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 unsigned int min_timer_period_us
= 500;
98 module_param(min_timer_period_us
, uint
, S_IRUGO
| S_IWUSR
);
100 bool kvm_has_tsc_control
;
101 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
102 u32 kvm_max_guest_tsc_khz
;
103 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
105 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
106 static u32 tsc_tolerance_ppm
= 250;
107 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
109 #define KVM_NR_SHARED_MSRS 16
111 struct kvm_shared_msrs_global
{
113 u32 msrs
[KVM_NR_SHARED_MSRS
];
116 struct kvm_shared_msrs
{
117 struct user_return_notifier urn
;
119 struct kvm_shared_msr_values
{
122 } values
[KVM_NR_SHARED_MSRS
];
125 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
126 static struct kvm_shared_msrs __percpu
*shared_msrs
;
128 struct kvm_stats_debugfs_item debugfs_entries
[] = {
129 { "pf_fixed", VCPU_STAT(pf_fixed
) },
130 { "pf_guest", VCPU_STAT(pf_guest
) },
131 { "tlb_flush", VCPU_STAT(tlb_flush
) },
132 { "invlpg", VCPU_STAT(invlpg
) },
133 { "exits", VCPU_STAT(exits
) },
134 { "io_exits", VCPU_STAT(io_exits
) },
135 { "mmio_exits", VCPU_STAT(mmio_exits
) },
136 { "signal_exits", VCPU_STAT(signal_exits
) },
137 { "irq_window", VCPU_STAT(irq_window_exits
) },
138 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
139 { "halt_exits", VCPU_STAT(halt_exits
) },
140 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
141 { "hypercalls", VCPU_STAT(hypercalls
) },
142 { "request_irq", VCPU_STAT(request_irq_exits
) },
143 { "irq_exits", VCPU_STAT(irq_exits
) },
144 { "host_state_reload", VCPU_STAT(host_state_reload
) },
145 { "efer_reload", VCPU_STAT(efer_reload
) },
146 { "fpu_reload", VCPU_STAT(fpu_reload
) },
147 { "insn_emulation", VCPU_STAT(insn_emulation
) },
148 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
149 { "irq_injections", VCPU_STAT(irq_injections
) },
150 { "nmi_injections", VCPU_STAT(nmi_injections
) },
151 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
152 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
153 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
154 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
155 { "mmu_flooded", VM_STAT(mmu_flooded
) },
156 { "mmu_recycled", VM_STAT(mmu_recycled
) },
157 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
158 { "mmu_unsync", VM_STAT(mmu_unsync
) },
159 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
160 { "largepages", VM_STAT(lpages
) },
164 u64 __read_mostly host_xcr0
;
166 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
168 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
171 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
172 vcpu
->arch
.apf
.gfns
[i
] = ~0;
175 static void kvm_on_user_return(struct user_return_notifier
*urn
)
178 struct kvm_shared_msrs
*locals
179 = container_of(urn
, struct kvm_shared_msrs
, urn
);
180 struct kvm_shared_msr_values
*values
;
182 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
183 values
= &locals
->values
[slot
];
184 if (values
->host
!= values
->curr
) {
185 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
186 values
->curr
= values
->host
;
189 locals
->registered
= false;
190 user_return_notifier_unregister(urn
);
193 static void shared_msr_update(unsigned slot
, u32 msr
)
196 unsigned int cpu
= smp_processor_id();
197 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
199 /* only read, and nobody should modify it at this time,
200 * so don't need lock */
201 if (slot
>= shared_msrs_global
.nr
) {
202 printk(KERN_ERR
"kvm: invalid MSR slot!");
205 rdmsrl_safe(msr
, &value
);
206 smsr
->values
[slot
].host
= value
;
207 smsr
->values
[slot
].curr
= value
;
210 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
212 if (slot
>= shared_msrs_global
.nr
)
213 shared_msrs_global
.nr
= slot
+ 1;
214 shared_msrs_global
.msrs
[slot
] = msr
;
215 /* we need ensured the shared_msr_global have been updated */
218 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
220 static void kvm_shared_msr_cpu_online(void)
224 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
225 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
228 int kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
230 unsigned int cpu
= smp_processor_id();
231 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
234 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
236 smsr
->values
[slot
].curr
= value
;
237 err
= wrmsrl_safe(shared_msrs_global
.msrs
[slot
], value
);
241 if (!smsr
->registered
) {
242 smsr
->urn
.on_user_return
= kvm_on_user_return
;
243 user_return_notifier_register(&smsr
->urn
);
244 smsr
->registered
= true;
248 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
250 static void drop_user_return_notifiers(void *ignore
)
252 unsigned int cpu
= smp_processor_id();
253 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
255 if (smsr
->registered
)
256 kvm_on_user_return(&smsr
->urn
);
259 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
261 return vcpu
->arch
.apic_base
;
263 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
265 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
267 /* TODO: reserve bits check */
268 kvm_lapic_set_base(vcpu
, data
);
270 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
272 asmlinkage
void kvm_spurious_fault(void)
274 /* Fault while not rebooting. We want the trace. */
277 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
279 #define EXCPT_BENIGN 0
280 #define EXCPT_CONTRIBUTORY 1
283 static int exception_class(int vector
)
293 return EXCPT_CONTRIBUTORY
;
300 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
301 unsigned nr
, bool has_error
, u32 error_code
,
307 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
309 if (!vcpu
->arch
.exception
.pending
) {
311 vcpu
->arch
.exception
.pending
= true;
312 vcpu
->arch
.exception
.has_error_code
= has_error
;
313 vcpu
->arch
.exception
.nr
= nr
;
314 vcpu
->arch
.exception
.error_code
= error_code
;
315 vcpu
->arch
.exception
.reinject
= reinject
;
319 /* to check exception */
320 prev_nr
= vcpu
->arch
.exception
.nr
;
321 if (prev_nr
== DF_VECTOR
) {
322 /* triple fault -> shutdown */
323 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
326 class1
= exception_class(prev_nr
);
327 class2
= exception_class(nr
);
328 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
329 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
330 /* generate double fault per SDM Table 5-5 */
331 vcpu
->arch
.exception
.pending
= true;
332 vcpu
->arch
.exception
.has_error_code
= true;
333 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
334 vcpu
->arch
.exception
.error_code
= 0;
336 /* replace previous exception with a new one in a hope
337 that instruction re-execution will regenerate lost
342 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
344 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
346 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
348 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
350 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
352 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
354 void kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
357 kvm_inject_gp(vcpu
, 0);
359 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
361 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
363 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
365 ++vcpu
->stat
.pf_guest
;
366 vcpu
->arch
.cr2
= fault
->address
;
367 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
369 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
371 void kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
373 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
374 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
376 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
379 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
381 atomic_inc(&vcpu
->arch
.nmi_queued
);
382 kvm_make_request(KVM_REQ_NMI
, vcpu
);
384 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
386 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
388 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
390 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
392 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
394 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
396 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
399 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
400 * a #GP and return false.
402 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
404 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
406 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
409 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
412 * This function will be used to read from the physical memory of the currently
413 * running guest. The difference to kvm_read_guest_page is that this function
414 * can read from guest physical or from the guest's guest physical memory.
416 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
417 gfn_t ngfn
, void *data
, int offset
, int len
,
423 ngpa
= gfn_to_gpa(ngfn
);
424 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
);
425 if (real_gfn
== UNMAPPED_GVA
)
428 real_gfn
= gpa_to_gfn(real_gfn
);
430 return kvm_read_guest_page(vcpu
->kvm
, real_gfn
, data
, offset
, len
);
432 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
434 int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
435 void *data
, int offset
, int len
, u32 access
)
437 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
438 data
, offset
, len
, access
);
442 * Load the pae pdptrs. Return true is they are all valid.
444 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
446 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
447 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
450 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
452 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
453 offset
* sizeof(u64
), sizeof(pdpte
),
454 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
459 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
460 if (is_present_gpte(pdpte
[i
]) &&
461 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
468 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
469 __set_bit(VCPU_EXREG_PDPTR
,
470 (unsigned long *)&vcpu
->arch
.regs_avail
);
471 __set_bit(VCPU_EXREG_PDPTR
,
472 (unsigned long *)&vcpu
->arch
.regs_dirty
);
477 EXPORT_SYMBOL_GPL(load_pdptrs
);
479 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
481 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
487 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
490 if (!test_bit(VCPU_EXREG_PDPTR
,
491 (unsigned long *)&vcpu
->arch
.regs_avail
))
494 gfn
= (kvm_read_cr3(vcpu
) & ~31u) >> PAGE_SHIFT
;
495 offset
= (kvm_read_cr3(vcpu
) & ~31u) & (PAGE_SIZE
- 1);
496 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
497 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
500 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
506 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
508 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
509 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
|
510 X86_CR0_CD
| X86_CR0_NW
;
515 if (cr0
& 0xffffffff00000000UL
)
519 cr0
&= ~CR0_RESERVED_BITS
;
521 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
524 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
527 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
529 if ((vcpu
->arch
.efer
& EFER_LME
)) {
534 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
539 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
544 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
547 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
549 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
550 kvm_clear_async_pf_completion_queue(vcpu
);
551 kvm_async_pf_hash_reset(vcpu
);
554 if ((cr0
^ old_cr0
) & update_bits
)
555 kvm_mmu_reset_context(vcpu
);
558 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
560 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
562 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
564 EXPORT_SYMBOL_GPL(kvm_lmsw
);
566 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
568 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
569 !vcpu
->guest_xcr0_loaded
) {
570 /* kvm_set_xcr() also depends on this */
571 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
572 vcpu
->guest_xcr0_loaded
= 1;
576 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
578 if (vcpu
->guest_xcr0_loaded
) {
579 if (vcpu
->arch
.xcr0
!= host_xcr0
)
580 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
581 vcpu
->guest_xcr0_loaded
= 0;
585 int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
589 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
590 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
593 if (!(xcr0
& XSTATE_FP
))
595 if ((xcr0
& XSTATE_YMM
) && !(xcr0
& XSTATE_SSE
))
597 if (xcr0
& ~host_xcr0
)
599 kvm_put_guest_xcr0(vcpu
);
600 vcpu
->arch
.xcr0
= xcr0
;
604 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
606 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 ||
607 __kvm_set_xcr(vcpu
, index
, xcr
)) {
608 kvm_inject_gp(vcpu
, 0);
613 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
615 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
617 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
618 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
|
619 X86_CR4_PAE
| X86_CR4_SMEP
;
620 if (cr4
& CR4_RESERVED_BITS
)
623 if (!guest_cpuid_has_xsave(vcpu
) && (cr4
& X86_CR4_OSXSAVE
))
626 if (!guest_cpuid_has_smep(vcpu
) && (cr4
& X86_CR4_SMEP
))
629 if (!guest_cpuid_has_fsgsbase(vcpu
) && (cr4
& X86_CR4_FSGSBASE
))
632 if (is_long_mode(vcpu
)) {
633 if (!(cr4
& X86_CR4_PAE
))
635 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
636 && ((cr4
^ old_cr4
) & pdptr_bits
)
637 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
641 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
642 if (!guest_cpuid_has_pcid(vcpu
))
645 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
646 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
650 if (kvm_x86_ops
->set_cr4(vcpu
, cr4
))
653 if (((cr4
^ old_cr4
) & pdptr_bits
) ||
654 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
655 kvm_mmu_reset_context(vcpu
);
657 if ((cr4
^ old_cr4
) & X86_CR4_OSXSAVE
)
658 kvm_update_cpuid(vcpu
);
662 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
664 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
666 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
667 kvm_mmu_sync_roots(vcpu
);
668 kvm_mmu_flush_tlb(vcpu
);
672 if (is_long_mode(vcpu
)) {
673 if (kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
)) {
674 if (cr3
& CR3_PCID_ENABLED_RESERVED_BITS
)
677 if (cr3
& CR3_L_MODE_RESERVED_BITS
)
681 if (cr3
& CR3_PAE_RESERVED_BITS
)
683 if (is_paging(vcpu
) &&
684 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
688 * We don't check reserved bits in nonpae mode, because
689 * this isn't enforced, and VMware depends on this.
694 * Does the new cr3 value map to physical memory? (Note, we
695 * catch an invalid cr3 even in real-mode, because it would
696 * cause trouble later on when we turn on paging anyway.)
698 * A real CPU would silently accept an invalid cr3 and would
699 * attempt to use it - with largely undefined (and often hard
700 * to debug) behavior on the guest side.
702 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
704 vcpu
->arch
.cr3
= cr3
;
705 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
706 vcpu
->arch
.mmu
.new_cr3(vcpu
);
709 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
711 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
713 if (cr8
& CR8_RESERVED_BITS
)
715 if (irqchip_in_kernel(vcpu
->kvm
))
716 kvm_lapic_set_tpr(vcpu
, cr8
);
718 vcpu
->arch
.cr8
= cr8
;
721 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
723 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
725 if (irqchip_in_kernel(vcpu
->kvm
))
726 return kvm_lapic_get_cr8(vcpu
);
728 return vcpu
->arch
.cr8
;
730 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
732 static void kvm_update_dr7(struct kvm_vcpu
*vcpu
)
736 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
737 dr7
= vcpu
->arch
.guest_debug_dr7
;
739 dr7
= vcpu
->arch
.dr7
;
740 kvm_x86_ops
->set_dr7(vcpu
, dr7
);
741 vcpu
->arch
.switch_db_regs
= (dr7
& DR7_BP_EN_MASK
);
744 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
748 vcpu
->arch
.db
[dr
] = val
;
749 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
750 vcpu
->arch
.eff_db
[dr
] = val
;
753 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
757 if (val
& 0xffffffff00000000ULL
)
759 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | DR6_FIXED_1
;
762 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
766 if (val
& 0xffffffff00000000ULL
)
768 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
769 kvm_update_dr7(vcpu
);
776 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
780 res
= __kvm_set_dr(vcpu
, dr
, val
);
782 kvm_queue_exception(vcpu
, UD_VECTOR
);
784 kvm_inject_gp(vcpu
, 0);
788 EXPORT_SYMBOL_GPL(kvm_set_dr
);
790 static int _kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
794 *val
= vcpu
->arch
.db
[dr
];
797 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
801 *val
= vcpu
->arch
.dr6
;
804 if (kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
808 *val
= vcpu
->arch
.dr7
;
815 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
817 if (_kvm_get_dr(vcpu
, dr
, val
)) {
818 kvm_queue_exception(vcpu
, UD_VECTOR
);
823 EXPORT_SYMBOL_GPL(kvm_get_dr
);
825 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
827 u32 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
831 err
= kvm_pmu_read_pmc(vcpu
, ecx
, &data
);
834 kvm_register_write(vcpu
, VCPU_REGS_RAX
, (u32
)data
);
835 kvm_register_write(vcpu
, VCPU_REGS_RDX
, data
>> 32);
838 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
841 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
842 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
844 * This list is modified at module load time to reflect the
845 * capabilities of the host cpu. This capabilities test skips MSRs that are
846 * kvm-specific. Those are put in the beginning of the list.
849 #define KVM_SAVE_MSRS_BEGIN 10
850 static u32 msrs_to_save
[] = {
851 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
852 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
853 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
854 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
856 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
859 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
861 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
864 static unsigned num_msrs_to_save
;
866 static const u32 emulated_msrs
[] = {
868 MSR_IA32_TSCDEADLINE
,
869 MSR_IA32_MISC_ENABLE
,
874 bool kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
876 if (efer
& efer_reserved_bits
)
879 if (efer
& EFER_FFXSR
) {
880 struct kvm_cpuid_entry2
*feat
;
882 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
883 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
)))
887 if (efer
& EFER_SVME
) {
888 struct kvm_cpuid_entry2
*feat
;
890 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
891 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
)))
897 EXPORT_SYMBOL_GPL(kvm_valid_efer
);
899 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
901 u64 old_efer
= vcpu
->arch
.efer
;
903 if (!kvm_valid_efer(vcpu
, efer
))
907 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
911 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
913 kvm_x86_ops
->set_efer(vcpu
, efer
);
915 /* Update reserved bits */
916 if ((efer
^ old_efer
) & EFER_NX
)
917 kvm_mmu_reset_context(vcpu
);
922 void kvm_enable_efer_bits(u64 mask
)
924 efer_reserved_bits
&= ~mask
;
926 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
929 * Writes msr value into into the appropriate "register".
930 * Returns 0 on success, non-0 otherwise.
931 * Assumes vcpu_load() was already called.
933 int kvm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
935 switch (msr
->index
) {
938 case MSR_KERNEL_GS_BASE
:
941 if (is_noncanonical_address(msr
->data
))
944 case MSR_IA32_SYSENTER_EIP
:
945 case MSR_IA32_SYSENTER_ESP
:
947 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
948 * non-canonical address is written on Intel but not on
949 * AMD (which ignores the top 32-bits, because it does
950 * not implement 64-bit SYSENTER).
952 * 64-bit code should hence be able to write a non-canonical
953 * value on AMD. Making the address canonical ensures that
954 * vmentry does not fail on Intel after writing a non-canonical
955 * value, and that something deterministic happens if the guest
956 * invokes 64-bit SYSENTER.
958 msr
->data
= get_canonical(msr
->data
);
960 return kvm_x86_ops
->set_msr(vcpu
, msr
);
962 EXPORT_SYMBOL_GPL(kvm_set_msr
);
965 * Adapt set_msr() to msr_io()'s calling convention
967 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
973 msr
.host_initiated
= true;
974 return kvm_set_msr(vcpu
, &msr
);
978 struct pvclock_gtod_data
{
981 struct { /* extract of a clocksource struct */
989 /* open coded 'struct timespec' */
990 u64 monotonic_time_snsec
;
991 time_t monotonic_time_sec
;
994 static struct pvclock_gtod_data pvclock_gtod_data
;
996 static void update_pvclock_gtod(struct timekeeper
*tk
)
998 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
1000 write_seqcount_begin(&vdata
->seq
);
1002 /* copy pvclock gtod data */
1003 vdata
->clock
.vclock_mode
= tk
->clock
->archdata
.vclock_mode
;
1004 vdata
->clock
.cycle_last
= tk
->clock
->cycle_last
;
1005 vdata
->clock
.mask
= tk
->clock
->mask
;
1006 vdata
->clock
.mult
= tk
->mult
;
1007 vdata
->clock
.shift
= tk
->shift
;
1009 vdata
->monotonic_time_sec
= tk
->xtime_sec
1010 + tk
->wall_to_monotonic
.tv_sec
;
1011 vdata
->monotonic_time_snsec
= tk
->xtime_nsec
1012 + (tk
->wall_to_monotonic
.tv_nsec
1014 while (vdata
->monotonic_time_snsec
>=
1015 (((u64
)NSEC_PER_SEC
) << tk
->shift
)) {
1016 vdata
->monotonic_time_snsec
-=
1017 ((u64
)NSEC_PER_SEC
) << tk
->shift
;
1018 vdata
->monotonic_time_sec
++;
1021 write_seqcount_end(&vdata
->seq
);
1026 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
1030 struct pvclock_wall_clock wc
;
1031 struct timespec boot
;
1036 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1041 ++version
; /* first time write, random junk */
1045 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1048 * The guest calculates current wall clock time by adding
1049 * system time (updated by kvm_guest_time_update below) to the
1050 * wall clock specified here. guest system time equals host
1051 * system time for us, thus we must fill in host boot time here.
