 vcpu
->arch
.exception
.nr
= nr
;
384 vcpu
->arch
.exception
.error_code
= error_code
;
385 vcpu
->arch
.exception
.reinject
= reinject
;
389 /* to check exception */
390 prev_nr
= vcpu
->arch
.exception
.nr
;
391 if (prev_nr
== DF_VECTOR
) {
392 /* triple fault -> shutdown */
393 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
396 class1
= exception_class(prev_nr
);
397 class2
= exception_class(nr
);
398 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
399 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
400 /* generate double fault per SDM Table 5-5 */
401 vcpu
->arch
.exception
.pending
= true;
402 vcpu
->arch
.exception
.has_error_code
= true;
403 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
404 vcpu
->arch
.exception
.error_code
= 0;
406 /* replace previous exception with a new one in a hope
407 that instruction re-execution will regenerate lost
412 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
414 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
416 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
418 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
420 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
422 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
424 void kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
427 kvm_inject_gp(vcpu
, 0);
429 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
431 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
433 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
435 ++vcpu
->stat
.pf_guest
;
436 vcpu
->arch
.cr2
= fault
->address
;
437 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
439 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
441 static bool kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
443 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
444 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
446 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
448 return fault
->nested_page_fault
;
451 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
453 atomic_inc(&vcpu
->arch
.nmi_queued
);
454 kvm_make_request(KVM_REQ_NMI
, vcpu
);
456 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
458 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
460 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
462 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
464 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
466 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
468 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
471 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
472 * a #GP and return false.
474 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
476 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
478 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
481 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
483 bool kvm_require_dr(struct kvm_vcpu
*vcpu
, int dr
)
485 if ((dr
!= 4 && dr
!= 5) || !kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
488 kvm_queue_exception(vcpu
, UD_VECTOR
);
491 EXPORT_SYMBOL_GPL(kvm_require_dr
);
494 * This function will be used to read from the physical memory of the currently
495 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
496 * can read from guest physical or from the guest's guest physical memory.
498 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
499 gfn_t ngfn
, void *data
, int offset
, int len
,
502 struct x86_exception exception
;
506 ngpa
= gfn_to_gpa(ngfn
);
507 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
, &exception
);
508 if (real_gfn
== UNMAPPED_GVA
)
511 real_gfn
= gpa_to_gfn(real_gfn
);
513 return kvm_vcpu_read_guest_page(vcpu
, real_gfn
, data
, offset
, len
);
515 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
517 static int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
518 void *data
, int offset
, int len
, u32 access
)
520 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
521 data
, offset
, len
, access
);
525 * Load the pae pdptrs. Return true is they are all valid.
527 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
529 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
530 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
533 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
535 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
536 offset
* sizeof(u64
), sizeof(pdpte
),
537 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
542 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
543 if ((pdpte
[i
] & PT_PRESENT_MASK
) &&
545 vcpu
->arch
.mmu
.guest_rsvd_check
.rsvd_bits_mask
[0][2])) {
552 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
553 __set_bit(VCPU_EXREG_PDPTR
,
554 (unsigned long *)&vcpu
->arch
.regs_avail
);
555 __set_bit(VCPU_EXREG_PDPTR
,
556 (unsigned long *)&vcpu
->arch
.regs_dirty
);
561 EXPORT_SYMBOL_GPL(load_pdptrs
);
563 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
565 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
571 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
574 if (!test_bit(VCPU_EXREG_PDPTR
,
575 (unsigned long *)&vcpu
->arch
.regs_avail
))
578 gfn
= (kvm_read_cr3(vcpu
) & ~31u) >> PAGE_SHIFT
;
579 offset
= (kvm_read_cr3(vcpu
) & ~31u) & (PAGE_SIZE
- 1);
580 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
581 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
584 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
590 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
592 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
593 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
;
598 if (cr0
& 0xffffffff00000000UL
)
602 cr0
&= ~CR0_RESERVED_BITS
;
604 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
607 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
610 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
612 if ((vcpu
->arch
.efer
& EFER_LME
)) {
617 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
622 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
627 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
630 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
632 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
633 kvm_clear_async_pf_completion_queue(vcpu
);
634 kvm_async_pf_hash_reset(vcpu
);
637 if ((cr0
^ old_cr0
) & update_bits
)
638 kvm_mmu_reset_context(vcpu
);
640 if (((cr0
^ old_cr0
) & X86_CR0_CD
) &&
641 kvm_arch_has_noncoherent_dma(vcpu
->kvm
) &&
642 !kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
643 kvm_zap_gfn_range(vcpu
->kvm
, 0, ~0ULL);
647 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
649 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
651 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
653 EXPORT_SYMBOL_GPL(kvm_lmsw
);
655 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
657 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
658 !vcpu
->guest_xcr0_loaded
) {
659 /* kvm_set_xcr() also depends on this */
660 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
661 vcpu
->guest_xcr0_loaded
= 1;
665 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
667 if (vcpu
->guest_xcr0_loaded
) {
668 if (vcpu
->arch
.xcr0
!= host_xcr0
)
669 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
670 vcpu
->guest_xcr0_loaded
= 0;
674 static int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
677 u64 old_xcr0
= vcpu
->arch
.xcr0
;
680 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
681 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
683 if (!(xcr0
& XFEATURE_MASK_FP
))
685 if ((xcr0
& XFEATURE_MASK_YMM
) && !(xcr0
& XFEATURE_MASK_SSE
))
689 * Do not allow the guest to set bits that we do not support
690 * saving. However, xcr0 bit 0 is always set, even if the
691 * emulated CPU does not support XSAVE (see fx_init).
693 valid_bits
= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FP
;
694 if (xcr0
& ~valid_bits
)
697 if ((!(xcr0
& XFEATURE_MASK_BNDREGS
)) !=
698 (!(xcr0
& XFEATURE_MASK_BNDCSR
)))
701 if (xcr0
& XFEATURE_MASK_AVX512
) {
702 if (!(xcr0
& XFEATURE_MASK_YMM
))
704 if ((xcr0
& XFEATURE_MASK_AVX512
) != XFEATURE_MASK_AVX512
)
707 vcpu
->arch
.xcr0
= xcr0
;
709 if ((xcr0
^ old_xcr0
) & XFEATURE_MASK_EXTEND
)
710 kvm_update_cpuid(vcpu
);
714 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
716 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 ||
717 __kvm_set_xcr(vcpu
, index
, xcr
)) {
718 kvm_inject_gp(vcpu
, 0);
723 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
725 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
727 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
728 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
|
729 X86_CR4_SMEP
| X86_CR4_SMAP
| X86_CR4_PKE
;
731 if (cr4
& CR4_RESERVED_BITS
)
734 if (!guest_cpuid_has_xsave(vcpu
) && (cr4
& X86_CR4_OSXSAVE
))
737 if (!guest_cpuid_has_smep(vcpu
) && (cr4
& X86_CR4_SMEP
))
740 if (!guest_cpuid_has_smap(vcpu
) && (cr4
& X86_CR4_SMAP
))
743 if (!guest_cpuid_has_fsgsbase(vcpu
) && (cr4
& X86_CR4_FSGSBASE
))
746 if (!guest_cpuid_has_pku(vcpu
) && (cr4
& X86_CR4_PKE
))
749 if (is_long_mode(vcpu
)) {
750 if (!(cr4
& X86_CR4_PAE
))
752 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
753 && ((cr4
^ old_cr4
) & pdptr_bits
)
754 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
758 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
759 if (!guest_cpuid_has_pcid(vcpu
))
762 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
763 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
767 if (kvm_x86_ops
->set_cr4(vcpu
, cr4
))
770 if (((cr4
^ old_cr4
) & pdptr_bits
) ||
771 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
772 kvm_mmu_reset_context(vcpu
);
774 if ((cr4
^ old_cr4
) & (X86_CR4_OSXSAVE
| X86_CR4_PKE
))
775 kvm_update_cpuid(vcpu
);
779 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
781 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
784 cr3
&= ~CR3_PCID_INVD
;
787 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
788 kvm_mmu_sync_roots(vcpu
);
789 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
793 if (is_long_mode(vcpu
)) {
794 if (cr3
& CR3_L_MODE_RESERVED_BITS
)
796 } else if (is_pae(vcpu
) && is_paging(vcpu
) &&
797 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
800 vcpu
->arch
.cr3
= cr3
;
801 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
802 kvm_mmu_new_cr3(vcpu
);
805 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
807 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
809 if (cr8
& CR8_RESERVED_BITS
)
811 if (lapic_in_kernel(vcpu
))
812 kvm_lapic_set_tpr(vcpu
, cr8
);
814 vcpu
->arch
.cr8
= cr8
;
817 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
819 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
821 if (lapic_in_kernel(vcpu
))
822 return kvm_lapic_get_cr8(vcpu
);
824 return vcpu
->arch
.cr8
;
826 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
828 static void kvm_update_dr0123(struct kvm_vcpu
*vcpu
)
832 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
833 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
834 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
835 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_RELOAD
;
839 static void kvm_update_dr6(struct kvm_vcpu
*vcpu
)
841 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
842 kvm_x86_ops
->set_dr6(vcpu
, vcpu
->arch
.dr6
);
845 static void kvm_update_dr7(struct kvm_vcpu
*vcpu
)
849 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
850 dr7
= vcpu
->arch
.guest_debug_dr7
;
852 dr7
= vcpu
->arch
.dr7
;
853 kvm_x86_ops
->set_dr7(vcpu
, dr7
);
854 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_BP_ENABLED
;
855 if (dr7
& DR7_BP_EN_MASK
)
856 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_BP_ENABLED
;
859 static u64
kvm_dr6_fixed(struct kvm_vcpu
*vcpu
)
861 u64 fixed
= DR6_FIXED_1
;
863 if (!guest_cpuid_has_rtm(vcpu
))
868 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
872 vcpu
->arch
.db
[dr
] = val
;
873 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
874 vcpu
->arch
.eff_db
[dr
] = val
;
879 if (val
& 0xffffffff00000000ULL
)
881 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | kvm_dr6_fixed(vcpu
);
882 kvm_update_dr6(vcpu
);
887 if (val
& 0xffffffff00000000ULL
)
889 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
890 kvm_update_dr7(vcpu
);
897 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
899 if (__kvm_set_dr(vcpu
, dr
, val
)) {
900 kvm_inject_gp(vcpu
, 0);
905 EXPORT_SYMBOL_GPL(kvm_set_dr
);
907 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
911 *val
= vcpu
->arch
.db
[dr
];
916 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
917 *val
= vcpu
->arch
.dr6
;
919 *val
= kvm_x86_ops
->get_dr6(vcpu
);
924 *val
= vcpu
->arch
.dr7
;
929 EXPORT_SYMBOL_GPL(kvm_get_dr
);
931 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
933 u32 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
937 err
= kvm_pmu_rdpmc(vcpu
, ecx
, &data
);
940 kvm_register_write(vcpu
, VCPU_REGS_RAX
, (u32
)data
);
941 kvm_register_write(vcpu
, VCPU_REGS_RDX
, data
>> 32);
944 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
947 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
948 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
950 * This list is modified at module load time to reflect the
951 * capabilities of the host cpu. This capabilities test skips MSRs that are
952 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
953 * may depend on host virtualization features rather than host cpu features.
956 static u32 msrs_to_save
[] = {
957 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
960 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
962 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
,
963 MSR_IA32_FEATURE_CONTROL
, MSR_IA32_BNDCFGS
, MSR_TSC_AUX
,
966 static unsigned num_msrs_to_save
;
968 static u32 emulated_msrs
[] = {
969 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
970 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
971 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
972 HV_X64_MSR_TIME_REF_COUNT
, HV_X64_MSR_REFERENCE_TSC
,
973 HV_X64_MSR_CRASH_P0
, HV_X64_MSR_CRASH_P1
, HV_X64_MSR_CRASH_P2
,
974 HV_X64_MSR_CRASH_P3
, HV_X64_MSR_CRASH_P4
, HV_X64_MSR_CRASH_CTL
,
977 HV_X64_MSR_VP_RUNTIME
,
979 HV_X64_MSR_STIMER0_CONFIG
,
980 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
984 MSR_IA32_TSCDEADLINE
,
985 MSR_IA32_MISC_ENABLE
,
988 MSR_IA32_MCG_EXT_CTL
,
992 static unsigned num_emulated_msrs
;
994 bool kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
996 if (efer
& efer_reserved_bits
)
999 if (efer
& EFER_FFXSR
) {
1000 struct kvm_cpuid_entry2
*feat
;
1002 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
1003 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
)))
1007 if (efer
& EFER_SVME
) {
1008 struct kvm_cpuid_entry2
*feat
;
1010 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
1011 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
)))
1017 EXPORT_SYMBOL_GPL(kvm_valid_efer
);
1019 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1021 u64 old_efer
= vcpu
->arch
.efer
;
1023 if (!kvm_valid_efer(vcpu
, efer
))
1027 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
1031 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
1033 kvm_x86_ops
->set_efer(vcpu
, efer
);
1035 /* Update reserved bits */
1036 if ((efer
^ old_efer
) & EFER_NX
)
1037 kvm_mmu_reset_context(vcpu
);
1042 void kvm_enable_efer_bits(u64 mask
)
1044 efer_reserved_bits
&= ~mask
;
1046 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
1049 * Writes msr value into into the appropriate "register".
1050 * Returns 0 on success, non-0 otherwise.
1051 * Assumes vcpu_load() was already called.
1053 int kvm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1055 switch (msr
->index
) {
1058 case MSR_KERNEL_GS_BASE
:
1061 if (is_noncanonical_address(msr
->data
))
1064 case MSR_IA32_SYSENTER_EIP
:
1065 case MSR_IA32_SYSENTER_ESP
:
1067 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1068 * non-canonical address is written on Intel but not on
1069 * AMD (which ignores the top 32-bits, because it does
1070 * not implement 64-bit SYSENTER).
1072 * 64-bit code should hence be able to write a non-canonical
1073 * value on AMD. Making the address canonical ensures that
1074 * vmentry does not fail on Intel after writing a non-canonical
1075 * value, and that something deterministic happens if the guest
1076 * invokes 64-bit SYSENTER.
1078 msr
->data
= get_canonical(msr
->data
);
1080 return kvm_x86_ops
->set_msr(vcpu
, msr
);
1082 EXPORT_SYMBOL_GPL(kvm_set_msr
);
1085 * Adapt set_msr() to msr_io()'s calling convention
1087 static int do_get_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1089 struct msr_data msr
;
1093 msr
.host_initiated
= true;
1094 r
= kvm_get_msr(vcpu
, &msr
);
1102 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1104 struct msr_data msr
;
1108 msr
.host_initiated
= true;
1109 return kvm_set_msr(vcpu
, &msr
);
1112 #ifdef CONFIG_X86_64
1113 struct pvclock_gtod_data
{
1116 struct { /* extract of a clocksource struct */
1128 static struct pvclock_gtod_data pvclock_gtod_data
;
1130 static void update_pvclock_gtod(struct timekeeper
*tk
)
1132 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
1135 boot_ns
= ktime_to_ns(ktime_add(tk
->tkr_mono
.base
, tk
->offs_boot
));
1137 write_seqcount_begin(&vdata
->seq
);
1139 /* copy pvclock gtod data */
1140 vdata
->clock
.vclock_mode
= tk
->tkr_mono
.clock
->archdata
.vclock_mode
;
1141 vdata
->clock
.cycle_last
= tk
->tkr_mono
.cycle_last
;
1142 vdata
->clock
.mask
= tk
->tkr_mono
.mask
;
1143 vdata
->clock
.mult
= tk
->tkr_mono
.mult
;
1144 vdata
->clock
.shift
= tk
->tkr_mono
.shift
;
1146 vdata
->boot_ns
= boot_ns
;
1147 vdata
->nsec_base
= tk
->tkr_mono
.xtime_nsec
;
1149 write_seqcount_end(&vdata
->seq
);
1153 void kvm_set_pending_timer(struct kvm_vcpu
*vcpu
)
1156 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1157 * vcpu_enter_guest. This function is only called from
1158 * the physical CPU that is running vcpu.
1160 kvm_make_request(KVM_REQ_PENDING_TIMER
, vcpu
);
1163 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
1167 struct pvclock_wall_clock wc
;
1168 struct timespec64 boot
;
1173 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1178 ++version
; /* first time write, random junk */
1182 if (kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
)))
1186 * The guest calculates current wall clock time by adding
1187 * system time (updated by kvm_guest_time_update below) to the
1188 * wall clock specified here. guest system time equals host
1189 * system time for us, thus we must fill in host boot time here.
1191 getboottime64(&boot
);
1193 if (kvm
->arch
.kvmclock_offset
) {
1194 struct timespec64 ts
= ns_to_timespec64(kvm
->arch
.kvmclock_offset
);
1195 boot
= timespec64_sub(boot
, ts
);
1197 wc
.sec
= (u32
)boot
.tv_sec
; /* overflow in 2106 guest time */
1198 wc
.nsec
= boot
.tv_nsec
;
1199 wc
.version
= version
;
1201 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1204 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1207 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
1209 do_shl32_div32(dividend
, divisor
);
1213 static void kvm_get_time_scale(uint64_t scaled_hz
, uint64_t base_hz
,
1214 s8
*pshift
, u32
*pmultiplier
)
1222 scaled64
= scaled_hz
;
1223 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
1228 tps32
= (uint32_t)tps64
;
1229 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
1230 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
1238 *pmultiplier
= div_frac(scaled64
, tps32
);
1240 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1241 __func__
, base_hz
, scaled_hz
, shift
, *pmultiplier
);
1244 #ifdef CONFIG_X86_64
1245 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
1248 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
1249 static unsigned long max_tsc_khz
;
1251 static inline u64
nsec_to_cycles(struct kvm_vcpu
*vcpu
, u64 nsec
)
1253 return pvclock_scale_delta(nsec
, vcpu
->arch
.virtual_tsc_mult
,
1254 vcpu
->arch
.virtual_tsc_shift
);
1257 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1259 u64 v
= (u64
)khz
* (1000000 + ppm
);
1264 static int set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
, bool scale
)
1268 /* Guest TSC same frequency as host TSC? */
1270 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1274 /* TSC scaling supported? */
1275 if (!kvm_has_tsc_control
) {
1276 if (user_tsc_khz
> tsc_khz
) {
1277 vcpu
->arch
.tsc_catchup
= 1;
1278 vcpu
->arch
.tsc_always_catchup
= 1;
1281 WARN(1, "user requested TSC rate below hardware speed\n");
1286 /* TSC scaling required - calculate ratio */
1287 ratio
= mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits
,
1288 user_tsc_khz
, tsc_khz
);
1290 if (ratio
== 0 || ratio
>= kvm_max_tsc_scaling_ratio
) {
1291 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1296 vcpu
->arch
.tsc_scaling_ratio
= ratio
;
1300 static int kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
)
1302 u32 thresh_lo
, thresh_hi
;
1303 int use_scaling
= 0;
1305 /* tsc_khz can be zero if TSC calibration fails */
1306 if (user_tsc_khz
== 0) {
1307 /* set tsc_scaling_ratio to a safe value */
1308 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1312 /* Compute a scale to convert nanoseconds in TSC cycles */
1313 kvm_get_time_scale(user_tsc_khz
* 1000LL, NSEC_PER_SEC
,
1314 &vcpu
->arch
.virtual_tsc_shift
,
1315 &vcpu
->arch
.virtual_tsc_mult
);
1316 vcpu
->arch
.virtual_tsc_khz
= user_tsc_khz
;
1319 * Compute the variation in TSC rate which is acceptable
1320 * within the range of tolerance and decide if the
1321 * rate being applied is within that bounds of the hardware
1322 * rate. If so, no scaling or compensation need be done.
1324 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1325 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1326 if (user_tsc_khz
< thresh_lo
|| user_tsc_khz
> thresh_hi
) {
1327 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz
, thresh_lo
, thresh_hi
);
1330 return set_tsc_khz(vcpu
, user_tsc_khz
, use_scaling
);
1333 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1335 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1336 vcpu
->arch
.virtual_tsc_mult
,
1337 vcpu
->arch
.virtual_tsc_shift
);
1338 tsc
+= vcpu
->arch
.this_tsc_write
;
1342 static void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
1344 #ifdef CONFIG_X86_64
1346 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1347 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1349 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1350 atomic_read(&vcpu
->kvm
->online_vcpus
));
1353 * Once the masterclock is enabled, always perform request in
1354 * order to update it.
1356 * In order to enable masterclock, the host clocksource must be TSC
1357 * and the vcpus need to have matched TSCs. When that happens,
1358 * perform request to enable masterclock.
1360 if (ka
->use_master_clock
||
1361 (gtod
->clock
.vclock_mode
== VCLOCK_TSC
&& vcpus_matched
))
1362 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1364 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
1365 atomic_read(&vcpu
->kvm
->online_vcpus
),
1366 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
1370 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu
*vcpu
, s64 offset
)
1372 u64 curr_offset
= kvm_x86_ops
->read_tsc_offset(vcpu
);
1373 vcpu
->arch
.ia32_tsc_adjust_msr
+= offset
- curr_offset
;
1377 * Multiply tsc by a fixed point number represented by ratio.
1379 * The most significant 64-N bits (mult) of ratio represent the
1380 * integral part of the fixed point number; the remaining N bits
1381 * (frac) represent the fractional part, ie. ratio represents a fixed
1382 * point number (mult + frac * 2^(-N)).
1384 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1386 static inline u64
__scale_tsc(u64 ratio
, u64 tsc
)
1388 return mul_u64_u64_shr(tsc
, ratio
, kvm_tsc_scaling_ratio_frac_bits
);
1391 u64
kvm_scale_tsc(struct kvm_vcpu
*vcpu
, u64 tsc
)
1394 u64 ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1396 if (ratio
!= kvm_default_tsc_scaling_ratio
)
1397 _tsc
= __scale_tsc(ratio
, tsc
);
1401 EXPORT_SYMBOL_GPL(kvm_scale_tsc
);
1403 static u64
kvm_compute_tsc_offset(struct kvm_vcpu
*vcpu
, u64 target_tsc
)
1407 tsc
= kvm_scale_tsc(vcpu
, rdtsc());
1409 return target_tsc
- tsc
;
1412 u64
kvm_read_l1_tsc(struct kvm_vcpu
*vcpu
, u64 host_tsc
)
1414 return kvm_x86_ops
->read_l1_tsc(vcpu
, kvm_scale_tsc(vcpu
, host_tsc
));
1416 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc
);
1418 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1420 struct kvm
*kvm
= vcpu
->kvm
;
1421 u64 offset
, ns
, elapsed
;
1422 unsigned long flags
;
1425 bool already_matched
;
1426 u64 data
= msr
->data
;
1428 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1429 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1430 ns
= get_kernel_ns();
1431 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1433 if (vcpu
->arch
.virtual_tsc_khz
) {
1436 /* n.b - signed multiplication and division required */
1437 usdiff
= data
- kvm
->arch
.last_tsc_write
;
1438 #ifdef CONFIG_X86_64
1439 usdiff
= (usdiff
* 1000) / vcpu
->arch
.virtual_tsc_khz
;
1441 /* do_div() only does unsigned */
1442 asm("1: idivl %[divisor]\n"
1443 "2: xor %%edx, %%edx\n"
1444 " movl $0, %[faulted]\n"
1446 ".section .fixup,\"ax\"\n"
1447 "4: movl $1, %[faulted]\n"
1451 _ASM_EXTABLE(1b
, 4b
)
1453 : "=A"(usdiff
), [faulted
] "=r" (faulted
)
1454 : "A"(usdiff
* 1000), [divisor
] "rm"(vcpu
->arch
.virtual_tsc_khz
));
1457 do_div(elapsed
, 1000);
1462 /* idivl overflow => difference is larger than USEC_PER_SEC */
1464 usdiff
= USEC_PER_SEC
;
1466 usdiff
= USEC_PER_SEC
; /* disable TSC match window below */
1469 * Special case: TSC write with a small delta (1 second) of virtual
1470 * cycle time against real time is interpreted as an attempt to
1471 * synchronize the CPU.
