2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <trace/events/kvm.h>
49 #define CREATE_TRACE_POINTS
52 #include <asm/debugreg.h>
59 #include <asm/pvclock.h>
60 #include <asm/div64.h>
62 #define MAX_IO_MSRS 256
63 #define KVM_MAX_MCE_BANKS 32
64 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
66 #define emul_to_vcpu(ctxt) \
67 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
70 * - enable syscall per default because its emulated by KVM
71 * - enable LME and LMA per default on 64 bit KVM
75 u64 __read_mostly efer_reserved_bits
= ~((u64
)(EFER_SCE
| EFER_LME
| EFER_LMA
));
77 static u64 __read_mostly efer_reserved_bits
= ~((u64
)EFER_SCE
);
80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
83 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
);
84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
85 struct kvm_cpuid_entry2 __user
*entries
);
87 struct kvm_x86_ops
*kvm_x86_ops
;
88 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
91 module_param_named(ignore_msrs
, ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
93 bool kvm_has_tsc_control
;
94 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
95 u32 kvm_max_guest_tsc_khz
;
96 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
98 #define KVM_NR_SHARED_MSRS 16
100 struct kvm_shared_msrs_global
{
102 u32 msrs
[KVM_NR_SHARED_MSRS
];
105 struct kvm_shared_msrs
{
106 struct user_return_notifier urn
;
108 struct kvm_shared_msr_values
{
111 } values
[KVM_NR_SHARED_MSRS
];
114 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
115 static DEFINE_PER_CPU(struct kvm_shared_msrs
, shared_msrs
);
117 struct kvm_stats_debugfs_item debugfs_entries
[] = {
118 { "pf_fixed", VCPU_STAT(pf_fixed
) },
119 { "pf_guest", VCPU_STAT(pf_guest
) },
120 { "tlb_flush", VCPU_STAT(tlb_flush
) },
121 { "invlpg", VCPU_STAT(invlpg
) },
122 { "exits", VCPU_STAT(exits
) },
123 { "io_exits", VCPU_STAT(io_exits
) },
124 { "mmio_exits", VCPU_STAT(mmio_exits
) },
125 { "signal_exits", VCPU_STAT(signal_exits
) },
126 { "irq_window", VCPU_STAT(irq_window_exits
) },
127 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
128 { "halt_exits", VCPU_STAT(halt_exits
) },
129 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
130 { "hypercalls", VCPU_STAT(hypercalls
) },
131 { "request_irq", VCPU_STAT(request_irq_exits
) },
132 { "irq_exits", VCPU_STAT(irq_exits
) },
133 { "host_state_reload", VCPU_STAT(host_state_reload
) },
134 { "efer_reload", VCPU_STAT(efer_reload
) },
135 { "fpu_reload", VCPU_STAT(fpu_reload
) },
136 { "insn_emulation", VCPU_STAT(insn_emulation
) },
137 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
138 { "irq_injections", VCPU_STAT(irq_injections
) },
139 { "nmi_injections", VCPU_STAT(nmi_injections
) },
140 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
141 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
142 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
143 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
144 { "mmu_flooded", VM_STAT(mmu_flooded
) },
145 { "mmu_recycled", VM_STAT(mmu_recycled
) },
146 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
147 { "mmu_unsync", VM_STAT(mmu_unsync
) },
148 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
149 { "largepages", VM_STAT(lpages
) },
153 u64 __read_mostly host_xcr0
;
155 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
158 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
159 vcpu
->arch
.apf
.gfns
[i
] = ~0;
162 static void kvm_on_user_return(struct user_return_notifier
*urn
)
165 struct kvm_shared_msrs
*locals
166 = container_of(urn
, struct kvm_shared_msrs
, urn
);
167 struct kvm_shared_msr_values
*values
;
169 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
170 values
= &locals
->values
[slot
];
171 if (values
->host
!= values
->curr
) {
172 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
173 values
->curr
= values
->host
;
176 locals
->registered
= false;
177 user_return_notifier_unregister(urn
);
180 static void shared_msr_update(unsigned slot
, u32 msr
)
182 struct kvm_shared_msrs
*smsr
;
185 smsr
= &__get_cpu_var(shared_msrs
);
186 /* only read, and nobody should modify it at this time,
187 * so don't need lock */
188 if (slot
>= shared_msrs_global
.nr
) {
189 printk(KERN_ERR
"kvm: invalid MSR slot!");
192 rdmsrl_safe(msr
, &value
);
193 smsr
->values
[slot
].host
= value
;
194 smsr
->values
[slot
].curr
= value
;
197 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
199 if (slot
>= shared_msrs_global
.nr
)
200 shared_msrs_global
.nr
= slot
+ 1;
201 shared_msrs_global
.msrs
[slot
] = msr
;
202 /* we need ensured the shared_msr_global have been updated */
205 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
207 static void kvm_shared_msr_cpu_online(void)
211 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
212 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
215 void kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
217 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
219 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
221 smsr
->values
[slot
].curr
= value
;
222 wrmsrl(shared_msrs_global
.msrs
[slot
], value
);
223 if (!smsr
->registered
) {
224 smsr
->urn
.on_user_return
= kvm_on_user_return
;
225 user_return_notifier_register(&smsr
->urn
);
226 smsr
->registered
= true;
229 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
231 static void drop_user_return_notifiers(void *ignore
)
233 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
235 if (smsr
->registered
)
236 kvm_on_user_return(&smsr
->urn
);
239 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
241 if (irqchip_in_kernel(vcpu
->kvm
))
242 return vcpu
->arch
.apic_base
;
244 return vcpu
->arch
.apic_base
;
246 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
248 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
250 /* TODO: reserve bits check */
251 if (irqchip_in_kernel(vcpu
->kvm
))
252 kvm_lapic_set_base(vcpu
, data
);
254 vcpu
->arch
.apic_base
= data
;
256 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
258 #define EXCPT_BENIGN 0
259 #define EXCPT_CONTRIBUTORY 1
262 static int exception_class(int vector
)
272 return EXCPT_CONTRIBUTORY
;
279 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
280 unsigned nr
, bool has_error
, u32 error_code
,
286 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
288 if (!vcpu
->arch
.exception
.pending
) {
290 vcpu
->arch
.exception
.pending
= true;
291 vcpu
->arch
.exception
.has_error_code
= has_error
;
292 vcpu
->arch
.exception
.nr
= nr
;
293 vcpu
->arch
.exception
.error_code
= error_code
;
294 vcpu
->arch
.exception
.reinject
= reinject
;
298 /* to check exception */
299 prev_nr
= vcpu
->arch
.exception
.nr
;
300 if (prev_nr
== DF_VECTOR
) {
301 /* triple fault -> shutdown */
302 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
305 class1
= exception_class(prev_nr
);
306 class2
= exception_class(nr
);
307 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
308 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
309 /* generate double fault per SDM Table 5-5 */
310 vcpu
->arch
.exception
.pending
= true;
311 vcpu
->arch
.exception
.has_error_code
= true;
312 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
313 vcpu
->arch
.exception
.error_code
= 0;
315 /* replace previous exception with a new one in a hope
316 that instruction re-execution will regenerate lost
321 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
323 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
325 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
327 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
329 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
331 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
333 void kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
336 kvm_inject_gp(vcpu
, 0);
338 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
340 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
342 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
344 ++vcpu
->stat
.pf_guest
;
345 vcpu
->arch
.cr2
= fault
->address
;
346 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
349 void kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
351 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
352 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
354 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
357 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
359 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
360 vcpu
->arch
.nmi_pending
= 1;
362 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
364 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
366 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
368 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
370 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
372 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
374 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
377 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
378 * a #GP and return false.
380 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
382 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
384 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
387 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
390 * This function will be used to read from the physical memory of the currently
391 * running guest. The difference to kvm_read_guest_page is that this function
392 * can read from guest physical or from the guest's guest physical memory.
394 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
395 gfn_t ngfn
, void *data
, int offset
, int len
,
401 ngpa
= gfn_to_gpa(ngfn
);
402 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
);
403 if (real_gfn
== UNMAPPED_GVA
)
406 real_gfn
= gpa_to_gfn(real_gfn
);
408 return kvm_read_guest_page(vcpu
->kvm
, real_gfn
, data
, offset
, len
);
410 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
412 int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
413 void *data
, int offset
, int len
, u32 access
)
415 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
416 data
, offset
, len
, access
);
420 * Load the pae pdptrs. Return true is they are all valid.
422 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
424 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
425 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
428 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
430 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
431 offset
* sizeof(u64
), sizeof(pdpte
),
432 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
437 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
438 if (is_present_gpte(pdpte
[i
]) &&
439 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
446 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
447 __set_bit(VCPU_EXREG_PDPTR
,
448 (unsigned long *)&vcpu
->arch
.regs_avail
);
449 __set_bit(VCPU_EXREG_PDPTR
,
450 (unsigned long *)&vcpu
->arch
.regs_dirty
);
455 EXPORT_SYMBOL_GPL(load_pdptrs
);
457 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
459 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
465 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
468 if (!test_bit(VCPU_EXREG_PDPTR
,
469 (unsigned long *)&vcpu
->arch
.regs_avail
))
472 gfn
= (kvm_read_cr3(vcpu
) & ~31u) >> PAGE_SHIFT
;
473 offset
= (kvm_read_cr3(vcpu
) & ~31u) & (PAGE_SIZE
- 1);
474 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
475 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
478 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
484 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
486 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
487 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
|
488 X86_CR0_CD
| X86_CR0_NW
;
493 if (cr0
& 0xffffffff00000000UL
)
497 cr0
&= ~CR0_RESERVED_BITS
;
499 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
502 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
505 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
507 if ((vcpu
->arch
.efer
& EFER_LME
)) {
512 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
517 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
522 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
524 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
525 kvm_clear_async_pf_completion_queue(vcpu
);
526 kvm_async_pf_hash_reset(vcpu
);
529 if ((cr0
^ old_cr0
) & update_bits
)
530 kvm_mmu_reset_context(vcpu
);
533 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
535 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
537 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
539 EXPORT_SYMBOL_GPL(kvm_lmsw
);
541 int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
545 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
546 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
549 if (kvm_x86_ops
->get_cpl(vcpu
) != 0)
551 if (!(xcr0
& XSTATE_FP
))
553 if ((xcr0
& XSTATE_YMM
) && !(xcr0
& XSTATE_SSE
))
555 if (xcr0
& ~host_xcr0
)
557 vcpu
->arch
.xcr0
= xcr0
;
558 vcpu
->guest_xcr0_loaded
= 0;
562 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
564 if (__kvm_set_xcr(vcpu
, index
, xcr
)) {
565 kvm_inject_gp(vcpu
, 0);
570 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
572 static bool guest_cpuid_has_xsave(struct kvm_vcpu
*vcpu
)
574 struct kvm_cpuid_entry2
*best
;
576 best
= kvm_find_cpuid_entry(vcpu
, 1, 0);
577 return best
&& (best
->ecx
& bit(X86_FEATURE_XSAVE
));
580 static void update_cpuid(struct kvm_vcpu
*vcpu
)
582 struct kvm_cpuid_entry2
*best
;
584 best
= kvm_find_cpuid_entry(vcpu
, 1, 0);
588 /* Update OSXSAVE bit */
589 if (cpu_has_xsave
&& best
->function
== 0x1) {
590 best
->ecx
&= ~(bit(X86_FEATURE_OSXSAVE
));
591 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
))
592 best
->ecx
|= bit(X86_FEATURE_OSXSAVE
);
596 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
598 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
599 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
;
601 if (cr4
& CR4_RESERVED_BITS
)
604 if (!guest_cpuid_has_xsave(vcpu
) && (cr4
& X86_CR4_OSXSAVE
))
607 if (is_long_mode(vcpu
)) {
608 if (!(cr4
& X86_CR4_PAE
))
610 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
611 && ((cr4
^ old_cr4
) & pdptr_bits
)
612 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
616 if (cr4
& X86_CR4_VMXE
)
619 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
621 if ((cr4
^ old_cr4
) & pdptr_bits
)
622 kvm_mmu_reset_context(vcpu
);
624 if ((cr4
^ old_cr4
) & X86_CR4_OSXSAVE
)
629 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
631 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
633 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
634 kvm_mmu_sync_roots(vcpu
);
635 kvm_mmu_flush_tlb(vcpu
);
639 if (is_long_mode(vcpu
)) {
640 if (cr3
& CR3_L_MODE_RESERVED_BITS
)
644 if (cr3
& CR3_PAE_RESERVED_BITS
)
646 if (is_paging(vcpu
) &&
647 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
651 * We don't check reserved bits in nonpae mode, because
652 * this isn't enforced, and VMware depends on this.
657 * Does the new cr3 value map to physical memory? (Note, we
658 * catch an invalid cr3 even in real-mode, because it would
659 * cause trouble later on when we turn on paging anyway.)
661 * A real CPU would silently accept an invalid cr3 and would
662 * attempt to use it - with largely undefined (and often hard
663 * to debug) behavior on the guest side.
665 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
667 vcpu
->arch
.cr3
= cr3
;
668 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
669 vcpu
->arch
.mmu
.new_cr3(vcpu
);
672 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
674 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
676 if (cr8
& CR8_RESERVED_BITS
)
678 if (irqchip_in_kernel(vcpu
->kvm
))
679 kvm_lapic_set_tpr(vcpu
, cr8
);
681 vcpu
->arch
.cr8
= cr8
;
684 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
686 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
688 if (irqchip_in_kernel(vcpu
->kvm
))
689 return kvm_lapic_get_cr8(vcpu
);
691 return vcpu
->arch
.cr8
;
693 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
695 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
699 vcpu
->arch
.db
[dr
] = val
;
700 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
701 vcpu
->arch
.eff_db
[dr
] = val
;
704 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
708 if (val
& 0xffffffff00000000ULL
)
710 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | DR6_FIXED_1
;
713 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
717 if (val
& 0xffffffff00000000ULL
)
719 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
720 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
721 kvm_x86_ops
->set_dr7(vcpu
, vcpu
->arch
.dr7
);
722 vcpu
->arch
.switch_db_regs
= (val
& DR7_BP_EN_MASK
);
730 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
734 res
= __kvm_set_dr(vcpu
, dr
, val
);
736 kvm_queue_exception(vcpu
, UD_VECTOR
);
738 kvm_inject_gp(vcpu
, 0);
742 EXPORT_SYMBOL_GPL(kvm_set_dr
);
744 static int _kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
748 *val
= vcpu
->arch
.db
[dr
];
751 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
755 *val
= vcpu
->arch
.dr6
;
758 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
762 *val
= vcpu
->arch
.dr7
;
769 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
771 if (_kvm_get_dr(vcpu
, dr
, val
)) {
772 kvm_queue_exception(vcpu
, UD_VECTOR
);
777 EXPORT_SYMBOL_GPL(kvm_get_dr
);
780 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
781 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
783 * This list is modified at module load time to reflect the
784 * capabilities of the host cpu. This capabilities test skips MSRs that are
785 * kvm-specific. Those are put in the beginning of the list.
788 #define KVM_SAVE_MSRS_BEGIN 8
789 static u32 msrs_to_save
[] = {
790 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
791 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
792 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
793 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
,
794 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
797 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
799 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
802 static unsigned num_msrs_to_save
;
804 static u32 emulated_msrs
[] = {
805 MSR_IA32_MISC_ENABLE
,
810 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
812 u64 old_efer
= vcpu
->arch
.efer
;
814 if (efer
& efer_reserved_bits
)
818 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
821 if (efer
& EFER_FFXSR
) {
822 struct kvm_cpuid_entry2
*feat
;
824 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
825 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
)))
829 if (efer
& EFER_SVME
) {
830 struct kvm_cpuid_entry2
*feat
;
832 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
833 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
)))
838 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
840 kvm_x86_ops
->set_efer(vcpu
, efer
);
842 vcpu
->arch
.mmu
.base_role
.nxe
= (efer
& EFER_NX
) && !tdp_enabled
;
844 /* Update reserved bits */
845 if ((efer
^ old_efer
) & EFER_NX
)
846 kvm_mmu_reset_context(vcpu
);
851 void kvm_enable_efer_bits(u64 mask
)
853 efer_reserved_bits
&= ~mask
;
855 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
859 * Writes msr value into into the appropriate "register".
860 * Returns 0 on success, non-0 otherwise.
861 * Assumes vcpu_load() was already called.
863 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
865 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
869 * Adapt set_msr() to msr_io()'s calling convention
871 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
873 return kvm_set_msr(vcpu
, index
, *data
);
876 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
880 struct pvclock_wall_clock wc
;
881 struct timespec boot
;
886 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
891 ++version
; /* first time write, random junk */
895 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
898 * The guest calculates current wall clock time by adding
899 * system time (updated by kvm_guest_time_update below) to the
900 * wall clock specified here. guest system time equals host
901 * system time for us, thus we must fill in host boot time here.
