2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <trace/events/kvm.h>
53 #define CREATE_TRACE_POINTS
56 #include <asm/debugreg.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 #define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
80 u64 __read_mostly efer_reserved_bits
= ~((u64
)(EFER_SCE
| EFER_LME
| EFER_LMA
));
82 static u64 __read_mostly efer_reserved_bits
= ~((u64
)EFER_SCE
);
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
);
89 static void process_nmi(struct kvm_vcpu
*vcpu
);
91 struct kvm_x86_ops
*kvm_x86_ops
;
92 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
94 static bool ignore_msrs
= 0;
95 module_param(ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
97 bool kvm_has_tsc_control
;
98 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
99 u32 kvm_max_guest_tsc_khz
;
100 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
102 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
103 static u32 tsc_tolerance_ppm
= 250;
104 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
106 #define KVM_NR_SHARED_MSRS 16
108 struct kvm_shared_msrs_global
{
110 u32 msrs
[KVM_NR_SHARED_MSRS
];
113 struct kvm_shared_msrs
{
114 struct user_return_notifier urn
;
116 struct kvm_shared_msr_values
{
119 } values
[KVM_NR_SHARED_MSRS
];
122 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
123 static DEFINE_PER_CPU(struct kvm_shared_msrs
, shared_msrs
);
125 struct kvm_stats_debugfs_item debugfs_entries
[] = {
126 { "pf_fixed", VCPU_STAT(pf_fixed
) },
127 { "pf_guest", VCPU_STAT(pf_guest
) },
128 { "tlb_flush", VCPU_STAT(tlb_flush
) },
129 { "invlpg", VCPU_STAT(invlpg
) },
130 { "exits", VCPU_STAT(exits
) },
131 { "io_exits", VCPU_STAT(io_exits
) },
132 { "mmio_exits", VCPU_STAT(mmio_exits
) },
133 { "signal_exits", VCPU_STAT(signal_exits
) },
134 { "irq_window", VCPU_STAT(irq_window_exits
) },
135 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
136 { "halt_exits", VCPU_STAT(halt_exits
) },
137 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
138 { "hypercalls", VCPU_STAT(hypercalls
) },
139 { "request_irq", VCPU_STAT(request_irq_exits
) },
140 { "irq_exits", VCPU_STAT(irq_exits
) },
141 { "host_state_reload", VCPU_STAT(host_state_reload
) },
142 { "efer_reload", VCPU_STAT(efer_reload
) },
143 { "fpu_reload", VCPU_STAT(fpu_reload
) },
144 { "insn_emulation", VCPU_STAT(insn_emulation
) },
145 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
146 { "irq_injections", VCPU_STAT(irq_injections
) },
147 { "nmi_injections", VCPU_STAT(nmi_injections
) },
148 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
149 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
150 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
151 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
152 { "mmu_flooded", VM_STAT(mmu_flooded
) },
153 { "mmu_recycled", VM_STAT(mmu_recycled
) },
154 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
155 { "mmu_unsync", VM_STAT(mmu_unsync
) },
156 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
157 { "largepages", VM_STAT(lpages
) },
161 u64 __read_mostly host_xcr0
;
163 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
165 static int kvm_vcpu_reset(struct kvm_vcpu
*vcpu
);
167 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
170 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
171 vcpu
->arch
.apf
.gfns
[i
] = ~0;
174 static void kvm_on_user_return(struct user_return_notifier
*urn
)
177 struct kvm_shared_msrs
*locals
178 = container_of(urn
, struct kvm_shared_msrs
, urn
);
179 struct kvm_shared_msr_values
*values
;
181 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
182 values
= &locals
->values
[slot
];
183 if (values
->host
!= values
->curr
) {
184 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
185 values
->curr
= values
->host
;
188 locals
->registered
= false;
189 user_return_notifier_unregister(urn
);
192 static void shared_msr_update(unsigned slot
, u32 msr
)
194 struct kvm_shared_msrs
*smsr
;
197 smsr
= &__get_cpu_var(shared_msrs
);
198 /* only read, and nobody should modify it at this time,
199 * so don't need lock */
200 if (slot
>= shared_msrs_global
.nr
) {
201 printk(KERN_ERR
"kvm: invalid MSR slot!");
204 rdmsrl_safe(msr
, &value
);
205 smsr
->values
[slot
].host
= value
;
206 smsr
->values
[slot
].curr
= value
;
209 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
211 if (slot
>= shared_msrs_global
.nr
)
212 shared_msrs_global
.nr
= slot
+ 1;
213 shared_msrs_global
.msrs
[slot
] = msr
;
214 /* we need ensured the shared_msr_global have been updated */
217 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
219 static void kvm_shared_msr_cpu_online(void)
223 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
224 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
227 void kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
229 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
231 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
233 smsr
->values
[slot
].curr
= value
;
234 wrmsrl(shared_msrs_global
.msrs
[slot
], value
);
235 if (!smsr
->registered
) {
236 smsr
->urn
.on_user_return
= kvm_on_user_return
;
237 user_return_notifier_register(&smsr
->urn
);
238 smsr
->registered
= true;
241 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
243 static void drop_user_return_notifiers(void *ignore
)
245 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
247 if (smsr
->registered
)
248 kvm_on_user_return(&smsr
->urn
);
251 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
253 return vcpu
->arch
.apic_base
;
255 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
257 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
259 /* TODO: reserve bits check */
260 kvm_lapic_set_base(vcpu
, data
);
262 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
264 #define EXCPT_BENIGN 0
265 #define EXCPT_CONTRIBUTORY 1
268 static int exception_class(int vector
)
278 return EXCPT_CONTRIBUTORY
;
285 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
286 unsigned nr
, bool has_error
, u32 error_code
,
292 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
294 if (!vcpu
->arch
.exception
.pending
) {
296 vcpu
->arch
.exception
.pending
= true;
297 vcpu
->arch
.exception
.has_error_code
= has_error
;
298 vcpu
->arch
.exception
.nr
= nr
;
299 vcpu
->arch
.exception
.error_code
= error_code
;
300 vcpu
->arch
.exception
.reinject
= reinject
;
304 /* to check exception */
305 prev_nr
= vcpu
->arch
.exception
.nr
;
306 if (prev_nr
== DF_VECTOR
) {
307 /* triple fault -> shutdown */
308 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
311 class1
= exception_class(prev_nr
);
312 class2
= exception_class(nr
);
313 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
314 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
315 /* generate double fault per SDM Table 5-5 */
316 vcpu
->arch
.exception
.pending
= true;
317 vcpu
->arch
.exception
.has_error_code
= true;
318 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
319 vcpu
->arch
.exception
.error_code
= 0;
321 /* replace previous exception with a new one in a hope
322 that instruction re-execution will regenerate lost
327 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
329 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
331 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
333 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
335 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
337 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
339 void kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
342 kvm_inject_gp(vcpu
, 0);
344 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
346 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
348 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
350 ++vcpu
->stat
.pf_guest
;
351 vcpu
->arch
.cr2
= fault
->address
;
352 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
354 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
356 void kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
358 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
359 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
361 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
364 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
366 atomic_inc(&vcpu
->arch
.nmi_queued
);
367 kvm_make_request(KVM_REQ_NMI
, vcpu
);
369 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
371 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
373 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
375 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
377 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
379 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
381 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
384 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
385 * a #GP and return false.
387 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
389 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
391 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
394 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
397 * This function will be used to read from the physical memory of the currently
398 * running guest. The difference to kvm_read_guest_page is that this function
399 * can read from guest physical or from the guest's guest physical memory.
401 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
402 gfn_t ngfn
, void *data
, int offset
, int len
,
408 ngpa
= gfn_to_gpa(ngfn
);
409 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
);
410 if (real_gfn
== UNMAPPED_GVA
)
413 real_gfn
= gpa_to_gfn(real_gfn
);
415 return kvm_read_guest_page(vcpu
->kvm
, real_gfn
, data
, offset
, len
);
417 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
419 int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
420 void *data
, int offset
, int len
, u32 access
)
422 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
423 data
, offset
, len
, access
);
427 * Load the pae pdptrs. Return true is they are all valid.
429 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
431 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
432 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
435 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
437 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
438 offset
* sizeof(u64
), sizeof(pdpte
),
439 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
444 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
445 if (is_present_gpte(pdpte
[i
]) &&
446 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
453 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
454 __set_bit(VCPU_EXREG_PDPTR
,
455 (unsigned long *)&vcpu
->arch
.regs_avail
);
456 __set_bit(VCPU_EXREG_PDPTR
,
457 (unsigned long *)&vcpu
->arch
.regs_dirty
);
462 EXPORT_SYMBOL_GPL(load_pdptrs
);
464 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
466 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
472 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
475 if (!test_bit(VCPU_EXREG_PDPTR
,
476 (unsigned long *)&vcpu
->arch
.regs_avail
))
479 gfn
= (kvm_read_cr3(vcpu
) & ~31u) >> PAGE_SHIFT
;
480 offset
= (kvm_read_cr3(vcpu
) & ~31u) & (PAGE_SIZE
- 1);
481 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
482 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
485 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
491 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
493 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
494 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
|
495 X86_CR0_CD
| X86_CR0_NW
;
500 if (cr0
& 0xffffffff00000000UL
)
504 cr0
&= ~CR0_RESERVED_BITS
;
506 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
509 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
512 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
514 if ((vcpu
->arch
.efer
& EFER_LME
)) {
519 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
524 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
529 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
532 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
534 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
535 kvm_clear_async_pf_completion_queue(vcpu
);
536 kvm_async_pf_hash_reset(vcpu
);
539 if ((cr0
^ old_cr0
) & update_bits
)
540 kvm_mmu_reset_context(vcpu
);
543 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
545 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
547 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
549 EXPORT_SYMBOL_GPL(kvm_lmsw
);
551 int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
555 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
556 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
559 if (kvm_x86_ops
->get_cpl(vcpu
) != 0)
561 if (!(xcr0
& XSTATE_FP
))
563 if ((xcr0
& XSTATE_YMM
) && !(xcr0
& XSTATE_SSE
))
565 if (xcr0
& ~host_xcr0
)
567 vcpu
->arch
.xcr0
= xcr0
;
568 vcpu
->guest_xcr0_loaded
= 0;
572 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
574 if (__kvm_set_xcr(vcpu
, index
, xcr
)) {
575 kvm_inject_gp(vcpu
, 0);
580 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
582 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
584 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
585 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
|
586 X86_CR4_PAE
| X86_CR4_SMEP
;
587 if (cr4
& CR4_RESERVED_BITS
)
590 if (!guest_cpuid_has_xsave(vcpu
) && (cr4
& X86_CR4_OSXSAVE
))
593 if (!guest_cpuid_has_smep(vcpu
) && (cr4
& X86_CR4_SMEP
))
596 if (!guest_cpuid_has_fsgsbase(vcpu
) && (cr4
& X86_CR4_RDWRGSFS
))
599 if (is_long_mode(vcpu
)) {
600 if (!(cr4
& X86_CR4_PAE
))
602 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
603 && ((cr4
^ old_cr4
) & pdptr_bits
)
604 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
608 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
609 if (!guest_cpuid_has_pcid(vcpu
))
612 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
613 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
617 if (kvm_x86_ops
->set_cr4(vcpu
, cr4
))
620 if (((cr4
^ old_cr4
) & pdptr_bits
) ||
621 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
622 kvm_mmu_reset_context(vcpu
);
624 if ((cr4
^ old_cr4
) & X86_CR4_OSXSAVE
)
625 kvm_update_cpuid(vcpu
);
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 (kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
)) {
641 if (cr3
& CR3_PCID_ENABLED_RESERVED_BITS
)
644 if (cr3
& CR3_L_MODE_RESERVED_BITS
)
648 if (cr3
& CR3_PAE_RESERVED_BITS
)
650 if (is_paging(vcpu
) &&
651 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
655 * We don't check reserved bits in nonpae mode, because
656 * this isn't enforced, and VMware depends on this.
661 * Does the new cr3 value map to physical memory? (Note, we
662 * catch an invalid cr3 even in real-mode, because it would
663 * cause trouble later on when we turn on paging anyway.)
665 * A real CPU would silently accept an invalid cr3 and would
666 * attempt to use it - with largely undefined (and often hard
667 * to debug) behavior on the guest side.
669 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
671 vcpu
->arch
.cr3
= cr3
;
672 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
673 vcpu
->arch
.mmu
.new_cr3(vcpu
);
676 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
678 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
680 if (cr8
& CR8_RESERVED_BITS
)
682 if (irqchip_in_kernel(vcpu
->kvm
))
683 kvm_lapic_set_tpr(vcpu
, cr8
);
685 vcpu
->arch
.cr8
= cr8
;
688 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
690 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
692 if (irqchip_in_kernel(vcpu
->kvm
))
693 return kvm_lapic_get_cr8(vcpu
);
695 return vcpu
->arch
.cr8
;
697 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
699 static void kvm_update_dr7(struct kvm_vcpu
*vcpu
)
703 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
704 dr7
= vcpu
->arch
.guest_debug_dr7
;
706 dr7
= vcpu
->arch
.dr7
;
707 kvm_x86_ops
->set_dr7(vcpu
, dr7
);
708 vcpu
->arch
.switch_db_regs
= (dr7
& DR7_BP_EN_MASK
);
711 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
715 vcpu
->arch
.db
[dr
] = val
;
716 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
717 vcpu
->arch
.eff_db
[dr
] = val
;
720 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
724 if (val
& 0xffffffff00000000ULL
)
726 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | DR6_FIXED_1
;
729 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
733 if (val
& 0xffffffff00000000ULL
)
735 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
736 kvm_update_dr7(vcpu
);
743 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
747 res
= __kvm_set_dr(vcpu
, dr
, val
);
749 kvm_queue_exception(vcpu
, UD_VECTOR
);
751 kvm_inject_gp(vcpu
, 0);
755 EXPORT_SYMBOL_GPL(kvm_set_dr
);
757 static int _kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
761 *val
= vcpu
->arch
.db
[dr
];
764 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
768 *val
= vcpu
->arch
.dr6
;
771 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
775 *val
= vcpu
->arch
.dr7
;
782 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
784 if (_kvm_get_dr(vcpu
, dr
, val
)) {
785 kvm_queue_exception(vcpu
, UD_VECTOR
);
790 EXPORT_SYMBOL_GPL(kvm_get_dr
);
792 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
794 u32 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
798 err
= kvm_pmu_read_pmc(vcpu
, ecx
, &data
);
801 kvm_register_write(vcpu
, VCPU_REGS_RAX
, (u32
)data
);
802 kvm_register_write(vcpu
, VCPU_REGS_RDX
, data
>> 32);
805 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
808 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
809 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
811 * This list is modified at module load time to reflect the
812 * capabilities of the host cpu. This capabilities test skips MSRs that are
813 * kvm-specific. Those are put in the beginning of the list.
816 #define KVM_SAVE_MSRS_BEGIN 10
817 static u32 msrs_to_save
[] = {
818 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
819 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
820 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
821 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
823 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
826 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
828 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
831 static unsigned num_msrs_to_save
;
833 static const u32 emulated_msrs
[] = {
834 MSR_IA32_TSCDEADLINE
,
835 MSR_IA32_MISC_ENABLE
,
840 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
842 u64 old_efer
= vcpu
->arch
.efer
;
844 if (efer
& efer_reserved_bits
)
848 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
851 if (efer
& EFER_FFXSR
) {
852 struct kvm_cpuid_entry2
*feat
;
854 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
855 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
)))
859 if (efer
& EFER_SVME
) {
860 struct kvm_cpuid_entry2
*feat
;
862 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
863 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
)))
868 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
870 kvm_x86_ops
->set_efer(vcpu
, efer
);
872 vcpu
->arch
.mmu
.base_role
.nxe
= (efer
& EFER_NX
) && !tdp_enabled
;
874 /* Update reserved bits */
875 if ((efer
^ old_efer
) & EFER_NX
)
876 kvm_mmu_reset_context(vcpu
);
881 void kvm_enable_efer_bits(u64 mask
)
883 efer_reserved_bits
&= ~mask
;
885 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
889 * Writes msr value into into the appropriate "register".
890 * Returns 0 on success, non-0 otherwise.
891 * Assumes vcpu_load() was already called.
893 int kvm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
895 return kvm_x86_ops
->set_msr(vcpu
, msr
);
899 * Adapt set_msr() to msr_io()'s calling convention
901 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
907 msr
.host_initiated
= true;
908 return kvm_set_msr(vcpu
, &msr
);
912 struct pvclock_gtod_data
{
915 struct { /* extract of a clocksource struct */
923 /* open coded 'struct timespec' */
924 u64 monotonic_time_snsec
;
925 time_t monotonic_time_sec
;
928 static struct pvclock_gtod_data pvclock_gtod_data
;
930 static void update_pvclock_gtod(struct timekeeper
*tk
)
932 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
934 write_seqcount_begin(&vdata
->seq
);
936 /* copy pvclock gtod data */
937 vdata
->clock
.vclock_mode
= tk
->clock
->archdata
.vclock_mode
;
938 vdata
->clock
.cycle_last
= tk
->clock
->cycle_last
;
939 vdata
->clock
.mask
= tk
->clock
->mask
;
940 vdata
->clock
.mult
= tk
->mult
;
941 vdata
->clock
.shift
= tk
->shift
;
943 vdata
->monotonic_time_sec
= tk
->xtime_sec
944 + tk
->wall_to_monotonic
.tv_sec
;
945 vdata
->monotonic_time_snsec
= tk
->xtime_nsec
946 + (tk
->wall_to_monotonic
.tv_nsec
948 while (vdata
->monotonic_time_snsec
>=
949 (((u64
)NSEC_PER_SEC
) << tk
->shift
)) {
950 vdata
->monotonic_time_snsec
-=
951 ((u64
)NSEC_PER_SEC
) << tk
->shift
;
952 vdata
->monotonic_time_sec
++;
955 write_seqcount_end(&vdata
->seq
);
960 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
964 struct pvclock_wall_clock wc
;
965 struct timespec boot
;
970 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
975 ++version
; /* first time write, random junk */
979 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
982 * The guest calculates current wall clock time by adding
983 * system time (updated by kvm_guest_time_update below) to the
984 * wall clock specified here. guest system time equals host
985 * system time for us, thus we must fill in host boot time here.