1473 * For a reliable TSC, we can match TSC offsets, and for an unstable
1474 * TSC, we add elapsed time in this computation. We could let the
1475 * compensation code attempt to catch up if we fall behind, but
1476 * it's better to try to match offsets from the beginning.
1478 if (usdiff
< USEC_PER_SEC
&&
1479 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1480 if (!check_tsc_unstable()) {
1481 offset
= kvm
->arch
.cur_tsc_offset
;
1482 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1484 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1486 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1487 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1490 already_matched
= (vcpu
->arch
.this_tsc_generation
== kvm
->arch
.cur_tsc_generation
);
1493 * We split periods of matched TSC writes into generations.
1494 * For each generation, we track the original measured
1495 * nanosecond time, offset, and write, so if TSCs are in
1496 * sync, we can match exact offset, and if not, we can match
1497 * exact software computation in compute_guest_tsc()
1499 * These values are tracked in kvm->arch.cur_xxx variables.
1501 kvm
->arch
.cur_tsc_generation
++;
1502 kvm
->arch
.cur_tsc_nsec
= ns
;
1503 kvm
->arch
.cur_tsc_write
= data
;
1504 kvm
->arch
.cur_tsc_offset
= offset
;
1506 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1507 kvm
->arch
.cur_tsc_generation
, data
);
1511 * We also track th most recent recorded KHZ, write and time to
1512 * allow the matching interval to be extended at each write.
1514 kvm
->arch
.last_tsc_nsec
= ns
;
1515 kvm
->arch
.last_tsc_write
= data
;
1516 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1518 vcpu
->arch
.last_guest_tsc
= data
;
1520 /* Keep track of which generation this VCPU has synchronized to */
1521 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1522 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1523 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1525 if (guest_cpuid_has_tsc_adjust(vcpu
) && !msr
->host_initiated
)
1526 update_ia32_tsc_adjust_msr(vcpu
, offset
);
1527 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1528 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1530 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1532 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
1533 } else if (!already_matched
) {
1534 kvm
->arch
.nr_vcpus_matched_tsc
++;
1537 kvm_track_tsc_matching(vcpu
);
1538 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1541 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1543 static inline void adjust_tsc_offset_guest(struct kvm_vcpu
*vcpu
,
1546 kvm_x86_ops
->adjust_tsc_offset_guest(vcpu
, adjustment
);
1549 static inline void adjust_tsc_offset_host(struct kvm_vcpu
*vcpu
, s64 adjustment
)
1551 if (vcpu
->arch
.tsc_scaling_ratio
!= kvm_default_tsc_scaling_ratio
)
1552 WARN_ON(adjustment
< 0);
1553 adjustment
= kvm_scale_tsc(vcpu
, (u64
) adjustment
);
1554 kvm_x86_ops
->adjust_tsc_offset_guest(vcpu
, adjustment
);
1557 #ifdef CONFIG_X86_64
1559 static cycle_t
read_tsc(void)
1561 cycle_t ret
= (cycle_t
)rdtsc_ordered();
1562 u64 last
= pvclock_gtod_data
.clock
.cycle_last
;
1564 if (likely(ret
>= last
))
1568 * GCC likes to generate cmov here, but this branch is extremely
1569 * predictable (it's just a function of time and the likely is
1570 * very likely) and there's a data dependence, so force GCC
1571 * to generate a branch instead. I don't barrier() because
1572 * we don't actually need a barrier, and if this function
1573 * ever gets inlined it will generate worse code.
1579 static inline u64
vgettsc(cycle_t
*cycle_now
)
1582 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1584 *cycle_now
= read_tsc();
1586 v
= (*cycle_now
- gtod
->clock
.cycle_last
) & gtod
->clock
.mask
;
1587 return v
* gtod
->clock
.mult
;
1590 static int do_monotonic_boot(s64
*t
, cycle_t
*cycle_now
)
1592 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1598 seq
= read_seqcount_begin(>od
->seq
);
1599 mode
= gtod
->clock
.vclock_mode
;
1600 ns
= gtod
->nsec_base
;
1601 ns
+= vgettsc(cycle_now
);
1602 ns
>>= gtod
->clock
.shift
;
1603 ns
+= gtod
->boot_ns
;
1604 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1610 /* returns true if host is using tsc clocksource */
1611 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, cycle_t
*cycle_now
)
1613 /* checked again under seqlock below */
1614 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1617 return do_monotonic_boot(kernel_ns
, cycle_now
) == VCLOCK_TSC
;
1623 * Assuming a stable TSC across physical CPUS, and a stable TSC
1624 * across virtual CPUs, the following condition is possible.
1625 * Each numbered line represents an event visible to both
1626 * CPUs at the next numbered event.
1628 * "timespecX" represents host monotonic time. "tscX" represents
1631 * VCPU0 on CPU0 | VCPU1 on CPU1
1633 * 1. read timespec0,tsc0
1634 * 2. | timespec1 = timespec0 + N
1636 * 3. transition to guest | transition to guest
1637 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1638 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1639 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1641 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1644 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1646 * - 0 < N - M => M < N
1648 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1649 * always the case (the difference between two distinct xtime instances
1650 * might be smaller then the difference between corresponding TSC reads,
1651 * when updating guest vcpus pvclock areas).
1653 * To avoid that problem, do not allow visibility of distinct
1654 * system_timestamp/tsc_timestamp values simultaneously: use a master
1655 * copy of host monotonic time values. Update that master copy
1658 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1662 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
1664 #ifdef CONFIG_X86_64
1665 struct kvm_arch
*ka
= &kvm
->arch
;
1667 bool host_tsc_clocksource
, vcpus_matched
;
1669 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1670 atomic_read(&kvm
->online_vcpus
));
1673 * If the host uses TSC clock, then passthrough TSC as stable
1676 host_tsc_clocksource
= kvm_get_time_and_clockread(
1677 &ka
->master_kernel_ns
,
1678 &ka
->master_cycle_now
);
1680 ka
->use_master_clock
= host_tsc_clocksource
&& vcpus_matched
1681 && !backwards_tsc_observed
1682 && !ka
->boot_vcpu_runs_old_kvmclock
;
1684 if (ka
->use_master_clock
)
1685 atomic_set(&kvm_guest_has_master_clock
, 1);
1687 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
1688 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
1693 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
1695 kvm_make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
1698 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
1700 #ifdef CONFIG_X86_64
1702 struct kvm_vcpu
*vcpu
;
1703 struct kvm_arch
*ka
= &kvm
->arch
;
1705 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1706 kvm_make_mclock_inprogress_request(kvm
);
1707 /* no guest entries from this point */
1708 pvclock_update_vm_gtod_copy(kvm
);
1710 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1711 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1713 /* guest entries allowed */
1714 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1715 clear_bit(KVM_REQ_MCLOCK_INPROGRESS
, &vcpu
->requests
);
1717 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1721 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1723 unsigned long flags
, tgt_tsc_khz
;
1724 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1725 struct kvm_arch
*ka
= &v
->kvm
->arch
;
1727 u64 tsc_timestamp
, host_tsc
;
1728 struct pvclock_vcpu_time_info guest_hv_clock
;
1730 bool use_master_clock
;
1736 * If the host uses TSC clock, then passthrough TSC as stable
1739 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1740 use_master_clock
= ka
->use_master_clock
;
1741 if (use_master_clock
) {
1742 host_tsc
= ka
->master_cycle_now
;
1743 kernel_ns
= ka
->master_kernel_ns
;
1745 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1747 /* Keep irq disabled to prevent changes to the clock */
1748 local_irq_save(flags
);
1749 tgt_tsc_khz
= __this_cpu_read(cpu_tsc_khz
);
1750 if (unlikely(tgt_tsc_khz
== 0)) {
1751 local_irq_restore(flags
);
1752 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1755 if (!use_master_clock
) {
1757 kernel_ns
= get_kernel_ns();
1760 tsc_timestamp
= kvm_read_l1_tsc(v
, host_tsc
);
1763 * We may have to catch up the TSC to match elapsed wall clock
1764 * time for two reasons, even if kvmclock is used.
1765 * 1) CPU could have been running below the maximum TSC rate
1766 * 2) Broken TSC compensation resets the base at each VCPU
1767 * entry to avoid unknown leaps of TSC even when running
1768 * again on the same CPU. This may cause apparent elapsed
1769 * time to disappear, and the guest to stand still or run
1772 if (vcpu
->tsc_catchup
) {
1773 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1774 if (tsc
> tsc_timestamp
) {
1775 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1776 tsc_timestamp
= tsc
;
1780 local_irq_restore(flags
);
1782 if (!vcpu
->pv_time_enabled
)
1785 if (kvm_has_tsc_control
)
1786 tgt_tsc_khz
= kvm_scale_tsc(v
, tgt_tsc_khz
);
1788 if (unlikely(vcpu
->hw_tsc_khz
!= tgt_tsc_khz
)) {
1789 kvm_get_time_scale(NSEC_PER_SEC
, tgt_tsc_khz
* 1000LL,
1790 &vcpu
->hv_clock
.tsc_shift
,
1791 &vcpu
->hv_clock
.tsc_to_system_mul
);
1792 vcpu
->hw_tsc_khz
= tgt_tsc_khz
;
1795 /* With all the info we got, fill in the values */
1796 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1797 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1798 vcpu
->last_guest_tsc
= tsc_timestamp
;
1800 if (unlikely(kvm_read_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1801 &guest_hv_clock
, sizeof(guest_hv_clock
))))
1804 /* This VCPU is paused, but it's legal for a guest to read another
1805 * VCPU's kvmclock, so we really have to follow the specification where
1806 * it says that version is odd if data is being modified, and even after
1809 * Version field updates must be kept separate. This is because
1810 * kvm_write_guest_cached might use a "rep movs" instruction, and
1811 * writes within a string instruction are weakly ordered. So there
1812 * are three writes overall.
1814 * As a small optimization, only write the version field in the first
1815 * and third write. The vcpu->pv_time cache is still valid, because the
1816 * version field is the first in the struct.
1818 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info
, version
) != 0);
1820 vcpu
->hv_clock
.version
= guest_hv_clock
.version
+ 1;
1821 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1823 sizeof(vcpu
->hv_clock
.version
));
1827 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1828 pvclock_flags
= (guest_hv_clock
.flags
& PVCLOCK_GUEST_STOPPED
);
1830 if (vcpu
->pvclock_set_guest_stopped_request
) {
1831 pvclock_flags
|= PVCLOCK_GUEST_STOPPED
;
1832 vcpu
->pvclock_set_guest_stopped_request
= false;
1835 /* If the host uses TSC clocksource, then it is stable */
1836 if (use_master_clock
)
1837 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
1839 vcpu
->hv_clock
.flags
= pvclock_flags
;
1841 trace_kvm_pvclock_update(v
->vcpu_id
, &vcpu
->hv_clock
);
1843 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1845 sizeof(vcpu
->hv_clock
));
1849 vcpu
->hv_clock
.version
++;
1850 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1852 sizeof(vcpu
->hv_clock
.version
));
1857 * kvmclock updates which are isolated to a given vcpu, such as
1858 * vcpu->cpu migration, should not allow system_timestamp from
1859 * the rest of the vcpus to remain static. Otherwise ntp frequency
1860 * correction applies to one vcpu's system_timestamp but not
1863 * So in those cases, request a kvmclock update for all vcpus.
1864 * We need to rate-limit these requests though, as they can
1865 * considerably slow guests that have a large number of vcpus.
1866 * The time for a remote vcpu to update its kvmclock is bound
1867 * by the delay we use to rate-limit the updates.
1870 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1872 static void kvmclock_update_fn(struct work_struct
*work
)
1875 struct delayed_work
*dwork
= to_delayed_work(work
);
1876 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
1877 kvmclock_update_work
);
1878 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
1879 struct kvm_vcpu
*vcpu
;
1881 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1882 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1883 kvm_vcpu_kick(vcpu
);
1887 static void kvm_gen_kvmclock_update(struct kvm_vcpu
*v
)
1889 struct kvm
*kvm
= v
->kvm
;
1891 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1892 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
,
1893 KVMCLOCK_UPDATE_DELAY
);
1896 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1898 static void kvmclock_sync_fn(struct work_struct
*work
)
1900 struct delayed_work
*dwork
= to_delayed_work(work
);
1901 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
1902 kvmclock_sync_work
);
1903 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
1905 if (!kvmclock_periodic_sync
)
1908 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
, 0);
1909 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
1910 KVMCLOCK_SYNC_PERIOD
);
1913 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1915 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1916 unsigned bank_num
= mcg_cap
& 0xff;
1919 case MSR_IA32_MCG_STATUS
:
1920 vcpu
->arch
.mcg_status
= data
;
1922 case MSR_IA32_MCG_CTL
:
1923 if (!(mcg_cap
& MCG_CTL_P
))
1925 if (data
!= 0 && data
!= ~(u64
)0)
1927 vcpu
->arch
.mcg_ctl
= data
;
1930 if (msr
>= MSR_IA32_MC0_CTL
&&
1931 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
1932 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1933 /* only 0 or all 1s can be written to IA32_MCi_CTL
1934 * some Linux kernels though clear bit 10 in bank 4 to
1935 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1936 * this to avoid an uncatched #GP in the guest
1938 if ((offset
& 0x3) == 0 &&
1939 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
1941 vcpu
->arch
.mce_banks
[offset
] = data
;
1949 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
1951 struct kvm
*kvm
= vcpu
->kvm
;
1952 int lm
= is_long_mode(vcpu
);
1953 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
1954 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
1955 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
1956 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
1957 u32 page_num
= data
& ~PAGE_MASK
;
1958 u64 page_addr
= data
& PAGE_MASK
;
1963 if (page_num
>= blob_size
)
1966 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
1971 if (kvm_vcpu_write_guest(vcpu
, page_addr
, page
, PAGE_SIZE
))
1980 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
1982 gpa_t gpa
= data
& ~0x3f;
1984 /* Bits 2:5 are reserved, Should be zero */
1988 vcpu
->arch
.apf
.msr_val
= data
;
1990 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
1991 kvm_clear_async_pf_completion_queue(vcpu
);
1992 kvm_async_pf_hash_reset(vcpu
);
1996 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
,
2000 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
2001 kvm_async_pf_wakeup_all(vcpu
);
2005 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
2007 vcpu
->arch
.pv_time_enabled
= false;
2010 static void record_steal_time(struct kvm_vcpu
*vcpu
)
2012 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
2015 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2016 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
2019 if (vcpu
->arch
.st
.steal
.version
& 1)
2020 vcpu
->arch
.st
.steal
.version
+= 1; /* first time write, random junk */
2022 vcpu
->arch
.st
.steal
.version
+= 1;
2024 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2025 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2029 vcpu
->arch
.st
.steal
.steal
+= current
->sched_info
.run_delay
-
2030 vcpu
->arch
.st
.last_steal
;
2031 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
2033 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2034 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2038 vcpu
->arch
.st
.steal
.version
+= 1;
2040 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2041 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2044 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2047 u32 msr
= msr_info
->index
;
2048 u64 data
= msr_info
->data
;
2051 case MSR_AMD64_NB_CFG
:
2052 case MSR_IA32_UCODE_REV
:
2053 case MSR_IA32_UCODE_WRITE
:
2054 case MSR_VM_HSAVE_PA
:
2055 case MSR_AMD64_PATCH_LOADER
:
2056 case MSR_AMD64_BU_CFG2
:
2060 return set_efer(vcpu
, data
);
2062 data
&= ~(u64
)0x40; /* ignore flush filter disable */
2063 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
2064 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
2065 data
&= ~(u64
)0x40000; /* ignore Mc status write enable */
2067 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
2072 case MSR_FAM10H_MMIO_CONF_BASE
:
2074 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
2079 case MSR_IA32_DEBUGCTLMSR
:
2081 /* We support the non-activated case already */
2083 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
2084 /* Values other than LBR and BTF are vendor-specific,
2085 thus reserved and should throw a #GP */
2088 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2091 case 0x200 ... 0x2ff:
2092 return kvm_mtrr_set_msr(vcpu
, msr
, data
);
2093 case MSR_IA32_APICBASE
:
2094 return kvm_set_apic_base(vcpu
, msr_info
);
2095 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2096 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
2097 case MSR_IA32_TSCDEADLINE
:
2098 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
2100 case MSR_IA32_TSC_ADJUST
:
2101 if (guest_cpuid_has_tsc_adjust(vcpu
)) {
2102 if (!msr_info
->host_initiated
) {
2103 s64 adj
= data
- vcpu
->arch
.ia32_tsc_adjust_msr
;
2104 adjust_tsc_offset_guest(vcpu
, adj
);
2106 vcpu
->arch
.ia32_tsc_adjust_msr
= data
;
2109 case MSR_IA32_MISC_ENABLE
:
2110 vcpu
->arch
.ia32_misc_enable_msr
= data
;
2112 case MSR_IA32_SMBASE
:
2113 if (!msr_info
->host_initiated
)
2115 vcpu
->arch
.smbase
= data
;
2117 case MSR_KVM_WALL_CLOCK_NEW
:
2118 case MSR_KVM_WALL_CLOCK
:
2119 vcpu
->kvm
->arch
.wall_clock
= data
;
2120 kvm_write_wall_clock(vcpu
->kvm
, data
);
2122 case MSR_KVM_SYSTEM_TIME_NEW
:
2123 case MSR_KVM_SYSTEM_TIME
: {
2125 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
2127 kvmclock_reset(vcpu
);
2129 if (vcpu
->vcpu_id
== 0 && !msr_info
->host_initiated
) {
2130 bool tmp
= (msr
== MSR_KVM_SYSTEM_TIME
);
2132 if (ka
->boot_vcpu_runs_old_kvmclock
!= tmp
)
2133 set_bit(KVM_REQ_MASTERCLOCK_UPDATE
,
2136 ka
->boot_vcpu_runs_old_kvmclock
= tmp
;
2139 vcpu
->arch
.time
= data
;
2140 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2142 /* we verify if the enable bit is set... */
2146 gpa_offset
= data
& ~(PAGE_MASK
| 1);
2148 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2149 &vcpu
->arch
.pv_time
, data
& ~1ULL,
2150 sizeof(struct pvclock_vcpu_time_info
)))
2151 vcpu
->arch
.pv_time_enabled
= false;
2153 vcpu
->arch
.pv_time_enabled
= true;
2157 case MSR_KVM_ASYNC_PF_EN
:
2158 if (kvm_pv_enable_async_pf(vcpu
, data
))
2161 case MSR_KVM_STEAL_TIME
:
2163 if (unlikely(!sched_info_on()))
2166 if (data
& KVM_STEAL_RESERVED_MASK
)
2169 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2170 data
& KVM_STEAL_VALID_BITS
,
2171 sizeof(struct kvm_steal_time
)))
2174 vcpu
->arch
.st
.msr_val
= data
;
2176 if (!(data
& KVM_MSR_ENABLED
))
2179 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2182 case MSR_KVM_PV_EOI_EN
:
2183 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
2187 case MSR_IA32_MCG_CTL
:
2188 case MSR_IA32_MCG_STATUS
:
2189 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2190 return set_msr_mce(vcpu
, msr
, data
);
2192 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2193 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2194 pr
= true; /* fall through */
2195 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2196 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2197 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2198 return kvm_pmu_set_msr(vcpu
, msr_info
);
2200 if (pr
|| data
!= 0)
2201 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
2202 "0x%x data 0x%llx\n", msr
, data
);
2204 case MSR_K7_CLK_CTL
:
2206 * Ignore all writes to this no longer documented MSR.
2207 * Writes are only relevant for old K7 processors,
2208 * all pre-dating SVM, but a recommended workaround from
2209 * AMD for these chips. It is possible to specify the
2210 * affected processor models on the command line, hence
2211 * the need to ignore the workaround.
2214 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2215 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2216 case HV_X64_MSR_CRASH_CTL
:
2217 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2218 return kvm_hv_set_msr_common(vcpu
, msr
, data
,
2219 msr_info
->host_initiated
);
2220 case MSR_IA32_BBL_CR_CTL3
:
2221 /* Drop writes to this legacy MSR -- see rdmsr
2222 * counterpart for further detail.
2224 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n", msr
, data
);
2226 case MSR_AMD64_OSVW_ID_LENGTH
:
2227 if (!guest_cpuid_has_osvw(vcpu
))
2229 vcpu
->arch
.osvw
.length
= data
;
2231 case MSR_AMD64_OSVW_STATUS
:
2232 if (!guest_cpuid_has_osvw(vcpu
))
2234 vcpu
->arch
.osvw
.status
= data
;
2237 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
2238 return xen_hvm_config(vcpu
, data
);
2239 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2240 return kvm_pmu_set_msr(vcpu
, msr_info
);
2242 vcpu_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
2246 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
2253 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
2257 * Reads an msr value (of 'msr_index') into 'pdata'.
2258 * Returns 0 on success, non-0 otherwise.
2259 * Assumes vcpu_load() was already called.
2261 int kvm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
2263 return kvm_x86_ops
->get_msr(vcpu
, msr
);
2265 EXPORT_SYMBOL_GPL(kvm_get_msr
);
2267 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2270 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2271 unsigned bank_num
= mcg_cap
& 0xff;
2274 case MSR_IA32_P5_MC_ADDR
:
2275 case MSR_IA32_P5_MC_TYPE
:
2278 case MSR_IA32_MCG_CAP
:
2279 data
= vcpu
->arch
.mcg_cap
;
2281 case MSR_IA32_MCG_CTL
:
2282 if (!(mcg_cap
& MCG_CTL_P
))
2284 data
= vcpu
->arch
.mcg_ctl
;
2286 case MSR_IA32_MCG_STATUS
:
2287 data
= vcpu
->arch
.mcg_status
;
2290 if (msr
>= MSR_IA32_MC0_CTL
&&
2291 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2292 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2293 data
= vcpu
->arch
.mce_banks
[offset
];
2302 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2304 switch (msr_info
->index
) {
2305 case MSR_IA32_PLATFORM_ID
:
2306 case MSR_IA32_EBL_CR_POWERON
:
2307 case MSR_IA32_DEBUGCTLMSR
:
2308 case MSR_IA32_LASTBRANCHFROMIP
:
2309 case MSR_IA32_LASTBRANCHTOIP
:
2310 case MSR_IA32_LASTINTFROMIP
:
2311 case MSR_IA32_LASTINTTOIP
:
2313 case MSR_K8_TSEG_ADDR
:
2314 case MSR_K8_TSEG_MASK
:
2316 case MSR_VM_HSAVE_PA
:
2317 case MSR_K8_INT_PENDING_MSG
:
2318 case MSR_AMD64_NB_CFG
:
2319 case MSR_FAM10H_MMIO_CONF_BASE
:
2320 case MSR_AMD64_BU_CFG2
:
2321 case MSR_IA32_PERF_CTL
:
2324 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2325 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2326 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2327 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2328 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2329 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2332 case MSR_IA32_UCODE_REV
:
2333 msr_info
->data
= 0x100000000ULL
;
2336 case 0x200 ... 0x2ff:
2337 return kvm_mtrr_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2338 case 0xcd: /* fsb frequency */
2342 * MSR_EBC_FREQUENCY_ID
2343 * Conservative value valid for even the basic CPU models.
2344 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2345 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2346 * and 266MHz for model 3, or 4. Set Core Clock
2347 * Frequency to System Bus Frequency Ratio to 1 (bits
2348 * 31:24) even though these are only valid for CPU
2349 * models > 2, however guests may end up dividing or
2350 * multiplying by zero otherwise.