905 wc
.sec
= boot
.tv_sec
;
906 wc
.nsec
= boot
.tv_nsec
;
907 wc
.version
= version
;
909 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
912 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
915 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
917 uint32_t quotient
, remainder
;
919 /* Don't try to replace with do_div(), this one calculates
920 * "(dividend << 32) / divisor" */
922 : "=a" (quotient
), "=d" (remainder
)
923 : "0" (0), "1" (dividend
), "r" (divisor
) );
927 static void kvm_get_time_scale(uint32_t scaled_khz
, uint32_t base_khz
,
928 s8
*pshift
, u32
*pmultiplier
)
935 tps64
= base_khz
* 1000LL;
936 scaled64
= scaled_khz
* 1000LL;
937 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
942 tps32
= (uint32_t)tps64
;
943 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
944 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
952 *pmultiplier
= div_frac(scaled64
, tps32
);
954 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
955 __func__
, base_khz
, scaled_khz
, shift
, *pmultiplier
);
958 static inline u64
get_kernel_ns(void)
962 WARN_ON(preemptible());
964 monotonic_to_bootbased(&ts
);
965 return timespec_to_ns(&ts
);
968 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
969 unsigned long max_tsc_khz
;
971 static inline int kvm_tsc_changes_freq(void)
974 int ret
= !boot_cpu_has(X86_FEATURE_CONSTANT_TSC
) &&
975 cpufreq_quick_get(cpu
) != 0;
980 static u64
vcpu_tsc_khz(struct kvm_vcpu
*vcpu
)
982 if (vcpu
->arch
.virtual_tsc_khz
)
983 return vcpu
->arch
.virtual_tsc_khz
;
985 return __this_cpu_read(cpu_tsc_khz
);
988 static inline u64
nsec_to_cycles(struct kvm_vcpu
*vcpu
, u64 nsec
)
992 WARN_ON(preemptible());
993 if (kvm_tsc_changes_freq())
994 printk_once(KERN_WARNING
995 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
996 ret
= nsec
* vcpu_tsc_khz(vcpu
);
997 do_div(ret
, USEC_PER_SEC
);
1001 static void kvm_init_tsc_catchup(struct kvm_vcpu
*vcpu
, u32 this_tsc_khz
)
1003 /* Compute a scale to convert nanoseconds in TSC cycles */
1004 kvm_get_time_scale(this_tsc_khz
, NSEC_PER_SEC
/ 1000,
1005 &vcpu
->arch
.tsc_catchup_shift
,
1006 &vcpu
->arch
.tsc_catchup_mult
);
1009 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1011 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.last_tsc_nsec
,
1012 vcpu
->arch
.tsc_catchup_mult
,
1013 vcpu
->arch
.tsc_catchup_shift
);
1014 tsc
+= vcpu
->arch
.last_tsc_write
;
1018 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, u64 data
)
1020 struct kvm
*kvm
= vcpu
->kvm
;
1021 u64 offset
, ns
, elapsed
;
1022 unsigned long flags
;
1025 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1026 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1027 ns
= get_kernel_ns();
1028 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1029 sdiff
= data
- kvm
->arch
.last_tsc_write
;
1034 * Special case: close write to TSC within 5 seconds of
1035 * another CPU is interpreted as an attempt to synchronize
1036 * The 5 seconds is to accommodate host load / swapping as
1037 * well as any reset of TSC during the boot process.
1039 * In that case, for a reliable TSC, we can match TSC offsets,
1040 * or make a best guest using elapsed value.
1042 if (sdiff
< nsec_to_cycles(vcpu
, 5ULL * NSEC_PER_SEC
) &&
1043 elapsed
< 5ULL * NSEC_PER_SEC
) {
1044 if (!check_tsc_unstable()) {
1045 offset
= kvm
->arch
.last_tsc_offset
;
1046 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1048 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1050 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1052 ns
= kvm
->arch
.last_tsc_nsec
;
1054 kvm
->arch
.last_tsc_nsec
= ns
;
1055 kvm
->arch
.last_tsc_write
= data
;
1056 kvm
->arch
.last_tsc_offset
= offset
;
1057 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1058 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1060 /* Reset of TSC must disable overshoot protection below */
1061 vcpu
->arch
.hv_clock
.tsc_timestamp
= 0;
1062 vcpu
->arch
.last_tsc_write
= data
;
1063 vcpu
->arch
.last_tsc_nsec
= ns
;
1065 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1067 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1069 unsigned long flags
;
1070 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1072 unsigned long this_tsc_khz
;
1073 s64 kernel_ns
, max_kernel_ns
;
1076 /* Keep irq disabled to prevent changes to the clock */
1077 local_irq_save(flags
);
1078 kvm_get_msr(v
, MSR_IA32_TSC
, &tsc_timestamp
);
1079 kernel_ns
= get_kernel_ns();
1080 this_tsc_khz
= vcpu_tsc_khz(v
);
1081 if (unlikely(this_tsc_khz
== 0)) {
1082 local_irq_restore(flags
);
1083 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1088 * We may have to catch up the TSC to match elapsed wall clock
1089 * time for two reasons, even if kvmclock is used.
1090 * 1) CPU could have been running below the maximum TSC rate
1091 * 2) Broken TSC compensation resets the base at each VCPU
1092 * entry to avoid unknown leaps of TSC even when running
1093 * again on the same CPU. This may cause apparent elapsed
1094 * time to disappear, and the guest to stand still or run
1097 if (vcpu
->tsc_catchup
) {
1098 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1099 if (tsc
> tsc_timestamp
) {
1100 kvm_x86_ops
->adjust_tsc_offset(v
, tsc
- tsc_timestamp
);
1101 tsc_timestamp
= tsc
;
1105 local_irq_restore(flags
);
1107 if (!vcpu
->time_page
)
1111 * Time as measured by the TSC may go backwards when resetting the base
1112 * tsc_timestamp. The reason for this is that the TSC resolution is
1113 * higher than the resolution of the other clock scales. Thus, many
1114 * possible measurments of the TSC correspond to one measurement of any
1115 * other clock, and so a spread of values is possible. This is not a
1116 * problem for the computation of the nanosecond clock; with TSC rates
1117 * around 1GHZ, there can only be a few cycles which correspond to one
1118 * nanosecond value, and any path through this code will inevitably
1119 * take longer than that. However, with the kernel_ns value itself,
1120 * the precision may be much lower, down to HZ granularity. If the
1121 * first sampling of TSC against kernel_ns ends in the low part of the
1122 * range, and the second in the high end of the range, we can get:
1124 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1126 * As the sampling errors potentially range in the thousands of cycles,
1127 * it is possible such a time value has already been observed by the
1128 * guest. To protect against this, we must compute the system time as
1129 * observed by the guest and ensure the new system time is greater.
1132 if (vcpu
->hv_clock
.tsc_timestamp
&& vcpu
->last_guest_tsc
) {
1133 max_kernel_ns
= vcpu
->last_guest_tsc
-
1134 vcpu
->hv_clock
.tsc_timestamp
;
1135 max_kernel_ns
= pvclock_scale_delta(max_kernel_ns
,
1136 vcpu
->hv_clock
.tsc_to_system_mul
,
1137 vcpu
->hv_clock
.tsc_shift
);
1138 max_kernel_ns
+= vcpu
->last_kernel_ns
;
1141 if (unlikely(vcpu
->hw_tsc_khz
!= this_tsc_khz
)) {
1142 kvm_get_time_scale(NSEC_PER_SEC
/ 1000, this_tsc_khz
,
1143 &vcpu
->hv_clock
.tsc_shift
,
1144 &vcpu
->hv_clock
.tsc_to_system_mul
);
1145 vcpu
->hw_tsc_khz
= this_tsc_khz
;
1148 if (max_kernel_ns
> kernel_ns
)
1149 kernel_ns
= max_kernel_ns
;
1151 /* With all the info we got, fill in the values */
1152 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1153 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1154 vcpu
->last_kernel_ns
= kernel_ns
;
1155 vcpu
->last_guest_tsc
= tsc_timestamp
;
1156 vcpu
->hv_clock
.flags
= 0;
1159 * The interface expects us to write an even number signaling that the
1160 * update is finished. Since the guest won't see the intermediate
1161 * state, we just increase by 2 at the end.
1163 vcpu
->hv_clock
.version
+= 2;
1165 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
1167 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
1168 sizeof(vcpu
->hv_clock
));
1170 kunmap_atomic(shared_kaddr
, KM_USER0
);
1172 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
1176 static bool msr_mtrr_valid(unsigned msr
)
1179 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
1180 case MSR_MTRRfix64K_00000
:
1181 case MSR_MTRRfix16K_80000
:
1182 case MSR_MTRRfix16K_A0000
:
1183 case MSR_MTRRfix4K_C0000
:
1184 case MSR_MTRRfix4K_C8000
:
1185 case MSR_MTRRfix4K_D0000
:
1186 case MSR_MTRRfix4K_D8000
:
1187 case MSR_MTRRfix4K_E0000
:
1188 case MSR_MTRRfix4K_E8000
:
1189 case MSR_MTRRfix4K_F0000
:
1190 case MSR_MTRRfix4K_F8000
:
1191 case MSR_MTRRdefType
:
1192 case MSR_IA32_CR_PAT
:
1200 static bool valid_pat_type(unsigned t
)
1202 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1205 static bool valid_mtrr_type(unsigned t
)
1207 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
1210 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1214 if (!msr_mtrr_valid(msr
))
1217 if (msr
== MSR_IA32_CR_PAT
) {
1218 for (i
= 0; i
< 8; i
++)
1219 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
1222 } else if (msr
== MSR_MTRRdefType
) {
1225 return valid_mtrr_type(data
& 0xff);
1226 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
1227 for (i
= 0; i
< 8 ; i
++)
1228 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
1233 /* variable MTRRs */
1234 return valid_mtrr_type(data
& 0xff);
1237 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1239 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1241 if (!mtrr_valid(vcpu
, msr
, data
))
1244 if (msr
== MSR_MTRRdefType
) {
1245 vcpu
->arch
.mtrr_state
.def_type
= data
;
1246 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
1247 } else if (msr
== MSR_MTRRfix64K_00000
)
1249 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1250 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
1251 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1252 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
1253 else if (msr
== MSR_IA32_CR_PAT
)
1254 vcpu
->arch
.pat
= data
;
1255 else { /* Variable MTRRs */
1256 int idx
, is_mtrr_mask
;
1259 idx
= (msr
- 0x200) / 2;
1260 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1263 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1266 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1270 kvm_mmu_reset_context(vcpu
);
1274 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1276 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1277 unsigned bank_num
= mcg_cap
& 0xff;
1280 case MSR_IA32_MCG_STATUS
:
1281 vcpu
->arch
.mcg_status
= data
;
1283 case MSR_IA32_MCG_CTL
:
1284 if (!(mcg_cap
& MCG_CTL_P
))
1286 if (data
!= 0 && data
!= ~(u64
)0)
1288 vcpu
->arch
.mcg_ctl
= data
;
1291 if (msr
>= MSR_IA32_MC0_CTL
&&
1292 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1293 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1294 /* only 0 or all 1s can be written to IA32_MCi_CTL
1295 * some Linux kernels though clear bit 10 in bank 4 to
1296 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1297 * this to avoid an uncatched #GP in the guest
1299 if ((offset
& 0x3) == 0 &&
1300 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
1302 vcpu
->arch
.mce_banks
[offset
] = data
;
1310 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
1312 struct kvm
*kvm
= vcpu
->kvm
;
1313 int lm
= is_long_mode(vcpu
);
1314 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
1315 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
1316 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
1317 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
1318 u32 page_num
= data
& ~PAGE_MASK
;
1319 u64 page_addr
= data
& PAGE_MASK
;
1324 if (page_num
>= blob_size
)
1327 page
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
1331 if (copy_from_user(page
, blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
))
1333 if (kvm_write_guest(kvm
, page_addr
, page
, PAGE_SIZE
))
1342 static bool kvm_hv_hypercall_enabled(struct kvm
*kvm
)
1344 return kvm
->arch
.hv_hypercall
& HV_X64_MSR_HYPERCALL_ENABLE
;
1347 static bool kvm_hv_msr_partition_wide(u32 msr
)
1351 case HV_X64_MSR_GUEST_OS_ID
:
1352 case HV_X64_MSR_HYPERCALL
:
1360 static int set_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1362 struct kvm
*kvm
= vcpu
->kvm
;
1365 case HV_X64_MSR_GUEST_OS_ID
:
1366 kvm
->arch
.hv_guest_os_id
= data
;
1367 /* setting guest os id to zero disables hypercall page */
1368 if (!kvm
->arch
.hv_guest_os_id
)
1369 kvm
->arch
.hv_hypercall
&= ~HV_X64_MSR_HYPERCALL_ENABLE
;
1371 case HV_X64_MSR_HYPERCALL
: {
1376 /* if guest os id is not set hypercall should remain disabled */
1377 if (!kvm
->arch
.hv_guest_os_id
)
1379 if (!(data
& HV_X64_MSR_HYPERCALL_ENABLE
)) {
1380 kvm
->arch
.hv_hypercall
= data
;
1383 gfn
= data
>> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT
;
1384 addr
= gfn_to_hva(kvm
, gfn
);
1385 if (kvm_is_error_hva(addr
))
1387 kvm_x86_ops
->patch_hypercall(vcpu
, instructions
);
1388 ((unsigned char *)instructions
)[3] = 0xc3; /* ret */
1389 if (copy_to_user((void __user
*)addr
, instructions
, 4))
1391 kvm
->arch
.hv_hypercall
= data
;
1395 pr_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1396 "data 0x%llx\n", msr
, data
);
1402 static int set_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1405 case HV_X64_MSR_APIC_ASSIST_PAGE
: {
1408 if (!(data
& HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE
)) {
1409 vcpu
->arch
.hv_vapic
= data
;
1412 addr
= gfn_to_hva(vcpu
->kvm
, data
>>
1413 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT
);
1414 if (kvm_is_error_hva(addr
))
1416 if (clear_user((void __user
*)addr
, PAGE_SIZE
))
1418 vcpu
->arch
.hv_vapic
= data
;
1421 case HV_X64_MSR_EOI
:
1422 return kvm_hv_vapic_msr_write(vcpu
, APIC_EOI
, data
);
1423 case HV_X64_MSR_ICR
:
1424 return kvm_hv_vapic_msr_write(vcpu
, APIC_ICR
, data
);
1425 case HV_X64_MSR_TPR
:
1426 return kvm_hv_vapic_msr_write(vcpu
, APIC_TASKPRI
, data
);
1428 pr_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1429 "data 0x%llx\n", msr
, data
);
1436 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
1438 gpa_t gpa
= data
& ~0x3f;
1440 /* Bits 2:5 are resrved, Should be zero */
1444 vcpu
->arch
.apf
.msr_val
= data
;
1446 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
1447 kvm_clear_async_pf_completion_queue(vcpu
);
1448 kvm_async_pf_hash_reset(vcpu
);
1452 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
))
1455 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
1456 kvm_async_pf_wakeup_all(vcpu
);
1460 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
1462 if (vcpu
->arch
.time_page
) {
1463 kvm_release_page_dirty(vcpu
->arch
.time_page
);
1464 vcpu
->arch
.time_page
= NULL
;
1468 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1472 return set_efer(vcpu
, data
);
1474 data
&= ~(u64
)0x40; /* ignore flush filter disable */
1475 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
1477 pr_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
1482 case MSR_FAM10H_MMIO_CONF_BASE
:
1484 pr_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
1489 case MSR_AMD64_NB_CFG
:
1491 case MSR_IA32_DEBUGCTLMSR
:
1493 /* We support the non-activated case already */
1495 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
1496 /* Values other than LBR and BTF are vendor-specific,
1497 thus reserved and should throw a #GP */
1500 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1503 case MSR_IA32_UCODE_REV
:
1504 case MSR_IA32_UCODE_WRITE
:
1505 case MSR_VM_HSAVE_PA
:
1506 case MSR_AMD64_PATCH_LOADER
:
1508 case 0x200 ... 0x2ff:
1509 return set_msr_mtrr(vcpu
, msr
, data
);
1510 case MSR_IA32_APICBASE
:
1511 kvm_set_apic_base(vcpu
, data
);
1513 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1514 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
1515 case MSR_IA32_MISC_ENABLE
:
1516 vcpu
->arch
.ia32_misc_enable_msr
= data
;
1518 case MSR_KVM_WALL_CLOCK_NEW
:
1519 case MSR_KVM_WALL_CLOCK
:
1520 vcpu
->kvm
->arch
.wall_clock
= data
;
1521 kvm_write_wall_clock(vcpu
->kvm
, data
);
1523 case MSR_KVM_SYSTEM_TIME_NEW
:
1524 case MSR_KVM_SYSTEM_TIME
: {
1525 kvmclock_reset(vcpu
);
1527 vcpu
->arch
.time
= data
;
1528 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1530 /* we verify if the enable bit is set... */
1534 /* ...but clean it before doing the actual write */
1535 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
1537 vcpu
->arch
.time_page
=
1538 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
1540 if (is_error_page(vcpu
->arch
.time_page
)) {
1541 kvm_release_page_clean(vcpu
->arch
.time_page
);
1542 vcpu
->arch
.time_page
= NULL
;
1546 case MSR_KVM_ASYNC_PF_EN
:
1547 if (kvm_pv_enable_async_pf(vcpu
, data
))
1550 case MSR_IA32_MCG_CTL
:
1551 case MSR_IA32_MCG_STATUS
:
1552 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1553 return set_msr_mce(vcpu
, msr
, data
);
1555 /* Performance counters are not protected by a CPUID bit,
1556 * so we should check all of them in the generic path for the sake of
1557 * cross vendor migration.
1558 * Writing a zero into the event select MSRs disables them,
1559 * which we perfectly emulate ;-). Any other value should be at least
1560 * reported, some guests depend on them.
1562 case MSR_P6_EVNTSEL0
:
1563 case MSR_P6_EVNTSEL1
:
1564 case MSR_K7_EVNTSEL0
:
1565 case MSR_K7_EVNTSEL1
:
1566 case MSR_K7_EVNTSEL2
:
1567 case MSR_K7_EVNTSEL3
:
1569 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
1570 "0x%x data 0x%llx\n", msr
, data
);
1572 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1573 * so we ignore writes to make it happy.
1575 case MSR_P6_PERFCTR0
:
1576 case MSR_P6_PERFCTR1
:
1577 case MSR_K7_PERFCTR0
:
1578 case MSR_K7_PERFCTR1
:
1579 case MSR_K7_PERFCTR2
:
1580 case MSR_K7_PERFCTR3
:
1581 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
1582 "0x%x data 0x%llx\n", msr
, data
);
1584 case MSR_K7_CLK_CTL
:
1586 * Ignore all writes to this no longer documented MSR.
1587 * Writes are only relevant for old K7 processors,
1588 * all pre-dating SVM, but a recommended workaround from
1589 * AMD for these chips. It is possible to speicify the
1590 * affected processor models on the command line, hence
1591 * the need to ignore the workaround.
