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"
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <trace/events/kvm.h>
53 #define CREATE_TRACE_POINTS
56 #include <asm/debugreg.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 #define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
80 u64 __read_mostly efer_reserved_bits
= ~((u64
)(EFER_SCE
| EFER_LME
| EFER_LMA
));
82 static u64 __read_mostly efer_reserved_bits
= ~((u64
)EFER_SCE
);
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
);
89 static void process_nmi(struct kvm_vcpu
*vcpu
);
91 struct kvm_x86_ops
*kvm_x86_ops
;
92 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
94 static bool ignore_msrs
= 0;
95 module_param(ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
97 bool kvm_has_tsc_control
;
98 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
99 u32 kvm_max_guest_tsc_khz
;
100 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
102 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
103 static u32 tsc_tolerance_ppm
= 250;
104 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
106 #define KVM_NR_SHARED_MSRS 16
108 struct kvm_shared_msrs_global
{
110 u32 msrs
[KVM_NR_SHARED_MSRS
];
113 struct kvm_shared_msrs
{
114 struct user_return_notifier urn
;
116 struct kvm_shared_msr_values
{
119 } values
[KVM_NR_SHARED_MSRS
];
122 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
123 static struct kvm_shared_msrs __percpu
*shared_msrs
;
125 struct kvm_stats_debugfs_item debugfs_entries
[] = {
126 { "pf_fixed", VCPU_STAT(pf_fixed
) },
127 { "pf_guest", VCPU_STAT(pf_guest
) },
128 { "tlb_flush", VCPU_STAT(tlb_flush
) },
129 { "invlpg", VCPU_STAT(invlpg
) },
130 { "exits", VCPU_STAT(exits
) },
131 { "io_exits", VCPU_STAT(io_exits
) },
132 { "mmio_exits", VCPU_STAT(mmio_exits
) },
133 { "signal_exits", VCPU_STAT(signal_exits
) },
134 { "irq_window", VCPU_STAT(irq_window_exits
) },
135 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
136 { "halt_exits", VCPU_STAT(halt_exits
) },
137 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
138 { "hypercalls", VCPU_STAT(hypercalls
) },
139 { "request_irq", VCPU_STAT(request_irq_exits
) },
140 { "irq_exits", VCPU_STAT(irq_exits
) },
141 { "host_state_reload", VCPU_STAT(host_state_reload
) },
142 { "efer_reload", VCPU_STAT(efer_reload
) },
143 { "fpu_reload", VCPU_STAT(fpu_reload
) },
144 { "insn_emulation", VCPU_STAT(insn_emulation
) },
145 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
146 { "irq_injections", VCPU_STAT(irq_injections
) },
147 { "nmi_injections", VCPU_STAT(nmi_injections
) },
148 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
149 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
150 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
151 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
152 { "mmu_flooded", VM_STAT(mmu_flooded
) },
153 { "mmu_recycled", VM_STAT(mmu_recycled
) },
154 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
155 { "mmu_unsync", VM_STAT(mmu_unsync
) },
156 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
157 { "largepages", VM_STAT(lpages
) },
161 u64 __read_mostly host_xcr0
;
163 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
165 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
168 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
169 vcpu
->arch
.apf
.gfns
[i
] = ~0;
172 static void kvm_on_user_return(struct user_return_notifier
*urn
)
175 struct kvm_shared_msrs
*locals
176 = container_of(urn
, struct kvm_shared_msrs
, urn
);
177 struct kvm_shared_msr_values
*values
;
179 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
180 values
= &locals
->values
[slot
];
181 if (values
->host
!= values
->curr
) {
182 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
183 values
->curr
= values
->host
;
186 locals
->registered
= false;
187 user_return_notifier_unregister(urn
);
190 static void shared_msr_update(unsigned slot
, u32 msr
)
193 unsigned int cpu
= smp_processor_id();
194 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
196 /* only read, and nobody should modify it at this time,
197 * so don't need lock */
198 if (slot
>= shared_msrs_global
.nr
) {
199 printk(KERN_ERR
"kvm: invalid MSR slot!");
202 rdmsrl_safe(msr
, &value
);
203 smsr
->values
[slot
].host
= value
;
204 smsr
->values
[slot
].curr
= value
;
207 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
209 if (slot
>= shared_msrs_global
.nr
)
210 shared_msrs_global
.nr
= slot
+ 1;
211 shared_msrs_global
.msrs
[slot
] = msr
;
212 /* we need ensured the shared_msr_global have been updated */
215 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
217 static void kvm_shared_msr_cpu_online(void)
221 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
222 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
225 void kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
227 unsigned int cpu
= smp_processor_id();
228 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
230 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
232 smsr
->values
[slot
].curr
= value
;
233 wrmsrl(shared_msrs_global
.msrs
[slot
], value
);
234 if (!smsr
->registered
) {
235 smsr
->urn
.on_user_return
= kvm_on_user_return
;
236 user_return_notifier_register(&smsr
->urn
);
237 smsr
->registered
= true;
240 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
242 static void drop_user_return_notifiers(void *ignore
)
244 unsigned int cpu
= smp_processor_id();
245 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
247 if (smsr
->registered
)
248 kvm_on_user_return(&smsr
->urn
);
251 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
253 return vcpu
->arch
.apic_base
;
255 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
257 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
259 /* TODO: reserve bits check */
260 kvm_lapic_set_base(vcpu
, data
);
262 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
264 asmlinkage
void kvm_spurious_fault(void)
266 /* Fault while not rebooting. We want the trace. */
269 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
271 #define EXCPT_BENIGN 0
272 #define EXCPT_CONTRIBUTORY 1
275 static int exception_class(int vector
)
285 return EXCPT_CONTRIBUTORY
;
292 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
293 unsigned nr
, bool has_error
, u32 error_code
,
299 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
301 if (!vcpu
->arch
.exception
.pending
) {
303 vcpu
->arch
.exception
.pending
= true;
304 vcpu
->arch
.exception
.has_error_code
= has_error
;
305 vcpu
->arch
.exception
.nr
= nr
;
306 vcpu
->arch
.exception
.error_code
= error_code
;
307 vcpu
->arch
.exception
.reinject
= reinject
;
311 /* to check exception */
312 prev_nr
= vcpu
->arch
.exception
.nr
;
313 if (prev_nr
== DF_VECTOR
) {
314 /* triple fault -> shutdown */
315 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
318 class1
= exception_class(prev_nr
);
319 class2
= exception_class(nr
);
320 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
321 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
322 /* generate double fault per SDM Table 5-5 */
323 vcpu
->arch
.exception
.pending
= true;
324 vcpu
->arch
.exception
.has_error_code
= true;
325 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
326 vcpu
->arch
.exception
.error_code
= 0;
328 /* replace previous exception with a new one in a hope
329 that instruction re-execution will regenerate lost
334 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
336 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
338 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
340 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
342 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
344 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
346 void kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
349 kvm_inject_gp(vcpu
, 0);
351 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
353 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
355 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
357 ++vcpu
->stat
.pf_guest
;
358 vcpu
->arch
.cr2
= fault
->address
;
359 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
361 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
363 void kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
365 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
366 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
368 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
371 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
373 atomic_inc(&vcpu
->arch
.nmi_queued
);
374 kvm_make_request(KVM_REQ_NMI
, vcpu
);
376 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
378 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
380 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
382 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
384 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
386 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
388 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
391 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
392 * a #GP and return false.
394 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
396 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
398 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
401 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
404 * This function will be used to read from the physical memory of the currently
405 * running guest. The difference to kvm_read_guest_page is that this function
406 * can read from guest physical or from the guest's guest physical memory.
408 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
409 gfn_t ngfn
, void *data
, int offset
, int len
,
415 ngpa
= gfn_to_gpa(ngfn
);
416 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
);
417 if (real_gfn
== UNMAPPED_GVA
)
420 real_gfn
= gpa_to_gfn(real_gfn
);
422 return kvm_read_guest_page(vcpu
->kvm
, real_gfn
, data
, offset
, len
);
424 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
426 int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
427 void *data
, int offset
, int len
, u32 access
)
429 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
430 data
, offset
, len
, access
);
434 * Load the pae pdptrs. Return true is they are all valid.
436 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
438 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
439 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
442 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
444 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
445 offset
* sizeof(u64
), sizeof(pdpte
),
446 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
451 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
452 if (is_present_gpte(pdpte
[i
]) &&
453 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
460 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
461 __set_bit(VCPU_EXREG_PDPTR
,
462 (unsigned long *)&vcpu
->arch
.regs_avail
);
463 __set_bit(VCPU_EXREG_PDPTR
,
464 (unsigned long *)&vcpu
->arch
.regs_dirty
);
469 EXPORT_SYMBOL_GPL(load_pdptrs
);
471 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
473 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
479 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
482 if (!test_bit(VCPU_EXREG_PDPTR
,
483 (unsigned long *)&vcpu
->arch
.regs_avail
))
486 gfn
= (kvm_read_cr3(vcpu
) & ~31u) >> PAGE_SHIFT
;
487 offset
= (kvm_read_cr3(vcpu
) & ~31u) & (PAGE_SIZE
- 1);
488 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
489 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
492 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
498 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
500 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
501 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
|
502 X86_CR0_CD
| X86_CR0_NW
;
507 if (cr0
& 0xffffffff00000000UL
)
511 cr0
&= ~CR0_RESERVED_BITS
;
513 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
516 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
519 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
521 if ((vcpu
->arch
.efer
& EFER_LME
)) {
526 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
531 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
536 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
539 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
541 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
542 kvm_clear_async_pf_completion_queue(vcpu
);
543 kvm_async_pf_hash_reset(vcpu
);
546 if ((cr0
^ old_cr0
) & update_bits
)
547 kvm_mmu_reset_context(vcpu
);
550 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
552 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
554 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
556 EXPORT_SYMBOL_GPL(kvm_lmsw
);
558 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
560 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
561 !vcpu
->guest_xcr0_loaded
) {
562 /* kvm_set_xcr() also depends on this */
563 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
564 vcpu
->guest_xcr0_loaded
= 1;
568 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
570 if (vcpu
->guest_xcr0_loaded
) {
571 if (vcpu
->arch
.xcr0
!= host_xcr0
)
572 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
573 vcpu
->guest_xcr0_loaded
= 0;
577 int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
581 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
582 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
585 if (kvm_x86_ops
->get_cpl(vcpu
) != 0)
587 if (!(xcr0
& XSTATE_FP
))
589 if ((xcr0
& XSTATE_YMM
) && !(xcr0
& XSTATE_SSE
))
591 if (xcr0
& ~host_xcr0
)
593 kvm_put_guest_xcr0(vcpu
);
594 vcpu
->arch
.xcr0
= xcr0
;
598 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
600 if (__kvm_set_xcr(vcpu
, index
, xcr
)) {
601 kvm_inject_gp(vcpu
, 0);
606 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
608 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
610 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
611 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
|
612 X86_CR4_PAE
| X86_CR4_SMEP
;
613 if (cr4
& CR4_RESERVED_BITS
)
616 if (!guest_cpuid_has_xsave(vcpu
) && (cr4
& X86_CR4_OSXSAVE
))
619 if (!guest_cpuid_has_smep(vcpu
) && (cr4
& X86_CR4_SMEP
))
622 if (!guest_cpuid_has_fsgsbase(vcpu
) && (cr4
& X86_CR4_FSGSBASE
))
625 if (is_long_mode(vcpu
)) {
626 if (!(cr4
& X86_CR4_PAE
))
628 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
629 && ((cr4
^ old_cr4
) & pdptr_bits
)
630 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
634 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
635 if (!guest_cpuid_has_pcid(vcpu
))
638 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
639 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
643 if (kvm_x86_ops
->set_cr4(vcpu
, cr4
))
646 if (((cr4
^ old_cr4
) & pdptr_bits
) ||
647 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
648 kvm_mmu_reset_context(vcpu
);
650 if ((cr4
^ old_cr4
) & X86_CR4_OSXSAVE
)
651 kvm_update_cpuid(vcpu
);
655 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
657 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
659 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
660 kvm_mmu_sync_roots(vcpu
);
661 kvm_mmu_flush_tlb(vcpu
);
665 if (is_long_mode(vcpu
)) {
666 if (kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
)) {
667 if (cr3
& CR3_PCID_ENABLED_RESERVED_BITS
)
670 if (cr3
& CR3_L_MODE_RESERVED_BITS
)
674 if (cr3
& CR3_PAE_RESERVED_BITS
)
676 if (is_paging(vcpu
) &&
677 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
681 * We don't check reserved bits in nonpae mode, because
682 * this isn't enforced, and VMware depends on this.
687 * Does the new cr3 value map to physical memory? (Note, we
688 * catch an invalid cr3 even in real-mode, because it would
689 * cause trouble later on when we turn on paging anyway.)
691 * A real CPU would silently accept an invalid cr3 and would
692 * attempt to use it - with largely undefined (and often hard
693 * to debug) behavior on the guest side.
695 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
697 vcpu
->arch
.cr3
= cr3
;
698 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
699 vcpu
->arch
.mmu
.new_cr3(vcpu
);
702 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
704 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
706 if (cr8
& CR8_RESERVED_BITS
)
708 if (irqchip_in_kernel(vcpu
->kvm
))
709 kvm_lapic_set_tpr(vcpu
, cr8
);
711 vcpu
->arch
.cr8
= cr8
;
714 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
716 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
718 if (irqchip_in_kernel(vcpu
->kvm
))
719 return kvm_lapic_get_cr8(vcpu
);
721 return vcpu
->arch
.cr8
;
723 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
725 static void kvm_update_dr7(struct kvm_vcpu
*vcpu
)
729 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
730 dr7
= vcpu
->arch
.guest_debug_dr7
;
732 dr7
= vcpu
->arch
.dr7
;
733 kvm_x86_ops
->set_dr7(vcpu
, dr7
);
734 vcpu
->arch
.switch_db_regs
= (dr7
& DR7_BP_EN_MASK
);
737 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
741 vcpu
->arch
.db
[dr
] = val
;
742 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
743 vcpu
->arch
.eff_db
[dr
] = val
;
746 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
750 if (val
& 0xffffffff00000000ULL
)
752 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | DR6_FIXED_1
;
755 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
759 if (val
& 0xffffffff00000000ULL
)
761 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
762 kvm_update_dr7(vcpu
);
769 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
773 res
= __kvm_set_dr(vcpu
, dr
, val
);
775 kvm_queue_exception(vcpu
, UD_VECTOR
);
777 kvm_inject_gp(vcpu
, 0);
781 EXPORT_SYMBOL_GPL(kvm_set_dr
);
783 static int _kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
787 *val
= vcpu
->arch
.db
[dr
];
790 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
794 *val
= vcpu
->arch
.dr6
;
797 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
801 *val
= vcpu
->arch
.dr7
;
808 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
810 if (_kvm_get_dr(vcpu
, dr
, val
)) {
811 kvm_queue_exception(vcpu
, UD_VECTOR
);
816 EXPORT_SYMBOL_GPL(kvm_get_dr
);
818 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
820 u32 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
824 err
= kvm_pmu_read_pmc(vcpu
, ecx
, &data
);
827 kvm_register_write(vcpu
, VCPU_REGS_RAX
, (u32
)data
);
828 kvm_register_write(vcpu
, VCPU_REGS_RDX
, data
>> 32);
831 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
834 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
835 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
837 * This list is modified at module load time to reflect the
838 * capabilities of the host cpu. This capabilities test skips MSRs that are
839 * kvm-specific. Those are put in the beginning of the list.
842 #define KVM_SAVE_MSRS_BEGIN 10
843 static u32 msrs_to_save
[] = {
844 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
845 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
846 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
847 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
849 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
852 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
854 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
857 static unsigned num_msrs_to_save
;
859 static const u32 emulated_msrs
[] = {
861 MSR_IA32_TSCDEADLINE
,
862 MSR_IA32_MISC_ENABLE
,
867 bool kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
869 if (efer
& efer_reserved_bits
)
872 if (efer
& EFER_FFXSR
) {
873 struct kvm_cpuid_entry2
*feat
;
875 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
876 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
)))
880 if (efer
& EFER_SVME
) {
881 struct kvm_cpuid_entry2
*feat
;
883 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
884 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
)))
890 EXPORT_SYMBOL_GPL(kvm_valid_efer
);
892 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
894 u64 old_efer
= vcpu
->arch
.efer
;
896 if (!kvm_valid_efer(vcpu
, efer
))
900 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
904 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
906 kvm_x86_ops
->set_efer(vcpu
, efer
);
908 /* Update reserved bits */
909 if ((efer
^ old_efer
) & EFER_NX
)
910 kvm_mmu_reset_context(vcpu
);
915 void kvm_enable_efer_bits(u64 mask
)
917 efer_reserved_bits
&= ~mask
;
919 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
923 * Writes msr value into into the appropriate "register".
924 * Returns 0 on success, non-0 otherwise.
925 * Assumes vcpu_load() was already called.
927 int kvm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
929 return kvm_x86_ops
->set_msr(vcpu
, msr
);
933 * Adapt set_msr() to msr_io()'s calling convention
935 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
941 msr
.host_initiated
= true;
942 return kvm_set_msr(vcpu
, &msr
);
946 struct pvclock_gtod_data
{
949 struct { /* extract of a clocksource struct */
957 /* open coded 'struct timespec' */
958 u64 monotonic_time_snsec
;
959 time_t monotonic_time_sec
;
962 static struct pvclock_gtod_data pvclock_gtod_data
;
964 static void update_pvclock_gtod(struct timekeeper
*tk
)
966 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
968 write_seqcount_begin(&vdata
->seq
);
970 /* copy pvclock gtod data */
971 vdata
->clock
.vclock_mode
= tk
->clock
->archdata
.vclock_mode
;
972 vdata
->clock
.cycle_last
= tk
->clock
->cycle_last
;
973 vdata
->clock
.mask
= tk
->clock
->mask
;
974 vdata
->clock
.mult
= tk
->mult
;
975 vdata
->clock
.shift
= tk
->shift
;
977 vdata
->monotonic_time_sec
= tk
->xtime_sec
978 + tk
->wall_to_monotonic
.tv_sec
;
979 vdata
->monotonic_time_snsec
= tk
->xtime_nsec
980 + (tk
->wall_to_monotonic
.tv_nsec
982 while (vdata
->monotonic_time_snsec
>=
983 (((u64
)NSEC_PER_SEC
) << tk
->shift
)) {
984 vdata
->monotonic_time_snsec
-=
985 ((u64
)NSEC_PER_SEC
) << tk
->shift
;
986 vdata
->monotonic_time_sec
++;
989 write_seqcount_end(&vdata
->seq
);
994 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
998 struct pvclock_wall_clock wc
;
999 struct timespec boot
;
1004 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1009 ++version
; /* first time write, random junk */
1013 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1016 * The guest calculates current wall clock time by adding
1017 * system time (updated by kvm_guest_time_update below) to the
1018 * wall clock specified here. guest system time equals host
1019 * system time for us, thus we must fill in host boot time here.