2352 case MSR_EBC_FREQUENCY_ID
:
2353 msr_info
->data
= 1 << 24;
2355 case MSR_IA32_APICBASE
:
2356 msr_info
->data
= kvm_get_apic_base(vcpu
);
2358 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2359 return kvm_x2apic_msr_read(vcpu
, msr_info
->index
, &msr_info
->data
);
2361 case MSR_IA32_TSCDEADLINE
:
2362 msr_info
->data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
2364 case MSR_IA32_TSC_ADJUST
:
2365 msr_info
->data
= (u64
)vcpu
->arch
.ia32_tsc_adjust_msr
;
2367 case MSR_IA32_MISC_ENABLE
:
2368 msr_info
->data
= vcpu
->arch
.ia32_misc_enable_msr
;
2370 case MSR_IA32_SMBASE
:
2371 if (!msr_info
->host_initiated
)
2373 msr_info
->data
= vcpu
->arch
.smbase
;
2375 case MSR_IA32_PERF_STATUS
:
2376 /* TSC increment by tick */
2377 msr_info
->data
= 1000ULL;
2378 /* CPU multiplier */
2379 msr_info
->data
|= (((uint64_t)4ULL) << 40);
2382 msr_info
->data
= vcpu
->arch
.efer
;
2384 case MSR_KVM_WALL_CLOCK
:
2385 case MSR_KVM_WALL_CLOCK_NEW
:
2386 msr_info
->data
= vcpu
->kvm
->arch
.wall_clock
;
2388 case MSR_KVM_SYSTEM_TIME
:
2389 case MSR_KVM_SYSTEM_TIME_NEW
:
2390 msr_info
->data
= vcpu
->arch
.time
;
2392 case MSR_KVM_ASYNC_PF_EN
:
2393 msr_info
->data
= vcpu
->arch
.apf
.msr_val
;
2395 case MSR_KVM_STEAL_TIME
:
2396 msr_info
->data
= vcpu
->arch
.st
.msr_val
;
2398 case MSR_KVM_PV_EOI_EN
:
2399 msr_info
->data
= vcpu
->arch
.pv_eoi
.msr_val
;
2401 case MSR_IA32_P5_MC_ADDR
:
2402 case MSR_IA32_P5_MC_TYPE
:
2403 case MSR_IA32_MCG_CAP
:
2404 case MSR_IA32_MCG_CTL
:
2405 case MSR_IA32_MCG_STATUS
:
2406 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2407 return get_msr_mce(vcpu
, msr_info
->index
, &msr_info
->data
);
2408 case MSR_K7_CLK_CTL
:
2410 * Provide expected ramp-up count for K7. All other
2411 * are set to zero, indicating minimum divisors for
2414 * This prevents guest kernels on AMD host with CPU
2415 * type 6, model 8 and higher from exploding due to
2416 * the rdmsr failing.
2418 msr_info
->data
= 0x20000000;
2420 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2421 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2422 case HV_X64_MSR_CRASH_CTL
:
2423 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2424 return kvm_hv_get_msr_common(vcpu
,
2425 msr_info
->index
, &msr_info
->data
);
2427 case MSR_IA32_BBL_CR_CTL3
:
2428 /* This legacy MSR exists but isn't fully documented in current
2429 * silicon. It is however accessed by winxp in very narrow
2430 * scenarios where it sets bit #19, itself documented as
2431 * a "reserved" bit. Best effort attempt to source coherent
2432 * read data here should the balance of the register be
2433 * interpreted by the guest:
2435 * L2 cache control register 3: 64GB range, 256KB size,
2436 * enabled, latency 0x1, configured
2438 msr_info
->data
= 0xbe702111;
2440 case MSR_AMD64_OSVW_ID_LENGTH
:
2441 if (!guest_cpuid_has_osvw(vcpu
))
2443 msr_info
->data
= vcpu
->arch
.osvw
.length
;
2445 case MSR_AMD64_OSVW_STATUS
:
2446 if (!guest_cpuid_has_osvw(vcpu
))
2448 msr_info
->data
= vcpu
->arch
.osvw
.status
;
2451 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2452 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2454 vcpu_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr_info
->index
);
2457 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr_info
->index
);
2464 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2467 * Read or write a bunch of msrs. All parameters are kernel addresses.
2469 * @return number of msrs set successfully.
2471 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2472 struct kvm_msr_entry
*entries
,
2473 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2474 unsigned index
, u64
*data
))
2478 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2479 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2480 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2482 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2488 * Read or write a bunch of msrs. Parameters are user addresses.
2490 * @return number of msrs set successfully.
2492 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2493 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2494 unsigned index
, u64
*data
),
2497 struct kvm_msrs msrs
;
2498 struct kvm_msr_entry
*entries
;
2503 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2507 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2510 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2511 entries
= memdup_user(user_msrs
->entries
, size
);
2512 if (IS_ERR(entries
)) {
2513 r
= PTR_ERR(entries
);
2517 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2522 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2533 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
2538 case KVM_CAP_IRQCHIP
:
2540 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2541 case KVM_CAP_SET_TSS_ADDR
:
2542 case KVM_CAP_EXT_CPUID
:
2543 case KVM_CAP_EXT_EMUL_CPUID
:
2544 case KVM_CAP_CLOCKSOURCE
:
2546 case KVM_CAP_NOP_IO_DELAY
:
2547 case KVM_CAP_MP_STATE
:
2548 case KVM_CAP_SYNC_MMU
:
2549 case KVM_CAP_USER_NMI
:
2550 case KVM_CAP_REINJECT_CONTROL
:
2551 case KVM_CAP_IRQ_INJECT_STATUS
:
2552 case KVM_CAP_IOEVENTFD
:
2553 case KVM_CAP_IOEVENTFD_NO_LENGTH
:
2555 case KVM_CAP_PIT_STATE2
:
2556 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2557 case KVM_CAP_XEN_HVM
:
2558 case KVM_CAP_ADJUST_CLOCK
:
2559 case KVM_CAP_VCPU_EVENTS
:
2560 case KVM_CAP_HYPERV
:
2561 case KVM_CAP_HYPERV_VAPIC
:
2562 case KVM_CAP_HYPERV_SPIN
:
2563 case KVM_CAP_HYPERV_SYNIC
:
2564 case KVM_CAP_PCI_SEGMENT
:
2565 case KVM_CAP_DEBUGREGS
:
2566 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2568 case KVM_CAP_ASYNC_PF
:
2569 case KVM_CAP_GET_TSC_KHZ
:
2570 case KVM_CAP_KVMCLOCK_CTRL
:
2571 case KVM_CAP_READONLY_MEM
:
2572 case KVM_CAP_HYPERV_TIME
:
2573 case KVM_CAP_IOAPIC_POLARITY_IGNORED
:
2574 case KVM_CAP_TSC_DEADLINE_TIMER
:
2575 case KVM_CAP_ENABLE_CAP_VM
:
2576 case KVM_CAP_DISABLE_QUIRKS
:
2577 case KVM_CAP_SET_BOOT_CPU_ID
:
2578 case KVM_CAP_SPLIT_IRQCHIP
:
2579 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2580 case KVM_CAP_ASSIGN_DEV_IRQ
:
2581 case KVM_CAP_PCI_2_3
:
2585 case KVM_CAP_X86_SMM
:
2586 /* SMBASE is usually relocated above 1M on modern chipsets,
2587 * and SMM handlers might indeed rely on 4G segment limits,
2588 * so do not report SMM to be available if real mode is
2589 * emulated via vm86 mode. Still, do not go to great lengths
2590 * to avoid userspace's usage of the feature, because it is a
2591 * fringe case that is not enabled except via specific settings
2592 * of the module parameters.
2594 r
= kvm_x86_ops
->cpu_has_high_real_mode_segbase();
2596 case KVM_CAP_COALESCED_MMIO
:
2597 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
2600 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2602 case KVM_CAP_NR_VCPUS
:
2603 r
= KVM_SOFT_MAX_VCPUS
;
2605 case KVM_CAP_MAX_VCPUS
:
2608 case KVM_CAP_NR_MEMSLOTS
:
2609 r
= KVM_USER_MEM_SLOTS
;
2611 case KVM_CAP_PV_MMU
: /* obsolete */
2614 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2616 r
= iommu_present(&pci_bus_type
);
2620 r
= KVM_MAX_MCE_BANKS
;
2623 r
= boot_cpu_has(X86_FEATURE_XSAVE
);
2625 case KVM_CAP_TSC_CONTROL
:
2626 r
= kvm_has_tsc_control
;
2636 long kvm_arch_dev_ioctl(struct file
*filp
,
2637 unsigned int ioctl
, unsigned long arg
)
2639 void __user
*argp
= (void __user
*)arg
;
2643 case KVM_GET_MSR_INDEX_LIST
: {
2644 struct kvm_msr_list __user
*user_msr_list
= argp
;
2645 struct kvm_msr_list msr_list
;
2649 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2652 msr_list
.nmsrs
= num_msrs_to_save
+ num_emulated_msrs
;
2653 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2656 if (n
< msr_list
.nmsrs
)
2659 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2660 num_msrs_to_save
* sizeof(u32
)))
2662 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2664 num_emulated_msrs
* sizeof(u32
)))
2669 case KVM_GET_SUPPORTED_CPUID
:
2670 case KVM_GET_EMULATED_CPUID
: {
2671 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2672 struct kvm_cpuid2 cpuid
;
2675 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2678 r
= kvm_dev_ioctl_get_cpuid(&cpuid
, cpuid_arg
->entries
,
2684 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2689 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2691 if (copy_to_user(argp
, &kvm_mce_cap_supported
,
2692 sizeof(kvm_mce_cap_supported
)))
2704 static void wbinvd_ipi(void *garbage
)
2709 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2711 return kvm_arch_has_noncoherent_dma(vcpu
->kvm
);
2714 static inline void kvm_migrate_timers(struct kvm_vcpu
*vcpu
)
2716 set_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
);
2719 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2721 /* Address WBINVD may be executed by guest */
2722 if (need_emulate_wbinvd(vcpu
)) {
2723 if (kvm_x86_ops
->has_wbinvd_exit())
2724 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2725 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2726 smp_call_function_single(vcpu
->cpu
,
2727 wbinvd_ipi
, NULL
, 1);
2730 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2732 /* Apply any externally detected TSC adjustments (due to suspend) */
2733 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2734 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2735 vcpu
->arch
.tsc_offset_adjustment
= 0;
2736 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2739 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2740 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2741 rdtsc() - vcpu
->arch
.last_host_tsc
;
2743 mark_tsc_unstable("KVM discovered backwards TSC");
2745 if (kvm_lapic_hv_timer_in_use(vcpu
) &&
2746 kvm_x86_ops
->set_hv_timer(vcpu
,
2747 kvm_get_lapic_tscdeadline_msr(vcpu
)))
2748 kvm_lapic_switch_to_sw_timer(vcpu
);
2749 if (check_tsc_unstable()) {
2750 u64 offset
= kvm_compute_tsc_offset(vcpu
,
2751 vcpu
->arch
.last_guest_tsc
);
2752 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
2753 vcpu
->arch
.tsc_catchup
= 1;
2756 * On a host with synchronized TSC, there is no need to update
2757 * kvmclock on vcpu->cpu migration
2759 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
2760 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2761 if (vcpu
->cpu
!= cpu
)
2762 kvm_migrate_timers(vcpu
);
2766 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2769 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2771 kvm_x86_ops
->vcpu_put(vcpu
);
2772 kvm_put_guest_fpu(vcpu
);
2773 vcpu
->arch
.last_host_tsc
= rdtsc();
2776 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
2777 struct kvm_lapic_state
*s
)
2779 if (vcpu
->arch
.apicv_active
)
2780 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
2782 return kvm_apic_get_state(vcpu
, s
);
2785 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
2786 struct kvm_lapic_state
*s
)
2790 r
= kvm_apic_set_state(vcpu
, s
);
2793 update_cr8_intercept(vcpu
);
2798 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu
*vcpu
)
2800 return (!lapic_in_kernel(vcpu
) ||
2801 kvm_apic_accept_pic_intr(vcpu
));
2805 * if userspace requested an interrupt window, check that the
2806 * interrupt window is open.
2808 * No need to exit to userspace if we already have an interrupt queued.
2810 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu
*vcpu
)
2812 return kvm_arch_interrupt_allowed(vcpu
) &&
2813 !kvm_cpu_has_interrupt(vcpu
) &&
2814 !kvm_event_needs_reinjection(vcpu
) &&
2815 kvm_cpu_accept_dm_intr(vcpu
);
2818 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2819 struct kvm_interrupt
*irq
)
2821 if (irq
->irq
>= KVM_NR_INTERRUPTS
)
2824 if (!irqchip_in_kernel(vcpu
->kvm
)) {
2825 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
2826 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2831 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2832 * fail for in-kernel 8259.
2834 if (pic_in_kernel(vcpu
->kvm
))
2837 if (vcpu
->arch
.pending_external_vector
!= -1)
2840 vcpu
->arch
.pending_external_vector
= irq
->irq
;
2841 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2845 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
2847 kvm_inject_nmi(vcpu
);
2852 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu
*vcpu
)
2854 kvm_make_request(KVM_REQ_SMI
, vcpu
);
2859 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
2860 struct kvm_tpr_access_ctl
*tac
)
2864 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
2868 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
2872 unsigned bank_num
= mcg_cap
& 0xff, bank
;
2875 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
2877 if (mcg_cap
& ~(kvm_mce_cap_supported
| 0xff | 0xff0000))
2880 vcpu
->arch
.mcg_cap
= mcg_cap
;
2881 /* Init IA32_MCG_CTL to all 1s */
2882 if (mcg_cap
& MCG_CTL_P
)
2883 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
2884 /* Init IA32_MCi_CTL to all 1s */
2885 for (bank
= 0; bank
< bank_num
; bank
++)
2886 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
2888 if (kvm_x86_ops
->setup_mce
)
2889 kvm_x86_ops
->setup_mce(vcpu
);
2894 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
2895 struct kvm_x86_mce
*mce
)
2897 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2898 unsigned bank_num
= mcg_cap
& 0xff;
2899 u64
*banks
= vcpu
->arch
.mce_banks
;
2901 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
2904 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2905 * reporting is disabled
2907 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
2908 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
2910 banks
+= 4 * mce
->bank
;
2912 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2913 * reporting is disabled for the bank
2915 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
2917 if (mce
->status
& MCI_STATUS_UC
) {
2918 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
2919 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
2920 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2923 if (banks
[1] & MCI_STATUS_VAL
)
2924 mce
->status
|= MCI_STATUS_OVER
;
2925 banks
[2] = mce
->addr
;
2926 banks
[3] = mce
->misc
;
2927 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
2928 banks
[1] = mce
->status
;
2929 kvm_queue_exception(vcpu
, MC_VECTOR
);
2930 } else if (!(banks
[1] & MCI_STATUS_VAL
)
2931 || !(banks
[1] & MCI_STATUS_UC
)) {
2932 if (banks
[1] & MCI_STATUS_VAL
)
2933 mce
->status
|= MCI_STATUS_OVER
;
2934 banks
[2] = mce
->addr
;
2935 banks
[3] = mce
->misc
;
2936 banks
[1] = mce
->status
;
2938 banks
[1] |= MCI_STATUS_OVER
;
2942 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
2943 struct kvm_vcpu_events
*events
)
2946 events
->exception
.injected
=
2947 vcpu
->arch
.exception
.pending
&&
2948 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
2949 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
2950 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
2951 events
->exception
.pad
= 0;
2952 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
2954 events
->interrupt
.injected
=
2955 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
2956 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
2957 events
->interrupt
.soft
= 0;
2958 events
->interrupt
.shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
2960 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
2961 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
2962 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
2963 events
->nmi
.pad
= 0;
2965 events
->sipi_vector
= 0; /* never valid when reporting to user space */
2967 events
->smi
.smm
= is_smm(vcpu
);
2968 events
->smi
.pending
= vcpu
->arch
.smi_pending
;
2969 events
->smi
.smm_inside_nmi
=
2970 !!(vcpu
->arch
.hflags
& HF_SMM_INSIDE_NMI_MASK
);
2971 events
->smi
.latched_init
= kvm_lapic_latched_init(vcpu
);
2973 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
2974 | KVM_VCPUEVENT_VALID_SHADOW
2975 | KVM_VCPUEVENT_VALID_SMM
);
2976 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
2979 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
2980 struct kvm_vcpu_events
*events
)
2982 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2983 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2984 | KVM_VCPUEVENT_VALID_SHADOW
2985 | KVM_VCPUEVENT_VALID_SMM
))
2988 if (events
->exception
.injected
&&
2989 (events
->exception
.nr
> 31 || events
->exception
.nr
== NMI_VECTOR
))
2993 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
2994 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
2995 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
2996 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
2998 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
2999 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
3000 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
3001 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
3002 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
3003 events
->interrupt
.shadow
);
3005 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
3006 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
3007 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
3008 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
3010 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
&&
3011 lapic_in_kernel(vcpu
))
3012 vcpu
->arch
.apic
->sipi_vector
= events
->sipi_vector
;
3014 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
) {
3015 if (events
->smi
.smm
)
3016 vcpu
->arch
.hflags
|= HF_SMM_MASK
;
3018 vcpu
->arch
.hflags
&= ~HF_SMM_MASK
;
3019 vcpu
->arch
.smi_pending
= events
->smi
.pending
;
3020 if (events
->smi
.smm_inside_nmi
)
3021 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
3023 vcpu
->arch
.hflags
&= ~HF_SMM_INSIDE_NMI_MASK
;
3024 if (lapic_in_kernel(vcpu
)) {
3025 if (events
->smi
.latched_init
)
3026 set_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3028 clear_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3032 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3037 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
3038 struct kvm_debugregs
*dbgregs
)
3042 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
3043 kvm_get_dr(vcpu
, 6, &val
);
3045 dbgregs
->dr7
= vcpu
->arch
.dr7
;
3047 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
3050 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
3051 struct kvm_debugregs
*dbgregs
)
3056 if (dbgregs
->dr6
& ~0xffffffffull
)
3058 if (dbgregs
->dr7
& ~0xffffffffull
)
3061 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
3062 kvm_update_dr0123(vcpu
);
3063 vcpu
->arch
.dr6
= dbgregs
->dr6
;
3064 kvm_update_dr6(vcpu
);
3065 vcpu
->arch
.dr7
= dbgregs
->dr7
;
3066 kvm_update_dr7(vcpu
);
3071 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3073 static void fill_xsave(u8
*dest
, struct kvm_vcpu
*vcpu
)
3075 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3076 u64 xstate_bv
= xsave
->header
.xfeatures
;
3080 * Copy legacy XSAVE area, to avoid complications with CPUID
3081 * leaves 0 and 1 in the loop below.
3083 memcpy(dest
, xsave
, XSAVE_HDR_OFFSET
);
3086 *(u64
*)(dest
+ XSAVE_HDR_OFFSET
) = xstate_bv
;
3089 * Copy each region from the possibly compacted offset to the
3090 * non-compacted offset.
3092 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3094 u64 feature
= valid
& -valid
;
3095 int index
= fls64(feature
) - 1;
3096 void *src
= get_xsave_addr(xsave
, feature
);
3099 u32 size
, offset
, ecx
, edx
;
3100 cpuid_count(XSTATE_CPUID
, index
,
3101 &size
, &offset
, &ecx
, &edx
);
3102 memcpy(dest
+ offset
, src
, size
);
3109 static void load_xsave(struct kvm_vcpu
*vcpu
, u8
*src
)
3111 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3112 u64 xstate_bv
= *(u64
*)(src
+ XSAVE_HDR_OFFSET
);
3116 * Copy legacy XSAVE area, to avoid complications with CPUID
3117 * leaves 0 and 1 in the loop below.
3119 memcpy(xsave
, src
, XSAVE_HDR_OFFSET
);
3121 /* Set XSTATE_BV and possibly XCOMP_BV. */
3122 xsave
->header
.xfeatures
= xstate_bv
;
3123 if (boot_cpu_has(X86_FEATURE_XSAVES
))
3124 xsave
->header
.xcomp_bv
= host_xcr0
| XSTATE_COMPACTION_ENABLED
;
3127 * Copy each region from the non-compacted offset to the
3128 * possibly compacted offset.
3130 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3132 u64 feature
= valid
& -valid
;
3133 int index
= fls64(feature
) - 1;
3134 void *dest
= get_xsave_addr(xsave
, feature
);
3137 u32 size
, offset
, ecx
, edx
;
3138 cpuid_count(XSTATE_CPUID
, index
,
3139 &size
, &offset
, &ecx
, &edx
);
3140 memcpy(dest
, src
+ offset
, size
);
3147 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
3148 struct kvm_xsave
*guest_xsave
)
3150 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3151 memset(guest_xsave
, 0, sizeof(struct kvm_xsave
));
3152 fill_xsave((u8
*) guest_xsave
->region
, vcpu
);
3154 memcpy(guest_xsave
->region
,
3155 &vcpu
->arch
.guest_fpu
.state
.fxsave
,
3156 sizeof(struct fxregs_state
));
3157 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
3158 XFEATURE_MASK_FPSSE
;
3162 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
3163 struct kvm_xsave
*guest_xsave
)
3166 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
3168 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3170 * Here we allow setting states that are not present in
3171 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3172 * with old userspace.
3174 if (xstate_bv
& ~kvm_supported_xcr0())
3176 load_xsave(vcpu
, (u8
*)guest_xsave
->region
);
3178 if (xstate_bv
& ~XFEATURE_MASK_FPSSE
)
3180 memcpy(&vcpu
->arch
.guest_fpu
.state
.fxsave
,
3181 guest_xsave
->region
, sizeof(struct fxregs_state
));
3186 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
3187 struct kvm_xcrs
*guest_xcrs
)
3189 if (!boot_cpu_has(X86_FEATURE_XSAVE
)) {
3190 guest_xcrs
->nr_xcrs
= 0;
3194 guest_xcrs
->nr_xcrs
= 1;
3195 guest_xcrs
->flags
= 0;
3196 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
3197 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
3200 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
3201 struct kvm_xcrs
*guest_xcrs
)
3205 if (!boot_cpu_has(X86_FEATURE_XSAVE
))
3208 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
3211 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
3212 /* Only support XCR0 currently */
3213 if (guest_xcrs
->xcrs
[i
].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
3214 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
3215 guest_xcrs
->xcrs
[i
].value
);
3224 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3225 * stopped by the hypervisor. This function will be called from the host only.
3226 * EINVAL is returned when the host attempts to set the flag for a guest that
3227 * does not support pv clocks.