1594 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
1595 if (kvm_hv_msr_partition_wide(msr
)) {
1597 mutex_lock(&vcpu
->kvm
->lock
);
1598 r
= set_msr_hyperv_pw(vcpu
, msr
, data
);
1599 mutex_unlock(&vcpu
->kvm
->lock
);
1602 return set_msr_hyperv(vcpu
, msr
, data
);
1604 case MSR_IA32_BBL_CR_CTL3
:
1605 /* Drop writes to this legacy MSR -- see rdmsr
1606 * counterpart for further detail.
1608 pr_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n", msr
, data
);
1611 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
1612 return xen_hvm_config(vcpu
, data
);
1614 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
1618 pr_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
1625 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1629 * Reads an msr value (of 'msr_index') into 'pdata'.
1630 * Returns 0 on success, non-0 otherwise.
1631 * Assumes vcpu_load() was already called.
1633 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1635 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
1638 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1640 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1642 if (!msr_mtrr_valid(msr
))
1645 if (msr
== MSR_MTRRdefType
)
1646 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
1647 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
1648 else if (msr
== MSR_MTRRfix64K_00000
)
1650 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1651 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
1652 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1653 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
1654 else if (msr
== MSR_IA32_CR_PAT
)
1655 *pdata
= vcpu
->arch
.pat
;
1656 else { /* Variable MTRRs */
1657 int idx
, is_mtrr_mask
;
1660 idx
= (msr
- 0x200) / 2;
1661 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1664 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1667 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1674 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1677 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1678 unsigned bank_num
= mcg_cap
& 0xff;
1681 case MSR_IA32_P5_MC_ADDR
:
1682 case MSR_IA32_P5_MC_TYPE
:
1685 case MSR_IA32_MCG_CAP
:
1686 data
= vcpu
->arch
.mcg_cap
;
1688 case MSR_IA32_MCG_CTL
:
1689 if (!(mcg_cap
& MCG_CTL_P
))
1691 data
= vcpu
->arch
.mcg_ctl
;
1693 case MSR_IA32_MCG_STATUS
:
1694 data
= vcpu
->arch
.mcg_status
;
1697 if (msr
>= MSR_IA32_MC0_CTL
&&
1698 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1699 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1700 data
= vcpu
->arch
.mce_banks
[offset
];
1709 static int get_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1712 struct kvm
*kvm
= vcpu
->kvm
;
1715 case HV_X64_MSR_GUEST_OS_ID
:
1716 data
= kvm
->arch
.hv_guest_os_id
;
1718 case HV_X64_MSR_HYPERCALL
:
1719 data
= kvm
->arch
.hv_hypercall
;
1722 pr_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
1730 static int get_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1735 case HV_X64_MSR_VP_INDEX
: {
1738 kvm_for_each_vcpu(r
, v
, vcpu
->kvm
)
1743 case HV_X64_MSR_EOI
:
1744 return kvm_hv_vapic_msr_read(vcpu
, APIC_EOI
, pdata
);
1745 case HV_X64_MSR_ICR
:
1746 return kvm_hv_vapic_msr_read(vcpu
, APIC_ICR
, pdata
);
1747 case HV_X64_MSR_TPR
:
1748 return kvm_hv_vapic_msr_read(vcpu
, APIC_TASKPRI
, pdata
);
1750 pr_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
1757 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1762 case MSR_IA32_PLATFORM_ID
:
1763 case MSR_IA32_UCODE_REV
:
1764 case MSR_IA32_EBL_CR_POWERON
:
1765 case MSR_IA32_DEBUGCTLMSR
:
1766 case MSR_IA32_LASTBRANCHFROMIP
:
1767 case MSR_IA32_LASTBRANCHTOIP
:
1768 case MSR_IA32_LASTINTFROMIP
:
1769 case MSR_IA32_LASTINTTOIP
:
1772 case MSR_VM_HSAVE_PA
:
1773 case MSR_P6_PERFCTR0
:
1774 case MSR_P6_PERFCTR1
:
1775 case MSR_P6_EVNTSEL0
:
1776 case MSR_P6_EVNTSEL1
:
1777 case MSR_K7_EVNTSEL0
:
1778 case MSR_K7_PERFCTR0
:
1779 case MSR_K8_INT_PENDING_MSG
:
1780 case MSR_AMD64_NB_CFG
:
1781 case MSR_FAM10H_MMIO_CONF_BASE
:
1785 data
= 0x500 | KVM_NR_VAR_MTRR
;
1787 case 0x200 ... 0x2ff:
1788 return get_msr_mtrr(vcpu
, msr
, pdata
);
1789 case 0xcd: /* fsb frequency */
1793 * MSR_EBC_FREQUENCY_ID
1794 * Conservative value valid for even the basic CPU models.
1795 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1796 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1797 * and 266MHz for model 3, or 4. Set Core Clock
1798 * Frequency to System Bus Frequency Ratio to 1 (bits
1799 * 31:24) even though these are only valid for CPU
1800 * models > 2, however guests may end up dividing or
1801 * multiplying by zero otherwise.
1803 case MSR_EBC_FREQUENCY_ID
:
1806 case MSR_IA32_APICBASE
:
1807 data
= kvm_get_apic_base(vcpu
);
1809 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1810 return kvm_x2apic_msr_read(vcpu
, msr
, pdata
);
1812 case MSR_IA32_MISC_ENABLE
:
1813 data
= vcpu
->arch
.ia32_misc_enable_msr
;
1815 case MSR_IA32_PERF_STATUS
:
1816 /* TSC increment by tick */
1818 /* CPU multiplier */
1819 data
|= (((uint64_t)4ULL) << 40);
1822 data
= vcpu
->arch
.efer
;
1824 case MSR_KVM_WALL_CLOCK
:
1825 case MSR_KVM_WALL_CLOCK_NEW
:
1826 data
= vcpu
->kvm
->arch
.wall_clock
;
1828 case MSR_KVM_SYSTEM_TIME
:
1829 case MSR_KVM_SYSTEM_TIME_NEW
:
1830 data
= vcpu
->arch
.time
;
1832 case MSR_KVM_ASYNC_PF_EN
:
1833 data
= vcpu
->arch
.apf
.msr_val
;
1835 case MSR_IA32_P5_MC_ADDR
:
1836 case MSR_IA32_P5_MC_TYPE
:
1837 case MSR_IA32_MCG_CAP
:
1838 case MSR_IA32_MCG_CTL
:
1839 case MSR_IA32_MCG_STATUS
:
1840 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1841 return get_msr_mce(vcpu
, msr
, pdata
);
1842 case MSR_K7_CLK_CTL
:
1844 * Provide expected ramp-up count for K7. All other
1845 * are set to zero, indicating minimum divisors for
1848 * This prevents guest kernels on AMD host with CPU
1849 * type 6, model 8 and higher from exploding due to
1850 * the rdmsr failing.
1854 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
1855 if (kvm_hv_msr_partition_wide(msr
)) {
1857 mutex_lock(&vcpu
->kvm
->lock
);
1858 r
= get_msr_hyperv_pw(vcpu
, msr
, pdata
);
1859 mutex_unlock(&vcpu
->kvm
->lock
);
1862 return get_msr_hyperv(vcpu
, msr
, pdata
);
1864 case MSR_IA32_BBL_CR_CTL3
:
1865 /* This legacy MSR exists but isn't fully documented in current
1866 * silicon. It is however accessed by winxp in very narrow
1867 * scenarios where it sets bit #19, itself documented as
1868 * a "reserved" bit. Best effort attempt to source coherent
1869 * read data here should the balance of the register be
1870 * interpreted by the guest:
1872 * L2 cache control register 3: 64GB range, 256KB size,
1873 * enabled, latency 0x1, configured
1879 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
1882 pr_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr
);
1890 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1893 * Read or write a bunch of msrs. All parameters are kernel addresses.
1895 * @return number of msrs set successfully.
1897 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
1898 struct kvm_msr_entry
*entries
,
1899 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1900 unsigned index
, u64
*data
))
1904 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
1905 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1906 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1908 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
1914 * Read or write a bunch of msrs. Parameters are user addresses.
1916 * @return number of msrs set successfully.
1918 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
1919 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1920 unsigned index
, u64
*data
),
1923 struct kvm_msrs msrs
;
1924 struct kvm_msr_entry
*entries
;
1929 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1933 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1937 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1938 entries
= kmalloc(size
, GFP_KERNEL
);
1943 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1946 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
1951 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1962 int kvm_dev_ioctl_check_extension(long ext
)
1967 case KVM_CAP_IRQCHIP
:
1969 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
1970 case KVM_CAP_SET_TSS_ADDR
:
1971 case KVM_CAP_EXT_CPUID
:
1972 case KVM_CAP_CLOCKSOURCE
:
1974 case KVM_CAP_NOP_IO_DELAY
:
1975 case KVM_CAP_MP_STATE
:
1976 case KVM_CAP_SYNC_MMU
:
1977 case KVM_CAP_USER_NMI
:
1978 case KVM_CAP_REINJECT_CONTROL
:
1979 case KVM_CAP_IRQ_INJECT_STATUS
:
1980 case KVM_CAP_ASSIGN_DEV_IRQ
:
1982 case KVM_CAP_IOEVENTFD
:
1984 case KVM_CAP_PIT_STATE2
:
1985 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
1986 case KVM_CAP_XEN_HVM
:
1987 case KVM_CAP_ADJUST_CLOCK
:
1988 case KVM_CAP_VCPU_EVENTS
:
1989 case KVM_CAP_HYPERV
:
1990 case KVM_CAP_HYPERV_VAPIC
:
1991 case KVM_CAP_HYPERV_SPIN
:
1992 case KVM_CAP_PCI_SEGMENT
:
1993 case KVM_CAP_DEBUGREGS
:
1994 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
1996 case KVM_CAP_ASYNC_PF
:
1997 case KVM_CAP_GET_TSC_KHZ
:
2000 case KVM_CAP_COALESCED_MMIO
:
2001 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
2004 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2006 case KVM_CAP_NR_VCPUS
:
2009 case KVM_CAP_NR_MEMSLOTS
:
2010 r
= KVM_MEMORY_SLOTS
;
2012 case KVM_CAP_PV_MMU
: /* obsolete */
2019 r
= KVM_MAX_MCE_BANKS
;
2024 case KVM_CAP_TSC_CONTROL
:
2025 r
= kvm_has_tsc_control
;
2035 long kvm_arch_dev_ioctl(struct file
*filp
,
2036 unsigned int ioctl
, unsigned long arg
)
2038 void __user
*argp
= (void __user
*)arg
;
2042 case KVM_GET_MSR_INDEX_LIST
: {
2043 struct kvm_msr_list __user
*user_msr_list
= argp
;
2044 struct kvm_msr_list msr_list
;
2048 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2051 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2052 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2055 if (n
< msr_list
.nmsrs
)
2058 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2059 num_msrs_to_save
* sizeof(u32
)))
2061 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2063 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2068 case KVM_GET_SUPPORTED_CPUID
: {
2069 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2070 struct kvm_cpuid2 cpuid
;
2073 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2075 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
2076 cpuid_arg
->entries
);
2081 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2086 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2089 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
2091 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
2103 static void wbinvd_ipi(void *garbage
)
2108 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2110 return vcpu
->kvm
->arch
.iommu_domain
&&
2111 !(vcpu
->kvm
->arch
.iommu_flags
& KVM_IOMMU_CACHE_COHERENCY
);
2114 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2116 /* Address WBINVD may be executed by guest */
2117 if (need_emulate_wbinvd(vcpu
)) {
2118 if (kvm_x86_ops
->has_wbinvd_exit())
2119 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2120 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2121 smp_call_function_single(vcpu
->cpu
,
2122 wbinvd_ipi
, NULL
, 1);
2125 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2126 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2127 /* Make sure TSC doesn't go backwards */
2131 kvm_get_msr(vcpu
, MSR_IA32_TSC
, &tsc
);
2132 tsc_delta
= !vcpu
->arch
.last_guest_tsc
? 0 :
2133 tsc
- vcpu
->arch
.last_guest_tsc
;
2136 mark_tsc_unstable("KVM discovered backwards TSC");
2137 if (check_tsc_unstable()) {
2138 kvm_x86_ops
->adjust_tsc_offset(vcpu
, -tsc_delta
);
2139 vcpu
->arch
.tsc_catchup
= 1;
2141 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2142 if (vcpu
->cpu
!= cpu
)
2143 kvm_migrate_timers(vcpu
);
2148 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2150 kvm_x86_ops
->vcpu_put(vcpu
);
2151 kvm_put_guest_fpu(vcpu
);
2152 kvm_get_msr(vcpu
, MSR_IA32_TSC
, &vcpu
->arch
.last_guest_tsc
);
2155 static int is_efer_nx(void)
2157 unsigned long long efer
= 0;
2159 rdmsrl_safe(MSR_EFER
, &efer
);
2160 return efer
& EFER_NX
;
2163 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
2166 struct kvm_cpuid_entry2
*e
, *entry
;
2169 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
2170 e
= &vcpu
->arch
.cpuid_entries
[i
];
2171 if (e
->function
== 0x80000001) {
2176 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
2177 entry
->edx
&= ~(1 << 20);
2178 printk(KERN_INFO
"kvm: guest NX capability removed\n");
2182 /* when an old userspace process fills a new kernel module */
2183 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
2184 struct kvm_cpuid
*cpuid
,
2185 struct kvm_cpuid_entry __user
*entries
)
2188 struct kvm_cpuid_entry
*cpuid_entries
;
2191 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2194 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
2198 if (copy_from_user(cpuid_entries
, entries
,
2199 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
2201 for (i
= 0; i
< cpuid
->nent
; i
++) {
2202 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
2203 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
2204 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
2205 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
2206 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
2207 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
2208 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
2209 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
2210 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
2211 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
2213 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
2214 cpuid_fix_nx_cap(vcpu
);
2216 kvm_apic_set_version(vcpu
);
2217 kvm_x86_ops
->cpuid_update(vcpu
);
2221 vfree(cpuid_entries
);
2226 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
2227 struct kvm_cpuid2
*cpuid
,
2228 struct kvm_cpuid_entry2 __user
*entries
)
2233 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2236 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
2237 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
2239 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
2240 kvm_apic_set_version(vcpu
);
2241 kvm_x86_ops
->cpuid_update(vcpu
);
2249 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
2250 struct kvm_cpuid2
*cpuid
,
2251 struct kvm_cpuid_entry2 __user
*entries
)
2256 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
2259 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
2260 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
2265 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
2269 static void cpuid_mask(u32
*word
, int wordnum
)
2271 *word
&= boot_cpu_data
.x86_capability
[wordnum
];
2274 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
2277 entry
->function
= function
;
2278 entry
->index
= index
;
2279 cpuid_count(entry
->function
, entry
->index
,
2280 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
2284 #define F(x) bit(X86_FEATURE_##x)
2286 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
2287 u32 index
, int *nent
, int maxnent
)
2289 unsigned f_nx
= is_efer_nx() ? F(NX
) : 0;
2290 #ifdef CONFIG_X86_64
2291 unsigned f_gbpages
= (kvm_x86_ops
->get_lpage_level() == PT_PDPE_LEVEL
)
2293 unsigned f_lm
= F(LM
);
2295 unsigned f_gbpages
= 0;
2298 unsigned f_rdtscp
= kvm_x86_ops
->rdtscp_supported() ? F(RDTSCP
) : 0;
2301 const u32 kvm_supported_word0_x86_features
=
2302 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
2303 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
2304 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SEP
) |
2305 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
2306 F(PAT
) | F(PSE36
) | 0 /* PSN */ | F(CLFLSH
) |
2307 0 /* Reserved, DS, ACPI */ | F(MMX
) |
2308 F(FXSR
) | F(XMM
) | F(XMM2
) | F(SELFSNOOP
) |
2309 0 /* HTT, TM, Reserved, PBE */;
2310 /* cpuid 0x80000001.edx */
2311 const u32 kvm_supported_word1_x86_features
=
2312 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
2313 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
2314 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SYSCALL
) |
2315 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
2316 F(PAT
) | F(PSE36
) | 0 /* Reserved */ |
2317 f_nx
| 0 /* Reserved */ | F(MMXEXT
) | F(MMX
) |
2318 F(FXSR
) | F(FXSR_OPT
) | f_gbpages
| f_rdtscp
|
2319 0 /* Reserved */ | f_lm
| F(3DNOWEXT
) | F(3DNOW
);
2321 const u32 kvm_supported_word4_x86_features
=
2322 F(XMM3
) | F(PCLMULQDQ
) | 0 /* DTES64, MONITOR */ |
2323 0 /* DS-CPL, VMX, SMX, EST */ |
2324 0 /* TM2 */ | F(SSSE3
) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2325 0 /* Reserved */ | F(CX16
) | 0 /* xTPR Update, PDCM */ |
2326 0 /* Reserved, DCA */ | F(XMM4_1
) |
2327 F(XMM4_2
) | F(X2APIC
) | F(MOVBE
) | F(POPCNT
) |
2328 0 /* Reserved*/ | F(AES
) | F(XSAVE
) | 0 /* OSXSAVE */ | F(AVX
) |
2330 /* cpuid 0x80000001.ecx */
2331 const u32 kvm_supported_word6_x86_features
=
2332 F(LAHF_LM
) | F(CMP_LEGACY
) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2333 F(CR8_LEGACY
) | F(ABM
) | F(SSE4A
) | F(MISALIGNSSE
) |
2334 F(3DNOWPREFETCH
) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP
) |
2335 0 /* SKINIT, WDT, LWP */ | F(FMA4
) | F(TBM
);
2337 /* all calls to cpuid_count() should be made on the same cpu */
2339 do_cpuid_1_ent(entry
, function
, index
);
2344 entry
->eax
= min(entry
->eax
, (u32
)0xd);
2347 entry
->edx
&= kvm_supported_word0_x86_features
;
2348 cpuid_mask(&entry
->edx
, 0);
2349 entry
->ecx
&= kvm_supported_word4_x86_features
;
2350 cpuid_mask(&entry
->ecx
, 4);
2351 /* we support x2apic emulation even if host does not support
2352 * it since we emulate x2apic in software */
2353 entry
->ecx
|= F(X2APIC
);
2355 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2356 * may return different values. This forces us to get_cpu() before
2357 * issuing the first command, and also to emulate this annoying behavior
2358 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2360 int t
, times
= entry
->eax
& 0xff;
2362 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
2363 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
2364 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
2365 do_cpuid_1_ent(&entry
[t
], function
, 0);
2366 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
2371 /* function 4 and 0xb have additional index. */
2375 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
2376 /* read more entries until cache_type is zero */
2377 for (i
= 1; *nent
< maxnent
; ++i
) {
2378 cache_type
= entry
[i
- 1].eax
& 0x1f;
2381 do_cpuid_1_ent(&entry
[i
], function
, i
);
2383 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
2391 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
2392 /* read more entries until level_type is zero */
2393 for (i
= 1; *nent
< maxnent
; ++i
) {
2394 level_type
= entry
[i
- 1].ecx
& 0xff00;
2397 do_cpuid_1_ent(&entry
[i
], function
, i
);
2399 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
2407 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
2408 for (i
= 1; *nent
< maxnent
&& i
< 64; ++i
) {
2409 if (entry
[i
].eax
== 0)
2411 do_cpuid_1_ent(&entry
[i
], function
, i
);
2413 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
2418 case KVM_CPUID_SIGNATURE
: {
2419 char signature
[12] = "KVMKVMKVM\0\0";
2420 u32
*sigptr
= (u32
*)signature
;
2422 entry
->ebx
= sigptr
[0];
2423 entry
->ecx
= sigptr
[1];
2424 entry
->edx
= sigptr
[2];
2427 case KVM_CPUID_FEATURES
:
2428 entry
->eax
= (1 << KVM_FEATURE_CLOCKSOURCE