1055 if (kvm
->arch
.kvmclock_offset
) {
1056 struct timespec ts
= ns_to_timespec(kvm
->arch
.kvmclock_offset
);
1057 boot
= timespec_sub(boot
, ts
);
1059 wc
.sec
= boot
.tv_sec
;
1060 wc
.nsec
= boot
.tv_nsec
;
1061 wc
.version
= version
;
1063 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1066 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1069 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
1071 uint32_t quotient
, remainder
;
1073 /* Don't try to replace with do_div(), this one calculates
1074 * "(dividend << 32) / divisor" */
1076 : "=a" (quotient
), "=d" (remainder
)
1077 : "0" (0), "1" (dividend
), "r" (divisor
) );
1081 static void kvm_get_time_scale(uint32_t scaled_khz
, uint32_t base_khz
,
1082 s8
*pshift
, u32
*pmultiplier
)
1089 tps64
= base_khz
* 1000LL;
1090 scaled64
= scaled_khz
* 1000LL;
1091 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
1096 tps32
= (uint32_t)tps64
;
1097 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
1098 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
1106 *pmultiplier
= div_frac(scaled64
, tps32
);
1108 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1109 __func__
, base_khz
, scaled_khz
, shift
, *pmultiplier
);
1112 static inline u64
get_kernel_ns(void)
1116 WARN_ON(preemptible());
1118 monotonic_to_bootbased(&ts
);
1119 return timespec_to_ns(&ts
);
1122 #ifdef CONFIG_X86_64
1123 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
1126 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
1127 unsigned long max_tsc_khz
;
1129 static inline u64
nsec_to_cycles(struct kvm_vcpu
*vcpu
, u64 nsec
)
1131 return pvclock_scale_delta(nsec
, vcpu
->arch
.virtual_tsc_mult
,
1132 vcpu
->arch
.virtual_tsc_shift
);
1135 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1137 u64 v
= (u64
)khz
* (1000000 + ppm
);
1142 static void kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 this_tsc_khz
)
1144 u32 thresh_lo
, thresh_hi
;
1145 int use_scaling
= 0;
1147 /* tsc_khz can be zero if TSC calibration fails */
1148 if (this_tsc_khz
== 0)
1151 /* Compute a scale to convert nanoseconds in TSC cycles */
1152 kvm_get_time_scale(this_tsc_khz
, NSEC_PER_SEC
/ 1000,
1153 &vcpu
->arch
.virtual_tsc_shift
,
1154 &vcpu
->arch
.virtual_tsc_mult
);
1155 vcpu
->arch
.virtual_tsc_khz
= this_tsc_khz
;
1158 * Compute the variation in TSC rate which is acceptable
1159 * within the range of tolerance and decide if the
1160 * rate being applied is within that bounds of the hardware
1161 * rate. If so, no scaling or compensation need be done.
1163 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1164 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1165 if (this_tsc_khz
< thresh_lo
|| this_tsc_khz
> thresh_hi
) {
1166 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz
, thresh_lo
, thresh_hi
);
1169 kvm_x86_ops
->set_tsc_khz(vcpu
, this_tsc_khz
, use_scaling
);
1172 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1174 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1175 vcpu
->arch
.virtual_tsc_mult
,
1176 vcpu
->arch
.virtual_tsc_shift
);
1177 tsc
+= vcpu
->arch
.this_tsc_write
;
1181 void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
1183 #ifdef CONFIG_X86_64
1185 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1186 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1188 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1189 atomic_read(&vcpu
->kvm
->online_vcpus
));
1192 * Once the masterclock is enabled, always perform request in
1193 * order to update it.
1195 * In order to enable masterclock, the host clocksource must be TSC
1196 * and the vcpus need to have matched TSCs. When that happens,
1197 * perform request to enable masterclock.
1199 if (ka
->use_master_clock
||
1200 (gtod
->clock
.vclock_mode
== VCLOCK_TSC
&& vcpus_matched
))
1201 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1203 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
1204 atomic_read(&vcpu
->kvm
->online_vcpus
),
1205 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
1209 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu
*vcpu
, s64 offset
)
1211 u64 curr_offset
= kvm_x86_ops
->read_tsc_offset(vcpu
);
1212 vcpu
->arch
.ia32_tsc_adjust_msr
+= offset
- curr_offset
;
1215 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1217 struct kvm
*kvm
= vcpu
->kvm
;
1218 u64 offset
, ns
, elapsed
;
1219 unsigned long flags
;
1222 u64 data
= msr
->data
;
1224 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1225 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1226 ns
= get_kernel_ns();
1227 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1229 if (vcpu
->arch
.virtual_tsc_khz
) {
1232 /* n.b - signed multiplication and division required */
1233 usdiff
= data
- kvm
->arch
.last_tsc_write
;
1234 #ifdef CONFIG_X86_64
1235 usdiff
= (usdiff
* 1000) / vcpu
->arch
.virtual_tsc_khz
;
1237 /* do_div() only does unsigned */
1238 asm("1: idivl %[divisor]\n"
1239 "2: xor %%edx, %%edx\n"
1240 " movl $0, %[faulted]\n"
1242 ".section .fixup,\"ax\"\n"
1243 "4: movl $1, %[faulted]\n"
1247 _ASM_EXTABLE(1b
, 4b
)
1249 : "=A"(usdiff
), [faulted
] "=r" (faulted
)
1250 : "A"(usdiff
* 1000), [divisor
] "rm"(vcpu
->arch
.virtual_tsc_khz
));
1253 do_div(elapsed
, 1000);
1258 /* idivl overflow => difference is larger than USEC_PER_SEC */
1260 usdiff
= USEC_PER_SEC
;
1262 usdiff
= USEC_PER_SEC
; /* disable TSC match window below */
1265 * Special case: TSC write with a small delta (1 second) of virtual
1266 * cycle time against real time is interpreted as an attempt to
1267 * synchronize the CPU.
1269 * For a reliable TSC, we can match TSC offsets, and for an unstable
1270 * TSC, we add elapsed time in this computation. We could let the
1271 * compensation code attempt to catch up if we fall behind, but
1272 * it's better to try to match offsets from the beginning.
1274 if (usdiff
< USEC_PER_SEC
&&
1275 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1276 if (!check_tsc_unstable()) {
1277 offset
= kvm
->arch
.cur_tsc_offset
;
1278 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1280 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1282 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
, data
);
1283 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1288 * We split periods of matched TSC writes into generations.
1289 * For each generation, we track the original measured
1290 * nanosecond time, offset, and write, so if TSCs are in
1291 * sync, we can match exact offset, and if not, we can match
1292 * exact software computation in compute_guest_tsc()
1294 * These values are tracked in kvm->arch.cur_xxx variables.
1296 kvm
->arch
.cur_tsc_generation
++;
1297 kvm
->arch
.cur_tsc_nsec
= ns
;
1298 kvm
->arch
.cur_tsc_write
= data
;
1299 kvm
->arch
.cur_tsc_offset
= offset
;
1301 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1302 kvm
->arch
.cur_tsc_generation
, data
);
1306 * We also track th most recent recorded KHZ, write and time to
1307 * allow the matching interval to be extended at each write.
1309 kvm
->arch
.last_tsc_nsec
= ns
;
1310 kvm
->arch
.last_tsc_write
= data
;
1311 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1313 /* Reset of TSC must disable overshoot protection below */
1314 vcpu
->arch
.hv_clock
.tsc_timestamp
= 0;
1315 vcpu
->arch
.last_guest_tsc
= data
;
1317 /* Keep track of which generation this VCPU has synchronized to */
1318 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1319 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1320 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1322 if (guest_cpuid_has_tsc_adjust(vcpu
) && !msr
->host_initiated
)
1323 update_ia32_tsc_adjust_msr(vcpu
, offset
);
1324 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1325 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1327 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1329 kvm
->arch
.nr_vcpus_matched_tsc
++;
1331 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
1333 kvm_track_tsc_matching(vcpu
);
1334 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1337 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1339 #ifdef CONFIG_X86_64
1341 static cycle_t
read_tsc(void)
1347 * Empirically, a fence (of type that depends on the CPU)
1348 * before rdtsc is enough to ensure that rdtsc is ordered
1349 * with respect to loads. The various CPU manuals are unclear
1350 * as to whether rdtsc can be reordered with later loads,
1351 * but no one has ever seen it happen.
1354 ret
= (cycle_t
)vget_cycles();
1356 last
= pvclock_gtod_data
.clock
.cycle_last
;
1358 if (likely(ret
>= last
))
1362 * GCC likes to generate cmov here, but this branch is extremely
1363 * predictable (it's just a funciton of time and the likely is
1364 * very likely) and there's a data dependence, so force GCC
1365 * to generate a branch instead. I don't barrier() because
1366 * we don't actually need a barrier, and if this function
1367 * ever gets inlined it will generate worse code.
1373 static inline u64
vgettsc(cycle_t
*cycle_now
)
1376 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1378 *cycle_now
= read_tsc();
1380 v
= (*cycle_now
- gtod
->clock
.cycle_last
) & gtod
->clock
.mask
;
1381 return v
* gtod
->clock
.mult
;
1384 static int do_monotonic(struct timespec
*ts
, cycle_t
*cycle_now
)
1389 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1393 seq
= read_seqcount_begin(>od
->seq
);
1394 mode
= gtod
->clock
.vclock_mode
;
1395 ts
->tv_sec
= gtod
->monotonic_time_sec
;
1396 ns
= gtod
->monotonic_time_snsec
;
1397 ns
+= vgettsc(cycle_now
);
1398 ns
>>= gtod
->clock
.shift
;
1399 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1400 timespec_add_ns(ts
, ns
);
1405 /* returns true if host is using tsc clocksource */
1406 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, cycle_t
*cycle_now
)
1410 /* checked again under seqlock below */
1411 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1414 if (do_monotonic(&ts
, cycle_now
) != VCLOCK_TSC
)
1417 monotonic_to_bootbased(&ts
);
1418 *kernel_ns
= timespec_to_ns(&ts
);
1426 * Assuming a stable TSC across physical CPUS, and a stable TSC
1427 * across virtual CPUs, the following condition is possible.
1428 * Each numbered line represents an event visible to both
1429 * CPUs at the next numbered event.
1431 * "timespecX" represents host monotonic time. "tscX" represents
1434 * VCPU0 on CPU0 | VCPU1 on CPU1
1436 * 1. read timespec0,tsc0
1437 * 2. | timespec1 = timespec0 + N
1439 * 3. transition to guest | transition to guest
1440 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1441 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1442 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1444 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1447 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1449 * - 0 < N - M => M < N
1451 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1452 * always the case (the difference between two distinct xtime instances
1453 * might be smaller then the difference between corresponding TSC reads,
1454 * when updating guest vcpus pvclock areas).
1456 * To avoid that problem, do not allow visibility of distinct
1457 * system_timestamp/tsc_timestamp values simultaneously: use a master
1458 * copy of host monotonic time values. Update that master copy
1461 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1465 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
1467 #ifdef CONFIG_X86_64
1468 struct kvm_arch
*ka
= &kvm
->arch
;
1470 bool host_tsc_clocksource
, vcpus_matched
;
1472 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1473 atomic_read(&kvm
->online_vcpus
));
1476 * If the host uses TSC clock, then passthrough TSC as stable
1479 host_tsc_clocksource
= kvm_get_time_and_clockread(
1480 &ka
->master_kernel_ns
,
1481 &ka
->master_cycle_now
);
1483 ka
->use_master_clock
= host_tsc_clocksource
& vcpus_matched
;
1485 if (ka
->use_master_clock
)
1486 atomic_set(&kvm_guest_has_master_clock
, 1);
1488 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
1489 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
1494 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1496 unsigned long flags
, this_tsc_khz
;
1497 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1498 struct kvm_arch
*ka
= &v
->kvm
->arch
;
1499 s64 kernel_ns
, max_kernel_ns
;
1500 u64 tsc_timestamp
, host_tsc
;
1501 struct pvclock_vcpu_time_info guest_hv_clock
;
1503 bool use_master_clock
;
1509 * If the host uses TSC clock, then passthrough TSC as stable
1512 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1513 use_master_clock
= ka
->use_master_clock
;
1514 if (use_master_clock
) {
1515 host_tsc
= ka
->master_cycle_now
;
1516 kernel_ns
= ka
->master_kernel_ns
;
1518 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1520 /* Keep irq disabled to prevent changes to the clock */
1521 local_irq_save(flags
);
1522 this_tsc_khz
= __get_cpu_var(cpu_tsc_khz
);
1523 if (unlikely(this_tsc_khz
== 0)) {
1524 local_irq_restore(flags
);
1525 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1528 if (!use_master_clock
) {
1529 host_tsc
= native_read_tsc();
1530 kernel_ns
= get_kernel_ns();
1533 tsc_timestamp
= kvm_x86_ops
->read_l1_tsc(v
, host_tsc
);
1536 * We may have to catch up the TSC to match elapsed wall clock
1537 * time for two reasons, even if kvmclock is used.
1538 * 1) CPU could have been running below the maximum TSC rate
1539 * 2) Broken TSC compensation resets the base at each VCPU
1540 * entry to avoid unknown leaps of TSC even when running
1541 * again on the same CPU. This may cause apparent elapsed
1542 * time to disappear, and the guest to stand still or run
1545 if (vcpu
->tsc_catchup
) {
1546 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1547 if (tsc
> tsc_timestamp
) {
1548 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1549 tsc_timestamp
= tsc
;
1553 local_irq_restore(flags
);
1555 if (!vcpu
->pv_time_enabled
)
1559 * Time as measured by the TSC may go backwards when resetting the base
1560 * tsc_timestamp. The reason for this is that the TSC resolution is
1561 * higher than the resolution of the other clock scales. Thus, many
1562 * possible measurments of the TSC correspond to one measurement of any
1563 * other clock, and so a spread of values is possible. This is not a
1564 * problem for the computation of the nanosecond clock; with TSC rates
1565 * around 1GHZ, there can only be a few cycles which correspond to one
1566 * nanosecond value, and any path through this code will inevitably
1567 * take longer than that. However, with the kernel_ns value itself,
1568 * the precision may be much lower, down to HZ granularity. If the
1569 * first sampling of TSC against kernel_ns ends in the low part of the
1570 * range, and the second in the high end of the range, we can get:
1572 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1574 * As the sampling errors potentially range in the thousands of cycles,
1575 * it is possible such a time value has already been observed by the
1576 * guest. To protect against this, we must compute the system time as
1577 * observed by the guest and ensure the new system time is greater.
1580 if (vcpu
->hv_clock
.tsc_timestamp
) {
1581 max_kernel_ns
= vcpu
->last_guest_tsc
-
1582 vcpu
->hv_clock
.tsc_timestamp
;
1583 max_kernel_ns
= pvclock_scale_delta(max_kernel_ns
,
1584 vcpu
->hv_clock
.tsc_to_system_mul
,
1585 vcpu
->hv_clock
.tsc_shift
);
1586 max_kernel_ns
+= vcpu
->last_kernel_ns
;
1589 if (unlikely(vcpu
->hw_tsc_khz
!= this_tsc_khz
)) {
1590 kvm_get_time_scale(NSEC_PER_SEC
/ 1000, this_tsc_khz
,
1591 &vcpu
->hv_clock
.tsc_shift
,
1592 &vcpu
->hv_clock
.tsc_to_system_mul
);
1593 vcpu
->hw_tsc_khz
= this_tsc_khz
;
1596 /* with a master <monotonic time, tsc value> tuple,
1597 * pvclock clock reads always increase at the (scaled) rate
1598 * of guest TSC - no need to deal with sampling errors.
1600 if (!use_master_clock
) {
1601 if (max_kernel_ns
> kernel_ns
)
1602 kernel_ns
= max_kernel_ns
;
1604 /* With all the info we got, fill in the values */
1605 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
1606 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
1607 vcpu
->last_kernel_ns
= kernel_ns
;
1608 vcpu
->last_guest_tsc
= tsc_timestamp
;
1611 * The interface expects us to write an even number signaling that the
1612 * update is finished. Since the guest won't see the intermediate
1613 * state, we just increase by 2 at the end.
