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 <trace/events/kvm.h>
51 #define CREATE_TRACE_POINTS
54 #include <asm/debugreg.h>
60 #include <asm/fpu-internal.h> /* Ugh! */
62 #include <asm/pvclock.h>
63 #include <asm/div64.h>
65 #define MAX_IO_MSRS 256
66 #define KVM_MAX_MCE_BANKS 32
67 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
69 #define emul_to_vcpu(ctxt) \
70 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
73 * - enable syscall per default because its emulated by KVM
74 * - enable LME and LMA per default on 64 bit KVM
78 u64 __read_mostly efer_reserved_bits
= ~((u64
)(EFER_SCE
| EFER_LME
| EFER_LMA
));
80 static u64 __read_mostly efer_reserved_bits
= ~((u64
)EFER_SCE
);
83 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
84 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
86 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
);
87 static void process_nmi(struct kvm_vcpu
*vcpu
);
89 struct kvm_x86_ops
*kvm_x86_ops
;
90 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
92 static bool ignore_msrs
= 0;
93 module_param(ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
95 bool kvm_has_tsc_control
;
96 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
97 u32 kvm_max_guest_tsc_khz
;
98 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
100 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
101 static u32 tsc_tolerance_ppm
= 250;
102 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
104 #define KVM_NR_SHARED_MSRS 16
106 struct kvm_shared_msrs_global
{
108 u32 msrs
[KVM_NR_SHARED_MSRS
];
111 struct kvm_shared_msrs
{
112 struct user_return_notifier urn
;
114 struct kvm_shared_msr_values
{
117 } values
[KVM_NR_SHARED_MSRS
];
120 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
121 static DEFINE_PER_CPU(struct kvm_shared_msrs
, shared_msrs
);
123 struct kvm_stats_debugfs_item debugfs_entries
[] = {
124 { "pf_fixed", VCPU_STAT(pf_fixed
) },
125 { "pf_guest", VCPU_STAT(pf_guest
) },
126 { "tlb_flush", VCPU_STAT(tlb_flush
) },
127 { "invlpg", VCPU_STAT(invlpg
) },
128 { "exits", VCPU_STAT(exits
) },
129 { "io_exits", VCPU_STAT(io_exits
) },
130 { "mmio_exits", VCPU_STAT(mmio_exits
) },
131 { "signal_exits", VCPU_STAT(signal_exits
) },
132 { "irq_window", VCPU_STAT(irq_window_exits
) },
133 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
134 { "halt_exits", VCPU_STAT(halt_exits
) },
135 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
136 { "hypercalls", VCPU_STAT(hypercalls
) },
137 { "request_irq", VCPU_STAT(request_irq_exits
) },
138 { "irq_exits", VCPU_STAT(irq_exits
) },
139 { "host_state_reload", VCPU_STAT(host_state_reload
) },
140 { "efer_reload", VCPU_STAT(efer_reload
) },
141 { "fpu_reload", VCPU_STAT(fpu_reload
) },
142 { "insn_emulation", VCPU_STAT(insn_emulation
) },
143 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
144 { "irq_injections", VCPU_STAT(irq_injections
) },
145 { "nmi_injections", VCPU_STAT(nmi_injections
) },
146 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
147 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
148 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
149 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
150 { "mmu_flooded", VM_STAT(mmu_flooded
) },
151 { "mmu_recycled", VM_STAT(mmu_recycled
) },
152 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
153 { "mmu_unsync", VM_STAT(mmu_unsync
) },
154 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
155 { "largepages", VM_STAT(lpages
) },
159 u64 __read_mostly host_xcr0
;
161 int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
163 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
166 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
167 vcpu
->arch
.apf
.gfns
[i
] = ~0;
170 static void kvm_on_user_return(struct user_return_notifier
*urn
)
173 struct kvm_shared_msrs
*locals
174 = container_of(urn
, struct kvm_shared_msrs
, urn
);
175 struct kvm_shared_msr_values
*values
;
177 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
178 values
= &locals
->values
[slot
];
179 if (values
->host
!= values
->curr
) {
180 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
181 values
->curr
= values
->host
;
184 locals
->registered
= false;
185 user_return_notifier_unregister(urn
);
188 static void shared_msr_update(unsigned slot
, u32 msr
)
190 struct kvm_shared_msrs
*smsr
;
193 smsr
= &__get_cpu_var(shared_msrs
);
194 /* only read, and nobody should modify it at this time,
195 * so don't need lock */
196 if (slot
>= shared_msrs_global
.nr
) {
197 printk(KERN_ERR
"kvm: invalid MSR slot!");
200 rdmsrl_safe(msr
, &value
);
201 smsr
->values
[slot
].host
= value
;
202 smsr
->values
[slot
].curr
= value
;
205 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
207 if (slot
>= shared_msrs_global
.nr
)
208 shared_msrs_global
.nr
= slot
+ 1;
209 shared_msrs_global
.msrs
[slot
] = msr
;
210 /* we need ensured the shared_msr_global have been updated */
213 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
215 static void kvm_shared_msr_cpu_online(void)
219 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
220 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
223 void kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
225 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
227 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
229 smsr
->values
[slot
].curr
= value
;
230 wrmsrl(shared_msrs_global
.msrs
[slot
], value
);
231 if (!smsr
->registered
) {
232 smsr
->urn
.on_user_return
= kvm_on_user_return
;
233 user_return_notifier_register(&smsr
->urn
);
234 smsr
->registered
= true;
237 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
239 static void drop_user_return_notifiers(void *ignore
)
241 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
243 if (smsr
->registered
)
244 kvm_on_user_return(&smsr
->urn
);
247 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
249 if (irqchip_in_kernel(vcpu
->kvm
))
250 return vcpu
->arch
.apic_base
;
252 return vcpu
->arch
.apic_base
;
254 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
256 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
258 /* TODO: reserve bits check */
259 if (irqchip_in_kernel(vcpu
->kvm
))
260 kvm_lapic_set_base(vcpu
, data
);
262 vcpu
->arch
.apic_base
= data
;
264 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
266 #define EXCPT_BENIGN 0
267 #define EXCPT_CONTRIBUTORY 1
270 static int exception_class(int vector
)
280 return EXCPT_CONTRIBUTORY
;
287 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
288 unsigned nr
, bool has_error
, u32 error_code
,
294 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
296 if (!vcpu
->arch
.exception
.pending
) {
298 vcpu
->arch
.exception
.pending
= true;
299 vcpu
->arch
.exception
.has_error_code
= has_error
;
300 vcpu
->arch
.exception
.nr
= nr
;
301 vcpu
->arch
.exception
.error_code
= error_code
;
302 vcpu
->arch
.exception
.reinject
= reinject
;
306 /* to check exception */
307 prev_nr
= vcpu
->arch
.exception
.nr
;
308 if (prev_nr
== DF_VECTOR
) {
309 /* triple fault -> shutdown */
310 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
313 class1
= exception_class(prev_nr
);
314 class2
= exception_class(nr
);
315 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
316 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
317 /* generate double fault per SDM Table 5-5 */
318 vcpu
->arch
.exception
.pending
= true;
319 vcpu
->arch
.exception
.has_error_code
= true;
320 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
321 vcpu
->arch
.exception
.error_code
= 0;
323 /* replace previous exception with a new one in a hope
324 that instruction re-execution will regenerate lost
329 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
331 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
333 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
335 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
337 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
339 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
341 void kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
344 kvm_inject_gp(vcpu
, 0);
346 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
348 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
350 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
352 ++vcpu
->stat
.pf_guest
;
353 vcpu
->arch
.cr2
= fault
->address
;
354 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
356 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
358 void kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
360 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
361 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
363 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
366 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
368 atomic_inc(&vcpu
->arch
.nmi_queued
);
369 kvm_make_request(KVM_REQ_NMI
, vcpu
);
371 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
373 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
375 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
377 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
379 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
381 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
383 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
386 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
387 * a #GP and return false.
389 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
391 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
393 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
396 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
399 * This function will be used to read from the physical memory of the currently
400 * running guest. The difference to kvm_read_guest_page is that this function
401 * can read from guest physical or from the guest's guest physical memory.
403 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
404 gfn_t ngfn
, void *data
, int offset
, int len
,
410 ngpa
= gfn_to_gpa(ngfn
);
411 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
);
412 if (real_gfn
== UNMAPPED_GVA
)
415 real_gfn
= gpa_to_gfn(real_gfn
);
417 return kvm_read_guest_page(vcpu
->kvm
, real_gfn
, data
, offset
, len
);
419 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
421 int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
422 void *data
, int offset
, int len
, u32 access
)
424 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
425 data
, offset
, len
, access
);
429 * Load the pae pdptrs. Return true is they are all valid.
431 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
433 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
434 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
437 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
439 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
440 offset
* sizeof(u64
), sizeof(pdpte
),
441 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
446 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
447 if (is_present_gpte(pdpte
[i
]) &&
448 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
455 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
456 __set_bit(VCPU_EXREG_PDPTR
,
457 (unsigned long *)&vcpu
->arch
.regs_avail
);
458 __set_bit(VCPU_EXREG_PDPTR
,
459 (unsigned long *)&vcpu
->arch
.regs_dirty
);
464 EXPORT_SYMBOL_GPL(load_pdptrs
);
466 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
468 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
474 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
477 if (!test_bit(VCPU_EXREG_PDPTR
,
478 (unsigned long *)&vcpu
->arch
.regs_avail
))
481 gfn
= (kvm_read_cr3(vcpu
) & ~31u) >> PAGE_SHIFT
;
482 offset
= (kvm_read_cr3(vcpu
) & ~31u) & (PAGE_SIZE
- 1);
483 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
484 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
487 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
493 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
495 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
496 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
|
497 X86_CR0_CD
| X86_CR0_NW
;
502 if (cr0
& 0xffffffff00000000UL
)
506 cr0
&= ~CR0_RESERVED_BITS
;
508 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
511 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
514 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
516 if ((vcpu
->arch
.efer
& EFER_LME
)) {
521 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
526 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
531 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
534 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
536 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
537 kvm_clear_async_pf_completion_queue(vcpu
);
538 kvm_async_pf_hash_reset(vcpu
);
541 if ((cr0
^ old_cr0
) & update_bits
)
542 kvm_mmu_reset_context(vcpu
);
545 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
547 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
549 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
551 EXPORT_SYMBOL_GPL(kvm_lmsw
);
553 int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
557 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
558 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
561 if (kvm_x86_ops
->get_cpl(vcpu
) != 0)
563 if (!(xcr0
& XSTATE_FP
))
565 if ((xcr0
& XSTATE_YMM
) && !(xcr0
& XSTATE_SSE
))
567 if (xcr0
& ~host_xcr0
)
569 vcpu
->arch
.xcr0
= xcr0
;
570 vcpu
->guest_xcr0_loaded
= 0;
574 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
576 if (__kvm_set_xcr(vcpu
, index
, xcr
)) {
577 kvm_inject_gp(vcpu
, 0);
582 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
584 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
586 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
587 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
|
588 X86_CR4_PAE
| X86_CR4_SMEP
;
589 if (cr4
& CR4_RESERVED_BITS
)
592 if (!guest_cpuid_has_xsave(vcpu
) && (cr4
& X86_CR4_OSXSAVE
))
595 if (!guest_cpuid_has_smep(vcpu
) && (cr4
& X86_CR4_SMEP
))
598 if (!guest_cpuid_has_fsgsbase(vcpu
) && (cr4
& X86_CR4_RDWRGSFS
))
601 if (is_long_mode(vcpu
)) {
602 if (!(cr4
& X86_CR4_PAE
))
604 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
605 && ((cr4
^ old_cr4
) & pdptr_bits
)
606 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
610 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
611 if (!guest_cpuid_has_pcid(vcpu
))
614 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
615 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
619 if (kvm_x86_ops
->set_cr4(vcpu
, cr4
))
622 if (((cr4
^ old_cr4
) & pdptr_bits
) ||
623 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
624 kvm_mmu_reset_context(vcpu
);
626 if ((cr4
^ old_cr4
) & X86_CR4_OSXSAVE
)
627 kvm_update_cpuid(vcpu
);
631 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
633 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
635 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
636 kvm_mmu_sync_roots(vcpu
);
637 kvm_mmu_flush_tlb(vcpu
);
641 if (is_long_mode(vcpu
)) {
642 if (kvm_read_cr4(vcpu
) & X86_CR4_PCIDE
) {
643 if (cr3
& CR3_PCID_ENABLED_RESERVED_BITS
)
646 if (cr3
& CR3_L_MODE_RESERVED_BITS
)
650 if (cr3
& CR3_PAE_RESERVED_BITS
)
652 if (is_paging(vcpu
) &&
653 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
657 * We don't check reserved bits in nonpae mode, because
658 * this isn't enforced, and VMware depends on this.
663 * Does the new cr3 value map to physical memory? (Note, we
664 * catch an invalid cr3 even in real-mode, because it would
665 * cause trouble later on when we turn on paging anyway.)
667 * A real CPU would silently accept an invalid cr3 and would
668 * attempt to use it - with largely undefined (and often hard
669 * to debug) behavior on the guest side.
671 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
673 vcpu
->arch
.cr3
= cr3
;
674 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
675 vcpu
->arch
.mmu
.new_cr3(vcpu
);
678 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
680 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
682 if (cr8
& CR8_RESERVED_BITS
)
684 if (irqchip_in_kernel(vcpu
->kvm
))
685 kvm_lapic_set_tpr(vcpu
, cr8
);
687 vcpu
->arch
.cr8
= cr8
;
690 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
692 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
694 if (irqchip_in_kernel(vcpu
->kvm
))
695 return kvm_lapic_get_cr8(vcpu
);
697 return vcpu
->arch
.cr8
;
699 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
701 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
705 vcpu
->arch
.db
[dr
] = val
;
706 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
707 vcpu
->arch
.eff_db
[dr
] = val
;
710 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
714 if (val
& 0xffffffff00000000ULL
)
716 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | DR6_FIXED_1
;
719 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
723 if (val
& 0xffffffff00000000ULL
)
725 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
726 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
727 kvm_x86_ops
->set_dr7(vcpu
, vcpu
->arch
.dr7
);
728 vcpu
->arch
.switch_db_regs
= (val
& DR7_BP_EN_MASK
);
736 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
740 res
= __kvm_set_dr(vcpu
, dr
, val
);
742 kvm_queue_exception(vcpu
, UD_VECTOR
);
744 kvm_inject_gp(vcpu
, 0);
748 EXPORT_SYMBOL_GPL(kvm_set_dr
);
750 static int _kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
754 *val
= vcpu
->arch
.db
[dr
];
757 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
761 *val
= vcpu
->arch
.dr6
;
764 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
768 *val
= vcpu
->arch
.dr7
;
775 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
777 if (_kvm_get_dr(vcpu
, dr
, val
)) {
778 kvm_queue_exception(vcpu
, UD_VECTOR
);
783 EXPORT_SYMBOL_GPL(kvm_get_dr
);
785 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
787 u32 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
791 err
= kvm_pmu_read_pmc(vcpu
, ecx
, &data
);
794 kvm_register_write(vcpu
, VCPU_REGS_RAX
, (u32
)data
);
795 kvm_register_write(vcpu
, VCPU_REGS_RDX
, data
>> 32);
798 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
801 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
802 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
804 * This list is modified at module load time to reflect the
805 * capabilities of the host cpu. This capabilities test skips MSRs that are
806 * kvm-specific. Those are put in the beginning of the list.
809 #define KVM_SAVE_MSRS_BEGIN 10
810 static u32 msrs_to_save
[] = {
811 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
812 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
813 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
814 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
816 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
819 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
821 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
824 static unsigned num_msrs_to_save
;
826 static u32 emulated_msrs
[] = {
827 MSR_IA32_TSCDEADLINE
,
828 MSR_IA32_MISC_ENABLE
,
833 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
835 u64 old_efer
= vcpu
->arch
.efer
;
837 if (efer
& efer_reserved_bits
)
841 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
844 if (efer
& EFER_FFXSR
) {
845 struct kvm_cpuid_entry2
*feat
;
847 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
848 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
)))
852 if (efer
& EFER_SVME
) {
853 struct kvm_cpuid_entry2
*feat
;
855 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
856 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
)))
861 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
863 kvm_x86_ops
->set_efer(vcpu
, efer
);
865 vcpu
->arch
.mmu
.base_role
.nxe
= (efer
& EFER_NX
) && !tdp_enabled
;
867 /* Update reserved bits */
868 if ((efer
^ old_efer
) & EFER_NX
)
869 kvm_mmu_reset_context(vcpu
);
874 void kvm_enable_efer_bits(u64 mask
)
876 efer_reserved_bits
&= ~mask
;
878 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
882 * Writes msr value into into the appropriate "register".
883 * Returns 0 on success, non-0 otherwise.
884 * Assumes vcpu_load() was already called.
886 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
888 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
892 * Adapt set_msr() to msr_io()'s calling convention
894 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
896 return kvm_set_msr(vcpu
, index
, *data
);
899 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
903 struct pvclock_wall_clock wc
;
904 struct timespec boot
;
909 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
914 ++version
; /* first time write, random junk */
918 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
921 * The guest calculates current wall clock time by adding
922 * system time (updated by kvm_guest_time_update below) to the
923 * wall clock specified here. guest system time equals host
924 * system time for us, thus we must fill in host boot time here.