) |
2429 (1 << KVM_FEATURE_NOP_IO_DELAY
) |
2430 (1 << KVM_FEATURE_CLOCKSOURCE2
) |
2431 (1 << KVM_FEATURE_ASYNC_PF
) |
2432 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT
);
2438 entry
->eax
= min(entry
->eax
, 0x8000001a);
2441 entry
->edx
&= kvm_supported_word1_x86_features
;
2442 cpuid_mask(&entry
->edx
, 1);
2443 entry
->ecx
&= kvm_supported_word6_x86_features
;
2444 cpuid_mask(&entry
->ecx
, 6);
2448 kvm_x86_ops
->set_supported_cpuid(function
, entry
);
2455 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
2456 struct kvm_cpuid_entry2 __user
*entries
)
2458 struct kvm_cpuid_entry2
*cpuid_entries
;
2459 int limit
, nent
= 0, r
= -E2BIG
;
2462 if (cpuid
->nent
< 1)
2464 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2465 cpuid
->nent
= KVM_MAX_CPUID_ENTRIES
;
2467 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
2471 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
2472 limit
= cpuid_entries
[0].eax
;
2473 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
2474 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
2475 &nent
, cpuid
->nent
);
2477 if (nent
>= cpuid
->nent
)
2480 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
2481 limit
= cpuid_entries
[nent
- 1].eax
;
2482 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
2483 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
2484 &nent
, cpuid
->nent
);
2489 if (nent
>= cpuid
->nent
)
2492 do_cpuid_ent(&cpuid_entries
[nent
], KVM_CPUID_SIGNATURE
, 0, &nent
,
2496 if (nent
>= cpuid
->nent
)
2499 do_cpuid_ent(&cpuid_entries
[nent
], KVM_CPUID_FEATURES
, 0, &nent
,
2503 if (nent
>= cpuid
->nent
)
2507 if (copy_to_user(entries
, cpuid_entries
,
2508 nent
* sizeof(struct kvm_cpuid_entry2
)))
2514 vfree(cpuid_entries
);
2519 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
2520 struct kvm_lapic_state
*s
)
2522 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
2527 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
2528 struct kvm_lapic_state
*s
)
2530 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
2531 kvm_apic_post_state_restore(vcpu
);
2532 update_cr8_intercept(vcpu
);
2537 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2538 struct kvm_interrupt
*irq
)
2540 if (irq
->irq
< 0 || irq
->irq
>= 256)
2542 if (irqchip_in_kernel(vcpu
->kvm
))
2545 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
2546 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2551 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
2553 kvm_inject_nmi(vcpu
);
2558 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
2559 struct kvm_tpr_access_ctl
*tac
)
2563 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
2567 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
2571 unsigned bank_num
= mcg_cap
& 0xff, bank
;
2574 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
2576 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
2579 vcpu
->arch
.mcg_cap
= mcg_cap
;
2580 /* Init IA32_MCG_CTL to all 1s */
2581 if (mcg_cap
& MCG_CTL_P
)
2582 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
2583 /* Init IA32_MCi_CTL to all 1s */
2584 for (bank
= 0; bank
< bank_num
; bank
++)
2585 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
2590 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
2591 struct kvm_x86_mce
*mce
)
2593 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2594 unsigned bank_num
= mcg_cap
& 0xff;
2595 u64
*banks
= vcpu
->arch
.mce_banks
;
2597 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
2600 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2601 * reporting is disabled
2603 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
2604 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
2606 banks
+= 4 * mce
->bank
;
2608 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2609 * reporting is disabled for the bank
2611 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
2613 if (mce
->status
& MCI_STATUS_UC
) {
2614 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
2615 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
2616 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2619 if (banks
[1] & MCI_STATUS_VAL
)
2620 mce
->status
|= MCI_STATUS_OVER
;
2621 banks
[2] = mce
->addr
;
2622 banks
[3] = mce
->misc
;
2623 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
2624 banks
[1] = mce
->status
;
2625 kvm_queue_exception(vcpu
, MC_VECTOR
);
2626 } else if (!(banks
[1] & MCI_STATUS_VAL
)
2627 || !(banks
[1] & MCI_STATUS_UC
)) {
2628 if (banks
[1] & MCI_STATUS_VAL
)
2629 mce
->status
|= MCI_STATUS_OVER
;
2630 banks
[2] = mce
->addr
;
2631 banks
[3] = mce
->misc
;
2632 banks
[1] = mce
->status
;
2634 banks
[1] |= MCI_STATUS_OVER
;
2638 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
2639 struct kvm_vcpu_events
*events
)
2641 events
->exception
.injected
=
2642 vcpu
->arch
.exception
.pending
&&
2643 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
2644 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
2645 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
2646 events
->exception
.pad
= 0;
2647 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
2649 events
->interrupt
.injected
=
2650 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
2651 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
2652 events
->interrupt
.soft
= 0;
2653 events
->interrupt
.shadow
=
2654 kvm_x86_ops
->get_interrupt_shadow(vcpu
,
2655 KVM_X86_SHADOW_INT_MOV_SS
| KVM_X86_SHADOW_INT_STI
);
2657 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
2658 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
;
2659 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
2660 events
->nmi
.pad
= 0;
2662 events
->sipi_vector
= vcpu
->arch
.sipi_vector
;
2664 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
2665 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2666 | KVM_VCPUEVENT_VALID_SHADOW
);
2667 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
2670 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
2671 struct kvm_vcpu_events
*events
)
2673 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2674 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2675 | KVM_VCPUEVENT_VALID_SHADOW
))
2678 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
2679 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
2680 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
2681 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
2683 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
2684 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
2685 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
2686 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
2687 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
2688 events
->interrupt
.shadow
);
2690 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
2691 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
2692 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
2693 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
2695 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
)
2696 vcpu
->arch
.sipi_vector
= events
->sipi_vector
;
2698 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2703 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
2704 struct kvm_debugregs
*dbgregs
)
2706 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
2707 dbgregs
->dr6
= vcpu
->arch
.dr6
;
2708 dbgregs
->dr7
= vcpu
->arch
.dr7
;
2710 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
2713 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
2714 struct kvm_debugregs
*dbgregs
)
2719 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
2720 vcpu
->arch
.dr6
= dbgregs
->dr6
;
2721 vcpu
->arch
.dr7
= dbgregs
->dr7
;
2726 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
2727 struct kvm_xsave
*guest_xsave
)
2730 memcpy(guest_xsave
->region
,
2731 &vcpu
->arch
.guest_fpu
.state
->xsave
,
2734 memcpy(guest_xsave
->region
,
2735 &vcpu
->arch
.guest_fpu
.state
->fxsave
,
2736 sizeof(struct i387_fxsave_struct
));
2737 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
2742 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
2743 struct kvm_xsave
*guest_xsave
)
2746 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
2749 memcpy(&vcpu
->arch
.guest_fpu
.state
->xsave
,
2750 guest_xsave
->region
, xstate_size
);
2752 if (xstate_bv
& ~XSTATE_FPSSE
)
2754 memcpy(&vcpu
->arch
.guest_fpu
.state
->fxsave
,
2755 guest_xsave
->region
, sizeof(struct i387_fxsave_struct
));
2760 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
2761 struct kvm_xcrs
*guest_xcrs
)
2763 if (!cpu_has_xsave
) {
2764 guest_xcrs
->nr_xcrs
= 0;
2768 guest_xcrs
->nr_xcrs
= 1;
2769 guest_xcrs
->flags
= 0;
2770 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
2771 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
2774 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
2775 struct kvm_xcrs
*guest_xcrs
)
2782 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
2785 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
2786 /* Only support XCR0 currently */
2787 if (guest_xcrs
->xcrs
[0].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
2788 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
2789 guest_xcrs
->xcrs
[0].value
);
2797 long kvm_arch_vcpu_ioctl(struct file
*filp
,
2798 unsigned int ioctl
, unsigned long arg
)
2800 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2801 void __user
*argp
= (void __user
*)arg
;
2804 struct kvm_lapic_state
*lapic
;
2805 struct kvm_xsave
*xsave
;
2806 struct kvm_xcrs
*xcrs
;
2812 case KVM_GET_LAPIC
: {
2814 if (!vcpu
->arch
.apic
)
2816 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
2821 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
2825 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
2830 case KVM_SET_LAPIC
: {
2832 if (!vcpu
->arch
.apic
)
2834 u
.lapic
= kmalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
2839 if (copy_from_user(u
.lapic
, argp
, sizeof(struct kvm_lapic_state
)))
2841 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
2847 case KVM_INTERRUPT
: {
2848 struct kvm_interrupt irq
;
2851 if (copy_from_user(&irq
, argp
, sizeof irq
))
2853 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2860 r
= kvm_vcpu_ioctl_nmi(vcpu
);
2866 case KVM_SET_CPUID
: {
2867 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2868 struct kvm_cpuid cpuid
;
2871 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2873 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2878 case KVM_SET_CPUID2
: {
2879 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2880 struct kvm_cpuid2 cpuid
;
2883 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2885 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
2886 cpuid_arg
->entries
);
2891 case KVM_GET_CPUID2
: {
2892 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2893 struct kvm_cpuid2 cpuid
;
2896 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2898 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
2899 cpuid_arg
->entries
);
2903 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2909 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
2912 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2914 case KVM_TPR_ACCESS_REPORTING
: {
2915 struct kvm_tpr_access_ctl tac
;
2918 if (copy_from_user(&tac
, argp
, sizeof tac
))
2920 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
2924 if (copy_to_user(argp
, &tac
, sizeof tac
))
2929 case KVM_SET_VAPIC_ADDR
: {
2930 struct kvm_vapic_addr va
;
2933 if (!irqchip_in_kernel(vcpu
->kvm
))
2936 if (copy_from_user(&va
, argp
, sizeof va
))
2939 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
2942 case KVM_X86_SETUP_MCE
: {
2946 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
2948 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
2951 case KVM_X86_SET_MCE
: {
2952 struct kvm_x86_mce mce
;
2955 if (copy_from_user(&mce
, argp
, sizeof mce
))
2957 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
2960 case KVM_GET_VCPU_EVENTS
: {
2961 struct kvm_vcpu_events events
;
2963 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
2966 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
2971 case KVM_SET_VCPU_EVENTS
: {
2972 struct kvm_vcpu_events events
;
2975 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
2978 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
2981 case KVM_GET_DEBUGREGS
: {
2982 struct kvm_debugregs dbgregs
;
2984 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
2987 if (copy_to_user(argp
, &dbgregs
,
2988 sizeof(struct kvm_debugregs
)))
2993 case KVM_SET_DEBUGREGS
: {
2994 struct kvm_debugregs dbgregs
;
2997 if (copy_from_user(&dbgregs
, argp
,
2998 sizeof(struct kvm_debugregs
)))
3001 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3004 case KVM_GET_XSAVE
: {
3005 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3010 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3013 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3018 case KVM_SET_XSAVE
: {
3019 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3025 if (copy_from_user(u
.xsave
, argp
, sizeof(struct kvm_xsave
)))
3028 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3031 case KVM_GET_XCRS
: {
3032 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3037 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3040 if (copy_to_user(argp
, u
.xcrs
,
3041 sizeof(struct kvm_xcrs
)))
3046 case KVM_SET_XCRS
: {
3047 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3053 if (copy_from_user(u
.xcrs
, argp
,
3054 sizeof(struct kvm_xcrs
)))
3057 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3060 case KVM_SET_TSC_KHZ
: {
3064 if (!kvm_has_tsc_control
)
3067 user_tsc_khz
= (u32
)arg
;
3069 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3072 kvm_x86_ops
->set_tsc_khz(vcpu
, user_tsc_khz
);
3077 case KVM_GET_TSC_KHZ
: {
3079 if (check_tsc_unstable())
3082 r
= vcpu_tsc_khz(vcpu
);
3094 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3098 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3100 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3104 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3107 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3111 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3112 u32 kvm_nr_mmu_pages
)
3114 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3117 mutex_lock(&kvm
->slots_lock
);
3118 spin_lock(&kvm
->mmu_lock
);
3120 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3121 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3123 spin_unlock(&kvm
->mmu_lock
);
3124 mutex_unlock(&kvm
->slots_lock
);
3128 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3130 return kvm
->arch
.n_max_mmu_pages
;
3133 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3138 switch (chip
->chip_id
) {
3139 case KVM_IRQCHIP_PIC_MASTER
:
3140 memcpy(&chip
->chip
.pic
,
3141 &pic_irqchip(kvm
)->pics
[0],
3142 sizeof(struct kvm_pic_state
));
3144 case KVM_IRQCHIP_PIC_SLAVE
:
3145 memcpy(&chip
->chip
.pic
,
3146 &pic_irqchip(kvm
)->pics
[1],
3147 sizeof(struct kvm_pic_state
));
3149 case KVM_IRQCHIP_IOAPIC
:
3150 r
= kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3159 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3164 switch (chip
->chip_id
) {
3165 case KVM_IRQCHIP_PIC_MASTER
:
3166 spin_lock(&pic_irqchip(kvm
)->lock
);
3167 memcpy(&pic_irqchip(kvm
)->pics
[0],
3169 sizeof(struct kvm_pic_state
));
3170 spin_unlock(&pic_irqchip(kvm
)->lock
);
3172 case KVM_IRQCHIP_PIC_SLAVE
:
3173 spin_lock(&pic_irqchip(kvm
)->lock
);
3174 memcpy(&pic_irqchip(kvm
)->pics
[1],
3176 sizeof(struct kvm_pic_state
));
3177 spin_unlock(&pic_irqchip(kvm
)->lock
);
3179 case KVM_IRQCHIP_IOAPIC
:
3180 r
= kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3186 kvm_pic_update_irq(pic_irqchip(kvm
));
3190 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3194 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3195 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
3196 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3200 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3204 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3205 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
3206 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
, 0);
3207 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3211 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3215 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3216 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3217 sizeof(ps
->channels
));
3218 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3219 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3220 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3224 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3226 int r
= 0, start
= 0;
3227 u32 prev_legacy
, cur_legacy
;
3228 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3229 prev_legacy
= kvm
->arch
.vpit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3230 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3231 if (!prev_legacy
&& cur_legacy
)
3233 memcpy(&kvm
->arch
.vpit
->pit_state
.channels
, &ps
->channels
,
3234 sizeof(kvm
->arch
.vpit
->pit_state
.channels
));
3235 kvm
->arch
.vpit
->pit_state
.flags
= ps
->flags
;
3236 kvm_pit_load_count(kvm
, 0, kvm
->arch
.vpit
->pit_state
.channels
[0].count
, start
);
3237 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3241 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3242 struct kvm_reinject_control
*control
)
3244 if (!kvm
->arch
.vpit
)
3246 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3247 kvm
->arch
.vpit
->pit_state
.pit_timer
.reinject
= control
->pit_reinject
;
3248 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3253 * Get (and clear) the dirty memory log for a memory slot.
3255 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
3256 struct kvm_dirty_log
*log
)
3259 struct kvm_memory_slot
*memslot
;
3261 unsigned long is_dirty
= 0;
3263 mutex_lock(&kvm
->slots_lock
);
3266 if (log
->slot
>= KVM_MEMORY_SLOTS
)
3269 memslot
= &kvm
->memslots
->memslots
[log
->slot
];
3271 if (!memslot
->dirty_bitmap
)
3274 n
= kvm_dirty_bitmap_bytes(memslot
);
3276 for (i
= 0; !is_dirty
&& i
< n
/sizeof(long); i
++)
3277 is_dirty
= memslot
->dirty_bitmap
[i
];
3279 /* If nothing is dirty, don't bother messing with page tables. */
3281 struct kvm_memslots
*slots
, *old_slots
;
3282 unsigned long *dirty_bitmap
;
3284 dirty_bitmap
= memslot
->dirty_bitmap_head
;
3285 if (memslot
->dirty_bitmap
== dirty_bitmap
)
3286 dirty_bitmap
+= n
/ sizeof(long);
3287 memset(dirty_bitmap
, 0, n
);
3290 slots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
3293 memcpy(slots
, kvm
->memslots
, sizeof(struct kvm_memslots
));
3294 slots
->memslots
[log
->slot
].dirty_bitmap
= dirty_bitmap
;
3295 slots
->generation
++;
3297 old_slots
= kvm
->memslots
;
3298 rcu_assign_pointer(kvm
->memslots
, slots
);
3299 synchronize_srcu_expedited(&kvm
->srcu
);
3300 dirty_bitmap
= old_slots
->memslots
[log
->slot
].dirty_bitmap
;
3303 spin_lock(&kvm
->mmu_lock
);
3304 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
3305 spin_unlock(&kvm
->mmu_lock
);
3308 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap
, n
))
3312 if (clear_user(log
->dirty_bitmap
, n
))
3318 mutex_unlock(&kvm
->slots_lock
);
3322 long kvm_arch_vm_ioctl(struct file
*filp
,
3323 unsigned int ioctl
, unsigned long arg
)
3325 struct kvm
*kvm
= filp
->private_data
;
3326 void __user
*argp
= (void __user
*)arg
;
3329 * This union makes it completely explicit to gcc-3.x
3330 * that these two variables' stack usage should be
3331 * combined, not added together.