1615 vcpu
->hv_clock
.version
+= 2;
1617 if (unlikely(kvm_read_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1618 &guest_hv_clock
, sizeof(guest_hv_clock
))))
1621 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1622 pvclock_flags
= (guest_hv_clock
.flags
& PVCLOCK_GUEST_STOPPED
);
1624 if (vcpu
->pvclock_set_guest_stopped_request
) {
1625 pvclock_flags
|= PVCLOCK_GUEST_STOPPED
;
1626 vcpu
->pvclock_set_guest_stopped_request
= false;
1629 /* If the host uses TSC clocksource, then it is stable */
1630 if (use_master_clock
)
1631 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
1633 vcpu
->hv_clock
.flags
= pvclock_flags
;
1635 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1637 sizeof(vcpu
->hv_clock
));
1641 static bool msr_mtrr_valid(unsigned msr
)
1644 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
1645 case MSR_MTRRfix64K_00000
:
1646 case MSR_MTRRfix16K_80000
:
1647 case MSR_MTRRfix16K_A0000
:
1648 case MSR_MTRRfix4K_C0000
:
1649 case MSR_MTRRfix4K_C8000
:
1650 case MSR_MTRRfix4K_D0000
:
1651 case MSR_MTRRfix4K_D8000
:
1652 case MSR_MTRRfix4K_E0000
:
1653 case MSR_MTRRfix4K_E8000
:
1654 case MSR_MTRRfix4K_F0000
:
1655 case MSR_MTRRfix4K_F8000
:
1656 case MSR_MTRRdefType
:
1657 case MSR_IA32_CR_PAT
:
1665 static bool valid_pat_type(unsigned t
)
1667 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1670 static bool valid_mtrr_type(unsigned t
)
1672 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
1675 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1679 if (!msr_mtrr_valid(msr
))
1682 if (msr
== MSR_IA32_CR_PAT
) {
1683 for (i
= 0; i
< 8; i
++)
1684 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
1687 } else if (msr
== MSR_MTRRdefType
) {
1690 return valid_mtrr_type(data
& 0xff);
1691 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
1692 for (i
= 0; i
< 8 ; i
++)
1693 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
1698 /* variable MTRRs */
1699 return valid_mtrr_type(data
& 0xff);
1702 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1704 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1706 if (!mtrr_valid(vcpu
, msr
, data
))
1709 if (msr
== MSR_MTRRdefType
) {
1710 vcpu
->arch
.mtrr_state
.def_type
= data
;
1711 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
1712 } else if (msr
== MSR_MTRRfix64K_00000
)
1714 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1715 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
1716 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1717 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
1718 else if (msr
== MSR_IA32_CR_PAT
)
1719 vcpu
->arch
.pat
= data
;
1720 else { /* Variable MTRRs */
1721 int idx
, is_mtrr_mask
;
1724 idx
= (msr
- 0x200) / 2;
1725 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1728 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1731 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1735 kvm_mmu_reset_context(vcpu
);
1739 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1741 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1742 unsigned bank_num
= mcg_cap
& 0xff;
1745 case MSR_IA32_MCG_STATUS
:
1746 vcpu
->arch
.mcg_status
= data
;
1748 case MSR_IA32_MCG_CTL
:
1749 if (!(mcg_cap
& MCG_CTL_P
))
1751 if (data
!= 0 && data
!= ~(u64
)0)
1753 vcpu
->arch
.mcg_ctl
= data
;
1756 if (msr
>= MSR_IA32_MC0_CTL
&&
1757 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1758 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1759 /* only 0 or all 1s can be written to IA32_MCi_CTL
1760 * some Linux kernels though clear bit 10 in bank 4 to
1761 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1762 * this to avoid an uncatched #GP in the guest
1764 if ((offset
& 0x3) == 0 &&
1765 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
1767 vcpu
->arch
.mce_banks
[offset
] = data
;
1775 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
1777 struct kvm
*kvm
= vcpu
->kvm
;
1778 int lm
= is_long_mode(vcpu
);
1779 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
1780 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
1781 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
1782 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
1783 u32 page_num
= data
& ~PAGE_MASK
;
1784 u64 page_addr
= data
& PAGE_MASK
;
1789 if (page_num
>= blob_size
)
1792 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
1797 if (kvm_write_guest(kvm
, page_addr
, page
, PAGE_SIZE
))
1806 static bool kvm_hv_hypercall_enabled(struct kvm
*kvm
)
1808 return kvm
->arch
.hv_hypercall
& HV_X64_MSR_HYPERCALL_ENABLE
;
1811 static bool kvm_hv_msr_partition_wide(u32 msr
)
1815 case HV_X64_MSR_GUEST_OS_ID
:
1816 case HV_X64_MSR_HYPERCALL
:
1824 static int set_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1826 struct kvm
*kvm
= vcpu
->kvm
;
1829 case HV_X64_MSR_GUEST_OS_ID
:
1830 kvm
->arch
.hv_guest_os_id
= data
;
1831 /* setting guest os id to zero disables hypercall page */
1832 if (!kvm
->arch
.hv_guest_os_id
)
1833 kvm
->arch
.hv_hypercall
&= ~HV_X64_MSR_HYPERCALL_ENABLE
;
1835 case HV_X64_MSR_HYPERCALL
: {
1840 /* if guest os id is not set hypercall should remain disabled */
1841 if (!kvm
->arch
.hv_guest_os_id
)
1843 if (!(data
& HV_X64_MSR_HYPERCALL_ENABLE
)) {
1844 kvm
->arch
.hv_hypercall
= data
;
1847 gfn
= data
>> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT
;
1848 addr
= gfn_to_hva(kvm
, gfn
);
1849 if (kvm_is_error_hva(addr
))
1851 kvm_x86_ops
->patch_hypercall(vcpu
, instructions
);
1852 ((unsigned char *)instructions
)[3] = 0xc3; /* ret */
1853 if (__copy_to_user((void __user
*)addr
, instructions
, 4))
1855 kvm
->arch
.hv_hypercall
= data
;
1859 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1860 "data 0x%llx\n", msr
, data
);
1866 static int set_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1869 case HV_X64_MSR_APIC_ASSIST_PAGE
: {
1872 if (!(data
& HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE
)) {
1873 vcpu
->arch
.hv_vapic
= data
;
1876 addr
= gfn_to_hva(vcpu
->kvm
, data
>>
1877 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT
);
1878 if (kvm_is_error_hva(addr
))
1880 if (__clear_user((void __user
*)addr
, PAGE_SIZE
))
1882 vcpu
->arch
.hv_vapic
= data
;
1885 case HV_X64_MSR_EOI
:
1886 return kvm_hv_vapic_msr_write(vcpu
, APIC_EOI
, data
);
1887 case HV_X64_MSR_ICR
:
1888 return kvm_hv_vapic_msr_write(vcpu
, APIC_ICR
, data
);
1889 case HV_X64_MSR_TPR
:
1890 return kvm_hv_vapic_msr_write(vcpu
, APIC_TASKPRI
, data
);
1892 vcpu_unimpl(vcpu
, "HYPER-V unimplemented wrmsr: 0x%x "
1893 "data 0x%llx\n", msr
, data
);
1900 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
1902 gpa_t gpa
= data
& ~0x3f;
1904 /* Bits 2:5 are reserved, Should be zero */
1908 vcpu
->arch
.apf
.msr_val
= data
;
1910 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
1911 kvm_clear_async_pf_completion_queue(vcpu
);
1912 kvm_async_pf_hash_reset(vcpu
);
1916 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
,
1920 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
1921 kvm_async_pf_wakeup_all(vcpu
);
1925 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
1927 vcpu
->arch
.pv_time_enabled
= false;
1930 static void accumulate_steal_time(struct kvm_vcpu
*vcpu
)
1934 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1937 delta
= current
->sched_info
.run_delay
- vcpu
->arch
.st
.last_steal
;
1938 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
1939 vcpu
->arch
.st
.accum_steal
= delta
;
1942 static void record_steal_time(struct kvm_vcpu
*vcpu
)
1944 accumulate_steal_time(vcpu
);
1946 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
1949 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1950 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
1953 vcpu
->arch
.st
.steal
.steal
+= vcpu
->arch
.st
.accum_steal
;
1954 vcpu
->arch
.st
.steal
.version
+= 2;
1955 vcpu
->arch
.st
.accum_steal
= 0;
1957 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
1958 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
1961 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
1964 u32 msr
= msr_info
->index
;
1965 u64 data
= msr_info
->data
;
1968 case MSR_AMD64_NB_CFG
:
1969 case MSR_IA32_UCODE_REV
:
1970 case MSR_IA32_UCODE_WRITE
:
1971 case MSR_VM_HSAVE_PA
:
1972 case MSR_AMD64_PATCH_LOADER
:
1973 case MSR_AMD64_BU_CFG2
:
1977 return set_efer(vcpu
, data
);
1979 data
&= ~(u64
)0x40; /* ignore flush filter disable */
1980 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
1981 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
1983 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
1988 case MSR_FAM10H_MMIO_CONF_BASE
:
1990 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
1995 case MSR_IA32_DEBUGCTLMSR
:
1997 /* We support the non-activated case already */
1999 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
2000 /* Values other than LBR and BTF are vendor-specific,
2001 thus reserved and should throw a #GP */
2004 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2007 case 0x200 ... 0x2ff:
2008 return set_msr_mtrr(vcpu
, msr
, data
);
2009 case MSR_IA32_APICBASE
:
2010 kvm_set_apic_base(vcpu
, data
);
2012 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2013 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
2014 case MSR_IA32_TSCDEADLINE
:
2015 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
2017 case MSR_IA32_TSC_ADJUST
:
2018 if (guest_cpuid_has_tsc_adjust(vcpu
)) {
2019 if (!msr_info
->host_initiated
) {
2020 u64 adj
= data
- vcpu
->arch
.ia32_tsc_adjust_msr
;
2021 kvm_x86_ops
->adjust_tsc_offset(vcpu
, adj
, true);
2023 vcpu
->arch
.ia32_tsc_adjust_msr
= data
;
2026 case MSR_IA32_MISC_ENABLE
:
2027 vcpu
->arch
.ia32_misc_enable_msr
= data
;
2029 case MSR_KVM_WALL_CLOCK_NEW
:
2030 case MSR_KVM_WALL_CLOCK
:
2031 vcpu
->kvm
->arch
.wall_clock
= data
;
2032 kvm_write_wall_clock(vcpu
->kvm
, data
);
2034 case MSR_KVM_SYSTEM_TIME_NEW
:
2035 case MSR_KVM_SYSTEM_TIME
: {
2037 kvmclock_reset(vcpu
);
2039 vcpu
->arch
.time
= data
;
2040 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2042 /* we verify if the enable bit is set... */
2046 gpa_offset
= data
& ~(PAGE_MASK
| 1);
2048 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2049 &vcpu
->arch
.pv_time
, data
& ~1ULL,
2050 sizeof(struct pvclock_vcpu_time_info
)))
2051 vcpu
->arch
.pv_time_enabled
= false;
2053 vcpu
->arch
.pv_time_enabled
= true;
2057 case MSR_KVM_ASYNC_PF_EN
:
2058 if (kvm_pv_enable_async_pf(vcpu
, data
))
2061 case MSR_KVM_STEAL_TIME
:
2063 if (unlikely(!sched_info_on()))
2066 if (data
& KVM_STEAL_RESERVED_MASK
)
2069 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2070 data
& KVM_STEAL_VALID_BITS
,
2071 sizeof(struct kvm_steal_time
)))
2074 vcpu
->arch
.st
.msr_val
= data
;
2076 if (!(data
& KVM_MSR_ENABLED
))
2079 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2082 case MSR_KVM_PV_EOI_EN
:
2083 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
2087 case MSR_IA32_MCG_CTL
:
2088 case MSR_IA32_MCG_STATUS
:
2089 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
2090 return set_msr_mce(vcpu
, msr
, data
);
2092 /* Performance counters are not protected by a CPUID bit,
2093 * so we should check all of them in the generic path for the sake of
2094 * cross vendor migration.
2095 * Writing a zero into the event select MSRs disables them,
2096 * which we perfectly emulate ;-). Any other value should be at least
2097 * reported, some guests depend on them.
2099 case MSR_K7_EVNTSEL0
:
2100 case MSR_K7_EVNTSEL1
:
2101 case MSR_K7_EVNTSEL2
:
2102 case MSR_K7_EVNTSEL3
:
2104 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
2105 "0x%x data 0x%llx\n", msr
, data
);
2107 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2108 * so we ignore writes to make it happy.
2110 case MSR_K7_PERFCTR0
:
2111 case MSR_K7_PERFCTR1
:
2112 case MSR_K7_PERFCTR2
:
2113 case MSR_K7_PERFCTR3
:
2114 vcpu_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
2115 "0x%x data 0x%llx\n", msr
, data
);
2117 case MSR_P6_PERFCTR0
:
2118 case MSR_P6_PERFCTR1
:
2120 case MSR_P6_EVNTSEL0
:
2121 case MSR_P6_EVNTSEL1
:
2122 if (kvm_pmu_msr(vcpu
, msr
))
2123 return kvm_pmu_set_msr(vcpu
, msr_info
);
2125 if (pr
|| data
!= 0)
2126 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
2127 "0x%x data 0x%llx\n", msr
, data
);
2129 case MSR_K7_CLK_CTL
:
2131 * Ignore all writes to this no longer documented MSR.
2132 * Writes are only relevant for old K7 processors,
2133 * all pre-dating SVM, but a recommended workaround from
2134 * AMD for these chips. It is possible to specify the
2135 * affected processor models on the command line, hence
2136 * the need to ignore the workaround.
2139 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2140 if (kvm_hv_msr_partition_wide(msr
)) {
2142 mutex_lock(&vcpu
->kvm
->lock
);
2143 r
= set_msr_hyperv_pw(vcpu
, msr
, data
);
2144 mutex_unlock(&vcpu
->kvm
->lock
);
2147 return set_msr_hyperv(vcpu
, msr
, data
);
2149 case MSR_IA32_BBL_CR_CTL3
:
2150 /* Drop writes to this legacy MSR -- see rdmsr
2151 * counterpart for further detail.
2153 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n", msr
, data
);
2155 case MSR_AMD64_OSVW_ID_LENGTH
:
2156 if (!guest_cpuid_has_osvw(vcpu
))
2158 vcpu
->arch
.osvw
.length
= data
;
2160 case MSR_AMD64_OSVW_STATUS
:
2161 if (!guest_cpuid_has_osvw(vcpu
))
2163 vcpu
->arch
.osvw
.status
= data
;
2166 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
2167 return xen_hvm_config(vcpu
, data
);
2168 if (kvm_pmu_msr(vcpu
, msr
))
2169 return kvm_pmu_set_msr(vcpu
, msr_info
);
2171 vcpu_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
2175 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
2182 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
2186 * Reads an msr value (of 'msr_index') into 'pdata'.
2187 * Returns 0 on success, non-0 otherwise.
2188 * Assumes vcpu_load() was already called.
2190 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
2192 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
2195 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2197 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
2199 if (!msr_mtrr_valid(msr
))
2202 if (msr
== MSR_MTRRdefType
)
2203 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
2204 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
2205 else if (msr
== MSR_MTRRfix64K_00000
)
2207 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
2208 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
2209 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
2210 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
2211 else if (msr
== MSR_IA32_CR_PAT
)
2212 *pdata
= vcpu
->arch
.pat
;
2213 else { /* Variable MTRRs */
2214 int idx
, is_mtrr_mask
;
2217 idx
= (msr
- 0x200) / 2;
2218 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
2221 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
2224 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
2231 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2234 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2235 unsigned bank_num
= mcg_cap
& 0xff;
2238 case MSR_IA32_P5_MC_ADDR
:
2239 case MSR_IA32_P5_MC_TYPE
:
2242 case MSR_IA32_MCG_CAP
:
2243 data
= vcpu
->arch
.mcg_cap
;
2245 case MSR_IA32_MCG_CTL
:
2246 if (!(mcg_cap
& MCG_CTL_P
))
2248 data
= vcpu
->arch
.mcg_ctl
;
2250 case MSR_IA32_MCG_STATUS
:
2251 data
= vcpu
->arch
.mcg_status
;
2254 if (msr
>= MSR_IA32_MC0_CTL
&&
2255 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
2256 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2257 data
= vcpu
->arch
.mce_banks
[offset
];
2266 static int get_msr_hyperv_pw(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2269 struct kvm
*kvm
= vcpu
->kvm
;
2272 case HV_X64_MSR_GUEST_OS_ID
:
2273 data
= kvm
->arch
.hv_guest_os_id
;
2275 case HV_X64_MSR_HYPERCALL
:
2276 data
= kvm
->arch
.hv_hypercall
;
2279 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
2287 static int get_msr_hyperv(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2292 case HV_X64_MSR_VP_INDEX
: {
2295 kvm_for_each_vcpu(r
, v
, vcpu
->kvm
)
2300 case HV_X64_MSR_EOI
:
2301 return kvm_hv_vapic_msr_read(vcpu
, APIC_EOI
, pdata
);
2302 case HV_X64_MSR_ICR
:
2303 return kvm_hv_vapic_msr_read(vcpu
, APIC_ICR
, pdata
);
2304 case HV_X64_MSR_TPR
:
2305 return kvm_hv_vapic_msr_read(vcpu
, APIC_TASKPRI
, pdata
);
2306 case HV_X64_MSR_APIC_ASSIST_PAGE
:
2307 data
= vcpu
->arch
.hv_vapic
;
2310 vcpu_unimpl(vcpu
, "Hyper-V unhandled rdmsr: 0x%x\n", msr
);
2317 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2322 case MSR_IA32_PLATFORM_ID
:
2323 case MSR_IA32_EBL_CR_POWERON
:
2324 case MSR_IA32_DEBUGCTLMSR
:
2325 case MSR_IA32_LASTBRANCHFROMIP
:
2326 case MSR_IA32_LASTBRANCHTOIP
:
2327 case MSR_IA32_LASTINTFROMIP
:
2328 case MSR_IA32_LASTINTTOIP
:
2331 case MSR_VM_HSAVE_PA
:
2332 case MSR_K7_EVNTSEL0
:
2333 case MSR_K7_PERFCTR0
:
2334 case MSR_K8_INT_PENDING_MSG
:
2335 case MSR_AMD64_NB_CFG
:
2336 case MSR_FAM10H_MMIO_CONF_BASE
:
2337 case MSR_AMD64_BU_CFG2
:
2340 case MSR_P6_PERFCTR0
:
2341 case MSR_P6_PERFCTR1
:
2342 case MSR_P6_EVNTSEL0
:
2343 case MSR_P6_EVNTSEL1
:
2344 if (kvm_pmu_msr(vcpu
, msr
))
2345 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
2348 case MSR_IA32_UCODE_REV
:
2349 data
= 0x100000000ULL
;
2352 data
= 0x500 | KVM_NR_VAR_MTRR
;
2354 case 0x200 ... 0x2ff:
2355 return get_msr_mtrr(vcpu
, msr
, pdata
);
2356 case 0xcd: /* fsb frequency */
2360 * MSR_EBC_FREQUENCY_ID
2361 * Conservative value valid for even the basic CPU models.
2362 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2363 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2364 * and 266MHz for model 3, or 4. Set Core Clock
2365 * Frequency to System Bus Frequency Ratio to 1 (bits
2366 * 31:24) even though these are only valid for CPU
2367 * models > 2, however guests may end up dividing or
2368 * multiplying by zero otherwise.
2370 case MSR_EBC_FREQUENCY_ID
:
2373 case MSR_IA32_APICBASE
:
2374 data
= kvm_get_apic_base(vcpu
);
2376 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2377 return kvm_x2apic_msr_read(vcpu
, msr
, pdata
);
2379 case MSR_IA32_TSCDEADLINE
:
2380 data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
2382 case MSR_IA32_TSC_ADJUST
:
2383 data
= (u64
)vcpu
->arch
.ia32_tsc_adjust_msr
;
2385 case MSR_IA32_MISC_ENABLE
:
2386 data
= vcpu
->arch
.ia32_misc_enable_msr
;
2388 case MSR_IA32_PERF_STATUS
:
2389 /* TSC increment by tick */
2391 /* CPU multiplier */
2392 data
|= (((uint64_t)4ULL) << 40);
2395 data
= vcpu
->arch
.efer
;
2397 case MSR_KVM_WALL_CLOCK
:
2398 case MSR_KVM_WALL_CLOCK_NEW
:
2399 data
= vcpu
->kvm
->arch
.wall_clock
;
2401 case MSR_KVM_SYSTEM_TIME
:
2402 case MSR_KVM_SYSTEM_TIME_NEW
:
2403 data
= vcpu
->arch
.time
;
2405 case MSR_KVM_ASYNC_PF_EN
:
2406 data
= vcpu
->arch
.apf
.msr_val
;
2408 case MSR_KVM_STEAL_TIME
:
2409 data
= vcpu
->arch
.st
.msr_val
;
2411 case MSR_KVM_PV_EOI_EN
:
2412 data
= vcpu
->arch
.pv_eoi
.msr_val
;
2414 case MSR_IA32_P5_MC_ADDR
:
2415 case MSR_IA32_P5_MC_TYPE
:
2416 case MSR_IA32_MCG_CAP
:
2417 case MSR_IA32_MCG_CTL
:
2418 case MSR_IA32_MCG_STATUS
:
2419 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
2420 return get_msr_mce(vcpu
, msr
, pdata
);
2421 case MSR_K7_CLK_CTL
:
2423 * Provide expected ramp-up count for K7. All other
2424 * are set to zero, indicating minimum divisors for
2427 * This prevents guest kernels on AMD host with CPU
2428 * type 6, model 8 and higher from exploding due to
2429 * the rdmsr failing.
2433 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2434 if (kvm_hv_msr_partition_wide(msr
)) {
2436 mutex_lock(&vcpu
->kvm
->lock
);
2437 r
= get_msr_hyperv_pw(vcpu
, msr
, pdata
);
2438 mutex_unlock(&vcpu
->kvm
->lock
);
2441 return get_msr_hyperv(vcpu
, msr
, pdata
);
2443 case MSR_IA32_BBL_CR_CTL3
:
2444 /* This legacy MSR exists but isn't fully documented in current
2445 * silicon. It is however accessed by winxp in very narrow
2446 * scenarios where it sets bit #19, itself documented as
2447 * a "reserved" bit. Best effort attempt to source coherent
2448 * read data here should the balance of the register be
2449 * interpreted by the guest:
2451 * L2 cache control register 3: 64GB range, 256KB size,
2452 * enabled, latency 0x1, configured
2456 case MSR_AMD64_OSVW_ID_LENGTH
:
2457 if (!guest_cpuid_has_osvw(vcpu
))
2459 data
= vcpu
->arch
.osvw
.length
;
2461 case MSR_AMD64_OSVW_STATUS
:
2462 if (!guest_cpuid_has_osvw(vcpu
))
2464 data
= vcpu
->arch
.osvw
.status
;
2467 if (kvm_pmu_msr(vcpu
, msr
))
2468 return kvm_pmu_get_msr(vcpu
, msr
, pdata
);
2470 vcpu_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
2473 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr
);
2481 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2484 * Read or write a bunch of msrs. All parameters are kernel addresses.
2486 * @return number of msrs set successfully.
2488 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2489 struct kvm_msr_entry
*entries
,
2490 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2491 unsigned index
, u64
*data
))
2495 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2496 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2497 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2499 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2505 * Read or write a bunch of msrs. Parameters are user addresses.
2507 * @return number of msrs set successfully.