3229 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
3231 if (!vcpu
->arch
.pv_time_enabled
)
3233 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
3234 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
3238 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
3239 struct kvm_enable_cap
*cap
)
3245 case KVM_CAP_HYPERV_SYNIC
:
3246 return kvm_hv_activate_synic(vcpu
);
3252 long kvm_arch_vcpu_ioctl(struct file
*filp
,
3253 unsigned int ioctl
, unsigned long arg
)
3255 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3256 void __user
*argp
= (void __user
*)arg
;
3259 struct kvm_lapic_state
*lapic
;
3260 struct kvm_xsave
*xsave
;
3261 struct kvm_xcrs
*xcrs
;
3267 case KVM_GET_LAPIC
: {
3269 if (!lapic_in_kernel(vcpu
))
3271 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
3276 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
3280 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
3285 case KVM_SET_LAPIC
: {
3287 if (!lapic_in_kernel(vcpu
))
3289 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
3290 if (IS_ERR(u
.lapic
))
3291 return PTR_ERR(u
.lapic
);
3293 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
3296 case KVM_INTERRUPT
: {
3297 struct kvm_interrupt irq
;
3300 if (copy_from_user(&irq
, argp
, sizeof irq
))
3302 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
3306 r
= kvm_vcpu_ioctl_nmi(vcpu
);
3310 r
= kvm_vcpu_ioctl_smi(vcpu
);
3313 case KVM_SET_CPUID
: {
3314 struct kvm_cpuid __user
*cpuid_arg
= argp
;
3315 struct kvm_cpuid cpuid
;
3318 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3320 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3323 case KVM_SET_CPUID2
: {
3324 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3325 struct kvm_cpuid2 cpuid
;
3328 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3330 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
3331 cpuid_arg
->entries
);
3334 case KVM_GET_CPUID2
: {
3335 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3336 struct kvm_cpuid2 cpuid
;
3339 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3341 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
3342 cpuid_arg
->entries
);
3346 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
3352 r
= msr_io(vcpu
, argp
, do_get_msr
, 1);
3355 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
3357 case KVM_TPR_ACCESS_REPORTING
: {
3358 struct kvm_tpr_access_ctl tac
;
3361 if (copy_from_user(&tac
, argp
, sizeof tac
))
3363 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
3367 if (copy_to_user(argp
, &tac
, sizeof tac
))
3372 case KVM_SET_VAPIC_ADDR
: {
3373 struct kvm_vapic_addr va
;
3376 if (!lapic_in_kernel(vcpu
))
3379 if (copy_from_user(&va
, argp
, sizeof va
))
3381 r
= kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
3384 case KVM_X86_SETUP_MCE
: {
3388 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
3390 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
3393 case KVM_X86_SET_MCE
: {
3394 struct kvm_x86_mce mce
;
3397 if (copy_from_user(&mce
, argp
, sizeof mce
))
3399 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
3402 case KVM_GET_VCPU_EVENTS
: {
3403 struct kvm_vcpu_events events
;
3405 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
3408 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
3413 case KVM_SET_VCPU_EVENTS
: {
3414 struct kvm_vcpu_events events
;
3417 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
3420 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
3423 case KVM_GET_DEBUGREGS
: {
3424 struct kvm_debugregs dbgregs
;
3426 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
3429 if (copy_to_user(argp
, &dbgregs
,
3430 sizeof(struct kvm_debugregs
)))
3435 case KVM_SET_DEBUGREGS
: {
3436 struct kvm_debugregs dbgregs
;
3439 if (copy_from_user(&dbgregs
, argp
,
3440 sizeof(struct kvm_debugregs
)))
3443 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3446 case KVM_GET_XSAVE
: {
3447 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3452 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3455 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3460 case KVM_SET_XSAVE
: {
3461 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
3462 if (IS_ERR(u
.xsave
))
3463 return PTR_ERR(u
.xsave
);
3465 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3468 case KVM_GET_XCRS
: {
3469 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3474 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3477 if (copy_to_user(argp
, u
.xcrs
,
3478 sizeof(struct kvm_xcrs
)))
3483 case KVM_SET_XCRS
: {
3484 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
3486 return PTR_ERR(u
.xcrs
);
3488 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3491 case KVM_SET_TSC_KHZ
: {
3495 user_tsc_khz
= (u32
)arg
;
3497 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3500 if (user_tsc_khz
== 0)
3501 user_tsc_khz
= tsc_khz
;
3503 if (!kvm_set_tsc_khz(vcpu
, user_tsc_khz
))
3508 case KVM_GET_TSC_KHZ
: {
3509 r
= vcpu
->arch
.virtual_tsc_khz
;
3512 case KVM_KVMCLOCK_CTRL
: {
3513 r
= kvm_set_guest_paused(vcpu
);
3516 case KVM_ENABLE_CAP
: {
3517 struct kvm_enable_cap cap
;
3520 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3522 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
3533 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
3535 return VM_FAULT_SIGBUS
;
3538 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3542 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3544 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3548 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3551 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3555 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3556 u32 kvm_nr_mmu_pages
)
3558 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3561 mutex_lock(&kvm
->slots_lock
);
3563 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3564 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3566 mutex_unlock(&kvm
->slots_lock
);
3570 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3572 return kvm
->arch
.n_max_mmu_pages
;
3575 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3580 switch (chip
->chip_id
) {
3581 case KVM_IRQCHIP_PIC_MASTER
:
3582 memcpy(&chip
->chip
.pic
,
3583 &pic_irqchip(kvm
)->pics
[0],
3584 sizeof(struct kvm_pic_state
));
3586 case KVM_IRQCHIP_PIC_SLAVE
:
3587 memcpy(&chip
->chip
.pic
,
3588 &pic_irqchip(kvm
)->pics
[1],
3589 sizeof(struct kvm_pic_state
));
3591 case KVM_IRQCHIP_IOAPIC
:
3592 r
= kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3601 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3606 switch (chip
->chip_id
) {
3607 case KVM_IRQCHIP_PIC_MASTER
:
3608 spin_lock(&pic_irqchip(kvm
)->lock
);
3609 memcpy(&pic_irqchip(kvm
)->pics
[0],
3611 sizeof(struct kvm_pic_state
));
3612 spin_unlock(&pic_irqchip(kvm
)->lock
);
3614 case KVM_IRQCHIP_PIC_SLAVE
:
3615 spin_lock(&pic_irqchip(kvm
)->lock
);
3616 memcpy(&pic_irqchip(kvm
)->pics
[1],
3618 sizeof(struct kvm_pic_state
));
3619 spin_unlock(&pic_irqchip(kvm
)->lock
);
3621 case KVM_IRQCHIP_IOAPIC
:
3622 r
= kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3628 kvm_pic_update_irq(pic_irqchip(kvm
));
3632 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3634 struct kvm_kpit_state
*kps
= &kvm
->arch
.vpit
->pit_state
;
3636 BUILD_BUG_ON(sizeof(*ps
) != sizeof(kps
->channels
));
3638 mutex_lock(&kps
->lock
);
3639 memcpy(ps
, &kps
->channels
, sizeof(*ps
));
3640 mutex_unlock(&kps
->lock
);
3644 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3647 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3649 mutex_lock(&pit
->pit_state
.lock
);
3650 memcpy(&pit
->pit_state
.channels
, ps
, sizeof(*ps
));
3651 for (i
= 0; i
< 3; i
++)
3652 kvm_pit_load_count(pit
, i
, ps
->channels
[i
].count
, 0);
3653 mutex_unlock(&pit
->pit_state
.lock
);
3657 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3659 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3660 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3661 sizeof(ps
->channels
));
3662 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3663 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3664 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3668 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3672 u32 prev_legacy
, cur_legacy
;
3673 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3675 mutex_lock(&pit
->pit_state
.lock
);
3676 prev_legacy
= pit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3677 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3678 if (!prev_legacy
&& cur_legacy
)
3680 memcpy(&pit
->pit_state
.channels
, &ps
->channels
,
3681 sizeof(pit
->pit_state
.channels
));
3682 pit
->pit_state
.flags
= ps
->flags
;
3683 for (i
= 0; i
< 3; i
++)
3684 kvm_pit_load_count(pit
, i
, pit
->pit_state
.channels
[i
].count
,
3686 mutex_unlock(&pit
->pit_state
.lock
);
3690 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3691 struct kvm_reinject_control
*control
)
3693 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3698 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3699 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3700 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3702 mutex_lock(&pit
->pit_state
.lock
);
3703 kvm_pit_set_reinject(pit
, control
->pit_reinject
);
3704 mutex_unlock(&pit
->pit_state
.lock
);
3710 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3711 * @kvm: kvm instance
3712 * @log: slot id and address to which we copy the log
3714 * Steps 1-4 below provide general overview of dirty page logging. See
3715 * kvm_get_dirty_log_protect() function description for additional details.
3717 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3718 * always flush the TLB (step 4) even if previous step failed and the dirty
3719 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3720 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3721 * writes will be marked dirty for next log read.
3723 * 1. Take a snapshot of the bit and clear it if needed.
3724 * 2. Write protect the corresponding page.
3725 * 3. Copy the snapshot to the userspace.
3726 * 4. Flush TLB's if needed.
3728 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
3730 bool is_dirty
= false;
3733 mutex_lock(&kvm
->slots_lock
);
3736 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3738 if (kvm_x86_ops
->flush_log_dirty
)
3739 kvm_x86_ops
->flush_log_dirty(kvm
);
3741 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
3744 * All the TLBs can be flushed out of mmu lock, see the comments in
3745 * kvm_mmu_slot_remove_write_access().
3747 lockdep_assert_held(&kvm
->slots_lock
);
3749 kvm_flush_remote_tlbs(kvm
);
3751 mutex_unlock(&kvm
->slots_lock
);
3755 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
3758 if (!irqchip_in_kernel(kvm
))
3761 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
3762 irq_event
->irq
, irq_event
->level
,
3767 static int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
3768 struct kvm_enable_cap
*cap
)
3776 case KVM_CAP_DISABLE_QUIRKS
:
3777 kvm
->arch
.disabled_quirks
= cap
->args
[0];
3780 case KVM_CAP_SPLIT_IRQCHIP
: {
3781 mutex_lock(&kvm
->lock
);
3783 if (cap
->args
[0] > MAX_NR_RESERVED_IOAPIC_PINS
)
3784 goto split_irqchip_unlock
;
3786 if (irqchip_in_kernel(kvm
))
3787 goto split_irqchip_unlock
;
3788 if (kvm
->created_vcpus
)
3789 goto split_irqchip_unlock
;
3790 r
= kvm_setup_empty_irq_routing(kvm
);
3792 goto split_irqchip_unlock
;
3793 /* Pairs with irqchip_in_kernel. */
3795 kvm
->arch
.irqchip_split
= true;
3796 kvm
->arch
.nr_reserved_ioapic_pins
= cap
->args
[0];
3798 split_irqchip_unlock
:
3799 mutex_unlock(&kvm
->lock
);
3809 long kvm_arch_vm_ioctl(struct file
*filp
,
3810 unsigned int ioctl
, unsigned long arg
)
3812 struct kvm
*kvm
= filp
->private_data
;
3813 void __user
*argp
= (void __user
*)arg
;
3816 * This union makes it completely explicit to gcc-3.x
3817 * that these two variables' stack usage should be
3818 * combined, not added together.
3821 struct kvm_pit_state ps
;
3822 struct kvm_pit_state2 ps2
;
3823 struct kvm_pit_config pit_config
;
3827 case KVM_SET_TSS_ADDR
:
3828 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
3830 case KVM_SET_IDENTITY_MAP_ADDR
: {
3834 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
3836 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
3839 case KVM_SET_NR_MMU_PAGES
:
3840 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
3842 case KVM_GET_NR_MMU_PAGES
:
3843 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
3845 case KVM_CREATE_IRQCHIP
: {
3846 struct kvm_pic
*vpic
;
3848 mutex_lock(&kvm
->lock
);
3851 goto create_irqchip_unlock
;
3853 if (kvm
->created_vcpus
)
3854 goto create_irqchip_unlock
;
3856 vpic
= kvm_create_pic(kvm
);
3858 r
= kvm_ioapic_init(kvm
);
3860 mutex_lock(&kvm
->slots_lock
);
3861 kvm_destroy_pic(vpic
);
3862 mutex_unlock(&kvm
->slots_lock
);
3863 goto create_irqchip_unlock
;
3866 goto create_irqchip_unlock
;
3867 r
= kvm_setup_default_irq_routing(kvm
);
3869 mutex_lock(&kvm
->slots_lock
);
3870 mutex_lock(&kvm
->irq_lock
);
3871 kvm_ioapic_destroy(kvm
);
3872 kvm_destroy_pic(vpic
);
3873 mutex_unlock(&kvm
->irq_lock
);
3874 mutex_unlock(&kvm
->slots_lock
);
3875 goto create_irqchip_unlock
;
3877 /* Write kvm->irq_routing before kvm->arch.vpic. */
3879 kvm
->arch
.vpic
= vpic
;
3880 create_irqchip_unlock
:
3881 mutex_unlock(&kvm
->lock
);
3884 case KVM_CREATE_PIT
:
3885 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
3887 case KVM_CREATE_PIT2
:
3889 if (copy_from_user(&u
.pit_config
, argp
,
3890 sizeof(struct kvm_pit_config
)))
3893 mutex_lock(&kvm
->lock
);
3896 goto create_pit_unlock
;
3898 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
3902 mutex_unlock(&kvm
->lock
);
3904 case KVM_GET_IRQCHIP
: {
3905 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3906 struct kvm_irqchip
*chip
;
3908 chip
= memdup_user(argp
, sizeof(*chip
));
3915 if (!irqchip_in_kernel(kvm
) || irqchip_split(kvm
))
3916 goto get_irqchip_out
;
3917 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
3919 goto get_irqchip_out
;
3921 if (copy_to_user(argp
, chip
, sizeof *chip
))
3922 goto get_irqchip_out
;
3928 case KVM_SET_IRQCHIP
: {
3929 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3930 struct kvm_irqchip
*chip
;
3932 chip
= memdup_user(argp
, sizeof(*chip
));
3939 if (!irqchip_in_kernel(kvm
) || irqchip_split(kvm
))
3940 goto set_irqchip_out
;
3941 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
3943 goto set_irqchip_out
;
3951 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
3954 if (!kvm
->arch
.vpit
)
3956 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
3960 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
3967 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
3970 if (!kvm
->arch
.vpit
)
3972 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
3975 case KVM_GET_PIT2
: {
3977 if (!kvm
->arch
.vpit
)
3979 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
3983 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
3988 case KVM_SET_PIT2
: {
3990 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
3993 if (!kvm
->arch
.vpit
)
3995 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
3998 case KVM_REINJECT_CONTROL
: {
3999 struct kvm_reinject_control control
;
4001 if (copy_from_user(&control
, argp
, sizeof(control
)))
4003 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
4006 case KVM_SET_BOOT_CPU_ID
:
4008 mutex_lock(&kvm
->lock
);
4009 if (kvm
->created_vcpus
)
4012 kvm
->arch
.bsp_vcpu_id
= arg
;
4013 mutex_unlock(&kvm
->lock
);
4015 case KVM_XEN_HVM_CONFIG
: {
4017 if (copy_from_user(&kvm
->arch
.xen_hvm_config
, argp
,
4018 sizeof(struct kvm_xen_hvm_config
)))
4021 if (kvm
->arch
.xen_hvm_config
.flags
)
4026 case KVM_SET_CLOCK
: {
4027 struct kvm_clock_data user_ns
;
4032 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
4040 local_irq_disable();
4041 now_ns
= get_kernel_ns();
4042 delta
= user_ns
.clock
- now_ns
;
4044 kvm
->arch
.kvmclock_offset
= delta
;
4045 kvm_gen_update_masterclock(kvm
);
4048 case KVM_GET_CLOCK
: {
4049 struct kvm_clock_data user_ns
;
4052 local_irq_disable();
4053 now_ns
= get_kernel_ns();
4054 user_ns
.clock
= kvm
->arch
.kvmclock_offset
+ now_ns
;
4057 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
4060 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
4065 case KVM_ENABLE_CAP
: {
4066 struct kvm_enable_cap cap
;
4069 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4071 r
= kvm_vm_ioctl_enable_cap(kvm
, &cap
);
4075 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
4081 static void kvm_init_msr_list(void)
4086 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
4087 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
4091 * Even MSRs that are valid in the host may not be exposed
4092 * to the guests in some cases.
4094 switch (msrs_to_save
[i
]) {
4095 case MSR_IA32_BNDCFGS
:
4096 if (!kvm_x86_ops
->mpx_supported())
4100 if (!kvm_x86_ops
->rdtscp_supported())
4108 msrs_to_save
[j
] = msrs_to_save
[i
];
4111 num_msrs_to_save
= j
;
4113 for (i
= j
= 0; i
< ARRAY_SIZE(emulated_msrs
); i
++) {
4114 switch (emulated_msrs
[i
]) {
4115 case MSR_IA32_SMBASE
:
4116 if (!kvm_x86_ops
->cpu_has_high_real_mode_segbase())
4124 emulated_msrs
[j
] = emulated_msrs
[i
];
4127 num_emulated_msrs
= j
;
4130 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
4138 if (!(lapic_in_kernel(vcpu
) &&
4139 !kvm_iodevice_write(vcpu
, &vcpu
->arch
.apic
->dev
, addr
, n
, v
))
4140 && kvm_io_bus_write(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4151 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
4158 if (!(lapic_in_kernel(vcpu
) &&
4159 !kvm_iodevice_read(vcpu
, &vcpu
->arch
.apic
->dev
,
4161 && kvm_io_bus_read(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4163 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, *(u64
*)v
);
4173 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
4174 struct kvm_segment
*var
, int seg
)
4176 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
4179 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
4180 struct kvm_segment
*var
, int seg
)
4182 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
4185 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
,
4186 struct x86_exception
*exception
)
4190 BUG_ON(!mmu_is_nested(vcpu
));
4192 /* NPT walks are always user-walks */
4193 access
|= PFERR_USER_MASK
;
4194 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, exception
);
4199 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
4200 struct x86_exception
*exception
)
4202 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4203 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4206 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
4207 struct x86_exception
*exception
)
4209 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4210 access
|= PFERR_FETCH_MASK
;
4211 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4214 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
4215 struct x86_exception
*exception
)
4217 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4218 access
|= PFERR_WRITE_MASK
;
4219 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4222 /* uses this to access any guest's mapped memory without checking CPL */
4223 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
4224 struct x86_exception
*exception
)
4226 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
4229 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
4230 struct kvm_vcpu
*vcpu
, u32 access
,
4231 struct x86_exception
*exception
)
4234 int r
= X86EMUL_CONTINUE
;
4237 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
4239 unsigned offset
= addr
& (PAGE_SIZE
-1);
4240 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4243 if (gpa
== UNMAPPED_GVA
)
4244 return X86EMUL_PROPAGATE_FAULT
;
4245 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, data
,
4248 r
= X86EMUL_IO_NEEDED
;
4260 /* used for instruction fetching */
4261 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
4262 gva_t addr
, void *val
, unsigned int bytes
,
4263 struct x86_exception
*exception
)
4265 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4266 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4270 /* Inline kvm_read_guest_virt_helper for speed. */
4271 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
|PFERR_FETCH_MASK
,
4273 if (unlikely(gpa
== UNMAPPED_GVA
))
4274 return X86EMUL_PROPAGATE_FAULT
;
4276 offset
= addr
& (PAGE_SIZE
-1);
4277 if (WARN_ON(offset
+ bytes
> PAGE_SIZE
))
4278 bytes
= (unsigned)PAGE_SIZE
- offset
;
4279 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, val
,
4281 if (unlikely(ret
< 0))
4282 return X86EMUL_IO_NEEDED
;
4284 return X86EMUL_CONTINUE
;
4287 int kvm_read_guest_virt(struct x86_emulate_ctxt
*ctxt
,
4288 gva_t addr
, void *val
, unsigned int bytes
,
4289 struct x86_exception
*exception
)
4291 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4292 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4294 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
4297 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
4299 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
4300 gva_t addr
, void *val
, unsigned int bytes
,
4301 struct x86_exception
*exception
)
4303 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4304 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, 0, exception
);
4307 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt
*ctxt
,
4308 unsigned long addr
, void *val
, unsigned int bytes
)
4310 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4311 int r
= kvm_vcpu_read_guest(vcpu
, addr
, val
, bytes
);
4313 return r
< 0 ? X86EMUL_IO_NEEDED
: X86EMUL_CONTINUE
;
4316 int kvm_write_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
4317 gva_t addr
, void *val
,
4319 struct x86_exception
*exception
)
4321 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4323 int r
= X86EMUL_CONTINUE
;
4326 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
4329 unsigned offset
= addr
& (PAGE_SIZE
-1);
4330 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4333 if (gpa
== UNMAPPED_GVA
)
4334 return X86EMUL_PROPAGATE_FAULT
;
4335 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, data
, towrite
);
4337 r
= X86EMUL_IO_NEEDED
;
4348 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
4350 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4351 gpa_t
*gpa
, struct x86_exception
*exception
,
4354 u32 access
= ((kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
4355 | (write
? PFERR_WRITE_MASK
: 0);
4358 * currently PKRU is only applied to ept enabled guest so
4359 * there is no pkey in EPT page table for L1 guest or EPT
4360 * shadow page table for L2 guest.
4362 if (vcpu_match_mmio_gva(vcpu
, gva
)
4363 && !permission_fault(vcpu
, vcpu
->arch
.walk_mmu
,
4364 vcpu
->arch
.access
, 0, access
)) {
4365 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
4366 (gva
& (PAGE_SIZE
- 1));
4367 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
4371 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4373 if (*gpa
== UNMAPPED_GVA
)
4376 /* For APIC access vmexit */
4377 if ((*gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4380 if (vcpu_match_mmio_gpa(vcpu
, *gpa
)) {
4381 trace_vcpu_match_mmio(gva
, *gpa
, write
, true);
4388 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4389 const void *val
, int bytes
)
4393 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, val
, bytes
);
4396 kvm_page_track_write(vcpu
, gpa
, val
, bytes
);
4400 struct read_write_emulator_ops
{
4401 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
4403 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4404 void *val
, int bytes
);
4405 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4406 int bytes
, void *val
);
4407 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4408 void *val
, int bytes
);
4412 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
4414 if (vcpu
->mmio_read_completed
) {
4415 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
4416 vcpu
->mmio_fragments
[0].gpa
, *(u64
*)val
);
4417 vcpu
->mmio_read_completed
= 0;
4424 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4425 void *val
, int bytes
)
4427 return !kvm_vcpu_read_guest(vcpu
, gpa
, val
, bytes
);
4430 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4431 void *val
, int bytes
)
4433 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
4436 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
4438 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
4439 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
4442 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4443 void *val
, int bytes
)
4445 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
4446 return X86EMUL_IO_NEEDED
;
4449 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4450 void *val
, int bytes
)
4452 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
4454 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
4455 return X86EMUL_CONTINUE
;
4458 static const struct read_write_emulator_ops read_emultor
= {
4459 .read_write_prepare
= read_prepare
,
4460 .read_write_emulate
= read_emulate
,
4461 .read_write_mmio
= vcpu_mmio_read
,
4462 .read_write_exit_mmio
= read_exit_mmio
,
4465 static const struct read_write_emulator_ops write_emultor
= {
4466 .read_write_emulate
= write_emulate
,
4467 .read_write_mmio
= write_mmio
,
4468 .read_write_exit_mmio
= write_exit_mmio
,
4472 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
4474 struct x86_exception
*exception
,
4475 struct kvm_vcpu
*vcpu
,
4476 const struct read_write_emulator_ops
*ops
)
4480 bool write
= ops
->write
;
4481 struct kvm_mmio_fragment
*frag
;
4483 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
4486 return X86EMUL_PROPAGATE_FAULT
;
4488 /* For APIC access vmexit */
4492 if (ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
4493 return X86EMUL_CONTINUE
;
4497 * Is this MMIO handled locally?