1023 if (kvm
->arch
.kvmclock_offset
) {
1024 struct timespec ts
= ns_to_timespec(kvm
->arch
.kvmclock_offset
);
1025 boot
= timespec_sub(boot
, ts
);
1027 wc
.sec
= boot
.tv_sec
;
1028 wc
.nsec
= boot
.tv_nsec
;
1029 wc
.version
= version
;
1031 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1034 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1037 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
1039 uint32_t quotient
, remainder
;
1041 /* Don't try to replace with do_div(), this one calculates
1042 * "(dividend << 32) / divisor" */
1044 : "=a" (quotient
), "=d" (remainder
)
1045 : "0" (0), "1" (dividend
), "r" (divisor
) );
1049 static void kvm_get_time_scale(uint32_t scaled_khz
, uint32_t base_khz
,
1050 s8
*pshift
, u32
*pmultiplier
)
1057 tps64
= base_khz
* 1000LL;
1058 scaled64
= scaled_khz
* 1000LL;
1059 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
1064 tps32
= (uint32_t)tps64
;
1065 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
1066 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
1074 *pmultiplier
= div_frac(scaled64
, tps32
);
1076 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1077 __func__
, base_khz
, scaled_khz
, shift
, *pmultiplier
);
1080 static inline u64
get_kernel_ns(void)
1084 WARN_ON(preemptible());
1086 monotonic_to_bootbased(&ts
);
1087 return timespec_to_ns(&ts
);
1090 #ifdef CONFIG_X86_64
1091 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
1094 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
1095 unsigned long max_tsc_khz
;
1097 static inline u64
nsec_to_cycles(struct kvm_vcpu
*vcpu
, u64 nsec
)
1099 return pvclock_scale_delta(nsec
, vcpu
->arch
.virtual_tsc_mult
,
1100 vcpu
->arch
.virtual_tsc_shift
);
1103 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1105 u64 v
= (u64
)khz
* (1000000 + ppm
);
1110 static void kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 this_tsc_khz
)
1112 u32 thresh_lo
, thresh_hi
;
1113 int use_scaling
= 0;
1115 /* tsc_khz can be zero if TSC calibration fails */
1116 if (this_tsc_khz
== 0)
1119 /* Compute a scale to convert nanoseconds in TSC cycles */
1120 kvm_get_time_scale(this_tsc_khz
, NSEC_PER_SEC
/ 1000,
1121 &vcpu
->arch
.virtual_tsc_shift
,
1122 &vcpu
->arch
.virtual_tsc_mult
);
1123 vcpu
->arch
.virtual_tsc_khz
= this_tsc_khz
;
1126 * Compute the variation in TSC rate which is acceptable
1127 * within the range of tolerance and decide if the
1128 * rate being applied is within that bounds of the hardware
1129 * rate. If so, no scaling or compensation need be done.
1131 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1132 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1133 if (this_tsc_khz
< thresh_lo
|| this_tsc_khz
> thresh_hi
) {
1134 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz
, thresh_lo
, thresh_hi
);
1137 kvm_x86_ops
->set_tsc_khz(vcpu
, this_tsc_khz
, use_scaling
);
1140 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1142 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1143 vcpu
->arch
.virtual_tsc_mult
,
1144 vcpu
->arch
.virtual_tsc_shift
);
1145 tsc
+= vcpu
->arch
.this_tsc_write
;
1149 void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
1151 #ifdef CONFIG_X86_64
1153 bool do_request
= false;
1154 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1155 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1157 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1158 atomic_read(&vcpu
->kvm
->online_vcpus
));
1160 if (vcpus_matched
&& gtod
->clock
.vclock_mode
== VCLOCK_TSC
)
1161 if (!ka
->use_master_clock
)
1164 if (!vcpus_matched
&& ka
->use_master_clock
)
1168 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1170 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
1171 atomic_read(&vcpu
->kvm
->online_vcpus
),
1172 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
1176 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu
*vcpu
, s64 offset
)
1178 u64 curr_offset
= kvm_x86_ops
->read_tsc_offset(vcpu
);
1179 vcpu
->arch
.ia32_tsc_adjust_msr
+= offset
- curr_offset
;
1182 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1184 struct kvm
*kvm
= vcpu
->kvm
;
1185 u64 offset
, ns
, elapsed
;
1186 unsigned long flags
;
1189 u64 data
= msr
->data
;
1191 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1192 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1193 ns
= get_kernel_ns();
1194 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1196 if (vcpu
->arch
.virtual_tsc_khz
) {
1197 /* n.b - signed multiplication and division required */
1198 usdiff
= data
- kvm
->arch
.last_tsc_write
;
1199 #ifdef CONFIG_X86_64
1200 usdiff
= (usdiff
* 1000) / vcpu
->arch
.virtual_tsc_khz
;
1202 /* do_div() only does unsigned */
1203 asm("idivl %2; xor %%edx, %%edx"
1205 : "A"(usdiff
* 1000), "rm"(vcpu
->arch
.virtual_tsc_khz
));
1207 do_div(elapsed
, 1000);
1212 usdiff
= USEC_PER_SEC
; /* disable TSC match window below */
1215 * Special case: TSC write with a small delta (1 second) of virtual
1216 * cycle time against real time is interpreted as an attempt to
1217 * synchronize the CPU.
1219 * For a reliable TSC, we can match TSC offsets, and for an unstable
1220 * TSC, we add elapsed time in this computation. We could let the
1221 * compensation code attempt to catch up if we fall behind, but
1222 * it's better to try to match offsets from the beginning.
1224 if (usdiff
< USEC_PER_SEC
&&
1225 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1226 if (!check_tsc_unstable()) {
1227 offset
= kvm
->arch
.cur_tsc_offset
;
1228 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1230 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1232 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1233 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1238 * We split periods of matched TSC writes into generations.
1239 * For each generation, we track the original measured
1240 * nanosecond time, offset, and write, so if TSCs are in
1241 * sync, we can match exact offset, and if not, we can match
1242 * exact software computation in compute_guest_tsc()
1244 * These values are tracked in kvm->arch.cur_xxx variables.
1246 kvm
->arch
.cur_tsc_generation
++;
1247 kvm
->arch
.cur_tsc_nsec
= ns
;
1248 kvm
->arch
.cur_tsc_write
= data
;
1249 kvm
->arch
.cur_tsc_offset
= offset
;
1251 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1252 kvm
->arch
.cur_tsc_generation
, data
);
1256 * We also track th most recent recorded KHZ, write and time to
1257 * allow the matching interval to be extended at each write.
1259 kvm
->arch
.last_tsc_nsec
= ns
;
1260 kvm
->arch
.last_tsc_write
= data
;
1261 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1263 /* Reset of TSC must disable overshoot protection below */
1264 vcpu
->arch
.hv_clock
.tsc_timestamp
= 0;
1265 vcpu
->arch
.last_guest_tsc
= data
;
1267 /* Keep track of which generation this VCPU has synchronized to */
1268 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1269 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1270 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1272 if (guest_cpuid_has_tsc_adjust(vcpu
) && !msr
->host_initiated
)
1273 update_ia32_tsc_adjust_msr(vcpu
, offset
);
1274 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1275 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1277 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1279 kvm
->arch
.nr_vcpus_matched_tsc
++;
1281 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
1283 kvm_track_tsc_matching(vcpu
);
1284 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1287 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1289 #ifdef CONFIG_X86_64
1291 static cycle_t
read_tsc(void)
1297 * Empirically, a fence (of type that depends on the CPU)
1298 * before rdtsc is enough to ensure that rdtsc is ordered
1299 * with respect to loads. The various CPU manuals are unclear
1300 * as to whether rdtsc can be reordered with later loads,
1301 * but no one has ever seen it happen.
1304 ret
= (cycle_t
)vget_cycles();
1306 last
= pvclock_gtod_data
.clock
.cycle_last
;
1308 if (likely(ret
>= last
))
1312 * GCC likes to generate cmov here, but this branch is extremely
1313 * predictable (it's just a funciton of time and the likely is
1314 * very likely) and there's a data dependence, so force GCC
1315 * to generate a branch instead. I don't barrier() because
1316 * we don't actually need a barrier, and if this function
1317 * ever gets inlined it will generate worse code.
1323 static inline u64
vgettsc(cycle_t
*cycle_now
)
1326 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1328 *cycle_now
= read_tsc();
1330 v
= (*cycle_now
- gtod
->clock
.cycle_last
) & gtod
->clock
.mask
;
1331 return v
* gtod
->clock
.mult
;
1334 static int do_monotonic(struct timespec
*ts
, cycle_t
*cycle_now
)
1339 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1343 seq
= read_seqcount_begin(>od
->seq
);
1344 mode
= gtod
->clock
.vclock_mode
;
1345 ts
->tv_sec
= gtod
->monotonic_time_sec
;
1346 ns
= gtod
->monotonic_time_snsec
;
1347 ns
+= vgettsc(cycle_now
);
1348 ns
>>= gtod
->clock
.shift
;
1349 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1350 timespec_add_ns(ts
, ns
);
1355 /* returns true if host is using tsc clocksource */
1356 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, cycle_t
*cycle_now
)
1360 /* checked again under seqlock below */
1361 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1364 if (do_monotonic(&ts
, cycle_now
) != VCLOCK_TSC
)
1367 monotonic_to_bootbased(&ts
);
1368 *kernel_ns
= timespec_to_ns(&ts
);
1376 * Assuming a stable TSC across physical CPUS, and a stable TSC
1377 * across virtual CPUs, the following condition is possible.
1378 * Each numbered line represents an event visible to both
1379 * CPUs at the next numbered event.
1381 * "timespecX" represents host monotonic time. "tscX" represents
1384 * VCPU0 on CPU0 | VCPU1 on CPU1
1386 * 1. read timespec0,tsc0
1387 * 2. | timespec1 = timespec0 + N
1389 * 3. transition to guest | transition to guest
1390 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1391 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1392 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1394 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1397 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1399 * - 0 < N - M => M < N
1401 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1402 * always the case (the difference between two distinct xtime instances
1403 * might be smaller then the difference between corresponding TSC reads,
1404 * when updating guest vcpus pvclock areas).
1406 * To avoid that problem, do not allow visibility of distinct
1407 * system_timestamp/tsc_timestamp values simultaneously: use a master
1408 * copy of host monotonic time values. Update that master copy
1411 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1415 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
1417 #ifdef CONFIG_X86_64
1418 struct kvm_arch
*ka
= &kvm
->arch
;
1420 bool host_tsc_clocksource
, vcpus_matched
;
1422 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1423 atomic_read(&kvm
->online_vcpus
));
1426 * If the host uses TSC clock, then passthrough TSC as stable
1429 host_tsc_clocksource
= kvm_get_time_and_clockread(
1430 &ka
->master_kernel_ns
,
1431 &ka
->master_cycle_now
);
1433 ka
->use_master_clock
= host_tsc_clocksource
& vcpus_matched
;
1435 if (ka
->use_master_clock
)
1436 atomic_set(&kvm_guest_has_master_clock
, 1);
1438 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
1439 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
1444 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1446 unsigned long flags
, this_tsc_khz
;
1447 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1448 struct kvm_arch
*ka
= &v
->kvm
->arch
;
1449 s64 kernel_ns
, max_kernel_ns
;
1450 u64 tsc_timestamp
, host_tsc
;
1451 struct pvclock_vcpu_time_info guest_hv_clock
;
1453 bool use_master_clock
;
1459 * If the host uses TSC clock, then passthrough TSC as stable
1462 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1463 use_master_clock
= ka
->use_master_clock
;
1464 if (use_master_clock
) {
1465 host_tsc
= ka
->master_cycle_now
;
1466 kernel_ns
= ka
->master_kernel_ns
;
1468 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1470 /* Keep irq disabled to prevent changes to the clock */
1471 local_irq_save(flags
);
1472 this_tsc_khz
= __get_cpu_var(cpu_tsc_khz
);
1473 if (unlikely(this_tsc_khz
== 0)) {
1474 local_irq_restore(flags
);
1475 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1478 if (!use_master_clock
) {
1479 host_tsc
= native_read_tsc();
1480 kernel_ns
= get_kernel_ns();
1483 tsc_timestamp
= kvm_x86_ops
->read_l1_tsc(v
, host_tsc
);
1486 * We may have to catch up the TSC to match elapsed wall clock
1487 * time for two reasons, even if kvmclock is used.
1488 * 1) CPU could have been running below the maximum TSC rate
1489 * 2) Broken TSC compensation resets the base at each VCPU
1490 * entry to avoid unknown leaps of TSC even when running
1491 * again on the same CPU. This may cause apparent elapsed
1492 * time to disappear, and the guest to stand still or run
1495 if (vcpu
->tsc_catchup
) {
1496 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1497 if (tsc
> tsc_timestamp
) {
1498 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1499 tsc_timestamp
= tsc
;
1503 local_irq_restore(flags
);
1505 if (!vcpu
->pv_time_enabled
)
1509 * Time as measured by the TSC may go backwards when resetting the base
1510 * tsc_timestamp. The reason for this is that the TSC resolution is
1511 * higher than the resolution of the other clock scales. Thus, many
1512 * possible measurments of the TSC correspond to one measurement of any
1513 * other clock, and so a spread of values is possible. This is not a
1514 * problem for the computation of the nanosecond clock; with TSC rates
1515 * around 1GHZ, there can only be a few cycles which correspond to one
1516 * nanosecond value, and any path through this code will inevitably
1517 * take longer than that. However, with the kernel_ns value itself,
1518 * the precision may be much lower, down to HZ granularity. If the
1519 * first sampling of TSC against kernel_ns ends in the low part of the
1520 * range, and the second in the high end of the range, we can get:
1522 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1524 * As the sampling errors potentially range in the thousands of cycles,
1525 * it is possible such a time value has already been observed by the
1526 * guest. To protect against this, we must compute the system time as
1527 * observed by the guest and ensure the new system time is greater.
1530 if (vcpu
->hv_clock
.tsc_timestamp
) {
1531 max_kernel_ns
= vcpu
->last_guest_tsc
-
1532 vcpu
->hv_clock
.tsc_timestamp
;
1533 max_kernel_ns
= pvclock_scale_delta(max_kernel_ns
,
1534 vcpu
->hv_clock
.tsc_to_system_mul
,
1535 vcpu
->hv_clock
.tsc_shift
);
1536 max_kernel_ns
+= vcpu
->last_kernel_ns
;
1539 if (unlikely(vcpu
->hw_tsc_khz
!= this_tsc_khz
)) {
1540 kvm_get_time_scale(NSEC_PER_SEC
/ 1000, this_tsc_khz
,
1541 &vcpu
->hv_clock
.tsc_shift
,
1542 &vcpu
->hv_clock
.tsc_to_system_mul
);
1543 vcpu
->hw_tsc_khz
= this_tsc_khz
;
1546 /* with a master <monotonic time, tsc value> tuple,
1547 * pvclock clock reads always increase at the (scaled) rate
1548 * of guest TSC - no need to deal with sampling errors.
1550 if (!use_master_clock
) {
1551 if (max_kernel_ns
> kernel_ns
)
1552 kernel_ns
= max_kernel_ns
;
1554 /* With all the info we got, fill in the values */
1555 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1556 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1557 vcpu
->last_kernel_ns
= kernel_ns
;
1558 vcpu
->last_guest_tsc
= tsc_timestamp
;
1561 * The interface expects us to write an even number signaling that the
1562 * update is finished. Since the guest won't see the intermediate
1563 * state, we just increase by 2 at the end.