928 if (kvm
->arch
.kvmclock_offset
) {
929 struct timespec ts
= ns_to_timespec(kvm
->arch
.kvmclock_offset
);
930 boot
= timespec_sub(boot
, ts
);
932 wc
.sec
= boot
.tv_sec
;
933 wc
.nsec
= boot
.tv_nsec
;
934 wc
.version
= version
;
936 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
939 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
942 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
944 uint32_t quotient
, remainder
;
946 /* Don't try to replace with do_div(), this one calculates
947 * "(dividend << 32) / divisor" */
949 : "=a" (quotient
), "=d" (remainder
)
950 : "0" (0), "1" (dividend
), "r" (divisor
) );
954 static void kvm_get_time_scale(uint32_t scaled_khz
, uint32_t base_khz
,
955 s8
*pshift
, u32
*pmultiplier
)
962 tps64
= base_khz
* 1000LL;
963 scaled64
= scaled_khz
* 1000LL;
964 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
969 tps32
= (uint32_t)tps64
;
970 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
971 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
979 *pmultiplier
= div_frac(scaled64
, tps32
);
981 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
982 __func__
, base_khz
, scaled_khz
, shift
, *pmultiplier
);
985 static inline u64
get_kernel_ns(void)
989 WARN_ON(preemptible());
991 monotonic_to_bootbased(&ts
);
992 return timespec_to_ns(&ts
);
995 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
996 unsigned long max_tsc_khz
;
998 static inline u64
nsec_to_cycles(struct kvm_vcpu
*vcpu
, u64 nsec
)
1000 return pvclock_scale_delta(nsec
, vcpu
->arch
.virtual_tsc_mult
,
1001 vcpu
->arch
.virtual_tsc_shift
);
1004 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1006 u64 v
= (u64
)khz
* (1000000 + ppm
);
1011 static void kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 this_tsc_khz
)
1013 u32 thresh_lo
, thresh_hi
;
1014 int use_scaling
= 0;
1016 /* Compute a scale to convert nanoseconds in TSC cycles */
1017 kvm_get_time_scale(this_tsc_khz
, NSEC_PER_SEC
/ 1000,
1018 &vcpu
->arch
.virtual_tsc_shift
,
1019 &vcpu
->arch
.virtual_tsc_mult
);
1020 vcpu
->arch
.virtual_tsc_khz
= this_tsc_khz
;
1023 * Compute the variation in TSC rate which is acceptable
1024 * within the range of tolerance and decide if the
1025 * rate being applied is within that bounds of the hardware
1026 * rate. If so, no scaling or compensation need be done.
1028 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1029 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1030 if (this_tsc_khz
< thresh_lo
|| this_tsc_khz
> thresh_hi
) {
1031 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz
, thresh_lo
, thresh_hi
);
1034 kvm_x86_ops
->set_tsc_khz(vcpu
, this_tsc_khz
, use_scaling
);
1037 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1039 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1040 vcpu
->arch
.virtual_tsc_mult
,
1041 vcpu
->arch
.virtual_tsc_shift
);
1042 tsc
+= vcpu
->arch
.this_tsc_write
;
1046 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, u64 data
)
1048 struct kvm
*kvm
= vcpu
->kvm
;
1049 u64 offset
, ns
, elapsed
;
1050 unsigned long flags
;
1053 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1054 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1055 ns
= get_kernel_ns();
1056 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1058 /* n.b - signed multiplication and division required */
1059 usdiff
= data
- kvm
->arch
.last_tsc_write
;
1060 #ifdef CONFIG_X86_64
1061 usdiff
= (usdiff
* 1000) / vcpu
->arch
.virtual_tsc_khz
;
1063 /* do_div() only does unsigned */
1064 asm("idivl %2; xor %%edx, %%edx"
1066 : "A"(usdiff
* 1000), "rm"(vcpu
->arch
.virtual_tsc_khz
));
1068 do_div(elapsed
, 1000);
1074 * Special case: TSC write with a small delta (1 second) of virtual
1075 * cycle time against real time is interpreted as an attempt to
1076 * synchronize the CPU.
1078 * For a reliable TSC, we can match TSC offsets, and for an unstable
1079 * TSC, we add elapsed time in this computation. We could let the
1080 * compensation code attempt to catch up if we fall behind, but
1081 * it's better to try to match offsets from the beginning.
1083 if (usdiff
< USEC_PER_SEC
&&
1084 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1085 if (!check_tsc_unstable()) {
1086 offset
= kvm
->arch
.cur_tsc_offset
;
1087 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1089 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1091 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1092 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1096 * We split periods of matched TSC writes into generations.
1097 * For each generation, we track the original measured
1098 * nanosecond time, offset, and write, so if TSCs are in
1099 * sync, we can match exact offset, and if not, we can match
1100 * exact software computation in compute_guest_tsc()
1102 * These values are tracked in kvm->arch.cur_xxx variables.
1104 kvm
->arch
.cur_tsc_generation
++;
1105 kvm
->arch
.cur_tsc_nsec
= ns
;
1106 kvm
->arch
.cur_tsc_write
= data
;
1107 kvm
->arch
.cur_tsc_offset
= offset
;
1108 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1109 kvm
->arch
.cur_tsc_generation
, data
);
1113 * We also track th most recent recorded KHZ, write and time to
1114 * allow the matching interval to be extended at each write.
1116 kvm
->arch
.last_tsc_nsec
= ns
;
1117 kvm
->arch
.last_tsc_write
= data
;
1118 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1120 /* Reset of TSC must disable overshoot protection below */
1121 vcpu
->arch
.hv_clock
.tsc_timestamp
= 0;
1122 vcpu
->arch
.last_guest_tsc
= data
;
1124 /* Keep track of which generation this VCPU has synchronized to */
1125 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1126 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1127 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1129 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1130 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1133 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1135 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1137 unsigned long flags
;
1138 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1140 unsigned long this_tsc_khz
;
1141 s64 kernel_ns
, max_kernel_ns
;
1144 /* Keep irq disabled to prevent changes to the clock */
1145 local_irq_save(flags
);
1146 tsc_timestamp
= kvm_x86_ops
->read_l1_tsc(v
);
1147 kernel_ns
= get_kernel_ns();
1148 this_tsc_khz
= __get_cpu_var(cpu_tsc_khz
);
1149 if (unlikely(this_tsc_khz
== 0)) {
1150 local_irq_restore(flags
);
1151 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1156 * We may have to catch up the TSC to match elapsed wall clock
1157 * time for two reasons, even if kvmclock is used.
1158 * 1) CPU could have been running below the maximum TSC rate
1159 * 2) Broken TSC compensation resets the base at each VCPU
1160 * entry to avoid unknown leaps of TSC even when running
1161 * again on the same CPU. This may cause apparent elapsed
1162 * time to disappear, and the guest to stand still or run
1165 if (vcpu
->tsc_catchup
) {
1166 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1167 if (tsc
> tsc_timestamp
) {
1168 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1169 tsc_timestamp
= tsc
;
1173 local_irq_restore(flags
);
1175 if (!vcpu
->time_page
)
1179 * Time as measured by the TSC may go backwards when resetting the base
1180 * tsc_timestamp. The reason for this is that the TSC resolution is
1181 * higher than the resolution of the other clock scales. Thus, many
1182 * possible measurments of the TSC correspond to one measurement of any
1183 * other clock, and so a spread of values is possible. This is not a
1184 * problem for the computation of the nanosecond clock; with TSC rates
1185 * around 1GHZ, there can only be a few cycles which correspond to one
1186 * nanosecond value, and any path through this code will inevitably
1187 * take longer than that. However, with the kernel_ns value itself,
1188 * the precision may be much lower, down to HZ granularity. If the
1189 * first sampling of TSC against kernel_ns ends in the low part of the
1190 * range, and the second in the high end of the range, we can get:
1192 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1194 * As the sampling errors potentially range in the thousands of cycles,
1195 * it is possible such a time value has already been observed by the
1196 * guest. To protect against this, we must compute the system time as
1197 * observed by the guest and ensure the new system time is greater.
1200 if (vcpu
->hv_clock
.tsc_timestamp
) {
1201 max_kernel_ns
= vcpu
->last_guest_tsc
-
1202 vcpu
->hv_clock
.tsc_timestamp
;
1203 max_kernel_ns
= pvclock_scale_delta(max_kernel_ns
,
1204 vcpu
->hv_clock
.tsc_to_system_mul
,
1205 vcpu
->hv_clock
.tsc_shift
);
1206 max_kernel_ns
+= vcpu
->last_kernel_ns
;
1209 if (unlikely(vcpu
->hw_tsc_khz
!= this_tsc_khz
)) {
1210 kvm_get_time_scale(NSEC_PER_SEC
/ 1000, this_tsc_khz
,
1211 &vcpu
->hv_clock
.tsc_shift
,
1212 &vcpu
->hv_clock
.tsc_to_system_mul
);
1213 vcpu
->hw_tsc_khz
= this_tsc_khz
;
1216 if (max_kernel_ns
> kernel_ns
)
1217 kernel_ns
= max_kernel_ns
;
1219 /* With all the info we got, fill in the values */
1220 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1221 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1222 vcpu
->last_kernel_ns
= kernel_ns
;
1223 vcpu
->last_guest_tsc
= tsc_timestamp
;
1224 vcpu
->hv_clock
.flags
= 0;
1227 * The interface expects us to write an even number signaling that the
1228 * update is finished. Since the guest won't see the intermediate
1229 * state, we just increase by 2 at the end.
1231 vcpu
->hv_clock
.version
+= 2;
1233 shared_kaddr
= kmap_atomic(vcpu
->time_page
);
1235 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
1236 sizeof(vcpu
->hv_clock
));
1238 kunmap_atomic(shared_kaddr
);
1240 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
1244 static bool msr_mtrr_valid(unsigned msr
)
1247 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
1248 case MSR_MTRRfix64K_00000
:
1249 case MSR_MTRRfix16K_80000
:
1250 case MSR_MTRRfix16K_A0000
:
1251 case MSR_MTRRfix4K_C0000
:
1252 case MSR_MTRRfix4K_C8000
:
1253 case MSR_MTRRfix4K_D0000
:
1254 case MSR_MTRRfix4K_D8000
:
1255 case MSR_MTRRfix4K_E0000
:
1256 case MSR_MTRRfix4K_E8000
:
1257 case MSR_MTRRfix4K_F0000
:
1258 case MSR_MTRRfix4K_F8000
:
1259 case MSR_MTRRdefType
:
1260 case MSR_IA32_CR_PAT
:
1268 static bool valid_pat_type(unsigned t
)
1270 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1273 static bool valid_mtrr_type(unsigned t
)
1275 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
1278 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1282 if (!msr_mtrr_valid(msr
))
1285 if (msr
== MSR_IA32_CR_PAT
) {
1286 for (i
= 0; i
< 8; i
++)
1287 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
1290 } else if (msr
== MSR_MTRRdefType
) {
1293 return valid_mtrr_type(data
& 0xff);
1294 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
1295 for (i
= 0; i
< 8 ; i
++)
1296 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
1301 /* variable MTRRs */
1302 return valid_mtrr_type(data
& 0xff);
1305 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1307 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1309 if (!mtrr_valid(vcpu
, msr
, data
))
1312 if (msr
== MSR_MTRRdefType
) {
1313 vcpu
->arch
.mtrr_state
.def_type
= data
;
1314 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
1315 } else if (msr
== MSR_MTRRfix64K_00000
)
1317 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1318 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
1319 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1320 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
1321 else if (msr
== MSR_IA32_CR_PAT
)
1322 vcpu
->arch
.pat
= data
;
1323 else { /* Variable MTRRs */
1324 int idx
, is_mtrr_mask
;
1327 idx
= (msr
- 0x200) / 2;
1328 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1331 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1334 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1338 kvm_mmu_reset_context(vcpu
);
1342 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1344 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1345 unsigned bank_num
= mcg_cap
& 0xff;
1348 case MSR_IA32_MCG_STATUS
:
1349 vcpu
->arch
.mcg_status
= data
;
1351 case MSR_IA32_MCG_CTL
:
1352 if (!(mcg_cap
& MCG_CTL_P
))
1354 if (data
!= 0 && data
!= ~(u64
)0)
1356 vcpu
->arch
.mcg_ctl
= data
;
1359 if (msr
>= MSR_IA32_MC0_CTL
&&
1360 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1361 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1362 /* only 0 or all 1s can be written to IA32_MCi_CTL
1363 * some Linux kernels though clear bit 10 in bank 4 to
1364 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1365 * this to avoid an uncatched #GP in the guest
1367 if ((offset
& 0x3) == 0 &&
1368 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
1370 vcpu
->arch
.mce_banks
[offset
] = data
;
1378 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
1380 struct kvm
*kvm
= vcpu
->kvm
;
1381 int lm
= is_long_mode(vcpu
);
1382 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
1383 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
1384 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
1385 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
1386 u32 page_num
= data
& ~PAGE_MASK
;
1387 u64 page_addr
= data
& PAGE_MASK
;
1392 if (page_num
>= blob_size
)
1395 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
1400 if (kvm_write_guest(kvm
, page_addr
, page
, PAGE_SIZE
))
1409 static bool kvm_hv_hypercall_enabled(struct kvm
*kvm
)
1411 return kvm
->arch
.hv_hypercall
& HV_X64_MSR_HYPERCALL_ENABLE
;
1414 static bool kvm_hv_msr_partition_wide(u32 msr
)
1418 case HV_X64_MSR_GUEST_OS_ID
:
1419 case HV_X64_MSR_HYPERCALL
:
1427 static int set_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1429 struct kvm
*kvm
= vcpu
->kvm
;
1432 case HV_X64_MSR_GUEST_OS_ID
:
1433 kvm
->arch
.hv_guest_os_id
= data
;
1434 /* setting guest os id to zero disables hypercall page */
1435 if (!kvm
->arch
.hv_guest_os_id
)
1436 kvm
->arch
.hv_hypercall
&= ~HV_X64_MSR_HYPERCALL_ENABLE
;
1438 case HV_X64_MSR_HYPERCALL
: {
1443 /* if guest os id is not set hypercall should remain disabled */
1444 if (!kvm
->arch
.hv_guest_os_id
)
1446 if (!(data
& HV_X64_MSR_HYPERCALL_ENABLE
)) {
1447 kvm
->arch
.hv_hypercall
= data
;
1450 gfn
= data
>> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT
;
1451 addr
= gfn_to_hva(kvm
, gfn
);
1452 if (kvm_is_error_hva(addr
))
1454 kvm_x86_ops
->patch_hypercall(vcpu
, instructions
);
1455 ((unsigned char *)instructions
)[3] = 0xc3; /* ret */
1456 if (__copy_to_user((void __user
*)addr
, instructions
, 4))
1458 kvm
->arch
.hv_hypercall
= data
;
1462 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1463 "data 0x%llx\n", msr
, data
);
1469 static int set_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1472 case HV_X64_MSR_APIC_ASSIST_PAGE
: {
1475 if (!(data
& HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE
)) {
1476 vcpu
->arch
.hv_vapic
= data
;
1479 addr
= gfn_to_hva(vcpu
->kvm
, data
>>
1480 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT
);
1481 if (kvm_is_error_hva(addr
))
1483 if (__clear_user((void __user
*)addr
, PAGE_SIZE
))
1485 vcpu
->arch
.hv_vapic
= data
;
1488 case HV_X64_MSR_EOI
:
1489 return kvm_hv_vapic_msr_write(vcpu
, APIC_EOI
, data
);
1490 case HV_X64_MSR_ICR
:
1491 return kvm_hv_vapic_msr_write(vcpu
, APIC_ICR
, data
);
1492 case HV_X64_MSR_TPR
:
1493 return kvm_hv_vapic_msr_write(vcpu
, APIC_TASKPRI
, data
);
1495 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1496 "data 0x%llx\n", msr
, data
);
1503 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
1505 gpa_t gpa
= data
& ~0x3f;
1507 /* Bits 2:5 are reserved, Should be zero */
1511 vcpu
->arch
.apf
.msr_val
= data
;
1513 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
1514 kvm_clear_async_pf_completion_queue(vcpu
);
1515 kvm_async_pf_hash_reset(vcpu
);
1519 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
))
1522 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
1523 kvm_async_pf_wakeup_all(vcpu
);
1527 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
1529 if (vcpu
->arch
.time_page
) {
1530 kvm_release_page_dirty(vcpu
->arch
.time_page
);
1531 vcpu
->arch
.time_page
= NULL
;
1535 static void accumulate_steal_time(struct kvm_vcpu
*vcpu
)
1539 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1542 delta
= current
->sched_info
.run_delay
- vcpu
->arch
.st
.last_steal
;
1543 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
1544 vcpu
->arch
.st
.accum_steal
= delta
;
1547 static void record_steal_time(struct kvm_vcpu
*vcpu
)
1549 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1552 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1553 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
1556 vcpu
->arch
.st
.steal
.steal
+= vcpu
->arch
.st
.accum_steal
;
1557 vcpu
->arch
.st
.steal
.version
+= 2;
1558 vcpu
->arch
.st
.accum_steal
= 0;
1560 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1561 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
1564 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1570 return set_efer(vcpu
, data
);
1572 data
&= ~(u64
)0x40; /* ignore flush filter disable */
1573 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
1574 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
1576 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
1581 case MSR_FAM10H_MMIO_CONF_BASE
:
1583 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
1588 case MSR_AMD64_NB_CFG
:
1590 case MSR_IA32_DEBUGCTLMSR
:
1592 /* We support the non-activated case already */
1594 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
1595 /* Values other than LBR and BTF are vendor-specific,
1596 thus reserved and should throw a #GP */
1599 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1602 case MSR_IA32_UCODE_REV
:
1603 case MSR_IA32_UCODE_WRITE
:
1604 case MSR_VM_HSAVE_PA
:
1605 case MSR_AMD64_PATCH_LOADER
:
1607 case 0x200 ... 0x2ff:
1608 return set_msr_mtrr(vcpu
, msr
, data
);
1609 case MSR_IA32_APICBASE
:
1610 kvm_set_apic_base(vcpu
, data
);
1612 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1613 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
1614 case MSR_IA32_TSCDEADLINE
:
1615 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
1617 case MSR_IA32_MISC_ENABLE
:
1618 vcpu
->arch
.ia32_misc_enable_msr
= data
;
1620 case MSR_KVM_WALL_CLOCK_NEW
:
1621 case MSR_KVM_WALL_CLOCK
:
1622 vcpu
->kvm
->arch
.wall_clock
= data
;
1623 kvm_write_wall_clock(vcpu
->kvm
, data
);
1625 case MSR_KVM_SYSTEM_TIME_NEW
:
1626 case MSR_KVM_SYSTEM_TIME
: {
1627 kvmclock_reset(vcpu
);
1629 vcpu
->arch
.time
= data
;
1630 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1632 /* we verify if the enable bit is set... */
1636 /* ...but clean it before doing the actual write */
1637 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
1639 vcpu
->arch
.time_page
=
1640 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
1642 if (is_error_page(vcpu
->arch
.time_page
)) {
1643 kvm_release_page_clean(vcpu
->arch
.time_page
);
1644 vcpu
->arch
.time_page
= NULL
;
1648 case MSR_KVM_ASYNC_PF_EN
:
1649 if (kvm_pv_enable_async_pf(vcpu
, data
))
1652 case MSR_KVM_STEAL_TIME
:
1654 if (unlikely(!sched_info_on()))
1657 if (data
& KVM_STEAL_RESERVED_MASK
)
1660 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1661 data
& KVM_STEAL_VALID_BITS
))
1664 vcpu
->arch
.st
.msr_val
= data
;
1666 if (!(data
& KVM_MSR_ENABLED
))
1669 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
1672 accumulate_steal_time(vcpu
);
1675 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
1678 case MSR_KVM_PV_EOI_EN
:
1679 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
1683 case MSR_IA32_MCG_CTL
:
1684 case MSR_IA32_MCG_STATUS
:
1685 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1686 return set_msr_mce(vcpu
, msr
, data
);
1688 /* Performance counters are not protected by a CPUID bit,
1689 * so we should check all of them in the generic path for the sake of
1690 * cross vendor migration.