989 if (kvm
->arch
.kvmclock_offset
) {
990 struct timespec ts
= ns_to_timespec(kvm
->arch
.kvmclock_offset
);
991 boot
= timespec_sub(boot
, ts
);
993 wc
.sec
= boot
.tv_sec
;
994 wc
.nsec
= boot
.tv_nsec
;
995 wc
.version
= version
;
997 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1000 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1003 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
1005 uint32_t quotient
, remainder
;
1007 /* Don't try to replace with do_div(), this one calculates
1008 * "(dividend << 32) / divisor" */
1010 : "=a" (quotient
), "=d" (remainder
)
1011 : "0" (0), "1" (dividend
), "r" (divisor
) );
1015 static void kvm_get_time_scale(uint32_t scaled_khz
, uint32_t base_khz
,
1016 s8
*pshift
, u32
*pmultiplier
)
1023 tps64
= base_khz
* 1000LL;
1024 scaled64
= scaled_khz
* 1000LL;
1025 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
1030 tps32
= (uint32_t)tps64
;
1031 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
1032 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
1040 *pmultiplier
= div_frac(scaled64
, tps32
);
1042 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1043 __func__
, base_khz
, scaled_khz
, shift
, *pmultiplier
);
1046 static inline u64
get_kernel_ns(void)
1050 WARN_ON(preemptible());
1052 monotonic_to_bootbased(&ts
);
1053 return timespec_to_ns(&ts
);
1056 #ifdef CONFIG_X86_64
1057 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
1060 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
1061 unsigned long max_tsc_khz
;
1063 static inline u64
nsec_to_cycles(struct kvm_vcpu
*vcpu
, u64 nsec
)
1065 return pvclock_scale_delta(nsec
, vcpu
->arch
.virtual_tsc_mult
,
1066 vcpu
->arch
.virtual_tsc_shift
);
1069 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1071 u64 v
= (u64
)khz
* (1000000 + ppm
);
1076 static void kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 this_tsc_khz
)
1078 u32 thresh_lo
, thresh_hi
;
1079 int use_scaling
= 0;
1081 /* Compute a scale to convert nanoseconds in TSC cycles */
1082 kvm_get_time_scale(this_tsc_khz
, NSEC_PER_SEC
/ 1000,
1083 &vcpu
->arch
.virtual_tsc_shift
,
1084 &vcpu
->arch
.virtual_tsc_mult
);
1085 vcpu
->arch
.virtual_tsc_khz
= this_tsc_khz
;
1088 * Compute the variation in TSC rate which is acceptable
1089 * within the range of tolerance and decide if the
1090 * rate being applied is within that bounds of the hardware
1091 * rate. If so, no scaling or compensation need be done.
1093 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1094 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1095 if (this_tsc_khz
< thresh_lo
|| this_tsc_khz
> thresh_hi
) {
1096 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz
, thresh_lo
, thresh_hi
);
1099 kvm_x86_ops
->set_tsc_khz(vcpu
, this_tsc_khz
, use_scaling
);
1102 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1104 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1105 vcpu
->arch
.virtual_tsc_mult
,
1106 vcpu
->arch
.virtual_tsc_shift
);
1107 tsc
+= vcpu
->arch
.this_tsc_write
;
1111 void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
1113 #ifdef CONFIG_X86_64
1115 bool do_request
= false;
1116 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1117 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1119 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1120 atomic_read(&vcpu
->kvm
->online_vcpus
));
1122 if (vcpus_matched
&& gtod
->clock
.vclock_mode
== VCLOCK_TSC
)
1123 if (!ka
->use_master_clock
)
1126 if (!vcpus_matched
&& ka
->use_master_clock
)
1130 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1132 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
1133 atomic_read(&vcpu
->kvm
->online_vcpus
),
1134 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
1138 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1140 struct kvm
*kvm
= vcpu
->kvm
;
1141 u64 offset
, ns
, elapsed
;
1142 unsigned long flags
;
1145 u64 data
= msr
->data
;
1147 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1148 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1149 ns
= get_kernel_ns();
1150 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1152 /* n.b - signed multiplication and division required */
1153 usdiff
= data
- kvm
->arch
.last_tsc_write
;
1154 #ifdef CONFIG_X86_64
1155 usdiff
= (usdiff
* 1000) / vcpu
->arch
.virtual_tsc_khz
;
1157 /* do_div() only does unsigned */
1158 asm("idivl %2; xor %%edx, %%edx"
1160 : "A"(usdiff
* 1000), "rm"(vcpu
->arch
.virtual_tsc_khz
));
1162 do_div(elapsed
, 1000);
1168 * Special case: TSC write with a small delta (1 second) of virtual
1169 * cycle time against real time is interpreted as an attempt to
1170 * synchronize the CPU.
1172 * For a reliable TSC, we can match TSC offsets, and for an unstable
1173 * TSC, we add elapsed time in this computation. We could let the
1174 * compensation code attempt to catch up if we fall behind, but
1175 * it's better to try to match offsets from the beginning.
1177 if (usdiff
< USEC_PER_SEC
&&
1178 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1179 if (!check_tsc_unstable()) {
1180 offset
= kvm
->arch
.cur_tsc_offset
;
1181 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1183 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1185 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1186 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1191 * We split periods of matched TSC writes into generations.
1192 * For each generation, we track the original measured
1193 * nanosecond time, offset, and write, so if TSCs are in
1194 * sync, we can match exact offset, and if not, we can match
1195 * exact software computation in compute_guest_tsc()
1197 * These values are tracked in kvm->arch.cur_xxx variables.
1199 kvm
->arch
.cur_tsc_generation
++;
1200 kvm
->arch
.cur_tsc_nsec
= ns
;
1201 kvm
->arch
.cur_tsc_write
= data
;
1202 kvm
->arch
.cur_tsc_offset
= offset
;
1204 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1205 kvm
->arch
.cur_tsc_generation
, data
);
1209 * We also track th most recent recorded KHZ, write and time to
1210 * allow the matching interval to be extended at each write.
1212 kvm
->arch
.last_tsc_nsec
= ns
;
1213 kvm
->arch
.last_tsc_write
= data
;
1214 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1216 /* Reset of TSC must disable overshoot protection below */
1217 vcpu
->arch
.hv_clock
.tsc_timestamp
= 0;
1218 vcpu
->arch
.last_guest_tsc
= data
;
1220 /* Keep track of which generation this VCPU has synchronized to */
1221 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1222 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1223 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1225 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1226 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1228 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1230 kvm
->arch
.nr_vcpus_matched_tsc
++;
1232 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
1234 kvm_track_tsc_matching(vcpu
);
1235 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1238 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1240 #ifdef CONFIG_X86_64
1242 static cycle_t
read_tsc(void)
1248 * Empirically, a fence (of type that depends on the CPU)
1249 * before rdtsc is enough to ensure that rdtsc is ordered
1250 * with respect to loads. The various CPU manuals are unclear
1251 * as to whether rdtsc can be reordered with later loads,
1252 * but no one has ever seen it happen.
1255 ret
= (cycle_t
)vget_cycles();
1257 last
= pvclock_gtod_data
.clock
.cycle_last
;
1259 if (likely(ret
>= last
))
1263 * GCC likes to generate cmov here, but this branch is extremely
1264 * predictable (it's just a funciton of time and the likely is
1265 * very likely) and there's a data dependence, so force GCC
1266 * to generate a branch instead. I don't barrier() because
1267 * we don't actually need a barrier, and if this function
1268 * ever gets inlined it will generate worse code.
1274 static inline u64
vgettsc(cycle_t
*cycle_now
)
1277 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1279 *cycle_now
= read_tsc();
1281 v
= (*cycle_now
- gtod
->clock
.cycle_last
) & gtod
->clock
.mask
;
1282 return v
* gtod
->clock
.mult
;
1285 static int do_monotonic(struct timespec
*ts
, cycle_t
*cycle_now
)
1290 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1294 seq
= read_seqcount_begin(>od
->seq
);
1295 mode
= gtod
->clock
.vclock_mode
;
1296 ts
->tv_sec
= gtod
->monotonic_time_sec
;
1297 ns
= gtod
->monotonic_time_snsec
;
1298 ns
+= vgettsc(cycle_now
);
1299 ns
>>= gtod
->clock
.shift
;
1300 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1301 timespec_add_ns(ts
, ns
);
1306 /* returns true if host is using tsc clocksource */
1307 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, cycle_t
*cycle_now
)
1311 /* checked again under seqlock below */
1312 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1315 if (do_monotonic(&ts
, cycle_now
) != VCLOCK_TSC
)
1318 monotonic_to_bootbased(&ts
);
1319 *kernel_ns
= timespec_to_ns(&ts
);
1327 * Assuming a stable TSC across physical CPUS, and a stable TSC
1328 * across virtual CPUs, the following condition is possible.
1329 * Each numbered line represents an event visible to both
1330 * CPUs at the next numbered event.
1332 * "timespecX" represents host monotonic time. "tscX" represents
1335 * VCPU0 on CPU0 | VCPU1 on CPU1
1337 * 1. read timespec0,tsc0
1338 * 2. | timespec1 = timespec0 + N
1340 * 3. transition to guest | transition to guest
1341 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1342 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1343 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1345 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1348 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1350 * - 0 < N - M => M < N
1352 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1353 * always the case (the difference between two distinct xtime instances
1354 * might be smaller then the difference between corresponding TSC reads,
1355 * when updating guest vcpus pvclock areas).
1357 * To avoid that problem, do not allow visibility of distinct
1358 * system_timestamp/tsc_timestamp values simultaneously: use a master
1359 * copy of host monotonic time values. Update that master copy
1362 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1366 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
1368 #ifdef CONFIG_X86_64
1369 struct kvm_arch
*ka
= &kvm
->arch
;
1371 bool host_tsc_clocksource
, vcpus_matched
;
1373 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1374 atomic_read(&kvm
->online_vcpus
));
1377 * If the host uses TSC clock, then passthrough TSC as stable
1380 host_tsc_clocksource
= kvm_get_time_and_clockread(
1381 &ka
->master_kernel_ns
,
1382 &ka
->master_cycle_now
);
1384 ka
->use_master_clock
= host_tsc_clocksource
& vcpus_matched
;
1386 if (ka
->use_master_clock
)
1387 atomic_set(&kvm_guest_has_master_clock
, 1);
1389 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
1390 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
1395 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1397 unsigned long flags
, this_tsc_khz
;
1398 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1399 struct kvm_arch
*ka
= &v
->kvm
->arch
;
1401 s64 kernel_ns
, max_kernel_ns
;
1402 u64 tsc_timestamp
, host_tsc
;
1403 struct pvclock_vcpu_time_info
*guest_hv_clock
;
1405 bool use_master_clock
;
1410 /* Keep irq disabled to prevent changes to the clock */
1411 local_irq_save(flags
);
1412 this_tsc_khz
= __get_cpu_var(cpu_tsc_khz
);
1413 if (unlikely(this_tsc_khz
== 0)) {
1414 local_irq_restore(flags
);
1415 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1420 * If the host uses TSC clock, then passthrough TSC as stable
1423 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1424 use_master_clock
= ka
->use_master_clock
;
1425 if (use_master_clock
) {
1426 host_tsc
= ka
->master_cycle_now
;
1427 kernel_ns
= ka
->master_kernel_ns
;
1429 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1430 if (!use_master_clock
) {
1431 host_tsc
= native_read_tsc();
1432 kernel_ns
= get_kernel_ns();
1435 tsc_timestamp
= kvm_x86_ops
->read_l1_tsc(v
, host_tsc
);
1438 * We may have to catch up the TSC to match elapsed wall clock
1439 * time for two reasons, even if kvmclock is used.
1440 * 1) CPU could have been running below the maximum TSC rate
1441 * 2) Broken TSC compensation resets the base at each VCPU
1442 * entry to avoid unknown leaps of TSC even when running
1443 * again on the same CPU. This may cause apparent elapsed
1444 * time to disappear, and the guest to stand still or run
1447 if (vcpu
->tsc_catchup
) {
1448 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1449 if (tsc
> tsc_timestamp
) {
1450 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1451 tsc_timestamp
= tsc
;
1455 local_irq_restore(flags
);
1457 if (!vcpu
->time_page
)
1461 * Time as measured by the TSC may go backwards when resetting the base
1462 * tsc_timestamp. The reason for this is that the TSC resolution is
1463 * higher than the resolution of the other clock scales. Thus, many
1464 * possible measurments of the TSC correspond to one measurement of any
1465 * other clock, and so a spread of values is possible. This is not a
1466 * problem for the computation of the nanosecond clock; with TSC rates
1467 * around 1GHZ, there can only be a few cycles which correspond to one
1468 * nanosecond value, and any path through this code will inevitably
1469 * take longer than that. However, with the kernel_ns value itself,
1470 * the precision may be much lower, down to HZ granularity. If the
1471 * first sampling of TSC against kernel_ns ends in the low part of the
1472 * range, and the second in the high end of the range, we can get:
1474 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1476 * As the sampling errors potentially range in the thousands of cycles,
1477 * it is possible such a time value has already been observed by the
1478 * guest. To protect against this, we must compute the system time as
1479 * observed by the guest and ensure the new system time is greater.
1482 if (vcpu
->hv_clock
.tsc_timestamp
) {
1483 max_kernel_ns
= vcpu
->last_guest_tsc
-
1484 vcpu
->hv_clock
.tsc_timestamp
;
1485 max_kernel_ns
= pvclock_scale_delta(max_kernel_ns
,
1486 vcpu
->hv_clock
.tsc_to_system_mul
,
1487 vcpu
->hv_clock
.tsc_shift
);
1488 max_kernel_ns
+= vcpu
->last_kernel_ns
;
1491 if (unlikely(vcpu
->hw_tsc_khz
!= this_tsc_khz
)) {
1492 kvm_get_time_scale(NSEC_PER_SEC
/ 1000, this_tsc_khz
,
1493 &vcpu
->hv_clock
.tsc_shift
,
1494 &vcpu
->hv_clock
.tsc_to_system_mul
);
1495 vcpu
->hw_tsc_khz
= this_tsc_khz
;
1498 /* with a master <monotonic time, tsc value> tuple,
1499 * pvclock clock reads always increase at the (scaled) rate
1500 * of guest TSC - no need to deal with sampling errors.
1502 if (!use_master_clock
) {
1503 if (max_kernel_ns
> kernel_ns
)
1504 kernel_ns
= max_kernel_ns
;
1506 /* With all the info we got, fill in the values */
1507 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1508 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1509 vcpu
->last_kernel_ns
= kernel_ns
;
1510 vcpu
->last_guest_tsc
= tsc_timestamp
;
1513 * The interface expects us to write an even number signaling that the
1514 * update is finished. Since the guest won't see the intermediate
1515 * state, we just increase by 2 at the end.