3334 struct kvm_pit_state ps
;
3335 struct kvm_pit_state2 ps2
;
3336 struct kvm_pit_config pit_config
;
3340 case KVM_SET_TSS_ADDR
:
3341 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
3345 case KVM_SET_IDENTITY_MAP_ADDR
: {
3349 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
3351 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
3356 case KVM_SET_NR_MMU_PAGES
:
3357 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
3361 case KVM_GET_NR_MMU_PAGES
:
3362 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
3364 case KVM_CREATE_IRQCHIP
: {
3365 struct kvm_pic
*vpic
;
3367 mutex_lock(&kvm
->lock
);
3370 goto create_irqchip_unlock
;
3372 vpic
= kvm_create_pic(kvm
);
3374 r
= kvm_ioapic_init(kvm
);
3376 mutex_lock(&kvm
->slots_lock
);
3377 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3379 mutex_unlock(&kvm
->slots_lock
);
3381 goto create_irqchip_unlock
;
3384 goto create_irqchip_unlock
;
3386 kvm
->arch
.vpic
= vpic
;
3388 r
= kvm_setup_default_irq_routing(kvm
);
3390 mutex_lock(&kvm
->slots_lock
);
3391 mutex_lock(&kvm
->irq_lock
);
3392 kvm_ioapic_destroy(kvm
);
3393 kvm_destroy_pic(kvm
);
3394 mutex_unlock(&kvm
->irq_lock
);
3395 mutex_unlock(&kvm
->slots_lock
);
3397 create_irqchip_unlock
:
3398 mutex_unlock(&kvm
->lock
);
3401 case KVM_CREATE_PIT
:
3402 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
3404 case KVM_CREATE_PIT2
:
3406 if (copy_from_user(&u
.pit_config
, argp
,
3407 sizeof(struct kvm_pit_config
)))
3410 mutex_lock(&kvm
->slots_lock
);
3413 goto create_pit_unlock
;
3415 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
3419 mutex_unlock(&kvm
->slots_lock
);
3421 case KVM_IRQ_LINE_STATUS
:
3422 case KVM_IRQ_LINE
: {
3423 struct kvm_irq_level irq_event
;
3426 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
3429 if (irqchip_in_kernel(kvm
)) {
3431 status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
3432 irq_event
.irq
, irq_event
.level
);
3433 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3435 irq_event
.status
= status
;
3436 if (copy_to_user(argp
, &irq_event
,
3444 case KVM_GET_IRQCHIP
: {
3445 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3446 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
3452 if (copy_from_user(chip
, argp
, sizeof *chip
))
3453 goto get_irqchip_out
;
3455 if (!irqchip_in_kernel(kvm
))
3456 goto get_irqchip_out
;
3457 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
3459 goto get_irqchip_out
;
3461 if (copy_to_user(argp
, chip
, sizeof *chip
))
3462 goto get_irqchip_out
;
3470 case KVM_SET_IRQCHIP
: {
3471 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3472 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
3478 if (copy_from_user(chip
, argp
, sizeof *chip
))
3479 goto set_irqchip_out
;
3481 if (!irqchip_in_kernel(kvm
))
3482 goto set_irqchip_out
;
3483 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
3485 goto set_irqchip_out
;
3495 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
3498 if (!kvm
->arch
.vpit
)
3500 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
3504 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
3511 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
3514 if (!kvm
->arch
.vpit
)
3516 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
3522 case KVM_GET_PIT2
: {
3524 if (!kvm
->arch
.vpit
)
3526 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
3530 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
3535 case KVM_SET_PIT2
: {
3537 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
3540 if (!kvm
->arch
.vpit
)
3542 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
3548 case KVM_REINJECT_CONTROL
: {
3549 struct kvm_reinject_control control
;
3551 if (copy_from_user(&control
, argp
, sizeof(control
)))
3553 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
3559 case KVM_XEN_HVM_CONFIG
: {
3561 if (copy_from_user(&kvm
->arch
.xen_hvm_config
, argp
,
3562 sizeof(struct kvm_xen_hvm_config
)))
3565 if (kvm
->arch
.xen_hvm_config
.flags
)
3570 case KVM_SET_CLOCK
: {
3571 struct kvm_clock_data user_ns
;
3576 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
3584 local_irq_disable();
3585 now_ns
= get_kernel_ns();
3586 delta
= user_ns
.clock
- now_ns
;
3588 kvm
->arch
.kvmclock_offset
= delta
;
3591 case KVM_GET_CLOCK
: {
3592 struct kvm_clock_data user_ns
;
3595 local_irq_disable();
3596 now_ns
= get_kernel_ns();
3597 user_ns
.clock
= kvm
->arch
.kvmclock_offset
+ now_ns
;
3600 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
3603 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
3616 static void kvm_init_msr_list(void)
3621 /* skip the first msrs in the list. KVM-specific */
3622 for (i
= j
= KVM_SAVE_MSRS_BEGIN
; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
3623 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
3626 msrs_to_save
[j
] = msrs_to_save
[i
];
3629 num_msrs_to_save
= j
;
3632 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
3640 if (!(vcpu
->arch
.apic
&&
3641 !kvm_iodevice_write(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3642 && kvm_io_bus_write(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3653 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
3660 if (!(vcpu
->arch
.apic
&&
3661 !kvm_iodevice_read(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3662 && kvm_io_bus_read(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3664 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, *(u64
*)v
);
3674 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3675 struct kvm_segment
*var
, int seg
)
3677 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3680 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3681 struct kvm_segment
*var
, int seg
)
3683 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3686 static gpa_t
translate_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
)
3691 static gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
)
3694 struct x86_exception exception
;
3696 BUG_ON(!mmu_is_nested(vcpu
));
3698 /* NPT walks are always user-walks */
3699 access
|= PFERR_USER_MASK
;
3700 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, &exception
);
3705 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
3706 struct x86_exception
*exception
)
3708 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3709 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3712 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
3713 struct x86_exception
*exception
)
3715 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3716 access
|= PFERR_FETCH_MASK
;
3717 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3720 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
3721 struct x86_exception
*exception
)
3723 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3724 access
|= PFERR_WRITE_MASK
;
3725 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3728 /* uses this to access any guest's mapped memory without checking CPL */
3729 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
3730 struct x86_exception
*exception
)
3732 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
3735 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
3736 struct kvm_vcpu
*vcpu
, u32 access
,
3737 struct x86_exception
*exception
)
3740 int r
= X86EMUL_CONTINUE
;
3743 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
3745 unsigned offset
= addr
& (PAGE_SIZE
-1);
3746 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3749 if (gpa
== UNMAPPED_GVA
)
3750 return X86EMUL_PROPAGATE_FAULT
;
3751 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
3753 r
= X86EMUL_IO_NEEDED
;
3765 /* used for instruction fetching */
3766 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3767 gva_t addr
, void *val
, unsigned int bytes
,
3768 struct x86_exception
*exception
)
3770 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3771 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3773 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
,
3774 access
| PFERR_FETCH_MASK
,
3778 static int kvm_read_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3779 gva_t addr
, void *val
, unsigned int bytes
,
3780 struct x86_exception
*exception
)
3782 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3783 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3785 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
3789 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3790 gva_t addr
, void *val
, unsigned int bytes
,
3791 struct x86_exception
*exception
)
3793 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3794 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, 0, exception
);
3797 static int kvm_write_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3798 gva_t addr
, void *val
,
3800 struct x86_exception
*exception
)
3802 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3804 int r
= X86EMUL_CONTINUE
;
3807 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
3810 unsigned offset
= addr
& (PAGE_SIZE
-1);
3811 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3814 if (gpa
== UNMAPPED_GVA
)
3815 return X86EMUL_PROPAGATE_FAULT
;
3816 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
3818 r
= X86EMUL_IO_NEEDED
;
3830 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
3834 struct x86_exception
*exception
)
3836 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3840 if (vcpu
->mmio_read_completed
) {
3841 memcpy(val
, vcpu
->mmio_data
, bytes
);
3842 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
3843 vcpu
->mmio_phys_addr
, *(u64
*)val
);
3844 vcpu
->mmio_read_completed
= 0;
3845 return X86EMUL_CONTINUE
;
3848 gpa
= kvm_mmu_gva_to_gpa_read(vcpu
, addr
, exception
);
3850 if (gpa
== UNMAPPED_GVA
)
3851 return X86EMUL_PROPAGATE_FAULT
;
3853 /* For APIC access vmexit */
3854 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
3857 if (kvm_read_guest_virt(ctxt
, addr
, val
, bytes
, exception
)
3858 == X86EMUL_CONTINUE
)
3859 return X86EMUL_CONTINUE
;
3863 * Is this MMIO handled locally?
3865 handled
= vcpu_mmio_read(vcpu
, gpa
, bytes
, val
);
3867 if (handled
== bytes
)
3868 return X86EMUL_CONTINUE
;
3874 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
3876 vcpu
->mmio_needed
= 1;
3877 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
3878 vcpu
->run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
= gpa
;
3879 vcpu
->mmio_size
= bytes
;
3880 vcpu
->run
->mmio
.len
= min(vcpu
->mmio_size
, 8);
3881 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= 0;
3882 vcpu
->mmio_index
= 0;
3884 return X86EMUL_IO_NEEDED
;
3887 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3888 const void *val
, int bytes
)
3892 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
3895 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
, 1);
3899 static int emulator_write_emulated_onepage(unsigned long addr
,
3902 struct x86_exception
*exception
,
3903 struct kvm_vcpu
*vcpu
)
3908 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, exception
);
3910 if (gpa
== UNMAPPED_GVA
)
3911 return X86EMUL_PROPAGATE_FAULT
;
3913 /* For APIC access vmexit */
3914 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
3917 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
3918 return X86EMUL_CONTINUE
;
3921 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
3923 * Is this MMIO handled locally?
3925 handled
= vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
3926 if (handled
== bytes
)
3927 return X86EMUL_CONTINUE
;
3933 vcpu
->mmio_needed
= 1;
3934 memcpy(vcpu
->mmio_data
, val
, bytes
);
3935 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
3936 vcpu
->run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
= gpa
;
3937 vcpu
->mmio_size
= bytes
;
3938 vcpu
->run
->mmio
.len
= min(vcpu
->mmio_size
, 8);
3939 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= 1;
3940 memcpy(vcpu
->run
->mmio
.data
, vcpu
->mmio_data
, 8);
3941 vcpu
->mmio_index
= 0;
3943 return X86EMUL_CONTINUE
;
3946 int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
3950 struct x86_exception
*exception
)
3952 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3954 /* Crossing a page boundary? */
3955 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
3958 now
= -addr
& ~PAGE_MASK
;
3959 rc
= emulator_write_emulated_onepage(addr
, val
, now
, exception
,
3961 if (rc
!= X86EMUL_CONTINUE
)
3967 return emulator_write_emulated_onepage(addr
, val
, bytes
, exception
,
3971 #define CMPXCHG_TYPE(t, ptr, old, new) \
3972 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3974 #ifdef CONFIG_X86_64
3975 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3977 # define CMPXCHG64(ptr, old, new) \
3978 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3981 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
3986 struct x86_exception
*exception
)
3988 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3994 /* guests cmpxchg8b have to be emulated atomically */
3995 if (bytes
> 8 || (bytes
& (bytes
- 1)))
3998 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
4000 if (gpa
== UNMAPPED_GVA
||
4001 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4004 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
4007 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
4008 if (is_error_page(page
)) {
4009 kvm_release_page_clean(page
);
4013 kaddr
= kmap_atomic(page
, KM_USER0
);
4014 kaddr
+= offset_in_page(gpa
);
4017 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
4020 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
4023 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
4026 exchanged
= CMPXCHG64(kaddr
, old
, new);
4031 kunmap_atomic(kaddr
, KM_USER0
);
4032 kvm_release_page_dirty(page
);
4035 return X86EMUL_CMPXCHG_FAILED
;
4037 kvm_mmu_pte_write(vcpu
, gpa
, new, bytes
, 1);
4039 return X86EMUL_CONTINUE
;
4042 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
4044 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
4047 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
4049 /* TODO: String I/O for in kernel device */
4052 if (vcpu
->arch
.pio
.in
)
4053 r
= kvm_io_bus_read(vcpu
->kvm
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
4054 vcpu
->arch
.pio
.size
, pd
);
4056 r
= kvm_io_bus_write(vcpu
->kvm
, KVM_PIO_BUS
,
4057 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
4063 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
4064 int size
, unsigned short port
, void *val
,
4067 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4069 if (vcpu
->arch
.pio
.count
)
4072 trace_kvm_pio(0, port
, size
, count
);
4074 vcpu
->arch
.pio
.port
= port
;
4075 vcpu
->arch
.pio
.in
= 1;
4076 vcpu
->arch
.pio
.count
= count
;
4077 vcpu
->arch
.pio
.size
= size
;
4079 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
4081 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
4082 vcpu
->arch
.pio
.count
= 0;
4086 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
4087 vcpu
->run
->io
.direction
= KVM_EXIT_IO_IN
;
4088 vcpu
->run
->io
.size
= size
;
4089 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
4090 vcpu
->run
->io
.count
= count
;
4091 vcpu
->run
->io
.port
= port
;
4096 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
4097 int size
, unsigned short port
,
4098 const void *val
, unsigned int count
)
4100 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4102 trace_kvm_pio(1, port
, size
, count
);
4104 vcpu
->arch
.pio
.port
= port
;
4105 vcpu
->arch
.pio
.in
= 0;
4106 vcpu
->arch
.pio
.count
= count
;
4107 vcpu
->arch
.pio
.size
= size
;
4109 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
4111 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
4112 vcpu
->arch
.pio
.count
= 0;
4116 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
4117 vcpu
->run
->io
.direction
= KVM_EXIT_IO_OUT
;
4118 vcpu
->run
->io
.size
= size
;
4119 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
4120 vcpu
->run
->io
.count
= count
;
4121 vcpu
->run
->io
.port
= port
;
4126 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
4128 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
4131 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
4133 kvm_mmu_invlpg(vcpu
, address
);
4134 return X86EMUL_CONTINUE
;
4137 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
4139 if (!need_emulate_wbinvd(vcpu
))
4140 return X86EMUL_CONTINUE
;
4142 if (kvm_x86_ops
->has_wbinvd_exit()) {
4143 int cpu
= get_cpu();
4145 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
4146 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
4147 wbinvd_ipi
, NULL
, 1);
4149 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
4152 return X86EMUL_CONTINUE
;
4154 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
4156 int emulate_clts(struct kvm_vcpu
*vcpu
)
4158 kvm_x86_ops
->set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~X86_CR0_TS
));
4159 kvm_x86_ops
->fpu_activate(vcpu
);
4160 return X86EMUL_CONTINUE
;
4163 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
4165 return _kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
4168 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
4171 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
4174 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
4176 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
4179 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
4181 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4182 unsigned long value
;
4186 value
= kvm_read_cr0(vcpu
);
4189 value
= vcpu
->arch
.cr2
;
4192 value
= kvm_read_cr3(vcpu
);
4195 value
= kvm_read_cr4(vcpu
);
4198 value
= kvm_get_cr8(vcpu
);
4201 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
4208 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
4210 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4215 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
4218 vcpu
->arch
.cr2
= val
;
4221 res
= kvm_set_cr3(vcpu
, val
);
4224 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
4227 res
= kvm_set_cr8(vcpu
, val
);
4230 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
4237 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
4239 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
4242 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4244 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
4247 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4249 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
4252 static unsigned long emulator_get_cached_segment_base(
4253 struct x86_emulate_ctxt
*ctxt
, int seg
)
4255 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
4258 static bool emulator_get_cached_descriptor(struct x86_emulate_ctxt
*ctxt
,
4259 struct desc_struct
*desc
, u32
*base3
,
4262 struct kvm_segment var
;
4264 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
4271 set_desc_limit(desc
, var
.limit
);
4272 set_desc_base(desc
, (unsigned long)var
.base
);
4273 #ifdef CONFIG_X86_64
4275 *base3
= var
.base
>> 32;
4277 desc
->type
= var
.type
;
4279 desc
->dpl
= var
.dpl
;
4280 desc
->p
= var
.present
;
4281 desc
->avl
= var
.avl
;
4289 static void emulator_set_cached_descriptor(struct x86_emulate_ctxt
*ctxt
,
4290 struct desc_struct
*desc
, u32 base3
,
4293 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4294 struct kvm_segment var
;
4296 /* needed to preserve selector */
4297 kvm_get_segment(vcpu
, &var
, seg
);
4299 var
.base
= get_desc_base(desc
);
4300 #ifdef CONFIG_X86_64
4301 var
.base
|= ((u64
)base3
) << 32;
4303 var
.limit
= get_desc_limit(desc
);
4305 var
.limit
= (var
.limit
<< 12) | 0xfff;
4306 var
.type
= desc
->type
;
4307 var
.present
= desc
->p
;
4308 var
.dpl
= desc
->dpl
;
4313 var
.avl
= desc
->avl
;
4314 var
.present
= desc
->p
;
4315 var
.unusable
= !var
.present
;
4318 kvm_set_segment(vcpu
, &var
, seg
);
4322 static u16
emulator_get_segment_selector(struct x86_emulate_ctxt
*ctxt
, int seg
)
4324 struct kvm_segment kvm_seg
;
4326 kvm_get_segment(emul_to_vcpu(ctxt
), &kvm_seg
, seg
);
4327 return kvm_seg
.selector
;
4330 static void emulator_set_segment_selector(struct x86_emulate_ctxt
*ctxt
,
4333 struct kvm_segment kvm_seg
;
4335 kvm_get_segment(emul_to_vcpu(ctxt
), &kvm_seg
, seg
);
4336 kvm_seg
.selector
= sel
;
4337 kvm_set_segment(emul_to_vcpu(ctxt
), &kvm_seg
, seg
);
4340 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
4341 u32 msr_index
, u64
*pdata
)
4343 return kvm_get_msr(emul_to_vcpu(ctxt
), msr_index
, pdata
);
4346 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
4347 u32 msr_index
, u64 data
)
4349 return kvm_set_msr(emul_to_vcpu(ctxt
), msr_index
, data
);
4352 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
4355 kvm_load_guest_fpu(ctxt
->vcpu
);
4357 * CR0.TS may reference the host fpu state, not the guest fpu state,
4358 * so it may be clear at this point.