1691 * Writing a zero into the event select MSRs disables them,
1692 * which we perfectly emulate ;-). Any other value should be at least
1693 * reported, some guests depend on them.
1695 case MSR_K7_EVNTSEL0
:
1696 case MSR_K7_EVNTSEL1
:
1697 case MSR_K7_EVNTSEL2
:
1698 case MSR_K7_EVNTSEL3
:
1700 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
1701 "0x%x data 0x%llx\n", msr
, data
);
1703 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1704 * so we ignore writes to make it happy.
1706 case MSR_K7_PERFCTR0
:
1707 case MSR_K7_PERFCTR1
:
1708 case MSR_K7_PERFCTR2
:
1709 case MSR_K7_PERFCTR3
:
1710 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
1711 "0x%x data 0x%llx\n", msr
, data
);
1713 case MSR_P6_PERFCTR0
:
1714 case MSR_P6_PERFCTR1
:
1716 case MSR_P6_EVNTSEL0
:
1717 case MSR_P6_EVNTSEL1
:
1718 if (kvm_pmu_msr(vcpu
, msr
))
1719 return kvm_pmu_set_msr(vcpu
, msr
, data
);
1721 if (pr
|| data
!= 0)
1722 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
1723 "0x%x data 0x%llx\n", msr
, data
);
1725 case MSR_K7_CLK_CTL
:
1727 * Ignore all writes to this no longer documented MSR.
1728 * Writes are only relevant for old K7 processors,
1729 * all pre-dating SVM, but a recommended workaround from
1730 * AMD for these chips. It is possible to specify the
1731 * affected processor models on the command line, hence
1732 * the need to ignore the workaround.
1735 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
1736 if (kvm_hv_msr_partition_wide(msr
)) {
1738 mutex_lock(&vcpu
->kvm
->lock
);
1739 r
= set_msr_hyperv_pw(vcpu
, msr
, data
);
1740 mutex_unlock(&vcpu
->kvm
->lock
);
1743 return set_msr_hyperv(vcpu
, msr
, data
);
1745 case MSR_IA32_BBL_CR_CTL3
:
1746 /* Drop writes to this legacy MSR -- see rdmsr
1747 * counterpart for further detail.
1749 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n", msr
, data
);
1751 case MSR_AMD64_OSVW_ID_LENGTH
:
1752 if (!guest_cpuid_has_osvw(vcpu
))
1754 vcpu
->arch
.osvw
.length
= data
;
1756 case MSR_AMD64_OSVW_STATUS
:
1757 if (!guest_cpuid_has_osvw(vcpu
))
1759 vcpu
->arch
.osvw
.status
= data
;
1762 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
1763 return xen_hvm_config(vcpu
, data
);
1764 if (kvm_pmu_msr(vcpu
, msr
))
1765 return kvm_pmu_set_msr(vcpu
, msr
, data
);
1767 vcpu_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
1771 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
1778 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1782 * Reads an msr value (of 'msr_index') into 'pdata'.
1783 * Returns 0 on success, non-0 otherwise.
1784 * Assumes vcpu_load() was already called.
1786 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1788 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
1791 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1793 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1795 if (!msr_mtrr_valid(msr
))
1798 if (msr
== MSR_MTRRdefType
)
1799 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
1800 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
1801 else if (msr
== MSR_MTRRfix64K_00000
)
1803 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1804 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
1805 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1806 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
1807 else if (msr
== MSR_IA32_CR_PAT
)
1808 *pdata
= vcpu
->arch
.pat
;
1809 else { /* Variable MTRRs */
1810 int idx
, is_mtrr_mask
;
1813 idx
= (msr
- 0x200) / 2;
1814 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1817 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1820 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1827 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1830 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1831 unsigned bank_num
= mcg_cap
& 0xff;
1834 case MSR_IA32_P5_MC_ADDR
:
1835 case MSR_IA32_P5_MC_TYPE
:
1838 case MSR_IA32_MCG_CAP
:
1839 data
= vcpu
->arch
.mcg_cap
;
1841 case MSR_IA32_MCG_CTL
:
1842 if (!(mcg_cap
& MCG_CTL_P
))
1844 data
= vcpu
->arch
.mcg_ctl
;
1846 case MSR_IA32_MCG_STATUS
:
1847 data
= vcpu
->arch
.mcg_status
;
1850 if (msr
>= MSR_IA32_MC0_CTL
&&
1851 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1852 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1853 data
= vcpu
->arch
.mce_banks
[offset
];
1862 static int get_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1865 struct kvm
*kvm
= vcpu
->kvm
;
1868 case HV_X64_MSR_GUEST_OS_ID
:
1869 data
= kvm
->arch
.hv_guest_os_id
;
1871 case HV_X64_MSR_HYPERCALL
:
1872 data
= kvm
->arch
.hv_hypercall
;
1875 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
1883 static int get_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1888 case HV_X64_MSR_VP_INDEX
: {
1891 kvm_for_each_vcpu(r
, v
, vcpu
->kvm
)
1896 case HV_X64_MSR_EOI
:
1897 return kvm_hv_vapic_msr_read(vcpu
, APIC_EOI
, pdata
);
1898 case HV_X64_MSR_ICR
:
1899 return kvm_hv_vapic_msr_read(vcpu
, APIC_ICR
, pdata
);
1900 case HV_X64_MSR_TPR
:
1901 return kvm_hv_vapic_msr_read(vcpu
, APIC_TASKPRI
, pdata
);
1902 case HV_X64_MSR_APIC_ASSIST_PAGE
:
1903 data
= vcpu
->arch
.hv_vapic
;
1906 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
1913 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1918 case MSR_IA32_PLATFORM_ID
:
1919 case MSR_IA32_EBL_CR_POWERON
:
1920 case MSR_IA32_DEBUGCTLMSR
:
1921 case MSR_IA32_LASTBRANCHFROMIP
:
1922 case MSR_IA32_LASTBRANCHTOIP
:
1923 case MSR_IA32_LASTINTFROMIP
:
1924 case MSR_IA32_LASTINTTOIP
:
1927 case MSR_VM_HSAVE_PA
:
1928 case MSR_K7_EVNTSEL0
:
1929 case MSR_K7_PERFCTR0
:
1930 case MSR_K8_INT_PENDING_MSG
:
1931 case MSR_AMD64_NB_CFG
:
1932 case MSR_FAM10H_MMIO_CONF_BASE
:
1935 case MSR_P6_PERFCTR0
:
1936 case MSR_P6_PERFCTR1
:
1937 case MSR_P6_EVNTSEL0
:
1938 case MSR_P6_EVNTSEL1
:
1939 if (kvm_pmu_msr(vcpu
, msr
))
1940 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
1943 case MSR_IA32_UCODE_REV
:
1944 data
= 0x100000000ULL
;
1947 data
= 0x500 | KVM_NR_VAR_MTRR
;
1949 case 0x200 ... 0x2ff:
1950 return get_msr_mtrr(vcpu
, msr
, pdata
);
1951 case 0xcd: /* fsb frequency */
1955 * MSR_EBC_FREQUENCY_ID
1956 * Conservative value valid for even the basic CPU models.
1957 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1958 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1959 * and 266MHz for model 3, or 4. Set Core Clock
1960 * Frequency to System Bus Frequency Ratio to 1 (bits
1961 * 31:24) even though these are only valid for CPU
1962 * models > 2, however guests may end up dividing or
1963 * multiplying by zero otherwise.
1965 case MSR_EBC_FREQUENCY_ID
:
1968 case MSR_IA32_APICBASE
:
1969 data
= kvm_get_apic_base(vcpu
);
1971 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1972 return kvm_x2apic_msr_read(vcpu
, msr
, pdata
);
1974 case MSR_IA32_TSCDEADLINE
:
1975 data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
1977 case MSR_IA32_MISC_ENABLE
:
1978 data
= vcpu
->arch
.ia32_misc_enable_msr
;
1980 case MSR_IA32_PERF_STATUS
:
1981 /* TSC increment by tick */
1983 /* CPU multiplier */
1984 data
|= (((uint64_t)4ULL) << 40);
1987 data
= vcpu
->arch
.efer
;
1989 case MSR_KVM_WALL_CLOCK
:
1990 case MSR_KVM_WALL_CLOCK_NEW
:
1991 data
= vcpu
->kvm
->arch
.wall_clock
;
1993 case MSR_KVM_SYSTEM_TIME
:
1994 case MSR_KVM_SYSTEM_TIME_NEW
:
1995 data
= vcpu
->arch
.time
;
1997 case MSR_KVM_ASYNC_PF_EN
:
1998 data
= vcpu
->arch
.apf
.msr_val
;
2000 case MSR_KVM_STEAL_TIME
:
2001 data
= vcpu
->arch
.st
.msr_val
;
2003 case MSR_IA32_P5_MC_ADDR
:
2004 case MSR_IA32_P5_MC_TYPE
:
2005 case MSR_IA32_MCG_CAP
:
2006 case MSR_IA32_MCG_CTL
:
2007 case MSR_IA32_MCG_STATUS
:
2008 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
2009 return get_msr_mce(vcpu
, msr
, pdata
);
2010 case MSR_K7_CLK_CTL
:
2012 * Provide expected ramp-up count for K7. All other
2013 * are set to zero, indicating minimum divisors for
2016 * This prevents guest kernels on AMD host with CPU
2017 * type 6, model 8 and higher from exploding due to
2018 * the rdmsr failing.
2022 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2023 if (kvm_hv_msr_partition_wide(msr
)) {
2025 mutex_lock(&vcpu
->kvm
->lock
);
2026 r
= get_msr_hyperv_pw(vcpu
, msr
, pdata
);
2027 mutex_unlock(&vcpu
->kvm
->lock
);
2030 return get_msr_hyperv(vcpu
, msr
, pdata
);
2032 case MSR_IA32_BBL_CR_CTL3
:
2033 /* This legacy MSR exists but isn't fully documented in current
2034 * silicon. It is however accessed by winxp in very narrow
2035 * scenarios where it sets bit #19, itself documented as
2036 * a "reserved" bit. Best effort attempt to source coherent
2037 * read data here should the balance of the register be
2038 * interpreted by the guest:
2040 * L2 cache control register 3: 64GB range, 256KB size,
2041 * enabled, latency 0x1, configured
2045 case MSR_AMD64_OSVW_ID_LENGTH
:
2046 if (!guest_cpuid_has_osvw(vcpu
))
2048 data
= vcpu
->arch
.osvw
.length
;
2050 case MSR_AMD64_OSVW_STATUS
:
2051 if (!guest_cpuid_has_osvw(vcpu
))
2053 data
= vcpu
->arch
.osvw
.status
;
2056 if (kvm_pmu_msr(vcpu
, msr
))
2057 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
2059 vcpu_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
2062 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr
);
2070 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2073 * Read or write a bunch of msrs. All parameters are kernel addresses.
2075 * @return number of msrs set successfully.
2077 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2078 struct kvm_msr_entry
*entries
,
2079 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2080 unsigned index
, u64
*data
))
2084 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2085 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2086 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2088 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2094 * Read or write a bunch of msrs. Parameters are user addresses.
2096 * @return number of msrs set successfully.