4499 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
4500 if (handled
== bytes
)
4501 return X86EMUL_CONTINUE
;
4507 WARN_ON(vcpu
->mmio_nr_fragments
>= KVM_MAX_MMIO_FRAGMENTS
);
4508 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
4512 return X86EMUL_CONTINUE
;
4515 static int emulator_read_write(struct x86_emulate_ctxt
*ctxt
,
4517 void *val
, unsigned int bytes
,
4518 struct x86_exception
*exception
,
4519 const struct read_write_emulator_ops
*ops
)
4521 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4525 if (ops
->read_write_prepare
&&
4526 ops
->read_write_prepare(vcpu
, val
, bytes
))
4527 return X86EMUL_CONTINUE
;
4529 vcpu
->mmio_nr_fragments
= 0;
4531 /* Crossing a page boundary? */
4532 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
4535 now
= -addr
& ~PAGE_MASK
;
4536 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
4539 if (rc
!= X86EMUL_CONTINUE
)
4542 if (ctxt
->mode
!= X86EMUL_MODE_PROT64
)
4548 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
4550 if (rc
!= X86EMUL_CONTINUE
)
4553 if (!vcpu
->mmio_nr_fragments
)
4556 gpa
= vcpu
->mmio_fragments
[0].gpa
;
4558 vcpu
->mmio_needed
= 1;
4559 vcpu
->mmio_cur_fragment
= 0;
4561 vcpu
->run
->mmio
.len
= min(8u, vcpu
->mmio_fragments
[0].len
);
4562 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
4563 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
4564 vcpu
->run
->mmio
.phys_addr
= gpa
;
4566 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
4569 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
4573 struct x86_exception
*exception
)
4575 return emulator_read_write(ctxt
, addr
, val
, bytes
,
4576 exception
, &read_emultor
);
4579 static int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
4583 struct x86_exception
*exception
)
4585 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
4586 exception
, &write_emultor
);
4589 #define CMPXCHG_TYPE(t, ptr, old, new) \
4590 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4592 #ifdef CONFIG_X86_64
4593 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4595 # define CMPXCHG64(ptr, old, new) \
4596 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4599 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
4604 struct x86_exception
*exception
)
4606 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4612 /* guests cmpxchg8b have to be emulated atomically */
4613 if (bytes
> 8 || (bytes
& (bytes
- 1)))
4616 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
4618 if (gpa
== UNMAPPED_GVA
||
4619 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4622 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
4625 page
= kvm_vcpu_gfn_to_page(vcpu
, gpa
>> PAGE_SHIFT
);
4626 if (is_error_page(page
))
4629 kaddr
= kmap_atomic(page
);
4630 kaddr
+= offset_in_page(gpa
);
4633 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
4636 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
4639 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
4642 exchanged
= CMPXCHG64(kaddr
, old
, new);
4647 kunmap_atomic(kaddr
);
4648 kvm_release_page_dirty(page
);
4651 return X86EMUL_CMPXCHG_FAILED
;
4653 kvm_vcpu_mark_page_dirty(vcpu
, gpa
>> PAGE_SHIFT
);
4654 kvm_page_track_write(vcpu
, gpa
, new, bytes
);
4656 return X86EMUL_CONTINUE
;
4659 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
4661 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
4664 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
4666 /* TODO: String I/O for in kernel device */
4669 if (vcpu
->arch
.pio
.in
)
4670 r
= kvm_io_bus_read(vcpu
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
4671 vcpu
->arch
.pio
.size
, pd
);
4673 r
= kvm_io_bus_write(vcpu
, KVM_PIO_BUS
,
4674 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
4679 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
4680 unsigned short port
, void *val
,
4681 unsigned int count
, bool in
)
4683 vcpu
->arch
.pio
.port
= port
;
4684 vcpu
->arch
.pio
.in
= in
;
4685 vcpu
->arch
.pio
.count
= count
;
4686 vcpu
->arch
.pio
.size
= size
;
4688 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
4689 vcpu
->arch
.pio
.count
= 0;
4693 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
4694 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
4695 vcpu
->run
->io
.size
= size
;
4696 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
4697 vcpu
->run
->io
.count
= count
;
4698 vcpu
->run
->io
.port
= port
;
4703 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
4704 int size
, unsigned short port
, void *val
,
4707 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4710 if (vcpu
->arch
.pio
.count
)
4713 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
4716 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
4717 trace_kvm_pio(KVM_PIO_IN
, port
, size
, count
, vcpu
->arch
.pio_data
);
4718 vcpu
->arch
.pio
.count
= 0;
4725 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
4726 int size
, unsigned short port
,
4727 const void *val
, unsigned int count
)
4729 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4731 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
4732 trace_kvm_pio(KVM_PIO_OUT
, port
, size
, count
, vcpu
->arch
.pio_data
);
4733 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
4736 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
4738 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
4741 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
4743 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
4746 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu
*vcpu
)
4748 if (!need_emulate_wbinvd(vcpu
))
4749 return X86EMUL_CONTINUE
;
4751 if (kvm_x86_ops
->has_wbinvd_exit()) {
4752 int cpu
= get_cpu();
4754 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
4755 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
4756 wbinvd_ipi
, NULL
, 1);
4758 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
4761 return X86EMUL_CONTINUE
;
4764 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
4766 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
4767 return kvm_emulate_wbinvd_noskip(vcpu
);
4769 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
4773 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
4775 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt
));
4778 static int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
4779 unsigned long *dest
)
4781 return kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
4784 static int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
4785 unsigned long value
)
4788 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
4791 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
4793 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
4796 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
4798 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4799 unsigned long value
;
4803 value
= kvm_read_cr0(vcpu
);
4806 value
= vcpu
->arch
.cr2
;
4809 value
= kvm_read_cr3(vcpu
);
4812 value
= kvm_read_cr4(vcpu
);
4815 value
= kvm_get_cr8(vcpu
);
4818 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4825 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
4827 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4832 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
4835 vcpu
->arch
.cr2
= val
;
4838 res
= kvm_set_cr3(vcpu
, val
);
4841 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
4844 res
= kvm_set_cr8(vcpu
, val
);
4847 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4854 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
4856 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
4859 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4861 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
4864 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4866 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
4869 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4871 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
4874 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4876 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
4879 static unsigned long emulator_get_cached_segment_base(
4880 struct x86_emulate_ctxt
*ctxt
, int seg
)
4882 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
4885 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
4886 struct desc_struct
*desc
, u32
*base3
,
4889 struct kvm_segment var
;
4891 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
4892 *selector
= var
.selector
;
4895 memset(desc
, 0, sizeof(*desc
));
4901 set_desc_limit(desc
, var
.limit
);
4902 set_desc_base(desc
, (unsigned long)var
.base
);
4903 #ifdef CONFIG_X86_64
4905 *base3
= var
.base
>> 32;
4907 desc
->type
= var
.type
;
4909 desc
->dpl
= var
.dpl
;
4910 desc
->p
= var
.present
;
4911 desc
->avl
= var
.avl
;
4919 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
4920 struct desc_struct
*desc
, u32 base3
,
4923 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4924 struct kvm_segment var
;
4926 var
.selector
= selector
;
4927 var
.base
= get_desc_base(desc
);
4928 #ifdef CONFIG_X86_64
4929 var
.base
|= ((u64
)base3
) << 32;
4931 var
.limit
= get_desc_limit(desc
);
4933 var
.limit
= (var
.limit
<< 12) | 0xfff;
4934 var
.type
= desc
->type
;
4935 var
.dpl
= desc
->dpl
;
4940 var
.avl
= desc
->avl
;
4941 var
.present
= desc
->p
;
4942 var
.unusable
= !var
.present
;
4945 kvm_set_segment(vcpu
, &var
, seg
);
4949 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
4950 u32 msr_index
, u64
*pdata
)
4952 struct msr_data msr
;
4955 msr
.index
= msr_index
;
4956 msr
.host_initiated
= false;
4957 r
= kvm_get_msr(emul_to_vcpu(ctxt
), &msr
);
4965 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
4966 u32 msr_index
, u64 data
)
4968 struct msr_data msr
;
4971 msr
.index
= msr_index
;
4972 msr
.host_initiated
= false;
4973 return kvm_set_msr(emul_to_vcpu(ctxt
), &msr
);
4976 static u64
emulator_get_smbase(struct x86_emulate_ctxt
*ctxt
)
4978 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4980 return vcpu
->arch
.smbase
;
4983 static void emulator_set_smbase(struct x86_emulate_ctxt
*ctxt
, u64 smbase
)
4985 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4987 vcpu
->arch
.smbase
= smbase
;
4990 static int emulator_check_pmc(struct x86_emulate_ctxt
*ctxt
,
4993 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt
), pmc
);
4996 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
4997 u32 pmc
, u64
*pdata
)
4999 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
5002 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
5004 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
5007 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
5010 kvm_load_guest_fpu(emul_to_vcpu(ctxt
));
5012 * CR0.TS may reference the host fpu state, not the guest fpu state,
5013 * so it may be clear at this point.
5018 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
5023 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
5024 struct x86_instruction_info
*info
,
5025 enum x86_intercept_stage stage
)
5027 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
5030 static void emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
5031 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
)
5033 kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
);
5036 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
5038 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
5041 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
5043 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
5046 static void emulator_set_nmi_mask(struct x86_emulate_ctxt
*ctxt
, bool masked
)
5048 kvm_x86_ops
->set_nmi_mask(emul_to_vcpu(ctxt
), masked
);
5051 static const struct x86_emulate_ops emulate_ops
= {
5052 .read_gpr
= emulator_read_gpr
,
5053 .write_gpr
= emulator_write_gpr
,
5054 .read_std
= kvm_read_guest_virt_system
,
5055 .write_std
= kvm_write_guest_virt_system
,
5056 .read_phys
= kvm_read_guest_phys_system
,
5057 .fetch
= kvm_fetch_guest_virt
,
5058 .read_emulated
= emulator_read_emulated
,
5059 .write_emulated
= emulator_write_emulated
,
5060 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
5061 .invlpg
= emulator_invlpg
,
5062 .pio_in_emulated
= emulator_pio_in_emulated
,
5063 .pio_out_emulated
= emulator_pio_out_emulated
,
5064 .get_segment
= emulator_get_segment
,
5065 .set_segment
= emulator_set_segment
,
5066 .get_cached_segment_base
= emulator_get_cached_segment_base
,
5067 .get_gdt
= emulator_get_gdt
,
5068 .get_idt
= emulator_get_idt
,
5069 .set_gdt
= emulator_set_gdt
,
5070 .set_idt
= emulator_set_idt
,
5071 .get_cr
= emulator_get_cr
,
5072 .set_cr
= emulator_set_cr
,
5073 .cpl
= emulator_get_cpl
,
5074 .get_dr
= emulator_get_dr
,
5075 .set_dr
= emulator_set_dr
,
5076 .get_smbase
= emulator_get_smbase
,
5077 .set_smbase
= emulator_set_smbase
,
5078 .set_msr
= emulator_set_msr
,
5079 .get_msr
= emulator_get_msr
,
5080 .check_pmc
= emulator_check_pmc
,
5081 .read_pmc
= emulator_read_pmc
,
5082 .halt
= emulator_halt
,
5083 .wbinvd
= emulator_wbinvd
,
5084 .fix_hypercall
= emulator_fix_hypercall
,
5085 .get_fpu
= emulator_get_fpu
,
5086 .put_fpu
= emulator_put_fpu
,
5087 .intercept
= emulator_intercept
,
5088 .get_cpuid
= emulator_get_cpuid
,
5089 .set_nmi_mask
= emulator_set_nmi_mask
,
5092 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
5094 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
5096 * an sti; sti; sequence only disable interrupts for the first
5097 * instruction. So, if the last instruction, be it emulated or
5098 * not, left the system with the INT_STI flag enabled, it
5099 * means that the last instruction is an sti. We should not
5100 * leave the flag on in this case. The same goes for mov ss
5102 if (int_shadow
& mask
)
5104 if (unlikely(int_shadow
|| mask
)) {
5105 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
5107 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5111 static bool inject_emulated_exception(struct kvm_vcpu
*vcpu
)
5113 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5114 if (ctxt
->exception
.vector
== PF_VECTOR
)
5115 return kvm_propagate_fault(vcpu
, &ctxt
->exception
);
5117 if (ctxt
->exception
.error_code_valid
)
5118 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
5119 ctxt
->exception
.error_code
);
5121 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
5125 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
5127 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5130 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
5132 ctxt
->eflags
= kvm_get_rflags(vcpu
);
5133 ctxt
->eip
= kvm_rip_read(vcpu
);
5134 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
5135 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
5136 (cs_l
&& is_long_mode(vcpu
)) ? X86EMUL_MODE_PROT64
:
5137 cs_db
? X86EMUL_MODE_PROT32
:
5138 X86EMUL_MODE_PROT16
;
5139 BUILD_BUG_ON(HF_GUEST_MASK
!= X86EMUL_GUEST_MASK
);
5140 BUILD_BUG_ON(HF_SMM_MASK
!= X86EMUL_SMM_MASK
);
5141 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK
!= X86EMUL_SMM_INSIDE_NMI_MASK
);
5142 ctxt
->emul_flags
= vcpu
->arch
.hflags
;
5144 init_decode_cache(ctxt
);
5145 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5148 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
5150 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5153 init_emulate_ctxt(vcpu
);
5157 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
5158 ret
= emulate_int_real(ctxt
, irq
);
5160 if (ret
!= X86EMUL_CONTINUE
)
5161 return EMULATE_FAIL
;
5163 ctxt
->eip
= ctxt
->_eip
;
5164 kvm_rip_write(vcpu
, ctxt
->eip
);
5165 kvm_set_rflags(vcpu
, ctxt
->eflags
);
5167 if (irq
== NMI_VECTOR
)
5168 vcpu
->arch
.nmi_pending
= 0;
5170 vcpu
->arch
.interrupt
.pending
= false;
5172 return EMULATE_DONE
;
5174 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
5176 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
5178 int r
= EMULATE_DONE
;
5180 ++vcpu
->stat
.insn_emulation_fail
;
5181 trace_kvm_emulate_insn_failed(vcpu
);
5182 if (!is_guest_mode(vcpu
) && kvm_x86_ops
->get_cpl(vcpu
) == 0) {
5183 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
5184 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
5185 vcpu
->run
->internal
.ndata
= 0;
5188 kvm_queue_exception(vcpu
, UD_VECTOR
);
5193 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t cr2
,
5194 bool write_fault_to_shadow_pgtable
,
5200 if (emulation_type
& EMULTYPE_NO_REEXECUTE
)
5203 if (!vcpu
->arch
.mmu
.direct_map
) {
5205 * Write permission should be allowed since only
5206 * write access need to be emulated.
5208 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5211 * If the mapping is invalid in guest, let cpu retry
5212 * it to generate fault.
5214 if (gpa
== UNMAPPED_GVA
)
5219 * Do not retry the unhandleable instruction if it faults on the
5220 * readonly host memory, otherwise it will goto a infinite loop:
5221 * retry instruction -> write #PF -> emulation fail -> retry
5222 * instruction -> ...
5224 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
5227 * If the instruction failed on the error pfn, it can not be fixed,
5228 * report the error to userspace.
5230 if (is_error_noslot_pfn(pfn
))
5233 kvm_release_pfn_clean(pfn
);
5235 /* The instructions are well-emulated on direct mmu. */
5236 if (vcpu
->arch
.mmu
.direct_map
) {
5237 unsigned int indirect_shadow_pages
;
5239 spin_lock(&vcpu
->kvm
->mmu_lock
);
5240 indirect_shadow_pages
= vcpu
->kvm
->arch
.indirect_shadow_pages
;
5241 spin_unlock(&vcpu
->kvm
->mmu_lock
);
5243 if (indirect_shadow_pages
)
5244 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5250 * if emulation was due to access to shadowed page table
5251 * and it failed try to unshadow page and re-enter the
5252 * guest to let CPU execute the instruction.
5254 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5257 * If the access faults on its page table, it can not
5258 * be fixed by unprotecting shadow page and it should
5259 * be reported to userspace.
5261 return !write_fault_to_shadow_pgtable
;
5264 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
5265 unsigned long cr2
, int emulation_type
)
5267 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5268 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
5270 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
5271 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
5274 * If the emulation is caused by #PF and it is non-page_table
5275 * writing instruction, it means the VM-EXIT is caused by shadow
5276 * page protected, we can zap the shadow page and retry this
5277 * instruction directly.
5279 * Note: if the guest uses a non-page-table modifying instruction
5280 * on the PDE that points to the instruction, then we will unmap
5281 * the instruction and go to an infinite loop. So, we cache the
5282 * last retried eip and the last fault address, if we meet the eip
5283 * and the address again, we can break out of the potential infinite
5286 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
5288 if (!(emulation_type
& EMULTYPE_RETRY
))
5291 if (x86_page_table_writing_insn(ctxt
))
5294 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
5297 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
5298 vcpu
->arch
.last_retry_addr
= cr2
;
5300 if (!vcpu
->arch
.mmu
.direct_map
)
5301 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5303 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5308 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
5309 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
5311 static void kvm_smm_changed(struct kvm_vcpu
*vcpu
)
5313 if (!(vcpu
->arch
.hflags
& HF_SMM_MASK
)) {
5314 /* This is a good place to trace that we are exiting SMM. */
5315 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, false);
5317 /* Process a latched INIT or SMI, if any. */
5318 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5321 kvm_mmu_reset_context(vcpu
);
5324 static void kvm_set_hflags(struct kvm_vcpu
*vcpu
, unsigned emul_flags
)
5326 unsigned changed
= vcpu
->arch
.hflags
^ emul_flags
;
5328 vcpu
->arch
.hflags
= emul_flags
;
5330 if (changed
& HF_SMM_MASK
)
5331 kvm_smm_changed(vcpu
);
5334 static int kvm_vcpu_check_hw_bp(unsigned long addr
, u32 type
, u32 dr7
,
5343 for (i
= 0; i
< 4; i
++, enable
>>= 2, rwlen
>>= 4)
5344 if ((enable
& 3) && (rwlen
& 15) == type
&& db
[i
] == addr
)
5349 static void kvm_vcpu_check_singlestep(struct kvm_vcpu
*vcpu
, unsigned long rflags
, int *r
)
5351 struct kvm_run
*kvm_run
= vcpu
->run
;
5354 * rflags is the old, "raw" value of the flags. The new value has
5355 * not been saved yet.
5357 * This is correct even for TF set by the guest, because "the
5358 * processor will not generate this exception after the instruction
5359 * that sets the TF flag".
5361 if (unlikely(rflags
& X86_EFLAGS_TF
)) {
5362 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
) {
5363 kvm_run
->debug
.arch
.dr6
= DR6_BS
| DR6_FIXED_1
|
5365 kvm_run
->debug
.arch
.pc
= vcpu
->arch
.singlestep_rip
;
5366 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5367 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5368 *r
= EMULATE_USER_EXIT
;
5370 vcpu
->arch
.emulate_ctxt
.eflags
&= ~X86_EFLAGS_TF
;
5372 * "Certain debug exceptions may clear bit 0-3. The
5373 * remaining contents of the DR6 register are never
5374 * cleared by the processor".
5376 vcpu
->arch
.dr6
&= ~15;
5377 vcpu
->arch
.dr6
|= DR6_BS
| DR6_RTM
;
5378 kvm_queue_exception(vcpu
, DB_VECTOR
);
5383 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu
*vcpu
, int *r
)
5385 if (unlikely(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) &&
5386 (vcpu
->arch
.guest_debug_dr7
& DR7_BP_EN_MASK
)) {
5387 struct kvm_run
*kvm_run
= vcpu
->run
;
5388 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5389 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5390 vcpu
->arch
.guest_debug_dr7
,
5394 kvm_run
->debug
.arch
.dr6
= dr6
| DR6_FIXED_1
| DR6_RTM
;
5395 kvm_run
->debug
.arch
.pc
= eip
;
5396 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5397 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5398 *r
= EMULATE_USER_EXIT
;
5403 if (unlikely(vcpu
->arch
.dr7
& DR7_BP_EN_MASK
) &&
5404 !(kvm_get_rflags(vcpu
) & X86_EFLAGS_RF
)) {
5405 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5406 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5411 vcpu
->arch
.dr6
&= ~15;
5412 vcpu
->arch
.dr6
|= dr6
| DR6_RTM
;
5413 kvm_queue_exception(vcpu
, DB_VECTOR
);
5422 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
5429 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5430 bool writeback
= true;
5431 bool write_fault_to_spt
= vcpu
->arch
.write_fault_to_shadow_pgtable
;
5434 * Clear write_fault_to_shadow_pgtable here to ensure it is
5437 vcpu
->arch
.write_fault_to_shadow_pgtable
= false;
5438 kvm_clear_exception_queue(vcpu
);
5440 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
5441 init_emulate_ctxt(vcpu
);
5444 * We will reenter on the same instruction since
5445 * we do not set complete_userspace_io. This does not
5446 * handle watchpoints yet, those would be handled in
5449 if (kvm_vcpu_check_breakpoint(vcpu
, &r
))
5452 ctxt
->interruptibility
= 0;
5453 ctxt
->have_exception
= false;
5454 ctxt
->exception
.vector
= -1;
5455 ctxt
->perm_ok
= false;
5457 ctxt
->ud
= emulation_type
& EMULTYPE_TRAP_UD
;
5459 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
5461 trace_kvm_emulate_insn_start(vcpu
);
5462 ++vcpu
->stat
.insn_emulation
;
5463 if (r
!= EMULATION_OK
) {
5464 if (emulation_type
& EMULTYPE_TRAP_UD
)
5465 return EMULATE_FAIL
;
5466 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5468 return EMULATE_DONE
;
5469 if (emulation_type
& EMULTYPE_SKIP
)
5470 return EMULATE_FAIL
;
5471 return handle_emulation_failure(vcpu
);
5475 if (emulation_type
& EMULTYPE_SKIP
) {
5476 kvm_rip_write(vcpu
, ctxt
->_eip
);
5477 if (ctxt
->eflags
& X86_EFLAGS_RF
)
5478 kvm_set_rflags(vcpu
, ctxt
->eflags
& ~X86_EFLAGS_RF
);
5479 return EMULATE_DONE
;
5482 if (retry_instruction(ctxt
, cr2
, emulation_type
))
5483 return EMULATE_DONE
;
5485 /* this is needed for vmware backdoor interface to work since it
5486 changes registers values during IO operation */
5487 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
5488 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5489 emulator_invalidate_register_cache(ctxt
);
5493 r
= x86_emulate_insn(ctxt
);
5495 if (r
== EMULATION_INTERCEPTED
)
5496 return EMULATE_DONE
;
5498 if (r
== EMULATION_FAILED
) {
5499 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5501 return EMULATE_DONE
;
5503 return handle_emulation_failure(vcpu
);
5506 if (ctxt
->have_exception
) {
5508 if (inject_emulated_exception(vcpu
))
5510 } else if (vcpu
->arch
.pio
.count
) {
5511 if (!vcpu
->arch
.pio
.in
) {
5512 /* FIXME: return into emulator if single-stepping. */
5513 vcpu
->arch
.pio
.count
= 0;
5516 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
5518 r
= EMULATE_USER_EXIT
;
5519 } else if (vcpu
->mmio_needed
) {
5520 if (!vcpu
->mmio_is_write
)
5522 r
= EMULATE_USER_EXIT
;
5523 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
5524 } else if (r
== EMULATION_RESTART
)
5530 unsigned long rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5531 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
5532 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5533 if (vcpu
->arch
.hflags
!= ctxt
->emul_flags
)
5534 kvm_set_hflags(vcpu
, ctxt
->emul_flags
);
5535 kvm_rip_write(vcpu
, ctxt
->eip
);
5536 if (r
== EMULATE_DONE
)
5537 kvm_vcpu_check_singlestep(vcpu
, rflags
, &r
);
5538 if (!ctxt
->have_exception
||
5539 exception_type(ctxt
->exception
.vector
) == EXCPT_TRAP
)
5540 __kvm_set_rflags(vcpu
, ctxt
->eflags
);
5543 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5544 * do nothing, and it will be requested again as soon as
5545 * the shadow expires. But we still need to check here,
5546 * because POPF has no interrupt shadow.
5548 if (unlikely((ctxt
->eflags
& ~rflags
) & X86_EFLAGS_IF
))
5549 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5551 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
5555 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
5557 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5559 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5560 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
5561 size
, port
, &val
, 1);
5562 /* do not return to emulator after return from userspace */
5563 vcpu
->arch
.pio
.count
= 0;
5566 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
5568 static void tsc_bad(void *info
)
5570 __this_cpu_write(cpu_tsc_khz
, 0);
5573 static void tsc_khz_changed(void *data
)
5575 struct cpufreq_freqs
*freq
= data
;
5576 unsigned long khz
= 0;
5580 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5581 khz
= cpufreq_quick_get(raw_smp_processor_id());
5584 __this_cpu_write(cpu_tsc_khz
, khz
);
5587 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
5590 struct cpufreq_freqs
*freq
= data
;
5592 struct kvm_vcpu
*vcpu
;
5593 int i
, send_ipi
= 0;
5596 * We allow guests to temporarily run on slowing clocks,
5597 * provided we notify them after, or to run on accelerating
5598 * clocks, provided we notify them before. Thus time never
5601 * However, we have a problem. We can't atomically update
5602 * the frequency of a given CPU from this function; it is
5603 * merely a notifier, which can be called from any CPU.
5604 * Changing the TSC frequency at arbitrary points in time
5605 * requires a recomputation of local variables related to
5606 * the TSC for each VCPU. We must flag these local variables
5607 * to be updated and be sure the update takes place with the
5608 * new frequency before any guests proceed.
5610 * Unfortunately, the combination of hotplug CPU and frequency
5611 * change creates an intractable locking scenario; the order
5612 * of when these callouts happen is undefined with respect to
5613 * CPU hotplug, and they can race with each other. As such,
5614 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5615 * undefined; you can actually have a CPU frequency change take
5616 * place in between the computation of X and the setting of the
5617 * variable. To protect against this problem, all updates of
5618 * the per_cpu tsc_khz variable are done in an interrupt
5619 * protected IPI, and all callers wishing to update the value
5620 * must wait for a synchronous IPI to complete (which is trivial
5621 * if the caller is on the CPU already). This establishes the
5622 * necessary total order on variable updates.