2509 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2510 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2511 unsigned index
, u64
*data
),
2514 struct kvm_msrs msrs
;
2515 struct kvm_msr_entry
*entries
;
2520 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2524 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2527 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2528 entries
= memdup_user(user_msrs
->entries
, size
);
2529 if (IS_ERR(entries
)) {
2530 r
= PTR_ERR(entries
);
2534 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2539 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2550 int kvm_dev_ioctl_check_extension(long ext
)
2555 case KVM_CAP_IRQCHIP
:
2557 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2558 case KVM_CAP_SET_TSS_ADDR
:
2559 case KVM_CAP_EXT_CPUID
:
2560 case KVM_CAP_CLOCKSOURCE
:
2562 case KVM_CAP_NOP_IO_DELAY
:
2563 case KVM_CAP_MP_STATE
:
2564 case KVM_CAP_SYNC_MMU
:
2565 case KVM_CAP_USER_NMI
:
2566 case KVM_CAP_REINJECT_CONTROL
:
2567 case KVM_CAP_IRQ_INJECT_STATUS
:
2569 case KVM_CAP_IOEVENTFD
:
2571 case KVM_CAP_PIT_STATE2
:
2572 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2573 case KVM_CAP_XEN_HVM
:
2574 case KVM_CAP_ADJUST_CLOCK
:
2575 case KVM_CAP_VCPU_EVENTS
:
2576 case KVM_CAP_HYPERV
:
2577 case KVM_CAP_HYPERV_VAPIC
:
2578 case KVM_CAP_HYPERV_SPIN
:
2579 case KVM_CAP_PCI_SEGMENT
:
2580 case KVM_CAP_DEBUGREGS
:
2581 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2583 case KVM_CAP_ASYNC_PF
:
2584 case KVM_CAP_GET_TSC_KHZ
:
2585 case KVM_CAP_KVMCLOCK_CTRL
:
2586 case KVM_CAP_READONLY_MEM
:
2587 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2588 case KVM_CAP_ASSIGN_DEV_IRQ
:
2589 case KVM_CAP_PCI_2_3
:
2593 case KVM_CAP_COALESCED_MMIO
:
2594 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
2597 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2599 case KVM_CAP_NR_VCPUS
:
2600 r
= KVM_SOFT_MAX_VCPUS
;
2602 case KVM_CAP_MAX_VCPUS
:
2605 case KVM_CAP_NR_MEMSLOTS
:
2606 r
= KVM_USER_MEM_SLOTS
;
2608 case KVM_CAP_PV_MMU
: /* obsolete */
2611 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2613 r
= iommu_present(&pci_bus_type
);
2617 r
= KVM_MAX_MCE_BANKS
;
2622 case KVM_CAP_TSC_CONTROL
:
2623 r
= kvm_has_tsc_control
;
2625 case KVM_CAP_TSC_DEADLINE_TIMER
:
2626 r
= boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER
);
2636 long kvm_arch_dev_ioctl(struct file
*filp
,
2637 unsigned int ioctl
, unsigned long arg
)
2639 void __user
*argp
= (void __user
*)arg
;
2643 case KVM_GET_MSR_INDEX_LIST
: {
2644 struct kvm_msr_list __user
*user_msr_list
= argp
;
2645 struct kvm_msr_list msr_list
;
2649 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2652 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2653 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2656 if (n
< msr_list
.nmsrs
)
2659 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2660 num_msrs_to_save
* sizeof(u32
)))
2662 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2664 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2669 case KVM_GET_SUPPORTED_CPUID
: {
2670 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2671 struct kvm_cpuid2 cpuid
;
2674 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2676 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
2677 cpuid_arg
->entries
);
2682 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2687 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2690 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
2692 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
2704 static void wbinvd_ipi(void *garbage
)
2709 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2711 return vcpu
->kvm
->arch
.iommu_domain
&&
2712 !(vcpu
->kvm
->arch
.iommu_flags
& KVM_IOMMU_CACHE_COHERENCY
);
2715 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2717 /* Address WBINVD may be executed by guest */
2718 if (need_emulate_wbinvd(vcpu
)) {
2719 if (kvm_x86_ops
->has_wbinvd_exit())
2720 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2721 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2722 smp_call_function_single(vcpu
->cpu
,
2723 wbinvd_ipi
, NULL
, 1);
2726 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2728 /* Apply any externally detected TSC adjustments (due to suspend) */
2729 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2730 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2731 vcpu
->arch
.tsc_offset_adjustment
= 0;
2732 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
2735 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2736 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2737 native_read_tsc() - vcpu
->arch
.last_host_tsc
;
2739 mark_tsc_unstable("KVM discovered backwards TSC");
2740 if (check_tsc_unstable()) {
2741 u64 offset
= kvm_x86_ops
->compute_tsc_offset(vcpu
,
2742 vcpu
->arch
.last_guest_tsc
);
2743 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
2744 vcpu
->arch
.tsc_catchup
= 1;
2747 * On a host with synchronized TSC, there is no need to update
2748 * kvmclock on vcpu->cpu migration
2750 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
2751 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2752 if (vcpu
->cpu
!= cpu
)
2753 kvm_migrate_timers(vcpu
);
2757 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2760 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
2762 kvm_x86_ops
->vcpu_put(vcpu
);
2763 kvm_put_guest_fpu(vcpu
);
2764 vcpu
->arch
.last_host_tsc
= native_read_tsc();
2767 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
2768 struct kvm_lapic_state
*s
)
2770 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
2771 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
2776 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
2777 struct kvm_lapic_state
*s
)
2779 kvm_apic_post_state_restore(vcpu
, s
);
2780 update_cr8_intercept(vcpu
);
2785 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2786 struct kvm_interrupt
*irq
)
2788 if (irq
->irq
>= KVM_NR_INTERRUPTS
)
2790 if (irqchip_in_kernel(vcpu
->kvm
))
2793 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
2794 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2799 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
2801 kvm_inject_nmi(vcpu
);
2806 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
2807 struct kvm_tpr_access_ctl
*tac
)
2811 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
2815 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
2819 unsigned bank_num
= mcg_cap
& 0xff, bank
;
2822 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
2824 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
2827 vcpu
->arch
.mcg_cap
= mcg_cap
;
2828 /* Init IA32_MCG_CTL to all 1s */
2829 if (mcg_cap
& MCG_CTL_P
)
2830 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
2831 /* Init IA32_MCi_CTL to all 1s */
2832 for (bank
= 0; bank
< bank_num
; bank
++)
2833 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
2838 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
2839 struct kvm_x86_mce
*mce
)
2841 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2842 unsigned bank_num
= mcg_cap
& 0xff;
2843 u64
*banks
= vcpu
->arch
.mce_banks
;
2845 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
2848 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2849 * reporting is disabled
2851 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
2852 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
2854 banks
+= 4 * mce
->bank
;
2856 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2857 * reporting is disabled for the bank
2859 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
2861 if (mce
->status
& MCI_STATUS_UC
) {
2862 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
2863 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
2864 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2867 if (banks
[1] & MCI_STATUS_VAL
)
2868 mce
->status
|= MCI_STATUS_OVER
;
2869 banks
[2] = mce
->addr
;
2870 banks
[3] = mce
->misc
;
2871 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
2872 banks
[1] = mce
->status
;
2873 kvm_queue_exception(vcpu
, MC_VECTOR
);
2874 } else if (!(banks
[1] & MCI_STATUS_VAL
)
2875 || !(banks
[1] & MCI_STATUS_UC
)) {
2876 if (banks
[1] & MCI_STATUS_VAL
)
2877 mce
->status
|= MCI_STATUS_OVER
;
2878 banks
[2] = mce
->addr
;
2879 banks
[3] = mce
->misc
;
2880 banks
[1] = mce
->status
;
2882 banks
[1] |= MCI_STATUS_OVER
;
2886 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
2887 struct kvm_vcpu_events
*events
)
2890 events
->exception
.injected
=
2891 vcpu
->arch
.exception
.pending
&&
2892 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
2893 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
2894 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
2895 events
->exception
.pad
= 0;
2896 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
2898 events
->interrupt
.injected
=
2899 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
2900 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
2901 events
->interrupt
.soft
= 0;
2902 events
->interrupt
.shadow
=
2903 kvm_x86_ops
->get_interrupt_shadow(vcpu
,
2904 KVM_X86_SHADOW_INT_MOV_SS
| KVM_X86_SHADOW_INT_STI
);
2906 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
2907 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
2908 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
2909 events
->nmi
.pad
= 0;
2911 events
->sipi_vector
= 0; /* never valid when reporting to user space */
2913 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
2914 | KVM_VCPUEVENT_VALID_SHADOW
);
2915 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
2918 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
2919 struct kvm_vcpu_events
*events
)
2921 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2922 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2923 | KVM_VCPUEVENT_VALID_SHADOW
))
2927 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
2928 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
2929 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
2930 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
2932 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
2933 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
2934 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
2935 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
2936 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
2937 events
->interrupt
.shadow
);
2939 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
2940 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
2941 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
2942 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
2944 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
&&
2945 kvm_vcpu_has_lapic(vcpu
))
2946 vcpu
->arch
.apic
->sipi_vector
= events
->sipi_vector
;
2948 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
2953 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
2954 struct kvm_debugregs
*dbgregs
)
2956 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
2957 dbgregs
->dr6
= vcpu
->arch
.dr6
;
2958 dbgregs
->dr7
= vcpu
->arch
.dr7
;
2960 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
2963 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
2964 struct kvm_debugregs
*dbgregs
)
2969 if (dbgregs
->dr6
& ~0xffffffffull
)
2971 if (dbgregs
->dr7
& ~0xffffffffull
)
2974 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
2975 vcpu
->arch
.dr6
= dbgregs
->dr6
;
2976 vcpu
->arch
.dr7
= dbgregs
->dr7
;
2981 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
2982 struct kvm_xsave
*guest_xsave
)
2985 memcpy(guest_xsave
->region
,
2986 &vcpu
->arch
.guest_fpu
.state
->xsave
,
2989 memcpy(guest_xsave
->region
,
2990 &vcpu
->arch
.guest_fpu
.state
->fxsave
,
2991 sizeof(struct i387_fxsave_struct
));
2992 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
2997 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
2998 struct kvm_xsave
*guest_xsave
)
3001 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
3004 memcpy(&vcpu
->arch
.guest_fpu
.state
->xsave
,
3005 guest_xsave
->region
, xstate_size
);
3007 if (xstate_bv
& ~XSTATE_FPSSE
)
3009 memcpy(&vcpu
->arch
.guest_fpu
.state
->fxsave
,
3010 guest_xsave
->region
, sizeof(struct i387_fxsave_struct
));
3015 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
3016 struct kvm_xcrs
*guest_xcrs
)
3018 if (!cpu_has_xsave
) {
3019 guest_xcrs
->nr_xcrs
= 0;
3023 guest_xcrs
->nr_xcrs
= 1;
3024 guest_xcrs
->flags
= 0;
3025 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
3026 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
3029 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
3030 struct kvm_xcrs
*guest_xcrs
)
3037 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
3040 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
3041 /* Only support XCR0 currently */
3042 if (guest_xcrs
->xcrs
[0].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
3043 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
3044 guest_xcrs
->xcrs
[0].value
);
3053 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3054 * stopped by the hypervisor. This function will be called from the host only.
3055 * EINVAL is returned when the host attempts to set the flag for a guest that
3056 * does not support pv clocks.
3058 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
3060 if (!vcpu
->arch
.pv_time_enabled
)
3062 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
3063 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
3067 long kvm_arch_vcpu_ioctl(struct file
*filp
,
3068 unsigned int ioctl
, unsigned long arg
)
3070 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3071 void __user
*argp
= (void __user
*)arg
;
3074 struct kvm_lapic_state
*lapic
;
3075 struct kvm_xsave
*xsave
;
3076 struct kvm_xcrs
*xcrs
;
3082 case KVM_GET_LAPIC
: {
3084 if (!vcpu
->arch
.apic
)
3086 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
3091 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
3095 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
3100 case KVM_SET_LAPIC
: {
3102 if (!vcpu
->arch
.apic
)
3104 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
3105 if (IS_ERR(u
.lapic
))
3106 return PTR_ERR(u
.lapic
);
3108 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
3111 case KVM_INTERRUPT
: {
3112 struct kvm_interrupt irq
;
3115 if (copy_from_user(&irq
, argp
, sizeof irq
))
3117 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
3121 r
= kvm_vcpu_ioctl_nmi(vcpu
);
3124 case KVM_SET_CPUID
: {
3125 struct kvm_cpuid __user
*cpuid_arg
= argp
;
3126 struct kvm_cpuid cpuid
;
3129 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3131 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3134 case KVM_SET_CPUID2
: {
3135 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3136 struct kvm_cpuid2 cpuid
;
3139 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3141 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
3142 cpuid_arg
->entries
);
3145 case KVM_GET_CPUID2
: {
3146 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3147 struct kvm_cpuid2 cpuid
;
3150 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3152 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
3153 cpuid_arg
->entries
);
3157 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
3163 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
3166 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
3168 case KVM_TPR_ACCESS_REPORTING
: {
3169 struct kvm_tpr_access_ctl tac
;
3172 if (copy_from_user(&tac
, argp
, sizeof tac
))
3174 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
3178 if (copy_to_user(argp
, &tac
, sizeof tac
))
3183 case KVM_SET_VAPIC_ADDR
: {
3184 struct kvm_vapic_addr va
;
3187 if (!irqchip_in_kernel(vcpu
->kvm
))
3190 if (copy_from_user(&va
, argp
, sizeof va
))
3192 r
= kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
3195 case KVM_X86_SETUP_MCE
: {
3199 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
3201 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
3204 case KVM_X86_SET_MCE
: {
3205 struct kvm_x86_mce mce
;
3208 if (copy_from_user(&mce
, argp
, sizeof mce
))
3210 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
3213 case KVM_GET_VCPU_EVENTS
: {
3214 struct kvm_vcpu_events events
;
3216 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
3219 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
3224 case KVM_SET_VCPU_EVENTS
: {
3225 struct kvm_vcpu_events events
;
3228 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
3231 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
3234 case KVM_GET_DEBUGREGS
: {
3235 struct kvm_debugregs dbgregs
;
3237 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
3240 if (copy_to_user(argp
, &dbgregs
,
3241 sizeof(struct kvm_debugregs
)))
3246 case KVM_SET_DEBUGREGS
: {
3247 struct kvm_debugregs dbgregs
;
3250 if (copy_from_user(&dbgregs
, argp
,
3251 sizeof(struct kvm_debugregs
)))
3254 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3257 case KVM_GET_XSAVE
: {
3258 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3263 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3266 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3271 case KVM_SET_XSAVE
: {
3272 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
3273 if (IS_ERR(u
.xsave
))
3274 return PTR_ERR(u
.xsave
);
3276 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3279 case KVM_GET_XCRS
: {
3280 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3285 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3288 if (copy_to_user(argp
, u
.xcrs
,
3289 sizeof(struct kvm_xcrs
)))
3294 case KVM_SET_XCRS
: {
3295 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
3297 return PTR_ERR(u
.xcrs
);
3299 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3302 case KVM_SET_TSC_KHZ
: {
3306 user_tsc_khz
= (u32
)arg
;
3308 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3311 if (user_tsc_khz
== 0)
3312 user_tsc_khz
= tsc_khz
;
3314 kvm_set_tsc_khz(vcpu
, user_tsc_khz
);
3319 case KVM_GET_TSC_KHZ
: {
3320 r
= vcpu
->arch
.virtual_tsc_khz
;
3323 case KVM_KVMCLOCK_CTRL
: {
3324 r
= kvm_set_guest_paused(vcpu
);
3335 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
3337 return VM_FAULT_SIGBUS
;
3340 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3344 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3346 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3350 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3353 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3357 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3358 u32 kvm_nr_mmu_pages
)
3360 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3363 mutex_lock(&kvm
->slots_lock
);
3365 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3366 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3368 mutex_unlock(&kvm
->slots_lock
);
3372 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3374 return kvm
->arch
.n_max_mmu_pages
;
3377 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3382 switch (chip
->chip_id
) {
3383 case KVM_IRQCHIP_PIC_MASTER
:
3384 memcpy(&chip
->chip
.pic
,
3385 &pic_irqchip(kvm
)->pics
[0],
3386 sizeof(struct kvm_pic_state
));
3388 case KVM_IRQCHIP_PIC_SLAVE
:
3389 memcpy(&chip
->chip
.pic
,
3390 &pic_irqchip(kvm
)->pics
[1],
3391 sizeof(struct kvm_pic_state
));
3393 case KVM_IRQCHIP_IOAPIC
:
3394 r
= kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3403 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3408 switch (chip
->chip_id
) {
3409 case KVM_IRQCHIP_PIC_MASTER
:
3410 spin_lock(&pic_irqchip(kvm
)->lock
);
3411 memcpy(&pic_irqchip(kvm
)->pics
[0],
3413 sizeof(struct kvm_pic_state
));
3414 spin_unlock(&pic_irqchip(kvm
)->lock
);
3416 case KVM_IRQCHIP_PIC_SLAVE
:
3417 spin_lock(&pic_irqchip(kvm
)->lock
);
3418 memcpy(&pic_irqchip(kvm
)->pics
[1],
3420 sizeof(struct kvm_pic_state
));
3421 spin_unlock(&pic_irqchip(kvm
)->lock
);
3423 case KVM_IRQCHIP_IOAPIC
:
3424 r
= kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3430 kvm_pic_update_irq(pic_irqchip(kvm
));
3434 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3438 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3439 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
3440 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3444 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3448 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3449 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
3450 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
, 0);
3451 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3455 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3459 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3460 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3461 sizeof(ps
->channels
));
3462 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3463 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3464 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3468 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3470 int r
= 0, start
= 0;
3471 u32 prev_legacy
, cur_legacy
;
3472 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3473 prev_legacy
= kvm
->arch
.vpit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3474 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3475 if (!prev_legacy
&& cur_legacy
)
3477 memcpy(&kvm
->arch
.vpit
->pit_state
.channels
, &ps
->channels
,
3478 sizeof(kvm
->arch
.vpit
->pit_state
.channels
));
3479 kvm
->arch
.vpit
->pit_state
.flags
= ps
->flags
;
3480 kvm_pit_load_count(kvm
, 0, kvm
->arch
.vpit
->pit_state
.channels
[0].count
, start
);
3481 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3485 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
3486 struct kvm_reinject_control
*control
)
3488 if (!kvm
->arch
.vpit
)
3490 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3491 kvm
->arch
.vpit
->pit_state
.reinject
= control
->pit_reinject
;
3492 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3497 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3498 * @kvm: kvm instance
3499 * @log: slot id and address to which we copy the log
3501 * We need to keep it in mind that VCPU threads can write to the bitmap
3502 * concurrently. So, to avoid losing data, we keep the following order for
3505 * 1. Take a snapshot of the bit and clear it if needed.
3506 * 2. Write protect the corresponding page.
3507 * 3. Flush TLB's if needed.
3508 * 4. Copy the snapshot to the userspace.
3510 * Between 2 and 3, the guest may write to the page using the remaining TLB
3511 * entry. This is not a problem because the page will be reported dirty at
3512 * step 4 using the snapshot taken before and step 3 ensures that successive
3513 * writes will be logged for the next call.