2098 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2099 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2100 unsigned index
, u64
*data
),
2103 struct kvm_msrs msrs
;
2104 struct kvm_msr_entry
*entries
;
2109 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2113 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2116 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2117 entries
= memdup_user(user_msrs
->entries
, size
);
2118 if (IS_ERR(entries
)) {
2119 r
= PTR_ERR(entries
);
2123 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2128 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2139 int kvm_dev_ioctl_check_extension(long ext
)
2144 case KVM_CAP_IRQCHIP
:
2146 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2147 case KVM_CAP_SET_TSS_ADDR
:
2148 case KVM_CAP_EXT_CPUID
:
2149 case KVM_CAP_CLOCKSOURCE
:
2151 case KVM_CAP_NOP_IO_DELAY
:
2152 case KVM_CAP_MP_STATE
:
2153 case KVM_CAP_SYNC_MMU
:
2154 case KVM_CAP_USER_NMI
:
2155 case KVM_CAP_REINJECT_CONTROL
:
2156 case KVM_CAP_IRQ_INJECT_STATUS
:
2157 case KVM_CAP_ASSIGN_DEV_IRQ
:
2159 case KVM_CAP_IOEVENTFD
:
2161 case KVM_CAP_PIT_STATE2
:
2162 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2163 case KVM_CAP_XEN_HVM
:
2164 case KVM_CAP_ADJUST_CLOCK
:
2165 case KVM_CAP_VCPU_EVENTS
:
2166 case KVM_CAP_HYPERV
:
2167 case KVM_CAP_HYPERV_VAPIC
:
2168 case KVM_CAP_HYPERV_SPIN
:
2169 case KVM_CAP_PCI_SEGMENT
:
2170 case KVM_CAP_DEBUGREGS
:
2171 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2173 case KVM_CAP_ASYNC_PF
:
2174 case KVM_CAP_GET_TSC_KHZ
:
2175 case KVM_CAP_PCI_2_3
:
2176 case KVM_CAP_KVMCLOCK_CTRL
:
2179 case KVM_CAP_COALESCED_MMIO
:
2180 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
2183 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2185 case KVM_CAP_NR_VCPUS
:
2186 r
= KVM_SOFT_MAX_VCPUS
;
2188 case KVM_CAP_MAX_VCPUS
:
2191 case KVM_CAP_NR_MEMSLOTS
:
2192 r
= KVM_MEMORY_SLOTS
;
2194 case KVM_CAP_PV_MMU
: /* obsolete */
2198 r
= iommu_present(&pci_bus_type
);
2201 r
= KVM_MAX_MCE_BANKS
;
2206 case KVM_CAP_TSC_CONTROL
:
2207 r
= kvm_has_tsc_control
;
2209 case KVM_CAP_TSC_DEADLINE_TIMER
:
2210 r
= boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER
);
2220 long kvm_arch_dev_ioctl(struct file
*filp
,
2221 unsigned int ioctl
, unsigned long arg
)
2223 void __user
*argp
= (void __user
*)arg
;
2227 case KVM_GET_MSR_INDEX_LIST
: {
2228 struct kvm_msr_list __user
*user_msr_list
= argp
;
2229 struct kvm_msr_list msr_list
;
2233 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2236 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2237 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2240 if (n
< msr_list
.nmsrs
)
2243 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2244 num_msrs_to_save
* sizeof(u32
)))
2246 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2248 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2253 case KVM_GET_SUPPORTED_CPUID
: {
2254 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2255 struct kvm_cpuid2 cpuid
;
2258 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2260 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
2261 cpuid_arg
->entries
);
2266 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2271 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2274 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
2276 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
2288 static void wbinvd_ipi(void *garbage
)
2293 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2295 return vcpu
->kvm
->arch
.iommu_domain
&&
2296 !(vcpu
->kvm
->arch
.iommu_flags
& KVM_IOMMU_CACHE_COHERENCY
);
2299 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2301 /* Address WBINVD may be executed by guest */
2302 if (need_emulate_wbinvd(vcpu
)) {
2303 if (kvm_x86_ops
->has_wbinvd_exit())
2304 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2305 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2306 smp_call_function_single(vcpu
->cpu
,
2307 wbinvd_ipi
, NULL
, 1);
2310 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2312 /* Apply any externally detected TSC adjustments (due to suspend) */
2313 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2314 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2315 vcpu
->arch
.tsc_offset_adjustment
= 0;
2316 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
2319 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2320 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2321 native_read_tsc() - vcpu
->arch
.last_host_tsc
;
2323 mark_tsc_unstable("KVM discovered backwards TSC");
2324 if (check_tsc_unstable()) {
2325 u64 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
,
2326 vcpu
->arch
.last_guest_tsc
);
2327 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
2328 vcpu
->arch
.tsc_catchup
= 1;
2330 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2331 if (vcpu
->cpu
!= cpu
)
2332 kvm_migrate_timers(vcpu
);
2336 accumulate_steal_time(vcpu
);
2337 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2340 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2342 kvm_x86_ops
->vcpu_put(vcpu
);
2343 kvm_put_guest_fpu(vcpu
);
2344 vcpu
->arch
.last_host_tsc
= native_read_tsc();
2347 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
2348 struct kvm_lapic_state
*s
)
2350 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
2355 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
2356 struct kvm_lapic_state
*s
)
2358 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
2359 kvm_apic_post_state_restore(vcpu
);
2360 update_cr8_intercept(vcpu
);
2365 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2366 struct kvm_interrupt
*irq
)
2368 if (irq
->irq
< 0 || irq
->irq
>= 256)
2370 if (irqchip_in_kernel(vcpu
->kvm
))
2373 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
2374 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2379 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
2381 kvm_inject_nmi(vcpu
);
2386 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
2387 struct kvm_tpr_access_ctl
*tac
)
2391 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
2395 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
2399 unsigned bank_num
= mcg_cap
& 0xff, bank
;
2402 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
2404 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
2407 vcpu
->arch
.mcg_cap
= mcg_cap
;
2408 /* Init IA32_MCG_CTL to all 1s */
2409 if (mcg_cap
& MCG_CTL_P
)
2410 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
2411 /* Init IA32_MCi_CTL to all 1s */
2412 for (bank
= 0; bank
< bank_num
; bank
++)
2413 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
2418 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
2419 struct kvm_x86_mce
*mce
)
2421 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2422 unsigned bank_num
= mcg_cap
& 0xff;
2423 u64
*banks
= vcpu
->arch
.mce_banks
;
2425 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
2428 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2429 * reporting is disabled
2431 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
2432 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
2434 banks
+= 4 * mce
->bank
;
2436 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2437 * reporting is disabled for the bank
2439 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
2441 if (mce
->status
& MCI_STATUS_UC
) {
2442 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
2443 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
2444 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2447 if (banks
[1] & MCI_STATUS_VAL
)
2448 mce
->status
|= MCI_STATUS_OVER
;
2449 banks
[2] = mce
->addr
;
2450 banks
[3] = mce
->misc
;
2451 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
2452 banks
[1] = mce
->status
;
2453 kvm_queue_exception(vcpu
, MC_VECTOR
);
2454 } else if (!(banks
[1] & MCI_STATUS_VAL
)
2455 || !(banks
[1] & MCI_STATUS_UC
)) {
2456 if (banks
[1] & MCI_STATUS_VAL
)
2457 mce
->status
|= MCI_STATUS_OVER
;
2458 banks
[2] = mce
->addr
;
2459 banks
[3] = mce
->misc
;
2460 banks
[1] = mce
->status
;
2462 banks
[1] |= MCI_STATUS_OVER
;
2466 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
2467 struct kvm_vcpu_events
*events
)
2470 events
->exception
.injected
=
2471 vcpu
->arch
.exception
.pending
&&
2472 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
2473 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
2474 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
2475 events
->exception
.pad
= 0;
2476 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
2478 events
->interrupt
.injected
=
2479 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
2480 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
2481 events
->interrupt
.soft
= 0;
2482 events
->interrupt
.shadow
=
2483 kvm_x86_ops
->get_interrupt_shadow(vcpu
,
2484 KVM_X86_SHADOW_INT_MOV_SS
| KVM_X86_SHADOW_INT_STI
);
2486 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
2487 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
2488 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
2489 events
->nmi
.pad
= 0;
2491 events
->sipi_vector
= vcpu
->arch
.sipi_vector
;
2493 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
2494 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2495 | KVM_VCPUEVENT_VALID_SHADOW
);
2496 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
2499 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
2500 struct kvm_vcpu_events
*events
)
2502 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2503 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2504 | KVM_VCPUEVENT_VALID_SHADOW
))
2508 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
2509 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
2510 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
2511 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
2513 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
2514 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
2515 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
2516 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
2517 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
2518 events
->interrupt
.shadow
);
2520 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
2521 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
2522 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
2523 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
2525 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
)
2526 vcpu
->arch
.sipi_vector
= events
->sipi_vector
;
2528 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2533 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
2534 struct kvm_debugregs
*dbgregs
)
2536 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
2537 dbgregs
->dr6
= vcpu
->arch
.dr6
;
2538 dbgregs
->dr7
= vcpu
->arch
.dr7
;
2540 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
2543 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
2544 struct kvm_debugregs
*dbgregs
)
2549 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
2550 vcpu
->arch
.dr6
= dbgregs
->dr6
;
2551 vcpu
->arch
.dr7
= dbgregs
->dr7
;
2556 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
2557 struct kvm_xsave
*guest_xsave
)
2560 memcpy(guest_xsave
->region
,
2561 &vcpu
->arch
.guest_fpu
.state
->xsave
,
2564 memcpy(guest_xsave
->region
,
2565 &vcpu
->arch
.guest_fpu
.state
->fxsave
,
2566 sizeof(struct i387_fxsave_struct
));
2567 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
2572 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
2573 struct kvm_xsave
*guest_xsave
)
2576 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
2579 memcpy(&vcpu
->arch
.guest_fpu
.state
->xsave
,
2580 guest_xsave
->region
, xstate_size
);
2582 if (xstate_bv
& ~XSTATE_FPSSE
)
2584 memcpy(&vcpu
->arch
.guest_fpu
.state
->fxsave
,
2585 guest_xsave
->region
, sizeof(struct i387_fxsave_struct
));
2590 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
2591 struct kvm_xcrs
*guest_xcrs
)
2593 if (!cpu_has_xsave
) {
2594 guest_xcrs
->nr_xcrs
= 0;
2598 guest_xcrs
->nr_xcrs
= 1;
2599 guest_xcrs
->flags
= 0;
2600 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
2601 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
2604 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
2605 struct kvm_xcrs
*guest_xcrs
)
2612 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
2615 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
2616 /* Only support XCR0 currently */
2617 if (guest_xcrs
->xcrs
[0].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
2618 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
2619 guest_xcrs
->xcrs
[0].value
);
2628 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2629 * stopped by the hypervisor. This function will be called from the host only.
2630 * EINVAL is returned when the host attempts to set the flag for a guest that
2631 * does not support pv clocks.
2633 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
2635 struct pvclock_vcpu_time_info
*src
= &vcpu
->arch
.hv_clock
;
2636 if (!vcpu
->arch
.time_page
)
2638 src
->flags
|= PVCLOCK_GUEST_STOPPED
;
2639 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2643 long kvm_arch_vcpu_ioctl(struct file
*filp
,
2644 unsigned int ioctl
, unsigned long arg
)
2646 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2647 void __user
*argp
= (void __user
*)arg
;
2650 struct kvm_lapic_state
*lapic
;
2651 struct kvm_xsave
*xsave
;
2652 struct kvm_xcrs
*xcrs
;
2658 case KVM_GET_LAPIC
: {
2660 if (!vcpu
->arch
.apic
)
2662 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
2667 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
2671 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
2676 case KVM_SET_LAPIC
: {
2678 if (!vcpu
->arch
.apic
)
2680 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
2681 if (IS_ERR(u
.lapic
)) {
2682 r
= PTR_ERR(u
.lapic
);
2686 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
2692 case KVM_INTERRUPT
: {
2693 struct kvm_interrupt irq
;
2696 if (copy_from_user(&irq
, argp
, sizeof irq
))
2698 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2705 r
= kvm_vcpu_ioctl_nmi(vcpu
);
2711 case KVM_SET_CPUID
: {
2712 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2713 struct kvm_cpuid cpuid
;
2716 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2718 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2723 case KVM_SET_CPUID2
: {
2724 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2725 struct kvm_cpuid2 cpuid
;
2728 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2730 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
2731 cpuid_arg
->entries
);
2736 case KVM_GET_CPUID2
: {
2737 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2738 struct kvm_cpuid2 cpuid
;
2741 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2743 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
2744 cpuid_arg
->entries
);
2748 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2754 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
2757 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2759 case KVM_TPR_ACCESS_REPORTING
: {
2760 struct kvm_tpr_access_ctl tac
;
2763 if (copy_from_user(&tac
, argp
, sizeof tac
))
2765 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
2769 if (copy_to_user(argp
, &tac
, sizeof tac
))
2774 case KVM_SET_VAPIC_ADDR
: {
2775 struct kvm_vapic_addr va
;
2778 if (!irqchip_in_kernel(vcpu
->kvm
))
2781 if (copy_from_user(&va
, argp
, sizeof va
))
2784 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
2787 case KVM_X86_SETUP_MCE
: {
2791 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
2793 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
2796 case KVM_X86_SET_MCE
: {
2797 struct kvm_x86_mce mce
;
2800 if (copy_from_user(&mce
, argp
, sizeof mce
))
2802 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
2805 case KVM_GET_VCPU_EVENTS
: {
2806 struct kvm_vcpu_events events
;
2808 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
2811 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
2816 case KVM_SET_VCPU_EVENTS
: {
2817 struct kvm_vcpu_events events
;
2820 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
2823 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
2826 case KVM_GET_DEBUGREGS
: {
2827 struct kvm_debugregs dbgregs
;
2829 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
2832 if (copy_to_user(argp
, &dbgregs
,
2833 sizeof(struct kvm_debugregs
)))
2838 case KVM_SET_DEBUGREGS
: {
2839 struct kvm_debugregs dbgregs
;
2842 if (copy_from_user(&dbgregs
, argp
,
2843 sizeof(struct kvm_debugregs
)))
2846 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
2849 case KVM_GET_XSAVE
: {
2850 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
2855 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
2858 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
2863 case KVM_SET_XSAVE
: {
2864 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
2865 if (IS_ERR(u
.xsave
)) {
2866 r
= PTR_ERR(u
.xsave
);
2870 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
2873 case KVM_GET_XCRS
: {
2874 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
2879 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
2882 if (copy_to_user(argp
, u
.xcrs
,
2883 sizeof(struct kvm_xcrs
)))
2888 case KVM_SET_XCRS
: {
2889 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
2890 if (IS_ERR(u
.xcrs
)) {
2891 r
= PTR_ERR(u
.xcrs
);
2895 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
2898 case KVM_SET_TSC_KHZ
: {
2902 user_tsc_khz
= (u32
)arg
;
2904 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
2907 if (user_tsc_khz
== 0)
2908 user_tsc_khz
= tsc_khz
;
2910 kvm_set_tsc_khz(vcpu
, user_tsc_khz
);
2915 case KVM_GET_TSC_KHZ
: {
2916 r
= vcpu
->arch
.virtual_tsc_khz
;
2919 case KVM_KVMCLOCK_CTRL
: {
2920 r
= kvm_set_guest_paused(vcpu
);
2931 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
2933 return VM_FAULT_SIGBUS
;
2936 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
2940 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
2942 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
2946 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
2949 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
2953 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
2954 u32 kvm_nr_mmu_pages
)
2956 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
2959 mutex_lock(&kvm
->slots_lock
);
2960 spin_lock(&kvm
->mmu_lock
);
2962 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
2963 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
2965 spin_unlock(&kvm
->mmu_lock
);
2966 mutex_unlock(&kvm
->slots_lock
);
2970 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
2972 return kvm
->arch
.n_max_mmu_pages
;
2975 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
2980 switch (chip
->chip_id
) {
2981 case KVM_IRQCHIP_PIC_MASTER
:
2982 memcpy(&chip
->chip
.pic
,
2983 &pic_irqchip(kvm
)->pics
[0],
2984 sizeof(struct kvm_pic_state
));
2986 case KVM_IRQCHIP_PIC_SLAVE
:
2987 memcpy(&chip
->chip
.pic
,
2988 &pic_irqchip(kvm
)->pics
[1],
2989 sizeof(struct kvm_pic_state
));
2991 case KVM_IRQCHIP_IOAPIC
:
2992 r
= kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3001 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3006 switch (chip
->chip_id
) {
3007 case KVM_IRQCHIP_PIC_MASTER
:
3008 spin_lock(&pic_irqchip(kvm
)->lock
);
3009 memcpy(&pic_irqchip(kvm
)->pics
[0],
3011 sizeof(struct kvm_pic_state
));
3012 spin_unlock(&pic_irqchip(kvm
)->lock
);
3014 case KVM_IRQCHIP_PIC_SLAVE
:
3015 spin_lock(&pic_irqchip(kvm
)->lock
);
3016 memcpy(&pic_irqchip(kvm
)->pics
[1],
3018 sizeof(struct kvm_pic_state
));
3019 spin_unlock(&pic_irqchip(kvm
)->lock
);
3021 case KVM_IRQCHIP_IOAPIC
:
3022 r
= kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3028 kvm_pic_update_irq(pic_irqchip(kvm
));
3032 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3036 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3037 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
3038 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3042 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3046 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3047 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
3048 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
, 0);
3049 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3053 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3057 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3058 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3059 sizeof(ps
->channels
));
3060 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3061 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3062 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3066 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3068 int r
= 0, start
= 0;
3069 u32 prev_legacy
, cur_legacy
;
3070 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3071 prev_legacy
= kvm
->arch
.vpit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3072 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3073 if (!prev_legacy
&& cur_legacy
)
3075 memcpy(&kvm
->arch
.vpit
->pit_state
.channels
, &ps
->channels
,
3076 sizeof(kvm
->arch
.vpit
->pit_state
.channels
));
3077 kvm
->arch
.vpit
->pit_state
.flags
= ps
->flags
;
3078 kvm_pit_load_count(kvm
, 0, kvm
->arch
.vpit
->pit_state
.channels
[0].count
, start
);
3079 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3083 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3084 struct kvm_reinject_control
*control
)
3086 if (!kvm
->arch
.vpit
)
3088 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3089 kvm
->arch
.vpit
->pit_state
.reinject
= control
->pit_reinject
;
3090 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3095 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3096 * @kvm: kvm instance
3097 * @log: slot id and address to which we copy the log
3099 * We need to keep it in mind that VCPU threads can write to the bitmap
3100 * concurrently. So, to avoid losing data, we keep the following order for
3103 * 1. Take a snapshot of the bit and clear it if needed.
3104 * 2. Write protect the corresponding page.
3105 * 3. Flush TLB's if needed.
3106 * 4. Copy the snapshot to the userspace.
3108 * Between 2 and 3, the guest may write to the page using the remaining TLB
3109 * entry. This is not a problem because the page will be reported dirty at
3110 * step 4 using the snapshot taken before and step 3 ensures that successive
3111 * writes will be logged for the next call.
3113 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
3116 struct kvm_memory_slot
*memslot
;
3118 unsigned long *dirty_bitmap
;
3119 unsigned long *dirty_bitmap_buffer
;
3120 bool is_dirty
= false;
3122 mutex_lock(&kvm
->slots_lock
);
3125 if (log
->slot
>= KVM_MEMORY_SLOTS
)
3128 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
3130 dirty_bitmap
= memslot
->dirty_bitmap
;
3135 n
= kvm_dirty_bitmap_bytes(memslot
);
3137 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
3138 memset(dirty_bitmap_buffer
, 0, n
);
3140 spin_lock(&kvm
->mmu_lock
);
3142 for (i
= 0; i
< n
/ sizeof(long); i
++) {
3146 if (!dirty_bitmap
[i
])
3151 mask
= xchg(&dirty_bitmap
[i
], 0);
3152 dirty_bitmap_buffer
[i
] = mask
;
3154 offset
= i
* BITS_PER_LONG
;
3155 kvm_mmu_write_protect_pt_masked(kvm
, memslot
, offset
, mask
);
3158 kvm_flush_remote_tlbs(kvm
);
3160 spin_unlock(&kvm
->mmu_lock
);
3163 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
3168 mutex_unlock(&kvm
->slots_lock
);
3172 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
)
3174 if (!irqchip_in_kernel(kvm
))
3177 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
3178 irq_event
->irq
, irq_event
->level
);
3182 long kvm_arch_vm_ioctl(struct file
*filp
,
3183 unsigned int ioctl
, unsigned long arg
)
3185 struct kvm
*kvm
= filp
->private_data
;
3186 void __user
*argp
= (void __user
*)arg
;
3189 * This union makes it completely explicit to gcc-3.x
3190 * that these two variables' stack usage should be
3191 * combined, not added together.