4363 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
4368 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
4369 struct x86_instruction_info
*info
,
4370 enum x86_intercept_stage stage
)
4372 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
4375 static struct x86_emulate_ops emulate_ops
= {
4376 .read_std
= kvm_read_guest_virt_system
,
4377 .write_std
= kvm_write_guest_virt_system
,
4378 .fetch
= kvm_fetch_guest_virt
,
4379 .read_emulated
= emulator_read_emulated
,
4380 .write_emulated
= emulator_write_emulated
,
4381 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
4382 .pio_in_emulated
= emulator_pio_in_emulated
,
4383 .pio_out_emulated
= emulator_pio_out_emulated
,
4384 .get_cached_descriptor
= emulator_get_cached_descriptor
,
4385 .set_cached_descriptor
= emulator_set_cached_descriptor
,
4386 .get_segment_selector
= emulator_get_segment_selector
,
4387 .set_segment_selector
= emulator_set_segment_selector
,
4388 .get_cached_segment_base
= emulator_get_cached_segment_base
,
4389 .get_gdt
= emulator_get_gdt
,
4390 .get_idt
= emulator_get_idt
,
4391 .get_cr
= emulator_get_cr
,
4392 .set_cr
= emulator_set_cr
,
4393 .cpl
= emulator_get_cpl
,
4394 .get_dr
= emulator_get_dr
,
4395 .set_dr
= emulator_set_dr
,
4396 .set_msr
= emulator_set_msr
,
4397 .get_msr
= emulator_get_msr
,
4398 .get_fpu
= emulator_get_fpu
,
4399 .put_fpu
= emulator_put_fpu
,
4400 .intercept
= emulator_intercept
,
4403 static void cache_all_regs(struct kvm_vcpu
*vcpu
)
4405 kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4406 kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4407 kvm_register_read(vcpu
, VCPU_REGS_RIP
);
4408 vcpu
->arch
.regs_dirty
= ~0;
4411 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
4413 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
, mask
);
4415 * an sti; sti; sequence only disable interrupts for the first
4416 * instruction. So, if the last instruction, be it emulated or
4417 * not, left the system with the INT_STI flag enabled, it
4418 * means that the last instruction is an sti. We should not
4419 * leave the flag on in this case. The same goes for mov ss
4421 if (!(int_shadow
& mask
))
4422 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
4425 static void inject_emulated_exception(struct kvm_vcpu
*vcpu
)
4427 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4428 if (ctxt
->exception
.vector
== PF_VECTOR
)
4429 kvm_propagate_fault(vcpu
, &ctxt
->exception
);
4430 else if (ctxt
->exception
.error_code_valid
)
4431 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
4432 ctxt
->exception
.error_code
);
4434 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
4437 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
4439 struct decode_cache
*c
= &vcpu
->arch
.emulate_ctxt
.decode
;
4442 cache_all_regs(vcpu
);
4444 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
4446 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
4447 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_get_rflags(vcpu
);
4448 vcpu
->arch
.emulate_ctxt
.eip
= kvm_rip_read(vcpu
);
4449 vcpu
->arch
.emulate_ctxt
.mode
=
4450 (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
4451 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
4452 ? X86EMUL_MODE_VM86
: cs_l
4453 ? X86EMUL_MODE_PROT64
: cs_db
4454 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
4455 vcpu
->arch
.emulate_ctxt
.guest_mode
= is_guest_mode(vcpu
);
4456 memset(c
, 0, sizeof(struct decode_cache
));
4457 memcpy(c
->regs
, vcpu
->arch
.regs
, sizeof c
->regs
);
4458 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4461 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
4463 struct decode_cache
*c
= &vcpu
->arch
.emulate_ctxt
.decode
;
4466 init_emulate_ctxt(vcpu
);
4468 vcpu
->arch
.emulate_ctxt
.decode
.op_bytes
= 2;
4469 vcpu
->arch
.emulate_ctxt
.decode
.ad_bytes
= 2;
4470 vcpu
->arch
.emulate_ctxt
.decode
.eip
= vcpu
->arch
.emulate_ctxt
.eip
+
4472 ret
= emulate_int_real(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
, irq
);
4474 if (ret
!= X86EMUL_CONTINUE
)
4475 return EMULATE_FAIL
;
4477 vcpu
->arch
.emulate_ctxt
.eip
= c
->eip
;
4478 memcpy(vcpu
->arch
.regs
, c
->regs
, sizeof c
->regs
);
4479 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.eip
);
4480 kvm_set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
4482 if (irq
== NMI_VECTOR
)
4483 vcpu
->arch
.nmi_pending
= false;
4485 vcpu
->arch
.interrupt
.pending
= false;
4487 return EMULATE_DONE
;
4489 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
4491 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
4493 int r
= EMULATE_DONE
;
4495 ++vcpu
->stat
.insn_emulation_fail
;
4496 trace_kvm_emulate_insn_failed(vcpu
);
4497 if (!is_guest_mode(vcpu
)) {
4498 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
4499 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
4500 vcpu
->run
->internal
.ndata
= 0;
4503 kvm_queue_exception(vcpu
, UD_VECTOR
);
4508 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t gva
)
4516 * if emulation was due to access to shadowed page table
4517 * and it failed try to unshadow page and re-entetr the
4518 * guest to let CPU execute the instruction.
4520 if (kvm_mmu_unprotect_page_virt(vcpu
, gva
))
4523 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, gva
, NULL
);
4525 if (gpa
== UNMAPPED_GVA
)
4526 return true; /* let cpu generate fault */
4528 if (!kvm_is_error_hva(gfn_to_hva(vcpu
->kvm
, gpa
>> PAGE_SHIFT
)))
4534 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
4541 struct decode_cache
*c
= &vcpu
->arch
.emulate_ctxt
.decode
;
4542 bool writeback
= true;
4544 kvm_clear_exception_queue(vcpu
);
4545 vcpu
->arch
.mmio_fault_cr2
= cr2
;
4547 * TODO: fix emulate.c to use guest_read/write_register
4548 * instead of direct ->regs accesses, can save hundred cycles
4549 * on Intel for instructions that don't read/change RSP, for
4552 cache_all_regs(vcpu
);
4554 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
4555 init_emulate_ctxt(vcpu
);
4556 vcpu
->arch
.emulate_ctxt
.interruptibility
= 0;
4557 vcpu
->arch
.emulate_ctxt
.have_exception
= false;
4558 vcpu
->arch
.emulate_ctxt
.perm_ok
= false;
4560 vcpu
->arch
.emulate_ctxt
.only_vendor_specific_insn
4561 = emulation_type
& EMULTYPE_TRAP_UD
;
4563 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, insn
, insn_len
);
4565 trace_kvm_emulate_insn_start(vcpu
);
4566 ++vcpu
->stat
.insn_emulation
;
4568 if (emulation_type
& EMULTYPE_TRAP_UD
)
4569 return EMULATE_FAIL
;
4570 if (reexecute_instruction(vcpu
, cr2
))
4571 return EMULATE_DONE
;
4572 if (emulation_type
& EMULTYPE_SKIP
)
4573 return EMULATE_FAIL
;
4574 return handle_emulation_failure(vcpu
);
4578 if (emulation_type
& EMULTYPE_SKIP
) {
4579 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.decode
.eip
);
4580 return EMULATE_DONE
;
4583 /* this is needed for vmware backdoor interface to work since it
4584 changes registers values during IO operation */
4585 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
4586 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4587 memcpy(c
->regs
, vcpu
->arch
.regs
, sizeof c
->regs
);
4591 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
);
4593 if (r
== EMULATION_INTERCEPTED
)
4594 return EMULATE_DONE
;
4596 if (r
== EMULATION_FAILED
) {
4597 if (reexecute_instruction(vcpu
, cr2
))
4598 return EMULATE_DONE
;
4600 return handle_emulation_failure(vcpu
);
4603 if (vcpu
->arch
.emulate_ctxt
.have_exception
) {
4604 inject_emulated_exception(vcpu
);
4606 } else if (vcpu
->arch
.pio
.count
) {
4607 if (!vcpu
->arch
.pio
.in
)
4608 vcpu
->arch
.pio
.count
= 0;
4611 r
= EMULATE_DO_MMIO
;
4612 } else if (vcpu
->mmio_needed
) {
4613 if (!vcpu
->mmio_is_write
)
4615 r
= EMULATE_DO_MMIO
;
4616 } else if (r
== EMULATION_RESTART
)
4622 toggle_interruptibility(vcpu
,
4623 vcpu
->arch
.emulate_ctxt
.interruptibility
);
4624 kvm_set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
4625 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
4626 memcpy(vcpu
->arch
.regs
, c
->regs
, sizeof c
->regs
);
4627 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
4628 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.eip
);
4630 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
4634 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
4636 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
4638 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4639 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
4640 size
, port
, &val
, 1);
4641 /* do not return to emulator after return from userspace */
4642 vcpu
->arch
.pio
.count
= 0;
4645 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
4647 static void tsc_bad(void *info
)
4649 __this_cpu_write(cpu_tsc_khz
, 0);
4652 static void tsc_khz_changed(void *data
)
4654 struct cpufreq_freqs
*freq
= data
;
4655 unsigned long khz
= 0;
4659 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
4660 khz
= cpufreq_quick_get(raw_smp_processor_id());
4663 __this_cpu_write(cpu_tsc_khz
, khz
);
4666 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
4669 struct cpufreq_freqs
*freq
= data
;
4671 struct kvm_vcpu
*vcpu
;
4672 int i
, send_ipi
= 0;
4675 * We allow guests to temporarily run on slowing clocks,
4676 * provided we notify them after, or to run on accelerating
4677 * clocks, provided we notify them before. Thus time never
4680 * However, we have a problem. We can't atomically update
4681 * the frequency of a given CPU from this function; it is
4682 * merely a notifier, which can be called from any CPU.
4683 * Changing the TSC frequency at arbitrary points in time
4684 * requires a recomputation of local variables related to
4685 * the TSC for each VCPU. We must flag these local variables
4686 * to be updated and be sure the update takes place with the
4687 * new frequency before any guests proceed.
4689 * Unfortunately, the combination of hotplug CPU and frequency
4690 * change creates an intractable locking scenario; the order
4691 * of when these callouts happen is undefined with respect to
4692 * CPU hotplug, and they can race with each other. As such,
4693 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4694 * undefined; you can actually have a CPU frequency change take
4695 * place in between the computation of X and the setting of the
4696 * variable. To protect against this problem, all updates of
4697 * the per_cpu tsc_khz variable are done in an interrupt
4698 * protected IPI, and all callers wishing to update the value
4699 * must wait for a synchronous IPI to complete (which is trivial
4700 * if the caller is on the CPU already). This establishes the
4701 * necessary total order on variable updates.
4703 * Note that because a guest time update may take place
4704 * anytime after the setting of the VCPU's request bit, the
4705 * correct TSC value must be set before the request. However,
4706 * to ensure the update actually makes it to any guest which
4707 * starts running in hardware virtualization between the set
4708 * and the acquisition of the spinlock, we must also ping the
4709 * CPU after setting the request bit.
4713 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
4715 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
4718 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
4720 raw_spin_lock(&kvm_lock
);
4721 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4722 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
4723 if (vcpu
->cpu
!= freq
->cpu
)
4725 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
4726 if (vcpu
->cpu
!= smp_processor_id())
4730 raw_spin_unlock(&kvm_lock
);
4732 if (freq
->old
< freq
->new && send_ipi
) {
4734 * We upscale the frequency. Must make the guest
4735 * doesn't see old kvmclock values while running with
4736 * the new frequency, otherwise we risk the guest sees
4737 * time go backwards.
4739 * In case we update the frequency for another cpu
4740 * (which might be in guest context) send an interrupt
4741 * to kick the cpu out of guest context. Next time
4742 * guest context is entered kvmclock will be updated,
4743 * so the guest will not see stale values.
4745 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
4750 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
4751 .notifier_call
= kvmclock_cpufreq_notifier
4754 static int kvmclock_cpu_notifier(struct notifier_block
*nfb
,
4755 unsigned long action
, void *hcpu
)
4757 unsigned int cpu
= (unsigned long)hcpu
;
4761 case CPU_DOWN_FAILED
:
4762 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
4764 case CPU_DOWN_PREPARE
:
4765 smp_call_function_single(cpu
, tsc_bad
, NULL
, 1);
4771 static struct notifier_block kvmclock_cpu_notifier_block
= {
4772 .notifier_call
= kvmclock_cpu_notifier
,
4773 .priority
= -INT_MAX
4776 static void kvm_timer_init(void)
4780 max_tsc_khz
= tsc_khz
;
4781 register_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
4782 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
4783 #ifdef CONFIG_CPU_FREQ
4784 struct cpufreq_policy policy
;
4785 memset(&policy
, 0, sizeof(policy
));
4787 cpufreq_get_policy(&policy
, cpu
);
4788 if (policy
.cpuinfo
.max_freq
)
4789 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
4792 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
4793 CPUFREQ_TRANSITION_NOTIFIER
);
4795 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
4796 for_each_online_cpu(cpu
)
4797 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
4800 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
4802 static int kvm_is_in_guest(void)
4804 return percpu_read(current_vcpu
) != NULL
;
4807 static int kvm_is_user_mode(void)
4811 if (percpu_read(current_vcpu
))
4812 user_mode
= kvm_x86_ops
->get_cpl(percpu_read(current_vcpu
));
4814 return user_mode
!= 0;
4817 static unsigned long kvm_get_guest_ip(void)
4819 unsigned long ip
= 0;
4821 if (percpu_read(current_vcpu
))
4822 ip
= kvm_rip_read(percpu_read(current_vcpu
));
4827 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
4828 .is_in_guest
= kvm_is_in_guest
,
4829 .is_user_mode
= kvm_is_user_mode
,
4830 .get_guest_ip
= kvm_get_guest_ip
,
4833 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
4835 percpu_write(current_vcpu
, vcpu
);
4837 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
4839 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
4841 percpu_write(current_vcpu
, NULL
);
4843 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
4845 int kvm_arch_init(void *opaque
)
4848 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
4851 printk(KERN_ERR
"kvm: already loaded the other module\n");
4856 if (!ops
->cpu_has_kvm_support()) {
4857 printk(KERN_ERR
"kvm: no hardware support\n");
4861 if (ops
->disabled_by_bios()) {
4862 printk(KERN_ERR
"kvm: disabled by bios\n");
4867 r
= kvm_mmu_module_init();
4871 kvm_init_msr_list();
4874 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4875 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
4876 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
4880 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
4883 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
4891 void kvm_arch_exit(void)
4893 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
4895 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
4896 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
4897 CPUFREQ_TRANSITION_NOTIFIER
);
4898 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
4900 kvm_mmu_module_exit();
4903 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
4905 ++vcpu
->stat
.halt_exits
;
4906 if (irqchip_in_kernel(vcpu
->kvm
)) {
4907 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
4910 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
4914 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
4916 static inline gpa_t
hc_gpa(struct kvm_vcpu
*vcpu
, unsigned long a0
,
4919 if (is_long_mode(vcpu
))
4922 return a0
| ((gpa_t
)a1
<< 32);
4925 int kvm_hv_hypercall(struct kvm_vcpu
*vcpu
)
4927 u64 param
, ingpa
, outgpa
, ret
;
4928 uint16_t code
, rep_idx
, rep_cnt
, res
= HV_STATUS_SUCCESS
, rep_done
= 0;
4929 bool fast
, longmode
;
4933 * hypercall generates UD from non zero cpl and real mode
4936 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 || !is_protmode(vcpu
)) {
4937 kvm_queue_exception(vcpu
, UD_VECTOR
);
4941 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
4942 longmode
= is_long_mode(vcpu
) && cs_l
== 1;
4945 param
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDX
) << 32) |
4946 (kvm_register_read(vcpu
, VCPU_REGS_RAX
) & 0xffffffff);
4947 ingpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RBX
) << 32) |
4948 (kvm_register_read(vcpu
, VCPU_REGS_RCX
) & 0xffffffff);
4949 outgpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDI
) << 32) |
4950 (kvm_register_read(vcpu
, VCPU_REGS_RSI
) & 0xffffffff);
4952 #ifdef CONFIG_X86_64
4954 param
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4955 ingpa
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4956 outgpa
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
4960 code
= param
& 0xffff;
4961 fast
= (param
>> 16) & 0x1;
4962 rep_cnt
= (param
>> 32) & 0xfff;
4963 rep_idx
= (param
>> 48) & 0xfff;
4965 trace_kvm_hv_hypercall(code
, fast
, rep_cnt
, rep_idx
, ingpa
, outgpa
);
4968 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT
:
4969 kvm_vcpu_on_spin(vcpu
);
4972 res
= HV_STATUS_INVALID_HYPERCALL_CODE
;
4976 ret
= res
| (((u64
)rep_done
& 0xfff) << 32);
4978 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
4980 kvm_register_write(vcpu
, VCPU_REGS_RDX
, ret
>> 32);
4981 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
& 0xffffffff);
4987 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
4989 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
4992 if (kvm_hv_hypercall_enabled(vcpu
->kvm
))
4993 return kvm_hv_hypercall(vcpu
);
4995 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4996 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4997 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4998 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4999 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5001 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
5003 if (!is_long_mode(vcpu
)) {
5011 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
5017 case KVM_HC_VAPIC_POLL_IRQ
:
5021 r
= kvm_pv_mmu_op(vcpu
, a0
, hc_gpa(vcpu
, a1
, a2
), &ret
);
5028 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5029 ++vcpu
->stat
.hypercalls
;
5032 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
5034 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
5036 char instruction
[3];
5037 unsigned long rip
= kvm_rip_read(vcpu
);
5040 * Blow out the MMU to ensure that no other VCPU has an active mapping
5041 * to ensure that the updated hypercall appears atomically across all
5044 kvm_mmu_zap_all(vcpu
->kvm
);
5046 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
5048 return emulator_write_emulated(&vcpu
->arch
.emulate_ctxt
,
5049 rip
, instruction
, 3, NULL
);
5052 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
5054 struct desc_ptr dt
= { limit
, base
};
5056 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
5059 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
5061 struct desc_ptr dt
= { limit
, base
};
5063 kvm_x86_ops
->set_idt(vcpu
, &dt
);
5066 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
5068 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
5069 int j
, nent
= vcpu
->arch
.cpuid_nent
;
5071 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
5072 /* when no next entry is found, the current entry[i] is reselected */
5073 for (j
= i
+ 1; ; j
= (j
+ 1) % nent
) {
5074 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
5075 if (ej
->function
== e
->function
) {
5076 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
5080 return 0; /* silence gcc, even though control never reaches here */
5083 /* find an entry with matching function, matching index (if needed), and that
5084 * should be read next (if it's stateful) */
5085 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
5086 u32 function
, u32 index
)
5088 if (e
->function
!= function
)
5090 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
5092 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
5093 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
5098 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
5099 u32 function
, u32 index
)
5102 struct kvm_cpuid_entry2
*best
= NULL
;
5104 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
5105 struct kvm_cpuid_entry2
*e
;
5107 e
= &vcpu
->arch
.cpuid_entries
[i
];
5108 if (is_matching_cpuid_entry(e
, function
, index
)) {
5109 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
5110 move_to_next_stateful_cpuid_entry(vcpu
, i
);
5117 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry
);
5119 int cpuid_maxphyaddr(struct kvm_vcpu
*vcpu
)
5121 struct kvm_cpuid_entry2
*best
;
5123 best
= kvm_find_cpuid_entry(vcpu
, 0x80000000, 0);
5124 if (!best
|| best
->eax
< 0x80000008)
5126 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
5128 return best
->eax
& 0xff;
5134 * If no match is found, check whether we exceed the vCPU's limit
5135 * and return the content of the highest valid _standard_ leaf instead.