3515 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
3518 struct kvm_memory_slot
*memslot
;
3520 unsigned long *dirty_bitmap
;
3521 unsigned long *dirty_bitmap_buffer
;
3522 bool is_dirty
= false;
3524 mutex_lock(&kvm
->slots_lock
);
3527 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
3530 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
3532 dirty_bitmap
= memslot
->dirty_bitmap
;
3537 n
= kvm_dirty_bitmap_bytes(memslot
);
3539 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
3540 memset(dirty_bitmap_buffer
, 0, n
);
3542 spin_lock(&kvm
->mmu_lock
);
3544 for (i
= 0; i
< n
/ sizeof(long); i
++) {
3548 if (!dirty_bitmap
[i
])
3553 mask
= xchg(&dirty_bitmap
[i
], 0);
3554 dirty_bitmap_buffer
[i
] = mask
;
3556 offset
= i
* BITS_PER_LONG
;
3557 kvm_mmu_write_protect_pt_masked(kvm
, memslot
, offset
, mask
);
3560 kvm_flush_remote_tlbs(kvm
);
3562 spin_unlock(&kvm
->mmu_lock
);
3565 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
3570 mutex_unlock(&kvm
->slots_lock
);
3574 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
3577 if (!irqchip_in_kernel(kvm
))
3580 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
3581 irq_event
->irq
, irq_event
->level
,
3586 long kvm_arch_vm_ioctl(struct file
*filp
,
3587 unsigned int ioctl
, unsigned long arg
)
3589 struct kvm
*kvm
= filp
->private_data
;
3590 void __user
*argp
= (void __user
*)arg
;
3593 * This union makes it completely explicit to gcc-3.x
3594 * that these two variables' stack usage should be
3595 * combined, not added together.
3598 struct kvm_pit_state ps
;
3599 struct kvm_pit_state2 ps2
;
3600 struct kvm_pit_config pit_config
;
3604 case KVM_SET_TSS_ADDR
:
3605 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
3607 case KVM_SET_IDENTITY_MAP_ADDR
: {
3611 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
3613 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
3616 case KVM_SET_NR_MMU_PAGES
:
3617 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
3619 case KVM_GET_NR_MMU_PAGES
:
3620 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
3622 case KVM_CREATE_IRQCHIP
: {
3623 struct kvm_pic
*vpic
;
3625 mutex_lock(&kvm
->lock
);
3628 goto create_irqchip_unlock
;
3630 if (atomic_read(&kvm
->online_vcpus
))
3631 goto create_irqchip_unlock
;
3633 vpic
= kvm_create_pic(kvm
);
3635 r
= kvm_ioapic_init(kvm
);
3637 mutex_lock(&kvm
->slots_lock
);
3638 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3640 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3642 kvm_io_bus_unregister_dev(kvm
, KVM_PIO_BUS
,
3644 mutex_unlock(&kvm
->slots_lock
);
3646 goto create_irqchip_unlock
;
3649 goto create_irqchip_unlock
;
3651 kvm
->arch
.vpic
= vpic
;
3653 r
= kvm_setup_default_irq_routing(kvm
);
3655 mutex_lock(&kvm
->slots_lock
);
3656 mutex_lock(&kvm
->irq_lock
);
3657 kvm_ioapic_destroy(kvm
);
3658 kvm_destroy_pic(kvm
);
3659 mutex_unlock(&kvm
->irq_lock
);
3660 mutex_unlock(&kvm
->slots_lock
);
3662 create_irqchip_unlock
:
3663 mutex_unlock(&kvm
->lock
);
3666 case KVM_CREATE_PIT
:
3667 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
3669 case KVM_CREATE_PIT2
:
3671 if (copy_from_user(&u
.pit_config
, argp
,
3672 sizeof(struct kvm_pit_config
)))
3675 mutex_lock(&kvm
->slots_lock
);
3678 goto create_pit_unlock
;
3680 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
3684 mutex_unlock(&kvm
->slots_lock
);
3686 case KVM_GET_IRQCHIP
: {
3687 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3688 struct kvm_irqchip
*chip
;
3690 chip
= memdup_user(argp
, sizeof(*chip
));
3697 if (!irqchip_in_kernel(kvm
))
3698 goto get_irqchip_out
;
3699 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
3701 goto get_irqchip_out
;
3703 if (copy_to_user(argp
, chip
, sizeof *chip
))
3704 goto get_irqchip_out
;
3710 case KVM_SET_IRQCHIP
: {
3711 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3712 struct kvm_irqchip
*chip
;
3714 chip
= memdup_user(argp
, sizeof(*chip
));
3721 if (!irqchip_in_kernel(kvm
))
3722 goto set_irqchip_out
;
3723 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
3725 goto set_irqchip_out
;
3733 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
3736 if (!kvm
->arch
.vpit
)
3738 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
3742 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
3749 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
3752 if (!kvm
->arch
.vpit
)
3754 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
3757 case KVM_GET_PIT2
: {
3759 if (!kvm
->arch
.vpit
)
3761 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
3765 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
3770 case KVM_SET_PIT2
: {
3772 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
3775 if (!kvm
->arch
.vpit
)
3777 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
3780 case KVM_REINJECT_CONTROL
: {
3781 struct kvm_reinject_control control
;
3783 if (copy_from_user(&control
, argp
, sizeof(control
)))
3785 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
3788 case KVM_XEN_HVM_CONFIG
: {
3790 if (copy_from_user(&kvm
->arch
.xen_hvm_config
, argp
,
3791 sizeof(struct kvm_xen_hvm_config
)))
3794 if (kvm
->arch
.xen_hvm_config
.flags
)
3799 case KVM_SET_CLOCK
: {
3800 struct kvm_clock_data user_ns
;
3805 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
3813 local_irq_disable();
3814 now_ns
= get_kernel_ns();
3815 delta
= user_ns
.clock
- now_ns
;
3817 kvm
->arch
.kvmclock_offset
= delta
;
3820 case KVM_GET_CLOCK
: {
3821 struct kvm_clock_data user_ns
;
3824 local_irq_disable();
3825 now_ns
= get_kernel_ns();
3826 user_ns
.clock
= kvm
->arch
.kvmclock_offset
+ now_ns
;
3829 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
3832 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
3845 static void kvm_init_msr_list(void)
3850 /* skip the first msrs in the list. KVM-specific */
3851 for (i
= j
= KVM_SAVE_MSRS_BEGIN
; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
3852 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
3855 msrs_to_save
[j
] = msrs_to_save
[i
];
3858 num_msrs_to_save
= j
;
3861 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
3869 if (!(vcpu
->arch
.apic
&&
3870 !kvm_iodevice_write(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3871 && kvm_io_bus_write(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3882 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
3889 if (!(vcpu
->arch
.apic
&&
3890 !kvm_iodevice_read(&vcpu
->arch
.apic
->dev
, addr
, n
, v
))
3891 && kvm_io_bus_read(vcpu
->kvm
, KVM_MMIO_BUS
, addr
, n
, v
))
3893 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, *(u64
*)v
);
3903 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3904 struct kvm_segment
*var
, int seg
)
3906 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3909 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3910 struct kvm_segment
*var
, int seg
)
3912 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3915 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
)
3918 struct x86_exception exception
;
3920 BUG_ON(!mmu_is_nested(vcpu
));
3922 /* NPT walks are always user-walks */
3923 access
|= PFERR_USER_MASK
;
3924 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, &exception
);
3929 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
3930 struct x86_exception
*exception
)
3932 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3933 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3936 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
3937 struct x86_exception
*exception
)
3939 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3940 access
|= PFERR_FETCH_MASK
;
3941 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3944 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
3945 struct x86_exception
*exception
)
3947 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3948 access
|= PFERR_WRITE_MASK
;
3949 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
3952 /* uses this to access any guest's mapped memory without checking CPL */
3953 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
3954 struct x86_exception
*exception
)
3956 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
3959 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
3960 struct kvm_vcpu
*vcpu
, u32 access
,
3961 struct x86_exception
*exception
)
3964 int r
= X86EMUL_CONTINUE
;
3967 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
3969 unsigned offset
= addr
& (PAGE_SIZE
-1);
3970 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
3973 if (gpa
== UNMAPPED_GVA
)
3974 return X86EMUL_PROPAGATE_FAULT
;
3975 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
3977 r
= X86EMUL_IO_NEEDED
;
3989 /* used for instruction fetching */
3990 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
3991 gva_t addr
, void *val
, unsigned int bytes
,
3992 struct x86_exception
*exception
)
3994 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
3995 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
3997 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
,
3998 access
| PFERR_FETCH_MASK
,
4002 int kvm_read_guest_virt(struct x86_emulate_ctxt
*ctxt
,
4003 gva_t addr
, void *val
, unsigned int bytes
,
4004 struct x86_exception
*exception
)
4006 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4007 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4009 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
4012 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
4014 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
4015 gva_t addr
, void *val
, unsigned int bytes
,
4016 struct x86_exception
*exception
)
4018 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4019 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, 0, exception
);
4022 int kvm_write_guest_virt_system(struct x86_emulate_ctxt
*ctxt
,
4023 gva_t addr
, void *val
,
4025 struct x86_exception
*exception
)
4027 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4029 int r
= X86EMUL_CONTINUE
;
4032 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
4035 unsigned offset
= addr
& (PAGE_SIZE
-1);
4036 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4039 if (gpa
== UNMAPPED_GVA
)
4040 return X86EMUL_PROPAGATE_FAULT
;
4041 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
4043 r
= X86EMUL_IO_NEEDED
;
4054 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
4056 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4057 gpa_t
*gpa
, struct x86_exception
*exception
,
4060 u32 access
= ((kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
4061 | (write
? PFERR_WRITE_MASK
: 0);
4063 if (vcpu_match_mmio_gva(vcpu
, gva
)
4064 && !permission_fault(vcpu
->arch
.walk_mmu
, vcpu
->arch
.access
, access
)) {
4065 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
4066 (gva
& (PAGE_SIZE
- 1));
4067 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
4071 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4073 if (*gpa
== UNMAPPED_GVA
)
4076 /* For APIC access vmexit */
4077 if ((*gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4080 if (vcpu_match_mmio_gpa(vcpu
, *gpa
)) {
4081 trace_vcpu_match_mmio(gva
, *gpa
, write
, true);
4088 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4089 const void *val
, int bytes
)
4093 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
4096 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
4100 struct read_write_emulator_ops
{
4101 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
4103 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4104 void *val
, int bytes
);
4105 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4106 int bytes
, void *val
);
4107 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4108 void *val
, int bytes
);
4112 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
4114 if (vcpu
->mmio_read_completed
) {
4115 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
4116 vcpu
->mmio_fragments
[0].gpa
, *(u64
*)val
);
4117 vcpu
->mmio_read_completed
= 0;
4124 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4125 void *val
, int bytes
)
4127 return !kvm_read_guest(vcpu
->kvm
, gpa
, val
, bytes
);
4130 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4131 void *val
, int bytes
)
4133 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
4136 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
4138 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
4139 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
4142 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4143 void *val
, int bytes
)
4145 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
4146 return X86EMUL_IO_NEEDED
;
4149 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4150 void *val
, int bytes
)
4152 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
4154 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
4155 return X86EMUL_CONTINUE
;
4158 static const struct read_write_emulator_ops read_emultor
= {
4159 .read_write_prepare
= read_prepare
,
4160 .read_write_emulate
= read_emulate
,
4161 .read_write_mmio
= vcpu_mmio_read
,
4162 .read_write_exit_mmio
= read_exit_mmio
,
4165 static const struct read_write_emulator_ops write_emultor
= {
4166 .read_write_emulate
= write_emulate
,
4167 .read_write_mmio
= write_mmio
,
4168 .read_write_exit_mmio
= write_exit_mmio
,
4172 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
4174 struct x86_exception
*exception
,
4175 struct kvm_vcpu
*vcpu
,
4176 const struct read_write_emulator_ops
*ops
)
4180 bool write
= ops
->write
;
4181 struct kvm_mmio_fragment
*frag
;
4183 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
4186 return X86EMUL_PROPAGATE_FAULT
;
4188 /* For APIC access vmexit */
4192 if (ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
4193 return X86EMUL_CONTINUE
;
4197 * Is this MMIO handled locally?
4199 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
4200 if (handled
== bytes
)
4201 return X86EMUL_CONTINUE
;
4207 WARN_ON(vcpu
->mmio_nr_fragments
>= KVM_MAX_MMIO_FRAGMENTS
);
4208 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
4212 return X86EMUL_CONTINUE
;
4215 int emulator_read_write(struct x86_emulate_ctxt
*ctxt
, unsigned long addr
,
4216 void *val
, unsigned int bytes
,
4217 struct x86_exception
*exception
,
4218 const struct read_write_emulator_ops
*ops
)
4220 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4224 if (ops
->read_write_prepare
&&
4225 ops
->read_write_prepare(vcpu
, val
, bytes
))
4226 return X86EMUL_CONTINUE
;
4228 vcpu
->mmio_nr_fragments
= 0;
4230 /* Crossing a page boundary? */
4231 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
4234 now
= -addr
& ~PAGE_MASK
;
4235 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
4238 if (rc
!= X86EMUL_CONTINUE
)
4245 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
4247 if (rc
!= X86EMUL_CONTINUE
)
4250 if (!vcpu
->mmio_nr_fragments
)
4253 gpa
= vcpu
->mmio_fragments
[0].gpa
;
4255 vcpu
->mmio_needed
= 1;
4256 vcpu
->mmio_cur_fragment
= 0;
4258 vcpu
->run
->mmio
.len
= min(8u, vcpu
->mmio_fragments
[0].len
);
4259 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
4260 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
4261 vcpu
->run
->mmio
.phys_addr
= gpa
;
4263 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
4266 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
4270 struct x86_exception
*exception
)
4272 return emulator_read_write(ctxt
, addr
, val
, bytes
,
4273 exception
, &read_emultor
);
4276 int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
4280 struct x86_exception
*exception
)
4282 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
4283 exception
, &write_emultor
);
4286 #define CMPXCHG_TYPE(t, ptr, old, new) \
4287 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4289 #ifdef CONFIG_X86_64
4290 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4292 # define CMPXCHG64(ptr, old, new) \
4293 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4296 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
4301 struct x86_exception
*exception
)
4303 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4309 /* guests cmpxchg8b have to be emulated atomically */
4310 if (bytes
> 8 || (bytes
& (bytes
- 1)))
4313 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
4315 if (gpa
== UNMAPPED_GVA
||
4316 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4319 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
4322 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
4323 if (is_error_page(page
))
4326 kaddr
= kmap_atomic(page
);
4327 kaddr
+= offset_in_page(gpa
);
4330 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
4333 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
4336 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
4339 exchanged
= CMPXCHG64(kaddr
, old
, new);
4344 kunmap_atomic(kaddr
);
4345 kvm_release_page_dirty(page
);
4348 return X86EMUL_CMPXCHG_FAILED
;
4350 kvm_mmu_pte_write(vcpu
, gpa
, new, bytes
);
4352 return X86EMUL_CONTINUE
;
4355 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
4357 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
4360 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
4362 /* TODO: String I/O for in kernel device */
4365 if (vcpu
->arch
.pio
.in
)
4366 r
= kvm_io_bus_read(vcpu
->kvm
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
4367 vcpu
->arch
.pio
.size
, pd
);
4369 r
= kvm_io_bus_write(vcpu
->kvm
, KVM_PIO_BUS
,
4370 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
4375 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
4376 unsigned short port
, void *val
,
4377 unsigned int count
, bool in
)
4379 trace_kvm_pio(!in
, port
, size
, count
);
4381 vcpu
->arch
.pio
.port
= port
;
4382 vcpu
->arch
.pio
.in
= in
;
4383 vcpu
->arch
.pio
.count
= count
;
4384 vcpu
->arch
.pio
.size
= size
;
4386 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
4387 vcpu
->arch
.pio
.count
= 0;
4391 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
4392 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
4393 vcpu
->run
->io
.size
= size
;
4394 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
4395 vcpu
->run
->io
.count
= count
;
4396 vcpu
->run
->io
.port
= port
;
4401 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
4402 int size
, unsigned short port
, void *val
,
4405 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4408 if (vcpu
->arch
.pio
.count
)
4411 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
4414 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
4415 vcpu
->arch
.pio
.count
= 0;
4422 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
4423 int size
, unsigned short port
,
4424 const void *val
, unsigned int count
)
4426 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4428 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
4429 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
4432 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
4434 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
4437 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
4439 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
4442 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
4444 if (!need_emulate_wbinvd(vcpu
))
4445 return X86EMUL_CONTINUE
;
4447 if (kvm_x86_ops
->has_wbinvd_exit()) {
4448 int cpu
= get_cpu();
4450 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
4451 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
4452 wbinvd_ipi
, NULL
, 1);
4454 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
4457 return X86EMUL_CONTINUE
;
4459 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
4461 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
4463 kvm_emulate_wbinvd(emul_to_vcpu(ctxt
));
4466 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
4468 return _kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
4471 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
4474 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
4477 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
4479 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
4482 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
4484 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4485 unsigned long value
;
4489 value
= kvm_read_cr0(vcpu
);
4492 value
= vcpu
->arch
.cr2
;
4495 value
= kvm_read_cr3(vcpu
);
4498 value
= kvm_read_cr4(vcpu
);
4501 value
= kvm_get_cr8(vcpu
);
4504 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4511 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
4513 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4518 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
4521 vcpu
->arch
.cr2
= val
;
4524 res
= kvm_set_cr3(vcpu
, val
);
4527 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
4530 res
= kvm_set_cr8(vcpu
, val
);
4533 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
4540 static void emulator_set_rflags(struct x86_emulate_ctxt
*ctxt
, ulong val
)
4542 kvm_set_rflags(emul_to_vcpu(ctxt
), val
);
4545 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
4547 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
4550 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4552 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
4555 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4557 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
4560 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4562 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
4565 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
4567 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
4570 static unsigned long emulator_get_cached_segment_base(
4571 struct x86_emulate_ctxt
*ctxt
, int seg
)
4573 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
4576 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
4577 struct desc_struct
*desc
, u32
*base3
,
4580 struct kvm_segment var
;
4582 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
4583 *selector
= var
.selector
;
4586 memset(desc
, 0, sizeof(*desc
));
4592 set_desc_limit(desc
, var
.limit
);
4593 set_desc_base(desc
, (unsigned long)var
.base
);
4594 #ifdef CONFIG_X86_64
4596 *base3
= var
.base
>> 32;
4598 desc
->type
= var
.type
;
4600 desc
->dpl
= var
.dpl
;
4601 desc
->p
= var
.present
;
4602 desc
->avl
= var
.avl
;
4610 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
4611 struct desc_struct
*desc
, u32 base3
,
4614 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4615 struct kvm_segment var
;
4617 var
.selector
= selector
;
4618 var
.base
= get_desc_base(desc
);
4619 #ifdef CONFIG_X86_64
4620 var
.base
|= ((u64
)base3
) << 32;
4622 var
.limit
= get_desc_limit(desc
);
4624 var
.limit
= (var
.limit
<< 12) | 0xfff;
4625 var
.type
= desc
->type
;
4626 var
.present
= desc
->p
;
4627 var
.dpl
= desc
->dpl
;
4632 var
.avl
= desc
->avl
;
4633 var
.present
= desc
->p
;
4634 var
.unusable
= !var
.present
;
4637 kvm_set_segment(vcpu
, &var
, seg
);
4641 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
4642 u32 msr_index
, u64
*pdata
)
4644 return kvm_get_msr(emul_to_vcpu(ctxt
), msr_index
, pdata
);
4647 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
4648 u32 msr_index
, u64 data
)
4650 struct msr_data msr
;
4653 msr
.index
= msr_index
;
4654 msr
.host_initiated
= false;
4655 return kvm_set_msr(emul_to_vcpu(ctxt
), &msr
);
4658 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
4659 u32 pmc
, u64
*pdata
)
4661 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
4664 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
4666 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
4669 static void emulator_get_fpu(struct x86_emulate_ctxt
*ctxt
)
4672 kvm_load_guest_fpu(emul_to_vcpu(ctxt
));
4674 * CR0.TS may reference the host fpu state, not the guest fpu state,
4675 * so it may be clear at this point.