3194 struct kvm_pit_state ps
;
3195 struct kvm_pit_state2 ps2
;
3196 struct kvm_pit_config pit_config
;
3200 case KVM_SET_TSS_ADDR
:
3201 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
3205 case KVM_SET_IDENTITY_MAP_ADDR
: {
3209 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
3211 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
3216 case KVM_SET_NR_MMU_PAGES
:
3217 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
3221 case KVM_GET_NR_MMU_PAGES
:
3222 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
3224 case KVM_CREATE_IRQCHIP
: {
3225 struct kvm_pic
*vpic
;
3227 mutex_lock(&kvm
->lock
);
3230 goto create_irqchip_unlock
;
3232 if (atomic_read(&kvm
->online_vcpus
))
3233 goto create_irqchip_unlock
;
3235 vpic
= kvm_create_pic(kvm
);
3237 r
= kvm_ioapic_init(kvm
);
3239 mutex_lock(&kvm
->slots_lock
);
3240 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3242 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3244 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3246 mutex_unlock(&kvm
->slots_lock
);
3248 goto create_irqchip_unlock
;
3251 goto create_irqchip_unlock
;
3253 kvm
->arch
.vpic
= vpic
;
3255 r
= kvm_setup_default_irq_routing(kvm
);
3257 mutex_lock(&kvm
->slots_lock
);
3258 mutex_lock(&kvm
->irq_lock
);
3259 kvm_ioapic_destroy(kvm
);
3260 kvm_destroy_pic(kvm
);
3261 mutex_unlock(&kvm
->irq_lock
);
3262 mutex_unlock(&kvm
->slots_lock
);
3264 create_irqchip_unlock
:
3265 mutex_unlock(&kvm
->lock
);
3268 case KVM_CREATE_PIT
:
3269 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
3271 case KVM_CREATE_PIT2
:
3273 if (copy_from_user(&u
.pit_config
, argp
,
3274 sizeof(struct kvm_pit_config
)))
3277 mutex_lock(&kvm
->slots_lock
);
3280 goto create_pit_unlock
;
3282 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
3286 mutex_unlock(&kvm
->slots_lock
);
3288 case KVM_GET_IRQCHIP
: {
3289 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3290 struct kvm_irqchip
*chip
;
3292 chip
= memdup_user(argp
, sizeof(*chip
));
3299 if (!irqchip_in_kernel(kvm
))
3300 goto get_irqchip_out
;
3301 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
3303 goto get_irqchip_out
;
3305 if (copy_to_user(argp
, chip
, sizeof *chip
))
3306 goto get_irqchip_out
;
3314 case KVM_SET_IRQCHIP
: {
3315 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3316 struct kvm_irqchip
*chip
;
3318 chip
= memdup_user(argp
, sizeof(*chip
));
3325 if (!irqchip_in_kernel(kvm
))
3326 goto set_irqchip_out
;
3327 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
3329 goto set_irqchip_out
;
3339 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
3342 if (!kvm
->arch
.vpit
)
3344 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
3348 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
3355 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
3358 if (!kvm
->arch
.vpit
)
3360 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
3366 case KVM_GET_PIT2
: {
3368 if (!kvm
->arch
.vpit
)
3370 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
3374 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
3379 case KVM_SET_PIT2
: {
3381 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
3384 if (!kvm
->arch
.vpit
)
3386 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
3392 case KVM_REINJECT_CONTROL
: {
3393 struct kvm_reinject_control control
;
3395 if (copy_from_user(&control
, argp
, sizeof(control
)))
3397 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
3403 case KVM_XEN_HVM_CONFIG
: {
3405 if (copy_from_user(&kvm
->arch
.xen_hvm_config
, argp
,
3406 sizeof(struct kvm_xen_hvm_config
)))
3409 if (kvm
->arch
.xen_hvm_config
.flags
)
3414 case KVM_SET_CLOCK
: {
3415 struct kvm_clock_data user_ns
;
3420 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
3428 local_irq_disable();
3429 now_ns
= get_kernel_ns();
3430 delta
= user_ns
.clock
- now_ns
;
3432 kvm
->arch
.kvmclock_offset
= delta
;
3435 case KVM_GET_CLOCK
: {
3436 struct kvm_clock_data user_ns
;
3439 local_irq_disable();
3440 now_ns
= get_kernel_ns();
3441 user_ns
.clock
= kvm
->arch
.kvmclock_offset
+ now_ns
;
3444 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
3447 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
3460 static void kvm_init_msr_list(void)
3465 /* skip the first msrs in the list. KVM-specific */
3466 for (i
= j
= KVM_SAVE_MSRS_BEGIN
; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
3467 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
3470 msrs_to_save
[j
] = msrs_to_save
[i
];
3473 num_msrs_to_save
= j
;
3476 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
3484 if (!(vcpu
->arch
.apic
&&
3485 !kvm_iodevice_write(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3486 && kvm_io_bus_write(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3497 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
3504 if (!(vcpu
->arch
.apic
&&
3505 !kvm_iodevice_read(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3506 && kvm_io_bus_read(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3508 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, *(u64
*)v
);
3518 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3519 struct kvm_segment
*var
, int seg
)
3521 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3524 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3525 struct kvm_segment
*var
, int seg
)
3527 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3530 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
)
3533 struct x86_exception exception
;
3535 BUG_ON(!mmu_is_nested(vcpu
));
3537 /* NPT walks are always user-walks */
3538 access
|= PFERR_USER_MASK
;
3539 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, &exception
);
3544 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
3545 struct x86_exception
*exception
)
3547 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3548 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3551 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
3552 struct x86_exception
*exception
)
3554 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3555 access
|= PFERR_FETCH_MASK
;
3556 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3559 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
3560 struct x86_exception
*exception
)
3562 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3563 access
|= PFERR_WRITE_MASK
;
3564 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3567 /* uses this to access any guest's mapped memory without checking CPL */
3568 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
3569 struct x86_exception
*exception
)
3571 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
3574 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
3575 struct kvm_vcpu
*vcpu
, u32 access
,
3576 struct x86_exception
*exception
)
3579 int r
= X86EMUL_CONTINUE
;
3582 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
3584 unsigned offset
= addr
& (PAGE_SIZE
-1);
3585 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3588 if (gpa
== UNMAPPED_GVA
)
3589 return X86EMUL_PROPAGATE_FAULT
;
3590 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
3592 r
= X86EMUL_IO_NEEDED
;
3604 /* used for instruction fetching */
3605 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3606 gva_t addr
, void *val
, unsigned int bytes
,
3607 struct x86_exception
*exception
)
3609 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3610 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3612 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
,
3613 access
| PFERR_FETCH_MASK
,
3617 int kvm_read_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3618 gva_t addr
, void *val
, unsigned int bytes
,
3619 struct x86_exception
*exception
)
3621 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3622 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3624 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
3627 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
3629 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3630 gva_t addr
, void *val
, unsigned int bytes
,
3631 struct x86_exception
*exception
)
3633 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3634 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, 0, exception
);
3637 int kvm_write_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3638 gva_t addr
, void *val
,
3640 struct x86_exception
*exception
)
3642 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3644 int r
= X86EMUL_CONTINUE
;
3647 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
3650 unsigned offset
= addr
& (PAGE_SIZE
-1);
3651 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3654 if (gpa
== UNMAPPED_GVA
)
3655 return X86EMUL_PROPAGATE_FAULT
;
3656 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
3658 r
= X86EMUL_IO_NEEDED
;
3669 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
3671 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
3672 gpa_t
*gpa
, struct x86_exception
*exception
,
3675 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3677 if (vcpu_match_mmio_gva(vcpu
, gva
) &&
3678 check_write_user_access(vcpu
, write
, access
,
3679 vcpu
->arch
.access
)) {
3680 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
3681 (gva
& (PAGE_SIZE
- 1));
3682 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
3687 access
|= PFERR_WRITE_MASK
;
3689 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3691 if (*gpa
== UNMAPPED_GVA
)
3694 /* For APIC access vmexit */
3695 if ((*gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
3698 if (vcpu_match_mmio_gpa(vcpu
, *gpa
)) {
3699 trace_vcpu_match_mmio(gva
, *gpa
, write
, true);
3706 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3707 const void *val
, int bytes
)
3711 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
3714 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
3718 struct read_write_emulator_ops
{
3719 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
3721 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3722 void *val
, int bytes
);
3723 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3724 int bytes
, void *val
);
3725 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3726 void *val
, int bytes
);
3730 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
3732 if (vcpu
->mmio_read_completed
) {
3733 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
3734 vcpu
->mmio_fragments
[0].gpa
, *(u64
*)val
);
3735 vcpu
->mmio_read_completed
= 0;
3742 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3743 void *val
, int bytes
)
3745 return !kvm_read_guest(vcpu
->kvm
, gpa
, val
, bytes
);
3748 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3749 void *val
, int bytes
)
3751 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
3754 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
3756 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
3757 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
3760 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3761 void *val
, int bytes
)
3763 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
3764 return X86EMUL_IO_NEEDED
;
3767 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3768 void *val
, int bytes
)
3770 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
3772 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, frag
->len
);
3773 return X86EMUL_CONTINUE
;
3776 static struct read_write_emulator_ops read_emultor
= {
3777 .read_write_prepare
= read_prepare
,
3778 .read_write_emulate
= read_emulate
,
3779 .read_write_mmio
= vcpu_mmio_read
,
3780 .read_write_exit_mmio
= read_exit_mmio
,
3783 static struct read_write_emulator_ops write_emultor
= {
3784 .read_write_emulate
= write_emulate
,
3785 .read_write_mmio
= write_mmio
,
3786 .read_write_exit_mmio
= write_exit_mmio
,
3790 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
3792 struct x86_exception
*exception
,
3793 struct kvm_vcpu
*vcpu
,
3794 struct read_write_emulator_ops
*ops
)
3798 bool write
= ops
->write
;
3799 struct kvm_mmio_fragment
*frag
;
3801 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
3804 return X86EMUL_PROPAGATE_FAULT
;
3806 /* For APIC access vmexit */
3810 if (ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
3811 return X86EMUL_CONTINUE
;
3815 * Is this MMIO handled locally?
3817 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
3818 if (handled
== bytes
)
3819 return X86EMUL_CONTINUE
;
3826 unsigned now
= min(bytes
, 8U);
3828 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
3837 return X86EMUL_CONTINUE
;
3840 int emulator_read_write(struct x86_emulate_ctxt
*ctxt
, unsigned long addr
,
3841 void *val
, unsigned int bytes
,
3842 struct x86_exception
*exception
,
3843 struct read_write_emulator_ops
*ops
)
3845 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3849 if (ops
->read_write_prepare
&&
3850 ops
->read_write_prepare(vcpu
, val
, bytes
))
3851 return X86EMUL_CONTINUE
;
3853 vcpu
->mmio_nr_fragments
= 0;
3855 /* Crossing a page boundary? */
3856 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
3859 now
= -addr
& ~PAGE_MASK
;
3860 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
3863 if (rc
!= X86EMUL_CONTINUE
)
3870 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
3872 if (rc
!= X86EMUL_CONTINUE
)
3875 if (!vcpu
->mmio_nr_fragments
)
3878 gpa
= vcpu
->mmio_fragments
[0].gpa
;
3880 vcpu
->mmio_needed
= 1;
3881 vcpu
->mmio_cur_fragment
= 0;
3883 vcpu
->run
->mmio
.len
= vcpu
->mmio_fragments
[0].len
;
3884 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
3885 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
3886 vcpu
->run
->mmio
.phys_addr
= gpa
;
3888 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
3891 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
3895 struct x86_exception
*exception
)
3897 return emulator_read_write(ctxt
, addr
, val
, bytes
,
3898 exception
, &read_emultor
);
3901 int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
3905 struct x86_exception
*exception
)
3907 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
3908 exception
, &write_emultor
);
3911 #define CMPXCHG_TYPE(t, ptr, old, new) \
3912 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3914 #ifdef CONFIG_X86_64
3915 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3917 # define CMPXCHG64(ptr, old, new) \
3918 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3921 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
3926 struct x86_exception
*exception
)
3928 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3934 /* guests cmpxchg8b have to be emulated atomically */
3935 if (bytes
> 8 || (bytes
& (bytes
- 1)))
3938 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
3940 if (gpa
== UNMAPPED_GVA
||
3941 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
3944 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
3947 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
3948 if (is_error_page(page
)) {
3949 kvm_release_page_clean(page
);
3953 kaddr
= kmap_atomic(page
);
3954 kaddr
+= offset_in_page(gpa
);
3957 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
3960 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
3963 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
3966 exchanged
= CMPXCHG64(kaddr
, old
, new);
3971 kunmap_atomic(kaddr
);
3972 kvm_release_page_dirty(page
);
3975 return X86EMUL_CMPXCHG_FAILED
;
3977 kvm_mmu_pte_write(vcpu
, gpa
, new, bytes
);
3979 return X86EMUL_CONTINUE
;
3982 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
3984 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
3987 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
3989 /* TODO: String I/O for in kernel device */
3992 if (vcpu
->arch
.pio
.in
)
3993 r
= kvm_io_bus_read(vcpu
->kvm
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
3994 vcpu
->arch
.pio
.size
, pd
);
3996 r
= kvm_io_bus_write(vcpu
->kvm
, KVM_PIO_BUS
,
3997 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
4002 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
4003 unsigned short port
, void *val
,
4004 unsigned int count
, bool in
)
4006 trace_kvm_pio(!in
, port
, size
, count
);
4008 vcpu
->arch
.pio
.port
= port
;
4009 vcpu
->arch
.pio
.in
= in
;
4010 vcpu
->arch
.pio
.count
= count
;
4011 vcpu
->arch
.pio
.size
= size
;
4013 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
4014 vcpu
->arch
.pio
.count
= 0;
4018 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
4019 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
4020 vcpu
->run
->io
.size
= size
;
4021 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
4022 vcpu
->run
->io
.count
= count
;
4023 vcpu
->run
->io
.port
= port
;
4028 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
4029 int size
, unsigned short port
, void *val
,
4032 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4035 if (vcpu
->arch
.pio
.count
)
4038 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
4041 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
4042 vcpu
->arch
.pio
.count
= 0;
4049 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
4050 int size
, unsigned short port
,
4051 const void *val
, unsigned int count
)
4053 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4055 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
4056 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
4059 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
4061 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
4064 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
4066 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
4069 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
4071 if (!need_emulate_wbinvd(vcpu
))
4072 return X86EMUL_CONTINUE
;
4074 if (kvm_x86_ops
->has_wbinvd_exit()) {
4075 int cpu
= get_cpu();
4077 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
4078 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
4079 wbinvd_ipi
, NULL
, 1);
4081 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
4084 return X86EMUL_CONTINUE
;
4086 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
4088 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
4090 kvm_emulate_wbinvd(emul_to_vcpu(ctxt
));
4093 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
4095 return _kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
4098 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
4101 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
4104 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
4106 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
4109 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
4111 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4112 unsigned long value
;
4116 value
= kvm_read_cr0(vcpu
);
4119 value
= vcpu
->arch
.cr2
;
4122 value
= kvm_read_cr3(vcpu
);
4125 value
= kvm_read_cr4(vcpu
);
4128 value
= kvm_get_cr8(vcpu
);
4131 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4138 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
4140 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4145 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
4148 vcpu
->arch
.cr2
= val
;
4151 res
= kvm_set_cr3(vcpu
, val
);
4154 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
4157 res
= kvm_set_cr8(vcpu
, val
);
4160 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4167 static void emulator_set_rflags(struct x86_emulate_ctxt
*ctxt
, ulong val
)
4169 kvm_set_rflags(emul_to_vcpu(ctxt
), val
);
4172 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
4174 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
4177 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4179 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
4182 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4184 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
4187 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4189 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
4192 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4194 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
4197 static unsigned long emulator_get_cached_segment_base(
4198 struct x86_emulate_ctxt
*ctxt
, int seg
)
4200 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
4203 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
4204 struct desc_struct
*desc
, u32
*base3
,
4207 struct kvm_segment var
;
4209 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
4210 *selector
= var
.selector
;
4217 set_desc_limit(desc
, var
.limit
);
4218 set_desc_base(desc
, (unsigned long)var
.base
);
4219 #ifdef CONFIG_X86_64
4221 *base3
= var
.base
>> 32;
4223 desc
->type
= var
.type
;
4225 desc
->dpl
= var
.dpl
;
4226 desc
->p
= var
.present
;
4227 desc
->avl
= var
.avl
;
4235 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
4236 struct desc_struct
*desc
, u32 base3
,
4239 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4240 struct kvm_segment var
;
4242 var
.selector
= selector
;
4243 var
.base
= get_desc_base(desc
);
4244 #ifdef CONFIG_X86_64
4245 var
.base
|= ((u64
)base3
) << 32;
4247 var
.limit
= get_desc_limit(desc
);
4249 var
.limit
= (var
.limit
<< 12) | 0xfff;
4250 var
.type
= desc
->type
;
4251 var
.present
= desc
->p
;
4252 var
.dpl
= desc
->dpl
;
4257 var
.avl
= desc
->avl
;
4258 var
.present
= desc
->p
;
4259 var
.unusable
= !var
.present
;
4262 kvm_set_segment(vcpu
, &var
, seg
);
4266 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
4267 u32 msr_index
, u64
*pdata
)
4269 return kvm_get_msr(emul_to_vcpu(ctxt
), msr_index
, pdata
);
4272 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
4273 u32 msr_index
, u64 data
)
4275 return kvm_set_msr(emul_to_vcpu(ctxt
), msr_index
, data
);
4278 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
4279 u32 pmc
, u64
*pdata
)
4281 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
4284 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
4286 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
4289 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
4292 kvm_load_guest_fpu(emul_to_vcpu(ctxt
));
4294 * CR0.TS may reference the host fpu state, not the guest fpu state,
4295 * so it may be clear at this point.