1517 vcpu
->hv_clock
.version
+= 2;
1519 shared_kaddr
= kmap_atomic(vcpu
->time_page
);
1521 guest_hv_clock
= shared_kaddr
+ vcpu
->time_offset
;
1523 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1524 pvclock_flags
= (guest_hv_clock
->flags
& PVCLOCK_GUEST_STOPPED
);
1526 if (vcpu
->pvclock_set_guest_stopped_request
) {
1527 pvclock_flags
|= PVCLOCK_GUEST_STOPPED
;
1528 vcpu
->pvclock_set_guest_stopped_request
= false;
1531 /* If the host uses TSC clocksource, then it is stable */
1532 if (use_master_clock
)
1533 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
1535 vcpu
->hv_clock
.flags
= pvclock_flags
;
1537 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
1538 sizeof(vcpu
->hv_clock
));
1540 kunmap_atomic(shared_kaddr
);
1542 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
1546 static bool msr_mtrr_valid(unsigned msr
)
1549 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
1550 case MSR_MTRRfix64K_00000
:
1551 case MSR_MTRRfix16K_80000
:
1552 case MSR_MTRRfix16K_A0000
:
1553 case MSR_MTRRfix4K_C0000
:
1554 case MSR_MTRRfix4K_C8000
:
1555 case MSR_MTRRfix4K_D0000
:
1556 case MSR_MTRRfix4K_D8000
:
1557 case MSR_MTRRfix4K_E0000
:
1558 case MSR_MTRRfix4K_E8000
:
1559 case MSR_MTRRfix4K_F0000
:
1560 case MSR_MTRRfix4K_F8000
:
1561 case MSR_MTRRdefType
:
1562 case MSR_IA32_CR_PAT
:
1570 static bool valid_pat_type(unsigned t
)
1572 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1575 static bool valid_mtrr_type(unsigned t
)
1577 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
1580 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1584 if (!msr_mtrr_valid(msr
))
1587 if (msr
== MSR_IA32_CR_PAT
) {
1588 for (i
= 0; i
< 8; i
++)
1589 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
1592 } else if (msr
== MSR_MTRRdefType
) {
1595 return valid_mtrr_type(data
& 0xff);
1596 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
1597 for (i
= 0; i
< 8 ; i
++)
1598 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
1603 /* variable MTRRs */
1604 return valid_mtrr_type(data
& 0xff);
1607 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1609 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1611 if (!mtrr_valid(vcpu
, msr
, data
))
1614 if (msr
== MSR_MTRRdefType
) {
1615 vcpu
->arch
.mtrr_state
.def_type
= data
;
1616 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
1617 } else if (msr
== MSR_MTRRfix64K_00000
)
1619 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1620 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
1621 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1622 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
1623 else if (msr
== MSR_IA32_CR_PAT
)
1624 vcpu
->arch
.pat
= data
;
1625 else { /* Variable MTRRs */
1626 int idx
, is_mtrr_mask
;
1629 idx
= (msr
- 0x200) / 2;
1630 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1633 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1636 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1640 kvm_mmu_reset_context(vcpu
);
1644 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1646 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1647 unsigned bank_num
= mcg_cap
& 0xff;
1650 case MSR_IA32_MCG_STATUS
:
1651 vcpu
->arch
.mcg_status
= data
;
1653 case MSR_IA32_MCG_CTL
:
1654 if (!(mcg_cap
& MCG_CTL_P
))
1656 if (data
!= 0 && data
!= ~(u64
)0)
1658 vcpu
->arch
.mcg_ctl
= data
;
1661 if (msr
>= MSR_IA32_MC0_CTL
&&
1662 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1663 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1664 /* only 0 or all 1s can be written to IA32_MCi_CTL
1665 * some Linux kernels though clear bit 10 in bank 4 to
1666 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1667 * this to avoid an uncatched #GP in the guest
1669 if ((offset
& 0x3) == 0 &&
1670 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
1672 vcpu
->arch
.mce_banks
[offset
] = data
;
1680 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
1682 struct kvm
*kvm
= vcpu
->kvm
;
1683 int lm
= is_long_mode(vcpu
);
1684 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
1685 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
1686 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
1687 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
1688 u32 page_num
= data
& ~PAGE_MASK
;
1689 u64 page_addr
= data
& PAGE_MASK
;
1694 if (page_num
>= blob_size
)
1697 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
1702 if (kvm_write_guest(kvm
, page_addr
, page
, PAGE_SIZE
))
1711 static bool kvm_hv_hypercall_enabled(struct kvm
*kvm
)
1713 return kvm
->arch
.hv_hypercall
& HV_X64_MSR_HYPERCALL_ENABLE
;
1716 static bool kvm_hv_msr_partition_wide(u32 msr
)
1720 case HV_X64_MSR_GUEST_OS_ID
:
1721 case HV_X64_MSR_HYPERCALL
:
1729 static int set_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1731 struct kvm
*kvm
= vcpu
->kvm
;
1734 case HV_X64_MSR_GUEST_OS_ID
:
1735 kvm
->arch
.hv_guest_os_id
= data
;
1736 /* setting guest os id to zero disables hypercall page */
1737 if (!kvm
->arch
.hv_guest_os_id
)
1738 kvm
->arch
.hv_hypercall
&= ~HV_X64_MSR_HYPERCALL_ENABLE
;
1740 case HV_X64_MSR_HYPERCALL
: {
1745 /* if guest os id is not set hypercall should remain disabled */
1746 if (!kvm
->arch
.hv_guest_os_id
)
1748 if (!(data
& HV_X64_MSR_HYPERCALL_ENABLE
)) {
1749 kvm
->arch
.hv_hypercall
= data
;
1752 gfn
= data
>> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT
;
1753 addr
= gfn_to_hva(kvm
, gfn
);
1754 if (kvm_is_error_hva(addr
))
1756 kvm_x86_ops
->patch_hypercall(vcpu
, instructions
);
1757 ((unsigned char *)instructions
)[3] = 0xc3; /* ret */
1758 if (__copy_to_user((void __user
*)addr
, instructions
, 4))
1760 kvm
->arch
.hv_hypercall
= data
;
1764 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1765 "data 0x%llx\n", msr
, data
);
1771 static int set_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1774 case HV_X64_MSR_APIC_ASSIST_PAGE
: {
1777 if (!(data
& HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE
)) {
1778 vcpu
->arch
.hv_vapic
= data
;
1781 addr
= gfn_to_hva(vcpu
->kvm
, data
>>
1782 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT
);
1783 if (kvm_is_error_hva(addr
))
1785 if (__clear_user((void __user
*)addr
, PAGE_SIZE
))
1787 vcpu
->arch
.hv_vapic
= data
;
1790 case HV_X64_MSR_EOI
:
1791 return kvm_hv_vapic_msr_write(vcpu
, APIC_EOI
, data
);
1792 case HV_X64_MSR_ICR
:
1793 return kvm_hv_vapic_msr_write(vcpu
, APIC_ICR
, data
);
1794 case HV_X64_MSR_TPR
:
1795 return kvm_hv_vapic_msr_write(vcpu
, APIC_TASKPRI
, data
);
1797 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1798 "data 0x%llx\n", msr
, data
);
1805 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
1807 gpa_t gpa
= data
& ~0x3f;
1809 /* Bits 2:5 are reserved, Should be zero */
1813 vcpu
->arch
.apf
.msr_val
= data
;
1815 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
1816 kvm_clear_async_pf_completion_queue(vcpu
);
1817 kvm_async_pf_hash_reset(vcpu
);
1821 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
))
1824 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
1825 kvm_async_pf_wakeup_all(vcpu
);
1829 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
1831 if (vcpu
->arch
.time_page
) {
1832 kvm_release_page_dirty(vcpu
->arch
.time_page
);
1833 vcpu
->arch
.time_page
= NULL
;
1837 static void accumulate_steal_time(struct kvm_vcpu
*vcpu
)
1841 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1844 delta
= current
->sched_info
.run_delay
- vcpu
->arch
.st
.last_steal
;
1845 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
1846 vcpu
->arch
.st
.accum_steal
= delta
;
1849 static void record_steal_time(struct kvm_vcpu
*vcpu
)
1851 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1854 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1855 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
1858 vcpu
->arch
.st
.steal
.steal
+= vcpu
->arch
.st
.accum_steal
;
1859 vcpu
->arch
.st
.steal
.version
+= 2;
1860 vcpu
->arch
.st
.accum_steal
= 0;
1862 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1863 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
1866 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
1869 u32 msr
= msr_info
->index
;
1870 u64 data
= msr_info
->data
;
1874 return set_efer(vcpu
, data
);
1876 data
&= ~(u64
)0x40; /* ignore flush filter disable */
1877 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
1878 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
1880 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
1885 case MSR_FAM10H_MMIO_CONF_BASE
:
1887 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
1892 case MSR_AMD64_NB_CFG
:
1894 case MSR_IA32_DEBUGCTLMSR
:
1896 /* We support the non-activated case already */
1898 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
1899 /* Values other than LBR and BTF are vendor-specific,
1900 thus reserved and should throw a #GP */
1903 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1906 case MSR_IA32_UCODE_REV
:
1907 case MSR_IA32_UCODE_WRITE
:
1908 case MSR_VM_HSAVE_PA
:
1909 case MSR_AMD64_PATCH_LOADER
:
1911 case 0x200 ... 0x2ff:
1912 return set_msr_mtrr(vcpu
, msr
, data
);
1913 case MSR_IA32_APICBASE
:
1914 kvm_set_apic_base(vcpu
, data
);
1916 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1917 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
1918 case MSR_IA32_TSCDEADLINE
:
1919 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
1921 case MSR_IA32_MISC_ENABLE
:
1922 vcpu
->arch
.ia32_misc_enable_msr
= data
;
1924 case MSR_KVM_WALL_CLOCK_NEW
:
1925 case MSR_KVM_WALL_CLOCK
:
1926 vcpu
->kvm
->arch
.wall_clock
= data
;
1927 kvm_write_wall_clock(vcpu
->kvm
, data
);
1929 case MSR_KVM_SYSTEM_TIME_NEW
:
1930 case MSR_KVM_SYSTEM_TIME
: {
1931 kvmclock_reset(vcpu
);
1933 vcpu
->arch
.time
= data
;
1934 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1936 /* we verify if the enable bit is set... */
1940 /* ...but clean it before doing the actual write */
1941 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
1943 vcpu
->arch
.time_page
=
1944 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
1946 if (is_error_page(vcpu
->arch
.time_page
))
1947 vcpu
->arch
.time_page
= NULL
;
1951 case MSR_KVM_ASYNC_PF_EN
:
1952 if (kvm_pv_enable_async_pf(vcpu
, data
))
1955 case MSR_KVM_STEAL_TIME
:
1957 if (unlikely(!sched_info_on()))
1960 if (data
& KVM_STEAL_RESERVED_MASK
)
1963 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1964 data
& KVM_STEAL_VALID_BITS
))
1967 vcpu
->arch
.st
.msr_val
= data
;
1969 if (!(data
& KVM_MSR_ENABLED
))
1972 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
1975 accumulate_steal_time(vcpu
);
1978 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
1981 case MSR_KVM_PV_EOI_EN
:
1982 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
1986 case MSR_IA32_MCG_CTL
:
1987 case MSR_IA32_MCG_STATUS
:
1988 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1989 return set_msr_mce(vcpu
, msr
, data
);
1991 /* Performance counters are not protected by a CPUID bit,
1992 * so we should check all of them in the generic path for the sake of
1993 * cross vendor migration.
1994 * Writing a zero into the event select MSRs disables them,
1995 * which we perfectly emulate ;-). Any other value should be at least
1996 * reported, some guests depend on them.
1998 case MSR_K7_EVNTSEL0
:
1999 case MSR_K7_EVNTSEL1
:
2000 case MSR_K7_EVNTSEL2
:
2001 case MSR_K7_EVNTSEL3
:
2003 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
2004 "0x%x data 0x%llx\n", msr
, data
);
2006 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2007 * so we ignore writes to make it happy.
2009 case MSR_K7_PERFCTR0
:
2010 case MSR_K7_PERFCTR1
:
2011 case MSR_K7_PERFCTR2
:
2012 case MSR_K7_PERFCTR3
:
2013 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
2014 "0x%x data 0x%llx\n", msr
, data
);
2016 case MSR_P6_PERFCTR0
:
2017 case MSR_P6_PERFCTR1
:
2019 case MSR_P6_EVNTSEL0
:
2020 case MSR_P6_EVNTSEL1
:
2021 if (kvm_pmu_msr(vcpu
, msr
))
2022 return kvm_pmu_set_msr(vcpu
, msr
, data
);
2024 if (pr
|| data
!= 0)
2025 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
2026 "0x%x data 0x%llx\n", msr
, data
);
2028 case MSR_K7_CLK_CTL
:
2030 * Ignore all writes to this no longer documented MSR.
2031 * Writes are only relevant for old K7 processors,
2032 * all pre-dating SVM, but a recommended workaround from
2033 * AMD for these chips. It is possible to specify the
2034 * affected processor models on the command line, hence
2035 * the need to ignore the workaround.
2038 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2039 if (kvm_hv_msr_partition_wide(msr
)) {
2041 mutex_lock(&vcpu
->kvm
->lock
);
2042 r
= set_msr_hyperv_pw(vcpu
, msr
, data
);
2043 mutex_unlock(&vcpu
->kvm
->lock
);
2046 return set_msr_hyperv(vcpu
, msr
, data
);
2048 case MSR_IA32_BBL_CR_CTL3
:
2049 /* Drop writes to this legacy MSR -- see rdmsr
2050 * counterpart for further detail.
2052 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n", msr
, data
);
2054 case MSR_AMD64_OSVW_ID_LENGTH
:
2055 if (!guest_cpuid_has_osvw(vcpu
))
2057 vcpu
->arch
.osvw
.length
= data
;
2059 case MSR_AMD64_OSVW_STATUS
:
2060 if (!guest_cpuid_has_osvw(vcpu
))
2062 vcpu
->arch
.osvw
.status
= data
;
2065 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
2066 return xen_hvm_config(vcpu
, data
);
2067 if (kvm_pmu_msr(vcpu
, msr
))
2068 return kvm_pmu_set_msr(vcpu
, msr
, data
);
2070 vcpu_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
2074 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
2081 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
2085 * Reads an msr value (of 'msr_index') into 'pdata'.
2086 * Returns 0 on success, non-0 otherwise.
2087 * Assumes vcpu_load() was already called.
2089 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
2091 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
2094 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2096 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
2098 if (!msr_mtrr_valid(msr
))
2101 if (msr
== MSR_MTRRdefType
)
2102 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
2103 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
2104 else if (msr
== MSR_MTRRfix64K_00000
)
2106 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
2107 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
2108 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
2109 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
2110 else if (msr
== MSR_IA32_CR_PAT
)
2111 *pdata
= vcpu
->arch
.pat
;
2112 else { /* Variable MTRRs */
2113 int idx
, is_mtrr_mask
;
2116 idx
= (msr
- 0x200) / 2;
2117 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
2120 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
2123 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
2130 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2133 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2134 unsigned bank_num
= mcg_cap
& 0xff;
2137 case MSR_IA32_P5_MC_ADDR
:
2138 case MSR_IA32_P5_MC_TYPE
:
2141 case MSR_IA32_MCG_CAP
:
2142 data
= vcpu
->arch
.mcg_cap
;
2144 case MSR_IA32_MCG_CTL
:
2145 if (!(mcg_cap
& MCG_CTL_P
))
2147 data
= vcpu
->arch
.mcg_ctl
;
2149 case MSR_IA32_MCG_STATUS
:
2150 data
= vcpu
->arch
.mcg_status
;
2153 if (msr
>= MSR_IA32_MC0_CTL
&&
2154 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
2155 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2156 data
= vcpu
->arch
.mce_banks
[offset
];
2165 static int get_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2168 struct kvm
*kvm
= vcpu
->kvm
;
2171 case HV_X64_MSR_GUEST_OS_ID
:
2172 data
= kvm
->arch
.hv_guest_os_id
;
2174 case HV_X64_MSR_HYPERCALL
:
2175 data
= kvm
->arch
.hv_hypercall
;
2178 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
2186 static int get_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2191 case HV_X64_MSR_VP_INDEX
: {
2194 kvm_for_each_vcpu(r
, v
, vcpu
->kvm
)
2199 case HV_X64_MSR_EOI
:
2200 return kvm_hv_vapic_msr_read(vcpu
, APIC_EOI
, pdata
);
2201 case HV_X64_MSR_ICR
:
2202 return kvm_hv_vapic_msr_read(vcpu
, APIC_ICR
, pdata
);
2203 case HV_X64_MSR_TPR
:
2204 return kvm_hv_vapic_msr_read(vcpu
, APIC_TASKPRI
, pdata
);
2205 case HV_X64_MSR_APIC_ASSIST_PAGE
:
2206 data
= vcpu
->arch
.hv_vapic
;
2209 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
2216 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2221 case MSR_IA32_PLATFORM_ID
:
2222 case MSR_IA32_EBL_CR_POWERON
:
2223 case MSR_IA32_DEBUGCTLMSR
:
2224 case MSR_IA32_LASTBRANCHFROMIP
:
2225 case MSR_IA32_LASTBRANCHTOIP
:
2226 case MSR_IA32_LASTINTFROMIP
:
2227 case MSR_IA32_LASTINTTOIP
:
2230 case MSR_VM_HSAVE_PA
:
2231 case MSR_K7_EVNTSEL0
:
2232 case MSR_K7_PERFCTR0
:
2233 case MSR_K8_INT_PENDING_MSG
:
2234 case MSR_AMD64_NB_CFG
:
2235 case MSR_FAM10H_MMIO_CONF_BASE
:
2238 case MSR_P6_PERFCTR0
:
2239 case MSR_P6_PERFCTR1
:
2240 case MSR_P6_EVNTSEL0
:
2241 case MSR_P6_EVNTSEL1
:
2242 if (kvm_pmu_msr(vcpu
, msr
))
2243 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
2246 case MSR_IA32_UCODE_REV
:
2247 data
= 0x100000000ULL
;
2250 data
= 0x500 | KVM_NR_VAR_MTRR
;
2252 case 0x200 ... 0x2ff:
2253 return get_msr_mtrr(vcpu
, msr
, pdata
);
2254 case 0xcd: /* fsb frequency */
2258 * MSR_EBC_FREQUENCY_ID
2259 * Conservative value valid for even the basic CPU models.
2260 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2261 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2262 * and 266MHz for model 3, or 4. Set Core Clock
2263 * Frequency to System Bus Frequency Ratio to 1 (bits
2264 * 31:24) even though these are only valid for CPU
2265 * models > 2, however guests may end up dividing or
2266 * multiplying by zero otherwise.
2268 case MSR_EBC_FREQUENCY_ID
:
2271 case MSR_IA32_APICBASE
:
2272 data
= kvm_get_apic_base(vcpu
);
2274 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2275 return kvm_x2apic_msr_read(vcpu
, msr
, pdata
);
2277 case MSR_IA32_TSCDEADLINE
:
2278 data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
2280 case MSR_IA32_MISC_ENABLE
:
2281 data
= vcpu
->arch
.ia32_misc_enable_msr
;
2283 case MSR_IA32_PERF_STATUS
:
2284 /* TSC increment by tick */
2286 /* CPU multiplier */
2287 data
|= (((uint64_t)4ULL) << 40);
2290 data
= vcpu
->arch
.efer
;
2292 case MSR_KVM_WALL_CLOCK
:
2293 case MSR_KVM_WALL_CLOCK_NEW
:
2294 data
= vcpu
->kvm
->arch
.wall_clock
;
2296 case MSR_KVM_SYSTEM_TIME
:
2297 case MSR_KVM_SYSTEM_TIME_NEW
:
2298 data
= vcpu
->arch
.time
;
2300 case MSR_KVM_ASYNC_PF_EN
:
2301 data
= vcpu
->arch
.apf
.msr_val
;
2303 case MSR_KVM_STEAL_TIME
:
2304 data
= vcpu
->arch
.st
.msr_val
;
2306 case MSR_KVM_PV_EOI_EN
:
2307 data
= vcpu
->arch
.pv_eoi
.msr_val
;
2309 case MSR_IA32_P5_MC_ADDR
:
2310 case MSR_IA32_P5_MC_TYPE
:
2311 case MSR_IA32_MCG_CAP
:
2312 case MSR_IA32_MCG_CTL
:
2313 case MSR_IA32_MCG_STATUS
:
2314 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
2315 return get_msr_mce(vcpu
, msr
, pdata
);
2316 case MSR_K7_CLK_CTL
:
2318 * Provide expected ramp-up count for K7. All other
2319 * are set to zero, indicating minimum divisors for
2322 * This prevents guest kernels on AMD host with CPU
2323 * type 6, model 8 and higher from exploding due to
2324 * the rdmsr failing.
2328 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2329 if (kvm_hv_msr_partition_wide(msr
)) {
2331 mutex_lock(&vcpu
->kvm
->lock
);
2332 r
= get_msr_hyperv_pw(vcpu
, msr
, pdata
);
2333 mutex_unlock(&vcpu
->kvm
->lock
);
2336 return get_msr_hyperv(vcpu
, msr
, pdata
);
2338 case MSR_IA32_BBL_CR_CTL3
:
2339 /* This legacy MSR exists but isn't fully documented in current
2340 * silicon. It is however accessed by winxp in very narrow
2341 * scenarios where it sets bit #19, itself documented as
2342 * a "reserved" bit. Best effort attempt to source coherent
2343 * read data here should the balance of the register be
2344 * interpreted by the guest:
2346 * L2 cache control register 3: 64GB range, 256KB size,
2347 * enabled, latency 0x1, configured
2351 case MSR_AMD64_OSVW_ID_LENGTH
:
2352 if (!guest_cpuid_has_osvw(vcpu
))
2354 data
= vcpu
->arch
.osvw
.length
;
2356 case MSR_AMD64_OSVW_STATUS
:
2357 if (!guest_cpuid_has_osvw(vcpu
))
2359 data
= vcpu
->arch
.osvw
.status
;
2362 if (kvm_pmu_msr(vcpu
, msr
))
2363 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
2365 vcpu_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
2368 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr
);
2376 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2379 * Read or write a bunch of msrs. All parameters are kernel addresses.
2381 * @return number of msrs set successfully.
2383 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2384 struct kvm_msr_entry
*entries
,
2385 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2386 unsigned index
, u64
*data
))
2390 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2391 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2392 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2394 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2400 * Read or write a bunch of msrs. Parameters are user addresses.
2402 * @return number of msrs set successfully.