4680 static void emulator_put_fpu(struct x86_emulate_ctxt
*ctxt
)
4685 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
4686 struct x86_instruction_info
*info
,
4687 enum x86_intercept_stage stage
)
4689 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
4692 static void emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
4693 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
)
4695 kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
);
4698 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
4700 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
4703 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
4705 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
4708 static const struct x86_emulate_ops emulate_ops
= {
4709 .read_gpr
= emulator_read_gpr
,
4710 .write_gpr
= emulator_write_gpr
,
4711 .read_std
= kvm_read_guest_virt_system
,
4712 .write_std
= kvm_write_guest_virt_system
,
4713 .fetch
= kvm_fetch_guest_virt
,
4714 .read_emulated
= emulator_read_emulated
,
4715 .write_emulated
= emulator_write_emulated
,
4716 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
4717 .invlpg
= emulator_invlpg
,
4718 .pio_in_emulated
= emulator_pio_in_emulated
,
4719 .pio_out_emulated
= emulator_pio_out_emulated
,
4720 .get_segment
= emulator_get_segment
,
4721 .set_segment
= emulator_set_segment
,
4722 .get_cached_segment_base
= emulator_get_cached_segment_base
,
4723 .get_gdt
= emulator_get_gdt
,
4724 .get_idt
= emulator_get_idt
,
4725 .set_gdt
= emulator_set_gdt
,
4726 .set_idt
= emulator_set_idt
,
4727 .get_cr
= emulator_get_cr
,
4728 .set_cr
= emulator_set_cr
,
4729 .set_rflags
= emulator_set_rflags
,
4730 .cpl
= emulator_get_cpl
,
4731 .get_dr
= emulator_get_dr
,
4732 .set_dr
= emulator_set_dr
,
4733 .set_msr
= emulator_set_msr
,
4734 .get_msr
= emulator_get_msr
,
4735 .read_pmc
= emulator_read_pmc
,
4736 .halt
= emulator_halt
,
4737 .wbinvd
= emulator_wbinvd
,
4738 .fix_hypercall
= emulator_fix_hypercall
,
4739 .get_fpu
= emulator_get_fpu
,
4740 .put_fpu
= emulator_put_fpu
,
4741 .intercept
= emulator_intercept
,
4742 .get_cpuid
= emulator_get_cpuid
,
4745 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
4747 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
, mask
);
4749 * an sti; sti; sequence only disable interrupts for the first
4750 * instruction. So, if the last instruction, be it emulated or
4751 * not, left the system with the INT_STI flag enabled, it
4752 * means that the last instruction is an sti. We should not
4753 * leave the flag on in this case. The same goes for mov ss
4755 if (!(int_shadow
& mask
))
4756 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
4759 static void inject_emulated_exception(struct kvm_vcpu
*vcpu
)
4761 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4762 if (ctxt
->exception
.vector
== PF_VECTOR
)
4763 kvm_propagate_fault(vcpu
, &ctxt
->exception
);
4764 else if (ctxt
->exception
.error_code_valid
)
4765 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
4766 ctxt
->exception
.error_code
);
4768 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
4771 static void init_decode_cache(struct x86_emulate_ctxt
*ctxt
)
4773 memset(&ctxt
->twobyte
, 0,
4774 (void *)&ctxt
->_regs
- (void *)&ctxt
->twobyte
);
4776 ctxt
->fetch
.start
= 0;
4777 ctxt
->fetch
.end
= 0;
4778 ctxt
->io_read
.pos
= 0;
4779 ctxt
->io_read
.end
= 0;
4780 ctxt
->mem_read
.pos
= 0;
4781 ctxt
->mem_read
.end
= 0;
4784 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
4786 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4789 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
4791 ctxt
->eflags
= kvm_get_rflags(vcpu
);
4792 ctxt
->eip
= kvm_rip_read(vcpu
);
4793 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
4794 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
4795 cs_l
? X86EMUL_MODE_PROT64
:
4796 cs_db
? X86EMUL_MODE_PROT32
:
4797 X86EMUL_MODE_PROT16
;
4798 ctxt
->guest_mode
= is_guest_mode(vcpu
);
4800 init_decode_cache(ctxt
);
4801 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
4804 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
4806 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4809 init_emulate_ctxt(vcpu
);
4813 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
4814 ret
= emulate_int_real(ctxt
, irq
);
4816 if (ret
!= X86EMUL_CONTINUE
)
4817 return EMULATE_FAIL
;
4819 ctxt
->eip
= ctxt
->_eip
;
4820 kvm_rip_write(vcpu
, ctxt
->eip
);
4821 kvm_set_rflags(vcpu
, ctxt
->eflags
);
4823 if (irq
== NMI_VECTOR
)
4824 vcpu
->arch
.nmi_pending
= 0;
4826 vcpu
->arch
.interrupt
.pending
= false;
4828 return EMULATE_DONE
;
4830 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
4832 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
4834 int r
= EMULATE_DONE
;
4836 ++vcpu
->stat
.insn_emulation_fail
;
4837 trace_kvm_emulate_insn_failed(vcpu
);
4838 if (!is_guest_mode(vcpu
) && kvm_x86_ops
->get_cpl(vcpu
) == 0) {
4839 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
4840 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
4841 vcpu
->run
->internal
.ndata
= 0;
4844 kvm_queue_exception(vcpu
, UD_VECTOR
);
4849 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t cr2
,
4850 bool write_fault_to_shadow_pgtable
,
4856 if (emulation_type
& EMULTYPE_NO_REEXECUTE
)
4859 if (!vcpu
->arch
.mmu
.direct_map
) {
4861 * Write permission should be allowed since only
4862 * write access need to be emulated.
4864 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
4867 * If the mapping is invalid in guest, let cpu retry
4868 * it to generate fault.
4870 if (gpa
== UNMAPPED_GVA
)
4875 * Do not retry the unhandleable instruction if it faults on the
4876 * readonly host memory, otherwise it will goto a infinite loop:
4877 * retry instruction -> write #PF -> emulation fail -> retry
4878 * instruction -> ...
4880 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
4883 * If the instruction failed on the error pfn, it can not be fixed,
4884 * report the error to userspace.
4886 if (is_error_noslot_pfn(pfn
))
4889 kvm_release_pfn_clean(pfn
);
4891 /* The instructions are well-emulated on direct mmu. */
4892 if (vcpu
->arch
.mmu
.direct_map
) {
4893 unsigned int indirect_shadow_pages
;
4895 spin_lock(&vcpu
->kvm
->mmu_lock
);
4896 indirect_shadow_pages
= vcpu
->kvm
->arch
.indirect_shadow_pages
;
4897 spin_unlock(&vcpu
->kvm
->mmu_lock
);
4899 if (indirect_shadow_pages
)
4900 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
4906 * if emulation was due to access to shadowed page table
4907 * and it failed try to unshadow page and re-enter the
4908 * guest to let CPU execute the instruction.
4910 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
4913 * If the access faults on its page table, it can not
4914 * be fixed by unprotecting shadow page and it should
4915 * be reported to userspace.
4917 return !write_fault_to_shadow_pgtable
;
4920 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
4921 unsigned long cr2
, int emulation_type
)
4923 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4924 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
4926 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
4927 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
4930 * If the emulation is caused by #PF and it is non-page_table
4931 * writing instruction, it means the VM-EXIT is caused by shadow
4932 * page protected, we can zap the shadow page and retry this
4933 * instruction directly.
4935 * Note: if the guest uses a non-page-table modifying instruction
4936 * on the PDE that points to the instruction, then we will unmap
4937 * the instruction and go to an infinite loop. So, we cache the
4938 * last retried eip and the last fault address, if we meet the eip
4939 * and the address again, we can break out of the potential infinite
4942 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
4944 if (!(emulation_type
& EMULTYPE_RETRY
))
4947 if (x86_page_table_writing_insn(ctxt
))
4950 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
4953 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
4954 vcpu
->arch
.last_retry_addr
= cr2
;
4956 if (!vcpu
->arch
.mmu
.direct_map
)
4957 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
4959 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
4964 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
4965 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
4967 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
4974 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4975 bool writeback
= true;
4976 bool write_fault_to_spt
= vcpu
->arch
.write_fault_to_shadow_pgtable
;
4979 * Clear write_fault_to_shadow_pgtable here to ensure it is
4982 vcpu
->arch
.write_fault_to_shadow_pgtable
= false;
4983 kvm_clear_exception_queue(vcpu
);
4985 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
4986 init_emulate_ctxt(vcpu
);
4987 ctxt
->interruptibility
= 0;
4988 ctxt
->have_exception
= false;
4989 ctxt
->perm_ok
= false;
4991 ctxt
->only_vendor_specific_insn
4992 = emulation_type
& EMULTYPE_TRAP_UD
;
4994 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
4996 trace_kvm_emulate_insn_start(vcpu
);
4997 ++vcpu
->stat
.insn_emulation
;
4998 if (r
!= EMULATION_OK
) {
4999 if (emulation_type
& EMULTYPE_TRAP_UD
)
5000 return EMULATE_FAIL
;
5001 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5003 return EMULATE_DONE
;
5004 if (emulation_type
& EMULTYPE_SKIP
)
5005 return EMULATE_FAIL
;
5006 return handle_emulation_failure(vcpu
);
5010 if (emulation_type
& EMULTYPE_SKIP
) {
5011 kvm_rip_write(vcpu
, ctxt
->_eip
);
5012 return EMULATE_DONE
;
5015 if (retry_instruction(ctxt
, cr2
, emulation_type
))
5016 return EMULATE_DONE
;
5018 /* this is needed for vmware backdoor interface to work since it
5019 changes registers values during IO operation */
5020 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
5021 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5022 emulator_invalidate_register_cache(ctxt
);
5026 r
= x86_emulate_insn(ctxt
);
5028 if (r
== EMULATION_INTERCEPTED
)
5029 return EMULATE_DONE
;
5031 if (r
== EMULATION_FAILED
) {
5032 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5034 return EMULATE_DONE
;
5036 return handle_emulation_failure(vcpu
);
5039 if (ctxt
->have_exception
) {
5040 inject_emulated_exception(vcpu
);
5042 } else if (vcpu
->arch
.pio
.count
) {
5043 if (!vcpu
->arch
.pio
.in
)
5044 vcpu
->arch
.pio
.count
= 0;
5047 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
5049 r
= EMULATE_DO_MMIO
;
5050 } else if (vcpu
->mmio_needed
) {
5051 if (!vcpu
->mmio_is_write
)
5053 r
= EMULATE_DO_MMIO
;
5054 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
5055 } else if (r
== EMULATION_RESTART
)
5061 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
5062 kvm_set_rflags(vcpu
, ctxt
->eflags
);
5063 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5064 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5065 kvm_rip_write(vcpu
, ctxt
->eip
);
5067 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
5071 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
5073 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5075 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5076 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
5077 size
, port
, &val
, 1);
5078 /* do not return to emulator after return from userspace */
5079 vcpu
->arch
.pio
.count
= 0;
5082 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
5084 static void tsc_bad(void *info
)
5086 __this_cpu_write(cpu_tsc_khz
, 0);
5089 static void tsc_khz_changed(void *data
)
5091 struct cpufreq_freqs
*freq
= data
;
5092 unsigned long khz
= 0;
5096 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5097 khz
= cpufreq_quick_get(raw_smp_processor_id());
5100 __this_cpu_write(cpu_tsc_khz
, khz
);
5103 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
5106 struct cpufreq_freqs
*freq
= data
;
5108 struct kvm_vcpu
*vcpu
;
5109 int i
, send_ipi
= 0;
5112 * We allow guests to temporarily run on slowing clocks,
5113 * provided we notify them after, or to run on accelerating
5114 * clocks, provided we notify them before. Thus time never
5117 * However, we have a problem. We can't atomically update
5118 * the frequency of a given CPU from this function; it is
5119 * merely a notifier, which can be called from any CPU.
5120 * Changing the TSC frequency at arbitrary points in time
5121 * requires a recomputation of local variables related to
5122 * the TSC for each VCPU. We must flag these local variables
5123 * to be updated and be sure the update takes place with the
5124 * new frequency before any guests proceed.
5126 * Unfortunately, the combination of hotplug CPU and frequency
5127 * change creates an intractable locking scenario; the order
5128 * of when these callouts happen is undefined with respect to
5129 * CPU hotplug, and they can race with each other. As such,
5130 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5131 * undefined; you can actually have a CPU frequency change take
5132 * place in between the computation of X and the setting of the
5133 * variable. To protect against this problem, all updates of
5134 * the per_cpu tsc_khz variable are done in an interrupt
5135 * protected IPI, and all callers wishing to update the value
5136 * must wait for a synchronous IPI to complete (which is trivial
5137 * if the caller is on the CPU already). This establishes the
5138 * necessary total order on variable updates.
5140 * Note that because a guest time update may take place
5141 * anytime after the setting of the VCPU's request bit, the
5142 * correct TSC value must be set before the request. However,
5143 * to ensure the update actually makes it to any guest which
5144 * starts running in hardware virtualization between the set
5145 * and the acquisition of the spinlock, we must also ping the
5146 * CPU after setting the request bit.
5150 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
5152 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
5155 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5157 raw_spin_lock(&kvm_lock
);
5158 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5159 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
5160 if (vcpu
->cpu
!= freq
->cpu
)
5162 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5163 if (vcpu
->cpu
!= smp_processor_id())
5167 raw_spin_unlock(&kvm_lock
);
5169 if (freq
->old
< freq
->new && send_ipi
) {
5171 * We upscale the frequency. Must make the guest
5172 * doesn't see old kvmclock values while running with
5173 * the new frequency, otherwise we risk the guest sees
5174 * time go backwards.
5176 * In case we update the frequency for another cpu
5177 * (which might be in guest context) send an interrupt
5178 * to kick the cpu out of guest context. Next time
5179 * guest context is entered kvmclock will be updated,
5180 * so the guest will not see stale values.
5182 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
5187 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
5188 .notifier_call
= kvmclock_cpufreq_notifier
5191 static int kvmclock_cpu_notifier(struct notifier_block
*nfb
,
5192 unsigned long action
, void *hcpu
)
5194 unsigned int cpu
= (unsigned long)hcpu
;
5198 case CPU_DOWN_FAILED
:
5199 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5201 case CPU_DOWN_PREPARE
:
5202 smp_call_function_single(cpu
, tsc_bad
, NULL
, 1);
5208 static struct notifier_block kvmclock_cpu_notifier_block
= {
5209 .notifier_call
= kvmclock_cpu_notifier
,
5210 .priority
= -INT_MAX
5213 static void kvm_timer_init(void)
5217 max_tsc_khz
= tsc_khz
;
5218 register_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5219 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
5220 #ifdef CONFIG_CPU_FREQ
5221 struct cpufreq_policy policy
;
5222 memset(&policy
, 0, sizeof(policy
));
5224 cpufreq_get_policy(&policy
, cpu
);
5225 if (policy
.cpuinfo
.max_freq
)
5226 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
5229 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
5230 CPUFREQ_TRANSITION_NOTIFIER
);
5232 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
5233 for_each_online_cpu(cpu
)
5234 smp_call_function_single(cpu
, tsc_khz_changed
, NULL
, 1);
5237 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
5239 int kvm_is_in_guest(void)
5241 return __this_cpu_read(current_vcpu
) != NULL
;
5244 static int kvm_is_user_mode(void)
5248 if (__this_cpu_read(current_vcpu
))
5249 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
5251 return user_mode
!= 0;
5254 static unsigned long kvm_get_guest_ip(void)
5256 unsigned long ip
= 0;
5258 if (__this_cpu_read(current_vcpu
))
5259 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
5264 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
5265 .is_in_guest
= kvm_is_in_guest
,
5266 .is_user_mode
= kvm_is_user_mode
,
5267 .get_guest_ip
= kvm_get_guest_ip
,
5270 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
5272 __this_cpu_write(current_vcpu
, vcpu
);
5274 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
5276 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
5278 __this_cpu_write(current_vcpu
, NULL
);
5280 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
5282 static void kvm_set_mmio_spte_mask(void)
5285 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
5288 * Set the reserved bits and the present bit of an paging-structure
5289 * entry to generate page fault with PFER.RSV = 1.
5291 mask
= ((1ull << (62 - maxphyaddr
+ 1)) - 1) << maxphyaddr
;
5294 #ifdef CONFIG_X86_64
5296 * If reserved bit is not supported, clear the present bit to disable
5299 if (maxphyaddr
== 52)
5303 kvm_mmu_set_mmio_spte_mask(mask
);
5306 #ifdef CONFIG_X86_64
5307 static void pvclock_gtod_update_fn(struct work_struct
*work
)
5311 struct kvm_vcpu
*vcpu
;
5314 raw_spin_lock(&kvm_lock
);
5315 list_for_each_entry(kvm
, &vm_list
, vm_list
)
5316 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5317 set_bit(KVM_REQ_MASTERCLOCK_UPDATE
, &vcpu
->requests
);
5318 atomic_set(&kvm_guest_has_master_clock
, 0);
5319 raw_spin_unlock(&kvm_lock
);
5322 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
5325 * Notification about pvclock gtod data update.
5327 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
5330 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
5331 struct timekeeper
*tk
= priv
;
5333 update_pvclock_gtod(tk
);
5335 /* disable master clock if host does not trust, or does not
5336 * use, TSC clocksource
5338 if (gtod
->clock
.vclock_mode
!= VCLOCK_TSC
&&
5339 atomic_read(&kvm_guest_has_master_clock
) != 0)
5340 queue_work(system_long_wq
, &pvclock_gtod_work
);
5345 static struct notifier_block pvclock_gtod_notifier
= {
5346 .notifier_call
= pvclock_gtod_notify
,
5350 int kvm_arch_init(void *opaque
)
5353 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
5356 printk(KERN_ERR
"kvm: already loaded the other module\n");
5361 if (!ops
->cpu_has_kvm_support()) {
5362 printk(KERN_ERR
"kvm: no hardware support\n");
5366 if (ops
->disabled_by_bios()) {
5367 printk(KERN_ERR
"kvm: disabled by bios\n");
5373 shared_msrs
= alloc_percpu(struct kvm_shared_msrs
);
5375 printk(KERN_ERR
"kvm: failed to allocate percpu kvm_shared_msrs\n");
5379 r
= kvm_mmu_module_init();
5381 goto out_free_percpu
;
5383 kvm_set_mmio_spte_mask();
5384 kvm_init_msr_list();
5387 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
5388 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
5392 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
5395 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
5398 #ifdef CONFIG_X86_64
5399 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
5405 free_percpu(shared_msrs
);
5410 void kvm_arch_exit(void)
5412 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
5414 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
5415 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
5416 CPUFREQ_TRANSITION_NOTIFIER
);
5417 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block
);
5418 #ifdef CONFIG_X86_64
5419 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
5422 kvm_mmu_module_exit();
5423 free_percpu(shared_msrs
);
5426 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
5428 ++vcpu
->stat
.halt_exits
;
5429 if (irqchip_in_kernel(vcpu
->kvm
)) {
5430 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
5433 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
5437 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
5439 int kvm_hv_hypercall(struct kvm_vcpu
*vcpu
)
5441 u64 param
, ingpa
, outgpa
, ret
;
5442 uint16_t code
, rep_idx
, rep_cnt
, res
= HV_STATUS_SUCCESS
, rep_done
= 0;
5443 bool fast
, longmode
;
5447 * hypercall generates UD from non zero cpl and real mode
5450 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 || !is_protmode(vcpu
)) {
5451 kvm_queue_exception(vcpu
, UD_VECTOR
);
5455 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
5456 longmode
= is_long_mode(vcpu
) && cs_l
== 1;
5459 param
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDX
) << 32) |
5460 (kvm_register_read(vcpu
, VCPU_REGS_RAX
) & 0xffffffff);
5461 ingpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RBX
) << 32) |
5462 (kvm_register_read(vcpu
, VCPU_REGS_RCX
) & 0xffffffff);
5463 outgpa
= ((u64
)kvm_register_read(vcpu
, VCPU_REGS_RDI
) << 32) |
5464 (kvm_register_read(vcpu
, VCPU_REGS_RSI
) & 0xffffffff);
5466 #ifdef CONFIG_X86_64
5468 param
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5469 ingpa
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5470 outgpa
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
5474 code
= param
& 0xffff;
5475 fast
= (param
>> 16) & 0x1;
5476 rep_cnt
= (param
>> 32) & 0xfff;
5477 rep_idx
= (param
>> 48) & 0xfff;
5479 trace_kvm_hv_hypercall(code
, fast
, rep_cnt
, rep_idx
, ingpa
, outgpa
);
5482 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT
:
5483 kvm_vcpu_on_spin(vcpu
);
5486 res
= HV_STATUS_INVALID_HYPERCALL_CODE
;
5490 ret
= res
| (((u64
)rep_done
& 0xfff) << 32);
5492 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5494 kvm_register_write(vcpu
, VCPU_REGS_RDX
, ret
>> 32);
5495 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
& 0xffffffff);
5501 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
5503 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
5506 if (kvm_hv_hypercall_enabled(vcpu
->kvm
))
5507 return kvm_hv_hypercall(vcpu
);
5509 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5510 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
5511 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
5512 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
5513 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
5515 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
5517 if (!is_long_mode(vcpu
)) {
5525 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
5531 case KVM_HC_VAPIC_POLL_IRQ
:
5539 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
5540 ++vcpu
->stat
.hypercalls
;
5543 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
5545 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
5547 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5548 char instruction
[3];
5549 unsigned long rip
= kvm_rip_read(vcpu
);
5552 * Blow out the MMU to ensure that no other VCPU has an active mapping
5553 * to ensure that the updated hypercall appears atomically across all
5556 kvm_mmu_zap_all(vcpu
->kvm
);
5558 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
5560 return emulator_write_emulated(ctxt
, rip
, instruction
, 3, NULL
);
5564 * Check if userspace requested an interrupt window, and that the
5565 * interrupt window is open.