4300 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
4305 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
4306 struct x86_instruction_info
*info
,
4307 enum x86_intercept_stage stage
)
4309 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
4312 static void emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
4313 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
)
4315 kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
);
4318 static struct x86_emulate_ops emulate_ops
= {
4319 .read_std
= kvm_read_guest_virt_system
,
4320 .write_std
= kvm_write_guest_virt_system
,
4321 .fetch
= kvm_fetch_guest_virt
,
4322 .read_emulated
= emulator_read_emulated
,
4323 .write_emulated
= emulator_write_emulated
,
4324 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
4325 .invlpg
= emulator_invlpg
,
4326 .pio_in_emulated
= emulator_pio_in_emulated
,
4327 .pio_out_emulated
= emulator_pio_out_emulated
,
4328 .get_segment
= emulator_get_segment
,
4329 .set_segment
= emulator_set_segment
,
4330 .get_cached_segment_base
= emulator_get_cached_segment_base
,
4331 .get_gdt
= emulator_get_gdt
,
4332 .get_idt
= emulator_get_idt
,
4333 .set_gdt
= emulator_set_gdt
,
4334 .set_idt
= emulator_set_idt
,
4335 .get_cr
= emulator_get_cr
,
4336 .set_cr
= emulator_set_cr
,
4337 .set_rflags
= emulator_set_rflags
,
4338 .cpl
= emulator_get_cpl
,
4339 .get_dr
= emulator_get_dr
,
4340 .set_dr
= emulator_set_dr
,
4341 .set_msr
= emulator_set_msr
,
4342 .get_msr
= emulator_get_msr
,
4343 .read_pmc
= emulator_read_pmc
,
4344 .halt
= emulator_halt
,
4345 .wbinvd
= emulator_wbinvd
,
4346 .fix_hypercall
= emulator_fix_hypercall
,
4347 .get_fpu
= emulator_get_fpu
,
4348 .put_fpu
= emulator_put_fpu
,
4349 .intercept
= emulator_intercept
,
4350 .get_cpuid
= emulator_get_cpuid
,
4353 static void cache_all_regs(struct kvm_vcpu
*vcpu
)
4355 kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4356 kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4357 kvm_register_read(vcpu
, VCPU_REGS_RIP
);
4358 vcpu
->arch
.regs_dirty
= ~0;
4361 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
4363 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
, mask
);
4365 * an sti; sti; sequence only disable interrupts for the first
4366 * instruction. So, if the last instruction, be it emulated or
4367 * not, left the system with the INT_STI flag enabled, it
4368 * means that the last instruction is an sti. We should not
4369 * leave the flag on in this case. The same goes for mov ss
4371 if (!(int_shadow
& mask
))
4372 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
4375 static void inject_emulated_exception(struct kvm_vcpu
*vcpu
)
4377 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4378 if (ctxt
->exception
.vector
== PF_VECTOR
)
4379 kvm_propagate_fault(vcpu
, &ctxt
->exception
);
4380 else if (ctxt
->exception
.error_code_valid
)
4381 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
4382 ctxt
->exception
.error_code
);
4384 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
4387 static void init_decode_cache(struct x86_emulate_ctxt
*ctxt
,
4388 const unsigned long *regs
)
4390 memset(&ctxt
->twobyte
, 0,
4391 (void *)&ctxt
->regs
- (void *)&ctxt
->twobyte
);
4392 memcpy(ctxt
->regs
, regs
, sizeof(ctxt
->regs
));
4394 ctxt
->fetch
.start
= 0;
4395 ctxt
->fetch
.end
= 0;
4396 ctxt
->io_read
.pos
= 0;
4397 ctxt
->io_read
.end
= 0;
4398 ctxt
->mem_read
.pos
= 0;
4399 ctxt
->mem_read
.end
= 0;
4402 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
4404 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4408 * TODO: fix emulate.c to use guest_read/write_register
4409 * instead of direct ->regs accesses, can save hundred cycles
4410 * on Intel for instructions that don't read/change RSP, for
4413 cache_all_regs(vcpu
);
4415 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
4417 ctxt
->eflags
= kvm_get_rflags(vcpu
);
4418 ctxt
->eip
= kvm_rip_read(vcpu
);
4419 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
4420 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
4421 cs_l
? X86EMUL_MODE_PROT64
:
4422 cs_db
? X86EMUL_MODE_PROT32
:
4423 X86EMUL_MODE_PROT16
;
4424 ctxt
->guest_mode
= is_guest_mode(vcpu
);
4426 init_decode_cache(ctxt
, vcpu
->arch
.regs
);
4427 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4430 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
4432 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4435 init_emulate_ctxt(vcpu
);
4439 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
4440 ret
= emulate_int_real(ctxt
, irq
);
4442 if (ret
!= X86EMUL_CONTINUE
)
4443 return EMULATE_FAIL
;
4445 ctxt
->eip
= ctxt
->_eip
;
4446 memcpy(vcpu
->arch
.regs
, ctxt
->regs
, sizeof ctxt
->regs
);
4447 kvm_rip_write(vcpu
, ctxt
->eip
);
4448 kvm_set_rflags(vcpu
, ctxt
->eflags
);
4450 if (irq
== NMI_VECTOR
)
4451 vcpu
->arch
.nmi_pending
= 0;
4453 vcpu
->arch
.interrupt
.pending
= false;
4455 return EMULATE_DONE
;
4457 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
4459 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
4461 int r
= EMULATE_DONE
;
4463 ++vcpu
->stat
.insn_emulation_fail
;
4464 trace_kvm_emulate_insn_failed(vcpu
);
4465 if (!is_guest_mode(vcpu
)) {
4466 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
4467 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
4468 vcpu
->run
->internal
.ndata
= 0;
4471 kvm_queue_exception(vcpu
, UD_VECTOR
);
4476 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t gva
)
4484 * if emulation was due to access to shadowed page table
4485 * and it failed try to unshadow page and re-enter the
4486 * guest to let CPU execute the instruction.
4488 if (kvm_mmu_unprotect_page_virt(vcpu
, gva
))
4491 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, gva
, NULL
);
4493 if (gpa
== UNMAPPED_GVA
)
4494 return true; /* let cpu generate fault */
4496 if (!kvm_is_error_hva(gfn_to_hva(vcpu
->kvm
, gpa
>> PAGE_SHIFT
)))
4502 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
4503 unsigned long cr2
, int emulation_type
)
4505 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4506 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
4508 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
4509 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
4512 * If the emulation is caused by #PF and it is non-page_table
4513 * writing instruction, it means the VM-EXIT is caused by shadow
4514 * page protected, we can zap the shadow page and retry this
4515 * instruction directly.
4517 * Note: if the guest uses a non-page-table modifying instruction
4518 * on the PDE that points to the instruction, then we will unmap
4519 * the instruction and go to an infinite loop. So, we cache the
4520 * last retried eip and the last fault address, if we meet the eip
4521 * and the address again, we can break out of the potential infinite
4524 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
4526 if (!(emulation_type
& EMULTYPE_RETRY
))
4529 if (x86_page_table_writing_insn(ctxt
))
4532 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
4535 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
4536 vcpu
->arch
.last_retry_addr
= cr2
;
4538 if (!vcpu
->arch
.mmu
.direct_map
)
4539 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
4541 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
4546 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
4553 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4554 bool writeback
= true;
4556 kvm_clear_exception_queue(vcpu
);
4558 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
4559 init_emulate_ctxt(vcpu
);
4560 ctxt
->interruptibility
= 0;
4561 ctxt
->have_exception
= false;
4562 ctxt
->perm_ok
= false;
4564 ctxt
->only_vendor_specific_insn
4565 = emulation_type
& EMULTYPE_TRAP_UD
;
4567 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
4569 trace_kvm_emulate_insn_start(vcpu
);
4570 ++vcpu
->stat
.insn_emulation
;
4571 if (r
!= EMULATION_OK
) {
4572 if (emulation_type
& EMULTYPE_TRAP_UD
)
4573 return EMULATE_FAIL
;
4574 if (reexecute_instruction(vcpu
, cr2
))
4575 return EMULATE_DONE
;
4576 if (emulation_type
& EMULTYPE_SKIP
)
4577 return EMULATE_FAIL
;
4578 return handle_emulation_failure(vcpu
);
4582 if (emulation_type
& EMULTYPE_SKIP
) {
4583 kvm_rip_write(vcpu
, ctxt
->_eip
);
4584 return EMULATE_DONE
;
4587 if (retry_instruction(ctxt
, cr2
, emulation_type
))
4588 return EMULATE_DONE
;
4590 /* this is needed for vmware backdoor interface to work since it
4591 changes registers values during IO operation */
4592 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
4593 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4594 memcpy(ctxt
->regs
, vcpu
->arch
.regs
, sizeof ctxt
->regs
);
4598 r
= x86_emulate_insn(ctxt
);
4600 if (r
== EMULATION_INTERCEPTED
)
4601 return EMULATE_DONE
;
4603 if (r
== EMULATION_FAILED
) {
4604 if (reexecute_instruction(vcpu
, cr2
))
4605 return EMULATE_DONE
;
4607 return handle_emulation_failure(vcpu
);
4610 if (ctxt
->have_exception
) {
4611 inject_emulated_exception(vcpu
);
4613 } else if (vcpu
->arch
.pio
.count
) {
4614 if (!vcpu
->arch
.pio
.in
)
4615 vcpu
->arch
.pio
.count
= 0;
4618 r
= EMULATE_DO_MMIO
;
4619 } else if (vcpu
->mmio_needed
) {
4620 if (!vcpu
->mmio_is_write
)
4622 r
= EMULATE_DO_MMIO
;
4623 } else if (r
== EMULATION_RESTART
)
4629 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
4630 kvm_set_rflags(vcpu
, ctxt
->eflags
);
4631 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
4632 memcpy(vcpu
->arch
.regs
, ctxt
->regs
, sizeof ctxt
->regs
);
4633 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
4634 kvm_rip_write(vcpu
, ctxt
->eip
);
4636 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
4640 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
4642 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
4644 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4645 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
4646 size
, port
, &val
, 1);
4647 /* do not return to emulator after return from userspace */
4648 vcpu
->arch
.pio
.count
= 0;
4651 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
4653 static void tsc_bad(void *info
)
4655 __this_cpu_write(cpu_tsc_khz
, 0);
4658 static void tsc_khz_changed(void *data
)
4660 struct cpufreq_freqs
*freq
= data
;
4661 unsigned long khz
= 0;
4665 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
4666 khz
= cpufreq_quick_get(raw_smp_processor_id());
4669 __this_cpu_write(cpu_tsc_khz
, khz
);
4672 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
4675 struct cpufreq_freqs
*freq
= data
;
4677 struct kvm_vcpu
*vcpu
;
4678 int i
, send_ipi
= 0;
4681 * We allow guests to temporarily run on slowing clocks,
4682 * provided we notify them after, or to run on accelerating
4683 * clocks, provided we notify them before. Thus time never
4686 * However, we have a problem. We can't atomically update
4687 * the frequency of a given CPU from this function; it is
4688 * merely a notifier, which can be called from any CPU.
4689 * Changing the TSC frequency at arbitrary points in time
4690 * requires a recomputation of local variables related to
4691 * the TSC for each VCPU. We must flag these local variables
4692 * to be updated and be sure the update takes place with the
4693 * new frequency before any guests proceed.
4695 * Unfortunately, the combination of hotplug CPU and frequency
4696 * change creates an intractable locking scenario; the order
4697 * of when these callouts happen is undefined with respect to
4698 * CPU hotplug, and they can race with each other. As such,
4699 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4700 * undefined; you can actually have a CPU frequency change take
4701 * place in between the computation of X and the setting of the
4702 * variable. To protect against this problem, all updates of
4703 * the per_cpu tsc_khz variable are done in an interrupt
4704 * protected IPI, and all callers wishing to update the value
4705 * must wait for a synchronous IPI to complete (which is trivial
4706 * if the caller is on the CPU already). This establishes the
4707 * necessary total order on variable updates.
4709 * Note that because a guest time update may take place
4710 * anytime after the setting of the VCPU's request bit, the
4711 * correct TSC value must be set before the request. However,
4712 * to ensure the update actually makes it to any guest which
4713 * starts running in hardware virtualization between the set
4714 * and the acquisition of the spinlock, we must also ping the
4715 * CPU after setting the request bit.
4719 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
4721 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
4724 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
4726 raw_spin_lock(&kvm_lock
);
4727 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4728 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
4729 if (vcpu
->cpu
!= freq
->cpu
)
4731 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
4732 if (vcpu
->cpu
!= smp_processor_id())
4736 raw_spin_unlock(&kvm_lock
);
4738 if (freq
->old
< freq
->new && send_ipi
) {
4740 * We upscale the frequency. Must make the guest
4741 * doesn't see old kvmclock values while running with
4742 * the new frequency, otherwise we risk the guest sees
4743 * time go backwards.
4745 * In case we update the frequency for another cpu
4746 * (which might be in guest context) send an interrupt
4747 * to kick the cpu out of guest context. Next time
4748 * guest context is entered kvmclock will be updated,
4749 * so the guest will not see stale values.
4751 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
4756 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
4757 .notifier_call
= kvmclock_cpufreq_notifier
4760 static int kvmclock_cpu_notifier(struct notifier_block
*nfb
,
4761 unsigned long action
, void *hcpu
)
4763 unsigned int cpu
= (unsigned long)hcpu
;
4767 case CPU_DOWN_FAILED
:
4768 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
4770 case CPU_DOWN_PREPARE
:
4771 smp_call_function_single(cpu
, tsc_bad
, NULL
, 1);
4777 static struct notifier_block kvmclock_cpu_notifier_block
= {
4778 .notifier_call
= kvmclock_cpu_notifier
,
4779 .priority
= -INT_MAX
4782 static void kvm_timer_init(void)
4786 max_tsc_khz
= tsc_khz
;
4787 register_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
4788 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
4789 #ifdef CONFIG_CPU_FREQ
4790 struct cpufreq_policy policy
;
4791 memset(&policy
, 0, sizeof(policy
));
4793 cpufreq_get_policy(&policy
, cpu
);
4794 if (policy
.cpuinfo
.max_freq
)
4795 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
4798 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
4799 CPUFREQ_TRANSITION_NOTIFIER
);
4801 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
4802 for_each_online_cpu(cpu
)
4803 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
4806 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
4808 int kvm_is_in_guest(void)
4810 return __this_cpu_read(current_vcpu
) != NULL
;
4813 static int kvm_is_user_mode(void)
4817 if (__this_cpu_read(current_vcpu
))
4818 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
4820 return user_mode
!= 0;
4823 static unsigned long kvm_get_guest_ip(void)
4825 unsigned long ip
= 0;
4827 if (__this_cpu_read(current_vcpu
))
4828 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
4833 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
4834 .is_in_guest
= kvm_is_in_guest
,
4835 .is_user_mode
= kvm_is_user_mode
,
4836 .get_guest_ip
= kvm_get_guest_ip
,
4839 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
4841 __this_cpu_write(current_vcpu
, vcpu
);
4843 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
4845 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
4847 __this_cpu_write(current_vcpu
, NULL
);
4849 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
4851 static void kvm_set_mmio_spte_mask(void)
4854 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
4857 * Set the reserved bits and the present bit of an paging-structure
4858 * entry to generate page fault with PFER.RSV = 1.
4860 mask
= ((1ull << (62 - maxphyaddr
+ 1)) - 1) << maxphyaddr
;
4863 #ifdef CONFIG_X86_64
4865 * If reserved bit is not supported, clear the present bit to disable
4868 if (maxphyaddr
== 52)
4872 kvm_mmu_set_mmio_spte_mask(mask
);
4875 int kvm_arch_init(void *opaque
)
4878 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
4881 printk(KERN_ERR
"kvm: already loaded the other module\n");
4886 if (!ops
->cpu_has_kvm_support()) {
4887 printk(KERN_ERR
"kvm: no hardware support\n");
4891 if (ops
->disabled_by_bios()) {
4892 printk(KERN_ERR
"kvm: disabled by bios\n");
4897 r
= kvm_mmu_module_init();
4901 kvm_set_mmio_spte_mask();
4902 kvm_init_msr_list();
4905 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
4906 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
4910 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
4913 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
4921 void kvm_arch_exit(void)
4923 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
4925 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
4926 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
4927 CPUFREQ_TRANSITION_NOTIFIER
);
4928 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
4930 kvm_mmu_module_exit();
4933 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
4935 ++vcpu
->stat
.halt_exits
;
4936 if (irqchip_in_kernel(vcpu
->kvm
)) {
4937 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
4940 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
4944 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
4946 int kvm_hv_hypercall(struct kvm_vcpu
*vcpu
)
4948 u64 param
, ingpa
, outgpa
, ret
;
4949 uint16_t code
, rep_idx
, rep_cnt
, res
= HV_STATUS_SUCCESS
, rep_done
= 0;
4950 bool fast
, longmode
;
4954 * hypercall generates UD from non zero cpl and real mode
4957 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 || !is_protmode(vcpu
)) {
4958 kvm_queue_exception(vcpu
, UD_VECTOR
);
4962 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
4963 longmode
= is_long_mode(vcpu
) && cs_l
== 1;
4966 param
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDX
) << 32) |
4967 (kvm_register_read(vcpu
, VCPU_REGS_RAX
) & 0xffffffff);
4968 ingpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RBX
) << 32) |
4969 (kvm_register_read(vcpu
, VCPU_REGS_RCX
) & 0xffffffff);
4970 outgpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDI
) << 32) |
4971 (kvm_register_read(vcpu
, VCPU_REGS_RSI
) & 0xffffffff);
4973 #ifdef CONFIG_X86_64
4975 param
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4976 ingpa
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4977 outgpa
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
4981 code
= param
& 0xffff;
4982 fast
= (param
>> 16) & 0x1;
4983 rep_cnt
= (param
>> 32) & 0xfff;
4984 rep_idx
= (param
>> 48) & 0xfff;
4986 trace_kvm_hv_hypercall(code
, fast
, rep_cnt
, rep_idx
, ingpa
, outgpa
);
4989 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT
:
4990 kvm_vcpu_on_spin(vcpu
);
4993 res
= HV_STATUS_INVALID_HYPERCALL_CODE
;
4997 ret
= res
| (((u64
)rep_done
& 0xfff) << 32);
4999 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5001 kvm_register_write(vcpu
, VCPU_REGS_RDX
, ret
>> 32);
5002 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
& 0xffffffff);
5008 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
5010 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
5013 if (kvm_hv_hypercall_enabled(vcpu
->kvm
))
5014 return kvm_hv_hypercall(vcpu
);
5016 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5017 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5018 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5019 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5020 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5022 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
5024 if (!is_long_mode(vcpu
)) {
5032 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
5038 case KVM_HC_VAPIC_POLL_IRQ
:
5046 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5047 ++vcpu
->stat
.hypercalls
;
5050 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
5052 int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
5054 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5055 char instruction
[3];
5056 unsigned long rip
= kvm_rip_read(vcpu
);
5059 * Blow out the MMU to ensure that no other VCPU has an active mapping
5060 * to ensure that the updated hypercall appears atomically across all
5063 kvm_mmu_zap_all(vcpu
->kvm
);
5065 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
5067 return emulator_write_emulated(ctxt
, rip
, instruction
, 3, NULL
);
5071 * Check if userspace requested an interrupt window, and that the
5072 * interrupt window is open.
5074 * No need to exit to userspace if we already have an interrupt queued.