2404 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2405 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2406 unsigned index
, u64
*data
),
2409 struct kvm_msrs msrs
;
2410 struct kvm_msr_entry
*entries
;
2415 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2419 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2422 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2423 entries
= memdup_user(user_msrs
->entries
, size
);
2424 if (IS_ERR(entries
)) {
2425 r
= PTR_ERR(entries
);
2429 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2434 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2445 int kvm_dev_ioctl_check_extension(long ext
)
2450 case KVM_CAP_IRQCHIP
:
2452 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2453 case KVM_CAP_SET_TSS_ADDR
:
2454 case KVM_CAP_EXT_CPUID
:
2455 case KVM_CAP_CLOCKSOURCE
:
2457 case KVM_CAP_NOP_IO_DELAY
:
2458 case KVM_CAP_MP_STATE
:
2459 case KVM_CAP_SYNC_MMU
:
2460 case KVM_CAP_USER_NMI
:
2461 case KVM_CAP_REINJECT_CONTROL
:
2462 case KVM_CAP_IRQ_INJECT_STATUS
:
2463 case KVM_CAP_ASSIGN_DEV_IRQ
:
2465 case KVM_CAP_IOEVENTFD
:
2467 case KVM_CAP_PIT_STATE2
:
2468 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2469 case KVM_CAP_XEN_HVM
:
2470 case KVM_CAP_ADJUST_CLOCK
:
2471 case KVM_CAP_VCPU_EVENTS
:
2472 case KVM_CAP_HYPERV
:
2473 case KVM_CAP_HYPERV_VAPIC
:
2474 case KVM_CAP_HYPERV_SPIN
:
2475 case KVM_CAP_PCI_SEGMENT
:
2476 case KVM_CAP_DEBUGREGS
:
2477 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2479 case KVM_CAP_ASYNC_PF
:
2480 case KVM_CAP_GET_TSC_KHZ
:
2481 case KVM_CAP_PCI_2_3
:
2482 case KVM_CAP_KVMCLOCK_CTRL
:
2483 case KVM_CAP_READONLY_MEM
:
2484 case KVM_CAP_IRQFD_RESAMPLE
:
2487 case KVM_CAP_COALESCED_MMIO
:
2488 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
2491 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2493 case KVM_CAP_NR_VCPUS
:
2494 r
= KVM_SOFT_MAX_VCPUS
;
2496 case KVM_CAP_MAX_VCPUS
:
2499 case KVM_CAP_NR_MEMSLOTS
:
2500 r
= KVM_MEMORY_SLOTS
;
2502 case KVM_CAP_PV_MMU
: /* obsolete */
2506 r
= iommu_present(&pci_bus_type
);
2509 r
= KVM_MAX_MCE_BANKS
;
2514 case KVM_CAP_TSC_CONTROL
:
2515 r
= kvm_has_tsc_control
;
2517 case KVM_CAP_TSC_DEADLINE_TIMER
:
2518 r
= boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER
);
2528 long kvm_arch_dev_ioctl(struct file
*filp
,
2529 unsigned int ioctl
, unsigned long arg
)
2531 void __user
*argp
= (void __user
*)arg
;
2535 case KVM_GET_MSR_INDEX_LIST
: {
2536 struct kvm_msr_list __user
*user_msr_list
= argp
;
2537 struct kvm_msr_list msr_list
;
2541 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2544 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2545 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2548 if (n
< msr_list
.nmsrs
)
2551 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2552 num_msrs_to_save
* sizeof(u32
)))
2554 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2556 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2561 case KVM_GET_SUPPORTED_CPUID
: {
2562 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2563 struct kvm_cpuid2 cpuid
;
2566 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2568 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
2569 cpuid_arg
->entries
);
2574 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2579 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2582 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
2584 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
2596 static void wbinvd_ipi(void *garbage
)
2601 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2603 return vcpu
->kvm
->arch
.iommu_domain
&&
2604 !(vcpu
->kvm
->arch
.iommu_flags
& KVM_IOMMU_CACHE_COHERENCY
);
2607 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2609 /* Address WBINVD may be executed by guest */
2610 if (need_emulate_wbinvd(vcpu
)) {
2611 if (kvm_x86_ops
->has_wbinvd_exit())
2612 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2613 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2614 smp_call_function_single(vcpu
->cpu
,
2615 wbinvd_ipi
, NULL
, 1);
2618 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2620 /* Apply any externally detected TSC adjustments (due to suspend) */
2621 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2622 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2623 vcpu
->arch
.tsc_offset_adjustment
= 0;
2624 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
2627 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2628 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2629 native_read_tsc() - vcpu
->arch
.last_host_tsc
;
2631 mark_tsc_unstable("KVM discovered backwards TSC");
2632 if (check_tsc_unstable()) {
2633 u64 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
,
2634 vcpu
->arch
.last_guest_tsc
);
2635 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
2636 vcpu
->arch
.tsc_catchup
= 1;
2639 * On a host with synchronized TSC, there is no need to update
2640 * kvmclock on vcpu->cpu migration
2642 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
2643 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2644 if (vcpu
->cpu
!= cpu
)
2645 kvm_migrate_timers(vcpu
);
2649 accumulate_steal_time(vcpu
);
2650 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2653 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2655 kvm_x86_ops
->vcpu_put(vcpu
);
2656 kvm_put_guest_fpu(vcpu
);
2657 vcpu
->arch
.last_host_tsc
= native_read_tsc();
2660 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
2661 struct kvm_lapic_state
*s
)
2663 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
2668 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
2669 struct kvm_lapic_state
*s
)
2671 kvm_apic_post_state_restore(vcpu
, s
);
2672 update_cr8_intercept(vcpu
);
2677 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2678 struct kvm_interrupt
*irq
)
2680 if (irq
->irq
< 0 || irq
->irq
>= KVM_NR_INTERRUPTS
)
2682 if (irqchip_in_kernel(vcpu
->kvm
))
2685 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
2686 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2691 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
2693 kvm_inject_nmi(vcpu
);
2698 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
2699 struct kvm_tpr_access_ctl
*tac
)
2703 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
2707 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
2711 unsigned bank_num
= mcg_cap
& 0xff, bank
;
2714 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
2716 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
2719 vcpu
->arch
.mcg_cap
= mcg_cap
;
2720 /* Init IA32_MCG_CTL to all 1s */
2721 if (mcg_cap
& MCG_CTL_P
)
2722 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
2723 /* Init IA32_MCi_CTL to all 1s */
2724 for (bank
= 0; bank
< bank_num
; bank
++)
2725 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
2730 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
2731 struct kvm_x86_mce
*mce
)
2733 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2734 unsigned bank_num
= mcg_cap
& 0xff;
2735 u64
*banks
= vcpu
->arch
.mce_banks
;
2737 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
2740 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2741 * reporting is disabled
2743 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
2744 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
2746 banks
+= 4 * mce
->bank
;
2748 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2749 * reporting is disabled for the bank
2751 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
2753 if (mce
->status
& MCI_STATUS_UC
) {
2754 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
2755 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
2756 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2759 if (banks
[1] & MCI_STATUS_VAL
)
2760 mce
->status
|= MCI_STATUS_OVER
;
2761 banks
[2] = mce
->addr
;
2762 banks
[3] = mce
->misc
;
2763 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
2764 banks
[1] = mce
->status
;
2765 kvm_queue_exception(vcpu
, MC_VECTOR
);
2766 } else if (!(banks
[1] & MCI_STATUS_VAL
)
2767 || !(banks
[1] & MCI_STATUS_UC
)) {
2768 if (banks
[1] & MCI_STATUS_VAL
)
2769 mce
->status
|= MCI_STATUS_OVER
;
2770 banks
[2] = mce
->addr
;
2771 banks
[3] = mce
->misc
;
2772 banks
[1] = mce
->status
;
2774 banks
[1] |= MCI_STATUS_OVER
;
2778 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
2779 struct kvm_vcpu_events
*events
)
2782 events
->exception
.injected
=
2783 vcpu
->arch
.exception
.pending
&&
2784 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
2785 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
2786 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
2787 events
->exception
.pad
= 0;
2788 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
2790 events
->interrupt
.injected
=
2791 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
2792 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
2793 events
->interrupt
.soft
= 0;
2794 events
->interrupt
.shadow
=
2795 kvm_x86_ops
->get_interrupt_shadow(vcpu
,
2796 KVM_X86_SHADOW_INT_MOV_SS
| KVM_X86_SHADOW_INT_STI
);
2798 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
2799 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
2800 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
2801 events
->nmi
.pad
= 0;
2803 events
->sipi_vector
= vcpu
->arch
.sipi_vector
;
2805 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
2806 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2807 | KVM_VCPUEVENT_VALID_SHADOW
);
2808 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
2811 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
2812 struct kvm_vcpu_events
*events
)
2814 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2815 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2816 | KVM_VCPUEVENT_VALID_SHADOW
))
2820 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
2821 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
2822 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
2823 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
2825 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
2826 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
2827 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
2828 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
2829 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
2830 events
->interrupt
.shadow
);
2832 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
2833 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
2834 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
2835 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
2837 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
)
2838 vcpu
->arch
.sipi_vector
= events
->sipi_vector
;
2840 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2845 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
2846 struct kvm_debugregs
*dbgregs
)
2848 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
2849 dbgregs
->dr6
= vcpu
->arch
.dr6
;
2850 dbgregs
->dr7
= vcpu
->arch
.dr7
;
2852 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
2855 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
2856 struct kvm_debugregs
*dbgregs
)
2861 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
2862 vcpu
->arch
.dr6
= dbgregs
->dr6
;
2863 vcpu
->arch
.dr7
= dbgregs
->dr7
;
2868 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
2869 struct kvm_xsave
*guest_xsave
)
2872 memcpy(guest_xsave
->region
,
2873 &vcpu
->arch
.guest_fpu
.state
->xsave
,
2876 memcpy(guest_xsave
->region
,
2877 &vcpu
->arch
.guest_fpu
.state
->fxsave
,
2878 sizeof(struct i387_fxsave_struct
));
2879 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
2884 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
2885 struct kvm_xsave
*guest_xsave
)
2888 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
2891 memcpy(&vcpu
->arch
.guest_fpu
.state
->xsave
,
2892 guest_xsave
->region
, xstate_size
);
2894 if (xstate_bv
& ~XSTATE_FPSSE
)
2896 memcpy(&vcpu
->arch
.guest_fpu
.state
->fxsave
,
2897 guest_xsave
->region
, sizeof(struct i387_fxsave_struct
));
2902 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
2903 struct kvm_xcrs
*guest_xcrs
)
2905 if (!cpu_has_xsave
) {
2906 guest_xcrs
->nr_xcrs
= 0;
2910 guest_xcrs
->nr_xcrs
= 1;
2911 guest_xcrs
->flags
= 0;
2912 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
2913 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
2916 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
2917 struct kvm_xcrs
*guest_xcrs
)
2924 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
2927 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
2928 /* Only support XCR0 currently */
2929 if (guest_xcrs
->xcrs
[0].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
2930 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
2931 guest_xcrs
->xcrs
[0].value
);
2940 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2941 * stopped by the hypervisor. This function will be called from the host only.
2942 * EINVAL is returned when the host attempts to set the flag for a guest that
2943 * does not support pv clocks.
2945 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
2947 if (!vcpu
->arch
.time_page
)
2949 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
2950 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2954 long kvm_arch_vcpu_ioctl(struct file
*filp
,
2955 unsigned int ioctl
, unsigned long arg
)
2957 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2958 void __user
*argp
= (void __user
*)arg
;
2961 struct kvm_lapic_state
*lapic
;
2962 struct kvm_xsave
*xsave
;
2963 struct kvm_xcrs
*xcrs
;
2969 case KVM_GET_LAPIC
: {
2971 if (!vcpu
->arch
.apic
)
2973 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
2978 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
2982 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
2987 case KVM_SET_LAPIC
: {
2988 if (!vcpu
->arch
.apic
)
2990 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
2991 if (IS_ERR(u
.lapic
))
2992 return PTR_ERR(u
.lapic
);
2994 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
2997 case KVM_INTERRUPT
: {
2998 struct kvm_interrupt irq
;
3001 if (copy_from_user(&irq
, argp
, sizeof irq
))
3003 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
3007 r
= kvm_vcpu_ioctl_nmi(vcpu
);
3010 case KVM_SET_CPUID
: {
3011 struct kvm_cpuid __user
*cpuid_arg
= argp
;
3012 struct kvm_cpuid cpuid
;
3015 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3017 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3020 case KVM_SET_CPUID2
: {
3021 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3022 struct kvm_cpuid2 cpuid
;
3025 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3027 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
3028 cpuid_arg
->entries
);
3031 case KVM_GET_CPUID2
: {
3032 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3033 struct kvm_cpuid2 cpuid
;
3036 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3038 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
3039 cpuid_arg
->entries
);
3043 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
3049 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
3052 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
3054 case KVM_TPR_ACCESS_REPORTING
: {
3055 struct kvm_tpr_access_ctl tac
;
3058 if (copy_from_user(&tac
, argp
, sizeof tac
))
3060 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
3064 if (copy_to_user(argp
, &tac
, sizeof tac
))
3069 case KVM_SET_VAPIC_ADDR
: {
3070 struct kvm_vapic_addr va
;
3073 if (!irqchip_in_kernel(vcpu
->kvm
))
3076 if (copy_from_user(&va
, argp
, sizeof va
))
3079 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
3082 case KVM_X86_SETUP_MCE
: {
3086 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
3088 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
3091 case KVM_X86_SET_MCE
: {
3092 struct kvm_x86_mce mce
;
3095 if (copy_from_user(&mce
, argp
, sizeof mce
))
3097 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
3100 case KVM_GET_VCPU_EVENTS
: {
3101 struct kvm_vcpu_events events
;
3103 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
3106 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
3111 case KVM_SET_VCPU_EVENTS
: {
3112 struct kvm_vcpu_events events
;
3115 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
3118 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
3121 case KVM_GET_DEBUGREGS
: {
3122 struct kvm_debugregs dbgregs
;
3124 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
3127 if (copy_to_user(argp
, &dbgregs
,
3128 sizeof(struct kvm_debugregs
)))
3133 case KVM_SET_DEBUGREGS
: {
3134 struct kvm_debugregs dbgregs
;
3137 if (copy_from_user(&dbgregs
, argp
,
3138 sizeof(struct kvm_debugregs
)))
3141 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3144 case KVM_GET_XSAVE
: {
3145 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3150 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3153 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3158 case KVM_SET_XSAVE
: {
3159 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
3160 if (IS_ERR(u
.xsave
))
3161 return PTR_ERR(u
.xsave
);
3163 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3166 case KVM_GET_XCRS
: {
3167 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3172 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3175 if (copy_to_user(argp
, u
.xcrs
,
3176 sizeof(struct kvm_xcrs
)))
3181 case KVM_SET_XCRS
: {
3182 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
3184 return PTR_ERR(u
.xcrs
);
3186 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3189 case KVM_SET_TSC_KHZ
: {
3193 user_tsc_khz
= (u32
)arg
;
3195 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3198 if (user_tsc_khz
== 0)
3199 user_tsc_khz
= tsc_khz
;
3201 kvm_set_tsc_khz(vcpu
, user_tsc_khz
);
3206 case KVM_GET_TSC_KHZ
: {
3207 r
= vcpu
->arch
.virtual_tsc_khz
;
3210 case KVM_KVMCLOCK_CTRL
: {
3211 r
= kvm_set_guest_paused(vcpu
);
3222 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
3224 return VM_FAULT_SIGBUS
;
3227 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3231 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3233 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3237 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3240 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3244 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3245 u32 kvm_nr_mmu_pages
)
3247 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3250 mutex_lock(&kvm
->slots_lock
);
3251 spin_lock(&kvm
->mmu_lock
);
3253 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3254 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3256 spin_unlock(&kvm
->mmu_lock
);
3257 mutex_unlock(&kvm
->slots_lock
);
3261 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3263 return kvm
->arch
.n_max_mmu_pages
;
3266 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3271 switch (chip
->chip_id
) {
3272 case KVM_IRQCHIP_PIC_MASTER
:
3273 memcpy(&chip
->chip
.pic
,
3274 &pic_irqchip(kvm
)->pics
[0],
3275 sizeof(struct kvm_pic_state
));
3277 case KVM_IRQCHIP_PIC_SLAVE
:
3278 memcpy(&chip
->chip
.pic
,
3279 &pic_irqchip(kvm
)->pics
[1],
3280 sizeof(struct kvm_pic_state
));
3282 case KVM_IRQCHIP_IOAPIC
:
3283 r
= kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3292 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3297 switch (chip
->chip_id
) {
3298 case KVM_IRQCHIP_PIC_MASTER
:
3299 spin_lock(&pic_irqchip(kvm
)->lock
);
3300 memcpy(&pic_irqchip(kvm
)->pics
[0],
3302 sizeof(struct kvm_pic_state
));
3303 spin_unlock(&pic_irqchip(kvm
)->lock
);
3305 case KVM_IRQCHIP_PIC_SLAVE
:
3306 spin_lock(&pic_irqchip(kvm
)->lock
);
3307 memcpy(&pic_irqchip(kvm
)->pics
[1],
3309 sizeof(struct kvm_pic_state
));
3310 spin_unlock(&pic_irqchip(kvm
)->lock
);
3312 case KVM_IRQCHIP_IOAPIC
:
3313 r
= kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3319 kvm_pic_update_irq(pic_irqchip(kvm
));
3323 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3327 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3328 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
3329 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3333 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3337 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3338 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
3339 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
, 0);
3340 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3344 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3348 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3349 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3350 sizeof(ps
->channels
));
3351 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3352 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3353 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3357 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3359 int r
= 0, start
= 0;
3360 u32 prev_legacy
, cur_legacy
;
3361 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3362 prev_legacy
= kvm
->arch
.vpit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3363 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3364 if (!prev_legacy
&& cur_legacy
)
3366 memcpy(&kvm
->arch
.vpit
->pit_state
.channels
, &ps
->channels
,
3367 sizeof(kvm
->arch
.vpit
->pit_state
.channels
));
3368 kvm
->arch
.vpit
->pit_state
.flags
= ps
->flags
;
3369 kvm_pit_load_count(kvm
, 0, kvm
->arch
.vpit
->pit_state
.channels
[0].count
, start
);
3370 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3374 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3375 struct kvm_reinject_control
*control
)
3377 if (!kvm
->arch
.vpit
)
3379 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3380 kvm
->arch
.vpit
->pit_state
.reinject
= control
->pit_reinject
;
3381 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3386 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3387 * @kvm: kvm instance
3388 * @log: slot id and address to which we copy the log
3390 * We need to keep it in mind that VCPU threads can write to the bitmap
3391 * concurrently. So, to avoid losing data, we keep the following order for
3394 * 1. Take a snapshot of the bit and clear it if needed.
3395 * 2. Write protect the corresponding page.
3396 * 3. Flush TLB's if needed.
3397 * 4. Copy the snapshot to the userspace.
3399 * Between 2 and 3, the guest may write to the page using the remaining TLB
3400 * entry. This is not a problem because the page will be reported dirty at
3401 * step 4 using the snapshot taken before and step 3 ensures that successive
3402 * writes will be logged for the next call.
3404 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
3407 struct kvm_memory_slot
*memslot
;
3409 unsigned long *dirty_bitmap
;
3410 unsigned long *dirty_bitmap_buffer
;
3411 bool is_dirty
= false;
3413 mutex_lock(&kvm
->slots_lock
);
3416 if (log
->slot
>= KVM_MEMORY_SLOTS
)
3419 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
3421 dirty_bitmap
= memslot
->dirty_bitmap
;
3426 n
= kvm_dirty_bitmap_bytes(memslot
);
3428 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
3429 memset(dirty_bitmap_buffer
, 0, n
);
3431 spin_lock(&kvm
->mmu_lock
);
3433 for (i
= 0; i
< n
/ sizeof(long); i
++) {
3437 if (!dirty_bitmap
[i
])
3442 mask
= xchg(&dirty_bitmap
[i
], 0);
3443 dirty_bitmap_buffer
[i
] = mask
;
3445 offset
= i
* BITS_PER_LONG
;
3446 kvm_mmu_write_protect_pt_masked(kvm
, memslot
, offset
, mask
);
3449 kvm_flush_remote_tlbs(kvm
);
3451 spin_unlock(&kvm
->mmu_lock
);
3454 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
3459 mutex_unlock(&kvm
->slots_lock
);
3463 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
)
3465 if (!irqchip_in_kernel(kvm
))
3468 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
3469 irq_event
->irq
, irq_event
->level
);
3473 long kvm_arch_vm_ioctl(struct file
*filp
,
3474 unsigned int ioctl
, unsigned long arg
)
3476 struct kvm
*kvm
= filp
->private_data
;
3477 void __user
*argp
= (void __user
*)arg
;
3480 * This union makes it completely explicit to gcc-3.x
3481 * that these two variables' stack usage should be
3482 * combined, not added together.