1565 vcpu
->hv_clock
.version
+= 2;
1567 if (unlikely(kvm_read_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1568 &guest_hv_clock
, sizeof(guest_hv_clock
))))
1571 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1572 pvclock_flags
= (guest_hv_clock
.flags
& PVCLOCK_GUEST_STOPPED
);
1574 if (vcpu
->pvclock_set_guest_stopped_request
) {
1575 pvclock_flags
|= PVCLOCK_GUEST_STOPPED
;
1576 vcpu
->pvclock_set_guest_stopped_request
= false;
1579 /* If the host uses TSC clocksource, then it is stable */
1580 if (use_master_clock
)
1581 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
1583 vcpu
->hv_clock
.flags
= pvclock_flags
;
1585 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1587 sizeof(vcpu
->hv_clock
));
1591 static bool msr_mtrr_valid(unsigned msr
)
1594 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
1595 case MSR_MTRRfix64K_00000
:
1596 case MSR_MTRRfix16K_80000
:
1597 case MSR_MTRRfix16K_A0000
:
1598 case MSR_MTRRfix4K_C0000
:
1599 case MSR_MTRRfix4K_C8000
:
1600 case MSR_MTRRfix4K_D0000
:
1601 case MSR_MTRRfix4K_D8000
:
1602 case MSR_MTRRfix4K_E0000
:
1603 case MSR_MTRRfix4K_E8000
:
1604 case MSR_MTRRfix4K_F0000
:
1605 case MSR_MTRRfix4K_F8000
:
1606 case MSR_MTRRdefType
:
1607 case MSR_IA32_CR_PAT
:
1615 static bool valid_pat_type(unsigned t
)
1617 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1620 static bool valid_mtrr_type(unsigned t
)
1622 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
1625 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1629 if (!msr_mtrr_valid(msr
))
1632 if (msr
== MSR_IA32_CR_PAT
) {
1633 for (i
= 0; i
< 8; i
++)
1634 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
1637 } else if (msr
== MSR_MTRRdefType
) {
1640 return valid_mtrr_type(data
& 0xff);
1641 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
1642 for (i
= 0; i
< 8 ; i
++)
1643 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
1648 /* variable MTRRs */
1649 return valid_mtrr_type(data
& 0xff);
1652 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1654 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1656 if (!mtrr_valid(vcpu
, msr
, data
))
1659 if (msr
== MSR_MTRRdefType
) {
1660 vcpu
->arch
.mtrr_state
.def_type
= data
;
1661 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
1662 } else if (msr
== MSR_MTRRfix64K_00000
)
1664 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1665 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
1666 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1667 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
1668 else if (msr
== MSR_IA32_CR_PAT
)
1669 vcpu
->arch
.pat
= data
;
1670 else { /* Variable MTRRs */
1671 int idx
, is_mtrr_mask
;
1674 idx
= (msr
- 0x200) / 2;
1675 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1678 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1681 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1685 kvm_mmu_reset_context(vcpu
);
1689 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1691 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1692 unsigned bank_num
= mcg_cap
& 0xff;
1695 case MSR_IA32_MCG_STATUS
:
1696 vcpu
->arch
.mcg_status
= data
;
1698 case MSR_IA32_MCG_CTL
:
1699 if (!(mcg_cap
& MCG_CTL_P
))
1701 if (data
!= 0 && data
!= ~(u64
)0)
1703 vcpu
->arch
.mcg_ctl
= data
;
1706 if (msr
>= MSR_IA32_MC0_CTL
&&
1707 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1708 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1709 /* only 0 or all 1s can be written to IA32_MCi_CTL
1710 * some Linux kernels though clear bit 10 in bank 4 to
1711 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1712 * this to avoid an uncatched #GP in the guest
1714 if ((offset
& 0x3) == 0 &&
1715 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
1717 vcpu
->arch
.mce_banks
[offset
] = data
;
1725 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
1727 struct kvm
*kvm
= vcpu
->kvm
;
1728 int lm
= is_long_mode(vcpu
);
1729 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
1730 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
1731 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
1732 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
1733 u32 page_num
= data
& ~PAGE_MASK
;
1734 u64 page_addr
= data
& PAGE_MASK
;
1739 if (page_num
>= blob_size
)
1742 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
1747 if (kvm_write_guest(kvm
, page_addr
, page
, PAGE_SIZE
))
1756 static bool kvm_hv_hypercall_enabled(struct kvm
*kvm
)
1758 return kvm
->arch
.hv_hypercall
& HV_X64_MSR_HYPERCALL_ENABLE
;
1761 static bool kvm_hv_msr_partition_wide(u32 msr
)
1765 case HV_X64_MSR_GUEST_OS_ID
:
1766 case HV_X64_MSR_HYPERCALL
:
1774 static int set_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1776 struct kvm
*kvm
= vcpu
->kvm
;
1779 case HV_X64_MSR_GUEST_OS_ID
:
1780 kvm
->arch
.hv_guest_os_id
= data
;
1781 /* setting guest os id to zero disables hypercall page */
1782 if (!kvm
->arch
.hv_guest_os_id
)
1783 kvm
->arch
.hv_hypercall
&= ~HV_X64_MSR_HYPERCALL_ENABLE
;
1785 case HV_X64_MSR_HYPERCALL
: {
1790 /* if guest os id is not set hypercall should remain disabled */
1791 if (!kvm
->arch
.hv_guest_os_id
)
1793 if (!(data
& HV_X64_MSR_HYPERCALL_ENABLE
)) {
1794 kvm
->arch
.hv_hypercall
= data
;
1797 gfn
= data
>> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT
;
1798 addr
= gfn_to_hva(kvm
, gfn
);
1799 if (kvm_is_error_hva(addr
))
1801 kvm_x86_ops
->patch_hypercall(vcpu
, instructions
);
1802 ((unsigned char *)instructions
)[3] = 0xc3; /* ret */
1803 if (__copy_to_user((void __user
*)addr
, instructions
, 4))
1805 kvm
->arch
.hv_hypercall
= data
;
1809 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1810 "data 0x%llx\n", msr
, data
);
1816 static int set_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1819 case HV_X64_MSR_APIC_ASSIST_PAGE
: {
1822 if (!(data
& HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE
)) {
1823 vcpu
->arch
.hv_vapic
= data
;
1826 addr
= gfn_to_hva(vcpu
->kvm
, data
>>
1827 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT
);
1828 if (kvm_is_error_hva(addr
))
1830 if (__clear_user((void __user
*)addr
, PAGE_SIZE
))
1832 vcpu
->arch
.hv_vapic
= data
;
1835 case HV_X64_MSR_EOI
:
1836 return kvm_hv_vapic_msr_write(vcpu
, APIC_EOI
, data
);
1837 case HV_X64_MSR_ICR
:
1838 return kvm_hv_vapic_msr_write(vcpu
, APIC_ICR
, data
);
1839 case HV_X64_MSR_TPR
:
1840 return kvm_hv_vapic_msr_write(vcpu
, APIC_TASKPRI
, data
);
1842 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1843 "data 0x%llx\n", msr
, data
);
1850 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
1852 gpa_t gpa
= data
& ~0x3f;
1854 /* Bits 2:5 are reserved, Should be zero */
1858 vcpu
->arch
.apf
.msr_val
= data
;
1860 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
1861 kvm_clear_async_pf_completion_queue(vcpu
);
1862 kvm_async_pf_hash_reset(vcpu
);
1866 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
,
1870 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
1871 kvm_async_pf_wakeup_all(vcpu
);
1875 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
1877 vcpu
->arch
.pv_time_enabled
= false;
1880 static void accumulate_steal_time(struct kvm_vcpu
*vcpu
)
1884 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1887 delta
= current
->sched_info
.run_delay
- vcpu
->arch
.st
.last_steal
;
1888 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
1889 vcpu
->arch
.st
.accum_steal
= delta
;
1892 static void record_steal_time(struct kvm_vcpu
*vcpu
)
1894 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1897 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1898 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
1901 vcpu
->arch
.st
.steal
.steal
+= vcpu
->arch
.st
.accum_steal
;
1902 vcpu
->arch
.st
.steal
.version
+= 2;
1903 vcpu
->arch
.st
.accum_steal
= 0;
1905 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1906 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
1909 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
1912 u32 msr
= msr_info
->index
;
1913 u64 data
= msr_info
->data
;
1916 case MSR_AMD64_NB_CFG
:
1917 case MSR_IA32_UCODE_REV
:
1918 case MSR_IA32_UCODE_WRITE
:
1919 case MSR_VM_HSAVE_PA
:
1920 case MSR_AMD64_PATCH_LOADER
:
1921 case MSR_AMD64_BU_CFG2
:
1925 return set_efer(vcpu
, data
);
1927 data
&= ~(u64
)0x40; /* ignore flush filter disable */
1928 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
1929 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
1931 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
1936 case MSR_FAM10H_MMIO_CONF_BASE
:
1938 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
1943 case MSR_IA32_DEBUGCTLMSR
:
1945 /* We support the non-activated case already */
1947 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
1948 /* Values other than LBR and BTF are vendor-specific,
1949 thus reserved and should throw a #GP */
1952 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1955 case 0x200 ... 0x2ff:
1956 return set_msr_mtrr(vcpu
, msr
, data
);
1957 case MSR_IA32_APICBASE
:
1958 kvm_set_apic_base(vcpu
, data
);
1960 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1961 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
1962 case MSR_IA32_TSCDEADLINE
:
1963 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
1965 case MSR_IA32_TSC_ADJUST
:
1966 if (guest_cpuid_has_tsc_adjust(vcpu
)) {
1967 if (!msr_info
->host_initiated
) {
1968 u64 adj
= data
- vcpu
->arch
.ia32_tsc_adjust_msr
;
1969 kvm_x86_ops
->adjust_tsc_offset(vcpu
, adj
, true);
1971 vcpu
->arch
.ia32_tsc_adjust_msr
= data
;
1974 case MSR_IA32_MISC_ENABLE
:
1975 vcpu
->arch
.ia32_misc_enable_msr
= data
;
1977 case MSR_KVM_WALL_CLOCK_NEW
:
1978 case MSR_KVM_WALL_CLOCK
:
1979 vcpu
->kvm
->arch
.wall_clock
= data
;
1980 kvm_write_wall_clock(vcpu
->kvm
, data
);
1982 case MSR_KVM_SYSTEM_TIME_NEW
:
1983 case MSR_KVM_SYSTEM_TIME
: {
1985 kvmclock_reset(vcpu
);
1987 vcpu
->arch
.time
= data
;
1988 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1990 /* we verify if the enable bit is set... */
1994 gpa_offset
= data
& ~(PAGE_MASK
| 1);
1996 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
1997 &vcpu
->arch
.pv_time
, data
& ~1ULL,
1998 sizeof(struct pvclock_vcpu_time_info
)))
1999 vcpu
->arch
.pv_time_enabled
= false;
2001 vcpu
->arch
.pv_time_enabled
= true;
2005 case MSR_KVM_ASYNC_PF_EN
:
2006 if (kvm_pv_enable_async_pf(vcpu
, data
))
2009 case MSR_KVM_STEAL_TIME
:
2011 if (unlikely(!sched_info_on()))
2014 if (data
& KVM_STEAL_RESERVED_MASK
)
2017 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2018 data
& KVM_STEAL_VALID_BITS
,
2019 sizeof(struct kvm_steal_time
)))
2022 vcpu
->arch
.st
.msr_val
= data
;
2024 if (!(data
& KVM_MSR_ENABLED
))
2027 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
2030 accumulate_steal_time(vcpu
);
2033 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2036 case MSR_KVM_PV_EOI_EN
:
2037 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
2041 case MSR_IA32_MCG_CTL
:
2042 case MSR_IA32_MCG_STATUS
:
2043 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
2044 return set_msr_mce(vcpu
, msr
, data
);
2046 /* Performance counters are not protected by a CPUID bit,
2047 * so we should check all of them in the generic path for the sake of
2048 * cross vendor migration.
2049 * Writing a zero into the event select MSRs disables them,
2050 * which we perfectly emulate ;-). Any other value should be at least
2051 * reported, some guests depend on them.
2053 case MSR_K7_EVNTSEL0
:
2054 case MSR_K7_EVNTSEL1
:
2055 case MSR_K7_EVNTSEL2
:
2056 case MSR_K7_EVNTSEL3
:
2058 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
2059 "0x%x data 0x%llx\n", msr
, data
);
2061 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2062 * so we ignore writes to make it happy.
2064 case MSR_K7_PERFCTR0
:
2065 case MSR_K7_PERFCTR1
:
2066 case MSR_K7_PERFCTR2
:
2067 case MSR_K7_PERFCTR3
:
2068 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
2069 "0x%x data 0x%llx\n", msr
, data
);
2071 case MSR_P6_PERFCTR0
:
2072 case MSR_P6_PERFCTR1
:
2074 case MSR_P6_EVNTSEL0
:
2075 case MSR_P6_EVNTSEL1
:
2076 if (kvm_pmu_msr(vcpu
, msr
))
2077 return kvm_pmu_set_msr(vcpu
, msr_info
);
2079 if (pr
|| data
!= 0)
2080 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
2081 "0x%x data 0x%llx\n", msr
, data
);
2083 case MSR_K7_CLK_CTL
:
2085 * Ignore all writes to this no longer documented MSR.
2086 * Writes are only relevant for old K7 processors,
2087 * all pre-dating SVM, but a recommended workaround from
2088 * AMD for these chips. It is possible to specify the
2089 * affected processor models on the command line, hence
2090 * the need to ignore the workaround.
2093 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2094 if (kvm_hv_msr_partition_wide(msr
)) {
2096 mutex_lock(&vcpu
->kvm
->lock
);
2097 r
= set_msr_hyperv_pw(vcpu
, msr
, data
);
2098 mutex_unlock(&vcpu
->kvm
->lock
);
2101 return set_msr_hyperv(vcpu
, msr
, data
);
2103 case MSR_IA32_BBL_CR_CTL3
:
2104 /* Drop writes to this legacy MSR -- see rdmsr
2105 * counterpart for further detail.
2107 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n", msr
, data
);
2109 case MSR_AMD64_OSVW_ID_LENGTH
:
2110 if (!guest_cpuid_has_osvw(vcpu
))
2112 vcpu
->arch
.osvw
.length
= data
;
2114 case MSR_AMD64_OSVW_STATUS
:
2115 if (!guest_cpuid_has_osvw(vcpu
))
2117 vcpu
->arch
.osvw
.status
= data
;
2120 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
2121 return xen_hvm_config(vcpu
, data
);
2122 if (kvm_pmu_msr(vcpu
, msr
))
2123 return kvm_pmu_set_msr(vcpu
, msr_info
);
2125 vcpu_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
2129 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
2136 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
2140 * Reads an msr value (of 'msr_index') into 'pdata'.
2141 * Returns 0 on success, non-0 otherwise.
2142 * Assumes vcpu_load() was already called.
2144 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
2146 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
2149 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2151 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
2153 if (!msr_mtrr_valid(msr
))
2156 if (msr
== MSR_MTRRdefType
)
2157 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
2158 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
2159 else if (msr
== MSR_MTRRfix64K_00000
)
2161 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
2162 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
2163 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
2164 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
2165 else if (msr
== MSR_IA32_CR_PAT
)
2166 *pdata
= vcpu
->arch
.pat
;
2167 else { /* Variable MTRRs */
2168 int idx
, is_mtrr_mask
;
2171 idx
= (msr
- 0x200) / 2;
2172 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
2175 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
2178 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
2185 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2188 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2189 unsigned bank_num
= mcg_cap
& 0xff;
2192 case MSR_IA32_P5_MC_ADDR
:
2193 case MSR_IA32_P5_MC_TYPE
:
2196 case MSR_IA32_MCG_CAP
:
2197 data
= vcpu
->arch
.mcg_cap
;
2199 case MSR_IA32_MCG_CTL
:
2200 if (!(mcg_cap
& MCG_CTL_P
))
2202 data
= vcpu
->arch
.mcg_ctl
;
2204 case MSR_IA32_MCG_STATUS
:
2205 data
= vcpu
->arch
.mcg_status
;
2208 if (msr
>= MSR_IA32_MC0_CTL
&&
2209 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
2210 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2211 data
= vcpu
->arch
.mce_banks
[offset
];
2220 static int get_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2223 struct kvm
*kvm
= vcpu
->kvm
;
2226 case HV_X64_MSR_GUEST_OS_ID
:
2227 data
= kvm
->arch
.hv_guest_os_id
;
2229 case HV_X64_MSR_HYPERCALL
:
2230 data
= kvm
->arch
.hv_hypercall
;
2233 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
2241 static int get_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2246 case HV_X64_MSR_VP_INDEX
: {
2249 kvm_for_each_vcpu(r
, v
, vcpu
->kvm
)
2254 case HV_X64_MSR_EOI
:
2255 return kvm_hv_vapic_msr_read(vcpu
, APIC_EOI
, pdata
);
2256 case HV_X64_MSR_ICR
:
2257 return kvm_hv_vapic_msr_read(vcpu
, APIC_ICR
, pdata
);
2258 case HV_X64_MSR_TPR
:
2259 return kvm_hv_vapic_msr_read(vcpu
, APIC_TASKPRI
, pdata
);
2260 case HV_X64_MSR_APIC_ASSIST_PAGE
:
2261 data
= vcpu
->arch
.hv_vapic
;
2264 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
2271 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2276 case MSR_IA32_PLATFORM_ID
:
2277 case MSR_IA32_EBL_CR_POWERON
:
2278 case MSR_IA32_DEBUGCTLMSR
:
2279 case MSR_IA32_LASTBRANCHFROMIP
:
2280 case MSR_IA32_LASTBRANCHTOIP
:
2281 case MSR_IA32_LASTINTFROMIP
:
2282 case MSR_IA32_LASTINTTOIP
:
2285 case MSR_VM_HSAVE_PA
:
2286 case MSR_K7_EVNTSEL0
:
2287 case MSR_K7_PERFCTR0
:
2288 case MSR_K8_INT_PENDING_MSG
:
2289 case MSR_AMD64_NB_CFG
:
2290 case MSR_FAM10H_MMIO_CONF_BASE
:
2291 case MSR_AMD64_BU_CFG2
:
2294 case MSR_P6_PERFCTR0
:
2295 case MSR_P6_PERFCTR1
:
2296 case MSR_P6_EVNTSEL0
:
2297 case MSR_P6_EVNTSEL1
:
2298 if (kvm_pmu_msr(vcpu
, msr
))
2299 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
2302 case MSR_IA32_UCODE_REV
:
2303 data
= 0x100000000ULL
;
2306 data
= 0x500 | KVM_NR_VAR_MTRR
;
2308 case 0x200 ... 0x2ff:
2309 return get_msr_mtrr(vcpu
, msr
, pdata
);
2310 case 0xcd: /* fsb frequency */
2314 * MSR_EBC_FREQUENCY_ID
2315 * Conservative value valid for even the basic CPU models.
2316 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2317 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2318 * and 266MHz for model 3, or 4. Set Core Clock
2319 * Frequency to System Bus Frequency Ratio to 1 (bits
2320 * 31:24) even though these are only valid for CPU
2321 * models > 2, however guests may end up dividing or
2322 * multiplying by zero otherwise.
2324 case MSR_EBC_FREQUENCY_ID
:
2327 case MSR_IA32_APICBASE
:
2328 data
= kvm_get_apic_base(vcpu
);
2330 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2331 return kvm_x2apic_msr_read(vcpu
, msr
, pdata
);
2333 case MSR_IA32_TSCDEADLINE
:
2334 data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
2336 case MSR_IA32_TSC_ADJUST
:
2337 data
= (u64
)vcpu
->arch
.ia32_tsc_adjust_msr
;
2339 case MSR_IA32_MISC_ENABLE
:
2340 data
= vcpu
->arch
.ia32_misc_enable_msr
;
2342 case MSR_IA32_PERF_STATUS
:
2343 /* TSC increment by tick */
2345 /* CPU multiplier */
2346 data
|= (((uint64_t)4ULL) << 40);
2349 data
= vcpu
->arch
.efer
;
2351 case MSR_KVM_WALL_CLOCK
:
2352 case MSR_KVM_WALL_CLOCK_NEW
:
2353 data
= vcpu
->kvm
->arch
.wall_clock
;
2355 case MSR_KVM_SYSTEM_TIME
:
2356 case MSR_KVM_SYSTEM_TIME_NEW
:
2357 data
= vcpu
->arch
.time
;
2359 case MSR_KVM_ASYNC_PF_EN
:
2360 data
= vcpu
->arch
.apf
.msr_val
;
2362 case MSR_KVM_STEAL_TIME
:
2363 data
= vcpu
->arch
.st
.msr_val
;
2365 case MSR_KVM_PV_EOI_EN
:
2366 data
= vcpu
->arch
.pv_eoi
.msr_val
;
2368 case MSR_IA32_P5_MC_ADDR
:
2369 case MSR_IA32_P5_MC_TYPE
:
2370 case MSR_IA32_MCG_CAP
:
2371 case MSR_IA32_MCG_CTL
:
2372 case MSR_IA32_MCG_STATUS
:
2373 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
2374 return get_msr_mce(vcpu
, msr
, pdata
);
2375 case MSR_K7_CLK_CTL
:
2377 * Provide expected ramp-up count for K7. All other
2378 * are set to zero, indicating minimum divisors for
2381 * This prevents guest kernels on AMD host with CPU
2382 * type 6, model 8 and higher from exploding due to
2383 * the rdmsr failing.
2387 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2388 if (kvm_hv_msr_partition_wide(msr
)) {
2390 mutex_lock(&vcpu
->kvm
->lock
);
2391 r
= get_msr_hyperv_pw(vcpu
, msr
, pdata
);
2392 mutex_unlock(&vcpu
->kvm
->lock
);
2395 return get_msr_hyperv(vcpu
, msr
, pdata
);
2397 case MSR_IA32_BBL_CR_CTL3
:
2398 /* This legacy MSR exists but isn't fully documented in current
2399 * silicon. It is however accessed by winxp in very narrow
2400 * scenarios where it sets bit #19, itself documented as
2401 * a "reserved" bit. Best effort attempt to source coherent
2402 * read data here should the balance of the register be
2403 * interpreted by the guest:
2405 * L2 cache control register 3: 64GB range, 256KB size,
2406 * enabled, latency 0x1, configured
2410 case MSR_AMD64_OSVW_ID_LENGTH
:
2411 if (!guest_cpuid_has_osvw(vcpu
))
2413 data
= vcpu
->arch
.osvw
.length
;
2415 case MSR_AMD64_OSVW_STATUS
:
2416 if (!guest_cpuid_has_osvw(vcpu
))
2418 data
= vcpu
->arch
.osvw
.status
;
2421 if (kvm_pmu_msr(vcpu
, msr
))
2422 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
2424 vcpu_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
2427 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr
);
2435 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2438 * Read or write a bunch of msrs. All parameters are kernel addresses.
2440 * @return number of msrs set successfully.
2442 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2443 struct kvm_msr_entry
*entries
,
2444 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2445 unsigned index
, u64
*data
))
2449 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2450 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2451 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2453 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2459 * Read or write a bunch of msrs. Parameters are user addresses.
2461 * @return number of msrs set successfully.