5624 * Note that because a guest time update may take place
5625 * anytime after the setting of the VCPU's request bit, the
5626 * correct TSC value must be set before the request. However,
5627 * to ensure the update actually makes it to any guest which
5628 * starts running in hardware virtualization between the set
5629 * and the acquisition of the spinlock, we must also ping the
5630 * CPU after setting the request bit.
5634 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
5636 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
5639 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5641 spin_lock(&kvm_lock
);
5642 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5643 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
5644 if (vcpu
->cpu
!= freq
->cpu
)
5646 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5647 if (vcpu
->cpu
!= smp_processor_id())
5651 spin_unlock(&kvm_lock
);
5653 if (freq
->old
< freq
->new && send_ipi
) {
5655 * We upscale the frequency. Must make the guest
5656 * doesn't see old kvmclock values while running with
5657 * the new frequency, otherwise we risk the guest sees
5658 * time go backwards.
5660 * In case we update the frequency for another cpu
5661 * (which might be in guest context) send an interrupt
5662 * to kick the cpu out of guest context. Next time
5663 * guest context is entered kvmclock will be updated,
5664 * so the guest will not see stale values.
5666 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5671 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
5672 .notifier_call
= kvmclock_cpufreq_notifier
5675 static int kvmclock_cpu_notifier(struct notifier_block
*nfb
,
5676 unsigned long action
, void *hcpu
)
5678 unsigned int cpu
= (unsigned long)hcpu
;
5682 case CPU_DOWN_FAILED
:
5683 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5685 case CPU_DOWN_PREPARE
:
5686 smp_call_function_single(cpu
, tsc_bad
, NULL
, 1);
5692 static struct notifier_block kvmclock_cpu_notifier_block
= {
5693 .notifier_call
= kvmclock_cpu_notifier
,
5694 .priority
= -INT_MAX
5697 static void kvm_timer_init(void)
5701 max_tsc_khz
= tsc_khz
;
5703 cpu_notifier_register_begin();
5704 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
5705 #ifdef CONFIG_CPU_FREQ
5706 struct cpufreq_policy policy
;
5707 memset(&policy
, 0, sizeof(policy
));
5709 cpufreq_get_policy(&policy
, cpu
);
5710 if (policy
.cpuinfo
.max_freq
)
5711 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
5714 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
5715 CPUFREQ_TRANSITION_NOTIFIER
);
5717 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
5718 for_each_online_cpu(cpu
)
5719 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5721 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5722 cpu_notifier_register_done();
5726 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
5728 int kvm_is_in_guest(void)
5730 return __this_cpu_read(current_vcpu
) != NULL
;
5733 static int kvm_is_user_mode(void)
5737 if (__this_cpu_read(current_vcpu
))
5738 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
5740 return user_mode
!= 0;
5743 static unsigned long kvm_get_guest_ip(void)
5745 unsigned long ip
= 0;
5747 if (__this_cpu_read(current_vcpu
))
5748 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
5753 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
5754 .is_in_guest
= kvm_is_in_guest
,
5755 .is_user_mode
= kvm_is_user_mode
,
5756 .get_guest_ip
= kvm_get_guest_ip
,
5759 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
5761 __this_cpu_write(current_vcpu
, vcpu
);
5763 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
5765 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
5767 __this_cpu_write(current_vcpu
, NULL
);
5769 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
5771 static void kvm_set_mmio_spte_mask(void)
5774 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
5777 * Set the reserved bits and the present bit of an paging-structure
5778 * entry to generate page fault with PFER.RSV = 1.
5780 /* Mask the reserved physical address bits. */
5781 mask
= rsvd_bits(maxphyaddr
, 51);
5783 /* Bit 62 is always reserved for 32bit host. */
5784 mask
|= 0x3ull
<< 62;
5786 /* Set the present bit. */
5789 #ifdef CONFIG_X86_64
5791 * If reserved bit is not supported, clear the present bit to disable
5794 if (maxphyaddr
== 52)
5798 kvm_mmu_set_mmio_spte_mask(mask
);
5801 #ifdef CONFIG_X86_64
5802 static void pvclock_gtod_update_fn(struct work_struct
*work
)
5806 struct kvm_vcpu
*vcpu
;
5809 spin_lock(&kvm_lock
);
5810 list_for_each_entry(kvm
, &vm_list
, vm_list
)
5811 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5812 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
5813 atomic_set(&kvm_guest_has_master_clock
, 0);
5814 spin_unlock(&kvm_lock
);
5817 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
5820 * Notification about pvclock gtod data update.
5822 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
5825 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
5826 struct timekeeper
*tk
= priv
;
5828 update_pvclock_gtod(tk
);
5830 /* disable master clock if host does not trust, or does not
5831 * use, TSC clocksource
5833 if (gtod
->clock
.vclock_mode
!= VCLOCK_TSC
&&
5834 atomic_read(&kvm_guest_has_master_clock
) != 0)
5835 queue_work(system_long_wq
, &pvclock_gtod_work
);
5840 static struct notifier_block pvclock_gtod_notifier
= {
5841 .notifier_call
= pvclock_gtod_notify
,
5845 int kvm_arch_init(void *opaque
)
5848 struct kvm_x86_ops
*ops
= opaque
;
5851 printk(KERN_ERR
"kvm: already loaded the other module\n");
5856 if (!ops
->cpu_has_kvm_support()) {
5857 printk(KERN_ERR
"kvm: no hardware support\n");
5861 if (ops
->disabled_by_bios()) {
5862 printk(KERN_ERR
"kvm: disabled by bios\n");
5868 shared_msrs
= alloc_percpu(struct kvm_shared_msrs
);
5870 printk(KERN_ERR
"kvm: failed to allocate percpu kvm_shared_msrs\n");
5874 r
= kvm_mmu_module_init();
5876 goto out_free_percpu
;
5878 kvm_set_mmio_spte_mask();
5882 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
5883 PT_DIRTY_MASK
, PT64_NX_MASK
, 0,
5887 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
5889 if (boot_cpu_has(X86_FEATURE_XSAVE
))
5890 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
5893 #ifdef CONFIG_X86_64
5894 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
5900 free_percpu(shared_msrs
);
5905 void kvm_arch_exit(void)
5907 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
5909 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5910 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
5911 CPUFREQ_TRANSITION_NOTIFIER
);
5912 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5913 #ifdef CONFIG_X86_64
5914 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
5917 kvm_mmu_module_exit();
5918 free_percpu(shared_msrs
);
5921 int kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
5923 ++vcpu
->stat
.halt_exits
;
5924 if (lapic_in_kernel(vcpu
)) {
5925 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
5928 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
5932 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
5934 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
5936 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
5937 return kvm_vcpu_halt(vcpu
);
5939 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
5942 * kvm_pv_kick_cpu_op: Kick a vcpu.
5944 * @apicid - apicid of vcpu to be kicked.
5946 static void kvm_pv_kick_cpu_op(struct kvm
*kvm
, unsigned long flags
, int apicid
)
5948 struct kvm_lapic_irq lapic_irq
;
5950 lapic_irq
.shorthand
= 0;
5951 lapic_irq
.dest_mode
= 0;
5952 lapic_irq
.dest_id
= apicid
;
5953 lapic_irq
.msi_redir_hint
= false;
5955 lapic_irq
.delivery_mode
= APIC_DM_REMRD
;
5956 kvm_irq_delivery_to_apic(kvm
, NULL
, &lapic_irq
, NULL
);
5959 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu
*vcpu
)
5961 vcpu
->arch
.apicv_active
= false;
5962 kvm_x86_ops
->refresh_apicv_exec_ctrl(vcpu
);
5965 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
5967 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
5968 int op_64_bit
, r
= 1;
5970 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
5972 if (kvm_hv_hypercall_enabled(vcpu
->kvm
))
5973 return kvm_hv_hypercall(vcpu
);
5975 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5976 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5977 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5978 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5979 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5981 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
5983 op_64_bit
= is_64_bit_mode(vcpu
);
5992 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
5998 case KVM_HC_VAPIC_POLL_IRQ
:
6001 case KVM_HC_KICK_CPU
:
6002 kvm_pv_kick_cpu_op(vcpu
->kvm
, a0
, a1
);
6012 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
6013 ++vcpu
->stat
.hypercalls
;
6016 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
6018 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
6020 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6021 char instruction
[3];
6022 unsigned long rip
= kvm_rip_read(vcpu
);
6024 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
6026 return emulator_write_emulated(ctxt
, rip
, instruction
, 3, NULL
);
6029 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
6031 return vcpu
->run
->request_interrupt_window
&&
6032 likely(!pic_in_kernel(vcpu
->kvm
));
6035 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
6037 struct kvm_run
*kvm_run
= vcpu
->run
;
6039 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
6040 kvm_run
->flags
= is_smm(vcpu
) ? KVM_RUN_X86_SMM
: 0;
6041 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
6042 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
6043 kvm_run
->ready_for_interrupt_injection
=
6044 pic_in_kernel(vcpu
->kvm
) ||
6045 kvm_vcpu_ready_for_interrupt_injection(vcpu
);
6048 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
6052 if (!kvm_x86_ops
->update_cr8_intercept
)
6055 if (!lapic_in_kernel(vcpu
))
6058 if (vcpu
->arch
.apicv_active
)
6061 if (!vcpu
->arch
.apic
->vapic_addr
)
6062 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
6069 tpr
= kvm_lapic_get_cr8(vcpu
);
6071 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
6074 static int inject_pending_event(struct kvm_vcpu
*vcpu
, bool req_int_win
)
6078 /* try to reinject previous events if any */
6079 if (vcpu
->arch
.exception
.pending
) {
6080 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
6081 vcpu
->arch
.exception
.has_error_code
,
6082 vcpu
->arch
.exception
.error_code
);
6084 if (exception_type(vcpu
->arch
.exception
.nr
) == EXCPT_FAULT
)
6085 __kvm_set_rflags(vcpu
, kvm_get_rflags(vcpu
) |
6088 if (vcpu
->arch
.exception
.nr
== DB_VECTOR
&&
6089 (vcpu
->arch
.dr7
& DR7_GD
)) {
6090 vcpu
->arch
.dr7
&= ~DR7_GD
;
6091 kvm_update_dr7(vcpu
);
6094 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
6095 vcpu
->arch
.exception
.has_error_code
,
6096 vcpu
->arch
.exception
.error_code
,
6097 vcpu
->arch
.exception
.reinject
);
6101 if (vcpu
->arch
.nmi_injected
) {
6102 kvm_x86_ops
->set_nmi(vcpu
);
6106 if (vcpu
->arch
.interrupt
.pending
) {
6107 kvm_x86_ops
->set_irq(vcpu
);
6111 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6112 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6117 /* try to inject new event if pending */
6118 if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
)) {
6119 vcpu
->arch
.smi_pending
= false;
6121 } else if (vcpu
->arch
.nmi_pending
&& kvm_x86_ops
->nmi_allowed(vcpu
)) {
6122 --vcpu
->arch
.nmi_pending
;
6123 vcpu
->arch
.nmi_injected
= true;
6124 kvm_x86_ops
->set_nmi(vcpu
);
6125 } else if (kvm_cpu_has_injectable_intr(vcpu
)) {
6127 * Because interrupts can be injected asynchronously, we are
6128 * calling check_nested_events again here to avoid a race condition.
6129 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6130 * proposal and current concerns. Perhaps we should be setting
6131 * KVM_REQ_EVENT only on certain events and not unconditionally?
6133 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6134 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6138 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
6139 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
6141 kvm_x86_ops
->set_irq(vcpu
);
6148 static void process_nmi(struct kvm_vcpu
*vcpu
)
6153 * x86 is limited to one NMI running, and one NMI pending after it.
6154 * If an NMI is already in progress, limit further NMIs to just one.
6155 * Otherwise, allow two (and we'll inject the first one immediately).
6157 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
6160 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
6161 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
6162 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6165 #define put_smstate(type, buf, offset, val) \
6166 *(type *)((buf) + (offset) - 0x7e00) = val
6168 static u32
enter_smm_get_segment_flags(struct kvm_segment
*seg
)
6171 flags
|= seg
->g
<< 23;
6172 flags
|= seg
->db
<< 22;
6173 flags
|= seg
->l
<< 21;
6174 flags
|= seg
->avl
<< 20;
6175 flags
|= seg
->present
<< 15;
6176 flags
|= seg
->dpl
<< 13;
6177 flags
|= seg
->s
<< 12;
6178 flags
|= seg
->type
<< 8;
6182 static void enter_smm_save_seg_32(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6184 struct kvm_segment seg
;
6187 kvm_get_segment(vcpu
, &seg
, n
);
6188 put_smstate(u32
, buf
, 0x7fa8 + n
* 4, seg
.selector
);
6191 offset
= 0x7f84 + n
* 12;
6193 offset
= 0x7f2c + (n
- 3) * 12;
6195 put_smstate(u32
, buf
, offset
+ 8, seg
.base
);
6196 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6197 put_smstate(u32
, buf
, offset
, enter_smm_get_segment_flags(&seg
));
6200 #ifdef CONFIG_X86_64
6201 static void enter_smm_save_seg_64(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6203 struct kvm_segment seg
;
6207 kvm_get_segment(vcpu
, &seg
, n
);
6208 offset
= 0x7e00 + n
* 16;
6210 flags
= enter_smm_get_segment_flags(&seg
) >> 8;
6211 put_smstate(u16
, buf
, offset
, seg
.selector
);
6212 put_smstate(u16
, buf
, offset
+ 2, flags
);
6213 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6214 put_smstate(u64
, buf
, offset
+ 8, seg
.base
);
6218 static void enter_smm_save_state_32(struct kvm_vcpu
*vcpu
, char *buf
)
6221 struct kvm_segment seg
;
6225 put_smstate(u32
, buf
, 0x7ffc, kvm_read_cr0(vcpu
));
6226 put_smstate(u32
, buf
, 0x7ff8, kvm_read_cr3(vcpu
));
6227 put_smstate(u32
, buf
, 0x7ff4, kvm_get_rflags(vcpu
));
6228 put_smstate(u32
, buf
, 0x7ff0, kvm_rip_read(vcpu
));
6230 for (i
= 0; i
< 8; i
++)
6231 put_smstate(u32
, buf
, 0x7fd0 + i
* 4, kvm_register_read(vcpu
, i
));
6233 kvm_get_dr(vcpu
, 6, &val
);
6234 put_smstate(u32
, buf
, 0x7fcc, (u32
)val
);
6235 kvm_get_dr(vcpu
, 7, &val
);
6236 put_smstate(u32
, buf
, 0x7fc8, (u32
)val
);
6238 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6239 put_smstate(u32
, buf
, 0x7fc4, seg
.selector
);
6240 put_smstate(u32
, buf
, 0x7f64, seg
.base
);
6241 put_smstate(u32
, buf
, 0x7f60, seg
.limit
);
6242 put_smstate(u32
, buf
, 0x7f5c, enter_smm_get_segment_flags(&seg
));
6244 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6245 put_smstate(u32
, buf
, 0x7fc0, seg
.selector
);
6246 put_smstate(u32
, buf
, 0x7f80, seg
.base
);
6247 put_smstate(u32
, buf
, 0x7f7c, seg
.limit
);
6248 put_smstate(u32
, buf
, 0x7f78, enter_smm_get_segment_flags(&seg
));
6250 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6251 put_smstate(u32
, buf
, 0x7f74, dt
.address
);
6252 put_smstate(u32
, buf
, 0x7f70, dt
.size
);
6254 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6255 put_smstate(u32
, buf
, 0x7f58, dt
.address
);
6256 put_smstate(u32
, buf
, 0x7f54, dt
.size
);
6258 for (i
= 0; i
< 6; i
++)
6259 enter_smm_save_seg_32(vcpu
, buf
, i
);
6261 put_smstate(u32
, buf
, 0x7f14, kvm_read_cr4(vcpu
));
6264 put_smstate(u32
, buf
, 0x7efc, 0x00020000);
6265 put_smstate(u32
, buf
, 0x7ef8, vcpu
->arch
.smbase
);
6268 static void enter_smm_save_state_64(struct kvm_vcpu
*vcpu
, char *buf
)
6270 #ifdef CONFIG_X86_64
6272 struct kvm_segment seg
;
6276 for (i
= 0; i
< 16; i
++)
6277 put_smstate(u64
, buf
, 0x7ff8 - i
* 8, kvm_register_read(vcpu
, i
));
6279 put_smstate(u64
, buf
, 0x7f78, kvm_rip_read(vcpu
));
6280 put_smstate(u32
, buf
, 0x7f70, kvm_get_rflags(vcpu
));
6282 kvm_get_dr(vcpu
, 6, &val
);
6283 put_smstate(u64
, buf
, 0x7f68, val
);
6284 kvm_get_dr(vcpu
, 7, &val
);
6285 put_smstate(u64
, buf
, 0x7f60, val
);
6287 put_smstate(u64
, buf
, 0x7f58, kvm_read_cr0(vcpu
));
6288 put_smstate(u64
, buf
, 0x7f50, kvm_read_cr3(vcpu
));
6289 put_smstate(u64
, buf
, 0x7f48, kvm_read_cr4(vcpu
));
6291 put_smstate(u32
, buf
, 0x7f00, vcpu
->arch
.smbase
);
6294 put_smstate(u32
, buf
, 0x7efc, 0x00020064);
6296 put_smstate(u64
, buf
, 0x7ed0, vcpu
->arch
.efer
);
6298 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6299 put_smstate(u16
, buf
, 0x7e90, seg
.selector
);
6300 put_smstate(u16
, buf
, 0x7e92, enter_smm_get_segment_flags(&seg
) >> 8);
6301 put_smstate(u32
, buf
, 0x7e94, seg
.limit
);
6302 put_smstate(u64
, buf
, 0x7e98, seg
.base
);
6304 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6305 put_smstate(u32
, buf
, 0x7e84, dt
.size
);
6306 put_smstate(u64
, buf
, 0x7e88, dt
.address
);
6308 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6309 put_smstate(u16
, buf
, 0x7e70, seg
.selector
);
6310 put_smstate(u16
, buf
, 0x7e72, enter_smm_get_segment_flags(&seg
) >> 8);
6311 put_smstate(u32
, buf
, 0x7e74, seg
.limit
);
6312 put_smstate(u64
, buf
, 0x7e78, seg
.base
);
6314 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6315 put_smstate(u32
, buf
, 0x7e64, dt
.size
);
6316 put_smstate(u64
, buf
, 0x7e68, dt
.address
);
6318 for (i
= 0; i
< 6; i
++)
6319 enter_smm_save_seg_64(vcpu
, buf
, i
);
6325 static void enter_smm(struct kvm_vcpu
*vcpu
)
6327 struct kvm_segment cs
, ds
;
6332 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, true);
6333 vcpu
->arch
.hflags
|= HF_SMM_MASK
;
6334 memset(buf
, 0, 512);
6335 if (guest_cpuid_has_longmode(vcpu
))
6336 enter_smm_save_state_64(vcpu
, buf
);
6338 enter_smm_save_state_32(vcpu
, buf
);
6340 kvm_vcpu_write_guest(vcpu
, vcpu
->arch
.smbase
+ 0xfe00, buf
, sizeof(buf
));
6342 if (kvm_x86_ops
->get_nmi_mask(vcpu
))
6343 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
6345 kvm_x86_ops
->set_nmi_mask(vcpu
, true);
6347 kvm_set_rflags(vcpu
, X86_EFLAGS_FIXED
);
6348 kvm_rip_write(vcpu
, 0x8000);
6350 cr0
= vcpu
->arch
.cr0
& ~(X86_CR0_PE
| X86_CR0_EM
| X86_CR0_TS
| X86_CR0_PG
);
6351 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
6352 vcpu
->arch
.cr0
= cr0
;
6354 kvm_x86_ops
->set_cr4(vcpu
, 0);
6356 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6357 dt
.address
= dt
.size
= 0;
6358 kvm_x86_ops
->set_idt(vcpu
, &dt
);
6360 __kvm_set_dr(vcpu
, 7, DR7_FIXED_1
);
6362 cs
.selector
= (vcpu
->arch
.smbase
>> 4) & 0xffff;
6363 cs
.base
= vcpu
->arch
.smbase
;
6368 cs
.limit
= ds
.limit
= 0xffffffff;
6369 cs
.type
= ds
.type
= 0x3;
6370 cs
.dpl
= ds
.dpl
= 0;
6375 cs
.avl
= ds
.avl
= 0;
6376 cs
.present
= ds
.present
= 1;
6377 cs
.unusable
= ds
.unusable
= 0;
6378 cs
.padding
= ds
.padding
= 0;
6380 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6381 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_DS
);
6382 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_ES
);
6383 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_FS
);
6384 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_GS
);
6385 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_SS
);
6387 if (guest_cpuid_has_longmode(vcpu
))
6388 kvm_x86_ops
->set_efer(vcpu
, 0);
6390 kvm_update_cpuid(vcpu
);
6391 kvm_mmu_reset_context(vcpu
);
6394 static void process_smi(struct kvm_vcpu
*vcpu
)
6396 vcpu
->arch
.smi_pending
= true;
6397 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6400 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
6402 kvm_make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
6405 static void vcpu_scan_ioapic(struct kvm_vcpu
*vcpu
)
6407 u64 eoi_exit_bitmap
[4];
6409 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
6412 bitmap_zero(vcpu
->arch
.ioapic_handled_vectors
, 256);
6414 if (irqchip_split(vcpu
->kvm
))
6415 kvm_scan_ioapic_routes(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6417 if (vcpu
->arch
.apicv_active
)
6418 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
6419 kvm_ioapic_scan_entry(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6421 bitmap_or((ulong
*)eoi_exit_bitmap
, vcpu
->arch
.ioapic_handled_vectors
,
6422 vcpu_to_synic(vcpu
)->vec_bitmap
, 256);
6423 kvm_x86_ops
->load_eoi_exitmap(vcpu
, eoi_exit_bitmap
);
6426 static void kvm_vcpu_flush_tlb(struct kvm_vcpu
*vcpu
)
6428 ++vcpu
->stat
.tlb_flush
;
6429 kvm_x86_ops
->tlb_flush(vcpu
);
6432 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu
*vcpu
)
6434 struct page
*page
= NULL
;
6436 if (!lapic_in_kernel(vcpu
))
6439 if (!kvm_x86_ops
->set_apic_access_page_addr
)
6442 page
= gfn_to_page(vcpu
->kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
6443 if (is_error_page(page
))
6445 kvm_x86_ops
->set_apic_access_page_addr(vcpu
, page_to_phys(page
));
6448 * Do not pin apic access page in memory, the MMU notifier
6449 * will call us again if it is migrated or swapped out.
6453 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page
);
6455 void kvm_arch_mmu_notifier_invalidate_page(struct kvm
*kvm
,
6456 unsigned long address
)
6459 * The physical address of apic access page is stored in the VMCS.
6460 * Update it when it becomes invalid.