5076 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
5078 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
5079 vcpu
->run
->request_interrupt_window
&&
5080 kvm_arch_interrupt_allowed(vcpu
));
5083 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
5085 struct kvm_run
*kvm_run
= vcpu
->run
;
5087 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
5088 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
5089 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
5090 if (irqchip_in_kernel(vcpu
->kvm
))
5091 kvm_run
->ready_for_interrupt_injection
= 1;
5093 kvm_run
->ready_for_interrupt_injection
=
5094 kvm_arch_interrupt_allowed(vcpu
) &&
5095 !kvm_cpu_has_interrupt(vcpu
) &&
5096 !kvm_event_needs_reinjection(vcpu
);
5099 static void vapic_enter(struct kvm_vcpu
*vcpu
)
5101 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5104 if (!apic
|| !apic
->vapic_addr
)
5107 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5109 vcpu
->arch
.apic
->vapic_page
= page
;
5112 static void vapic_exit(struct kvm_vcpu
*vcpu
)
5114 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5117 if (!apic
|| !apic
->vapic_addr
)
5120 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5121 kvm_release_page_dirty(apic
->vapic_page
);
5122 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5123 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5126 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
5130 if (!kvm_x86_ops
->update_cr8_intercept
)
5133 if (!vcpu
->arch
.apic
)
5136 if (!vcpu
->arch
.apic
->vapic_addr
)
5137 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
5144 tpr
= kvm_lapic_get_cr8(vcpu
);
5146 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
5149 static void inject_pending_event(struct kvm_vcpu
*vcpu
)
5151 /* try to reinject previous events if any */
5152 if (vcpu
->arch
.exception
.pending
) {
5153 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
5154 vcpu
->arch
.exception
.has_error_code
,
5155 vcpu
->arch
.exception
.error_code
);
5156 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
5157 vcpu
->arch
.exception
.has_error_code
,
5158 vcpu
->arch
.exception
.error_code
,
5159 vcpu
->arch
.exception
.reinject
);
5163 if (vcpu
->arch
.nmi_injected
) {
5164 kvm_x86_ops
->set_nmi(vcpu
);
5168 if (vcpu
->arch
.interrupt
.pending
) {
5169 kvm_x86_ops
->set_irq(vcpu
);
5173 /* try to inject new event if pending */
5174 if (vcpu
->arch
.nmi_pending
) {
5175 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
5176 --vcpu
->arch
.nmi_pending
;
5177 vcpu
->arch
.nmi_injected
= true;
5178 kvm_x86_ops
->set_nmi(vcpu
);
5180 } else if (kvm_cpu_has_interrupt(vcpu
)) {
5181 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
5182 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
5184 kvm_x86_ops
->set_irq(vcpu
);
5189 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
5191 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
5192 !vcpu
->guest_xcr0_loaded
) {
5193 /* kvm_set_xcr() also depends on this */
5194 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
5195 vcpu
->guest_xcr0_loaded
= 1;
5199 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
5201 if (vcpu
->guest_xcr0_loaded
) {
5202 if (vcpu
->arch
.xcr0
!= host_xcr0
)
5203 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
5204 vcpu
->guest_xcr0_loaded
= 0;
5208 static void process_nmi(struct kvm_vcpu
*vcpu
)
5213 * x86 is limited to one NMI running, and one NMI pending after it.
5214 * If an NMI is already in progress, limit further NMIs to just one.
5215 * Otherwise, allow two (and we'll inject the first one immediately).
5217 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
5220 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
5221 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
5222 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5225 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
5228 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
5229 vcpu
->run
->request_interrupt_window
;
5230 bool req_immediate_exit
= 0;
5232 if (vcpu
->requests
) {
5233 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
5234 kvm_mmu_unload(vcpu
);
5235 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
5236 __kvm_migrate_timers(vcpu
);
5237 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
5238 r
= kvm_guest_time_update(vcpu
);
5242 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
5243 kvm_mmu_sync_roots(vcpu
);
5244 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
5245 kvm_x86_ops
->tlb_flush(vcpu
);
5246 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
5247 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
5251 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
5252 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
5256 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
)) {
5257 vcpu
->fpu_active
= 0;
5258 kvm_x86_ops
->fpu_deactivate(vcpu
);
5260 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
5261 /* Page is swapped out. Do synthetic halt */
5262 vcpu
->arch
.apf
.halted
= true;
5266 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
5267 record_steal_time(vcpu
);
5268 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
5270 req_immediate_exit
=
5271 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT
, vcpu
);
5272 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
5273 kvm_handle_pmu_event(vcpu
);
5274 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
5275 kvm_deliver_pmi(vcpu
);
5278 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
5279 inject_pending_event(vcpu
);
5281 /* enable NMI/IRQ window open exits if needed */
5282 if (vcpu
->arch
.nmi_pending
)
5283 kvm_x86_ops
->enable_nmi_window(vcpu
);
5284 else if (kvm_cpu_has_interrupt(vcpu
) || req_int_win
)
5285 kvm_x86_ops
->enable_irq_window(vcpu
);
5287 if (kvm_lapic_enabled(vcpu
)) {
5288 update_cr8_intercept(vcpu
);
5289 kvm_lapic_sync_to_vapic(vcpu
);
5293 r
= kvm_mmu_reload(vcpu
);
5295 goto cancel_injection
;
5300 kvm_x86_ops
->prepare_guest_switch(vcpu
);
5301 if (vcpu
->fpu_active
)
5302 kvm_load_guest_fpu(vcpu
);
5303 kvm_load_guest_xcr0(vcpu
);
5305 vcpu
->mode
= IN_GUEST_MODE
;
5307 /* We should set ->mode before check ->requests,
5308 * see the comment in make_all_cpus_request.
5312 local_irq_disable();
5314 if (vcpu
->mode
== EXITING_GUEST_MODE
|| vcpu
->requests
5315 || need_resched() || signal_pending(current
)) {
5316 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5321 goto cancel_injection
;
5324 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5326 if (req_immediate_exit
)
5327 smp_send_reschedule(vcpu
->cpu
);
5331 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
5333 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
5334 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
5335 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
5336 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
5339 trace_kvm_entry(vcpu
->vcpu_id
);
5340 kvm_x86_ops
->run(vcpu
);
5343 * If the guest has used debug registers, at least dr7
5344 * will be disabled while returning to the host.
5345 * If we don't have active breakpoints in the host, we don't
5346 * care about the messed up debug address registers. But if
5347 * we have some of them active, restore the old state.
5349 if (hw_breakpoint_active())
5350 hw_breakpoint_restore();
5352 vcpu
->arch
.last_guest_tsc
= kvm_x86_ops
->read_l1_tsc(vcpu
);
5354 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5361 * We must have an instruction between local_irq_enable() and
5362 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5363 * the interrupt shadow. The stat.exits increment will do nicely.
5364 * But we need to prevent reordering, hence this barrier():
5372 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5375 * Profile KVM exit RIPs:
5377 if (unlikely(prof_on
== KVM_PROFILING
)) {
5378 unsigned long rip
= kvm_rip_read(vcpu
);
5379 profile_hit(KVM_PROFILING
, (void *)rip
);
5382 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
5383 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5385 if (vcpu
->arch
.apic_attention
)
5386 kvm_lapic_sync_from_vapic(vcpu
);
5388 r
= kvm_x86_ops
->handle_exit(vcpu
);
5392 kvm_x86_ops
->cancel_injection(vcpu
);
5393 if (unlikely(vcpu
->arch
.apic_attention
))
5394 kvm_lapic_sync_from_vapic(vcpu
);
5400 static int __vcpu_run(struct kvm_vcpu
*vcpu
)
5403 struct kvm
*kvm
= vcpu
->kvm
;
5405 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
5406 pr_debug("vcpu %d received sipi with vector # %x\n",
5407 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
5408 kvm_lapic_reset(vcpu
);
5409 r
= kvm_arch_vcpu_reset(vcpu
);
5412 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
5415 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5420 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
5421 !vcpu
->arch
.apf
.halted
)
5422 r
= vcpu_enter_guest(vcpu
);
5424 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5425 kvm_vcpu_block(vcpu
);
5426 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5427 if (kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
5429 switch(vcpu
->arch
.mp_state
) {
5430 case KVM_MP_STATE_HALTED
:
5431 vcpu
->arch
.mp_state
=
5432 KVM_MP_STATE_RUNNABLE
;
5433 case KVM_MP_STATE_RUNNABLE
:
5434 vcpu
->arch
.apf
.halted
= false;
5436 case KVM_MP_STATE_SIPI_RECEIVED
:
5447 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
5448 if (kvm_cpu_has_pending_timer(vcpu
))
5449 kvm_inject_pending_timer_irqs(vcpu
);
5451 if (dm_request_for_irq_injection(vcpu
)) {
5453 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5454 ++vcpu
->stat
.request_irq_exits
;
5457 kvm_check_async_pf_completion(vcpu
);
5459 if (signal_pending(current
)) {
5461 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5462 ++vcpu
->stat
.signal_exits
;
5464 if (need_resched()) {
5465 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5467 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5471 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5479 * Implements the following, as a state machine:
5494 static int complete_mmio(struct kvm_vcpu
*vcpu
)
5496 struct kvm_run
*run
= vcpu
->run
;
5497 struct kvm_mmio_fragment
*frag
;
5500 if (!(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
))
5503 if (vcpu
->mmio_needed
) {
5504 /* Complete previous fragment */
5505 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
++];
5506 if (!vcpu
->mmio_is_write
)
5507 memcpy(frag
->data
, run
->mmio
.data
, frag
->len
);
5508 if (vcpu
->mmio_cur_fragment
== vcpu
->mmio_nr_fragments
) {
5509 vcpu
->mmio_needed
= 0;
5510 if (vcpu
->mmio_is_write
)
5512 vcpu
->mmio_read_completed
= 1;
5515 /* Initiate next fragment */
5517 run
->exit_reason
= KVM_EXIT_MMIO
;
5518 run
->mmio
.phys_addr
= frag
->gpa
;
5519 if (vcpu
->mmio_is_write
)
5520 memcpy(run
->mmio
.data
, frag
->data
, frag
->len
);
5521 run
->mmio
.len
= frag
->len
;
5522 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
5527 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5528 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
5529 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5530 if (r
!= EMULATE_DONE
)
5535 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
5540 if (!tsk_used_math(current
) && init_fpu(current
))
5543 if (vcpu
->sigset_active
)
5544 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
5546 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
5547 kvm_vcpu_block(vcpu
);
5548 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
5553 /* re-sync apic's tpr */
5554 if (!irqchip_in_kernel(vcpu
->kvm
)) {
5555 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
5561 r
= complete_mmio(vcpu
);
5565 r
= __vcpu_run(vcpu
);
5568 post_kvm_run_save(vcpu
);
5569 if (vcpu
->sigset_active
)
5570 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
5575 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
5577 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
5579 * We are here if userspace calls get_regs() in the middle of
5580 * instruction emulation. Registers state needs to be copied
5581 * back from emulation context to vcpu. Userspace shouldn't do
5582 * that usually, but some bad designed PV devices (vmware
5583 * backdoor interface) need this to work
5585 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5586 memcpy(vcpu
->arch
.regs
, ctxt
->regs
, sizeof ctxt
->regs
);
5587 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5589 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5590 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5591 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5592 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5593 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5594 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
5595 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
5596 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
5597 #ifdef CONFIG_X86_64
5598 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
5599 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
5600 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
5601 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
5602 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
5603 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
5604 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
5605 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
5608 regs
->rip
= kvm_rip_read(vcpu
);
5609 regs
->rflags
= kvm_get_rflags(vcpu
);
5614 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
5616 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
5617 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5619 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
5620 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
5621 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
5622 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
5623 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
5624 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
5625 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
5626 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
5627 #ifdef CONFIG_X86_64
5628 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
5629 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
5630 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
5631 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
5632 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
5633 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
5634 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
5635 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
5638 kvm_rip_write(vcpu
, regs
->rip
);
5639 kvm_set_rflags(vcpu
, regs
->rflags
);
5641 vcpu
->arch
.exception
.pending
= false;
5643 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5648 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
5650 struct kvm_segment cs
;
5652 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
5656 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
5658 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
5659 struct kvm_sregs
*sregs
)
5663 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
5664 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
5665 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
5666 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
5667 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
5668 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
5670 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
5671 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
5673 kvm_x86_ops
->get_idt(vcpu
, &dt
);
5674 sregs
->idt
.limit
= dt
.size
;
5675 sregs
->idt
.base
= dt
.address
;
5676 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
5677 sregs
->gdt
.limit
= dt
.size
;
5678 sregs
->gdt
.base
= dt
.address
;
5680 sregs
->cr0
= kvm_read_cr0(vcpu
);
5681 sregs
->cr2
= vcpu
->arch
.cr2
;
5682 sregs
->cr3
= kvm_read_cr3(vcpu
);
5683 sregs
->cr4
= kvm_read_cr4(vcpu
);
5684 sregs
->cr8
= kvm_get_cr8(vcpu
);
5685 sregs
->efer
= vcpu
->arch
.efer
;
5686 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
5688 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
5690 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
5691 set_bit(vcpu
->arch
.interrupt
.nr
,
5692 (unsigned long *)sregs
->interrupt_bitmap
);
5697 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
5698 struct kvm_mp_state
*mp_state
)
5700 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
5704 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
5705 struct kvm_mp_state
*mp_state
)
5707 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
5708 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5712 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
5713 int reason
, bool has_error_code
, u32 error_code
)
5715 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5718 init_emulate_ctxt(vcpu
);
5720 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
5721 has_error_code
, error_code
);
5724 return EMULATE_FAIL
;
5726 memcpy(vcpu
->arch
.regs
, ctxt
->regs
, sizeof ctxt
->regs
);
5727 kvm_rip_write(vcpu
, ctxt
->eip
);
5728 kvm_set_rflags(vcpu
, ctxt
->eflags
);
5729 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5730 return EMULATE_DONE
;
5732 EXPORT_SYMBOL_GPL(kvm_task_switch
);
5734 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
5735 struct kvm_sregs
*sregs
)
5737 int mmu_reset_needed
= 0;
5738 int pending_vec
, max_bits
, idx
;
5741 dt
.size
= sregs
->idt
.limit
;
5742 dt
.address
= sregs
->idt
.base
;
5743 kvm_x86_ops
->set_idt(vcpu
, &dt
);
5744 dt
.size
= sregs
->gdt
.limit
;
5745 dt
.address
= sregs
->gdt
.base
;
5746 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
5748 vcpu
->arch
.cr2
= sregs
->cr2
;
5749 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
5750 vcpu
->arch
.cr3
= sregs
->cr3
;
5751 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
5753 kvm_set_cr8(vcpu
, sregs
->cr8
);
5755 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
5756 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
5757 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
5759 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
5760 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
5761 vcpu
->arch
.cr0
= sregs
->cr0
;
5763 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
5764 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
5765 if (sregs
->cr4
& X86_CR4_OSXSAVE
)
5766 kvm_update_cpuid(vcpu
);
5768 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5769 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
5770 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
5771 mmu_reset_needed
= 1;
5773 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5775 if (mmu_reset_needed
)
5776 kvm_mmu_reset_context(vcpu
);
5778 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
5779 pending_vec
= find_first_bit(
5780 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
5781 if (pending_vec
< max_bits
) {
5782 kvm_queue_interrupt(vcpu
, pending_vec
, false);
5783 pr_debug("Set back pending irq %d\n", pending_vec
);
5786 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
5787 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
5788 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
5789 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
5790 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
5791 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
5793 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
5794 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
5796 update_cr8_intercept(vcpu
);
5798 /* Older userspace won't unhalt the vcpu on reset. */
5799 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
5800 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
5802 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
5804 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5809 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
5810 struct kvm_guest_debug
*dbg
)
5812 unsigned long rflags
;
5815 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
5817 if (vcpu
->arch
.exception
.pending
)
5819 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
5820 kvm_queue_exception(vcpu
, DB_VECTOR
);
5822 kvm_queue_exception(vcpu
, BP_VECTOR
);
5826 * Read rflags as long as potentially injected trace flags are still
5829 rflags
= kvm_get_rflags(vcpu
);
5831 vcpu
->guest_debug
= dbg
->control
;
5832 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
5833 vcpu
->guest_debug
= 0;
5835 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
5836 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
5837 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
5838 vcpu
->arch
.switch_db_regs
=
5839 (dbg
->arch
.debugreg
[7] & DR7_BP_EN_MASK
);
5841 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
5842 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
5843 vcpu
->arch
.switch_db_regs
= (vcpu
->arch
.dr7
& DR7_BP_EN_MASK
);
5846 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
5847 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
5848 get_segment_base(vcpu
, VCPU_SREG_CS
);
5851 * Trigger an rflags update that will inject or remove the trace
5854 kvm_set_rflags(vcpu
, rflags
);
5856 kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
5866 * Translate a guest virtual address to a guest physical address.
5868 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
5869 struct kvm_translation
*tr
)
5871 unsigned long vaddr
= tr
->linear_address
;
5875 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5876 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
5877 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5878 tr
->physical_address
= gpa
;
5879 tr
->valid
= gpa
!= UNMAPPED_GVA
;
5886 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
5888 struct i387_fxsave_struct
*fxsave
=
5889 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
5891 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
5892 fpu
->fcw
= fxsave
->cwd
;
5893 fpu
->fsw
= fxsave
->swd
;
5894 fpu
->ftwx
= fxsave
->twd
;
5895 fpu
->last_opcode
= fxsave
->fop
;
5896 fpu
->last_ip
= fxsave
->rip
;
5897 fpu
->last_dp
= fxsave
->rdp
;
5898 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
5903 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
5905 struct i387_fxsave_struct
*fxsave
=
5906 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
5908 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
5909 fxsave
->cwd
= fpu
->fcw
;
5910 fxsave
->swd
= fpu
->fsw
;
5911 fxsave
->twd
= fpu
->ftwx
;
5912 fxsave
->fop
= fpu
->last_opcode
;
5913 fxsave
->rip
= fpu
->last_ip
;
5914 fxsave
->rdp
= fpu
->last_dp
;
5915 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
5920 int fx_init(struct kvm_vcpu
*vcpu
)
5924 err
= fpu_alloc(&vcpu
->arch
.guest_fpu
);
5928 fpu_finit(&vcpu
->arch
.guest_fpu
);
5931 * Ensure guest xcr0 is valid for loading
5933 vcpu
->arch
.xcr0
= XSTATE_FP
;
5935 vcpu
->arch
.cr0
|= X86_CR0_ET
;
5939 EXPORT_SYMBOL_GPL(fx_init
);
5941 static void fx_free(struct kvm_vcpu
*vcpu
)
5943 fpu_free(&vcpu
->arch
.guest_fpu
);
5946 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
5948 if (vcpu
->guest_fpu_loaded
)
5952 * Restore all possible states in the guest,
5953 * and assume host would use all available bits.