2463 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2464 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2465 unsigned index
, u64
*data
),
2468 struct kvm_msrs msrs
;
2469 struct kvm_msr_entry
*entries
;
2474 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2478 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2481 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2482 entries
= memdup_user(user_msrs
->entries
, size
);
2483 if (IS_ERR(entries
)) {
2484 r
= PTR_ERR(entries
);
2488 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2493 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2504 int kvm_dev_ioctl_check_extension(long ext
)
2509 case KVM_CAP_IRQCHIP
:
2511 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2512 case KVM_CAP_SET_TSS_ADDR
:
2513 case KVM_CAP_EXT_CPUID
:
2514 case KVM_CAP_CLOCKSOURCE
:
2516 case KVM_CAP_NOP_IO_DELAY
:
2517 case KVM_CAP_MP_STATE
:
2518 case KVM_CAP_SYNC_MMU
:
2519 case KVM_CAP_USER_NMI
:
2520 case KVM_CAP_REINJECT_CONTROL
:
2521 case KVM_CAP_IRQ_INJECT_STATUS
:
2523 case KVM_CAP_IOEVENTFD
:
2525 case KVM_CAP_PIT_STATE2
:
2526 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2527 case KVM_CAP_XEN_HVM
:
2528 case KVM_CAP_ADJUST_CLOCK
:
2529 case KVM_CAP_VCPU_EVENTS
:
2530 case KVM_CAP_HYPERV
:
2531 case KVM_CAP_HYPERV_VAPIC
:
2532 case KVM_CAP_HYPERV_SPIN
:
2533 case KVM_CAP_PCI_SEGMENT
:
2534 case KVM_CAP_DEBUGREGS
:
2535 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2537 case KVM_CAP_ASYNC_PF
:
2538 case KVM_CAP_GET_TSC_KHZ
:
2539 case KVM_CAP_KVMCLOCK_CTRL
:
2540 case KVM_CAP_READONLY_MEM
:
2541 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2542 case KVM_CAP_ASSIGN_DEV_IRQ
:
2543 case KVM_CAP_PCI_2_3
:
2547 case KVM_CAP_COALESCED_MMIO
:
2548 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
2551 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2553 case KVM_CAP_NR_VCPUS
:
2554 r
= KVM_SOFT_MAX_VCPUS
;
2556 case KVM_CAP_MAX_VCPUS
:
2559 case KVM_CAP_NR_MEMSLOTS
:
2560 r
= KVM_USER_MEM_SLOTS
;
2562 case KVM_CAP_PV_MMU
: /* obsolete */
2565 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2567 r
= iommu_present(&pci_bus_type
);
2571 r
= KVM_MAX_MCE_BANKS
;
2576 case KVM_CAP_TSC_CONTROL
:
2577 r
= kvm_has_tsc_control
;
2579 case KVM_CAP_TSC_DEADLINE_TIMER
:
2580 r
= boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER
);
2590 long kvm_arch_dev_ioctl(struct file
*filp
,
2591 unsigned int ioctl
, unsigned long arg
)
2593 void __user
*argp
= (void __user
*)arg
;
2597 case KVM_GET_MSR_INDEX_LIST
: {
2598 struct kvm_msr_list __user
*user_msr_list
= argp
;
2599 struct kvm_msr_list msr_list
;
2603 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2606 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2607 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2610 if (n
< msr_list
.nmsrs
)
2613 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2614 num_msrs_to_save
* sizeof(u32
)))
2616 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2618 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2623 case KVM_GET_SUPPORTED_CPUID
: {
2624 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2625 struct kvm_cpuid2 cpuid
;
2628 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2630 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
2631 cpuid_arg
->entries
);
2636 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2641 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2644 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
2646 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
2658 static void wbinvd_ipi(void *garbage
)
2663 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2665 return vcpu
->kvm
->arch
.iommu_domain
&&
2666 !(vcpu
->kvm
->arch
.iommu_flags
& KVM_IOMMU_CACHE_COHERENCY
);
2669 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2671 /* Address WBINVD may be executed by guest */
2672 if (need_emulate_wbinvd(vcpu
)) {
2673 if (kvm_x86_ops
->has_wbinvd_exit())
2674 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2675 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2676 smp_call_function_single(vcpu
->cpu
,
2677 wbinvd_ipi
, NULL
, 1);
2680 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2682 /* Apply any externally detected TSC adjustments (due to suspend) */
2683 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2684 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2685 vcpu
->arch
.tsc_offset_adjustment
= 0;
2686 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
2689 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2690 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2691 native_read_tsc() - vcpu
->arch
.last_host_tsc
;
2693 mark_tsc_unstable("KVM discovered backwards TSC");
2694 if (check_tsc_unstable()) {
2695 u64 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
,
2696 vcpu
->arch
.last_guest_tsc
);
2697 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
2698 vcpu
->arch
.tsc_catchup
= 1;
2701 * On a host with synchronized TSC, there is no need to update
2702 * kvmclock on vcpu->cpu migration
2704 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
2705 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2706 if (vcpu
->cpu
!= cpu
)
2707 kvm_migrate_timers(vcpu
);
2711 accumulate_steal_time(vcpu
);
2712 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2715 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2717 kvm_x86_ops
->vcpu_put(vcpu
);
2718 kvm_put_guest_fpu(vcpu
);
2719 vcpu
->arch
.last_host_tsc
= native_read_tsc();
2722 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
2723 struct kvm_lapic_state
*s
)
2725 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
2726 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
2731 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
2732 struct kvm_lapic_state
*s
)
2734 kvm_apic_post_state_restore(vcpu
, s
);
2735 update_cr8_intercept(vcpu
);
2740 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2741 struct kvm_interrupt
*irq
)
2743 if (irq
->irq
>= KVM_NR_INTERRUPTS
)
2745 if (irqchip_in_kernel(vcpu
->kvm
))
2748 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
2749 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2754 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
2756 kvm_inject_nmi(vcpu
);
2761 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
2762 struct kvm_tpr_access_ctl
*tac
)
2766 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
2770 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
2774 unsigned bank_num
= mcg_cap
& 0xff, bank
;
2777 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
2779 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
2782 vcpu
->arch
.mcg_cap
= mcg_cap
;
2783 /* Init IA32_MCG_CTL to all 1s */
2784 if (mcg_cap
& MCG_CTL_P
)
2785 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
2786 /* Init IA32_MCi_CTL to all 1s */
2787 for (bank
= 0; bank
< bank_num
; bank
++)
2788 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
2793 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
2794 struct kvm_x86_mce
*mce
)
2796 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2797 unsigned bank_num
= mcg_cap
& 0xff;
2798 u64
*banks
= vcpu
->arch
.mce_banks
;
2800 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
2803 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2804 * reporting is disabled
2806 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
2807 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
2809 banks
+= 4 * mce
->bank
;
2811 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2812 * reporting is disabled for the bank
2814 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
2816 if (mce
->status
& MCI_STATUS_UC
) {
2817 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
2818 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
2819 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2822 if (banks
[1] & MCI_STATUS_VAL
)
2823 mce
->status
|= MCI_STATUS_OVER
;
2824 banks
[2] = mce
->addr
;
2825 banks
[3] = mce
->misc
;
2826 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
2827 banks
[1] = mce
->status
;
2828 kvm_queue_exception(vcpu
, MC_VECTOR
);
2829 } else if (!(banks
[1] & MCI_STATUS_VAL
)
2830 || !(banks
[1] & MCI_STATUS_UC
)) {
2831 if (banks
[1] & MCI_STATUS_VAL
)
2832 mce
->status
|= MCI_STATUS_OVER
;
2833 banks
[2] = mce
->addr
;
2834 banks
[3] = mce
->misc
;
2835 banks
[1] = mce
->status
;
2837 banks
[1] |= MCI_STATUS_OVER
;
2841 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
2842 struct kvm_vcpu_events
*events
)
2845 events
->exception
.injected
=
2846 vcpu
->arch
.exception
.pending
&&
2847 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
2848 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
2849 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
2850 events
->exception
.pad
= 0;
2851 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
2853 events
->interrupt
.injected
=
2854 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
2855 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
2856 events
->interrupt
.soft
= 0;
2857 events
->interrupt
.shadow
=
2858 kvm_x86_ops
->get_interrupt_shadow(vcpu
,
2859 KVM_X86_SHADOW_INT_MOV_SS
| KVM_X86_SHADOW_INT_STI
);
2861 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
2862 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
2863 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
2864 events
->nmi
.pad
= 0;
2866 events
->sipi_vector
= 0; /* never valid when reporting to user space */
2868 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
2869 | KVM_VCPUEVENT_VALID_SHADOW
);
2870 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
2873 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
2874 struct kvm_vcpu_events
*events
)
2876 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2877 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2878 | KVM_VCPUEVENT_VALID_SHADOW
))
2882 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
2883 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
2884 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
2885 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
2887 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
2888 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
2889 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
2890 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
2891 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
2892 events
->interrupt
.shadow
);
2894 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
2895 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
2896 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
2897 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
2899 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
&&
2900 kvm_vcpu_has_lapic(vcpu
))
2901 vcpu
->arch
.apic
->sipi_vector
= events
->sipi_vector
;
2903 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2908 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
2909 struct kvm_debugregs
*dbgregs
)
2911 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
2912 dbgregs
->dr6
= vcpu
->arch
.dr6
;
2913 dbgregs
->dr7
= vcpu
->arch
.dr7
;
2915 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
2918 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
2919 struct kvm_debugregs
*dbgregs
)
2924 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
2925 vcpu
->arch
.dr6
= dbgregs
->dr6
;
2926 vcpu
->arch
.dr7
= dbgregs
->dr7
;
2931 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
2932 struct kvm_xsave
*guest_xsave
)
2935 memcpy(guest_xsave
->region
,
2936 &vcpu
->arch
.guest_fpu
.state
->xsave
,
2939 memcpy(guest_xsave
->region
,
2940 &vcpu
->arch
.guest_fpu
.state
->fxsave
,
2941 sizeof(struct i387_fxsave_struct
));
2942 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
2947 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
2948 struct kvm_xsave
*guest_xsave
)
2951 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
2954 memcpy(&vcpu
->arch
.guest_fpu
.state
->xsave
,
2955 guest_xsave
->region
, xstate_size
);
2957 if (xstate_bv
& ~XSTATE_FPSSE
)
2959 memcpy(&vcpu
->arch
.guest_fpu
.state
->fxsave
,
2960 guest_xsave
->region
, sizeof(struct i387_fxsave_struct
));
2965 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
2966 struct kvm_xcrs
*guest_xcrs
)
2968 if (!cpu_has_xsave
) {
2969 guest_xcrs
->nr_xcrs
= 0;
2973 guest_xcrs
->nr_xcrs
= 1;
2974 guest_xcrs
->flags
= 0;
2975 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
2976 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
2979 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
2980 struct kvm_xcrs
*guest_xcrs
)
2987 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
2990 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
2991 /* Only support XCR0 currently */
2992 if (guest_xcrs
->xcrs
[0].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
2993 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
2994 guest_xcrs
->xcrs
[0].value
);
3003 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3004 * stopped by the hypervisor. This function will be called from the host only.
3005 * EINVAL is returned when the host attempts to set the flag for a guest that
3006 * does not support pv clocks.
3008 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
3010 if (!vcpu
->arch
.pv_time_enabled
)
3012 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
3013 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
3017 long kvm_arch_vcpu_ioctl(struct file
*filp
,
3018 unsigned int ioctl
, unsigned long arg
)
3020 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3021 void __user
*argp
= (void __user
*)arg
;
3024 struct kvm_lapic_state
*lapic
;
3025 struct kvm_xsave
*xsave
;
3026 struct kvm_xcrs
*xcrs
;
3032 case KVM_GET_LAPIC
: {
3034 if (!vcpu
->arch
.apic
)
3036 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
3041 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
3045 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
3050 case KVM_SET_LAPIC
: {
3052 if (!vcpu
->arch
.apic
)
3054 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
3055 if (IS_ERR(u
.lapic
))
3056 return PTR_ERR(u
.lapic
);
3058 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
3061 case KVM_INTERRUPT
: {
3062 struct kvm_interrupt irq
;
3065 if (copy_from_user(&irq
, argp
, sizeof irq
))
3067 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
3071 r
= kvm_vcpu_ioctl_nmi(vcpu
);
3074 case KVM_SET_CPUID
: {
3075 struct kvm_cpuid __user
*cpuid_arg
= argp
;
3076 struct kvm_cpuid cpuid
;
3079 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3081 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3084 case KVM_SET_CPUID2
: {
3085 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3086 struct kvm_cpuid2 cpuid
;
3089 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3091 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
3092 cpuid_arg
->entries
);
3095 case KVM_GET_CPUID2
: {
3096 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3097 struct kvm_cpuid2 cpuid
;
3100 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3102 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
3103 cpuid_arg
->entries
);
3107 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
3113 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
3116 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
3118 case KVM_TPR_ACCESS_REPORTING
: {
3119 struct kvm_tpr_access_ctl tac
;
3122 if (copy_from_user(&tac
, argp
, sizeof tac
))
3124 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
3128 if (copy_to_user(argp
, &tac
, sizeof tac
))
3133 case KVM_SET_VAPIC_ADDR
: {
3134 struct kvm_vapic_addr va
;
3137 if (!irqchip_in_kernel(vcpu
->kvm
))
3140 if (copy_from_user(&va
, argp
, sizeof va
))
3143 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
3146 case KVM_X86_SETUP_MCE
: {
3150 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
3152 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
3155 case KVM_X86_SET_MCE
: {
3156 struct kvm_x86_mce mce
;
3159 if (copy_from_user(&mce
, argp
, sizeof mce
))
3161 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
3164 case KVM_GET_VCPU_EVENTS
: {
3165 struct kvm_vcpu_events events
;
3167 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
3170 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
3175 case KVM_SET_VCPU_EVENTS
: {
3176 struct kvm_vcpu_events events
;
3179 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
3182 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
3185 case KVM_GET_DEBUGREGS
: {
3186 struct kvm_debugregs dbgregs
;
3188 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
3191 if (copy_to_user(argp
, &dbgregs
,
3192 sizeof(struct kvm_debugregs
)))
3197 case KVM_SET_DEBUGREGS
: {
3198 struct kvm_debugregs dbgregs
;
3201 if (copy_from_user(&dbgregs
, argp
,
3202 sizeof(struct kvm_debugregs
)))
3205 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3208 case KVM_GET_XSAVE
: {
3209 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3214 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3217 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3222 case KVM_SET_XSAVE
: {
3223 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
3224 if (IS_ERR(u
.xsave
))
3225 return PTR_ERR(u
.xsave
);
3227 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3230 case KVM_GET_XCRS
: {
3231 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3236 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3239 if (copy_to_user(argp
, u
.xcrs
,
3240 sizeof(struct kvm_xcrs
)))
3245 case KVM_SET_XCRS
: {
3246 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
3248 return PTR_ERR(u
.xcrs
);
3250 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3253 case KVM_SET_TSC_KHZ
: {
3257 user_tsc_khz
= (u32
)arg
;
3259 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3262 if (user_tsc_khz
== 0)
3263 user_tsc_khz
= tsc_khz
;
3265 kvm_set_tsc_khz(vcpu
, user_tsc_khz
);
3270 case KVM_GET_TSC_KHZ
: {
3271 r
= vcpu
->arch
.virtual_tsc_khz
;
3274 case KVM_KVMCLOCK_CTRL
: {
3275 r
= kvm_set_guest_paused(vcpu
);
3286 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
3288 return VM_FAULT_SIGBUS
;
3291 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3295 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3297 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3301 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3304 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3308 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3309 u32 kvm_nr_mmu_pages
)
3311 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3314 mutex_lock(&kvm
->slots_lock
);
3316 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3317 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3319 mutex_unlock(&kvm
->slots_lock
);
3323 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3325 return kvm
->arch
.n_max_mmu_pages
;
3328 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3333 switch (chip
->chip_id
) {
3334 case KVM_IRQCHIP_PIC_MASTER
:
3335 memcpy(&chip
->chip
.pic
,
3336 &pic_irqchip(kvm
)->pics
[0],
3337 sizeof(struct kvm_pic_state
));
3339 case KVM_IRQCHIP_PIC_SLAVE
:
3340 memcpy(&chip
->chip
.pic
,
3341 &pic_irqchip(kvm
)->pics
[1],
3342 sizeof(struct kvm_pic_state
));
3344 case KVM_IRQCHIP_IOAPIC
:
3345 r
= kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3354 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3359 switch (chip
->chip_id
) {
3360 case KVM_IRQCHIP_PIC_MASTER
:
3361 spin_lock(&pic_irqchip(kvm
)->lock
);
3362 memcpy(&pic_irqchip(kvm
)->pics
[0],
3364 sizeof(struct kvm_pic_state
));
3365 spin_unlock(&pic_irqchip(kvm
)->lock
);
3367 case KVM_IRQCHIP_PIC_SLAVE
:
3368 spin_lock(&pic_irqchip(kvm
)->lock
);
3369 memcpy(&pic_irqchip(kvm
)->pics
[1],
3371 sizeof(struct kvm_pic_state
));
3372 spin_unlock(&pic_irqchip(kvm
)->lock
);
3374 case KVM_IRQCHIP_IOAPIC
:
3375 r
= kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3381 kvm_pic_update_irq(pic_irqchip(kvm
));
3385 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3389 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3390 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
3391 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3395 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3399 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3400 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
3401 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
, 0);
3402 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3406 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3410 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3411 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3412 sizeof(ps
->channels
));
3413 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3414 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3415 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3419 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3421 int r
= 0, start
= 0;
3422 u32 prev_legacy
, cur_legacy
;
3423 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3424 prev_legacy
= kvm
->arch
.vpit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3425 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3426 if (!prev_legacy
&& cur_legacy
)
3428 memcpy(&kvm
->arch
.vpit
->pit_state
.channels
, &ps
->channels
,
3429 sizeof(kvm
->arch
.vpit
->pit_state
.channels
));
3430 kvm
->arch
.vpit
->pit_state
.flags
= ps
->flags
;
3431 kvm_pit_load_count(kvm
, 0, kvm
->arch
.vpit
->pit_state
.channels
[0].count
, start
);
3432 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3436 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3437 struct kvm_reinject_control
*control
)
3439 if (!kvm
->arch
.vpit
)
3441 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3442 kvm
->arch
.vpit
->pit_state
.reinject
= control
->pit_reinject
;
3443 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3448 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3449 * @kvm: kvm instance
3450 * @log: slot id and address to which we copy the log
3452 * We need to keep it in mind that VCPU threads can write to the bitmap
3453 * concurrently. So, to avoid losing data, we keep the following order for
3456 * 1. Take a snapshot of the bit and clear it if needed.
3457 * 2. Write protect the corresponding page.
3458 * 3. Flush TLB's if needed.
3459 * 4. Copy the snapshot to the userspace.
3461 * Between 2 and 3, the guest may write to the page using the remaining TLB
3462 * entry. This is not a problem because the page will be reported dirty at
3463 * step 4 using the snapshot taken before and step 3 ensures that successive
3464 * writes will be logged for the next call.
3466 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
3469 struct kvm_memory_slot
*memslot
;
3471 unsigned long *dirty_bitmap
;
3472 unsigned long *dirty_bitmap_buffer
;
3473 bool is_dirty
= false;
3475 mutex_lock(&kvm
->slots_lock
);
3478 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
3481 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
3483 dirty_bitmap
= memslot
->dirty_bitmap
;
3488 n
= kvm_dirty_bitmap_bytes(memslot
);
3490 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
3491 memset(dirty_bitmap_buffer
, 0, n
);
3493 spin_lock(&kvm
->mmu_lock
);
3495 for (i
= 0; i
< n
/ sizeof(long); i
++) {
3499 if (!dirty_bitmap
[i
])
3504 mask
= xchg(&dirty_bitmap
[i
], 0);
3505 dirty_bitmap_buffer
[i
] = mask
;
3507 offset
= i
* BITS_PER_LONG
;
3508 kvm_mmu_write_protect_pt_masked(kvm
, memslot
, offset
, mask
);
3511 kvm_flush_remote_tlbs(kvm
);
3513 spin_unlock(&kvm
->mmu_lock
);
3516 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
3521 mutex_unlock(&kvm
->slots_lock
);
3525 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
3528 if (!irqchip_in_kernel(kvm
))
3531 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
3532 irq_event
->irq
, irq_event
->level
,
3537 long kvm_arch_vm_ioctl(struct file
*filp
,
3538 unsigned int ioctl
, unsigned long arg
)
3540 struct kvm
*kvm
= filp
->private_data
;
3541 void __user
*argp
= (void __user
*)arg
;
3544 * This union makes it completely explicit to gcc-3.x
3545 * that these two variables' stack usage should be
3546 * combined, not added together.