3485 struct kvm_pit_state ps
;
3486 struct kvm_pit_state2 ps2
;
3487 struct kvm_pit_config pit_config
;
3491 case KVM_SET_TSS_ADDR
:
3492 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
3494 case KVM_SET_IDENTITY_MAP_ADDR
: {
3498 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
3500 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
3503 case KVM_SET_NR_MMU_PAGES
:
3504 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
3506 case KVM_GET_NR_MMU_PAGES
:
3507 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
3509 case KVM_CREATE_IRQCHIP
: {
3510 struct kvm_pic
*vpic
;
3512 mutex_lock(&kvm
->lock
);
3515 goto create_irqchip_unlock
;
3517 if (atomic_read(&kvm
->online_vcpus
))
3518 goto create_irqchip_unlock
;
3520 vpic
= kvm_create_pic(kvm
);
3522 r
= kvm_ioapic_init(kvm
);
3524 mutex_lock(&kvm
->slots_lock
);
3525 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3527 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3529 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3531 mutex_unlock(&kvm
->slots_lock
);
3533 goto create_irqchip_unlock
;
3536 goto create_irqchip_unlock
;
3538 kvm
->arch
.vpic
= vpic
;
3540 r
= kvm_setup_default_irq_routing(kvm
);
3542 mutex_lock(&kvm
->slots_lock
);
3543 mutex_lock(&kvm
->irq_lock
);
3544 kvm_ioapic_destroy(kvm
);
3545 kvm_destroy_pic(kvm
);
3546 mutex_unlock(&kvm
->irq_lock
);
3547 mutex_unlock(&kvm
->slots_lock
);
3549 create_irqchip_unlock
:
3550 mutex_unlock(&kvm
->lock
);
3553 case KVM_CREATE_PIT
:
3554 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
3556 case KVM_CREATE_PIT2
:
3558 if (copy_from_user(&u
.pit_config
, argp
,
3559 sizeof(struct kvm_pit_config
)))
3562 mutex_lock(&kvm
->slots_lock
);
3565 goto create_pit_unlock
;
3567 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
3571 mutex_unlock(&kvm
->slots_lock
);
3573 case KVM_GET_IRQCHIP
: {
3574 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3575 struct kvm_irqchip
*chip
;
3577 chip
= memdup_user(argp
, sizeof(*chip
));
3584 if (!irqchip_in_kernel(kvm
))
3585 goto get_irqchip_out
;
3586 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
3588 goto get_irqchip_out
;
3590 if (copy_to_user(argp
, chip
, sizeof *chip
))
3591 goto get_irqchip_out
;
3597 case KVM_SET_IRQCHIP
: {
3598 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3599 struct kvm_irqchip
*chip
;
3601 chip
= memdup_user(argp
, sizeof(*chip
));
3608 if (!irqchip_in_kernel(kvm
))
3609 goto set_irqchip_out
;
3610 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
3612 goto set_irqchip_out
;
3620 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
3623 if (!kvm
->arch
.vpit
)
3625 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
3629 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
3636 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
3639 if (!kvm
->arch
.vpit
)
3641 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
3644 case KVM_GET_PIT2
: {
3646 if (!kvm
->arch
.vpit
)
3648 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
3652 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
3657 case KVM_SET_PIT2
: {
3659 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
3662 if (!kvm
->arch
.vpit
)
3664 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
3667 case KVM_REINJECT_CONTROL
: {
3668 struct kvm_reinject_control control
;
3670 if (copy_from_user(&control
, argp
, sizeof(control
)))
3672 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
3675 case KVM_XEN_HVM_CONFIG
: {
3677 if (copy_from_user(&kvm
->arch
.xen_hvm_config
, argp
,
3678 sizeof(struct kvm_xen_hvm_config
)))
3681 if (kvm
->arch
.xen_hvm_config
.flags
)
3686 case KVM_SET_CLOCK
: {
3687 struct kvm_clock_data user_ns
;
3692 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
3700 local_irq_disable();
3701 now_ns
= get_kernel_ns();
3702 delta
= user_ns
.clock
- now_ns
;
3704 kvm
->arch
.kvmclock_offset
= delta
;
3707 case KVM_GET_CLOCK
: {
3708 struct kvm_clock_data user_ns
;
3711 local_irq_disable();
3712 now_ns
= get_kernel_ns();
3713 user_ns
.clock
= kvm
->arch
.kvmclock_offset
+ now_ns
;
3716 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
3719 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
3732 static void kvm_init_msr_list(void)
3737 /* skip the first msrs in the list. KVM-specific */
3738 for (i
= j
= KVM_SAVE_MSRS_BEGIN
; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
3739 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
3742 msrs_to_save
[j
] = msrs_to_save
[i
];
3745 num_msrs_to_save
= j
;
3748 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
3756 if (!(vcpu
->arch
.apic
&&
3757 !kvm_iodevice_write(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3758 && kvm_io_bus_write(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3769 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
3776 if (!(vcpu
->arch
.apic
&&
3777 !kvm_iodevice_read(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3778 && kvm_io_bus_read(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3780 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, *(u64
*)v
);
3790 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3791 struct kvm_segment
*var
, int seg
)
3793 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3796 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3797 struct kvm_segment
*var
, int seg
)
3799 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3802 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
)
3805 struct x86_exception exception
;
3807 BUG_ON(!mmu_is_nested(vcpu
));
3809 /* NPT walks are always user-walks */
3810 access
|= PFERR_USER_MASK
;
3811 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, &exception
);
3816 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
3817 struct x86_exception
*exception
)
3819 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3820 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3823 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
3824 struct x86_exception
*exception
)
3826 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3827 access
|= PFERR_FETCH_MASK
;
3828 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3831 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
3832 struct x86_exception
*exception
)
3834 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3835 access
|= PFERR_WRITE_MASK
;
3836 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3839 /* uses this to access any guest's mapped memory without checking CPL */
3840 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
3841 struct x86_exception
*exception
)
3843 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
3846 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
3847 struct kvm_vcpu
*vcpu
, u32 access
,
3848 struct x86_exception
*exception
)
3851 int r
= X86EMUL_CONTINUE
;
3854 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
3856 unsigned offset
= addr
& (PAGE_SIZE
-1);
3857 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3860 if (gpa
== UNMAPPED_GVA
)
3861 return X86EMUL_PROPAGATE_FAULT
;
3862 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
3864 r
= X86EMUL_IO_NEEDED
;
3876 /* used for instruction fetching */
3877 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3878 gva_t addr
, void *val
, unsigned int bytes
,
3879 struct x86_exception
*exception
)
3881 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3882 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3884 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
,
3885 access
| PFERR_FETCH_MASK
,
3889 int kvm_read_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3890 gva_t addr
, void *val
, unsigned int bytes
,
3891 struct x86_exception
*exception
)
3893 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3894 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3896 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
3899 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
3901 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3902 gva_t addr
, void *val
, unsigned int bytes
,
3903 struct x86_exception
*exception
)
3905 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3906 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, 0, exception
);
3909 int kvm_write_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
3910 gva_t addr
, void *val
,
3912 struct x86_exception
*exception
)
3914 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3916 int r
= X86EMUL_CONTINUE
;
3919 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
3922 unsigned offset
= addr
& (PAGE_SIZE
-1);
3923 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3926 if (gpa
== UNMAPPED_GVA
)
3927 return X86EMUL_PROPAGATE_FAULT
;
3928 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
3930 r
= X86EMUL_IO_NEEDED
;
3941 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
3943 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
3944 gpa_t
*gpa
, struct x86_exception
*exception
,
3947 u32 access
= ((kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
3948 | (write
? PFERR_WRITE_MASK
: 0);
3950 if (vcpu_match_mmio_gva(vcpu
, gva
)
3951 && !permission_fault(vcpu
->arch
.walk_mmu
, vcpu
->arch
.access
, access
)) {
3952 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
3953 (gva
& (PAGE_SIZE
- 1));
3954 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
3958 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3960 if (*gpa
== UNMAPPED_GVA
)
3963 /* For APIC access vmexit */
3964 if ((*gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
3967 if (vcpu_match_mmio_gpa(vcpu
, *gpa
)) {
3968 trace_vcpu_match_mmio(gva
, *gpa
, write
, true);
3975 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3976 const void *val
, int bytes
)
3980 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
3983 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
3987 struct read_write_emulator_ops
{
3988 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
3990 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3991 void *val
, int bytes
);
3992 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3993 int bytes
, void *val
);
3994 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3995 void *val
, int bytes
);
3999 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
4001 if (vcpu
->mmio_read_completed
) {
4002 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
4003 vcpu
->mmio_fragments
[0].gpa
, *(u64
*)val
);
4004 vcpu
->mmio_read_completed
= 0;
4011 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4012 void *val
, int bytes
)
4014 return !kvm_read_guest(vcpu
->kvm
, gpa
, val
, bytes
);
4017 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4018 void *val
, int bytes
)
4020 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
4023 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
4025 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
4026 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
4029 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4030 void *val
, int bytes
)
4032 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
4033 return X86EMUL_IO_NEEDED
;
4036 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4037 void *val
, int bytes
)
4039 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
4041 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, frag
->len
);
4042 return X86EMUL_CONTINUE
;
4045 static const struct read_write_emulator_ops read_emultor
= {
4046 .read_write_prepare
= read_prepare
,
4047 .read_write_emulate
= read_emulate
,
4048 .read_write_mmio
= vcpu_mmio_read
,
4049 .read_write_exit_mmio
= read_exit_mmio
,
4052 static const struct read_write_emulator_ops write_emultor
= {
4053 .read_write_emulate
= write_emulate
,
4054 .read_write_mmio
= write_mmio
,
4055 .read_write_exit_mmio
= write_exit_mmio
,
4059 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
4061 struct x86_exception
*exception
,
4062 struct kvm_vcpu
*vcpu
,
4063 const struct read_write_emulator_ops
*ops
)
4067 bool write
= ops
->write
;
4068 struct kvm_mmio_fragment
*frag
;
4070 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
4073 return X86EMUL_PROPAGATE_FAULT
;
4075 /* For APIC access vmexit */
4079 if (ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
4080 return X86EMUL_CONTINUE
;
4084 * Is this MMIO handled locally?
4086 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
4087 if (handled
== bytes
)
4088 return X86EMUL_CONTINUE
;
4095 unsigned now
= min(bytes
, 8U);
4097 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
4106 return X86EMUL_CONTINUE
;
4109 int emulator_read_write(struct x86_emulate_ctxt
*ctxt
, unsigned long addr
,
4110 void *val
, unsigned int bytes
,
4111 struct x86_exception
*exception
,
4112 const struct read_write_emulator_ops
*ops
)
4114 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4118 if (ops
->read_write_prepare
&&
4119 ops
->read_write_prepare(vcpu
, val
, bytes
))
4120 return X86EMUL_CONTINUE
;
4122 vcpu
->mmio_nr_fragments
= 0;
4124 /* Crossing a page boundary? */
4125 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
4128 now
= -addr
& ~PAGE_MASK
;
4129 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
4132 if (rc
!= X86EMUL_CONTINUE
)
4139 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
4141 if (rc
!= X86EMUL_CONTINUE
)
4144 if (!vcpu
->mmio_nr_fragments
)
4147 gpa
= vcpu
->mmio_fragments
[0].gpa
;
4149 vcpu
->mmio_needed
= 1;
4150 vcpu
->mmio_cur_fragment
= 0;
4152 vcpu
->run
->mmio
.len
= vcpu
->mmio_fragments
[0].len
;
4153 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
4154 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
4155 vcpu
->run
->mmio
.phys_addr
= gpa
;
4157 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
4160 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
4164 struct x86_exception
*exception
)
4166 return emulator_read_write(ctxt
, addr
, val
, bytes
,
4167 exception
, &read_emultor
);
4170 int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
4174 struct x86_exception
*exception
)
4176 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
4177 exception
, &write_emultor
);
4180 #define CMPXCHG_TYPE(t, ptr, old, new) \
4181 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4183 #ifdef CONFIG_X86_64
4184 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4186 # define CMPXCHG64(ptr, old, new) \
4187 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4190 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
4195 struct x86_exception
*exception
)
4197 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4203 /* guests cmpxchg8b have to be emulated atomically */
4204 if (bytes
> 8 || (bytes
& (bytes
- 1)))
4207 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
4209 if (gpa
== UNMAPPED_GVA
||
4210 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4213 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
4216 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
4217 if (is_error_page(page
))
4220 kaddr
= kmap_atomic(page
);
4221 kaddr
+= offset_in_page(gpa
);
4224 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
4227 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
4230 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
4233 exchanged
= CMPXCHG64(kaddr
, old
, new);
4238 kunmap_atomic(kaddr
);
4239 kvm_release_page_dirty(page
);
4242 return X86EMUL_CMPXCHG_FAILED
;
4244 kvm_mmu_pte_write(vcpu
, gpa
, new, bytes
);
4246 return X86EMUL_CONTINUE
;
4249 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
4251 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
4254 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
4256 /* TODO: String I/O for in kernel device */
4259 if (vcpu
->arch
.pio
.in
)
4260 r
= kvm_io_bus_read(vcpu
->kvm
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
4261 vcpu
->arch
.pio
.size
, pd
);
4263 r
= kvm_io_bus_write(vcpu
->kvm
, KVM_PIO_BUS
,
4264 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
4269 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
4270 unsigned short port
, void *val
,
4271 unsigned int count
, bool in
)
4273 trace_kvm_pio(!in
, port
, size
, count
);
4275 vcpu
->arch
.pio
.port
= port
;
4276 vcpu
->arch
.pio
.in
= in
;
4277 vcpu
->arch
.pio
.count
= count
;
4278 vcpu
->arch
.pio
.size
= size
;
4280 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
4281 vcpu
->arch
.pio
.count
= 0;
4285 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
4286 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
4287 vcpu
->run
->io
.size
= size
;
4288 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
4289 vcpu
->run
->io
.count
= count
;
4290 vcpu
->run
->io
.port
= port
;
4295 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
4296 int size
, unsigned short port
, void *val
,
4299 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4302 if (vcpu
->arch
.pio
.count
)
4305 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
4308 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
4309 vcpu
->arch
.pio
.count
= 0;
4316 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
4317 int size
, unsigned short port
,
4318 const void *val
, unsigned int count
)
4320 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4322 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
4323 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
4326 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
4328 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
4331 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
4333 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
4336 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
4338 if (!need_emulate_wbinvd(vcpu
))
4339 return X86EMUL_CONTINUE
;
4341 if (kvm_x86_ops
->has_wbinvd_exit()) {
4342 int cpu
= get_cpu();
4344 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
4345 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
4346 wbinvd_ipi
, NULL
, 1);
4348 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
4351 return X86EMUL_CONTINUE
;
4353 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
4355 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
4357 kvm_emulate_wbinvd(emul_to_vcpu(ctxt
));
4360 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
4362 return _kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
4365 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
4368 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
4371 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
4373 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
4376 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
4378 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4379 unsigned long value
;
4383 value
= kvm_read_cr0(vcpu
);
4386 value
= vcpu
->arch
.cr2
;
4389 value
= kvm_read_cr3(vcpu
);
4392 value
= kvm_read_cr4(vcpu
);
4395 value
= kvm_get_cr8(vcpu
);
4398 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4405 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
4407 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4412 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
4415 vcpu
->arch
.cr2
= val
;
4418 res
= kvm_set_cr3(vcpu
, val
);
4421 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
4424 res
= kvm_set_cr8(vcpu
, val
);
4427 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4434 static void emulator_set_rflags(struct x86_emulate_ctxt
*ctxt
, ulong val
)
4436 kvm_set_rflags(emul_to_vcpu(ctxt
), val
);
4439 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
4441 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
4444 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4446 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
4449 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4451 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
4454 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4456 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
4459 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4461 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
4464 static unsigned long emulator_get_cached_segment_base(
4465 struct x86_emulate_ctxt
*ctxt
, int seg
)
4467 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
4470 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
4471 struct desc_struct
*desc
, u32
*base3
,
4474 struct kvm_segment var
;
4476 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
4477 *selector
= var
.selector
;
4484 set_desc_limit(desc
, var
.limit
);
4485 set_desc_base(desc
, (unsigned long)var
.base
);
4486 #ifdef CONFIG_X86_64
4488 *base3
= var
.base
>> 32;
4490 desc
->type
= var
.type
;
4492 desc
->dpl
= var
.dpl
;
4493 desc
->p
= var
.present
;
4494 desc
->avl
= var
.avl
;
4502 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
4503 struct desc_struct
*desc
, u32 base3
,
4506 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4507 struct kvm_segment var
;
4509 var
.selector
= selector
;
4510 var
.base
= get_desc_base(desc
);
4511 #ifdef CONFIG_X86_64
4512 var
.base
|= ((u64
)base3
) << 32;
4514 var
.limit
= get_desc_limit(desc
);
4516 var
.limit
= (var
.limit
<< 12) | 0xfff;
4517 var
.type
= desc
->type
;
4518 var
.present
= desc
->p
;
4519 var
.dpl
= desc
->dpl
;
4524 var
.avl
= desc
->avl
;
4525 var
.present
= desc
->p
;
4526 var
.unusable
= !var
.present
;
4529 kvm_set_segment(vcpu
, &var
, seg
);
4533 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
4534 u32 msr_index
, u64
*pdata
)
4536 return kvm_get_msr(emul_to_vcpu(ctxt
), msr_index
, pdata
);
4539 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
4540 u32 msr_index
, u64 data
)
4542 struct msr_data msr
;
4545 msr
.index
= msr_index
;
4546 msr
.host_initiated
= false;
4547 return kvm_set_msr(emul_to_vcpu(ctxt
), &msr
);
4550 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
4551 u32 pmc
, u64
*pdata
)
4553 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
4556 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
4558 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
4561 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
4564 kvm_load_guest_fpu(emul_to_vcpu(ctxt
));
4566 * CR0.TS may reference the host fpu state, not the guest fpu state,
4567 * so it may be clear at this point.