5567 * No need to exit to userspace if we already have an interrupt queued.
5569 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
5571 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
5572 vcpu
->run
->request_interrupt_window
&&
5573 kvm_arch_interrupt_allowed(vcpu
));
5576 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
5578 struct kvm_run
*kvm_run
= vcpu
->run
;
5580 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
5581 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
5582 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
5583 if (irqchip_in_kernel(vcpu
->kvm
))
5584 kvm_run
->ready_for_interrupt_injection
= 1;
5586 kvm_run
->ready_for_interrupt_injection
=
5587 kvm_arch_interrupt_allowed(vcpu
) &&
5588 !kvm_cpu_has_interrupt(vcpu
) &&
5589 !kvm_event_needs_reinjection(vcpu
);
5592 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
5596 if (!kvm_x86_ops
->update_cr8_intercept
)
5599 if (!vcpu
->arch
.apic
)
5602 if (!vcpu
->arch
.apic
->vapic_addr
)
5603 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
5610 tpr
= kvm_lapic_get_cr8(vcpu
);
5612 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
5615 static void inject_pending_event(struct kvm_vcpu
*vcpu
)
5617 /* try to reinject previous events if any */
5618 if (vcpu
->arch
.exception
.pending
) {
5619 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
5620 vcpu
->arch
.exception
.has_error_code
,
5621 vcpu
->arch
.exception
.error_code
);
5622 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
5623 vcpu
->arch
.exception
.has_error_code
,
5624 vcpu
->arch
.exception
.error_code
,
5625 vcpu
->arch
.exception
.reinject
);
5629 if (vcpu
->arch
.nmi_injected
) {
5630 kvm_x86_ops
->set_nmi(vcpu
);
5634 if (vcpu
->arch
.interrupt
.pending
) {
5635 kvm_x86_ops
->set_irq(vcpu
);
5639 /* try to inject new event if pending */
5640 if (vcpu
->arch
.nmi_pending
) {
5641 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
5642 --vcpu
->arch
.nmi_pending
;
5643 vcpu
->arch
.nmi_injected
= true;
5644 kvm_x86_ops
->set_nmi(vcpu
);
5646 } else if (kvm_cpu_has_injectable_intr(vcpu
)) {
5647 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
5648 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
5650 kvm_x86_ops
->set_irq(vcpu
);
5655 static void process_nmi(struct kvm_vcpu
*vcpu
)
5660 * x86 is limited to one NMI running, and one NMI pending after it.
5661 * If an NMI is already in progress, limit further NMIs to just one.
5662 * Otherwise, allow two (and we'll inject the first one immediately).
5664 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
5667 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
5668 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
5669 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5672 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
5674 #ifdef CONFIG_X86_64
5676 struct kvm_vcpu
*vcpu
;
5677 struct kvm_arch
*ka
= &kvm
->arch
;
5679 spin_lock(&ka
->pvclock_gtod_sync_lock
);
5680 kvm_make_mclock_inprogress_request(kvm
);
5681 /* no guest entries from this point */
5682 pvclock_update_vm_gtod_copy(kvm
);
5684 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5685 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
5687 /* guest entries allowed */
5688 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5689 clear_bit(KVM_REQ_MCLOCK_INPROGRESS
, &vcpu
->requests
);
5691 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
5695 static void vcpu_scan_ioapic(struct kvm_vcpu
*vcpu
)
5697 u64 eoi_exit_bitmap
[4];
5700 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
5703 memset(eoi_exit_bitmap
, 0, 32);
5706 kvm_ioapic_scan_entry(vcpu
, eoi_exit_bitmap
, tmr
);
5707 kvm_x86_ops
->load_eoi_exitmap(vcpu
, eoi_exit_bitmap
);
5708 kvm_apic_update_tmr(vcpu
, tmr
);
5711 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
5714 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
5715 vcpu
->run
->request_interrupt_window
;
5716 bool req_immediate_exit
= false;
5718 if (vcpu
->requests
) {
5719 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
5720 kvm_mmu_unload(vcpu
);
5721 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
5722 __kvm_migrate_timers(vcpu
);
5723 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
5724 kvm_gen_update_masterclock(vcpu
->kvm
);
5725 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
5726 r
= kvm_guest_time_update(vcpu
);
5730 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
5731 kvm_mmu_sync_roots(vcpu
);
5732 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
5733 kvm_x86_ops
->tlb_flush(vcpu
);
5734 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
5735 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
5739 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
5740 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
5744 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
)) {
5745 vcpu
->fpu_active
= 0;
5746 kvm_x86_ops
->fpu_deactivate(vcpu
);
5748 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
5749 /* Page is swapped out. Do synthetic halt */
5750 vcpu
->arch
.apf
.halted
= true;
5754 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
5755 record_steal_time(vcpu
);
5756 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
5758 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
5759 kvm_handle_pmu_event(vcpu
);
5760 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
5761 kvm_deliver_pmi(vcpu
);
5762 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC
, vcpu
))
5763 vcpu_scan_ioapic(vcpu
);
5766 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
5767 kvm_apic_accept_events(vcpu
);
5768 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
5773 inject_pending_event(vcpu
);
5775 /* enable NMI/IRQ window open exits if needed */
5776 if (vcpu
->arch
.nmi_pending
)
5777 req_immediate_exit
=
5778 kvm_x86_ops
->enable_nmi_window(vcpu
) != 0;
5779 else if (kvm_cpu_has_injectable_intr(vcpu
) || req_int_win
)
5780 req_immediate_exit
=
5781 kvm_x86_ops
->enable_irq_window(vcpu
) != 0;
5783 if (kvm_lapic_enabled(vcpu
)) {
5785 * Update architecture specific hints for APIC
5786 * virtual interrupt delivery.
5788 if (kvm_x86_ops
->hwapic_irr_update
)
5789 kvm_x86_ops
->hwapic_irr_update(vcpu
,
5790 kvm_lapic_find_highest_irr(vcpu
));
5791 update_cr8_intercept(vcpu
);
5792 kvm_lapic_sync_to_vapic(vcpu
);
5796 r
= kvm_mmu_reload(vcpu
);
5798 goto cancel_injection
;
5803 kvm_x86_ops
->prepare_guest_switch(vcpu
);
5804 if (vcpu
->fpu_active
)
5805 kvm_load_guest_fpu(vcpu
);
5806 kvm_load_guest_xcr0(vcpu
);
5808 vcpu
->mode
= IN_GUEST_MODE
;
5810 /* We should set ->mode before check ->requests,
5811 * see the comment in make_all_cpus_request.
5815 local_irq_disable();
5817 if (vcpu
->mode
== EXITING_GUEST_MODE
|| vcpu
->requests
5818 || need_resched() || signal_pending(current
)) {
5819 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5824 goto cancel_injection
;
5827 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5829 if (req_immediate_exit
)
5830 smp_send_reschedule(vcpu
->cpu
);
5834 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
5836 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
5837 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
5838 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
5839 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
5842 trace_kvm_entry(vcpu
->vcpu_id
);
5843 kvm_x86_ops
->run(vcpu
);
5846 * If the guest has used debug registers, at least dr7
5847 * will be disabled while returning to the host.
5848 * If we don't have active breakpoints in the host, we don't
5849 * care about the messed up debug address registers. But if
5850 * we have some of them active, restore the old state.
5852 if (hw_breakpoint_active())
5853 hw_breakpoint_restore();
5855 vcpu
->arch
.last_guest_tsc
= kvm_x86_ops
->read_l1_tsc(vcpu
,
5858 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
5861 /* Interrupt is enabled by handle_external_intr() */
5862 kvm_x86_ops
->handle_external_intr(vcpu
);
5867 * We must have an instruction between local_irq_enable() and
5868 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5869 * the interrupt shadow. The stat.exits increment will do nicely.
5870 * But we need to prevent reordering, hence this barrier():
5878 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5881 * Profile KVM exit RIPs:
5883 if (unlikely(prof_on
== KVM_PROFILING
)) {
5884 unsigned long rip
= kvm_rip_read(vcpu
);
5885 profile_hit(KVM_PROFILING
, (void *)rip
);
5888 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
5889 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
5891 if (vcpu
->arch
.apic_attention
)
5892 kvm_lapic_sync_from_vapic(vcpu
);
5894 r
= kvm_x86_ops
->handle_exit(vcpu
);
5898 kvm_x86_ops
->cancel_injection(vcpu
);
5899 if (unlikely(vcpu
->arch
.apic_attention
))
5900 kvm_lapic_sync_from_vapic(vcpu
);
5906 static int __vcpu_run(struct kvm_vcpu
*vcpu
)
5909 struct kvm
*kvm
= vcpu
->kvm
;
5911 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5915 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
5916 !vcpu
->arch
.apf
.halted
)
5917 r
= vcpu_enter_guest(vcpu
);
5919 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5920 kvm_vcpu_block(vcpu
);
5921 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5922 if (kvm_check_request(KVM_REQ_UNHALT
, vcpu
)) {
5923 kvm_apic_accept_events(vcpu
);
5924 switch(vcpu
->arch
.mp_state
) {
5925 case KVM_MP_STATE_HALTED
:
5926 vcpu
->arch
.mp_state
=
5927 KVM_MP_STATE_RUNNABLE
;
5928 case KVM_MP_STATE_RUNNABLE
:
5929 vcpu
->arch
.apf
.halted
= false;
5931 case KVM_MP_STATE_INIT_RECEIVED
:
5943 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
5944 if (kvm_cpu_has_pending_timer(vcpu
))
5945 kvm_inject_pending_timer_irqs(vcpu
);
5947 if (dm_request_for_irq_injection(vcpu
)) {
5949 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5950 ++vcpu
->stat
.request_irq_exits
;
5953 kvm_check_async_pf_completion(vcpu
);
5955 if (signal_pending(current
)) {
5957 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
5958 ++vcpu
->stat
.signal_exits
;
5960 if (need_resched()) {
5961 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5963 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5967 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
5972 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
5975 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5976 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
5977 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
5978 if (r
!= EMULATE_DONE
)
5983 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
5985 BUG_ON(!vcpu
->arch
.pio
.count
);
5987 return complete_emulated_io(vcpu
);
5991 * Implements the following, as a state machine:
5995 * for each mmio piece in the fragment
6003 * for each mmio piece in the fragment
6008 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
6010 struct kvm_run
*run
= vcpu
->run
;
6011 struct kvm_mmio_fragment
*frag
;
6014 BUG_ON(!vcpu
->mmio_needed
);
6016 /* Complete previous fragment */
6017 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
6018 len
= min(8u, frag
->len
);
6019 if (!vcpu
->mmio_is_write
)
6020 memcpy(frag
->data
, run
->mmio
.data
, len
);
6022 if (frag
->len
<= 8) {
6023 /* Switch to the next fragment. */
6025 vcpu
->mmio_cur_fragment
++;
6027 /* Go forward to the next mmio piece. */
6033 if (vcpu
->mmio_cur_fragment
>= vcpu
->mmio_nr_fragments
) {
6034 vcpu
->mmio_needed
= 0;
6035 if (vcpu
->mmio_is_write
)
6037 vcpu
->mmio_read_completed
= 1;
6038 return complete_emulated_io(vcpu
);
6041 run
->exit_reason
= KVM_EXIT_MMIO
;
6042 run
->mmio
.phys_addr
= frag
->gpa
;
6043 if (vcpu
->mmio_is_write
)
6044 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
6045 run
->mmio
.len
= min(8u, frag
->len
);
6046 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
6047 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
6052 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
6057 if (!tsk_used_math(current
) && init_fpu(current
))
6060 if (vcpu
->sigset_active
)
6061 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
6063 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
6064 kvm_vcpu_block(vcpu
);
6065 kvm_apic_accept_events(vcpu
);
6066 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
6071 /* re-sync apic's tpr */
6072 if (!irqchip_in_kernel(vcpu
->kvm
)) {
6073 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
6079 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
6080 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
6081 vcpu
->arch
.complete_userspace_io
= NULL
;
6086 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
6088 r
= __vcpu_run(vcpu
);
6091 post_kvm_run_save(vcpu
);
6092 if (vcpu
->sigset_active
)
6093 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
6098 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
6100 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
6102 * We are here if userspace calls get_regs() in the middle of
6103 * instruction emulation. Registers state needs to be copied
6104 * back from emulation context to vcpu. Userspace shouldn't do
6105 * that usually, but some bad designed PV devices (vmware
6106 * backdoor interface) need this to work
6108 emulator_writeback_register_cache(&vcpu
->arch
.emulate_ctxt
);
6109 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
6111 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
6112 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
6113 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
6114 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
6115 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
6116 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
6117 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
6118 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
6119 #ifdef CONFIG_X86_64
6120 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
6121 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
6122 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
6123 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
6124 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
6125 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
6126 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
6127 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
6130 regs
->rip
= kvm_rip_read(vcpu
);
6131 regs
->rflags
= kvm_get_rflags(vcpu
);
6136 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
6138 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
6139 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
6141 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
6142 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
6143 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
6144 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
6145 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
6146 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
6147 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
6148 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
6149 #ifdef CONFIG_X86_64
6150 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
6151 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
6152 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
6153 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
6154 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
6155 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
6156 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
6157 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
6160 kvm_rip_write(vcpu
, regs
->rip
);
6161 kvm_set_rflags(vcpu
, regs
->rflags
);
6163 vcpu
->arch
.exception
.pending
= false;
6165 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6170 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
6172 struct kvm_segment cs
;
6174 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6178 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
6180 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
6181 struct kvm_sregs
*sregs
)
6185 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
6186 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
6187 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
6188 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
6189 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
6190 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
6192 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
6193 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
6195 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6196 sregs
->idt
.limit
= dt
.size
;
6197 sregs
->idt
.base
= dt
.address
;
6198 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6199 sregs
->gdt
.limit
= dt
.size
;
6200 sregs
->gdt
.base
= dt
.address
;
6202 sregs
->cr0
= kvm_read_cr0(vcpu
);
6203 sregs
->cr2
= vcpu
->arch
.cr2
;
6204 sregs
->cr3
= kvm_read_cr3(vcpu
);
6205 sregs
->cr4
= kvm_read_cr4(vcpu
);
6206 sregs
->cr8
= kvm_get_cr8(vcpu
);
6207 sregs
->efer
= vcpu
->arch
.efer
;
6208 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
6210 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
6212 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
6213 set_bit(vcpu
->arch
.interrupt
.nr
,
6214 (unsigned long *)sregs
->interrupt_bitmap
);
6219 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
6220 struct kvm_mp_state
*mp_state
)
6222 kvm_apic_accept_events(vcpu
);
6223 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
6227 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
6228 struct kvm_mp_state
*mp_state
)
6230 if (!kvm_vcpu_has_lapic(vcpu
) &&
6231 mp_state
->mp_state
!= KVM_MP_STATE_RUNNABLE
)
6234 if (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
) {
6235 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
6236 set_bit(KVM_APIC_SIPI
, &vcpu
->arch
.apic
->pending_events
);
6238 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
6239 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6243 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
6244 int reason
, bool has_error_code
, u32 error_code
)
6246 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
6249 init_emulate_ctxt(vcpu
);
6251 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
6252 has_error_code
, error_code
);
6255 return EMULATE_FAIL
;
6257 kvm_rip_write(vcpu
, ctxt
->eip
);
6258 kvm_set_rflags(vcpu
, ctxt
->eflags
);
6259 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6260 return EMULATE_DONE
;
6262 EXPORT_SYMBOL_GPL(kvm_task_switch
);
6264 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
6265 struct kvm_sregs
*sregs
)
6267 int mmu_reset_needed
= 0;
6268 int pending_vec
, max_bits
, idx
;
6271 if (!guest_cpuid_has_xsave(vcpu
) && (sregs
->cr4
& X86_CR4_OSXSAVE
))
6274 dt
.size
= sregs
->idt
.limit
;
6275 dt
.address
= sregs
->idt
.base
;
6276 kvm_x86_ops
->set_idt(vcpu
, &dt
);
6277 dt
.size
= sregs
->gdt
.limit
;
6278 dt
.address
= sregs
->gdt
.base
;
6279 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
6281 vcpu
->arch
.cr2
= sregs
->cr2
;
6282 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
6283 vcpu
->arch
.cr3
= sregs
->cr3
;
6284 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
6286 kvm_set_cr8(vcpu
, sregs
->cr8
);
6288 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
6289 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
6290 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
6292 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
6293 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
6294 vcpu
->arch
.cr0
= sregs
->cr0
;
6296 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
6297 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
6298 if (sregs
->cr4
& X86_CR4_OSXSAVE
)
6299 kvm_update_cpuid(vcpu
);
6301 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6302 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
6303 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
6304 mmu_reset_needed
= 1;
6306 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6308 if (mmu_reset_needed
)
6309 kvm_mmu_reset_context(vcpu
);
6311 max_bits
= KVM_NR_INTERRUPTS
;
6312 pending_vec
= find_first_bit(
6313 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
6314 if (pending_vec
< max_bits
) {
6315 kvm_queue_interrupt(vcpu
, pending_vec
, false);
6316 pr_debug("Set back pending irq %d\n", pending_vec
);
6319 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
6320 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
6321 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
6322 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
6323 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
6324 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
6326 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
6327 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
6329 update_cr8_intercept(vcpu
);
6331 /* Older userspace won't unhalt the vcpu on reset. */
6332 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
6333 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
6335 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6337 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6342 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
6343 struct kvm_guest_debug
*dbg
)
6345 unsigned long rflags
;
6348 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
6350 if (vcpu
->arch
.exception
.pending
)
6352 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
6353 kvm_queue_exception(vcpu
, DB_VECTOR
);
6355 kvm_queue_exception(vcpu
, BP_VECTOR
);
6359 * Read rflags as long as potentially injected trace flags are still
6362 rflags
= kvm_get_rflags(vcpu
);
6364 vcpu
->guest_debug
= dbg
->control
;
6365 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
6366 vcpu
->guest_debug
= 0;
6368 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
6369 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
6370 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
6371 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
6373 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
6374 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
6376 kvm_update_dr7(vcpu
);
6378 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
6379 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
6380 get_segment_base(vcpu
, VCPU_SREG_CS
);
6383 * Trigger an rflags update that will inject or remove the trace
6386 kvm_set_rflags(vcpu
, rflags
);
6388 kvm_x86_ops
->update_db_bp_intercept(vcpu
);
6398 * Translate a guest virtual address to a guest physical address.