6462 if (address
== gfn_to_hva(kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
))
6463 kvm_make_all_cpus_request(kvm
, KVM_REQ_APIC_PAGE_RELOAD
);
6467 * Returns 1 to let vcpu_run() continue the guest execution loop without
6468 * exiting to the userspace. Otherwise, the value will be returned to the
6471 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
6475 dm_request_for_irq_injection(vcpu
) &&
6476 kvm_cpu_accept_dm_intr(vcpu
);
6478 bool req_immediate_exit
= false;
6480 if (vcpu
->requests
) {
6481 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
6482 kvm_mmu_unload(vcpu
);
6483 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
6484 __kvm_migrate_timers(vcpu
);
6485 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
6486 kvm_gen_update_masterclock(vcpu
->kvm
);
6487 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
))
6488 kvm_gen_kvmclock_update(vcpu
);
6489 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
6490 r
= kvm_guest_time_update(vcpu
);
6494 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
6495 kvm_mmu_sync_roots(vcpu
);
6496 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
6497 kvm_vcpu_flush_tlb(vcpu
);
6498 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
6499 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
6503 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
6504 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
6508 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
)) {
6509 vcpu
->fpu_active
= 0;
6510 kvm_x86_ops
->fpu_deactivate(vcpu
);
6512 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
6513 /* Page is swapped out. Do synthetic halt */
6514 vcpu
->arch
.apf
.halted
= true;
6518 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
6519 record_steal_time(vcpu
);
6520 if (kvm_check_request(KVM_REQ_SMI
, vcpu
))
6522 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
6524 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
6525 kvm_pmu_handle_event(vcpu
);
6526 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
6527 kvm_pmu_deliver_pmi(vcpu
);
6528 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT
, vcpu
)) {
6529 BUG_ON(vcpu
->arch
.pending_ioapic_eoi
> 255);
6530 if (test_bit(vcpu
->arch
.pending_ioapic_eoi
,
6531 vcpu
->arch
.ioapic_handled_vectors
)) {
6532 vcpu
->run
->exit_reason
= KVM_EXIT_IOAPIC_EOI
;
6533 vcpu
->run
->eoi
.vector
=
6534 vcpu
->arch
.pending_ioapic_eoi
;
6539 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC
, vcpu
))
6540 vcpu_scan_ioapic(vcpu
);
6541 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD
, vcpu
))
6542 kvm_vcpu_reload_apic_access_page(vcpu
);
6543 if (kvm_check_request(KVM_REQ_HV_CRASH
, vcpu
)) {
6544 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
6545 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_CRASH
;
6549 if (kvm_check_request(KVM_REQ_HV_RESET
, vcpu
)) {
6550 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
6551 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_RESET
;
6555 if (kvm_check_request(KVM_REQ_HV_EXIT
, vcpu
)) {
6556 vcpu
->run
->exit_reason
= KVM_EXIT_HYPERV
;
6557 vcpu
->run
->hyperv
= vcpu
->arch
.hyperv
.exit
;
6563 * KVM_REQ_HV_STIMER has to be processed after
6564 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6565 * depend on the guest clock being up-to-date
6567 if (kvm_check_request(KVM_REQ_HV_STIMER
, vcpu
))
6568 kvm_hv_process_stimers(vcpu
);
6572 * KVM_REQ_EVENT is not set when posted interrupts are set by
6573 * VT-d hardware, so we have to update RVI unconditionally.
6575 if (kvm_lapic_enabled(vcpu
)) {
6577 * Update architecture specific hints for APIC
6578 * virtual interrupt delivery.
6580 if (vcpu
->arch
.apicv_active
)
6581 kvm_x86_ops
->hwapic_irr_update(vcpu
,
6582 kvm_lapic_find_highest_irr(vcpu
));
6585 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
6586 kvm_apic_accept_events(vcpu
);
6587 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
6592 if (inject_pending_event(vcpu
, req_int_win
) != 0)
6593 req_immediate_exit
= true;
6595 /* Enable NMI/IRQ window open exits if needed.
6597 * SMIs have two cases: 1) they can be nested, and
6598 * then there is nothing to do here because RSM will
6599 * cause a vmexit anyway; 2) or the SMI can be pending
6600 * because inject_pending_event has completed the
6601 * injection of an IRQ or NMI from the previous vmexit,
6602 * and then we request an immediate exit to inject the SMI.
6604 if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
))
6605 req_immediate_exit
= true;
6606 if (vcpu
->arch
.nmi_pending
)
6607 kvm_x86_ops
->enable_nmi_window(vcpu
);
6608 if (kvm_cpu_has_injectable_intr(vcpu
) || req_int_win
)
6609 kvm_x86_ops
->enable_irq_window(vcpu
);
6612 if (kvm_lapic_enabled(vcpu
)) {
6613 update_cr8_intercept(vcpu
);
6614 kvm_lapic_sync_to_vapic(vcpu
);
6618 r
= kvm_mmu_reload(vcpu
);
6620 goto cancel_injection
;
6625 kvm_x86_ops
->prepare_guest_switch(vcpu
);
6626 if (vcpu
->fpu_active
)
6627 kvm_load_guest_fpu(vcpu
);
6628 vcpu
->mode
= IN_GUEST_MODE
;
6630 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
6633 * We should set ->mode before check ->requests,
6634 * Please see the comment in kvm_make_all_cpus_request.
6635 * This also orders the write to mode from any reads
6636 * to the page tables done while the VCPU is running.
6637 * Please see the comment in kvm_flush_remote_tlbs.
6639 smp_mb__after_srcu_read_unlock();
6641 local_irq_disable();
6643 if (vcpu
->mode
== EXITING_GUEST_MODE
|| vcpu
->requests
6644 || need_resched() || signal_pending(current
)) {
6645 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
6649 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6651 goto cancel_injection
;
6654 kvm_load_guest_xcr0(vcpu
);
6656 if (req_immediate_exit
) {
6657 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6658 smp_send_reschedule(vcpu
->cpu
);
6661 trace_kvm_entry(vcpu
->vcpu_id
);
6662 wait_lapic_expire(vcpu
);
6663 guest_enter_irqoff();
6665 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
6667 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
6668 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
6669 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
6670 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
6671 set_debugreg(vcpu
->arch
.dr6
, 6);
6672 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
6675 kvm_x86_ops
->run(vcpu
);
6678 * Do this here before restoring debug registers on the host. And
6679 * since we do this before handling the vmexit, a DR access vmexit
6680 * can (a) read the correct value of the debug registers, (b) set
6681 * KVM_DEBUGREG_WONT_EXIT again.
6683 if (unlikely(vcpu
->arch
.switch_db_regs
& KVM_DEBUGREG_WONT_EXIT
)) {
6684 WARN_ON(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
);
6685 kvm_x86_ops
->sync_dirty_debug_regs(vcpu
);
6686 kvm_update_dr0123(vcpu
);
6687 kvm_update_dr6(vcpu
);
6688 kvm_update_dr7(vcpu
);
6689 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
6693 * If the guest has used debug registers, at least dr7
6694 * will be disabled while returning to the host.
6695 * If we don't have active breakpoints in the host, we don't
6696 * care about the messed up debug address registers. But if
6697 * we have some of them active, restore the old state.
6699 if (hw_breakpoint_active())
6700 hw_breakpoint_restore();
6702 vcpu
->arch
.last_guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
6704 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
6707 kvm_put_guest_xcr0(vcpu
);
6709 /* Interrupt is enabled by handle_external_intr() */
6710 kvm_x86_ops
->handle_external_intr(vcpu
);
6714 guest_exit_irqoff();
6719 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6722 * Profile KVM exit RIPs:
6724 if (unlikely(prof_on
== KVM_PROFILING
)) {
6725 unsigned long rip
= kvm_rip_read(vcpu
);
6726 profile_hit(KVM_PROFILING
, (void *)rip
);
6729 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
6730 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
6732 if (vcpu
->arch
.apic_attention
)
6733 kvm_lapic_sync_from_vapic(vcpu
);
6735 r
= kvm_x86_ops
->handle_exit(vcpu
);
6739 kvm_x86_ops
->cancel_injection(vcpu
);
6740 if (unlikely(vcpu
->arch
.apic_attention
))
6741 kvm_lapic_sync_from_vapic(vcpu
);
6746 static inline int vcpu_block(struct kvm
*kvm
, struct kvm_vcpu
*vcpu
)
6748 if (!kvm_arch_vcpu_runnable(vcpu
) &&
6749 (!kvm_x86_ops
->pre_block
|| kvm_x86_ops
->pre_block(vcpu
) == 0)) {
6750 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
6751 kvm_vcpu_block(vcpu
);
6752 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
6754 if (kvm_x86_ops
->post_block
)
6755 kvm_x86_ops
->post_block(vcpu
);
6757 if (!kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
6761 kvm_apic_accept_events(vcpu
);
6762 switch(vcpu
->arch
.mp_state
) {
6763 case KVM_MP_STATE_HALTED
:
6764 vcpu
->arch
.pv
.pv_unhalted
= false;
6765 vcpu
->arch
.mp_state
=
6766 KVM_MP_STATE_RUNNABLE
;
6767 case KVM_MP_STATE_RUNNABLE
:
6768 vcpu
->arch
.apf
.halted
= false;
6770 case KVM_MP_STATE_INIT_RECEIVED
:
6779 static inline bool kvm_vcpu_running(struct kvm_vcpu
*vcpu
)
6781 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
6782 !vcpu
->arch
.apf
.halted
);
6785 static int vcpu_run(struct kvm_vcpu
*vcpu
)
6788 struct kvm
*kvm
= vcpu
->kvm
;
6790 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
6793 if (kvm_vcpu_running(vcpu
)) {
6794 r
= vcpu_enter_guest(vcpu
);
6796 r
= vcpu_block(kvm
, vcpu
);
6802 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
6803 if (kvm_cpu_has_pending_timer(vcpu
))
6804 kvm_inject_pending_timer_irqs(vcpu
);
6806 if (dm_request_for_irq_injection(vcpu
) &&
6807 kvm_vcpu_ready_for_interrupt_injection(vcpu
)) {
6809 vcpu
->run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
6810 ++vcpu
->stat
.request_irq_exits
;
6814 kvm_check_async_pf_completion(vcpu
);
6816 if (signal_pending(current
)) {
6818 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
6819 ++vcpu
->stat
.signal_exits
;
6822 if (need_resched()) {
6823 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
6825 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
6829 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
6834 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
6837 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6838 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
6839 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
6840 if (r
!= EMULATE_DONE
)
6845 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
6847 BUG_ON(!vcpu
->arch
.pio
.count
);
6849 return complete_emulated_io(vcpu
);
6853 * Implements the following, as a state machine:
6857 * for each mmio piece in the fragment
6865 * for each mmio piece in the fragment
6870 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
6872 struct kvm_run
*run
= vcpu
->run
;
6873 struct kvm_mmio_fragment
*frag
;
6876 BUG_ON(!vcpu
->mmio_needed
);
6878 /* Complete previous fragment */
6879 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
6880 len
= min(8u, frag
->len
);
6881 if (!vcpu
->mmio_is_write
)
6882 memcpy(frag
->data
, run
->mmio
.data
, len
);
6884 if (frag
->len
<= 8) {
6885 /* Switch to the next fragment. */
6887 vcpu
->mmio_cur_fragment
++;
6889 /* Go forward to the next mmio piece. */
6895 if (vcpu
->mmio_cur_fragment
>= vcpu
->mmio_nr_fragments
) {
6896 vcpu
->mmio_needed
= 0;
6898 /* FIXME: return into emulator if single-stepping. */
6899 if (vcpu
->mmio_is_write
)
6901 vcpu
->mmio_read_completed
= 1;
6902 return complete_emulated_io(vcpu
);
6905 run
->exit_reason
= KVM_EXIT_MMIO
;
6906 run
->mmio
.phys_addr
= frag
->gpa
;
6907 if (vcpu
->mmio_is_write
)
6908 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
6909 run
->mmio
.len
= min(8u, frag
->len
);
6910 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
6911 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
6916 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
6918 struct fpu
*fpu
= ¤t
->thread
.fpu
;
6922 fpu__activate_curr(fpu
);
6924 if (vcpu
->sigset_active
)
6925 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
6927 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
6928 kvm_vcpu_block(vcpu
);
6929 kvm_apic_accept_events(vcpu
);
6930 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
6935 /* re-sync apic's tpr */
6936 if (!lapic_in_kernel(vcpu
)) {
6937 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
6943 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
6944 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
6945 vcpu
->arch
.complete_userspace_io
= NULL
;
6950 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
6955 post_kvm_run_save(vcpu
);
6956 if (vcpu
->sigset_active
)
6957 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
6962 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
6964 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
6966 * We are here if userspace calls get_regs() in the middle of
6967 * instruction emulation. Registers state needs to be copied
6968 * back from emulation context to vcpu. Userspace shouldn't do
6969 * that usually, but some bad designed PV devices (vmware
6970 * backdoor interface) need this to work
6972 emulator_writeback_register_cache(&vcpu
->arch
.emulate_ctxt
);
6973 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
6975 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
6976 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
6977 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
6978 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
6979 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
6980 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
6981 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
6982 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
6983 #ifdef CONFIG_X86_64
6984 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
6985 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
6986 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
6987 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
6988 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
6989 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
6990 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
6991 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
6994 regs
->rip
= kvm_rip_read(vcpu
);
6995 regs
->rflags
= kvm_get_rflags(vcpu
);
7000 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
7002 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
7003 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7005 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
7006 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
7007 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
7008 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
7009 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
7010 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
7011 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
7012 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
7013 #ifdef CONFIG_X86_64
7014 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
7015 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
7016 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
7017 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
7018 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
7019 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
7020 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
7021 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
7024 kvm_rip_write(vcpu
, regs
->rip
);
7025 kvm_set_rflags(vcpu
, regs
->rflags
);
7027 vcpu
->arch
.exception
.pending
= false;
7029 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7034 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
7036 struct kvm_segment cs
;
7038 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7042 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
7044 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
7045 struct kvm_sregs
*sregs
)
7049 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7050 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7051 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7052 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7053 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7054 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7056 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7057 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7059 kvm_x86_ops
->get_idt(vcpu
, &dt
);
7060 sregs
->idt
.limit
= dt
.size
;
7061 sregs
->idt
.base
= dt
.address
;
7062 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
7063 sregs
->gdt
.limit
= dt
.size
;
7064 sregs
->gdt
.base
= dt
.address
;
7066 sregs
->cr0
= kvm_read_cr0(vcpu
);
7067 sregs
->cr2
= vcpu
->arch
.cr2
;
7068 sregs
->cr3
= kvm_read_cr3(vcpu
);
7069 sregs
->cr4
= kvm_read_cr4(vcpu
);
7070 sregs
->cr8
= kvm_get_cr8(vcpu
);
7071 sregs
->efer
= vcpu
->arch
.efer
;
7072 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
7074 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
7076 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
7077 set_bit(vcpu
->arch
.interrupt
.nr
,
7078 (unsigned long *)sregs
->interrupt_bitmap
);
7083 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
7084 struct kvm_mp_state
*mp_state
)
7086 kvm_apic_accept_events(vcpu
);
7087 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_HALTED
&&
7088 vcpu
->arch
.pv
.pv_unhalted
)
7089 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
7091 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
7096 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
7097 struct kvm_mp_state
*mp_state
)
7099 if (!lapic_in_kernel(vcpu
) &&
7100 mp_state
->mp_state
!= KVM_MP_STATE_RUNNABLE
)
7103 if (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
) {
7104 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
7105 set_bit(KVM_APIC_SIPI
, &vcpu
->arch
.apic
->pending_events
);
7107 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
7108 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7112 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
7113 int reason
, bool has_error_code
, u32 error_code
)
7115 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
7118 init_emulate_ctxt(vcpu
);
7120 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
7121 has_error_code
, error_code
);
7124 return EMULATE_FAIL
;
7126 kvm_rip_write(vcpu
, ctxt
->eip
);
7127 kvm_set_rflags(vcpu
, ctxt
->eflags
);
7128 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7129 return EMULATE_DONE
;
7131 EXPORT_SYMBOL_GPL(kvm_task_switch
);
7133 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
7134 struct kvm_sregs
*sregs
)
7136 struct msr_data apic_base_msr
;
7137 int mmu_reset_needed
= 0;
7138 int pending_vec
, max_bits
, idx
;
7141 if (!guest_cpuid_has_xsave(vcpu
) && (sregs
->cr4
& X86_CR4_OSXSAVE
))
7144 dt
.size
= sregs
->idt
.limit
;
7145 dt
.address
= sregs
->idt
.base
;
7146 kvm_x86_ops
->set_idt(vcpu
, &dt
);
7147 dt
.size
= sregs
->gdt
.limit
;
7148 dt
.address
= sregs
->gdt
.base
;
7149 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
7151 vcpu
->arch
.cr2
= sregs
->cr2
;
7152 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
7153 vcpu
->arch
.cr3
= sregs
->cr3
;
7154 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
7156 kvm_set_cr8(vcpu
, sregs
->cr8
);
7158 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
7159 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
7160 apic_base_msr
.data
= sregs
->apic_base
;
7161 apic_base_msr
.host_initiated
= true;
7162 kvm_set_apic_base(vcpu
, &apic_base_msr
);
7164 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
7165 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
7166 vcpu
->arch
.cr0
= sregs
->cr0
;
7168 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
7169 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
7170 if (sregs
->cr4
& (X86_CR4_OSXSAVE
| X86_CR4_PKE
))
7171 kvm_update_cpuid(vcpu
);
7173 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7174 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
7175 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
7176 mmu_reset_needed
= 1;
7178 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7180 if (mmu_reset_needed
)
7181 kvm_mmu_reset_context(vcpu
);
7183 max_bits
= KVM_NR_INTERRUPTS
;
7184 pending_vec
= find_first_bit(
7185 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
7186 if (pending_vec
< max_bits
) {
7187 kvm_queue_interrupt(vcpu
, pending_vec
, false);
7188 pr_debug("Set back pending irq %d\n", pending_vec
);
7191 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7192 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7193 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7194 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7195 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7196 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7198 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7199 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7201 update_cr8_intercept(vcpu
);
7203 /* Older userspace won't unhalt the vcpu on reset. */
7204 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
7205 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
7207 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7209 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7214 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
7215 struct kvm_guest_debug
*dbg
)
7217 unsigned long rflags
;
7220 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
7222 if (vcpu
->arch
.exception
.pending
)
7224 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
7225 kvm_queue_exception(vcpu
, DB_VECTOR
);
7227 kvm_queue_exception(vcpu
, BP_VECTOR
);
7231 * Read rflags as long as potentially injected trace flags are still
7234 rflags
= kvm_get_rflags(vcpu
);
7236 vcpu
->guest_debug
= dbg
->control
;
7237 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
7238 vcpu
->guest_debug
= 0;
7240 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
7241 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
7242 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
7243 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
7245 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
7246 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
7248 kvm_update_dr7(vcpu
);
7250 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
7251 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
7252 get_segment_base(vcpu
, VCPU_SREG_CS
);
7255 * Trigger an rflags update that will inject or remove the trace
7258 kvm_set_rflags(vcpu
, rflags
);
7260 kvm_x86_ops
->update_bp_intercept(vcpu
);
7270 * Translate a guest virtual address to a guest physical address.
7272 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
7273 struct kvm_translation
*tr
)
7275 unsigned long vaddr
= tr
->linear_address
;
7279 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7280 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
7281 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7282 tr
->physical_address
= gpa
;
7283 tr
->valid
= gpa
!= UNMAPPED_GVA
;
7290 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7292 struct fxregs_state
*fxsave
=
7293 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7295 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
7296 fpu
->fcw
= fxsave
->cwd
;
7297 fpu
->fsw
= fxsave
->swd
;
7298 fpu
->ftwx
= fxsave
->twd
;
7299 fpu
->last_opcode
= fxsave
->fop
;
7300 fpu
->last_ip
= fxsave
->rip
;
7301 fpu
->last_dp
= fxsave
->rdp
;
7302 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
7307 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7309 struct fxregs_state
*fxsave
=
7310 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7312 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
7313 fxsave
->cwd
= fpu
->fcw
;
7314 fxsave
->swd
= fpu
->fsw
;
7315 fxsave
->twd
= fpu
->ftwx
;
7316 fxsave
->fop
= fpu
->last_opcode
;
7317 fxsave
->rip
= fpu
->last_ip
;
7318 fxsave
->rdp
= fpu
->last_dp
;
7319 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
7324 static void fx_init(struct kvm_vcpu
*vcpu
)
7326 fpstate_init(&vcpu
->arch
.guest_fpu
.state
);
7327 if (boot_cpu_has(X86_FEATURE_XSAVES
))
7328 vcpu
->arch
.guest_fpu
.state
.xsave
.header
.xcomp_bv
=
7329 host_xcr0
| XSTATE_COMPACTION_ENABLED
;
7332 * Ensure guest xcr0 is valid for loading
7334 vcpu
->arch
.xcr0
= XFEATURE_MASK_FP
;
7336 vcpu
->arch
.cr0
|= X86_CR0_ET
;
7339 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
7341 if (vcpu
->guest_fpu_loaded
)
7345 * Restore all possible states in the guest,
7346 * and assume host would use all available bits.
7347 * Guest xcr0 would be loaded later.
7349 vcpu
->guest_fpu_loaded
= 1;
7350 __kernel_fpu_begin();
7351 __copy_kernel_to_fpregs(&vcpu
->arch
.guest_fpu
.state
);
7355 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
7357 if (!vcpu
->guest_fpu_loaded
) {
7358 vcpu
->fpu_counter
= 0;
7362 vcpu
->guest_fpu_loaded
= 0;
7363 copy_fpregs_to_fpstate(&vcpu
->arch
.guest_fpu
);
7365 ++vcpu
->stat
.fpu_reload
;
7367 * If using eager FPU mode, or if the guest is a frequent user
7368 * of the FPU, just leave the FPU active for next time.
7369 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7370 * the FPU in bursts will revert to loading it on demand.
7372 if (!use_eager_fpu()) {
7373 if (++vcpu
->fpu_counter
< 5)
7374 kvm_make_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
);
7379 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
7381 kvmclock_reset(vcpu
);
7383 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
7384 kvm_x86_ops
->vcpu_free(vcpu
);
7387 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
7390 struct kvm_vcpu
*vcpu
;
7392 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
7393 printk_once(KERN_WARNING
7394 "kvm: SMP vm created on host with unstable TSC; "
7395 "guest TSC will not be reliable\n");
7397 vcpu
= kvm_x86_ops
->vcpu_create(kvm
, id
);
7402 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
7406 kvm_vcpu_mtrr_init(vcpu
);
7407 r
= vcpu_load(vcpu
);
7410 kvm_vcpu_reset(vcpu
, false);
7411 kvm_mmu_setup(vcpu
);
7416 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
7418 struct msr_data msr
;
7419 struct kvm
*kvm
= vcpu
->kvm
;
7421 if (vcpu_load(vcpu
))
7424 msr
.index
= MSR_IA32_TSC
;
7425 msr
.host_initiated
= true;
7426 kvm_write_tsc(vcpu
, &msr
);
7429 if (!kvmclock_periodic_sync
)
7432 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
7433 KVMCLOCK_SYNC_PERIOD
);
7436 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
7439 vcpu
->arch
.apf
.msr_val
= 0;
7441 r
= vcpu_load(vcpu
);
7443 kvm_mmu_unload(vcpu
);
7446 kvm_x86_ops
->vcpu_free(vcpu
);
7449 void kvm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
7451 vcpu
->arch
.hflags
= 0;
7453 vcpu
->arch
.smi_pending
= 0;
7454 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
7455 vcpu
->arch
.nmi_pending
= 0;
7456 vcpu
->arch
.nmi_injected
= false;
7457 kvm_clear_interrupt_queue(vcpu
);
7458 kvm_clear_exception_queue(vcpu
);
7460 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
7461 kvm_update_dr0123(vcpu
);
7462 vcpu
->arch
.dr6
= DR6_INIT
;
7463 kvm_update_dr6(vcpu
);
7464 vcpu
->arch
.dr7
= DR7_FIXED_1
;
7465 kvm_update_dr7(vcpu
);
7469 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7470 vcpu
->arch
.apf
.msr_val
= 0;
7471 vcpu
->arch
.st
.msr_val
= 0;
7473 kvmclock_reset(vcpu
);
7475 kvm_clear_async_pf_completion_queue(vcpu
);
7476 kvm_async_pf_hash_reset(vcpu
);
7477 vcpu
->arch
.apf
.halted
= false;
7480 kvm_pmu_reset(vcpu
);
7481 vcpu
->arch
.smbase
= 0x30000;
7484 memset(vcpu
->arch
.regs
, 0, sizeof(vcpu
->arch
.regs
));
7485 vcpu
->arch
.regs_avail
= ~0;
7486 vcpu
->arch
.regs_dirty
= ~0;
7488 kvm_x86_ops
->vcpu_reset(vcpu
, init_event
);
7491 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu
*vcpu
, u8 vector
)
7493 struct kvm_segment cs
;
7495 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7496 cs
.selector
= vector
<< 8;
7497 cs
.base
= vector
<< 12;
7498 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7499 kvm_rip_write(vcpu
, 0);
7502 int kvm_arch_hardware_enable(void)
7505 struct kvm_vcpu
*vcpu
;
7510 bool stable
, backwards_tsc
= false;
7512 kvm_shared_msr_cpu_online();
7513 ret
= kvm_x86_ops
->hardware_enable();
7517 local_tsc
= rdtsc();
7518 stable
= !check_tsc_unstable();
7519 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
7520 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
7521 if (!stable
&& vcpu
->cpu
== smp_processor_id())
7522 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
7523 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
7524 backwards_tsc
= true;
7525 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
7526 max_tsc
= vcpu
->arch
.last_host_tsc
;
7532 * Sometimes, even reliable TSCs go backwards. This happens on
7533 * platforms that reset TSC during suspend or hibernate actions, but
7534 * maintain synchronization. We must compensate. Fortunately, we can
7535 * detect that condition here, which happens early in CPU bringup,
7536 * before any KVM threads can be running. Unfortunately, we can't
7537 * bring the TSCs fully up to date with real time, as we aren't yet far
7538 * enough into CPU bringup that we know how much real time has actually
7539 * elapsed; our helper function, get_kernel_ns() will be using boot
7540 * variables that haven't been updated yet.