4572 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
4577 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
4578 struct x86_instruction_info
*info
,
4579 enum x86_intercept_stage stage
)
4581 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
4584 static void emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
4585 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
)
4587 kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
);
4590 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
4592 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
4595 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
4597 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
4600 static const struct x86_emulate_ops emulate_ops
= {
4601 .read_gpr
= emulator_read_gpr
,
4602 .write_gpr
= emulator_write_gpr
,
4603 .read_std
= kvm_read_guest_virt_system
,
4604 .write_std
= kvm_write_guest_virt_system
,
4605 .fetch
= kvm_fetch_guest_virt
,
4606 .read_emulated
= emulator_read_emulated
,
4607 .write_emulated
= emulator_write_emulated
,
4608 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
4609 .invlpg
= emulator_invlpg
,
4610 .pio_in_emulated
= emulator_pio_in_emulated
,
4611 .pio_out_emulated
= emulator_pio_out_emulated
,
4612 .get_segment
= emulator_get_segment
,
4613 .set_segment
= emulator_set_segment
,
4614 .get_cached_segment_base
= emulator_get_cached_segment_base
,
4615 .get_gdt
= emulator_get_gdt
,
4616 .get_idt
= emulator_get_idt
,
4617 .set_gdt
= emulator_set_gdt
,
4618 .set_idt
= emulator_set_idt
,
4619 .get_cr
= emulator_get_cr
,
4620 .set_cr
= emulator_set_cr
,
4621 .set_rflags
= emulator_set_rflags
,
4622 .cpl
= emulator_get_cpl
,
4623 .get_dr
= emulator_get_dr
,
4624 .set_dr
= emulator_set_dr
,
4625 .set_msr
= emulator_set_msr
,
4626 .get_msr
= emulator_get_msr
,
4627 .read_pmc
= emulator_read_pmc
,
4628 .halt
= emulator_halt
,
4629 .wbinvd
= emulator_wbinvd
,
4630 .fix_hypercall
= emulator_fix_hypercall
,
4631 .get_fpu
= emulator_get_fpu
,
4632 .put_fpu
= emulator_put_fpu
,
4633 .intercept
= emulator_intercept
,
4634 .get_cpuid
= emulator_get_cpuid
,
4637 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
4639 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
, mask
);
4641 * an sti; sti; sequence only disable interrupts for the first
4642 * instruction. So, if the last instruction, be it emulated or
4643 * not, left the system with the INT_STI flag enabled, it
4644 * means that the last instruction is an sti. We should not
4645 * leave the flag on in this case. The same goes for mov ss
4647 if (!(int_shadow
& mask
))
4648 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
4651 static void inject_emulated_exception(struct kvm_vcpu
*vcpu
)
4653 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4654 if (ctxt
->exception
.vector
== PF_VECTOR
)
4655 kvm_propagate_fault(vcpu
, &ctxt
->exception
);
4656 else if (ctxt
->exception
.error_code_valid
)
4657 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
4658 ctxt
->exception
.error_code
);
4660 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
4663 static void init_decode_cache(struct x86_emulate_ctxt
*ctxt
)
4665 memset(&ctxt
->twobyte
, 0,
4666 (void *)&ctxt
->_regs
- (void *)&ctxt
->twobyte
);
4668 ctxt
->fetch
.start
= 0;
4669 ctxt
->fetch
.end
= 0;
4670 ctxt
->io_read
.pos
= 0;
4671 ctxt
->io_read
.end
= 0;
4672 ctxt
->mem_read
.pos
= 0;
4673 ctxt
->mem_read
.end
= 0;
4676 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
4678 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4681 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
4683 ctxt
->eflags
= kvm_get_rflags(vcpu
);
4684 ctxt
->eip
= kvm_rip_read(vcpu
);
4685 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
4686 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
4687 cs_l
? X86EMUL_MODE_PROT64
:
4688 cs_db
? X86EMUL_MODE_PROT32
:
4689 X86EMUL_MODE_PROT16
;
4690 ctxt
->guest_mode
= is_guest_mode(vcpu
);
4692 init_decode_cache(ctxt
);
4693 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4696 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
4698 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4701 init_emulate_ctxt(vcpu
);
4705 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
4706 ret
= emulate_int_real(ctxt
, irq
);
4708 if (ret
!= X86EMUL_CONTINUE
)
4709 return EMULATE_FAIL
;
4711 ctxt
->eip
= ctxt
->_eip
;
4712 kvm_rip_write(vcpu
, ctxt
->eip
);
4713 kvm_set_rflags(vcpu
, ctxt
->eflags
);
4715 if (irq
== NMI_VECTOR
)
4716 vcpu
->arch
.nmi_pending
= 0;
4718 vcpu
->arch
.interrupt
.pending
= false;
4720 return EMULATE_DONE
;
4722 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
4724 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
4726 int r
= EMULATE_DONE
;
4728 ++vcpu
->stat
.insn_emulation_fail
;
4729 trace_kvm_emulate_insn_failed(vcpu
);
4730 if (!is_guest_mode(vcpu
)) {
4731 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
4732 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
4733 vcpu
->run
->internal
.ndata
= 0;
4736 kvm_queue_exception(vcpu
, UD_VECTOR
);
4741 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t gva
)
4750 * if emulation was due to access to shadowed page table
4751 * and it failed try to unshadow page and re-enter the
4752 * guest to let CPU execute the instruction.
4754 if (kvm_mmu_unprotect_page_virt(vcpu
, gva
))
4757 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, gva
, NULL
);
4759 if (gpa
== UNMAPPED_GVA
)
4760 return true; /* let cpu generate fault */
4763 * Do not retry the unhandleable instruction if it faults on the
4764 * readonly host memory, otherwise it will goto a infinite loop:
4765 * retry instruction -> write #PF -> emulation fail -> retry
4766 * instruction -> ...
4768 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
4769 if (!is_error_noslot_pfn(pfn
)) {
4770 kvm_release_pfn_clean(pfn
);
4777 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
4778 unsigned long cr2
, int emulation_type
)
4780 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4781 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
4783 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
4784 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
4787 * If the emulation is caused by #PF and it is non-page_table
4788 * writing instruction, it means the VM-EXIT is caused by shadow
4789 * page protected, we can zap the shadow page and retry this
4790 * instruction directly.
4792 * Note: if the guest uses a non-page-table modifying instruction
4793 * on the PDE that points to the instruction, then we will unmap
4794 * the instruction and go to an infinite loop. So, we cache the
4795 * last retried eip and the last fault address, if we meet the eip
4796 * and the address again, we can break out of the potential infinite
4799 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
4801 if (!(emulation_type
& EMULTYPE_RETRY
))
4804 if (x86_page_table_writing_insn(ctxt
))
4807 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
4810 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
4811 vcpu
->arch
.last_retry_addr
= cr2
;
4813 if (!vcpu
->arch
.mmu
.direct_map
)
4814 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
4816 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
4821 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
4822 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
4824 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
4831 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4832 bool writeback
= true;
4834 kvm_clear_exception_queue(vcpu
);
4836 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
4837 init_emulate_ctxt(vcpu
);
4838 ctxt
->interruptibility
= 0;
4839 ctxt
->have_exception
= false;
4840 ctxt
->perm_ok
= false;
4842 ctxt
->only_vendor_specific_insn
4843 = emulation_type
& EMULTYPE_TRAP_UD
;
4845 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
4847 trace_kvm_emulate_insn_start(vcpu
);
4848 ++vcpu
->stat
.insn_emulation
;
4849 if (r
!= EMULATION_OK
) {
4850 if (emulation_type
& EMULTYPE_TRAP_UD
)
4851 return EMULATE_FAIL
;
4852 if (reexecute_instruction(vcpu
, cr2
))
4853 return EMULATE_DONE
;
4854 if (emulation_type
& EMULTYPE_SKIP
)
4855 return EMULATE_FAIL
;
4856 return handle_emulation_failure(vcpu
);
4860 if (emulation_type
& EMULTYPE_SKIP
) {
4861 kvm_rip_write(vcpu
, ctxt
->_eip
);
4862 return EMULATE_DONE
;
4865 if (retry_instruction(ctxt
, cr2
, emulation_type
))
4866 return EMULATE_DONE
;
4868 /* this is needed for vmware backdoor interface to work since it
4869 changes registers values during IO operation */
4870 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
4871 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4872 emulator_invalidate_register_cache(ctxt
);
4876 r
= x86_emulate_insn(ctxt
);
4878 if (r
== EMULATION_INTERCEPTED
)
4879 return EMULATE_DONE
;
4881 if (r
== EMULATION_FAILED
) {
4882 if (reexecute_instruction(vcpu
, cr2
))
4883 return EMULATE_DONE
;
4885 return handle_emulation_failure(vcpu
);
4888 if (ctxt
->have_exception
) {
4889 inject_emulated_exception(vcpu
);
4891 } else if (vcpu
->arch
.pio
.count
) {
4892 if (!vcpu
->arch
.pio
.in
)
4893 vcpu
->arch
.pio
.count
= 0;
4896 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
4898 r
= EMULATE_DO_MMIO
;
4899 } else if (vcpu
->mmio_needed
) {
4900 if (!vcpu
->mmio_is_write
)
4902 r
= EMULATE_DO_MMIO
;
4903 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
4904 } else if (r
== EMULATION_RESTART
)
4910 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
4911 kvm_set_rflags(vcpu
, ctxt
->eflags
);
4912 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
4913 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
4914 kvm_rip_write(vcpu
, ctxt
->eip
);
4916 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
4920 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
4922 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
4924 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4925 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
4926 size
, port
, &val
, 1);
4927 /* do not return to emulator after return from userspace */
4928 vcpu
->arch
.pio
.count
= 0;
4931 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
4933 static void tsc_bad(void *info
)
4935 __this_cpu_write(cpu_tsc_khz
, 0);
4938 static void tsc_khz_changed(void *data
)
4940 struct cpufreq_freqs
*freq
= data
;
4941 unsigned long khz
= 0;
4945 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
4946 khz
= cpufreq_quick_get(raw_smp_processor_id());
4949 __this_cpu_write(cpu_tsc_khz
, khz
);
4952 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
4955 struct cpufreq_freqs
*freq
= data
;
4957 struct kvm_vcpu
*vcpu
;
4958 int i
, send_ipi
= 0;
4961 * We allow guests to temporarily run on slowing clocks,
4962 * provided we notify them after, or to run on accelerating
4963 * clocks, provided we notify them before. Thus time never
4966 * However, we have a problem. We can't atomically update
4967 * the frequency of a given CPU from this function; it is
4968 * merely a notifier, which can be called from any CPU.
4969 * Changing the TSC frequency at arbitrary points in time
4970 * requires a recomputation of local variables related to
4971 * the TSC for each VCPU. We must flag these local variables
4972 * to be updated and be sure the update takes place with the
4973 * new frequency before any guests proceed.
4975 * Unfortunately, the combination of hotplug CPU and frequency
4976 * change creates an intractable locking scenario; the order
4977 * of when these callouts happen is undefined with respect to
4978 * CPU hotplug, and they can race with each other. As such,
4979 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4980 * undefined; you can actually have a CPU frequency change take
4981 * place in between the computation of X and the setting of the
4982 * variable. To protect against this problem, all updates of
4983 * the per_cpu tsc_khz variable are done in an interrupt
4984 * protected IPI, and all callers wishing to update the value
4985 * must wait for a synchronous IPI to complete (which is trivial
4986 * if the caller is on the CPU already). This establishes the
4987 * necessary total order on variable updates.
4989 * Note that because a guest time update may take place
4990 * anytime after the setting of the VCPU's request bit, the
4991 * correct TSC value must be set before the request. However,
4992 * to ensure the update actually makes it to any guest which
4993 * starts running in hardware virtualization between the set
4994 * and the acquisition of the spinlock, we must also ping the
4995 * CPU after setting the request bit.
4999 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
5001 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
5004 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5006 raw_spin_lock(&kvm_lock
);
5007 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5008 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
5009 if (vcpu
->cpu
!= freq
->cpu
)
5011 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5012 if (vcpu
->cpu
!= smp_processor_id())
5016 raw_spin_unlock(&kvm_lock
);
5018 if (freq
->old
< freq
->new && send_ipi
) {
5020 * We upscale the frequency. Must make the guest
5021 * doesn't see old kvmclock values while running with
5022 * the new frequency, otherwise we risk the guest sees
5023 * time go backwards.
5025 * In case we update the frequency for another cpu
5026 * (which might be in guest context) send an interrupt
5027 * to kick the cpu out of guest context. Next time
5028 * guest context is entered kvmclock will be updated,
5029 * so the guest will not see stale values.
5031 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5036 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
5037 .notifier_call
= kvmclock_cpufreq_notifier
5040 static int kvmclock_cpu_notifier(struct notifier_block
*nfb
,
5041 unsigned long action
, void *hcpu
)
5043 unsigned int cpu
= (unsigned long)hcpu
;
5047 case CPU_DOWN_FAILED
:
5048 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5050 case CPU_DOWN_PREPARE
:
5051 smp_call_function_single(cpu
, tsc_bad
, NULL
, 1);
5057 static struct notifier_block kvmclock_cpu_notifier_block
= {
5058 .notifier_call
= kvmclock_cpu_notifier
,
5059 .priority
= -INT_MAX
5062 static void kvm_timer_init(void)
5066 max_tsc_khz
= tsc_khz
;
5067 register_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5068 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
5069 #ifdef CONFIG_CPU_FREQ
5070 struct cpufreq_policy policy
;
5071 memset(&policy
, 0, sizeof(policy
));
5073 cpufreq_get_policy(&policy
, cpu
);
5074 if (policy
.cpuinfo
.max_freq
)
5075 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
5078 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
5079 CPUFREQ_TRANSITION_NOTIFIER
);
5081 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
5082 for_each_online_cpu(cpu
)
5083 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5086 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
5088 int kvm_is_in_guest(void)
5090 return __this_cpu_read(current_vcpu
) != NULL
;
5093 static int kvm_is_user_mode(void)
5097 if (__this_cpu_read(current_vcpu
))
5098 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
5100 return user_mode
!= 0;
5103 static unsigned long kvm_get_guest_ip(void)
5105 unsigned long ip
= 0;
5107 if (__this_cpu_read(current_vcpu
))
5108 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
5113 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
5114 .is_in_guest
= kvm_is_in_guest
,
5115 .is_user_mode
= kvm_is_user_mode
,
5116 .get_guest_ip
= kvm_get_guest_ip
,
5119 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
5121 __this_cpu_write(current_vcpu
, vcpu
);
5123 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
5125 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
5127 __this_cpu_write(current_vcpu
, NULL
);
5129 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
5131 static void kvm_set_mmio_spte_mask(void)
5134 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
5137 * Set the reserved bits and the present bit of an paging-structure
5138 * entry to generate page fault with PFER.RSV = 1.
5140 mask
= ((1ull << (62 - maxphyaddr
+ 1)) - 1) << maxphyaddr
;
5143 #ifdef CONFIG_X86_64
5145 * If reserved bit is not supported, clear the present bit to disable
5148 if (maxphyaddr
== 52)
5152 kvm_mmu_set_mmio_spte_mask(mask
);
5155 #ifdef CONFIG_X86_64
5156 static void pvclock_gtod_update_fn(struct work_struct
*work
)
5160 struct kvm_vcpu
*vcpu
;
5163 raw_spin_lock(&kvm_lock
);
5164 list_for_each_entry(kvm
, &vm_list
, vm_list
)
5165 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5166 set_bit(KVM_REQ_MASTERCLOCK_UPDATE
, &vcpu
->requests
);
5167 atomic_set(&kvm_guest_has_master_clock
, 0);
5168 raw_spin_unlock(&kvm_lock
);
5171 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
5174 * Notification about pvclock gtod data update.
5176 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
5179 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
5180 struct timekeeper
*tk
= priv
;
5182 update_pvclock_gtod(tk
);
5184 /* disable master clock if host does not trust, or does not
5185 * use, TSC clocksource
5187 if (gtod
->clock
.vclock_mode
!= VCLOCK_TSC
&&
5188 atomic_read(&kvm_guest_has_master_clock
) != 0)
5189 queue_work(system_long_wq
, &pvclock_gtod_work
);
5194 static struct notifier_block pvclock_gtod_notifier
= {
5195 .notifier_call
= pvclock_gtod_notify
,
5199 int kvm_arch_init(void *opaque
)
5202 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
5205 printk(KERN_ERR
"kvm: already loaded the other module\n");
5210 if (!ops
->cpu_has_kvm_support()) {
5211 printk(KERN_ERR
"kvm: no hardware support\n");
5215 if (ops
->disabled_by_bios()) {
5216 printk(KERN_ERR
"kvm: disabled by bios\n");
5221 r
= kvm_mmu_module_init();
5225 kvm_set_mmio_spte_mask();
5226 kvm_init_msr_list();
5229 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
5230 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
5234 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
5237 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
5240 #ifdef CONFIG_X86_64
5241 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
5250 void kvm_arch_exit(void)
5252 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
5254 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5255 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
5256 CPUFREQ_TRANSITION_NOTIFIER
);
5257 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5258 #ifdef CONFIG_X86_64
5259 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
5262 kvm_mmu_module_exit();
5265 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
5267 ++vcpu
->stat
.halt_exits
;
5268 if (irqchip_in_kernel(vcpu
->kvm
)) {
5269 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
5272 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
5276 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
5278 int kvm_hv_hypercall(struct kvm_vcpu
*vcpu
)
5280 u64 param
, ingpa
, outgpa
, ret
;
5281 uint16_t code
, rep_idx
, rep_cnt
, res
= HV_STATUS_SUCCESS
, rep_done
= 0;
5282 bool fast
, longmode
;
5286 * hypercall generates UD from non zero cpl and real mode
5289 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 || !is_protmode(vcpu
)) {
5290 kvm_queue_exception(vcpu
, UD_VECTOR
);
5294 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
5295 longmode
= is_long_mode(vcpu
) && cs_l
== 1;
5298 param
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDX
) << 32) |
5299 (kvm_register_read(vcpu
, VCPU_REGS_RAX
) & 0xffffffff);
5300 ingpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RBX
) << 32) |
5301 (kvm_register_read(vcpu
, VCPU_REGS_RCX
) & 0xffffffff);
5302 outgpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDI
) << 32) |
5303 (kvm_register_read(vcpu
, VCPU_REGS_RSI
) & 0xffffffff);
5305 #ifdef CONFIG_X86_64
5307 param
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5308 ingpa
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5309 outgpa
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
5313 code
= param
& 0xffff;
5314 fast
= (param
>> 16) & 0x1;
5315 rep_cnt
= (param
>> 32) & 0xfff;
5316 rep_idx
= (param
>> 48) & 0xfff;
5318 trace_kvm_hv_hypercall(code
, fast
, rep_cnt
, rep_idx
, ingpa
, outgpa
);
5321 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT
:
5322 kvm_vcpu_on_spin(vcpu
);
5325 res
= HV_STATUS_INVALID_HYPERCALL_CODE
;
5329 ret
= res
| (((u64
)rep_done
& 0xfff) << 32);
5331 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5333 kvm_register_write(vcpu
, VCPU_REGS_RDX
, ret
>> 32);
5334 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
& 0xffffffff);
5340 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
5342 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
5345 if (kvm_hv_hypercall_enabled(vcpu
->kvm
))
5346 return kvm_hv_hypercall(vcpu
);
5348 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5349 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5350 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5351 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5352 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5354 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
5356 if (!is_long_mode(vcpu
)) {
5364 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
5370 case KVM_HC_VAPIC_POLL_IRQ
:
5378 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5379 ++vcpu
->stat
.hypercalls
;
5382 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
5384 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
5386 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5387 char instruction
[3];
5388 unsigned long rip
= kvm_rip_read(vcpu
);
5391 * Blow out the MMU to ensure that no other VCPU has an active mapping
5392 * to ensure that the updated hypercall appears atomically across all
5395 kvm_mmu_zap_all(vcpu
->kvm
);
5397 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
5399 return emulator_write_emulated(ctxt
, rip
, instruction
, 3, NULL
);
5403 * Check if userspace requested an interrupt window, and that the
5404 * interrupt window is open.