5136 * This is to satisfy the CPUID specification.
5138 static struct kvm_cpuid_entry2
* check_cpuid_limit(struct kvm_vcpu
*vcpu
,
5139 u32 function
, u32 index
)
5141 struct kvm_cpuid_entry2
*maxlevel
;
5143 maxlevel
= kvm_find_cpuid_entry(vcpu
, function
& 0x80000000, 0);
5144 if (!maxlevel
|| maxlevel
->eax
>= function
)
5146 if (function
& 0x80000000) {
5147 maxlevel
= kvm_find_cpuid_entry(vcpu
, 0, 0);
5151 return kvm_find_cpuid_entry(vcpu
, maxlevel
->eax
, index
);
5154 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
5156 u32 function
, index
;
5157 struct kvm_cpuid_entry2
*best
;
5159 function
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5160 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5161 kvm_register_write(vcpu
, VCPU_REGS_RAX
, 0);
5162 kvm_register_write(vcpu
, VCPU_REGS_RBX
, 0);
5163 kvm_register_write(vcpu
, VCPU_REGS_RCX
, 0);
5164 kvm_register_write(vcpu
, VCPU_REGS_RDX
, 0);
5165 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
5168 best
= check_cpuid_limit(vcpu
, function
, index
);
5171 kvm_register_write(vcpu
, VCPU_REGS_RAX
, best
->eax
);
5172 kvm_register_write(vcpu
, VCPU_REGS_RBX
, best
->ebx
);
5173 kvm_register_write(vcpu
, VCPU_REGS_RCX
, best
->ecx
);
5174 kvm_register_write(vcpu
, VCPU_REGS_RDX
, best
->edx
);
5176 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
5177 trace_kvm_cpuid(function
,
5178 kvm_register_read(vcpu
, VCPU_REGS_RAX
),
5179 kvm_register_read(vcpu
, VCPU_REGS_RBX
),
5180 kvm_register_read(vcpu
, VCPU_REGS_RCX
),
5181 kvm_register_read(vcpu
, VCPU_REGS_RDX
));
5183 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
5186 * Check if userspace requested an interrupt window, and that the
5187 * interrupt window is open.
5189 * No need to exit to userspace if we already have an interrupt queued.
5191 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
5193 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
5194 vcpu
->run
->request_interrupt_window
&&
5195 kvm_arch_interrupt_allowed(vcpu
));
5198 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
5200 struct kvm_run
*kvm_run
= vcpu
->run
;
5202 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
5203 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
5204 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
5205 if (irqchip_in_kernel(vcpu
->kvm
))
5206 kvm_run
->ready_for_interrupt_injection
= 1;
5208 kvm_run
->ready_for_interrupt_injection
=
5209 kvm_arch_interrupt_allowed(vcpu
) &&
5210 !kvm_cpu_has_interrupt(vcpu
) &&
5211 !kvm_event_needs_reinjection(vcpu
);
5214 static void vapic_enter(struct kvm_vcpu
*vcpu
)
5216 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5219 if (!apic
|| !apic
->vapic_addr
)
5222 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5224 vcpu
->arch
.apic
->vapic_page
= page
;
5227 static void vapic_exit(struct kvm_vcpu
*vcpu
)
5229 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5232 if (!apic
|| !apic
->vapic_addr
)
5235 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5236 kvm_release_page_dirty(apic
->vapic_page
);
5237 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5238 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5241 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
5245 if (!kvm_x86_ops
->update_cr8_intercept
)
5248 if (!vcpu
->arch
.apic
)
5251 if (!vcpu
->arch
.apic
->vapic_addr
)
5252 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
5259 tpr
= kvm_lapic_get_cr8(vcpu
);
5261 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
5264 static void inject_pending_event(struct kvm_vcpu
*vcpu
)
5266 /* try to reinject previous events if any */
5267 if (vcpu
->arch
.exception
.pending
) {
5268 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
5269 vcpu
->arch
.exception
.has_error_code
,
5270 vcpu
->arch
.exception
.error_code
);
5271 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
5272 vcpu
->arch
.exception
.has_error_code
,
5273 vcpu
->arch
.exception
.error_code
,
5274 vcpu
->arch
.exception
.reinject
);
5278 if (vcpu
->arch
.nmi_injected
) {
5279 kvm_x86_ops
->set_nmi(vcpu
);
5283 if (vcpu
->arch
.interrupt
.pending
) {
5284 kvm_x86_ops
->set_irq(vcpu
);
5288 /* try to inject new event if pending */
5289 if (vcpu
->arch
.nmi_pending
) {
5290 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
5291 vcpu
->arch
.nmi_pending
= false;
5292 vcpu
->arch
.nmi_injected
= true;
5293 kvm_x86_ops
->set_nmi(vcpu
);
5295 } else if (kvm_cpu_has_interrupt(vcpu
)) {
5296 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
5297 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
5299 kvm_x86_ops
->set_irq(vcpu
);
5304 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
5306 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
5307 !vcpu
->guest_xcr0_loaded
) {
5308 /* kvm_set_xcr() also depends on this */
5309 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
5310 vcpu
->guest_xcr0_loaded
= 1;
5314 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
5316 if (vcpu
->guest_xcr0_loaded
) {
5317 if (vcpu
->arch
.xcr0
!= host_xcr0
)
5318 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
5319 vcpu
->guest_xcr0_loaded
= 0;
5323 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
5327 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
5328 vcpu
->run
->request_interrupt_window
;
5330 if (vcpu
->requests
) {
5331 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
5332 kvm_mmu_unload(vcpu
);
5333 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
5334 __kvm_migrate_timers(vcpu
);
5335 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
5336 r
= kvm_guest_time_update(vcpu
);
5340 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
5341 kvm_mmu_sync_roots(vcpu
);
5342 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
5343 kvm_x86_ops
->tlb_flush(vcpu
);
5344 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
5345 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
5349 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
5350 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
5354 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
)) {
5355 vcpu
->fpu_active
= 0;
5356 kvm_x86_ops
->fpu_deactivate(vcpu
);
5358 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
5359 /* Page is swapped out. Do synthetic halt */
5360 vcpu
->arch
.apf
.halted
= true;
5366 r
= kvm_mmu_reload(vcpu
);
5371 * An NMI can be injected between local nmi_pending read and
5372 * vcpu->arch.nmi_pending read inside inject_pending_event().
5373 * But in that case, KVM_REQ_EVENT will be set, which makes
5374 * the race described above benign.
5376 nmi_pending
= ACCESS_ONCE(vcpu
->arch
.nmi_pending
);
5378 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
5379 inject_pending_event(vcpu
);
5381 /* enable NMI/IRQ window open exits if needed */
5383 kvm_x86_ops
->enable_nmi_window(vcpu
);
5384 else if (kvm_cpu_has_interrupt(vcpu
) || req_int_win
)
5385 kvm_x86_ops
->enable_irq_window(vcpu
);
5387 if (kvm_lapic_enabled(vcpu
)) {
5388 update_cr8_intercept(vcpu
);
5389 kvm_lapic_sync_to_vapic(vcpu
);
5395 kvm_x86_ops
->prepare_guest_switch(vcpu
);
5396 if (vcpu
->fpu_active
)
5397 kvm_load_guest_fpu(vcpu
);
5398 kvm_load_guest_xcr0(vcpu
);
5400 vcpu
->mode
= IN_GUEST_MODE
;
5402 /* We should set ->mode before check ->requests,
5403 * see the comment in make_all_cpus_request.
5407 local_irq_disable();
5409 if (vcpu
->mode
== EXITING_GUEST_MODE
|| vcpu
->requests
5410 || need_resched() || signal_pending(current
)) {
5411 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5415 kvm_x86_ops
->cancel_injection(vcpu
);
5420 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5424 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
5426 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
5427 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
5428 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
5429 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
5432 trace_kvm_entry(vcpu
->vcpu_id
);
5433 kvm_x86_ops
->run(vcpu
);
5436 * If the guest has used debug registers, at least dr7
5437 * will be disabled while returning to the host.
5438 * If we don't have active breakpoints in the host, we don't
5439 * care about the messed up debug address registers. But if
5440 * we have some of them active, restore the old state.
5442 if (hw_breakpoint_active())
5443 hw_breakpoint_restore();
5445 kvm_get_msr(vcpu
, MSR_IA32_TSC
, &vcpu
->arch
.last_guest_tsc
);
5447 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5454 * We must have an instruction between local_irq_enable() and
5455 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5456 * the interrupt shadow. The stat.exits increment will do nicely.
5457 * But we need to prevent reordering, hence this barrier():
5465 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5468 * Profile KVM exit RIPs:
5470 if (unlikely(prof_on
== KVM_PROFILING
)) {
5471 unsigned long rip
= kvm_rip_read(vcpu
);
5472 profile_hit(KVM_PROFILING
, (void *)rip
);
5476 kvm_lapic_sync_from_vapic(vcpu
);
5478 r
= kvm_x86_ops
->handle_exit(vcpu
);
5484 static int __vcpu_run(struct kvm_vcpu
*vcpu
)
5487 struct kvm
*kvm
= vcpu
->kvm
;
5489 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
5490 pr_debug("vcpu %d received sipi with vector # %x\n",
5491 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
5492 kvm_lapic_reset(vcpu
);
5493 r
= kvm_arch_vcpu_reset(vcpu
);
5496 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
5499 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5504 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
5505 !vcpu
->arch
.apf
.halted
)
5506 r
= vcpu_enter_guest(vcpu
);
5508 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5509 kvm_vcpu_block(vcpu
);
5510 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5511 if (kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
5513 switch(vcpu
->arch
.mp_state
) {
5514 case KVM_MP_STATE_HALTED
:
5515 vcpu
->arch
.mp_state
=
5516 KVM_MP_STATE_RUNNABLE
;
5517 case KVM_MP_STATE_RUNNABLE
:
5518 vcpu
->arch
.apf
.halted
= false;
5520 case KVM_MP_STATE_SIPI_RECEIVED
:
5531 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
5532 if (kvm_cpu_has_pending_timer(vcpu
))
5533 kvm_inject_pending_timer_irqs(vcpu
);
5535 if (dm_request_for_irq_injection(vcpu
)) {
5537 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5538 ++vcpu
->stat
.request_irq_exits
;
5541 kvm_check_async_pf_completion(vcpu
);
5543 if (signal_pending(current
)) {
5545 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5546 ++vcpu
->stat
.signal_exits
;
5548 if (need_resched()) {
5549 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5551 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5555 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5562 static int complete_mmio(struct kvm_vcpu
*vcpu
)
5564 struct kvm_run
*run
= vcpu
->run
;
5567 if (!(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
))
5570 if (vcpu
->mmio_needed
) {
5571 vcpu
->mmio_needed
= 0;
5572 if (!vcpu
->mmio_is_write
)
5573 memcpy(vcpu
->mmio_data
, run
->mmio
.data
, 8);
5574 vcpu
->mmio_index
+= 8;
5575 if (vcpu
->mmio_index
< vcpu
->mmio_size
) {
5576 run
->exit_reason
= KVM_EXIT_MMIO
;
5577 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
+ vcpu
->mmio_index
;
5578 memcpy(run
->mmio
.data
, vcpu
->mmio_data
+ vcpu
->mmio_index
, 8);
5579 run
->mmio
.len
= min(vcpu
->mmio_size
- vcpu
->mmio_index
, 8);
5580 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
5581 vcpu
->mmio_needed
= 1;
5584 if (vcpu
->mmio_is_write
)
5586 vcpu
->mmio_read_completed
= 1;
5588 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5589 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
5590 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5591 if (r
!= EMULATE_DONE
)
5596 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
5601 if (!tsk_used_math(current
) && init_fpu(current
))
5604 if (vcpu
->sigset_active
)
5605 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
5607 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
5608 kvm_vcpu_block(vcpu
);
5609 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
5614 /* re-sync apic's tpr */
5615 if (!irqchip_in_kernel(vcpu
->kvm
)) {
5616 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
5622 r
= complete_mmio(vcpu
);
5626 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
)
5627 kvm_register_write(vcpu
, VCPU_REGS_RAX
,
5628 kvm_run
->hypercall
.ret
);
5630 r
= __vcpu_run(vcpu
);
5633 post_kvm_run_save(vcpu
);
5634 if (vcpu
->sigset_active
)
5635 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
5640 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
5642 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
5644 * We are here if userspace calls get_regs() in the middle of
5645 * instruction emulation. Registers state needs to be copied
5646 * back from emulation context to vcpu. Usrapace shouldn't do
5647 * that usually, but some bad designed PV devices (vmware
5648 * backdoor interface) need this to work
5650 struct decode_cache
*c
= &vcpu
->arch
.emulate_ctxt
.decode
;
5651 memcpy(vcpu
->arch
.regs
, c
->regs
, sizeof c
->regs
);
5652 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5654 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5655 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5656 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5657 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5658 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5659 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
5660 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
5661 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
5662 #ifdef CONFIG_X86_64
5663 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
5664 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
5665 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
5666 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
5667 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
5668 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
5669 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
5670 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
5673 regs
->rip
= kvm_rip_read(vcpu
);
5674 regs
->rflags
= kvm_get_rflags(vcpu
);
5679 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
5681 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
5682 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5684 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
5685 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
5686 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
5687 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
5688 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
5689 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
5690 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
5691 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
5692 #ifdef CONFIG_X86_64
5693 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
5694 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
5695 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
5696 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
5697 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
5698 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
5699 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
5700 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
5703 kvm_rip_write(vcpu
, regs
->rip
);
5704 kvm_set_rflags(vcpu
, regs
->rflags
);
5706 vcpu
->arch
.exception
.pending
= false;
5708 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5713 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
5715 struct kvm_segment cs
;
5717 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
5721 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
5723 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
5724 struct kvm_sregs
*sregs
)
5728 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
5729 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
5730 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
5731 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
5732 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
5733 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
5735 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
5736 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
5738 kvm_x86_ops
->get_idt(vcpu
, &dt
);
5739 sregs
->idt
.limit
= dt
.size
;
5740 sregs
->idt
.base
= dt
.address
;
5741 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
5742 sregs
->gdt
.limit
= dt
.size
;
5743 sregs
->gdt
.base
= dt
.address
;
5745 sregs
->cr0
= kvm_read_cr0(vcpu
);
5746 sregs
->cr2
= vcpu
->arch
.cr2
;
5747 sregs
->cr3
= kvm_read_cr3(vcpu
);
5748 sregs
->cr4
= kvm_read_cr4(vcpu
);
5749 sregs
->cr8
= kvm_get_cr8(vcpu
);
5750 sregs
->efer
= vcpu
->arch
.efer
;
5751 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
5753 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
5755 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
5756 set_bit(vcpu
->arch
.interrupt
.nr
,
5757 (unsigned long *)sregs
->interrupt_bitmap
);
5762 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
5763 struct kvm_mp_state
*mp_state
)
5765 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
5769 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
5770 struct kvm_mp_state
*mp_state
)
5772 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
5773 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5777 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int reason
,
5778 bool has_error_code
, u32 error_code
)
5780 struct decode_cache
*c
= &vcpu
->arch
.emulate_ctxt
.decode
;
5783 init_emulate_ctxt(vcpu
);
5785 ret
= emulator_task_switch(&vcpu
->arch
.emulate_ctxt
,
5786 tss_selector
, reason
, has_error_code
,
5790 return EMULATE_FAIL
;
5792 memcpy(vcpu
->arch
.regs
, c
->regs
, sizeof c
->regs
);
5793 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.eip
);
5794 kvm_set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
5795 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5796 return EMULATE_DONE
;
5798 EXPORT_SYMBOL_GPL(kvm_task_switch
);
5800 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
5801 struct kvm_sregs
*sregs
)
5803 int mmu_reset_needed
= 0;
5804 int pending_vec
, max_bits
, idx
;
5807 dt
.size
= sregs
->idt
.limit
;
5808 dt
.address
= sregs
->idt
.base
;
5809 kvm_x86_ops
->set_idt(vcpu
, &dt
);
5810 dt
.size
= sregs
->gdt
.limit
;
5811 dt
.address
= sregs
->gdt
.base
;
5812 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
5814 vcpu
->arch
.cr2
= sregs
->cr2
;
5815 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
5816 vcpu
->arch
.cr3
= sregs
->cr3
;
5817 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
5819 kvm_set_cr8(vcpu
, sregs
->cr8
);
5821 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
5822 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
5823 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
5825 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
5826 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
5827 vcpu
->arch
.cr0
= sregs
->cr0
;
5829 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
5830 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
5831 if (sregs
->cr4
& X86_CR4_OSXSAVE
)
5834 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5835 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
5836 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
5837 mmu_reset_needed
= 1;
5839 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5841 if (mmu_reset_needed
)
5842 kvm_mmu_reset_context(vcpu
);
5844 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
5845 pending_vec
= find_first_bit(
5846 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
5847 if (pending_vec
< max_bits
) {
5848 kvm_queue_interrupt(vcpu
, pending_vec
, false);
5849 pr_debug("Set back pending irq %d\n", pending_vec
);
5852 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
5853 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
5854 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
5855 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
5856 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
5857 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
5859 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
5860 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
5862 update_cr8_intercept(vcpu
);
5864 /* Older userspace won't unhalt the vcpu on reset. */
5865 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
5866 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
5868 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
5870 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5875 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
5876 struct kvm_guest_debug
*dbg
)
5878 unsigned long rflags
;
5881 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
5883 if (vcpu
->arch
.exception
.pending
)
5885 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
5886 kvm_queue_exception(vcpu
, DB_VECTOR
);
5888 kvm_queue_exception(vcpu
, BP_VECTOR
);
5892 * Read rflags as long as potentially injected trace flags are still
5895 rflags
= kvm_get_rflags(vcpu
);
5897 vcpu
->guest_debug
= dbg
->control
;
5898 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
5899 vcpu
->guest_debug
= 0;
5901 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
5902 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
5903 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
5904 vcpu
->arch
.switch_db_regs
=
5905 (dbg
->arch
.debugreg
[7] & DR7_BP_EN_MASK
);
5907 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
5908 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
5909 vcpu
->arch
.switch_db_regs
= (vcpu
->arch
.dr7
& DR7_BP_EN_MASK
);
5912 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
5913 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
5914 get_segment_base(vcpu
, VCPU_SREG_CS
);
5917 * Trigger an rflags update that will inject or remove the trace
5920 kvm_set_rflags(vcpu
, rflags
);
5922 kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
5932 * Translate a guest virtual address to a guest physical address.