5954 * Guest xcr0 would be loaded later.
5956 kvm_put_guest_xcr0(vcpu
);
5957 vcpu
->guest_fpu_loaded
= 1;
5958 unlazy_fpu(current
);
5959 fpu_restore_checking(&vcpu
->arch
.guest_fpu
);
5963 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
5965 kvm_put_guest_xcr0(vcpu
);
5967 if (!vcpu
->guest_fpu_loaded
)
5970 vcpu
->guest_fpu_loaded
= 0;
5971 fpu_save_init(&vcpu
->arch
.guest_fpu
);
5972 ++vcpu
->stat
.fpu_reload
;
5973 kvm_make_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
);
5977 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
5979 kvmclock_reset(vcpu
);
5981 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
5983 kvm_x86_ops
->vcpu_free(vcpu
);
5986 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
5989 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
5990 printk_once(KERN_WARNING
5991 "kvm: SMP vm created on host with unstable TSC; "
5992 "guest TSC will not be reliable\n");
5993 return kvm_x86_ops
->vcpu_create(kvm
, id
);
5996 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
6000 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
6002 r
= kvm_arch_vcpu_reset(vcpu
);
6004 r
= kvm_mmu_setup(vcpu
);
6010 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
6012 vcpu
->arch
.apf
.msr_val
= 0;
6015 kvm_mmu_unload(vcpu
);
6019 kvm_x86_ops
->vcpu_free(vcpu
);
6022 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
6024 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
6025 vcpu
->arch
.nmi_pending
= 0;
6026 vcpu
->arch
.nmi_injected
= false;
6028 vcpu
->arch
.switch_db_regs
= 0;
6029 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
6030 vcpu
->arch
.dr6
= DR6_FIXED_1
;
6031 vcpu
->arch
.dr7
= DR7_FIXED_1
;
6033 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6034 vcpu
->arch
.apf
.msr_val
= 0;
6035 vcpu
->arch
.st
.msr_val
= 0;
6037 kvmclock_reset(vcpu
);
6039 kvm_clear_async_pf_completion_queue(vcpu
);
6040 kvm_async_pf_hash_reset(vcpu
);
6041 vcpu
->arch
.apf
.halted
= false;
6043 kvm_pmu_reset(vcpu
);
6045 return kvm_x86_ops
->vcpu_reset(vcpu
);
6048 int kvm_arch_hardware_enable(void *garbage
)
6051 struct kvm_vcpu
*vcpu
;
6056 bool stable
, backwards_tsc
= false;
6058 kvm_shared_msr_cpu_online();
6059 ret
= kvm_x86_ops
->hardware_enable(garbage
);
6063 local_tsc
= native_read_tsc();
6064 stable
= !check_tsc_unstable();
6065 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6066 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6067 if (!stable
&& vcpu
->cpu
== smp_processor_id())
6068 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
6069 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
6070 backwards_tsc
= true;
6071 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
6072 max_tsc
= vcpu
->arch
.last_host_tsc
;
6078 * Sometimes, even reliable TSCs go backwards. This happens on
6079 * platforms that reset TSC during suspend or hibernate actions, but
6080 * maintain synchronization. We must compensate. Fortunately, we can
6081 * detect that condition here, which happens early in CPU bringup,
6082 * before any KVM threads can be running. Unfortunately, we can't
6083 * bring the TSCs fully up to date with real time, as we aren't yet far
6084 * enough into CPU bringup that we know how much real time has actually
6085 * elapsed; our helper function, get_kernel_ns() will be using boot
6086 * variables that haven't been updated yet.
6088 * So we simply find the maximum observed TSC above, then record the
6089 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6090 * the adjustment will be applied. Note that we accumulate
6091 * adjustments, in case multiple suspend cycles happen before some VCPU
6092 * gets a chance to run again. In the event that no KVM threads get a
6093 * chance to run, we will miss the entire elapsed period, as we'll have
6094 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6095 * loose cycle time. This isn't too big a deal, since the loss will be
6096 * uniform across all VCPUs (not to mention the scenario is extremely
6097 * unlikely). It is possible that a second hibernate recovery happens
6098 * much faster than a first, causing the observed TSC here to be
6099 * smaller; this would require additional padding adjustment, which is
6100 * why we set last_host_tsc to the local tsc observed here.
6102 * N.B. - this code below runs only on platforms with reliable TSC,
6103 * as that is the only way backwards_tsc is set above. Also note
6104 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6105 * have the same delta_cyc adjustment applied if backwards_tsc
6106 * is detected. Note further, this adjustment is only done once,
6107 * as we reset last_host_tsc on all VCPUs to stop this from being
6108 * called multiple times (one for each physical CPU bringup).
6110 * Platforms with unreliable TSCs don't have to deal with this, they
6111 * will be compensated by the logic in vcpu_load, which sets the TSC to
6112 * catchup mode. This will catchup all VCPUs to real time, but cannot
6113 * guarantee that they stay in perfect synchronization.
6115 if (backwards_tsc
) {
6116 u64 delta_cyc
= max_tsc
- local_tsc
;
6117 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6118 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6119 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
6120 vcpu
->arch
.last_host_tsc
= local_tsc
;
6124 * We have to disable TSC offset matching.. if you were
6125 * booting a VM while issuing an S4 host suspend....
6126 * you may have some problem. Solving this issue is
6127 * left as an exercise to the reader.
6129 kvm
->arch
.last_tsc_nsec
= 0;
6130 kvm
->arch
.last_tsc_write
= 0;
6137 void kvm_arch_hardware_disable(void *garbage
)
6139 kvm_x86_ops
->hardware_disable(garbage
);
6140 drop_user_return_notifiers(garbage
);
6143 int kvm_arch_hardware_setup(void)
6145 return kvm_x86_ops
->hardware_setup();
6148 void kvm_arch_hardware_unsetup(void)
6150 kvm_x86_ops
->hardware_unsetup();
6153 void kvm_arch_check_processor_compat(void *rtn
)
6155 kvm_x86_ops
->check_processor_compatibility(rtn
);
6158 bool kvm_vcpu_compatible(struct kvm_vcpu
*vcpu
)
6160 return irqchip_in_kernel(vcpu
->kvm
) == (vcpu
->arch
.apic
!= NULL
);
6163 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
6169 BUG_ON(vcpu
->kvm
== NULL
);
6172 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
6173 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
6174 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6176 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
6178 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
6183 vcpu
->arch
.pio_data
= page_address(page
);
6185 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
6187 r
= kvm_mmu_create(vcpu
);
6189 goto fail_free_pio_data
;
6191 if (irqchip_in_kernel(kvm
)) {
6192 r
= kvm_create_lapic(vcpu
);
6194 goto fail_mmu_destroy
;
6197 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
6199 if (!vcpu
->arch
.mce_banks
) {
6201 goto fail_free_lapic
;
6203 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
6205 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
))
6206 goto fail_free_mce_banks
;
6208 kvm_async_pf_hash_reset(vcpu
);
6212 fail_free_mce_banks
:
6213 kfree(vcpu
->arch
.mce_banks
);
6215 kvm_free_lapic(vcpu
);
6217 kvm_mmu_destroy(vcpu
);
6219 free_page((unsigned long)vcpu
->arch
.pio_data
);
6224 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
6228 kvm_pmu_destroy(vcpu
);
6229 kfree(vcpu
->arch
.mce_banks
);
6230 kvm_free_lapic(vcpu
);
6231 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6232 kvm_mmu_destroy(vcpu
);
6233 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6234 free_page((unsigned long)vcpu
->arch
.pio_data
);
6237 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
6242 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
6243 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
6245 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6246 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
6248 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
6253 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
6256 kvm_mmu_unload(vcpu
);
6260 static void kvm_free_vcpus(struct kvm
*kvm
)
6263 struct kvm_vcpu
*vcpu
;
6266 * Unpin any mmu pages first.
6268 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6269 kvm_clear_async_pf_completion_queue(vcpu
);
6270 kvm_unload_vcpu_mmu(vcpu
);
6272 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6273 kvm_arch_vcpu_free(vcpu
);
6275 mutex_lock(&kvm
->lock
);
6276 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
6277 kvm
->vcpus
[i
] = NULL
;
6279 atomic_set(&kvm
->online_vcpus
, 0);
6280 mutex_unlock(&kvm
->lock
);
6283 void kvm_arch_sync_events(struct kvm
*kvm
)
6285 kvm_free_all_assigned_devices(kvm
);
6289 void kvm_arch_destroy_vm(struct kvm
*kvm
)
6291 kvm_iommu_unmap_guest(kvm
);
6292 kfree(kvm
->arch
.vpic
);
6293 kfree(kvm
->arch
.vioapic
);
6294 kvm_free_vcpus(kvm
);
6295 if (kvm
->arch
.apic_access_page
)
6296 put_page(kvm
->arch
.apic_access_page
);
6297 if (kvm
->arch
.ept_identity_pagetable
)
6298 put_page(kvm
->arch
.ept_identity_pagetable
);
6301 void kvm_arch_free_memslot(struct kvm_memory_slot
*free
,
6302 struct kvm_memory_slot
*dont
)
6306 for (i
= 0; i
< KVM_NR_PAGE_SIZES
- 1; ++i
) {
6307 if (!dont
|| free
->arch
.rmap_pde
[i
] != dont
->arch
.rmap_pde
[i
]) {
6308 kvm_kvfree(free
->arch
.rmap_pde
[i
]);
6309 free
->arch
.rmap_pde
[i
] = NULL
;
6311 if (!dont
|| free
->arch
.lpage_info
[i
] != dont
->arch
.lpage_info
[i
]) {
6312 kvm_kvfree(free
->arch
.lpage_info
[i
]);
6313 free
->arch
.lpage_info
[i
] = NULL
;
6318 int kvm_arch_create_memslot(struct kvm_memory_slot
*slot
, unsigned long npages
)
6322 for (i
= 0; i
< KVM_NR_PAGE_SIZES
- 1; ++i
) {
6327 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
6328 slot
->base_gfn
, level
) + 1;
6330 slot
->arch
.rmap_pde
[i
] =
6331 kvm_kvzalloc(lpages
* sizeof(*slot
->arch
.rmap_pde
[i
]));
6332 if (!slot
->arch
.rmap_pde
[i
])
6335 slot
->arch
.lpage_info
[i
] =
6336 kvm_kvzalloc(lpages
* sizeof(*slot
->arch
.lpage_info
[i
]));
6337 if (!slot
->arch
.lpage_info
[i
])
6340 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
6341 slot
->arch
.lpage_info
[i
][0].write_count
= 1;
6342 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
6343 slot
->arch
.lpage_info
[i
][lpages
- 1].write_count
= 1;
6344 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
6346 * If the gfn and userspace address are not aligned wrt each
6347 * other, or if explicitly asked to, disable large page
6348 * support for this slot
6350 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
6351 !kvm_largepages_enabled()) {
6354 for (j
= 0; j
< lpages
; ++j
)
6355 slot
->arch
.lpage_info
[i
][j
].write_count
= 1;
6362 for (i
= 0; i
< KVM_NR_PAGE_SIZES
- 1; ++i
) {
6363 kvm_kvfree(slot
->arch
.rmap_pde
[i
]);
6364 kvm_kvfree(slot
->arch
.lpage_info
[i
]);
6365 slot
->arch
.rmap_pde
[i
] = NULL
;
6366 slot
->arch
.lpage_info
[i
] = NULL
;
6371 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
6372 struct kvm_memory_slot
*memslot
,
6373 struct kvm_memory_slot old
,
6374 struct kvm_userspace_memory_region
*mem
,
6377 int npages
= memslot
->npages
;
6378 int map_flags
= MAP_PRIVATE
| MAP_ANONYMOUS
;
6380 /* Prevent internal slot pages from being moved by fork()/COW. */
6381 if (memslot
->id
>= KVM_MEMORY_SLOTS
)
6382 map_flags
= MAP_SHARED
| MAP_ANONYMOUS
;
6384 /*To keep backward compatibility with older userspace,
6385 *x86 needs to handle !user_alloc case.
6388 if (npages
&& !old
.rmap
) {
6389 unsigned long userspace_addr
;
6391 userspace_addr
= vm_mmap(NULL
, 0,
6393 PROT_READ
| PROT_WRITE
,
6397 if (IS_ERR((void *)userspace_addr
))
6398 return PTR_ERR((void *)userspace_addr
);
6400 memslot
->userspace_addr
= userspace_addr
;
6408 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
6409 struct kvm_userspace_memory_region
*mem
,
6410 struct kvm_memory_slot old
,
6414 int nr_mmu_pages
= 0, npages
= mem
->memory_size
>> PAGE_SHIFT
;
6416 if (!user_alloc
&& !old
.user_alloc
&& old
.rmap
&& !npages
) {
6419 ret
= vm_munmap(old
.userspace_addr
,
6420 old
.npages
* PAGE_SIZE
);
6423 "kvm_vm_ioctl_set_memory_region: "
6424 "failed to munmap memory\n");
6427 if (!kvm
->arch
.n_requested_mmu_pages
)
6428 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
6430 spin_lock(&kvm
->mmu_lock
);
6432 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
6433 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
6434 spin_unlock(&kvm
->mmu_lock
);
6437 void kvm_arch_flush_shadow(struct kvm
*kvm
)
6439 kvm_mmu_zap_all(kvm
);
6440 kvm_reload_remote_mmus(kvm
);
6443 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
6445 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
6446 !vcpu
->arch
.apf
.halted
)
6447 || !list_empty_careful(&vcpu
->async_pf
.done
)
6448 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
6449 || atomic_read(&vcpu
->arch
.nmi_queued
) ||
6450 (kvm_arch_interrupt_allowed(vcpu
) &&
6451 kvm_cpu_has_interrupt(vcpu
));
6454 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
6456 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
6459 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
6461 return kvm_x86_ops
->interrupt_allowed(vcpu
);
6464 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
6466 unsigned long current_rip
= kvm_rip_read(vcpu
) +
6467 get_segment_base(vcpu
, VCPU_SREG_CS
);
6469 return current_rip
== linear_rip
;
6471 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
6473 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
6475 unsigned long rflags
;
6477 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
6478 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
6479 rflags
&= ~X86_EFLAGS_TF
;
6482 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
6484 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
6486 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
6487 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
6488 rflags
|= X86_EFLAGS_TF
;
6489 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
6490 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6492 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
6494 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
6498 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
6499 is_error_page(work
->page
))
6502 r
= kvm_mmu_reload(vcpu
);
6506 if (!vcpu
->arch
.mmu
.direct_map
&&
6507 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
6510 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
6513 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
6515 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
6518 static inline u32
kvm_async_pf_next_probe(u32 key
)
6520 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
6523 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6525 u32 key
= kvm_async_pf_hash_fn(gfn
);
6527 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
6528 key
= kvm_async_pf_next_probe(key
);
6530 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
6533 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6536 u32 key
= kvm_async_pf_hash_fn(gfn
);
6538 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
6539 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
6540 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
6541 key
= kvm_async_pf_next_probe(key
);
6546 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6548 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
6551 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6555 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
6557 vcpu
->arch
.apf
.gfns
[i
] = ~0;
6559 j
= kvm_async_pf_next_probe(j
);
6560 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
6562 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
6564 * k lies cyclically in ]i,j]
6566 * |....j i.k.| or |.k..j i...|
6568 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
6569 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
6574 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
6577 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
6581 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
6582 struct kvm_async_pf
*work
)
6584 struct x86_exception fault
;
6586 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
6587 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
6589 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
6590 (vcpu
->arch
.apf
.send_user_only
&&
6591 kvm_x86_ops
->get_cpl(vcpu
) == 0))
6592 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
6593 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
6594 fault
.vector
= PF_VECTOR
;
6595 fault
.error_code_valid
= true;
6596 fault
.error_code
= 0;
6597 fault
.nested_page_fault
= false;
6598 fault
.address
= work
->arch
.token
;
6599 kvm_inject_page_fault(vcpu
, &fault
);
6603 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
6604 struct kvm_async_pf
*work
)
6606 struct x86_exception fault
;
6608 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
6609 if (is_error_page(work
->page
))
6610 work
->arch
.token
= ~0; /* broadcast wakeup */
6612 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
6614 if ((vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) &&
6615 !apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
6616 fault
.vector
= PF_VECTOR
;
6617 fault
.error_code_valid
= true;
6618 fault
.error_code
= 0;
6619 fault
.nested_page_fault
= false;
6620 fault
.address
= work
->arch
.token
;
6621 kvm_inject_page_fault(vcpu
, &fault
);
6623 vcpu
->arch
.apf
.halted
= false;
6624 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6627 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
6629 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
6632 return !kvm_event_needs_reinjection(vcpu
) &&
6633 kvm_x86_ops
->interrupt_allowed(vcpu
);
6636 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
6637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
6638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
6639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
6640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
6641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
6642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
6643 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
6644 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
6645 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
6646 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
6647 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);