3549 struct kvm_pit_state ps
;
3550 struct kvm_pit_state2 ps2
;
3551 struct kvm_pit_config pit_config
;
3555 case KVM_SET_TSS_ADDR
:
3556 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
3558 case KVM_SET_IDENTITY_MAP_ADDR
: {
3562 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
3564 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
3567 case KVM_SET_NR_MMU_PAGES
:
3568 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
3570 case KVM_GET_NR_MMU_PAGES
:
3571 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
3573 case KVM_CREATE_IRQCHIP
: {
3574 struct kvm_pic
*vpic
;
3576 mutex_lock(&kvm
->lock
);
3579 goto create_irqchip_unlock
;
3581 if (atomic_read(&kvm
->online_vcpus
))
3582 goto create_irqchip_unlock
;
3584 vpic
= kvm_create_pic(kvm
);
3586 r
= kvm_ioapic_init(kvm
);
3588 mutex_lock(&kvm
->slots_lock
);
3589 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3591 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3593 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3595 mutex_unlock(&kvm
->slots_lock
);
3597 goto create_irqchip_unlock
;
3600 goto create_irqchip_unlock
;
3602 kvm
->arch
.vpic
= vpic
;
3604 r
= kvm_setup_default_irq_routing(kvm
);
3606 mutex_lock(&kvm
->slots_lock
);
3607 mutex_lock(&kvm
->irq_lock
);
3608 kvm_ioapic_destroy(kvm
);
3609 kvm_destroy_pic(kvm
);
3610 mutex_unlock(&kvm
->irq_lock
);
3611 mutex_unlock(&kvm
->slots_lock
);
3613 create_irqchip_unlock
:
3614 mutex_unlock(&kvm
->lock
);
3617 case KVM_CREATE_PIT
:
3618 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
3620 case KVM_CREATE_PIT2
:
3622 if (copy_from_user(&u
.pit_config
, argp
,
3623 sizeof(struct kvm_pit_config
)))
3626 mutex_lock(&kvm
->slots_lock
);
3629 goto create_pit_unlock
;
3631 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
3635 mutex_unlock(&kvm
->slots_lock
);
3637 case KVM_GET_IRQCHIP
: {
3638 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3639 struct kvm_irqchip
*chip
;
3641 chip
= memdup_user(argp
, sizeof(*chip
));
3648 if (!irqchip_in_kernel(kvm
))
3649 goto get_irqchip_out
;
3650 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
3652 goto get_irqchip_out
;
3654 if (copy_to_user(argp
, chip
, sizeof *chip
))
3655 goto get_irqchip_out
;
3661 case KVM_SET_IRQCHIP
: {
3662 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3663 struct kvm_irqchip
*chip
;
3665 chip
= memdup_user(argp
, sizeof(*chip
));
3672 if (!irqchip_in_kernel(kvm
))
3673 goto set_irqchip_out
;
3674 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
3676 goto set_irqchip_out
;
3684 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
3687 if (!kvm
->arch
.vpit
)
3689 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
3693 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
3700 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
3703 if (!kvm
->arch
.vpit
)
3705 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
3708 case KVM_GET_PIT2
: {
3710 if (!kvm
->arch
.vpit
)
3712 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
3716 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
3721 case KVM_SET_PIT2
: {
3723 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
3726 if (!kvm
->arch
.vpit
)
3728 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
3731 case KVM_REINJECT_CONTROL
: {
3732 struct kvm_reinject_control control
;
3734 if (copy_from_user(&control
, argp
, sizeof(control
)))
3736 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
3739 case KVM_XEN_HVM_CONFIG
: {
3741 if (copy_from_user(&kvm
->arch
.xen_hvm_config
, argp
,
3742 sizeof(struct kvm_xen_hvm_config
)))
3745 if (kvm
->arch
.xen_hvm_config
.flags
)
3750 case KVM_SET_CLOCK
: {
3751 struct kvm_clock_data user_ns
;
3756 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
3764 local_irq_disable();
3765 now_ns
= get_kernel_ns();
3766 delta
= user_ns
.clock
- now_ns
;
3768 kvm
->arch
.kvmclock_offset
= delta
;
3771 case KVM_GET_CLOCK
: {
3772 struct kvm_clock_data user_ns
;
3775 local_irq_disable();
3776 now_ns
= get_kernel_ns();
3777 user_ns
.clock
= kvm
->arch
.kvmclock_offset
+ now_ns
;
3780 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
3783 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
3796 static void kvm_init_msr_list(void)
3801 /* skip the first msrs in the list. KVM-specific */
3802 for (i
= j
= KVM_SAVE_MSRS_BEGIN
; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
3803 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
3806 msrs_to_save
[j
] = msrs_to_save
[i
];
3809 num_msrs_to_save
= j
;
3812 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
3820 if (!(vcpu
->arch
.apic
&&
3821 !kvm_iodevice_write(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3822 && kvm_io_bus_write(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3833 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
3840 if (!(vcpu
->arch
.apic
&&
3841 !kvm_iodevice_read(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3842 && kvm_io_bus_read(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3844 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, *(u64
*)v
);
3854 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3855 struct kvm_segment
*var
, int seg
)
3857 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3860 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3861 struct kvm_segment
*var
, int seg
)
3863 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3866 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
)
3869 struct x86_exception exception
;
3871 BUG_ON(!mmu_is_nested(vcpu
));
3873 /* NPT walks are always user-walks */
3874 access
|= PFERR_USER_MASK
;
3875 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, &exception
);
3880 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
3881 struct x86_exception
*exception
)
3883 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3884 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3887 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
3888 struct x86_exception
*exception
)
3890 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3891 access
|= PFERR_FETCH_MASK
;
3892 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3895 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
3896 struct x86_exception
*exception
)
3898 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3899 access
|= PFERR_WRITE_MASK
;
3900 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3903 /* uses this to access any guest's mapped memory without checking CPL */
3904 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
3905 struct x86_exception
*exception
)
3907 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
3910 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
3911 struct kvm_vcpu
*vcpu
, u32 access
,
3912 struct x86_exception
*exception
)
3915 int r
= X86EMUL_CONTINUE
;
3918 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
3920 unsigned offset
= addr
& (PAGE_SIZE
-1);
3921 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3924 if (gpa
== UNMAPPED_GVA
)
3925 return X86EMUL_PROPAGATE_FAULT
;
3926 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
3928 r
= X86EMUL_IO_NEEDED
;
3940 /* used for instruction fetching */
3941 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3942 gva_t addr
, void *val
, unsigned int bytes
,
3943 struct x86_exception
*exception
)
3945 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3946 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3948 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
,
3949 access
| PFERR_FETCH_MASK
,
3953 int kvm_read_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3954 gva_t addr
, void *val
, unsigned int bytes
,
3955 struct x86_exception
*exception
)
3957 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3958 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3960 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
3963 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
3965 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3966 gva_t addr
, void *val
, unsigned int bytes
,
3967 struct x86_exception
*exception
)
3969 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3970 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, 0, exception
);
3973 int kvm_write_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3974 gva_t addr
, void *val
,
3976 struct x86_exception
*exception
)
3978 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3980 int r
= X86EMUL_CONTINUE
;
3983 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
3986 unsigned offset
= addr
& (PAGE_SIZE
-1);
3987 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3990 if (gpa
== UNMAPPED_GVA
)
3991 return X86EMUL_PROPAGATE_FAULT
;
3992 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
3994 r
= X86EMUL_IO_NEEDED
;
4005 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
4007 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4008 gpa_t
*gpa
, struct x86_exception
*exception
,
4011 u32 access
= ((kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
4012 | (write
? PFERR_WRITE_MASK
: 0);
4014 if (vcpu_match_mmio_gva(vcpu
, gva
)
4015 && !permission_fault(vcpu
->arch
.walk_mmu
, vcpu
->arch
.access
, access
)) {
4016 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
4017 (gva
& (PAGE_SIZE
- 1));
4018 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
4022 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4024 if (*gpa
== UNMAPPED_GVA
)
4027 /* For APIC access vmexit */
4028 if ((*gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4031 if (vcpu_match_mmio_gpa(vcpu
, *gpa
)) {
4032 trace_vcpu_match_mmio(gva
, *gpa
, write
, true);
4039 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4040 const void *val
, int bytes
)
4044 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
4047 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
4051 struct read_write_emulator_ops
{
4052 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
4054 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4055 void *val
, int bytes
);
4056 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4057 int bytes
, void *val
);
4058 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4059 void *val
, int bytes
);
4063 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
4065 if (vcpu
->mmio_read_completed
) {
4066 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
4067 vcpu
->mmio_fragments
[0].gpa
, *(u64
*)val
);
4068 vcpu
->mmio_read_completed
= 0;
4075 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4076 void *val
, int bytes
)
4078 return !kvm_read_guest(vcpu
->kvm
, gpa
, val
, bytes
);
4081 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4082 void *val
, int bytes
)
4084 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
4087 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
4089 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
4090 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
4093 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4094 void *val
, int bytes
)
4096 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
4097 return X86EMUL_IO_NEEDED
;
4100 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4101 void *val
, int bytes
)
4103 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
4105 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
4106 return X86EMUL_CONTINUE
;
4109 static const struct read_write_emulator_ops read_emultor
= {
4110 .read_write_prepare
= read_prepare
,
4111 .read_write_emulate
= read_emulate
,
4112 .read_write_mmio
= vcpu_mmio_read
,
4113 .read_write_exit_mmio
= read_exit_mmio
,
4116 static const struct read_write_emulator_ops write_emultor
= {
4117 .read_write_emulate
= write_emulate
,
4118 .read_write_mmio
= write_mmio
,
4119 .read_write_exit_mmio
= write_exit_mmio
,
4123 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
4125 struct x86_exception
*exception
,
4126 struct kvm_vcpu
*vcpu
,
4127 const struct read_write_emulator_ops
*ops
)
4131 bool write
= ops
->write
;
4132 struct kvm_mmio_fragment
*frag
;
4134 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
4137 return X86EMUL_PROPAGATE_FAULT
;
4139 /* For APIC access vmexit */
4143 if (ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
4144 return X86EMUL_CONTINUE
;
4148 * Is this MMIO handled locally?
4150 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
4151 if (handled
== bytes
)
4152 return X86EMUL_CONTINUE
;
4158 WARN_ON(vcpu
->mmio_nr_fragments
>= KVM_MAX_MMIO_FRAGMENTS
);
4159 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
4163 return X86EMUL_CONTINUE
;
4166 int emulator_read_write(struct x86_emulate_ctxt
*ctxt
, unsigned long addr
,
4167 void *val
, unsigned int bytes
,
4168 struct x86_exception
*exception
,
4169 const struct read_write_emulator_ops
*ops
)
4171 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4175 if (ops
->read_write_prepare
&&
4176 ops
->read_write_prepare(vcpu
, val
, bytes
))
4177 return X86EMUL_CONTINUE
;
4179 vcpu
->mmio_nr_fragments
= 0;
4181 /* Crossing a page boundary? */
4182 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
4185 now
= -addr
& ~PAGE_MASK
;
4186 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
4189 if (rc
!= X86EMUL_CONTINUE
)
4196 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
4198 if (rc
!= X86EMUL_CONTINUE
)
4201 if (!vcpu
->mmio_nr_fragments
)
4204 gpa
= vcpu
->mmio_fragments
[0].gpa
;
4206 vcpu
->mmio_needed
= 1;
4207 vcpu
->mmio_cur_fragment
= 0;
4209 vcpu
->run
->mmio
.len
= min(8u, vcpu
->mmio_fragments
[0].len
);
4210 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
4211 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
4212 vcpu
->run
->mmio
.phys_addr
= gpa
;
4214 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
4217 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
4221 struct x86_exception
*exception
)
4223 return emulator_read_write(ctxt
, addr
, val
, bytes
,
4224 exception
, &read_emultor
);
4227 int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
4231 struct x86_exception
*exception
)
4233 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
4234 exception
, &write_emultor
);
4237 #define CMPXCHG_TYPE(t, ptr, old, new) \
4238 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4240 #ifdef CONFIG_X86_64
4241 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4243 # define CMPXCHG64(ptr, old, new) \
4244 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4247 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
4252 struct x86_exception
*exception
)
4254 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4260 /* guests cmpxchg8b have to be emulated atomically */
4261 if (bytes
> 8 || (bytes
& (bytes
- 1)))
4264 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
4266 if (gpa
== UNMAPPED_GVA
||
4267 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4270 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
4273 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
4274 if (is_error_page(page
))
4277 kaddr
= kmap_atomic(page
);
4278 kaddr
+= offset_in_page(gpa
);
4281 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
4284 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
4287 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
4290 exchanged
= CMPXCHG64(kaddr
, old
, new);
4295 kunmap_atomic(kaddr
);
4296 kvm_release_page_dirty(page
);
4299 return X86EMUL_CMPXCHG_FAILED
;
4301 kvm_mmu_pte_write(vcpu
, gpa
, new, bytes
);
4303 return X86EMUL_CONTINUE
;
4306 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
4308 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
4311 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
4313 /* TODO: String I/O for in kernel device */
4316 if (vcpu
->arch
.pio
.in
)
4317 r
= kvm_io_bus_read(vcpu
->kvm
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
4318 vcpu
->arch
.pio
.size
, pd
);
4320 r
= kvm_io_bus_write(vcpu
->kvm
, KVM_PIO_BUS
,
4321 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
4326 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
4327 unsigned short port
, void *val
,
4328 unsigned int count
, bool in
)
4330 trace_kvm_pio(!in
, port
, size
, count
);
4332 vcpu
->arch
.pio
.port
= port
;
4333 vcpu
->arch
.pio
.in
= in
;
4334 vcpu
->arch
.pio
.count
= count
;
4335 vcpu
->arch
.pio
.size
= size
;
4337 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
4338 vcpu
->arch
.pio
.count
= 0;
4342 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
4343 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
4344 vcpu
->run
->io
.size
= size
;
4345 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
4346 vcpu
->run
->io
.count
= count
;
4347 vcpu
->run
->io
.port
= port
;
4352 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
4353 int size
, unsigned short port
, void *val
,
4356 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4359 if (vcpu
->arch
.pio
.count
)
4362 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
4365 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
4366 vcpu
->arch
.pio
.count
= 0;
4373 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
4374 int size
, unsigned short port
,
4375 const void *val
, unsigned int count
)
4377 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4379 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
4380 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
4383 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
4385 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
4388 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
4390 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
4393 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
4395 if (!need_emulate_wbinvd(vcpu
))
4396 return X86EMUL_CONTINUE
;
4398 if (kvm_x86_ops
->has_wbinvd_exit()) {
4399 int cpu
= get_cpu();
4401 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
4402 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
4403 wbinvd_ipi
, NULL
, 1);
4405 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
4408 return X86EMUL_CONTINUE
;
4410 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
4412 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
4414 kvm_emulate_wbinvd(emul_to_vcpu(ctxt
));
4417 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
4419 return _kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
4422 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
4425 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
4428 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
4430 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
4433 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
4435 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4436 unsigned long value
;
4440 value
= kvm_read_cr0(vcpu
);
4443 value
= vcpu
->arch
.cr2
;
4446 value
= kvm_read_cr3(vcpu
);
4449 value
= kvm_read_cr4(vcpu
);
4452 value
= kvm_get_cr8(vcpu
);
4455 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4462 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
4464 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4469 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
4472 vcpu
->arch
.cr2
= val
;
4475 res
= kvm_set_cr3(vcpu
, val
);
4478 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
4481 res
= kvm_set_cr8(vcpu
, val
);
4484 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4491 static void emulator_set_rflags(struct x86_emulate_ctxt
*ctxt
, ulong val
)
4493 kvm_set_rflags(emul_to_vcpu(ctxt
), val
);
4496 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
4498 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
4501 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4503 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
4506 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4508 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
4511 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4513 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
4516 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4518 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
4521 static unsigned long emulator_get_cached_segment_base(
4522 struct x86_emulate_ctxt
*ctxt
, int seg
)
4524 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
4527 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
4528 struct desc_struct
*desc
, u32
*base3
,
4531 struct kvm_segment var
;
4533 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
4534 *selector
= var
.selector
;
4537 memset(desc
, 0, sizeof(*desc
));
4543 set_desc_limit(desc
, var
.limit
);
4544 set_desc_base(desc
, (unsigned long)var
.base
);
4545 #ifdef CONFIG_X86_64
4547 *base3
= var
.base
>> 32;
4549 desc
->type
= var
.type
;
4551 desc
->dpl
= var
.dpl
;
4552 desc
->p
= var
.present
;
4553 desc
->avl
= var
.avl
;
4561 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
4562 struct desc_struct
*desc
, u32 base3
,
4565 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4566 struct kvm_segment var
;
4568 var
.selector
= selector
;
4569 var
.base
= get_desc_base(desc
);
4570 #ifdef CONFIG_X86_64
4571 var
.base
|= ((u64
)base3
) << 32;
4573 var
.limit
= get_desc_limit(desc
);
4575 var
.limit
= (var
.limit
<< 12) | 0xfff;
4576 var
.type
= desc
->type
;
4577 var
.present
= desc
->p
;
4578 var
.dpl
= desc
->dpl
;
4583 var
.avl
= desc
->avl
;
4584 var
.present
= desc
->p
;
4585 var
.unusable
= !var
.present
;
4588 kvm_set_segment(vcpu
, &var
, seg
);
4592 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
4593 u32 msr_index
, u64
*pdata
)
4595 return kvm_get_msr(emul_to_vcpu(ctxt
), msr_index
, pdata
);
4598 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
4599 u32 msr_index
, u64 data
)
4601 struct msr_data msr
;
4604 msr
.index
= msr_index
;
4605 msr
.host_initiated
= false;
4606 return kvm_set_msr(emul_to_vcpu(ctxt
), &msr
);
4609 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
4610 u32 pmc
, u64
*pdata
)
4612 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
4615 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
4617 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
4620 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
4623 kvm_load_guest_fpu(emul_to_vcpu(ctxt
));
4625 * CR0.TS may reference the host fpu state, not the guest fpu state,
4626 * so it may be clear at this point.