7542 * So we simply find the maximum observed TSC above, then record the
7543 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7544 * the adjustment will be applied. Note that we accumulate
7545 * adjustments, in case multiple suspend cycles happen before some VCPU
7546 * gets a chance to run again. In the event that no KVM threads get a
7547 * chance to run, we will miss the entire elapsed period, as we'll have
7548 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7549 * loose cycle time. This isn't too big a deal, since the loss will be
7550 * uniform across all VCPUs (not to mention the scenario is extremely
7551 * unlikely). It is possible that a second hibernate recovery happens
7552 * much faster than a first, causing the observed TSC here to be
7553 * smaller; this would require additional padding adjustment, which is
7554 * why we set last_host_tsc to the local tsc observed here.
7556 * N.B. - this code below runs only on platforms with reliable TSC,
7557 * as that is the only way backwards_tsc is set above. Also note
7558 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7559 * have the same delta_cyc adjustment applied if backwards_tsc
7560 * is detected. Note further, this adjustment is only done once,
7561 * as we reset last_host_tsc on all VCPUs to stop this from being
7562 * called multiple times (one for each physical CPU bringup).
7564 * Platforms with unreliable TSCs don't have to deal with this, they
7565 * will be compensated by the logic in vcpu_load, which sets the TSC to
7566 * catchup mode. This will catchup all VCPUs to real time, but cannot
7567 * guarantee that they stay in perfect synchronization.
7569 if (backwards_tsc
) {
7570 u64 delta_cyc
= max_tsc
- local_tsc
;
7571 backwards_tsc_observed
= true;
7572 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
7573 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
7574 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
7575 vcpu
->arch
.last_host_tsc
= local_tsc
;
7576 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
7580 * We have to disable TSC offset matching.. if you were
7581 * booting a VM while issuing an S4 host suspend....
7582 * you may have some problem. Solving this issue is
7583 * left as an exercise to the reader.
7585 kvm
->arch
.last_tsc_nsec
= 0;
7586 kvm
->arch
.last_tsc_write
= 0;
7593 void kvm_arch_hardware_disable(void)
7595 kvm_x86_ops
->hardware_disable();
7596 drop_user_return_notifiers();
7599 int kvm_arch_hardware_setup(void)
7603 r
= kvm_x86_ops
->hardware_setup();
7607 if (kvm_has_tsc_control
) {
7609 * Make sure the user can only configure tsc_khz values that
7610 * fit into a signed integer.
7611 * A min value is not calculated needed because it will always
7612 * be 1 on all machines.
7614 u64 max
= min(0x7fffffffULL
,
7615 __scale_tsc(kvm_max_tsc_scaling_ratio
, tsc_khz
));
7616 kvm_max_guest_tsc_khz
= max
;
7618 kvm_default_tsc_scaling_ratio
= 1ULL << kvm_tsc_scaling_ratio_frac_bits
;
7621 kvm_init_msr_list();
7625 void kvm_arch_hardware_unsetup(void)
7627 kvm_x86_ops
->hardware_unsetup();
7630 void kvm_arch_check_processor_compat(void *rtn
)
7632 kvm_x86_ops
->check_processor_compatibility(rtn
);
7635 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu
*vcpu
)
7637 return vcpu
->kvm
->arch
.bsp_vcpu_id
== vcpu
->vcpu_id
;
7639 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp
);
7641 bool kvm_vcpu_is_bsp(struct kvm_vcpu
*vcpu
)
7643 return (vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_BSP
) != 0;
7646 struct static_key kvm_no_apic_vcpu __read_mostly
;
7647 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu
);
7649 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
7655 BUG_ON(vcpu
->kvm
== NULL
);
7658 vcpu
->arch
.apicv_active
= kvm_x86_ops
->get_enable_apicv();
7659 vcpu
->arch
.pv
.pv_unhalted
= false;
7660 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
7661 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_reset_bsp(vcpu
))
7662 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7664 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
7666 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
7671 vcpu
->arch
.pio_data
= page_address(page
);
7673 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
7675 r
= kvm_mmu_create(vcpu
);
7677 goto fail_free_pio_data
;
7679 if (irqchip_in_kernel(kvm
)) {
7680 r
= kvm_create_lapic(vcpu
);
7682 goto fail_mmu_destroy
;
7684 static_key_slow_inc(&kvm_no_apic_vcpu
);
7686 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
7688 if (!vcpu
->arch
.mce_banks
) {
7690 goto fail_free_lapic
;
7692 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
7694 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
)) {
7696 goto fail_free_mce_banks
;
7701 vcpu
->arch
.ia32_tsc_adjust_msr
= 0x0;
7702 vcpu
->arch
.pv_time_enabled
= false;
7704 vcpu
->arch
.guest_supported_xcr0
= 0;
7705 vcpu
->arch
.guest_xstate_size
= XSAVE_HDR_SIZE
+ XSAVE_HDR_OFFSET
;
7707 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
7709 vcpu
->arch
.pat
= MSR_IA32_CR_PAT_DEFAULT
;
7711 kvm_async_pf_hash_reset(vcpu
);
7714 vcpu
->arch
.pending_external_vector
= -1;
7716 kvm_hv_vcpu_init(vcpu
);
7720 fail_free_mce_banks
:
7721 kfree(vcpu
->arch
.mce_banks
);
7723 kvm_free_lapic(vcpu
);
7725 kvm_mmu_destroy(vcpu
);
7727 free_page((unsigned long)vcpu
->arch
.pio_data
);
7732 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
7736 kvm_hv_vcpu_uninit(vcpu
);
7737 kvm_pmu_destroy(vcpu
);
7738 kfree(vcpu
->arch
.mce_banks
);
7739 kvm_free_lapic(vcpu
);
7740 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7741 kvm_mmu_destroy(vcpu
);
7742 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7743 free_page((unsigned long)vcpu
->arch
.pio_data
);
7744 if (!lapic_in_kernel(vcpu
))
7745 static_key_slow_dec(&kvm_no_apic_vcpu
);
7748 void kvm_arch_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
7750 kvm_x86_ops
->sched_in(vcpu
, cpu
);
7753 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
7758 INIT_HLIST_HEAD(&kvm
->arch
.mask_notifier_list
);
7759 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
7760 INIT_LIST_HEAD(&kvm
->arch
.zapped_obsolete_pages
);
7761 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
7762 atomic_set(&kvm
->arch
.noncoherent_dma_count
, 0);
7764 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7765 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
7766 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7767 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
7768 &kvm
->arch
.irq_sources_bitmap
);
7770 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
7771 mutex_init(&kvm
->arch
.apic_map_lock
);
7772 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
7774 pvclock_update_vm_gtod_copy(kvm
);
7776 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_update_work
, kvmclock_update_fn
);
7777 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_sync_work
, kvmclock_sync_fn
);
7779 kvm_page_track_init(kvm
);
7780 kvm_mmu_init_vm(kvm
);
7782 if (kvm_x86_ops
->vm_init
)
7783 return kvm_x86_ops
->vm_init(kvm
);
7788 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
7791 r
= vcpu_load(vcpu
);
7793 kvm_mmu_unload(vcpu
);
7797 static void kvm_free_vcpus(struct kvm
*kvm
)
7800 struct kvm_vcpu
*vcpu
;
7803 * Unpin any mmu pages first.
7805 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
7806 kvm_clear_async_pf_completion_queue(vcpu
);
7807 kvm_unload_vcpu_mmu(vcpu
);
7809 kvm_for_each_vcpu(i
, vcpu
, kvm
)
7810 kvm_arch_vcpu_free(vcpu
);
7812 mutex_lock(&kvm
->lock
);
7813 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
7814 kvm
->vcpus
[i
] = NULL
;
7816 atomic_set(&kvm
->online_vcpus
, 0);
7817 mutex_unlock(&kvm
->lock
);
7820 void kvm_arch_sync_events(struct kvm
*kvm
)
7822 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_sync_work
);
7823 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_update_work
);
7824 kvm_free_all_assigned_devices(kvm
);
7828 int __x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
7832 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
7833 struct kvm_memory_slot
*slot
, old
;
7835 /* Called with kvm->slots_lock held. */
7836 if (WARN_ON(id
>= KVM_MEM_SLOTS_NUM
))
7839 slot
= id_to_memslot(slots
, id
);
7845 * MAP_SHARED to prevent internal slot pages from being moved
7848 hva
= vm_mmap(NULL
, 0, size
, PROT_READ
| PROT_WRITE
,
7849 MAP_SHARED
| MAP_ANONYMOUS
, 0);
7850 if (IS_ERR((void *)hva
))
7851 return PTR_ERR((void *)hva
);
7860 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
7861 struct kvm_userspace_memory_region m
;
7863 m
.slot
= id
| (i
<< 16);
7865 m
.guest_phys_addr
= gpa
;
7866 m
.userspace_addr
= hva
;
7867 m
.memory_size
= size
;
7868 r
= __kvm_set_memory_region(kvm
, &m
);
7874 r
= vm_munmap(old
.userspace_addr
, old
.npages
* PAGE_SIZE
);
7880 EXPORT_SYMBOL_GPL(__x86_set_memory_region
);
7882 int x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
7886 mutex_lock(&kvm
->slots_lock
);
7887 r
= __x86_set_memory_region(kvm
, id
, gpa
, size
);
7888 mutex_unlock(&kvm
->slots_lock
);
7892 EXPORT_SYMBOL_GPL(x86_set_memory_region
);
7894 void kvm_arch_destroy_vm(struct kvm
*kvm
)
7896 if (current
->mm
== kvm
->mm
) {
7898 * Free memory regions allocated on behalf of userspace,
7899 * unless the the memory map has changed due to process exit
7902 x86_set_memory_region(kvm
, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
, 0, 0);
7903 x86_set_memory_region(kvm
, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
, 0, 0);
7904 x86_set_memory_region(kvm
, TSS_PRIVATE_MEMSLOT
, 0, 0);
7906 if (kvm_x86_ops
->vm_destroy
)
7907 kvm_x86_ops
->vm_destroy(kvm
);
7908 kvm_iommu_unmap_guest(kvm
);
7909 kfree(kvm
->arch
.vpic
);
7910 kfree(kvm
->arch
.vioapic
);
7911 kvm_free_vcpus(kvm
);
7912 kfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
7913 kvm_mmu_uninit_vm(kvm
);
7916 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
7917 struct kvm_memory_slot
*dont
)
7921 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
7922 if (!dont
|| free
->arch
.rmap
[i
] != dont
->arch
.rmap
[i
]) {
7923 kvfree(free
->arch
.rmap
[i
]);
7924 free
->arch
.rmap
[i
] = NULL
;
7929 if (!dont
|| free
->arch
.lpage_info
[i
- 1] !=
7930 dont
->arch
.lpage_info
[i
- 1]) {
7931 kvfree(free
->arch
.lpage_info
[i
- 1]);
7932 free
->arch
.lpage_info
[i
- 1] = NULL
;
7936 kvm_page_track_free_memslot(free
, dont
);
7939 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
7940 unsigned long npages
)
7944 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
7945 struct kvm_lpage_info
*linfo
;
7950 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
7951 slot
->base_gfn
, level
) + 1;
7953 slot
->arch
.rmap
[i
] =
7954 kvm_kvzalloc(lpages
* sizeof(*slot
->arch
.rmap
[i
]));
7955 if (!slot
->arch
.rmap
[i
])
7960 linfo
= kvm_kvzalloc(lpages
* sizeof(*linfo
));
7964 slot
->arch
.lpage_info
[i
- 1] = linfo
;
7966 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
7967 linfo
[0].disallow_lpage
= 1;
7968 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
7969 linfo
[lpages
- 1].disallow_lpage
= 1;
7970 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
7972 * If the gfn and userspace address are not aligned wrt each
7973 * other, or if explicitly asked to, disable large page
7974 * support for this slot
7976 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
7977 !kvm_largepages_enabled()) {
7980 for (j
= 0; j
< lpages
; ++j
)
7981 linfo
[j
].disallow_lpage
= 1;
7985 if (kvm_page_track_create_memslot(slot
, npages
))
7991 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
7992 kvfree(slot
->arch
.rmap
[i
]);
7993 slot
->arch
.rmap
[i
] = NULL
;
7997 kvfree(slot
->arch
.lpage_info
[i
- 1]);
7998 slot
->arch
.lpage_info
[i
- 1] = NULL
;
8003 void kvm_arch_memslots_updated(struct kvm
*kvm
, struct kvm_memslots
*slots
)
8006 * memslots->generation has been incremented.
8007 * mmio generation may have reached its maximum value.
8009 kvm_mmu_invalidate_mmio_sptes(kvm
, slots
);
8012 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
8013 struct kvm_memory_slot
*memslot
,
8014 const struct kvm_userspace_memory_region
*mem
,
8015 enum kvm_mr_change change
)
8020 static void kvm_mmu_slot_apply_flags(struct kvm
*kvm
,
8021 struct kvm_memory_slot
*new)
8023 /* Still write protect RO slot */
8024 if (new->flags
& KVM_MEM_READONLY
) {
8025 kvm_mmu_slot_remove_write_access(kvm
, new);
8030 * Call kvm_x86_ops dirty logging hooks when they are valid.
8032 * kvm_x86_ops->slot_disable_log_dirty is called when:
8034 * - KVM_MR_CREATE with dirty logging is disabled
8035 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8037 * The reason is, in case of PML, we need to set D-bit for any slots
8038 * with dirty logging disabled in order to eliminate unnecessary GPA
8039 * logging in PML buffer (and potential PML buffer full VMEXT). This
8040 * guarantees leaving PML enabled during guest's lifetime won't have
8041 * any additonal overhead from PML when guest is running with dirty
8042 * logging disabled for memory slots.
8044 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8045 * to dirty logging mode.
8047 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8049 * In case of write protect:
8051 * Write protect all pages for dirty logging.
8053 * All the sptes including the large sptes which point to this
8054 * slot are set to readonly. We can not create any new large
8055 * spte on this slot until the end of the logging.
8057 * See the comments in fast_page_fault().
8059 if (new->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
8060 if (kvm_x86_ops
->slot_enable_log_dirty
)
8061 kvm_x86_ops
->slot_enable_log_dirty(kvm
, new);
8063 kvm_mmu_slot_remove_write_access(kvm
, new);
8065 if (kvm_x86_ops
->slot_disable_log_dirty
)
8066 kvm_x86_ops
->slot_disable_log_dirty(kvm
, new);
8070 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
8071 const struct kvm_userspace_memory_region
*mem
,
8072 const struct kvm_memory_slot
*old
,
8073 const struct kvm_memory_slot
*new,
8074 enum kvm_mr_change change
)
8076 int nr_mmu_pages
= 0;
8078 if (!kvm
->arch
.n_requested_mmu_pages
)
8079 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
8082 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
8085 * Dirty logging tracks sptes in 4k granularity, meaning that large
8086 * sptes have to be split. If live migration is successful, the guest
8087 * in the source machine will be destroyed and large sptes will be
8088 * created in the destination. However, if the guest continues to run
8089 * in the source machine (for example if live migration fails), small
8090 * sptes will remain around and cause bad performance.
8092 * Scan sptes if dirty logging has been stopped, dropping those
8093 * which can be collapsed into a single large-page spte. Later
8094 * page faults will create the large-page sptes.
8096 if ((change
!= KVM_MR_DELETE
) &&
8097 (old
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) &&
8098 !(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
8099 kvm_mmu_zap_collapsible_sptes(kvm
, new);
8102 * Set up write protection and/or dirty logging for the new slot.
8104 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8105 * been zapped so no dirty logging staff is needed for old slot. For
8106 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8107 * new and it's also covered when dealing with the new slot.
8109 * FIXME: const-ify all uses of struct kvm_memory_slot.
8111 if (change
!= KVM_MR_DELETE
)
8112 kvm_mmu_slot_apply_flags(kvm
, (struct kvm_memory_slot
*) new);
8115 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
8117 kvm_mmu_invalidate_zap_all_pages(kvm
);
8120 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
8121 struct kvm_memory_slot
*slot
)
8123 kvm_mmu_invalidate_zap_all_pages(kvm
);
8126 static inline bool kvm_vcpu_has_events(struct kvm_vcpu
*vcpu
)
8128 if (!list_empty_careful(&vcpu
->async_pf
.done
))
8131 if (kvm_apic_has_events(vcpu
))
8134 if (vcpu
->arch
.pv
.pv_unhalted
)
8137 if (atomic_read(&vcpu
->arch
.nmi_queued
))
8140 if (test_bit(KVM_REQ_SMI
, &vcpu
->requests
))
8143 if (kvm_arch_interrupt_allowed(vcpu
) &&
8144 kvm_cpu_has_interrupt(vcpu
))
8147 if (kvm_hv_has_stimer_pending(vcpu
))
8153 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
8155 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
)
8156 kvm_x86_ops
->check_nested_events(vcpu
, false);
8158 return kvm_vcpu_running(vcpu
) || kvm_vcpu_has_events(vcpu
);
8161 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
8163 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
8166 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
8168 return kvm_x86_ops
->interrupt_allowed(vcpu
);
8171 unsigned long kvm_get_linear_rip(struct kvm_vcpu
*vcpu
)
8173 if (is_64_bit_mode(vcpu
))
8174 return kvm_rip_read(vcpu
);
8175 return (u32
)(get_segment_base(vcpu
, VCPU_SREG_CS
) +
8176 kvm_rip_read(vcpu
));
8178 EXPORT_SYMBOL_GPL(kvm_get_linear_rip
);
8180 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
8182 return kvm_get_linear_rip(vcpu
) == linear_rip
;
8184 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
8186 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
8188 unsigned long rflags
;
8190 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
8191 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
8192 rflags
&= ~X86_EFLAGS_TF
;
8195 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
8197 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8199 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
8200 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
8201 rflags
|= X86_EFLAGS_TF
;
8202 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
8205 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8207 __kvm_set_rflags(vcpu
, rflags
);
8208 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8210 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
8212 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
8216 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
8220 r
= kvm_mmu_reload(vcpu
);
8224 if (!vcpu
->arch
.mmu
.direct_map
&&
8225 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
8228 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
8231 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
8233 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
8236 static inline u32
kvm_async_pf_next_probe(u32 key
)
8238 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
8241 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8243 u32 key
= kvm_async_pf_hash_fn(gfn
);
8245 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
8246 key
= kvm_async_pf_next_probe(key
);
8248 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
8251 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8254 u32 key
= kvm_async_pf_hash_fn(gfn
);
8256 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
8257 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
8258 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
8259 key
= kvm_async_pf_next_probe(key
);
8264 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8266 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
8269 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8273 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
8275 vcpu
->arch
.apf
.gfns
[i
] = ~0;
8277 j
= kvm_async_pf_next_probe(j
);
8278 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
8280 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
8282 * k lies cyclically in ]i,j]
8284 * |....j i.k.| or |.k..j i...|
8286 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
8287 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
8292 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
8295 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
8299 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
8300 struct kvm_async_pf
*work
)
8302 struct x86_exception fault
;
8304 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
8305 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8307 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
8308 (vcpu
->arch
.apf
.send_user_only
&&
8309 kvm_x86_ops
->get_cpl(vcpu
) == 0))
8310 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
8311 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
8312 fault
.vector
= PF_VECTOR
;
8313 fault
.error_code_valid
= true;
8314 fault
.error_code
= 0;
8315 fault
.nested_page_fault
= false;
8316 fault
.address
= work
->arch
.token
;
8317 kvm_inject_page_fault(vcpu
, &fault
);
8321 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
8322 struct kvm_async_pf
*work
)
8324 struct x86_exception fault
;
8326 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
8327 if (work
->wakeup_all
)
8328 work
->arch
.token
= ~0; /* broadcast wakeup */
8330 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8332 if ((vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) &&
8333 !apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
8334 fault
.vector
= PF_VECTOR
;
8335 fault
.error_code_valid
= true;
8336 fault
.error_code
= 0;
8337 fault
.nested_page_fault
= false;
8338 fault
.address
= work
->arch
.token
;
8339 kvm_inject_page_fault(vcpu
, &fault
);
8341 vcpu
->arch
.apf
.halted
= false;
8342 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
8345 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
8347 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
8350 return !kvm_event_needs_reinjection(vcpu
) &&
8351 kvm_x86_ops
->interrupt_allowed(vcpu
);
8354 void kvm_arch_start_assignment(struct kvm
*kvm
)
8356 atomic_inc(&kvm
->arch
.assigned_device_count
);
8358 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment
);
8360 void kvm_arch_end_assignment(struct kvm
*kvm
)
8362 atomic_dec(&kvm
->arch
.assigned_device_count
);
8364 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment
);
8366 bool kvm_arch_has_assigned_device(struct kvm
*kvm
)
8368 return atomic_read(&kvm
->arch
.assigned_device_count
);
8370 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device
);
8372 void kvm_arch_register_noncoherent_dma(struct kvm
*kvm
)
8374 atomic_inc(&kvm
->arch
.noncoherent_dma_count
);
8376 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma
);
8378 void kvm_arch_unregister_noncoherent_dma(struct kvm
*kvm
)
8380 atomic_dec(&kvm
->arch
.noncoherent_dma_count
);
8382 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma
);
8384 bool kvm_arch_has_noncoherent_dma(struct kvm
*kvm
)
8386 return atomic_read(&kvm
->arch
.noncoherent_dma_count
);
8388 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma
);
8390 bool kvm_arch_has_irq_bypass(void)
8392 return kvm_x86_ops
->update_pi_irte
!= NULL
;
8395 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
8396 struct irq_bypass_producer
*prod
)
8398 struct kvm_kernel_irqfd
*irqfd
=
8399 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
8401 irqfd
->producer
= prod
;
8403 return kvm_x86_ops
->update_pi_irte(irqfd
->kvm
,
8404 prod
->irq
, irqfd
->gsi
, 1);
8407 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
8408 struct irq_bypass_producer
*prod
)
8411 struct kvm_kernel_irqfd
*irqfd
=
8412 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
8414 WARN_ON(irqfd
->producer
!= prod
);
8415 irqfd
->producer
= NULL
;
8418 * When producer of consumer is unregistered, we change back to
8419 * remapped mode, so we can re-use the current implementation
8420 * when the irq is masked/disabled or the consumer side (KVM
8421 * int this case doesn't want to receive the interrupts.
8423 ret
= kvm_x86_ops
->update_pi_irte(irqfd
->kvm
, prod
->irq
, irqfd
->gsi
, 0);
8425 printk(KERN_INFO
"irq bypass consumer (token %p) unregistration"
8426 " fails: %d\n", irqfd
->consumer
.token
, ret
);
8429 int kvm_arch_update_irqfd_routing(struct kvm
*kvm
, unsigned int host_irq
,
8430 uint32_t guest_irq
, bool set
)
8432 if (!kvm_x86_ops
->update_pi_irte
)
8435 return kvm_x86_ops
->update_pi_irte(kvm
, host_irq
, guest_irq
, set
);
8438 bool kvm_vector_hashing_enabled(void)
8440 return vector_hashing
;
8442 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled
);
8444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
8445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio
);
8446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
8447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
8448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
8449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
8450 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
8451 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
8452 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
8453 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
8454 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
8455 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
8456 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);
8457 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset
);
8458 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window
);
8459 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full
);
8460 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update
);
8461 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access
);
8462 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi
);