5934 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
5935 struct kvm_translation
*tr
)
5937 unsigned long vaddr
= tr
->linear_address
;
5941 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5942 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
5943 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5944 tr
->physical_address
= gpa
;
5945 tr
->valid
= gpa
!= UNMAPPED_GVA
;
5952 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
5954 struct i387_fxsave_struct
*fxsave
=
5955 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
5957 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
5958 fpu
->fcw
= fxsave
->cwd
;
5959 fpu
->fsw
= fxsave
->swd
;
5960 fpu
->ftwx
= fxsave
->twd
;
5961 fpu
->last_opcode
= fxsave
->fop
;
5962 fpu
->last_ip
= fxsave
->rip
;
5963 fpu
->last_dp
= fxsave
->rdp
;
5964 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
5969 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
5971 struct i387_fxsave_struct
*fxsave
=
5972 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
5974 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
5975 fxsave
->cwd
= fpu
->fcw
;
5976 fxsave
->swd
= fpu
->fsw
;
5977 fxsave
->twd
= fpu
->ftwx
;
5978 fxsave
->fop
= fpu
->last_opcode
;
5979 fxsave
->rip
= fpu
->last_ip
;
5980 fxsave
->rdp
= fpu
->last_dp
;
5981 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
5986 int fx_init(struct kvm_vcpu
*vcpu
)
5990 err
= fpu_alloc(&vcpu
->arch
.guest_fpu
);
5994 fpu_finit(&vcpu
->arch
.guest_fpu
);
5997 * Ensure guest xcr0 is valid for loading
5999 vcpu
->arch
.xcr0
= XSTATE_FP
;
6001 vcpu
->arch
.cr0
|= X86_CR0_ET
;
6005 EXPORT_SYMBOL_GPL(fx_init
);
6007 static void fx_free(struct kvm_vcpu
*vcpu
)
6009 fpu_free(&vcpu
->arch
.guest_fpu
);
6012 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
6014 if (vcpu
->guest_fpu_loaded
)
6018 * Restore all possible states in the guest,
6019 * and assume host would use all available bits.
6020 * Guest xcr0 would be loaded later.
6022 kvm_put_guest_xcr0(vcpu
);
6023 vcpu
->guest_fpu_loaded
= 1;
6024 unlazy_fpu(current
);
6025 fpu_restore_checking(&vcpu
->arch
.guest_fpu
);
6029 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
6031 kvm_put_guest_xcr0(vcpu
);
6033 if (!vcpu
->guest_fpu_loaded
)
6036 vcpu
->guest_fpu_loaded
= 0;
6037 fpu_save_init(&vcpu
->arch
.guest_fpu
);
6038 ++vcpu
->stat
.fpu_reload
;
6039 kvm_make_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
);
6043 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
6045 kvmclock_reset(vcpu
);
6047 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
6049 kvm_x86_ops
->vcpu_free(vcpu
);
6052 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
6055 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
6056 printk_once(KERN_WARNING
6057 "kvm: SMP vm created on host with unstable TSC; "
6058 "guest TSC will not be reliable\n");
6059 return kvm_x86_ops
->vcpu_create(kvm
, id
);
6062 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
6066 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
6068 r
= kvm_arch_vcpu_reset(vcpu
);
6070 r
= kvm_mmu_setup(vcpu
);
6077 kvm_x86_ops
->vcpu_free(vcpu
);
6081 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
6083 vcpu
->arch
.apf
.msr_val
= 0;
6086 kvm_mmu_unload(vcpu
);
6090 kvm_x86_ops
->vcpu_free(vcpu
);
6093 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
6095 vcpu
->arch
.nmi_pending
= false;
6096 vcpu
->arch
.nmi_injected
= false;
6098 vcpu
->arch
.switch_db_regs
= 0;
6099 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
6100 vcpu
->arch
.dr6
= DR6_FIXED_1
;
6101 vcpu
->arch
.dr7
= DR7_FIXED_1
;
6103 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6104 vcpu
->arch
.apf
.msr_val
= 0;
6106 kvmclock_reset(vcpu
);
6108 kvm_clear_async_pf_completion_queue(vcpu
);
6109 kvm_async_pf_hash_reset(vcpu
);
6110 vcpu
->arch
.apf
.halted
= false;
6112 return kvm_x86_ops
->vcpu_reset(vcpu
);
6115 int kvm_arch_hardware_enable(void *garbage
)
6118 struct kvm_vcpu
*vcpu
;
6121 kvm_shared_msr_cpu_online();
6122 list_for_each_entry(kvm
, &vm_list
, vm_list
)
6123 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6124 if (vcpu
->cpu
== smp_processor_id())
6125 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
6126 return kvm_x86_ops
->hardware_enable(garbage
);
6129 void kvm_arch_hardware_disable(void *garbage
)
6131 kvm_x86_ops
->hardware_disable(garbage
);
6132 drop_user_return_notifiers(garbage
);
6135 int kvm_arch_hardware_setup(void)
6137 return kvm_x86_ops
->hardware_setup();
6140 void kvm_arch_hardware_unsetup(void)
6142 kvm_x86_ops
->hardware_unsetup();
6145 void kvm_arch_check_processor_compat(void *rtn
)
6147 kvm_x86_ops
->check_processor_compatibility(rtn
);
6150 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
6156 BUG_ON(vcpu
->kvm
== NULL
);
6159 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
6160 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
6161 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
6162 vcpu
->arch
.mmu
.translate_gpa
= translate_gpa
;
6163 vcpu
->arch
.nested_mmu
.translate_gpa
= translate_nested_gpa
;
6164 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
6165 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6167 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
6169 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
6174 vcpu
->arch
.pio_data
= page_address(page
);
6176 kvm_init_tsc_catchup(vcpu
, max_tsc_khz
);
6178 r
= kvm_mmu_create(vcpu
);
6180 goto fail_free_pio_data
;
6182 if (irqchip_in_kernel(kvm
)) {
6183 r
= kvm_create_lapic(vcpu
);
6185 goto fail_mmu_destroy
;
6188 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
6190 if (!vcpu
->arch
.mce_banks
) {
6192 goto fail_free_lapic
;
6194 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
6196 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
))
6197 goto fail_free_mce_banks
;
6199 kvm_async_pf_hash_reset(vcpu
);
6202 fail_free_mce_banks
:
6203 kfree(vcpu
->arch
.mce_banks
);
6205 kvm_free_lapic(vcpu
);
6207 kvm_mmu_destroy(vcpu
);
6209 free_page((unsigned long)vcpu
->arch
.pio_data
);
6214 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
6218 kfree(vcpu
->arch
.mce_banks
);
6219 kvm_free_lapic(vcpu
);
6220 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6221 kvm_mmu_destroy(vcpu
);
6222 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6223 free_page((unsigned long)vcpu
->arch
.pio_data
);
6226 int kvm_arch_init_vm(struct kvm
*kvm
)
6228 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
6229 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
6231 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6232 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
6234 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
6239 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
6242 kvm_mmu_unload(vcpu
);
6246 static void kvm_free_vcpus(struct kvm
*kvm
)
6249 struct kvm_vcpu
*vcpu
;
6252 * Unpin any mmu pages first.
6254 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6255 kvm_clear_async_pf_completion_queue(vcpu
);
6256 kvm_unload_vcpu_mmu(vcpu
);
6258 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6259 kvm_arch_vcpu_free(vcpu
);
6261 mutex_lock(&kvm
->lock
);
6262 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
6263 kvm
->vcpus
[i
] = NULL
;
6265 atomic_set(&kvm
->online_vcpus
, 0);
6266 mutex_unlock(&kvm
->lock
);
6269 void kvm_arch_sync_events(struct kvm
*kvm
)
6271 kvm_free_all_assigned_devices(kvm
);
6275 void kvm_arch_destroy_vm(struct kvm
*kvm
)
6277 kvm_iommu_unmap_guest(kvm
);
6278 kfree(kvm
->arch
.vpic
);
6279 kfree(kvm
->arch
.vioapic
);
6280 kvm_free_vcpus(kvm
);
6281 if (kvm
->arch
.apic_access_page
)
6282 put_page(kvm
->arch
.apic_access_page
);
6283 if (kvm
->arch
.ept_identity_pagetable
)
6284 put_page(kvm
->arch
.ept_identity_pagetable
);
6287 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
6288 struct kvm_memory_slot
*memslot
,
6289 struct kvm_memory_slot old
,
6290 struct kvm_userspace_memory_region
*mem
,
6293 int npages
= memslot
->npages
;
6294 int map_flags
= MAP_PRIVATE
| MAP_ANONYMOUS
;
6296 /* Prevent internal slot pages from being moved by fork()/COW. */
6297 if (memslot
->id
>= KVM_MEMORY_SLOTS
)
6298 map_flags
= MAP_SHARED
| MAP_ANONYMOUS
;
6300 /*To keep backward compatibility with older userspace,
6301 *x86 needs to hanlde !user_alloc case.
6304 if (npages
&& !old
.rmap
) {
6305 unsigned long userspace_addr
;
6307 down_write(¤t
->mm
->mmap_sem
);
6308 userspace_addr
= do_mmap(NULL
, 0,
6310 PROT_READ
| PROT_WRITE
,
6313 up_write(¤t
->mm
->mmap_sem
);
6315 if (IS_ERR((void *)userspace_addr
))
6316 return PTR_ERR((void *)userspace_addr
);
6318 memslot
->userspace_addr
= userspace_addr
;
6326 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
6327 struct kvm_userspace_memory_region
*mem
,
6328 struct kvm_memory_slot old
,
6332 int nr_mmu_pages
= 0, npages
= mem
->memory_size
>> PAGE_SHIFT
;
6334 if (!user_alloc
&& !old
.user_alloc
&& old
.rmap
&& !npages
) {
6337 down_write(¤t
->mm
->mmap_sem
);
6338 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
6339 old
.npages
* PAGE_SIZE
);
6340 up_write(¤t
->mm
->mmap_sem
);
6343 "kvm_vm_ioctl_set_memory_region: "
6344 "failed to munmap memory\n");
6347 if (!kvm
->arch
.n_requested_mmu_pages
)
6348 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
6350 spin_lock(&kvm
->mmu_lock
);
6352 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
6353 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
6354 spin_unlock(&kvm
->mmu_lock
);
6357 void kvm_arch_flush_shadow(struct kvm
*kvm
)
6359 kvm_mmu_zap_all(kvm
);
6360 kvm_reload_remote_mmus(kvm
);
6363 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
6365 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
6366 !vcpu
->arch
.apf
.halted
)
6367 || !list_empty_careful(&vcpu
->async_pf
.done
)
6368 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
6369 || vcpu
->arch
.nmi_pending
||
6370 (kvm_arch_interrupt_allowed(vcpu
) &&
6371 kvm_cpu_has_interrupt(vcpu
));
6374 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
6377 int cpu
= vcpu
->cpu
;
6379 if (waitqueue_active(&vcpu
->wq
)) {
6380 wake_up_interruptible(&vcpu
->wq
);
6381 ++vcpu
->stat
.halt_wakeup
;
6385 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
6386 if (kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
)
6387 smp_send_reschedule(cpu
);
6391 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
6393 return kvm_x86_ops
->interrupt_allowed(vcpu
);
6396 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
6398 unsigned long current_rip
= kvm_rip_read(vcpu
) +
6399 get_segment_base(vcpu
, VCPU_SREG_CS
);
6401 return current_rip
== linear_rip
;
6403 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
6405 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
6407 unsigned long rflags
;
6409 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
6410 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
6411 rflags
&= ~X86_EFLAGS_TF
;
6414 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
6416 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
6418 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
6419 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
6420 rflags
|= X86_EFLAGS_TF
;
6421 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
6422 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6424 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
6426 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
6430 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
6431 is_error_page(work
->page
))
6434 r
= kvm_mmu_reload(vcpu
);
6438 if (!vcpu
->arch
.mmu
.direct_map
&&
6439 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
6442 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
6445 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
6447 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
6450 static inline u32
kvm_async_pf_next_probe(u32 key
)
6452 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
6455 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6457 u32 key
= kvm_async_pf_hash_fn(gfn
);
6459 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
6460 key
= kvm_async_pf_next_probe(key
);
6462 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
6465 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6468 u32 key
= kvm_async_pf_hash_fn(gfn
);
6470 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
6471 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
6472 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
6473 key
= kvm_async_pf_next_probe(key
);
6478 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6480 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
6483 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6487 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
6489 vcpu
->arch
.apf
.gfns
[i
] = ~0;
6491 j
= kvm_async_pf_next_probe(j
);
6492 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
6494 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
6496 * k lies cyclically in ]i,j]
6498 * |....j i.k.| or |.k..j i...|
6500 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
6501 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
6506 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
6509 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
6513 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
6514 struct kvm_async_pf
*work
)
6516 struct x86_exception fault
;
6518 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
6519 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
6521 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
6522 (vcpu
->arch
.apf
.send_user_only
&&
6523 kvm_x86_ops
->get_cpl(vcpu
) == 0))
6524 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
6525 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
6526 fault
.vector
= PF_VECTOR
;
6527 fault
.error_code_valid
= true;
6528 fault
.error_code
= 0;
6529 fault
.nested_page_fault
= false;
6530 fault
.address
= work
->arch
.token
;
6531 kvm_inject_page_fault(vcpu
, &fault
);
6535 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
6536 struct kvm_async_pf
*work
)
6538 struct x86_exception fault
;
6540 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
6541 if (is_error_page(work
->page
))
6542 work
->arch
.token
= ~0; /* broadcast wakeup */
6544 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
6546 if ((vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) &&
6547 !apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
6548 fault
.vector
= PF_VECTOR
;
6549 fault
.error_code_valid
= true;
6550 fault
.error_code
= 0;
6551 fault
.nested_page_fault
= false;
6552 fault
.address
= work
->arch
.token
;
6553 kvm_inject_page_fault(vcpu
, &fault
);
6555 vcpu
->arch
.apf
.halted
= false;
6558 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
6560 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
6563 return !kvm_event_needs_reinjection(vcpu
) &&
6564 kvm_x86_ops
->interrupt_allowed(vcpu
);
6567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
6568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
6569 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
6570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
6571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
6572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
6573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
6574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
6575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
6576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
6577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
6578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);