4631 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
4636 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
4637 struct x86_instruction_info
*info
,
4638 enum x86_intercept_stage stage
)
4640 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
4643 static void emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
4644 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
)
4646 kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
);
4649 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
4651 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
4654 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
4656 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
4659 static const struct x86_emulate_ops emulate_ops
= {
4660 .read_gpr
= emulator_read_gpr
,
4661 .write_gpr
= emulator_write_gpr
,
4662 .read_std
= kvm_read_guest_virt_system
,
4663 .write_std
= kvm_write_guest_virt_system
,
4664 .fetch
= kvm_fetch_guest_virt
,
4665 .read_emulated
= emulator_read_emulated
,
4666 .write_emulated
= emulator_write_emulated
,
4667 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
4668 .invlpg
= emulator_invlpg
,
4669 .pio_in_emulated
= emulator_pio_in_emulated
,
4670 .pio_out_emulated
= emulator_pio_out_emulated
,
4671 .get_segment
= emulator_get_segment
,
4672 .set_segment
= emulator_set_segment
,
4673 .get_cached_segment_base
= emulator_get_cached_segment_base
,
4674 .get_gdt
= emulator_get_gdt
,
4675 .get_idt
= emulator_get_idt
,
4676 .set_gdt
= emulator_set_gdt
,
4677 .set_idt
= emulator_set_idt
,
4678 .get_cr
= emulator_get_cr
,
4679 .set_cr
= emulator_set_cr
,
4680 .set_rflags
= emulator_set_rflags
,
4681 .cpl
= emulator_get_cpl
,
4682 .get_dr
= emulator_get_dr
,
4683 .set_dr
= emulator_set_dr
,
4684 .set_msr
= emulator_set_msr
,
4685 .get_msr
= emulator_get_msr
,
4686 .read_pmc
= emulator_read_pmc
,
4687 .halt
= emulator_halt
,
4688 .wbinvd
= emulator_wbinvd
,
4689 .fix_hypercall
= emulator_fix_hypercall
,
4690 .get_fpu
= emulator_get_fpu
,
4691 .put_fpu
= emulator_put_fpu
,
4692 .intercept
= emulator_intercept
,
4693 .get_cpuid
= emulator_get_cpuid
,
4696 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
4698 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
, mask
);
4700 * an sti; sti; sequence only disable interrupts for the first
4701 * instruction. So, if the last instruction, be it emulated or
4702 * not, left the system with the INT_STI flag enabled, it
4703 * means that the last instruction is an sti. We should not
4704 * leave the flag on in this case. The same goes for mov ss
4706 if (!(int_shadow
& mask
))
4707 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
4710 static void inject_emulated_exception(struct kvm_vcpu
*vcpu
)
4712 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4713 if (ctxt
->exception
.vector
== PF_VECTOR
)
4714 kvm_propagate_fault(vcpu
, &ctxt
->exception
);
4715 else if (ctxt
->exception
.error_code_valid
)
4716 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
4717 ctxt
->exception
.error_code
);
4719 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
4722 static void init_decode_cache(struct x86_emulate_ctxt
*ctxt
)
4724 memset(&ctxt
->twobyte
, 0,
4725 (void *)&ctxt
->_regs
- (void *)&ctxt
->twobyte
);
4727 ctxt
->fetch
.start
= 0;
4728 ctxt
->fetch
.end
= 0;
4729 ctxt
->io_read
.pos
= 0;
4730 ctxt
->io_read
.end
= 0;
4731 ctxt
->mem_read
.pos
= 0;
4732 ctxt
->mem_read
.end
= 0;
4735 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
4737 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4740 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
4742 ctxt
->eflags
= kvm_get_rflags(vcpu
);
4743 ctxt
->eip
= kvm_rip_read(vcpu
);
4744 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
4745 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
4746 cs_l
? X86EMUL_MODE_PROT64
:
4747 cs_db
? X86EMUL_MODE_PROT32
:
4748 X86EMUL_MODE_PROT16
;
4749 ctxt
->guest_mode
= is_guest_mode(vcpu
);
4751 init_decode_cache(ctxt
);
4752 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4755 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
4757 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4760 init_emulate_ctxt(vcpu
);
4764 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
4765 ret
= emulate_int_real(ctxt
, irq
);
4767 if (ret
!= X86EMUL_CONTINUE
)
4768 return EMULATE_FAIL
;
4770 ctxt
->eip
= ctxt
->_eip
;
4771 kvm_rip_write(vcpu
, ctxt
->eip
);
4772 kvm_set_rflags(vcpu
, ctxt
->eflags
);
4774 if (irq
== NMI_VECTOR
)
4775 vcpu
->arch
.nmi_pending
= 0;
4777 vcpu
->arch
.interrupt
.pending
= false;
4779 return EMULATE_DONE
;
4781 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
4783 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
4785 int r
= EMULATE_DONE
;
4787 ++vcpu
->stat
.insn_emulation_fail
;
4788 trace_kvm_emulate_insn_failed(vcpu
);
4789 if (!is_guest_mode(vcpu
)) {
4790 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
4791 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
4792 vcpu
->run
->internal
.ndata
= 0;
4795 kvm_queue_exception(vcpu
, UD_VECTOR
);
4800 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t cr2
,
4801 bool write_fault_to_shadow_pgtable
,
4807 if (emulation_type
& EMULTYPE_NO_REEXECUTE
)
4810 if (!vcpu
->arch
.mmu
.direct_map
) {
4812 * Write permission should be allowed since only
4813 * write access need to be emulated.
4815 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
4818 * If the mapping is invalid in guest, let cpu retry
4819 * it to generate fault.
4821 if (gpa
== UNMAPPED_GVA
)
4826 * Do not retry the unhandleable instruction if it faults on the
4827 * readonly host memory, otherwise it will goto a infinite loop:
4828 * retry instruction -> write #PF -> emulation fail -> retry
4829 * instruction -> ...
4831 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
4834 * If the instruction failed on the error pfn, it can not be fixed,
4835 * report the error to userspace.
4837 if (is_error_noslot_pfn(pfn
))
4840 kvm_release_pfn_clean(pfn
);
4842 /* The instructions are well-emulated on direct mmu. */
4843 if (vcpu
->arch
.mmu
.direct_map
) {
4844 unsigned int indirect_shadow_pages
;
4846 spin_lock(&vcpu
->kvm
->mmu_lock
);
4847 indirect_shadow_pages
= vcpu
->kvm
->arch
.indirect_shadow_pages
;
4848 spin_unlock(&vcpu
->kvm
->mmu_lock
);
4850 if (indirect_shadow_pages
)
4851 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
4857 * if emulation was due to access to shadowed page table
4858 * and it failed try to unshadow page and re-enter the
4859 * guest to let CPU execute the instruction.
4861 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
4864 * If the access faults on its page table, it can not
4865 * be fixed by unprotecting shadow page and it should
4866 * be reported to userspace.
4868 return !write_fault_to_shadow_pgtable
;
4871 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
4872 unsigned long cr2
, int emulation_type
)
4874 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4875 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
4877 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
4878 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
4881 * If the emulation is caused by #PF and it is non-page_table
4882 * writing instruction, it means the VM-EXIT is caused by shadow
4883 * page protected, we can zap the shadow page and retry this
4884 * instruction directly.
4886 * Note: if the guest uses a non-page-table modifying instruction
4887 * on the PDE that points to the instruction, then we will unmap
4888 * the instruction and go to an infinite loop. So, we cache the
4889 * last retried eip and the last fault address, if we meet the eip
4890 * and the address again, we can break out of the potential infinite
4893 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
4895 if (!(emulation_type
& EMULTYPE_RETRY
))
4898 if (x86_page_table_writing_insn(ctxt
))
4901 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
4904 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
4905 vcpu
->arch
.last_retry_addr
= cr2
;
4907 if (!vcpu
->arch
.mmu
.direct_map
)
4908 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
4910 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
4915 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
4916 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
4918 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
4925 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4926 bool writeback
= true;
4927 bool write_fault_to_spt
= vcpu
->arch
.write_fault_to_shadow_pgtable
;
4930 * Clear write_fault_to_shadow_pgtable here to ensure it is
4933 vcpu
->arch
.write_fault_to_shadow_pgtable
= false;
4934 kvm_clear_exception_queue(vcpu
);
4936 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
4937 init_emulate_ctxt(vcpu
);
4938 ctxt
->interruptibility
= 0;
4939 ctxt
->have_exception
= false;
4940 ctxt
->perm_ok
= false;
4942 ctxt
->only_vendor_specific_insn
4943 = emulation_type
& EMULTYPE_TRAP_UD
;
4945 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
4947 trace_kvm_emulate_insn_start(vcpu
);
4948 ++vcpu
->stat
.insn_emulation
;
4949 if (r
!= EMULATION_OK
) {
4950 if (emulation_type
& EMULTYPE_TRAP_UD
)
4951 return EMULATE_FAIL
;
4952 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
4954 return EMULATE_DONE
;
4955 if (emulation_type
& EMULTYPE_SKIP
)
4956 return EMULATE_FAIL
;
4957 return handle_emulation_failure(vcpu
);
4961 if (emulation_type
& EMULTYPE_SKIP
) {
4962 kvm_rip_write(vcpu
, ctxt
->_eip
);
4963 return EMULATE_DONE
;
4966 if (retry_instruction(ctxt
, cr2
, emulation_type
))
4967 return EMULATE_DONE
;
4969 /* this is needed for vmware backdoor interface to work since it
4970 changes registers values during IO operation */
4971 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
4972 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4973 emulator_invalidate_register_cache(ctxt
);
4977 r
= x86_emulate_insn(ctxt
);
4979 if (r
== EMULATION_INTERCEPTED
)
4980 return EMULATE_DONE
;
4982 if (r
== EMULATION_FAILED
) {
4983 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
4985 return EMULATE_DONE
;
4987 return handle_emulation_failure(vcpu
);
4990 if (ctxt
->have_exception
) {
4991 inject_emulated_exception(vcpu
);
4993 } else if (vcpu
->arch
.pio
.count
) {
4994 if (!vcpu
->arch
.pio
.in
)
4995 vcpu
->arch
.pio
.count
= 0;
4998 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
5000 r
= EMULATE_DO_MMIO
;
5001 } else if (vcpu
->mmio_needed
) {
5002 if (!vcpu
->mmio_is_write
)
5004 r
= EMULATE_DO_MMIO
;
5005 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
5006 } else if (r
== EMULATION_RESTART
)
5012 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
5013 kvm_set_rflags(vcpu
, ctxt
->eflags
);
5014 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5015 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5016 kvm_rip_write(vcpu
, ctxt
->eip
);
5018 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
5022 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
5024 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5026 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5027 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
5028 size
, port
, &val
, 1);
5029 /* do not return to emulator after return from userspace */
5030 vcpu
->arch
.pio
.count
= 0;
5033 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
5035 static void tsc_bad(void *info
)
5037 __this_cpu_write(cpu_tsc_khz
, 0);
5040 static void tsc_khz_changed(void *data
)
5042 struct cpufreq_freqs
*freq
= data
;
5043 unsigned long khz
= 0;
5047 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5048 khz
= cpufreq_quick_get(raw_smp_processor_id());
5051 __this_cpu_write(cpu_tsc_khz
, khz
);
5054 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
5057 struct cpufreq_freqs
*freq
= data
;
5059 struct kvm_vcpu
*vcpu
;
5060 int i
, send_ipi
= 0;
5063 * We allow guests to temporarily run on slowing clocks,
5064 * provided we notify them after, or to run on accelerating
5065 * clocks, provided we notify them before. Thus time never
5068 * However, we have a problem. We can't atomically update
5069 * the frequency of a given CPU from this function; it is
5070 * merely a notifier, which can be called from any CPU.
5071 * Changing the TSC frequency at arbitrary points in time
5072 * requires a recomputation of local variables related to
5073 * the TSC for each VCPU. We must flag these local variables
5074 * to be updated and be sure the update takes place with the
5075 * new frequency before any guests proceed.
5077 * Unfortunately, the combination of hotplug CPU and frequency
5078 * change creates an intractable locking scenario; the order
5079 * of when these callouts happen is undefined with respect to
5080 * CPU hotplug, and they can race with each other. As such,
5081 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5082 * undefined; you can actually have a CPU frequency change take
5083 * place in between the computation of X and the setting of the
5084 * variable. To protect against this problem, all updates of
5085 * the per_cpu tsc_khz variable are done in an interrupt
5086 * protected IPI, and all callers wishing to update the value
5087 * must wait for a synchronous IPI to complete (which is trivial
5088 * if the caller is on the CPU already). This establishes the
5089 * necessary total order on variable updates.
5091 * Note that because a guest time update may take place
5092 * anytime after the setting of the VCPU's request bit, the
5093 * correct TSC value must be set before the request. However,
5094 * to ensure the update actually makes it to any guest which
5095 * starts running in hardware virtualization between the set
5096 * and the acquisition of the spinlock, we must also ping the
5097 * CPU after setting the request bit.
5101 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
5103 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
5106 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5108 raw_spin_lock(&kvm_lock
);
5109 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5110 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
5111 if (vcpu
->cpu
!= freq
->cpu
)
5113 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5114 if (vcpu
->cpu
!= smp_processor_id())
5118 raw_spin_unlock(&kvm_lock
);
5120 if (freq
->old
< freq
->new && send_ipi
) {
5122 * We upscale the frequency. Must make the guest
5123 * doesn't see old kvmclock values while running with
5124 * the new frequency, otherwise we risk the guest sees
5125 * time go backwards.
5127 * In case we update the frequency for another cpu
5128 * (which might be in guest context) send an interrupt
5129 * to kick the cpu out of guest context. Next time
5130 * guest context is entered kvmclock will be updated,
5131 * so the guest will not see stale values.
5133 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5138 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
5139 .notifier_call
= kvmclock_cpufreq_notifier
5142 static int kvmclock_cpu_notifier(struct notifier_block
*nfb
,
5143 unsigned long action
, void *hcpu
)
5145 unsigned int cpu
= (unsigned long)hcpu
;
5149 case CPU_DOWN_FAILED
:
5150 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5152 case CPU_DOWN_PREPARE
:
5153 smp_call_function_single(cpu
, tsc_bad
, NULL
, 1);
5159 static struct notifier_block kvmclock_cpu_notifier_block
= {
5160 .notifier_call
= kvmclock_cpu_notifier
,
5161 .priority
= -INT_MAX
5164 static void kvm_timer_init(void)
5168 max_tsc_khz
= tsc_khz
;
5169 register_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5170 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
5171 #ifdef CONFIG_CPU_FREQ
5172 struct cpufreq_policy policy
;
5173 memset(&policy
, 0, sizeof(policy
));
5175 cpufreq_get_policy(&policy
, cpu
);
5176 if (policy
.cpuinfo
.max_freq
)
5177 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
5180 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
5181 CPUFREQ_TRANSITION_NOTIFIER
);
5183 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
5184 for_each_online_cpu(cpu
)
5185 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5188 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
5190 int kvm_is_in_guest(void)
5192 return __this_cpu_read(current_vcpu
) != NULL
;
5195 static int kvm_is_user_mode(void)
5199 if (__this_cpu_read(current_vcpu
))
5200 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
5202 return user_mode
!= 0;
5205 static unsigned long kvm_get_guest_ip(void)
5207 unsigned long ip
= 0;
5209 if (__this_cpu_read(current_vcpu
))
5210 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
5215 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
5216 .is_in_guest
= kvm_is_in_guest
,
5217 .is_user_mode
= kvm_is_user_mode
,
5218 .get_guest_ip
= kvm_get_guest_ip
,
5221 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
5223 __this_cpu_write(current_vcpu
, vcpu
);
5225 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
5227 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
5229 __this_cpu_write(current_vcpu
, NULL
);
5231 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
5233 static void kvm_set_mmio_spte_mask(void)
5236 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
5239 * Set the reserved bits and the present bit of an paging-structure
5240 * entry to generate page fault with PFER.RSV = 1.
5242 mask
= ((1ull << (62 - maxphyaddr
+ 1)) - 1) << maxphyaddr
;
5245 #ifdef CONFIG_X86_64
5247 * If reserved bit is not supported, clear the present bit to disable
5250 if (maxphyaddr
== 52)
5254 kvm_mmu_set_mmio_spte_mask(mask
);
5257 #ifdef CONFIG_X86_64
5258 static void pvclock_gtod_update_fn(struct work_struct
*work
)
5262 struct kvm_vcpu
*vcpu
;
5265 raw_spin_lock(&kvm_lock
);
5266 list_for_each_entry(kvm
, &vm_list
, vm_list
)
5267 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5268 set_bit(KVM_REQ_MASTERCLOCK_UPDATE
, &vcpu
->requests
);
5269 atomic_set(&kvm_guest_has_master_clock
, 0);
5270 raw_spin_unlock(&kvm_lock
);
5273 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
5276 * Notification about pvclock gtod data update.
5278 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
5281 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
5282 struct timekeeper
*tk
= priv
;
5284 update_pvclock_gtod(tk
);
5286 /* disable master clock if host does not trust, or does not
5287 * use, TSC clocksource
5289 if (gtod
->clock
.vclock_mode
!= VCLOCK_TSC
&&
5290 atomic_read(&kvm_guest_has_master_clock
) != 0)
5291 queue_work(system_long_wq
, &pvclock_gtod_work
);
5296 static struct notifier_block pvclock_gtod_notifier
= {
5297 .notifier_call
= pvclock_gtod_notify
,
5301 int kvm_arch_init(void *opaque
)
5304 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
5307 printk(KERN_ERR
"kvm: already loaded the other module\n");
5312 if (!ops
->cpu_has_kvm_support()) {
5313 printk(KERN_ERR
"kvm: no hardware support\n");
5317 if (ops
->disabled_by_bios()) {
5318 printk(KERN_ERR
"kvm: disabled by bios\n");
5324 shared_msrs
= alloc_percpu(struct kvm_shared_msrs
);
5326 printk(KERN_ERR
"kvm: failed to allocate percpu kvm_shared_msrs\n");
5330 r
= kvm_mmu_module_init();
5332 goto out_free_percpu
;
5334 kvm_set_mmio_spte_mask();
5335 kvm_init_msr_list();
5338 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
5339 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
5343 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
5346 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
5349 #ifdef CONFIG_X86_64
5350 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
5356 free_percpu(shared_msrs
);
5361 void kvm_arch_exit(void)
5363 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
5365 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5366 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
5367 CPUFREQ_TRANSITION_NOTIFIER
);
5368 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5369 #ifdef CONFIG_X86_64
5370 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
5373 kvm_mmu_module_exit();
5374 free_percpu(shared_msrs
);
5377 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
5379 ++vcpu
->stat
.halt_exits
;
5380 if (irqchip_in_kernel(vcpu
->kvm
)) {
5381 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
5384 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
5388 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
5390 int kvm_hv_hypercall(struct kvm_vcpu
*vcpu
)
5392 u64 param
, ingpa
, outgpa
, ret
;
5393 uint16_t code
, rep_idx
, rep_cnt
, res
= HV_STATUS_SUCCESS
, rep_done
= 0;
5394 bool fast
, longmode
;
5398 * hypercall generates UD from non zero cpl and real mode
5401 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 || !is_protmode(vcpu
)) {
5402 kvm_queue_exception(vcpu
, UD_VECTOR
);
5406 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
5407 longmode
= is_long_mode(vcpu
) && cs_l
== 1;
5410 param
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDX
) << 32) |
5411 (kvm_register_read(vcpu
, VCPU_REGS_RAX
) & 0xffffffff);
5412 ingpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RBX
) << 32) |
5413 (kvm_register_read(vcpu
, VCPU_REGS_RCX
) & 0xffffffff);
5414 outgpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDI
) << 32) |
5415 (kvm_register_read(vcpu
, VCPU_REGS_RSI
) & 0xffffffff);
5417 #ifdef CONFIG_X86_64
5419 param
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5420 ingpa
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5421 outgpa
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
5425 code
= param
& 0xffff;
5426 fast
= (param
>> 16) & 0x1;
5427 rep_cnt
= (param
>> 32) & 0xfff;
5428 rep_idx
= (param
>> 48) & 0xfff;
5430 trace_kvm_hv_hypercall(code
, fast
, rep_cnt
, rep_idx
, ingpa
, outgpa
);
5433 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT
:
5434 kvm_vcpu_on_spin(vcpu
);
5437 res
= HV_STATUS_INVALID_HYPERCALL_CODE
;
5441 ret
= res
| (((u64
)rep_done
& 0xfff) << 32);
5443 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5445 kvm_register_write(vcpu
, VCPU_REGS_RDX
, ret
>> 32);
5446 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
& 0xffffffff);
5452 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
5454 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
5457 if (kvm_hv_hypercall_enabled(vcpu
->kvm
))
5458 return kvm_hv_hypercall(vcpu
);
5460 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5461 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5462 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5463 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5464 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5466 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
5468 if (!is_long_mode(vcpu
)) {
5476 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
5482 case KVM_HC_VAPIC_POLL_IRQ
:
5490 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5491 ++vcpu
->stat
.hypercalls
;
5494 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
5496 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
5498 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5499 char instruction
[3];
5500 unsigned long rip
= kvm_rip_read(vcpu
);
5503 * Blow out the MMU to ensure that no other VCPU has an active mapping
5504 * to ensure that the updated hypercall appears atomically across all
5507 kvm_mmu_zap_all(vcpu
->kvm
);
5509 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
5511 return emulator_write_emulated(ctxt
, rip
, instruction
, 3, NULL
);
5515 * Check if userspace requested an interrupt window, and that the
5516 * interrupt window is open.
5518 * No need to exit to userspace if we already have an interrupt queued.