5406 * No need to exit to userspace if we already have an interrupt queued.
5408 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
5410 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
5411 vcpu
->run
->request_interrupt_window
&&
5412 kvm_arch_interrupt_allowed(vcpu
));
5415 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
5417 struct kvm_run
*kvm_run
= vcpu
->run
;
5419 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
5420 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
5421 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
5422 if (irqchip_in_kernel(vcpu
->kvm
))
5423 kvm_run
->ready_for_interrupt_injection
= 1;
5425 kvm_run
->ready_for_interrupt_injection
=
5426 kvm_arch_interrupt_allowed(vcpu
) &&
5427 !kvm_cpu_has_interrupt(vcpu
) &&
5428 !kvm_event_needs_reinjection(vcpu
);
5431 static int vapic_enter(struct kvm_vcpu
*vcpu
)
5433 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5436 if (!apic
|| !apic
->vapic_addr
)
5439 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5440 if (is_error_page(page
))
5443 vcpu
->arch
.apic
->vapic_page
= page
;
5447 static void vapic_exit(struct kvm_vcpu
*vcpu
)
5449 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
5452 if (!apic
|| !apic
->vapic_addr
)
5455 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5456 kvm_release_page_dirty(apic
->vapic_page
);
5457 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
5458 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5461 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
5465 if (!kvm_x86_ops
->update_cr8_intercept
)
5468 if (!vcpu
->arch
.apic
)
5471 if (!vcpu
->arch
.apic
->vapic_addr
)
5472 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
5479 tpr
= kvm_lapic_get_cr8(vcpu
);
5481 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
5484 static void inject_pending_event(struct kvm_vcpu
*vcpu
)
5486 /* try to reinject previous events if any */
5487 if (vcpu
->arch
.exception
.pending
) {
5488 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
5489 vcpu
->arch
.exception
.has_error_code
,
5490 vcpu
->arch
.exception
.error_code
);
5491 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
5492 vcpu
->arch
.exception
.has_error_code
,
5493 vcpu
->arch
.exception
.error_code
,
5494 vcpu
->arch
.exception
.reinject
);
5498 if (vcpu
->arch
.nmi_injected
) {
5499 kvm_x86_ops
->set_nmi(vcpu
);
5503 if (vcpu
->arch
.interrupt
.pending
) {
5504 kvm_x86_ops
->set_irq(vcpu
);
5508 /* try to inject new event if pending */
5509 if (vcpu
->arch
.nmi_pending
) {
5510 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
5511 --vcpu
->arch
.nmi_pending
;
5512 vcpu
->arch
.nmi_injected
= true;
5513 kvm_x86_ops
->set_nmi(vcpu
);
5515 } else if (kvm_cpu_has_interrupt(vcpu
)) {
5516 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
5517 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
5519 kvm_x86_ops
->set_irq(vcpu
);
5524 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
5526 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
5527 !vcpu
->guest_xcr0_loaded
) {
5528 /* kvm_set_xcr() also depends on this */
5529 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
5530 vcpu
->guest_xcr0_loaded
= 1;
5534 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
5536 if (vcpu
->guest_xcr0_loaded
) {
5537 if (vcpu
->arch
.xcr0
!= host_xcr0
)
5538 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
5539 vcpu
->guest_xcr0_loaded
= 0;
5543 static void process_nmi(struct kvm_vcpu
*vcpu
)
5548 * x86 is limited to one NMI running, and one NMI pending after it.
5549 * If an NMI is already in progress, limit further NMIs to just one.
5550 * Otherwise, allow two (and we'll inject the first one immediately).
5552 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
5555 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
5556 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
5557 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5560 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
5562 #ifdef CONFIG_X86_64
5564 struct kvm_vcpu
*vcpu
;
5565 struct kvm_arch
*ka
= &kvm
->arch
;
5567 spin_lock(&ka
->pvclock_gtod_sync_lock
);
5568 kvm_make_mclock_inprogress_request(kvm
);
5569 /* no guest entries from this point */
5570 pvclock_update_vm_gtod_copy(kvm
);
5572 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5573 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
5575 /* guest entries allowed */
5576 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5577 clear_bit(KVM_REQ_MCLOCK_INPROGRESS
, &vcpu
->requests
);
5579 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
5583 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
5586 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
5587 vcpu
->run
->request_interrupt_window
;
5588 bool req_immediate_exit
= 0;
5590 if (vcpu
->requests
) {
5591 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
5592 kvm_mmu_unload(vcpu
);
5593 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
5594 __kvm_migrate_timers(vcpu
);
5595 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
5596 kvm_gen_update_masterclock(vcpu
->kvm
);
5597 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
5598 r
= kvm_guest_time_update(vcpu
);
5602 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
5603 kvm_mmu_sync_roots(vcpu
);
5604 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
5605 kvm_x86_ops
->tlb_flush(vcpu
);
5606 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
5607 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
5611 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
5612 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
5616 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
)) {
5617 vcpu
->fpu_active
= 0;
5618 kvm_x86_ops
->fpu_deactivate(vcpu
);
5620 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
5621 /* Page is swapped out. Do synthetic halt */
5622 vcpu
->arch
.apf
.halted
= true;
5626 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
5627 record_steal_time(vcpu
);
5628 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
5630 req_immediate_exit
=
5631 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT
, vcpu
);
5632 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
5633 kvm_handle_pmu_event(vcpu
);
5634 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
5635 kvm_deliver_pmi(vcpu
);
5638 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
5639 inject_pending_event(vcpu
);
5641 /* enable NMI/IRQ window open exits if needed */
5642 if (vcpu
->arch
.nmi_pending
)
5643 kvm_x86_ops
->enable_nmi_window(vcpu
);
5644 else if (kvm_cpu_has_interrupt(vcpu
) || req_int_win
)
5645 kvm_x86_ops
->enable_irq_window(vcpu
);
5647 if (kvm_lapic_enabled(vcpu
)) {
5648 update_cr8_intercept(vcpu
);
5649 kvm_lapic_sync_to_vapic(vcpu
);
5653 r
= kvm_mmu_reload(vcpu
);
5655 goto cancel_injection
;
5660 kvm_x86_ops
->prepare_guest_switch(vcpu
);
5661 if (vcpu
->fpu_active
)
5662 kvm_load_guest_fpu(vcpu
);
5663 kvm_load_guest_xcr0(vcpu
);
5665 vcpu
->mode
= IN_GUEST_MODE
;
5667 /* We should set ->mode before check ->requests,
5668 * see the comment in make_all_cpus_request.
5672 local_irq_disable();
5674 if (vcpu
->mode
== EXITING_GUEST_MODE
|| vcpu
->requests
5675 || need_resched() || signal_pending(current
)) {
5676 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5681 goto cancel_injection
;
5684 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5686 if (req_immediate_exit
)
5687 smp_send_reschedule(vcpu
->cpu
);
5691 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
5693 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
5694 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
5695 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
5696 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
5699 trace_kvm_entry(vcpu
->vcpu_id
);
5700 kvm_x86_ops
->run(vcpu
);
5703 * If the guest has used debug registers, at least dr7
5704 * will be disabled while returning to the host.
5705 * If we don't have active breakpoints in the host, we don't
5706 * care about the messed up debug address registers. But if
5707 * we have some of them active, restore the old state.
5709 if (hw_breakpoint_active())
5710 hw_breakpoint_restore();
5712 vcpu
->arch
.last_guest_tsc
= kvm_x86_ops
->read_l1_tsc(vcpu
,
5715 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5722 * We must have an instruction between local_irq_enable() and
5723 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5724 * the interrupt shadow. The stat.exits increment will do nicely.
5725 * But we need to prevent reordering, hence this barrier():
5733 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5736 * Profile KVM exit RIPs:
5738 if (unlikely(prof_on
== KVM_PROFILING
)) {
5739 unsigned long rip
= kvm_rip_read(vcpu
);
5740 profile_hit(KVM_PROFILING
, (void *)rip
);
5743 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
5744 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5746 if (vcpu
->arch
.apic_attention
)
5747 kvm_lapic_sync_from_vapic(vcpu
);
5749 r
= kvm_x86_ops
->handle_exit(vcpu
);
5753 kvm_x86_ops
->cancel_injection(vcpu
);
5754 if (unlikely(vcpu
->arch
.apic_attention
))
5755 kvm_lapic_sync_from_vapic(vcpu
);
5761 static int __vcpu_run(struct kvm_vcpu
*vcpu
)
5764 struct kvm
*kvm
= vcpu
->kvm
;
5766 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
5767 pr_debug("vcpu %d received sipi with vector # %x\n",
5768 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
5769 kvm_lapic_reset(vcpu
);
5770 r
= kvm_vcpu_reset(vcpu
);
5773 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
5776 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5777 r
= vapic_enter(vcpu
);
5779 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5785 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
5786 !vcpu
->arch
.apf
.halted
)
5787 r
= vcpu_enter_guest(vcpu
);
5789 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5790 kvm_vcpu_block(vcpu
);
5791 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5792 if (kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
5794 switch(vcpu
->arch
.mp_state
) {
5795 case KVM_MP_STATE_HALTED
:
5796 vcpu
->arch
.mp_state
=
5797 KVM_MP_STATE_RUNNABLE
;
5798 case KVM_MP_STATE_RUNNABLE
:
5799 vcpu
->arch
.apf
.halted
= false;
5801 case KVM_MP_STATE_SIPI_RECEIVED
:
5812 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
5813 if (kvm_cpu_has_pending_timer(vcpu
))
5814 kvm_inject_pending_timer_irqs(vcpu
);
5816 if (dm_request_for_irq_injection(vcpu
)) {
5818 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5819 ++vcpu
->stat
.request_irq_exits
;
5822 kvm_check_async_pf_completion(vcpu
);
5824 if (signal_pending(current
)) {
5826 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5827 ++vcpu
->stat
.signal_exits
;
5829 if (need_resched()) {
5830 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5832 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5836 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5843 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
5846 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5847 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
5848 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5849 if (r
!= EMULATE_DONE
)
5854 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
5856 BUG_ON(!vcpu
->arch
.pio
.count
);
5858 return complete_emulated_io(vcpu
);
5862 * Implements the following, as a state machine:
5877 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
5879 struct kvm_run
*run
= vcpu
->run
;
5880 struct kvm_mmio_fragment
*frag
;
5882 BUG_ON(!vcpu
->mmio_needed
);
5884 /* Complete previous fragment */
5885 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
++];
5886 if (!vcpu
->mmio_is_write
)
5887 memcpy(frag
->data
, run
->mmio
.data
, frag
->len
);
5888 if (vcpu
->mmio_cur_fragment
== vcpu
->mmio_nr_fragments
) {
5889 vcpu
->mmio_needed
= 0;
5890 if (vcpu
->mmio_is_write
)
5892 vcpu
->mmio_read_completed
= 1;
5893 return complete_emulated_io(vcpu
);
5895 /* Initiate next fragment */
5897 run
->exit_reason
= KVM_EXIT_MMIO
;
5898 run
->mmio
.phys_addr
= frag
->gpa
;
5899 if (vcpu
->mmio_is_write
)
5900 memcpy(run
->mmio
.data
, frag
->data
, frag
->len
);
5901 run
->mmio
.len
= frag
->len
;
5902 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
5903 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
5908 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
5913 if (!tsk_used_math(current
) && init_fpu(current
))
5916 if (vcpu
->sigset_active
)
5917 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
5919 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
5920 kvm_vcpu_block(vcpu
);
5921 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
5926 /* re-sync apic's tpr */
5927 if (!irqchip_in_kernel(vcpu
->kvm
)) {
5928 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
5934 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
5935 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
5936 vcpu
->arch
.complete_userspace_io
= NULL
;
5941 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
5943 r
= __vcpu_run(vcpu
);
5946 post_kvm_run_save(vcpu
);
5947 if (vcpu
->sigset_active
)
5948 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
5953 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
5955 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
5957 * We are here if userspace calls get_regs() in the middle of
5958 * instruction emulation. Registers state needs to be copied
5959 * back from emulation context to vcpu. Userspace shouldn't do
5960 * that usually, but some bad designed PV devices (vmware
5961 * backdoor interface) need this to work
5963 emulator_writeback_register_cache(&vcpu
->arch
.emulate_ctxt
);
5964 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5966 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5967 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5968 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5969 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5970 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5971 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
5972 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
5973 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
5974 #ifdef CONFIG_X86_64
5975 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
5976 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
5977 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
5978 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
5979 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
5980 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
5981 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
5982 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
5985 regs
->rip
= kvm_rip_read(vcpu
);
5986 regs
->rflags
= kvm_get_rflags(vcpu
);
5991 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
5993 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
5994 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5996 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
5997 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
5998 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
5999 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
6000 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
6001 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
6002 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
6003 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
6004 #ifdef CONFIG_X86_64
6005 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
6006 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
6007 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
6008 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
6009 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
6010 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
6011 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
6012 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
6015 kvm_rip_write(vcpu
, regs
->rip
);
6016 kvm_set_rflags(vcpu
, regs
->rflags
);
6018 vcpu
->arch
.exception
.pending
= false;
6020 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6025 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
6027 struct kvm_segment cs
;
6029 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6033 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
6035 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
6036 struct kvm_sregs
*sregs
)
6040 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
6041 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
6042 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
6043 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
6044 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
6045 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
6047 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
6048 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
6050 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6051 sregs
->idt
.limit
= dt
.size
;
6052 sregs
->idt
.base
= dt
.address
;
6053 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6054 sregs
->gdt
.limit
= dt
.size
;
6055 sregs
->gdt
.base
= dt
.address
;
6057 sregs
->cr0
= kvm_read_cr0(vcpu
);
6058 sregs
->cr2
= vcpu
->arch
.cr2
;
6059 sregs
->cr3
= kvm_read_cr3(vcpu
);
6060 sregs
->cr4
= kvm_read_cr4(vcpu
);
6061 sregs
->cr8
= kvm_get_cr8(vcpu
);
6062 sregs
->efer
= vcpu
->arch
.efer
;
6063 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
6065 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
6067 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
6068 set_bit(vcpu
->arch
.interrupt
.nr
,
6069 (unsigned long *)sregs
->interrupt_bitmap
);
6074 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
6075 struct kvm_mp_state
*mp_state
)
6077 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
6081 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
6082 struct kvm_mp_state
*mp_state
)
6084 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
6085 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6089 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
6090 int reason
, bool has_error_code
, u32 error_code
)
6092 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
6095 init_emulate_ctxt(vcpu
);
6097 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
6098 has_error_code
, error_code
);
6101 return EMULATE_FAIL
;
6103 kvm_rip_write(vcpu
, ctxt
->eip
);
6104 kvm_set_rflags(vcpu
, ctxt
->eflags
);
6105 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6106 return EMULATE_DONE
;
6108 EXPORT_SYMBOL_GPL(kvm_task_switch
);
6110 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
6111 struct kvm_sregs
*sregs
)
6113 int mmu_reset_needed
= 0;
6114 int pending_vec
, max_bits
, idx
;
6117 dt
.size
= sregs
->idt
.limit
;
6118 dt
.address
= sregs
->idt
.base
;
6119 kvm_x86_ops
->set_idt(vcpu
, &dt
);
6120 dt
.size
= sregs
->gdt
.limit
;
6121 dt
.address
= sregs
->gdt
.base
;
6122 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
6124 vcpu
->arch
.cr2
= sregs
->cr2
;
6125 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
6126 vcpu
->arch
.cr3
= sregs
->cr3
;
6127 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
6129 kvm_set_cr8(vcpu
, sregs
->cr8
);
6131 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
6132 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
6133 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
6135 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
6136 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
6137 vcpu
->arch
.cr0
= sregs
->cr0
;
6139 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
6140 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
6141 if (sregs
->cr4
& X86_CR4_OSXSAVE
)
6142 kvm_update_cpuid(vcpu
);
6144 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6145 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
6146 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
6147 mmu_reset_needed
= 1;
6149 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6151 if (mmu_reset_needed
)
6152 kvm_mmu_reset_context(vcpu
);
6154 max_bits
= KVM_NR_INTERRUPTS
;
6155 pending_vec
= find_first_bit(
6156 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
6157 if (pending_vec
< max_bits
) {
6158 kvm_queue_interrupt(vcpu
, pending_vec
, false);
6159 pr_debug("Set back pending irq %d\n", pending_vec
);
6162 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
6163 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
6164 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
6165 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
6166 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
6167 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
6169 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
6170 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
6172 update_cr8_intercept(vcpu
);
6174 /* Older userspace won't unhalt the vcpu on reset. */
6175 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
6176 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
6178 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6180 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6185 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
6186 struct kvm_guest_debug
*dbg
)
6188 unsigned long rflags
;
6191 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
6193 if (vcpu
->arch
.exception
.pending
)
6195 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
6196 kvm_queue_exception(vcpu
, DB_VECTOR
);
6198 kvm_queue_exception(vcpu
, BP_VECTOR
);
6202 * Read rflags as long as potentially injected trace flags are still
6205 rflags
= kvm_get_rflags(vcpu
);
6207 vcpu
->guest_debug
= dbg
->control
;
6208 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
6209 vcpu
->guest_debug
= 0;
6211 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
6212 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
6213 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
6214 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
6216 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
6217 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
6219 kvm_update_dr7(vcpu
);
6221 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
6222 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
6223 get_segment_base(vcpu
, VCPU_SREG_CS
);
6226 * Trigger an rflags update that will inject or remove the trace
6229 kvm_set_rflags(vcpu
, rflags
);
6231 kvm_x86_ops
->update_db_bp_intercept(vcpu
);
6241 * Translate a guest virtual address to a guest physical address.
6243 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
6244 struct kvm_translation
*tr
)
6246 unsigned long vaddr
= tr
->linear_address
;
6250 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6251 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
6252 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6253 tr
->physical_address
= gpa
;
6254 tr
->valid
= gpa
!= UNMAPPED_GVA
;
6261 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
6263 struct i387_fxsave_struct
*fxsave
=
6264 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
6266 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
6267 fpu
->fcw
= fxsave
->cwd
;
6268 fpu
->fsw
= fxsave
->swd
;
6269 fpu
->ftwx
= fxsave
->twd
;
6270 fpu
->last_opcode
= fxsave
->fop
;
6271 fpu
->last_ip
= fxsave
->rip
;
6272 fpu
->last_dp
= fxsave
->rdp
;
6273 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
6278 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
6280 struct i387_fxsave_struct
*fxsave
=
6281 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
6283 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
6284 fxsave
->cwd
= fpu
->fcw
;
6285 fxsave
->swd
= fpu
->fsw
;
6286 fxsave
->twd
= fpu
->ftwx
;
6287 fxsave
->fop
= fpu
->last_opcode
;
6288 fxsave
->rip
= fpu
->last_ip
;
6289 fxsave
->rdp
= fpu
->last_dp
;
6290 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
6295 int fx_init(struct kvm_vcpu
*vcpu
)
6299 err
= fpu_alloc(&vcpu
->arch
.guest_fpu
);
6303 fpu_finit(&vcpu
->arch
.guest_fpu
);
6306 * Ensure guest xcr0 is valid for loading
6308 vcpu
->arch
.xcr0
= XSTATE_FP
;
6310 vcpu
->arch
.cr0
|= X86_CR0_ET
;
6314 EXPORT_SYMBOL_GPL(fx_init
);
6316 static void fx_free(struct kvm_vcpu
*vcpu
)
6318 fpu_free(&vcpu
->arch
.guest_fpu
);
6321 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
6323 if (vcpu
->guest_fpu_loaded
)
6327 * Restore all possible states in the guest,
6328 * and assume host would use all available bits.