5520 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
5522 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
5523 vcpu
->run
->request_interrupt_window
&&
5524 kvm_arch_interrupt_allowed(vcpu
));
5527 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
5529 struct kvm_run
*kvm_run
= vcpu
->run
;
5531 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
5532 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
5533 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
5534 if (irqchip_in_kernel(vcpu
->kvm
))
5535 kvm_run
->ready_for_interrupt_injection
= 1;
5537 kvm_run
->ready_for_interrupt_injection
=
5538 kvm_arch_interrupt_allowed(vcpu
) &&
5539 !kvm_cpu_has_interrupt(vcpu
) &&
5540 !kvm_event_needs_reinjection(vcpu
);
5543 static int vapic_enter(struct kvm_vcpu
*vcpu
)
5545 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5548 if (!apic
|| !apic
->vapic_addr
)
5551 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5552 if (is_error_page(page
))
5555 vcpu
->arch
.apic
->vapic_page
= page
;
5559 static void vapic_exit(struct kvm_vcpu
*vcpu
)
5561 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5564 if (!apic
|| !apic
->vapic_addr
)
5567 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5568 kvm_release_page_dirty(apic
->vapic_page
);
5569 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5570 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5573 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
5577 if (!kvm_x86_ops
->update_cr8_intercept
)
5580 if (!vcpu
->arch
.apic
)
5583 if (!vcpu
->arch
.apic
->vapic_addr
)
5584 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
5591 tpr
= kvm_lapic_get_cr8(vcpu
);
5593 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
5596 static void inject_pending_event(struct kvm_vcpu
*vcpu
)
5598 /* try to reinject previous events if any */
5599 if (vcpu
->arch
.exception
.pending
) {
5600 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
5601 vcpu
->arch
.exception
.has_error_code
,
5602 vcpu
->arch
.exception
.error_code
);
5603 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
5604 vcpu
->arch
.exception
.has_error_code
,
5605 vcpu
->arch
.exception
.error_code
,
5606 vcpu
->arch
.exception
.reinject
);
5610 if (vcpu
->arch
.nmi_injected
) {
5611 kvm_x86_ops
->set_nmi(vcpu
);
5615 if (vcpu
->arch
.interrupt
.pending
) {
5616 kvm_x86_ops
->set_irq(vcpu
);
5620 /* try to inject new event if pending */
5621 if (vcpu
->arch
.nmi_pending
) {
5622 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
5623 --vcpu
->arch
.nmi_pending
;
5624 vcpu
->arch
.nmi_injected
= true;
5625 kvm_x86_ops
->set_nmi(vcpu
);
5627 } else if (kvm_cpu_has_injectable_intr(vcpu
)) {
5628 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
5629 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
5631 kvm_x86_ops
->set_irq(vcpu
);
5636 static void process_nmi(struct kvm_vcpu
*vcpu
)
5641 * x86 is limited to one NMI running, and one NMI pending after it.
5642 * If an NMI is already in progress, limit further NMIs to just one.
5643 * Otherwise, allow two (and we'll inject the first one immediately).
5645 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
5648 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
5649 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
5650 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5653 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
5655 #ifdef CONFIG_X86_64
5657 struct kvm_vcpu
*vcpu
;
5658 struct kvm_arch
*ka
= &kvm
->arch
;
5660 spin_lock(&ka
->pvclock_gtod_sync_lock
);
5661 kvm_make_mclock_inprogress_request(kvm
);
5662 /* no guest entries from this point */
5663 pvclock_update_vm_gtod_copy(kvm
);
5665 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5666 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
5668 /* guest entries allowed */
5669 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5670 clear_bit(KVM_REQ_MCLOCK_INPROGRESS
, &vcpu
->requests
);
5672 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
5676 static void vcpu_scan_ioapic(struct kvm_vcpu
*vcpu
)
5678 u64 eoi_exit_bitmap
[4];
5681 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
5684 memset(eoi_exit_bitmap
, 0, 32);
5687 kvm_ioapic_scan_entry(vcpu
, eoi_exit_bitmap
, tmr
);
5688 kvm_x86_ops
->load_eoi_exitmap(vcpu
, eoi_exit_bitmap
);
5689 kvm_apic_update_tmr(vcpu
, tmr
);
5692 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
5695 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
5696 vcpu
->run
->request_interrupt_window
;
5697 bool req_immediate_exit
= false;
5699 if (vcpu
->requests
) {
5700 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
5701 kvm_mmu_unload(vcpu
);
5702 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
5703 __kvm_migrate_timers(vcpu
);
5704 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
5705 kvm_gen_update_masterclock(vcpu
->kvm
);
5706 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
5707 r
= kvm_guest_time_update(vcpu
);
5711 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
5712 kvm_mmu_sync_roots(vcpu
);
5713 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
5714 kvm_x86_ops
->tlb_flush(vcpu
);
5715 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
5716 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
5720 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
5721 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
5725 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
)) {
5726 vcpu
->fpu_active
= 0;
5727 kvm_x86_ops
->fpu_deactivate(vcpu
);
5729 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
5730 /* Page is swapped out. Do synthetic halt */
5731 vcpu
->arch
.apf
.halted
= true;
5735 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
5736 record_steal_time(vcpu
);
5737 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
5739 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
5740 kvm_handle_pmu_event(vcpu
);
5741 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
5742 kvm_deliver_pmi(vcpu
);
5743 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC
, vcpu
))
5744 vcpu_scan_ioapic(vcpu
);
5747 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
5748 kvm_apic_accept_events(vcpu
);
5749 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
5754 inject_pending_event(vcpu
);
5756 /* enable NMI/IRQ window open exits if needed */
5757 if (vcpu
->arch
.nmi_pending
)
5758 req_immediate_exit
=
5759 kvm_x86_ops
->enable_nmi_window(vcpu
) != 0;
5760 else if (kvm_cpu_has_injectable_intr(vcpu
) || req_int_win
)
5761 req_immediate_exit
=
5762 kvm_x86_ops
->enable_irq_window(vcpu
) != 0;
5764 if (kvm_lapic_enabled(vcpu
)) {
5766 * Update architecture specific hints for APIC
5767 * virtual interrupt delivery.
5769 if (kvm_x86_ops
->hwapic_irr_update
)
5770 kvm_x86_ops
->hwapic_irr_update(vcpu
,
5771 kvm_lapic_find_highest_irr(vcpu
));
5772 update_cr8_intercept(vcpu
);
5773 kvm_lapic_sync_to_vapic(vcpu
);
5777 r
= kvm_mmu_reload(vcpu
);
5779 goto cancel_injection
;
5784 kvm_x86_ops
->prepare_guest_switch(vcpu
);
5785 if (vcpu
->fpu_active
)
5786 kvm_load_guest_fpu(vcpu
);
5787 kvm_load_guest_xcr0(vcpu
);
5789 vcpu
->mode
= IN_GUEST_MODE
;
5791 /* We should set ->mode before check ->requests,
5792 * see the comment in make_all_cpus_request.
5796 local_irq_disable();
5798 if (vcpu
->mode
== EXITING_GUEST_MODE
|| vcpu
->requests
5799 || need_resched() || signal_pending(current
)) {
5800 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5805 goto cancel_injection
;
5808 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5810 if (req_immediate_exit
)
5811 smp_send_reschedule(vcpu
->cpu
);
5815 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
5817 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
5818 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
5819 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
5820 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
5823 trace_kvm_entry(vcpu
->vcpu_id
);
5824 kvm_x86_ops
->run(vcpu
);
5827 * If the guest has used debug registers, at least dr7
5828 * will be disabled while returning to the host.
5829 * If we don't have active breakpoints in the host, we don't
5830 * care about the messed up debug address registers. But if
5831 * we have some of them active, restore the old state.
5833 if (hw_breakpoint_active())
5834 hw_breakpoint_restore();
5836 vcpu
->arch
.last_guest_tsc
= kvm_x86_ops
->read_l1_tsc(vcpu
,
5839 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5842 /* Interrupt is enabled by handle_external_intr() */
5843 kvm_x86_ops
->handle_external_intr(vcpu
);
5848 * We must have an instruction between local_irq_enable() and
5849 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5850 * the interrupt shadow. The stat.exits increment will do nicely.
5851 * But we need to prevent reordering, hence this barrier():
5859 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5862 * Profile KVM exit RIPs:
5864 if (unlikely(prof_on
== KVM_PROFILING
)) {
5865 unsigned long rip
= kvm_rip_read(vcpu
);
5866 profile_hit(KVM_PROFILING
, (void *)rip
);
5869 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
5870 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5872 if (vcpu
->arch
.apic_attention
)
5873 kvm_lapic_sync_from_vapic(vcpu
);
5875 r
= kvm_x86_ops
->handle_exit(vcpu
);
5879 kvm_x86_ops
->cancel_injection(vcpu
);
5880 if (unlikely(vcpu
->arch
.apic_attention
))
5881 kvm_lapic_sync_from_vapic(vcpu
);
5887 static int __vcpu_run(struct kvm_vcpu
*vcpu
)
5890 struct kvm
*kvm
= vcpu
->kvm
;
5892 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5893 r
= vapic_enter(vcpu
);
5895 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5901 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
5902 !vcpu
->arch
.apf
.halted
)
5903 r
= vcpu_enter_guest(vcpu
);
5905 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5906 kvm_vcpu_block(vcpu
);
5907 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5908 if (kvm_check_request(KVM_REQ_UNHALT
, vcpu
)) {
5909 kvm_apic_accept_events(vcpu
);
5910 switch(vcpu
->arch
.mp_state
) {
5911 case KVM_MP_STATE_HALTED
:
5912 vcpu
->arch
.mp_state
=
5913 KVM_MP_STATE_RUNNABLE
;
5914 case KVM_MP_STATE_RUNNABLE
:
5915 vcpu
->arch
.apf
.halted
= false;
5917 case KVM_MP_STATE_INIT_RECEIVED
:
5929 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
5930 if (kvm_cpu_has_pending_timer(vcpu
))
5931 kvm_inject_pending_timer_irqs(vcpu
);
5933 if (dm_request_for_irq_injection(vcpu
)) {
5935 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5936 ++vcpu
->stat
.request_irq_exits
;
5939 kvm_check_async_pf_completion(vcpu
);
5941 if (signal_pending(current
)) {
5943 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5944 ++vcpu
->stat
.signal_exits
;
5946 if (need_resched()) {
5947 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5949 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5953 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5960 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
5963 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5964 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
5965 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5966 if (r
!= EMULATE_DONE
)
5971 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
5973 BUG_ON(!vcpu
->arch
.pio
.count
);
5975 return complete_emulated_io(vcpu
);
5979 * Implements the following, as a state machine:
5983 * for each mmio piece in the fragment
5991 * for each mmio piece in the fragment
5996 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
5998 struct kvm_run
*run
= vcpu
->run
;
5999 struct kvm_mmio_fragment
*frag
;
6002 BUG_ON(!vcpu
->mmio_needed
);
6004 /* Complete previous fragment */
6005 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
6006 len
= min(8u, frag
->len
);
6007 if (!vcpu
->mmio_is_write
)
6008 memcpy(frag
->data
, run
->mmio
.data
, len
);
6010 if (frag
->len
<= 8) {
6011 /* Switch to the next fragment. */
6013 vcpu
->mmio_cur_fragment
++;
6015 /* Go forward to the next mmio piece. */
6021 if (vcpu
->mmio_cur_fragment
== vcpu
->mmio_nr_fragments
) {
6022 vcpu
->mmio_needed
= 0;
6023 if (vcpu
->mmio_is_write
)
6025 vcpu
->mmio_read_completed
= 1;
6026 return complete_emulated_io(vcpu
);
6029 run
->exit_reason
= KVM_EXIT_MMIO
;
6030 run
->mmio
.phys_addr
= frag
->gpa
;
6031 if (vcpu
->mmio_is_write
)
6032 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
6033 run
->mmio
.len
= min(8u, frag
->len
);
6034 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
6035 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
6040 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
6045 if (!tsk_used_math(current
) && init_fpu(current
))
6048 if (vcpu
->sigset_active
)
6049 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
6051 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
6052 kvm_vcpu_block(vcpu
);
6053 kvm_apic_accept_events(vcpu
);
6054 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
6059 /* re-sync apic's tpr */
6060 if (!irqchip_in_kernel(vcpu
->kvm
)) {
6061 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
6067 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
6068 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
6069 vcpu
->arch
.complete_userspace_io
= NULL
;
6074 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
6076 r
= __vcpu_run(vcpu
);
6079 post_kvm_run_save(vcpu
);
6080 if (vcpu
->sigset_active
)
6081 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
6086 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
6088 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
6090 * We are here if userspace calls get_regs() in the middle of
6091 * instruction emulation. Registers state needs to be copied
6092 * back from emulation context to vcpu. Userspace shouldn't do
6093 * that usually, but some bad designed PV devices (vmware
6094 * backdoor interface) need this to work
6096 emulator_writeback_register_cache(&vcpu
->arch
.emulate_ctxt
);
6097 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
6099 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
6100 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
6101 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
6102 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
6103 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
6104 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
6105 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
6106 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
6107 #ifdef CONFIG_X86_64
6108 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
6109 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
6110 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
6111 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
6112 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
6113 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
6114 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
6115 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
6118 regs
->rip
= kvm_rip_read(vcpu
);
6119 regs
->rflags
= kvm_get_rflags(vcpu
);
6124 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
6126 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
6127 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
6129 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
6130 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
6131 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
6132 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
6133 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
6134 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
6135 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
6136 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
6137 #ifdef CONFIG_X86_64
6138 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
6139 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
6140 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
6141 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
6142 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
6143 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
6144 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
6145 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
6148 kvm_rip_write(vcpu
, regs
->rip
);
6149 kvm_set_rflags(vcpu
, regs
->rflags
);
6151 vcpu
->arch
.exception
.pending
= false;
6153 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6158 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
6160 struct kvm_segment cs
;
6162 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6166 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
6168 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
6169 struct kvm_sregs
*sregs
)
6173 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
6174 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
6175 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
6176 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
6177 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
6178 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
6180 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
6181 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
6183 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6184 sregs
->idt
.limit
= dt
.size
;
6185 sregs
->idt
.base
= dt
.address
;
6186 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6187 sregs
->gdt
.limit
= dt
.size
;
6188 sregs
->gdt
.base
= dt
.address
;
6190 sregs
->cr0
= kvm_read_cr0(vcpu
);
6191 sregs
->cr2
= vcpu
->arch
.cr2
;
6192 sregs
->cr3
= kvm_read_cr3(vcpu
);
6193 sregs
->cr4
= kvm_read_cr4(vcpu
);
6194 sregs
->cr8
= kvm_get_cr8(vcpu
);
6195 sregs
->efer
= vcpu
->arch
.efer
;
6196 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
6198 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
6200 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
6201 set_bit(vcpu
->arch
.interrupt
.nr
,
6202 (unsigned long *)sregs
->interrupt_bitmap
);
6207 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
6208 struct kvm_mp_state
*mp_state
)
6210 kvm_apic_accept_events(vcpu
);
6211 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
6215 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
6216 struct kvm_mp_state
*mp_state
)
6218 if (!kvm_vcpu_has_lapic(vcpu
) &&
6219 mp_state
->mp_state
!= KVM_MP_STATE_RUNNABLE
)
6222 if (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
) {
6223 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
6224 set_bit(KVM_APIC_SIPI
, &vcpu
->arch
.apic
->pending_events
);
6226 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
6227 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6231 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
6232 int reason
, bool has_error_code
, u32 error_code
)
6234 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
6237 init_emulate_ctxt(vcpu
);
6239 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
6240 has_error_code
, error_code
);
6243 return EMULATE_FAIL
;
6245 kvm_rip_write(vcpu
, ctxt
->eip
);
6246 kvm_set_rflags(vcpu
, ctxt
->eflags
);
6247 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6248 return EMULATE_DONE
;
6250 EXPORT_SYMBOL_GPL(kvm_task_switch
);
6252 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
6253 struct kvm_sregs
*sregs
)
6255 int mmu_reset_needed
= 0;
6256 int pending_vec
, max_bits
, idx
;
6259 if (!guest_cpuid_has_xsave(vcpu
) && (sregs
->cr4
& X86_CR4_OSXSAVE
))
6262 dt
.size
= sregs
->idt
.limit
;
6263 dt
.address
= sregs
->idt
.base
;
6264 kvm_x86_ops
->set_idt(vcpu
, &dt
);
6265 dt
.size
= sregs
->gdt
.limit
;
6266 dt
.address
= sregs
->gdt
.base
;
6267 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
6269 vcpu
->arch
.cr2
= sregs
->cr2
;
6270 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
6271 vcpu
->arch
.cr3
= sregs
->cr3
;
6272 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
6274 kvm_set_cr8(vcpu
, sregs
->cr8
);
6276 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
6277 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
6278 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
6280 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
6281 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
6282 vcpu
->arch
.cr0
= sregs
->cr0
;
6284 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
6285 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
6286 if (sregs
->cr4
& X86_CR4_OSXSAVE
)
6287 kvm_update_cpuid(vcpu
);
6289 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6290 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
6291 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
6292 mmu_reset_needed
= 1;
6294 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6296 if (mmu_reset_needed
)
6297 kvm_mmu_reset_context(vcpu
);
6299 max_bits
= KVM_NR_INTERRUPTS
;
6300 pending_vec
= find_first_bit(
6301 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
6302 if (pending_vec
< max_bits
) {
6303 kvm_queue_interrupt(vcpu
, pending_vec
, false);
6304 pr_debug("Set back pending irq %d\n", pending_vec
);
6307 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
6308 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
6309 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
6310 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
6311 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
6312 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
6314 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
6315 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
6317 update_cr8_intercept(vcpu
);
6319 /* Older userspace won't unhalt the vcpu on reset. */
6320 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
6321 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
6323 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6325 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6330 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
6331 struct kvm_guest_debug
*dbg
)
6333 unsigned long rflags
;
6336 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
6338 if (vcpu
->arch
.exception
.pending
)
6340 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
6341 kvm_queue_exception(vcpu
, DB_VECTOR
);
6343 kvm_queue_exception(vcpu
, BP_VECTOR
);
6347 * Read rflags as long as potentially injected trace flags are still
6350 rflags
= kvm_get_rflags(vcpu
);
6352 vcpu
->guest_debug
= dbg
->control
;
6353 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
6354 vcpu
->guest_debug
= 0;
6356 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
6357 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
6358 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
6359 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
6361 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
6362 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
6364 kvm_update_dr7(vcpu
);
6366 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
6367 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
6368 get_segment_base(vcpu
, VCPU_SREG_CS
);
6371 * Trigger an rflags update that will inject or remove the trace
6374 kvm_set_rflags(vcpu
, rflags
);
6376 kvm_x86_ops
->update_db_bp_intercept(vcpu
);
6386 * Translate a guest virtual address to a guest physical address.
6388 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
6389 struct kvm_translation
*tr
)
6391 unsigned long vaddr
= tr
->linear_address
;
6395 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6396 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
6397 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6398 tr
->physical_address
= gpa
;
6399 tr
->valid
= gpa
!= UNMAPPED_GVA
;
6406 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
6408 struct i387_fxsave_struct
*fxsave
=
6409 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
6411 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
6412 fpu
->fcw
= fxsave
->cwd
;
6413 fpu
->fsw
= fxsave
->swd
;
6414 fpu
->ftwx
= fxsave
->twd
;
6415 fpu
->last_opcode
= fxsave
->fop
;
6416 fpu
->last_ip
= fxsave
->rip
;
6417 fpu
->last_dp
= fxsave
->rdp
;
6418 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
6423 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
6425 struct i387_fxsave_struct
*fxsave
=
6426 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
6428 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
6429 fxsave
->cwd
= fpu
->fcw
;
6430 fxsave
->swd
= fpu
->fsw
;
6431 fxsave
->twd
= fpu
->ftwx
;
6432 fxsave
->fop
= fpu
->last_opcode
;
6433 fxsave
->rip
= fpu
->last_ip
;
6434 fxsave
->rdp
= fpu
->last_dp
;
6435 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
6440 int fx_init(struct kvm_vcpu
*vcpu
)
6444 err
= fpu_alloc(&vcpu
->arch
.guest_fpu
);
6448 fpu_finit(&vcpu
->arch
.guest_fpu
);
6451 * Ensure guest xcr0 is valid for loading
6453 vcpu
->arch
.xcr0
= XSTATE_FP
;
6455 vcpu
->arch
.cr0
|= X86_CR0_ET
;
6459 EXPORT_SYMBOL_GPL(fx_init
);
6461 static void fx_free(struct kvm_vcpu
*vcpu
)
6463 fpu_free(&vcpu
->arch
.guest_fpu
);
6466 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
6468 if (vcpu
->guest_fpu_loaded
)
6472 * Restore all possible states in the guest,
6473 * and assume host would use all available bits.
6474 * Guest xcr0 would be loaded later.
6476 kvm_put_guest_xcr0(vcpu
);
6477 vcpu
->guest_fpu_loaded
= 1;
6478 __kernel_fpu_begin();
6479 fpu_restore_checking(&vcpu
->arch
.guest_fpu
);
6483 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
6485 kvm_put_guest_xcr0(vcpu
);
6487 if (!vcpu
->guest_fpu_loaded
)
6490 vcpu
->guest_fpu_loaded
= 0;
6491 fpu_save_init(&vcpu
->arch
.guest_fpu
);
6493 ++vcpu
->stat
.fpu_reload
;
6494 kvm_make_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
);
6498 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
6500 kvmclock_reset(vcpu
);
6502 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
6504 kvm_x86_ops
->vcpu_free(vcpu
);
6507 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
6510 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
6511 printk_once(KERN_WARNING
6512 "kvm: SMP vm created on host with unstable TSC; "
6513 "guest TSC will not be reliable\n");
6514 return kvm_x86_ops
->vcpu_create(kvm
, id
);
6517 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
6521 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
6522 r
= vcpu_load(vcpu
);
6525 kvm_vcpu_reset(vcpu
);
6526 r
= kvm_mmu_setup(vcpu
);
6532 int kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
6535 struct msr_data msr
;
6537 r
= vcpu_load(vcpu
);
6541 msr
.index
= MSR_IA32_TSC
;
6542 msr
.host_initiated
= true;
6543 kvm_write_tsc(vcpu
, &msr
);
6549 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
6552 vcpu
->arch
.apf
.msr_val
= 0;
6554 r
= vcpu_load(vcpu
);
6556 kvm_mmu_unload(vcpu
);
6560 kvm_x86_ops
->vcpu_free(vcpu
);
6563 void kvm_vcpu_reset(struct kvm_vcpu
*vcpu
)
6565 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
6566 vcpu
->arch
.nmi_pending
= 0;
6567 vcpu
->arch
.nmi_injected
= false;
6569 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
6570 vcpu
->arch
.dr6
= DR6_FIXED_1
;
6571 vcpu
->arch
.dr7
= DR7_FIXED_1
;
6572 kvm_update_dr7(vcpu
);
6574 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6575 vcpu
->arch
.apf
.msr_val
= 0;
6576 vcpu
->arch
.st
.msr_val
= 0;
6578 kvmclock_reset(vcpu
);
6580 kvm_clear_async_pf_completion_queue(vcpu
);
6581 kvm_async_pf_hash_reset(vcpu
);
6582 vcpu
->arch
.apf
.halted
= false;
6584 kvm_pmu_reset(vcpu
);
6586 memset(vcpu
->arch
.regs
, 0, sizeof(vcpu
->arch
.regs
));
6587 vcpu
->arch
.regs_avail
= ~0;
6588 vcpu
->arch
.regs_dirty
= ~0;
6590 kvm_x86_ops
->vcpu_reset(vcpu
);
6593 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu
*vcpu
, unsigned int vector
)
6595 struct kvm_segment cs
;
6597 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6598 cs
.selector
= vector
<< 8;
6599 cs
.base
= vector
<< 12;
6600 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6601 kvm_rip_write(vcpu
, 0);
6604 int kvm_arch_hardware_enable(void *garbage
)
6607 struct kvm_vcpu
*vcpu
;
6612 bool stable
, backwards_tsc
= false;
6614 kvm_shared_msr_cpu_online();
6615 ret
= kvm_x86_ops
->hardware_enable(garbage
);
6619 local_tsc
= native_read_tsc();
6620 stable
= !check_tsc_unstable();
6621 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6622 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6623 if (!stable
&& vcpu
->cpu
== smp_processor_id())
6624 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
6625 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
6626 backwards_tsc
= true;
6627 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
6628 max_tsc
= vcpu
->arch
.last_host_tsc
;
6634 * Sometimes, even reliable TSCs go backwards. This happens on
6635 * platforms that reset TSC during suspend or hibernate actions, but
6636 * maintain synchronization. We must compensate. Fortunately, we can
6637 * detect that condition here, which happens early in CPU bringup,
6638 * before any KVM threads can be running. Unfortunately, we can't
6639 * bring the TSCs fully up to date with real time, as we aren't yet far
6640 * enough into CPU bringup that we know how much real time has actually
6641 * elapsed; our helper function, get_kernel_ns() will be using boot
6642 * variables that haven't been updated yet.
6644 * So we simply find the maximum observed TSC above, then record the
6645 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6646 * the adjustment will be applied. Note that we accumulate
6647 * adjustments, in case multiple suspend cycles happen before some VCPU
6648 * gets a chance to run again. In the event that no KVM threads get a
6649 * chance to run, we will miss the entire elapsed period, as we'll have
6650 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6651 * loose cycle time. This isn't too big a deal, since the loss will be
6652 * uniform across all VCPUs (not to mention the scenario is extremely
6653 * unlikely). It is possible that a second hibernate recovery happens
6654 * much faster than a first, causing the observed TSC here to be
6655 * smaller; this would require additional padding adjustment, which is
6656 * why we set last_host_tsc to the local tsc observed here.
6658 * N.B. - this code below runs only on platforms with reliable TSC,
6659 * as that is the only way backwards_tsc is set above. Also note
6660 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6661 * have the same delta_cyc adjustment applied if backwards_tsc
6662 * is detected. Note further, this adjustment is only done once,
6663 * as we reset last_host_tsc on all VCPUs to stop this from being
6664 * called multiple times (one for each physical CPU bringup).
6666 * Platforms with unreliable TSCs don't have to deal with this, they
6667 * will be compensated by the logic in vcpu_load, which sets the TSC to
6668 * catchup mode. This will catchup all VCPUs to real time, but cannot
6669 * guarantee that they stay in perfect synchronization.
6671 if (backwards_tsc
) {
6672 u64 delta_cyc
= max_tsc
- local_tsc
;
6673 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6674 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6675 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
6676 vcpu
->arch
.last_host_tsc
= local_tsc
;
6677 set_bit(KVM_REQ_MASTERCLOCK_UPDATE
,
6682 * We have to disable TSC offset matching.. if you were
6683 * booting a VM while issuing an S4 host suspend....
6684 * you may have some problem. Solving this issue is
6685 * left as an exercise to the reader.
6687 kvm
->arch
.last_tsc_nsec
= 0;
6688 kvm
->arch
.last_tsc_write
= 0;
6695 void kvm_arch_hardware_disable(void *garbage
)
6697 kvm_x86_ops
->hardware_disable(garbage
);
6698 drop_user_return_notifiers(garbage
);
6701 int kvm_arch_hardware_setup(void)
6703 return kvm_x86_ops
->hardware_setup();
6706 void kvm_arch_hardware_unsetup(void)
6708 kvm_x86_ops
->hardware_unsetup();
6711 void kvm_arch_check_processor_compat(void *rtn
)
6713 kvm_x86_ops
->check_processor_compatibility(rtn
);
6716 bool kvm_vcpu_compatible(struct kvm_vcpu
*vcpu
)
6718 return irqchip_in_kernel(vcpu
->kvm
) == (vcpu
->arch
.apic
!= NULL
);
6721 struct static_key kvm_no_apic_vcpu __read_mostly
;
6723 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
6729 BUG_ON(vcpu
->kvm
== NULL
);
6732 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
6733 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
6734 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6736 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
6738 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
6743 vcpu
->arch
.pio_data
= page_address(page
);
6745 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
6747 r
= kvm_mmu_create(vcpu
);
6749 goto fail_free_pio_data
;
6751 if (irqchip_in_kernel(kvm
)) {
6752 r
= kvm_create_lapic(vcpu
);
6754 goto fail_mmu_destroy
;
6756 static_key_slow_inc(&kvm_no_apic_vcpu
);
6758 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
6760 if (!vcpu
->arch
.mce_banks
) {
6762 goto fail_free_lapic
;
6764 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
6766 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
)) {
6768 goto fail_free_mce_banks
;
6773 goto fail_free_wbinvd_dirty_mask
;
6775 vcpu
->arch
.ia32_tsc_adjust_msr
= 0x0;
6776 vcpu
->arch
.pv_time_enabled
= false;
6777 kvm_async_pf_hash_reset(vcpu
);
6781 fail_free_wbinvd_dirty_mask
:
6782 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
6783 fail_free_mce_banks
:
6784 kfree(vcpu
->arch
.mce_banks
);
6786 kvm_free_lapic(vcpu
);
6788 kvm_mmu_destroy(vcpu
);
6790 free_page((unsigned long)vcpu
->arch
.pio_data
);
6795 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
6799 kvm_pmu_destroy(vcpu
);
6800 kfree(vcpu
->arch
.mce_banks
);
6801 kvm_free_lapic(vcpu
);
6802 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6803 kvm_mmu_destroy(vcpu
);
6804 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6805 free_page((unsigned long)vcpu
->arch
.pio_data
);
6806 if (!irqchip_in_kernel(vcpu
->kvm
))
6807 static_key_slow_dec(&kvm_no_apic_vcpu
);
6810 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
6815 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
6816 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
6818 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6819 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
6820 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6821 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
6822 &kvm
->arch
.irq_sources_bitmap
);
6824 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
6825 mutex_init(&kvm
->arch
.apic_map_lock
);
6826 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
6828 pvclock_update_vm_gtod_copy(kvm
);
6833 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
6836 r
= vcpu_load(vcpu
);
6838 kvm_mmu_unload(vcpu
);
6842 static void kvm_free_vcpus(struct kvm
*kvm
)
6845 struct kvm_vcpu
*vcpu
;
6848 * Unpin any mmu pages first.
6850 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6851 kvm_clear_async_pf_completion_queue(vcpu
);
6852 kvm_unload_vcpu_mmu(vcpu
);
6854 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6855 kvm_arch_vcpu_free(vcpu
);
6857 mutex_lock(&kvm
->lock
);
6858 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
6859 kvm
->vcpus
[i
] = NULL
;
6861 atomic_set(&kvm
->online_vcpus
, 0);
6862 mutex_unlock(&kvm
->lock
);
6865 void kvm_arch_sync_events(struct kvm
*kvm
)
6867 kvm_free_all_assigned_devices(kvm
);
6871 void kvm_arch_destroy_vm(struct kvm
*kvm
)
6873 if (current
->mm
== kvm
->mm
) {
6875 * Free memory regions allocated on behalf of userspace,
6876 * unless the the memory map has changed due to process exit
6879 struct kvm_userspace_memory_region mem
;
6880 memset(&mem
, 0, sizeof(mem
));
6881 mem
.slot
= APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
;
6882 kvm_set_memory_region(kvm
, &mem
);
6884 mem
.slot
= IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
;
6885 kvm_set_memory_region(kvm
, &mem
);
6887 mem
.slot
= TSS_PRIVATE_MEMSLOT
;
6888 kvm_set_memory_region(kvm
, &mem
);
6890 kvm_iommu_unmap_guest(kvm
);
6891 kfree(kvm
->arch
.vpic
);
6892 kfree(kvm
->arch
.vioapic
);
6893 kvm_free_vcpus(kvm
);
6894 if (kvm
->arch
.apic_access_page
)
6895 put_page(kvm
->arch
.apic_access_page
);
6896 if (kvm
->arch
.ept_identity_pagetable
)
6897 put_page(kvm
->arch
.ept_identity_pagetable
);
6898 kfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
6901 void kvm_arch_free_memslot(struct kvm_memory_slot
*free
,
6902 struct kvm_memory_slot
*dont
)
6906 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6907 if (!dont
|| free
->arch
.rmap
[i
] != dont
->arch
.rmap
[i
]) {
6908 kvm_kvfree(free
->arch
.rmap
[i
]);
6909 free
->arch
.rmap
[i
] = NULL
;
6914 if (!dont
|| free
->arch
.lpage_info
[i
- 1] !=
6915 dont
->arch
.lpage_info
[i
- 1]) {
6916 kvm_kvfree(free
->arch
.lpage_info
[i
- 1]);
6917 free
->arch
.lpage_info
[i
- 1] = NULL
;
6922 int kvm_arch_create_memslot(struct kvm_memory_slot
*slot
, unsigned long npages
)
6926 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6931 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
6932 slot
->base_gfn
, level
) + 1;
6934 slot
->arch
.rmap
[i
] =
6935 kvm_kvzalloc(lpages
* sizeof(*slot
->arch
.rmap
[i
]));
6936 if (!slot
->arch
.rmap
[i
])
6941 slot
->arch
.lpage_info
[i
- 1] = kvm_kvzalloc(lpages
*
6942 sizeof(*slot
->arch
.lpage_info
[i
- 1]));
6943 if (!slot
->arch
.lpage_info
[i
- 1])
6946 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
6947 slot
->arch
.lpage_info
[i
- 1][0].write_count
= 1;
6948 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
6949 slot
->arch
.lpage_info
[i
- 1][lpages
- 1].write_count
= 1;
6950 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
6952 * If the gfn and userspace address are not aligned wrt each
6953 * other, or if explicitly asked to, disable large page
6954 * support for this slot
6956 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
6957 !kvm_largepages_enabled()) {
6960 for (j
= 0; j
< lpages
; ++j
)
6961 slot
->arch
.lpage_info
[i
- 1][j
].write_count
= 1;
6968 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6969 kvm_kvfree(slot
->arch
.rmap
[i
]);
6970 slot
->arch
.rmap
[i
] = NULL
;
6974 kvm_kvfree(slot
->arch
.lpage_info
[i
- 1]);
6975 slot
->arch
.lpage_info
[i
- 1] = NULL
;
6980 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
6981 struct kvm_memory_slot
*memslot
,
6982 struct kvm_userspace_memory_region
*mem
,
6983 enum kvm_mr_change change
)
6986 * Only private memory slots need to be mapped here since
6987 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
6989 if ((memslot
->id
>= KVM_USER_MEM_SLOTS
) && (change
== KVM_MR_CREATE
)) {
6990 unsigned long userspace_addr
;
6993 * MAP_SHARED to prevent internal slot pages from being moved
6996 userspace_addr
= vm_mmap(NULL
, 0, memslot
->npages
* PAGE_SIZE
,
6997 PROT_READ
| PROT_WRITE
,
6998 MAP_SHARED
| MAP_ANONYMOUS
, 0);
7000 if (IS_ERR((void *)userspace_addr
))
7001 return PTR_ERR((void *)userspace_addr
);
7003 memslot
->userspace_addr
= userspace_addr
;
7009 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
7010 struct kvm_userspace_memory_region
*mem
,
7011 const struct kvm_memory_slot
*old
,
7012 enum kvm_mr_change change
)
7015 int nr_mmu_pages
= 0;
7017 if ((mem
->slot
>= KVM_USER_MEM_SLOTS
) && (change
== KVM_MR_DELETE
)) {
7020 ret
= vm_munmap(old
->userspace_addr
,
7021 old
->npages
* PAGE_SIZE
);
7024 "kvm_vm_ioctl_set_memory_region: "
7025 "failed to munmap memory\n");
7028 if (!kvm
->arch
.n_requested_mmu_pages
)
7029 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
7032 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
7034 * Write protect all pages for dirty logging.
7035 * Existing largepage mappings are destroyed here and new ones will
7036 * not be created until the end of the logging.
7038 if ((change
!= KVM_MR_DELETE
) && (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
7039 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
7041 * If memory slot is created, or moved, we need to clear all
7044 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
7045 kvm_mmu_zap_mmio_sptes(kvm
);
7046 kvm_reload_remote_mmus(kvm
);
7050 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
7052 kvm_mmu_zap_all(kvm
);
7053 kvm_reload_remote_mmus(kvm
);
7056 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
7057 struct kvm_memory_slot
*slot
)
7059 kvm_arch_flush_shadow_all(kvm
);
7062 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
7064 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
7065 !vcpu
->arch
.apf
.halted
)
7066 || !list_empty_careful(&vcpu
->async_pf
.done
)
7067 || kvm_apic_has_events(vcpu
)
7068 || atomic_read(&vcpu
->arch
.nmi_queued
) ||
7069 (kvm_arch_interrupt_allowed(vcpu
) &&
7070 kvm_cpu_has_interrupt(vcpu
));
7073 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
7075 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
7078 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
7080 return kvm_x86_ops
->interrupt_allowed(vcpu
);
7083 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
7085 unsigned long current_rip
= kvm_rip_read(vcpu
) +
7086 get_segment_base(vcpu
, VCPU_SREG_CS
);
7088 return current_rip
== linear_rip
;
7090 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
7092 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
7094 unsigned long rflags
;
7096 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
7097 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
7098 rflags
&= ~X86_EFLAGS_TF
;
7101 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
7103 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
7105 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
7106 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
7107 rflags
|= X86_EFLAGS_TF
;
7108 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
7109 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7111 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
7113 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
7117 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
7118 is_error_page(work
->page
))
7121 r
= kvm_mmu_reload(vcpu
);
7125 if (!vcpu
->arch
.mmu
.direct_map
&&
7126 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
7129 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
7132 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
7134 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
7137 static inline u32
kvm_async_pf_next_probe(u32 key
)
7139 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
7142 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7144 u32 key
= kvm_async_pf_hash_fn(gfn
);
7146 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
7147 key
= kvm_async_pf_next_probe(key
);
7149 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
7152 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7155 u32 key
= kvm_async_pf_hash_fn(gfn
);
7157 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
7158 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
7159 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
7160 key
= kvm_async_pf_next_probe(key
);
7165 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7167 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
7170 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7174 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
7176 vcpu
->arch
.apf
.gfns
[i
] = ~0;
7178 j
= kvm_async_pf_next_probe(j
);
7179 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
7181 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
7183 * k lies cyclically in ]i,j]
7185 * |....j i.k.| or |.k..j i...|
7187 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
7188 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
7193 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
7196 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
7200 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
7201 struct kvm_async_pf
*work
)
7203 struct x86_exception fault
;
7205 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
7206 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
7208 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
7209 (vcpu
->arch
.apf
.send_user_only
&&
7210 kvm_x86_ops
->get_cpl(vcpu
) == 0))
7211 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
7212 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
7213 fault
.vector
= PF_VECTOR
;
7214 fault
.error_code_valid
= true;
7215 fault
.error_code
= 0;
7216 fault
.nested_page_fault
= false;
7217 fault
.address
= work
->arch
.token
;
7218 kvm_inject_page_fault(vcpu
, &fault
);
7222 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
7223 struct kvm_async_pf
*work
)
7225 struct x86_exception fault
;
7227 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
7228 if (is_error_page(work
->page
))
7229 work
->arch
.token
= ~0; /* broadcast wakeup */
7231 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
7233 if ((vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) &&
7234 !apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
7235 fault
.vector
= PF_VECTOR
;
7236 fault
.error_code_valid
= true;
7237 fault
.error_code
= 0;
7238 fault
.nested_page_fault
= false;
7239 fault
.address
= work
->arch
.token
;
7240 kvm_inject_page_fault(vcpu
, &fault
);
7242 vcpu
->arch
.apf
.halted
= false;
7243 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7246 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
7248 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
7251 return !kvm_event_needs_reinjection(vcpu
) &&
7252 kvm_x86_ops
->interrupt_allowed(vcpu
);
7255 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
7256 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
7257 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
7258 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
7259 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
7260 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
7261 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
7262 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
7263 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
7264 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
7265 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
7266 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);