6329 * Guest xcr0 would be loaded later.
6331 kvm_put_guest_xcr0(vcpu
);
6332 vcpu
->guest_fpu_loaded
= 1;
6333 __kernel_fpu_begin();
6334 fpu_restore_checking(&vcpu
->arch
.guest_fpu
);
6338 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
6340 kvm_put_guest_xcr0(vcpu
);
6342 if (!vcpu
->guest_fpu_loaded
)
6345 vcpu
->guest_fpu_loaded
= 0;
6346 fpu_save_init(&vcpu
->arch
.guest_fpu
);
6348 ++vcpu
->stat
.fpu_reload
;
6349 kvm_make_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
);
6353 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
6355 kvmclock_reset(vcpu
);
6357 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
6359 kvm_x86_ops
->vcpu_free(vcpu
);
6362 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
6365 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
6366 printk_once(KERN_WARNING
6367 "kvm: SMP vm created on host with unstable TSC; "
6368 "guest TSC will not be reliable\n");
6369 return kvm_x86_ops
->vcpu_create(kvm
, id
);
6372 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
6376 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
6377 r
= vcpu_load(vcpu
);
6380 r
= kvm_vcpu_reset(vcpu
);
6382 r
= kvm_mmu_setup(vcpu
);
6388 int kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
6391 struct msr_data msr
;
6393 r
= vcpu_load(vcpu
);
6397 msr
.index
= MSR_IA32_TSC
;
6398 msr
.host_initiated
= true;
6399 kvm_write_tsc(vcpu
, &msr
);
6405 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
6408 vcpu
->arch
.apf
.msr_val
= 0;
6410 r
= vcpu_load(vcpu
);
6412 kvm_mmu_unload(vcpu
);
6416 kvm_x86_ops
->vcpu_free(vcpu
);
6419 static int kvm_vcpu_reset(struct kvm_vcpu
*vcpu
)
6421 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
6422 vcpu
->arch
.nmi_pending
= 0;
6423 vcpu
->arch
.nmi_injected
= false;
6425 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
6426 vcpu
->arch
.dr6
= DR6_FIXED_1
;
6427 vcpu
->arch
.dr7
= DR7_FIXED_1
;
6428 kvm_update_dr7(vcpu
);
6430 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6431 vcpu
->arch
.apf
.msr_val
= 0;
6432 vcpu
->arch
.st
.msr_val
= 0;
6434 kvmclock_reset(vcpu
);
6436 kvm_clear_async_pf_completion_queue(vcpu
);
6437 kvm_async_pf_hash_reset(vcpu
);
6438 vcpu
->arch
.apf
.halted
= false;
6440 kvm_pmu_reset(vcpu
);
6442 return kvm_x86_ops
->vcpu_reset(vcpu
);
6445 int kvm_arch_hardware_enable(void *garbage
)
6448 struct kvm_vcpu
*vcpu
;
6453 bool stable
, backwards_tsc
= false;
6455 kvm_shared_msr_cpu_online();
6456 ret
= kvm_x86_ops
->hardware_enable(garbage
);
6460 local_tsc
= native_read_tsc();
6461 stable
= !check_tsc_unstable();
6462 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6463 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6464 if (!stable
&& vcpu
->cpu
== smp_processor_id())
6465 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
6466 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
6467 backwards_tsc
= true;
6468 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
6469 max_tsc
= vcpu
->arch
.last_host_tsc
;
6475 * Sometimes, even reliable TSCs go backwards. This happens on
6476 * platforms that reset TSC during suspend or hibernate actions, but
6477 * maintain synchronization. We must compensate. Fortunately, we can
6478 * detect that condition here, which happens early in CPU bringup,
6479 * before any KVM threads can be running. Unfortunately, we can't
6480 * bring the TSCs fully up to date with real time, as we aren't yet far
6481 * enough into CPU bringup that we know how much real time has actually
6482 * elapsed; our helper function, get_kernel_ns() will be using boot
6483 * variables that haven't been updated yet.
6485 * So we simply find the maximum observed TSC above, then record the
6486 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6487 * the adjustment will be applied. Note that we accumulate
6488 * adjustments, in case multiple suspend cycles happen before some VCPU
6489 * gets a chance to run again. In the event that no KVM threads get a
6490 * chance to run, we will miss the entire elapsed period, as we'll have
6491 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6492 * loose cycle time. This isn't too big a deal, since the loss will be
6493 * uniform across all VCPUs (not to mention the scenario is extremely
6494 * unlikely). It is possible that a second hibernate recovery happens
6495 * much faster than a first, causing the observed TSC here to be
6496 * smaller; this would require additional padding adjustment, which is
6497 * why we set last_host_tsc to the local tsc observed here.
6499 * N.B. - this code below runs only on platforms with reliable TSC,
6500 * as that is the only way backwards_tsc is set above. Also note
6501 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6502 * have the same delta_cyc adjustment applied if backwards_tsc
6503 * is detected. Note further, this adjustment is only done once,
6504 * as we reset last_host_tsc on all VCPUs to stop this from being
6505 * called multiple times (one for each physical CPU bringup).
6507 * Platforms with unreliable TSCs don't have to deal with this, they
6508 * will be compensated by the logic in vcpu_load, which sets the TSC to
6509 * catchup mode. This will catchup all VCPUs to real time, but cannot
6510 * guarantee that they stay in perfect synchronization.
6512 if (backwards_tsc
) {
6513 u64 delta_cyc
= max_tsc
- local_tsc
;
6514 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6515 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6516 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
6517 vcpu
->arch
.last_host_tsc
= local_tsc
;
6518 set_bit(KVM_REQ_MASTERCLOCK_UPDATE
,
6523 * We have to disable TSC offset matching.. if you were
6524 * booting a VM while issuing an S4 host suspend....
6525 * you may have some problem. Solving this issue is
6526 * left as an exercise to the reader.
6528 kvm
->arch
.last_tsc_nsec
= 0;
6529 kvm
->arch
.last_tsc_write
= 0;
6536 void kvm_arch_hardware_disable(void *garbage
)
6538 kvm_x86_ops
->hardware_disable(garbage
);
6539 drop_user_return_notifiers(garbage
);
6542 int kvm_arch_hardware_setup(void)
6544 return kvm_x86_ops
->hardware_setup();
6547 void kvm_arch_hardware_unsetup(void)
6549 kvm_x86_ops
->hardware_unsetup();
6552 void kvm_arch_check_processor_compat(void *rtn
)
6554 kvm_x86_ops
->check_processor_compatibility(rtn
);
6557 bool kvm_vcpu_compatible(struct kvm_vcpu
*vcpu
)
6559 return irqchip_in_kernel(vcpu
->kvm
) == (vcpu
->arch
.apic
!= NULL
);
6562 struct static_key kvm_no_apic_vcpu __read_mostly
;
6564 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
6570 BUG_ON(vcpu
->kvm
== NULL
);
6573 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
6574 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
6575 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6577 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
6579 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
6584 vcpu
->arch
.pio_data
= page_address(page
);
6586 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
6588 r
= kvm_mmu_create(vcpu
);
6590 goto fail_free_pio_data
;
6592 if (irqchip_in_kernel(kvm
)) {
6593 r
= kvm_create_lapic(vcpu
);
6595 goto fail_mmu_destroy
;
6597 static_key_slow_inc(&kvm_no_apic_vcpu
);
6599 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
6601 if (!vcpu
->arch
.mce_banks
) {
6603 goto fail_free_lapic
;
6605 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
6607 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
))
6608 goto fail_free_mce_banks
;
6610 kvm_async_pf_hash_reset(vcpu
);
6614 fail_free_mce_banks
:
6615 kfree(vcpu
->arch
.mce_banks
);
6617 kvm_free_lapic(vcpu
);
6619 kvm_mmu_destroy(vcpu
);
6621 free_page((unsigned long)vcpu
->arch
.pio_data
);
6626 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
6630 kvm_pmu_destroy(vcpu
);
6631 kfree(vcpu
->arch
.mce_banks
);
6632 kvm_free_lapic(vcpu
);
6633 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6634 kvm_mmu_destroy(vcpu
);
6635 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6636 free_page((unsigned long)vcpu
->arch
.pio_data
);
6637 if (!irqchip_in_kernel(vcpu
->kvm
))
6638 static_key_slow_dec(&kvm_no_apic_vcpu
);
6641 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
6646 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
6647 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
6649 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6650 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
6651 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6652 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
6653 &kvm
->arch
.irq_sources_bitmap
);
6655 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
6656 mutex_init(&kvm
->arch
.apic_map_lock
);
6657 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
6659 pvclock_update_vm_gtod_copy(kvm
);
6664 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
6667 r
= vcpu_load(vcpu
);
6669 kvm_mmu_unload(vcpu
);
6673 static void kvm_free_vcpus(struct kvm
*kvm
)
6676 struct kvm_vcpu
*vcpu
;
6679 * Unpin any mmu pages first.
6681 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6682 kvm_clear_async_pf_completion_queue(vcpu
);
6683 kvm_unload_vcpu_mmu(vcpu
);
6685 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6686 kvm_arch_vcpu_free(vcpu
);
6688 mutex_lock(&kvm
->lock
);
6689 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
6690 kvm
->vcpus
[i
] = NULL
;
6692 atomic_set(&kvm
->online_vcpus
, 0);
6693 mutex_unlock(&kvm
->lock
);
6696 void kvm_arch_sync_events(struct kvm
*kvm
)
6698 kvm_free_all_assigned_devices(kvm
);
6702 void kvm_arch_destroy_vm(struct kvm
*kvm
)
6704 kvm_iommu_unmap_guest(kvm
);
6705 kfree(kvm
->arch
.vpic
);
6706 kfree(kvm
->arch
.vioapic
);
6707 kvm_free_vcpus(kvm
);
6708 if (kvm
->arch
.apic_access_page
)
6709 put_page(kvm
->arch
.apic_access_page
);
6710 if (kvm
->arch
.ept_identity_pagetable
)
6711 put_page(kvm
->arch
.ept_identity_pagetable
);
6712 kfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
6715 void kvm_arch_free_memslot(struct kvm_memory_slot
*free
,
6716 struct kvm_memory_slot
*dont
)
6720 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6721 if (!dont
|| free
->arch
.rmap
[i
] != dont
->arch
.rmap
[i
]) {
6722 kvm_kvfree(free
->arch
.rmap
[i
]);
6723 free
->arch
.rmap
[i
] = NULL
;
6728 if (!dont
|| free
->arch
.lpage_info
[i
- 1] !=
6729 dont
->arch
.lpage_info
[i
- 1]) {
6730 kvm_kvfree(free
->arch
.lpage_info
[i
- 1]);
6731 free
->arch
.lpage_info
[i
- 1] = NULL
;
6736 int kvm_arch_create_memslot(struct kvm_memory_slot
*slot
, unsigned long npages
)
6740 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6745 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
6746 slot
->base_gfn
, level
) + 1;
6748 slot
->arch
.rmap
[i
] =
6749 kvm_kvzalloc(lpages
* sizeof(*slot
->arch
.rmap
[i
]));
6750 if (!slot
->arch
.rmap
[i
])
6755 slot
->arch
.lpage_info
[i
- 1] = kvm_kvzalloc(lpages
*
6756 sizeof(*slot
->arch
.lpage_info
[i
- 1]));
6757 if (!slot
->arch
.lpage_info
[i
- 1])
6760 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
6761 slot
->arch
.lpage_info
[i
- 1][0].write_count
= 1;
6762 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
6763 slot
->arch
.lpage_info
[i
- 1][lpages
- 1].write_count
= 1;
6764 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
6766 * If the gfn and userspace address are not aligned wrt each
6767 * other, or if explicitly asked to, disable large page
6768 * support for this slot
6770 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
6771 !kvm_largepages_enabled()) {
6774 for (j
= 0; j
< lpages
; ++j
)
6775 slot
->arch
.lpage_info
[i
- 1][j
].write_count
= 1;
6782 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6783 kvm_kvfree(slot
->arch
.rmap
[i
]);
6784 slot
->arch
.rmap
[i
] = NULL
;
6788 kvm_kvfree(slot
->arch
.lpage_info
[i
- 1]);
6789 slot
->arch
.lpage_info
[i
- 1] = NULL
;
6794 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
6795 struct kvm_memory_slot
*memslot
,
6796 struct kvm_memory_slot old
,
6797 struct kvm_userspace_memory_region
*mem
,
6800 int npages
= memslot
->npages
;
6801 int map_flags
= MAP_PRIVATE
| MAP_ANONYMOUS
;
6803 /* Prevent internal slot pages from being moved by fork()/COW. */
6804 if (memslot
->id
>= KVM_MEMORY_SLOTS
)
6805 map_flags
= MAP_SHARED
| MAP_ANONYMOUS
;
6807 /*To keep backward compatibility with older userspace,
6808 *x86 needs to handle !user_alloc case.
6811 if (npages
&& !old
.npages
) {
6812 unsigned long userspace_addr
;
6814 userspace_addr
= vm_mmap(NULL
, 0,
6816 PROT_READ
| PROT_WRITE
,
6820 if (IS_ERR((void *)userspace_addr
))
6821 return PTR_ERR((void *)userspace_addr
);
6823 memslot
->userspace_addr
= userspace_addr
;
6831 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
6832 struct kvm_userspace_memory_region
*mem
,
6833 struct kvm_memory_slot old
,
6837 int nr_mmu_pages
= 0, npages
= mem
->memory_size
>> PAGE_SHIFT
;
6839 if (!user_alloc
&& !old
.user_alloc
&& old
.npages
&& !npages
) {
6842 ret
= vm_munmap(old
.userspace_addr
,
6843 old
.npages
* PAGE_SIZE
);
6846 "kvm_vm_ioctl_set_memory_region: "
6847 "failed to munmap memory\n");
6850 if (!kvm
->arch
.n_requested_mmu_pages
)
6851 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
6853 spin_lock(&kvm
->mmu_lock
);
6855 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
6856 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
6857 spin_unlock(&kvm
->mmu_lock
);
6859 * If memory slot is created, or moved, we need to clear all
6862 if (npages
&& old
.base_gfn
!= mem
->guest_phys_addr
>> PAGE_SHIFT
) {
6863 kvm_mmu_zap_all(kvm
);
6864 kvm_reload_remote_mmus(kvm
);
6868 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
6870 kvm_mmu_zap_all(kvm
);
6871 kvm_reload_remote_mmus(kvm
);
6874 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
6875 struct kvm_memory_slot
*slot
)
6877 kvm_arch_flush_shadow_all(kvm
);
6880 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
6882 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
6883 !vcpu
->arch
.apf
.halted
)
6884 || !list_empty_careful(&vcpu
->async_pf
.done
)
6885 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
6886 || atomic_read(&vcpu
->arch
.nmi_queued
) ||
6887 (kvm_arch_interrupt_allowed(vcpu
) &&
6888 kvm_cpu_has_interrupt(vcpu
));
6891 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
6893 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
6896 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
6898 return kvm_x86_ops
->interrupt_allowed(vcpu
);
6901 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
6903 unsigned long current_rip
= kvm_rip_read(vcpu
) +
6904 get_segment_base(vcpu
, VCPU_SREG_CS
);
6906 return current_rip
== linear_rip
;
6908 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
6910 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
6912 unsigned long rflags
;
6914 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
6915 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
6916 rflags
&= ~X86_EFLAGS_TF
;
6919 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
6921 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
6923 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
6924 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
6925 rflags
|= X86_EFLAGS_TF
;
6926 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
6927 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6929 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
6931 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
6935 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
6936 is_error_page(work
->page
))
6939 r
= kvm_mmu_reload(vcpu
);
6943 if (!vcpu
->arch
.mmu
.direct_map
&&
6944 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
6947 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
6950 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
6952 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
6955 static inline u32
kvm_async_pf_next_probe(u32 key
)
6957 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
6960 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6962 u32 key
= kvm_async_pf_hash_fn(gfn
);
6964 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
6965 key
= kvm_async_pf_next_probe(key
);
6967 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
6970 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6973 u32 key
= kvm_async_pf_hash_fn(gfn
);
6975 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
6976 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
6977 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
6978 key
= kvm_async_pf_next_probe(key
);
6983 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6985 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
6988 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
6992 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
6994 vcpu
->arch
.apf
.gfns
[i
] = ~0;
6996 j
= kvm_async_pf_next_probe(j
);
6997 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
6999 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
7001 * k lies cyclically in ]i,j]
7003 * |....j i.k.| or |.k..j i...|
7005 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
7006 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
7011 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
7014 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
7018 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
7019 struct kvm_async_pf
*work
)
7021 struct x86_exception fault
;
7023 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
7024 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
7026 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
7027 (vcpu
->arch
.apf
.send_user_only
&&
7028 kvm_x86_ops
->get_cpl(vcpu
) == 0))
7029 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
7030 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
7031 fault
.vector
= PF_VECTOR
;
7032 fault
.error_code_valid
= true;
7033 fault
.error_code
= 0;
7034 fault
.nested_page_fault
= false;
7035 fault
.address
= work
->arch
.token
;
7036 kvm_inject_page_fault(vcpu
, &fault
);
7040 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
7041 struct kvm_async_pf
*work
)
7043 struct x86_exception fault
;
7045 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
7046 if (is_error_page(work
->page
))
7047 work
->arch
.token
= ~0; /* broadcast wakeup */
7049 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
7051 if ((vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) &&
7052 !apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
7053 fault
.vector
= PF_VECTOR
;
7054 fault
.error_code_valid
= true;
7055 fault
.error_code
= 0;
7056 fault
.nested_page_fault
= false;
7057 fault
.address
= work
->arch
.token
;
7058 kvm_inject_page_fault(vcpu
, &fault
);
7060 vcpu
->arch
.apf
.halted
= false;
7061 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7064 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
7066 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
7069 return !kvm_event_needs_reinjection(vcpu
) &&
7070 kvm_x86_ops
->interrupt_allowed(vcpu
);
7073 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
7074 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
7075 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
7076 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
7077 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
7078 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
7079 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
7080 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
7081 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
7082 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
7083 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
7084 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);