6400 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
6401 struct kvm_translation
*tr
)
6403 unsigned long vaddr
= tr
->linear_address
;
6407 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6408 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
6409 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6410 tr
->physical_address
= gpa
;
6411 tr
->valid
= gpa
!= UNMAPPED_GVA
;
6418 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
6420 struct i387_fxsave_struct
*fxsave
=
6421 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
6423 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
6424 fpu
->fcw
= fxsave
->cwd
;
6425 fpu
->fsw
= fxsave
->swd
;
6426 fpu
->ftwx
= fxsave
->twd
;
6427 fpu
->last_opcode
= fxsave
->fop
;
6428 fpu
->last_ip
= fxsave
->rip
;
6429 fpu
->last_dp
= fxsave
->rdp
;
6430 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
6435 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
6437 struct i387_fxsave_struct
*fxsave
=
6438 &vcpu
->arch
.guest_fpu
.state
->fxsave
;
6440 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
6441 fxsave
->cwd
= fpu
->fcw
;
6442 fxsave
->swd
= fpu
->fsw
;
6443 fxsave
->twd
= fpu
->ftwx
;
6444 fxsave
->fop
= fpu
->last_opcode
;
6445 fxsave
->rip
= fpu
->last_ip
;
6446 fxsave
->rdp
= fpu
->last_dp
;
6447 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
6452 int fx_init(struct kvm_vcpu
*vcpu
)
6456 err
= fpu_alloc(&vcpu
->arch
.guest_fpu
);
6460 fpu_finit(&vcpu
->arch
.guest_fpu
);
6463 * Ensure guest xcr0 is valid for loading
6465 vcpu
->arch
.xcr0
= XSTATE_FP
;
6467 vcpu
->arch
.cr0
|= X86_CR0_ET
;
6471 EXPORT_SYMBOL_GPL(fx_init
);
6473 static void fx_free(struct kvm_vcpu
*vcpu
)
6475 fpu_free(&vcpu
->arch
.guest_fpu
);
6478 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
6480 if (vcpu
->guest_fpu_loaded
)
6484 * Restore all possible states in the guest,
6485 * and assume host would use all available bits.
6486 * Guest xcr0 would be loaded later.
6488 kvm_put_guest_xcr0(vcpu
);
6489 vcpu
->guest_fpu_loaded
= 1;
6490 __kernel_fpu_begin();
6491 fpu_restore_checking(&vcpu
->arch
.guest_fpu
);
6495 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
6497 kvm_put_guest_xcr0(vcpu
);
6499 if (!vcpu
->guest_fpu_loaded
)
6502 vcpu
->guest_fpu_loaded
= 0;
6503 fpu_save_init(&vcpu
->arch
.guest_fpu
);
6505 ++vcpu
->stat
.fpu_reload
;
6506 kvm_make_request(KVM_REQ_DEACTIVATE_FPU
, vcpu
);
6510 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
6512 kvmclock_reset(vcpu
);
6514 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
6516 kvm_x86_ops
->vcpu_free(vcpu
);
6519 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
6522 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
6523 printk_once(KERN_WARNING
6524 "kvm: SMP vm created on host with unstable TSC; "
6525 "guest TSC will not be reliable\n");
6526 return kvm_x86_ops
->vcpu_create(kvm
, id
);
6529 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
6533 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
6534 r
= vcpu_load(vcpu
);
6537 kvm_vcpu_reset(vcpu
);
6538 r
= kvm_mmu_setup(vcpu
);
6544 int kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
6547 struct msr_data msr
;
6549 r
= vcpu_load(vcpu
);
6553 msr
.index
= MSR_IA32_TSC
;
6554 msr
.host_initiated
= true;
6555 kvm_write_tsc(vcpu
, &msr
);
6561 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
6564 vcpu
->arch
.apf
.msr_val
= 0;
6566 r
= vcpu_load(vcpu
);
6568 kvm_mmu_unload(vcpu
);
6572 kvm_x86_ops
->vcpu_free(vcpu
);
6575 void kvm_vcpu_reset(struct kvm_vcpu
*vcpu
)
6577 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
6578 vcpu
->arch
.nmi_pending
= 0;
6579 vcpu
->arch
.nmi_injected
= false;
6581 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
6582 vcpu
->arch
.dr6
= DR6_FIXED_1
;
6583 vcpu
->arch
.dr7
= DR7_FIXED_1
;
6584 kvm_update_dr7(vcpu
);
6586 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6587 vcpu
->arch
.apf
.msr_val
= 0;
6588 vcpu
->arch
.st
.msr_val
= 0;
6590 kvmclock_reset(vcpu
);
6592 kvm_clear_async_pf_completion_queue(vcpu
);
6593 kvm_async_pf_hash_reset(vcpu
);
6594 vcpu
->arch
.apf
.halted
= false;
6596 kvm_pmu_reset(vcpu
);
6598 memset(vcpu
->arch
.regs
, 0, sizeof(vcpu
->arch
.regs
));
6599 vcpu
->arch
.regs_avail
= ~0;
6600 vcpu
->arch
.regs_dirty
= ~0;
6602 kvm_x86_ops
->vcpu_reset(vcpu
);
6605 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu
*vcpu
, unsigned int vector
)
6607 struct kvm_segment cs
;
6609 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6610 cs
.selector
= vector
<< 8;
6611 cs
.base
= vector
<< 12;
6612 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6613 kvm_rip_write(vcpu
, 0);
6616 int kvm_arch_hardware_enable(void *garbage
)
6619 struct kvm_vcpu
*vcpu
;
6624 bool stable
, backwards_tsc
= false;
6626 kvm_shared_msr_cpu_online();
6627 ret
= kvm_x86_ops
->hardware_enable(garbage
);
6631 local_tsc
= native_read_tsc();
6632 stable
= !check_tsc_unstable();
6633 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6634 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6635 if (!stable
&& vcpu
->cpu
== smp_processor_id())
6636 set_bit(KVM_REQ_CLOCK_UPDATE
, &vcpu
->requests
);
6637 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
6638 backwards_tsc
= true;
6639 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
6640 max_tsc
= vcpu
->arch
.last_host_tsc
;
6646 * Sometimes, even reliable TSCs go backwards. This happens on
6647 * platforms that reset TSC during suspend or hibernate actions, but
6648 * maintain synchronization. We must compensate. Fortunately, we can
6649 * detect that condition here, which happens early in CPU bringup,
6650 * before any KVM threads can be running. Unfortunately, we can't
6651 * bring the TSCs fully up to date with real time, as we aren't yet far
6652 * enough into CPU bringup that we know how much real time has actually
6653 * elapsed; our helper function, get_kernel_ns() will be using boot
6654 * variables that haven't been updated yet.
6656 * So we simply find the maximum observed TSC above, then record the
6657 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6658 * the adjustment will be applied. Note that we accumulate
6659 * adjustments, in case multiple suspend cycles happen before some VCPU
6660 * gets a chance to run again. In the event that no KVM threads get a
6661 * chance to run, we will miss the entire elapsed period, as we'll have
6662 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6663 * loose cycle time. This isn't too big a deal, since the loss will be
6664 * uniform across all VCPUs (not to mention the scenario is extremely
6665 * unlikely). It is possible that a second hibernate recovery happens
6666 * much faster than a first, causing the observed TSC here to be
6667 * smaller; this would require additional padding adjustment, which is
6668 * why we set last_host_tsc to the local tsc observed here.
6670 * N.B. - this code below runs only on platforms with reliable TSC,
6671 * as that is the only way backwards_tsc is set above. Also note
6672 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6673 * have the same delta_cyc adjustment applied if backwards_tsc
6674 * is detected. Note further, this adjustment is only done once,
6675 * as we reset last_host_tsc on all VCPUs to stop this from being
6676 * called multiple times (one for each physical CPU bringup).
6678 * Platforms with unreliable TSCs don't have to deal with this, they
6679 * will be compensated by the logic in vcpu_load, which sets the TSC to
6680 * catchup mode. This will catchup all VCPUs to real time, but cannot
6681 * guarantee that they stay in perfect synchronization.
6683 if (backwards_tsc
) {
6684 u64 delta_cyc
= max_tsc
- local_tsc
;
6685 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6686 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6687 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
6688 vcpu
->arch
.last_host_tsc
= local_tsc
;
6689 set_bit(KVM_REQ_MASTERCLOCK_UPDATE
,
6694 * We have to disable TSC offset matching.. if you were
6695 * booting a VM while issuing an S4 host suspend....
6696 * you may have some problem. Solving this issue is
6697 * left as an exercise to the reader.
6699 kvm
->arch
.last_tsc_nsec
= 0;
6700 kvm
->arch
.last_tsc_write
= 0;
6707 void kvm_arch_hardware_disable(void *garbage
)
6709 kvm_x86_ops
->hardware_disable(garbage
);
6710 drop_user_return_notifiers(garbage
);
6713 int kvm_arch_hardware_setup(void)
6715 return kvm_x86_ops
->hardware_setup();
6718 void kvm_arch_hardware_unsetup(void)
6720 kvm_x86_ops
->hardware_unsetup();
6723 void kvm_arch_check_processor_compat(void *rtn
)
6725 kvm_x86_ops
->check_processor_compatibility(rtn
);
6728 bool kvm_vcpu_compatible(struct kvm_vcpu
*vcpu
)
6730 return irqchip_in_kernel(vcpu
->kvm
) == (vcpu
->arch
.apic
!= NULL
);
6733 struct static_key kvm_no_apic_vcpu __read_mostly
;
6735 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
6741 BUG_ON(vcpu
->kvm
== NULL
);
6744 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
6745 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
6746 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
6748 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
6750 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
6755 vcpu
->arch
.pio_data
= page_address(page
);
6757 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
6759 r
= kvm_mmu_create(vcpu
);
6761 goto fail_free_pio_data
;
6763 if (irqchip_in_kernel(kvm
)) {
6764 r
= kvm_create_lapic(vcpu
);
6766 goto fail_mmu_destroy
;
6768 static_key_slow_inc(&kvm_no_apic_vcpu
);
6770 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
6772 if (!vcpu
->arch
.mce_banks
) {
6774 goto fail_free_lapic
;
6776 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
6778 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
)) {
6780 goto fail_free_mce_banks
;
6785 goto fail_free_wbinvd_dirty_mask
;
6787 vcpu
->arch
.ia32_tsc_adjust_msr
= 0x0;
6788 vcpu
->arch
.pv_time_enabled
= false;
6789 kvm_async_pf_hash_reset(vcpu
);
6793 fail_free_wbinvd_dirty_mask
:
6794 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
6795 fail_free_mce_banks
:
6796 kfree(vcpu
->arch
.mce_banks
);
6798 kvm_free_lapic(vcpu
);
6800 kvm_mmu_destroy(vcpu
);
6802 free_page((unsigned long)vcpu
->arch
.pio_data
);
6807 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
6811 kvm_pmu_destroy(vcpu
);
6812 kfree(vcpu
->arch
.mce_banks
);
6813 kvm_free_lapic(vcpu
);
6814 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
6815 kvm_mmu_destroy(vcpu
);
6816 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
6817 free_page((unsigned long)vcpu
->arch
.pio_data
);
6818 if (!irqchip_in_kernel(vcpu
->kvm
))
6819 static_key_slow_dec(&kvm_no_apic_vcpu
);
6822 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
6827 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
6828 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
6830 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6831 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
6832 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6833 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
6834 &kvm
->arch
.irq_sources_bitmap
);
6836 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
6837 mutex_init(&kvm
->arch
.apic_map_lock
);
6838 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
6840 pvclock_update_vm_gtod_copy(kvm
);
6845 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
6848 r
= vcpu_load(vcpu
);
6850 kvm_mmu_unload(vcpu
);
6854 static void kvm_free_vcpus(struct kvm
*kvm
)
6857 struct kvm_vcpu
*vcpu
;
6860 * Unpin any mmu pages first.
6862 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6863 kvm_clear_async_pf_completion_queue(vcpu
);
6864 kvm_unload_vcpu_mmu(vcpu
);
6866 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6867 kvm_arch_vcpu_free(vcpu
);
6869 mutex_lock(&kvm
->lock
);
6870 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
6871 kvm
->vcpus
[i
] = NULL
;
6873 atomic_set(&kvm
->online_vcpus
, 0);
6874 mutex_unlock(&kvm
->lock
);
6877 void kvm_arch_sync_events(struct kvm
*kvm
)
6879 kvm_free_all_assigned_devices(kvm
);
6883 void kvm_arch_destroy_vm(struct kvm
*kvm
)
6885 if (current
->mm
== kvm
->mm
) {
6887 * Free memory regions allocated on behalf of userspace,
6888 * unless the the memory map has changed due to process exit
6891 struct kvm_userspace_memory_region mem
;
6892 memset(&mem
, 0, sizeof(mem
));
6893 mem
.slot
= APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
;
6894 kvm_set_memory_region(kvm
, &mem
);
6896 mem
.slot
= IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
;
6897 kvm_set_memory_region(kvm
, &mem
);
6899 mem
.slot
= TSS_PRIVATE_MEMSLOT
;
6900 kvm_set_memory_region(kvm
, &mem
);
6902 kvm_iommu_unmap_guest(kvm
);
6903 kfree(kvm
->arch
.vpic
);
6904 kfree(kvm
->arch
.vioapic
);
6905 kvm_free_vcpus(kvm
);
6906 if (kvm
->arch
.apic_access_page
)
6907 put_page(kvm
->arch
.apic_access_page
);
6908 if (kvm
->arch
.ept_identity_pagetable
)
6909 put_page(kvm
->arch
.ept_identity_pagetable
);
6910 kfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
6913 void kvm_arch_free_memslot(struct kvm_memory_slot
*free
,
6914 struct kvm_memory_slot
*dont
)
6918 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6919 if (!dont
|| free
->arch
.rmap
[i
] != dont
->arch
.rmap
[i
]) {
6920 kvm_kvfree(free
->arch
.rmap
[i
]);
6921 free
->arch
.rmap
[i
] = NULL
;
6926 if (!dont
|| free
->arch
.lpage_info
[i
- 1] !=
6927 dont
->arch
.lpage_info
[i
- 1]) {
6928 kvm_kvfree(free
->arch
.lpage_info
[i
- 1]);
6929 free
->arch
.lpage_info
[i
- 1] = NULL
;
6934 int kvm_arch_create_memslot(struct kvm_memory_slot
*slot
, unsigned long npages
)
6938 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6943 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
6944 slot
->base_gfn
, level
) + 1;
6946 slot
->arch
.rmap
[i
] =
6947 kvm_kvzalloc(lpages
* sizeof(*slot
->arch
.rmap
[i
]));
6948 if (!slot
->arch
.rmap
[i
])
6953 slot
->arch
.lpage_info
[i
- 1] = kvm_kvzalloc(lpages
*
6954 sizeof(*slot
->arch
.lpage_info
[i
- 1]));
6955 if (!slot
->arch
.lpage_info
[i
- 1])
6958 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
6959 slot
->arch
.lpage_info
[i
- 1][0].write_count
= 1;
6960 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
6961 slot
->arch
.lpage_info
[i
- 1][lpages
- 1].write_count
= 1;
6962 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
6964 * If the gfn and userspace address are not aligned wrt each
6965 * other, or if explicitly asked to, disable large page
6966 * support for this slot
6968 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
6969 !kvm_largepages_enabled()) {
6972 for (j
= 0; j
< lpages
; ++j
)
6973 slot
->arch
.lpage_info
[i
- 1][j
].write_count
= 1;
6980 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
6981 kvm_kvfree(slot
->arch
.rmap
[i
]);
6982 slot
->arch
.rmap
[i
] = NULL
;
6986 kvm_kvfree(slot
->arch
.lpage_info
[i
- 1]);
6987 slot
->arch
.lpage_info
[i
- 1] = NULL
;
6992 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
6993 struct kvm_memory_slot
*memslot
,
6994 struct kvm_userspace_memory_region
*mem
,
6995 enum kvm_mr_change change
)
6998 * Only private memory slots need to be mapped here since
6999 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7001 if ((memslot
->id
>= KVM_USER_MEM_SLOTS
) && (change
== KVM_MR_CREATE
)) {
7002 unsigned long userspace_addr
;
7005 * MAP_SHARED to prevent internal slot pages from being moved
7008 userspace_addr
= vm_mmap(NULL
, 0, memslot
->npages
* PAGE_SIZE
,
7009 PROT_READ
| PROT_WRITE
,
7010 MAP_SHARED
| MAP_ANONYMOUS
, 0);
7012 if (IS_ERR((void *)userspace_addr
))
7013 return PTR_ERR((void *)userspace_addr
);
7015 memslot
->userspace_addr
= userspace_addr
;
7021 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
7022 struct kvm_userspace_memory_region
*mem
,
7023 const struct kvm_memory_slot
*old
,
7024 enum kvm_mr_change change
)
7027 int nr_mmu_pages
= 0;
7029 if ((mem
->slot
>= KVM_USER_MEM_SLOTS
) && (change
== KVM_MR_DELETE
)) {
7032 ret
= vm_munmap(old
->userspace_addr
,
7033 old
->npages
* PAGE_SIZE
);
7036 "kvm_vm_ioctl_set_memory_region: "
7037 "failed to munmap memory\n");
7040 if (!kvm
->arch
.n_requested_mmu_pages
)
7041 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
7044 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
7046 * Write protect all pages for dirty logging.
7047 * Existing largepage mappings are destroyed here and new ones will
7048 * not be created until the end of the logging.
7050 if ((change
!= KVM_MR_DELETE
) && (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
7051 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
7053 * If memory slot is created, or moved, we need to clear all
7056 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
7057 kvm_mmu_zap_mmio_sptes(kvm
);
7058 kvm_reload_remote_mmus(kvm
);
7062 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
7064 kvm_mmu_zap_all(kvm
);
7065 kvm_reload_remote_mmus(kvm
);
7068 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
7069 struct kvm_memory_slot
*slot
)
7071 kvm_arch_flush_shadow_all(kvm
);
7074 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
7076 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
7077 !vcpu
->arch
.apf
.halted
)
7078 || !list_empty_careful(&vcpu
->async_pf
.done
)
7079 || kvm_apic_has_events(vcpu
)
7080 || atomic_read(&vcpu
->arch
.nmi_queued
) ||
7081 (kvm_arch_interrupt_allowed(vcpu
) &&
7082 kvm_cpu_has_interrupt(vcpu
));
7085 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
7087 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
7090 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
7092 return kvm_x86_ops
->interrupt_allowed(vcpu
);
7095 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
7097 unsigned long current_rip
= kvm_rip_read(vcpu
) +
7098 get_segment_base(vcpu
, VCPU_SREG_CS
);
7100 return current_rip
== linear_rip
;
7102 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
7104 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
7106 unsigned long rflags
;
7108 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
7109 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
7110 rflags
&= ~X86_EFLAGS_TF
;
7113 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
7115 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
7117 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
7118 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
7119 rflags
|= X86_EFLAGS_TF
;
7120 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
7121 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7123 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
7125 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
7129 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
7130 is_error_page(work
->page
))
7133 r
= kvm_mmu_reload(vcpu
);
7137 if (!vcpu
->arch
.mmu
.direct_map
&&
7138 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
7141 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
7144 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
7146 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
7149 static inline u32
kvm_async_pf_next_probe(u32 key
)
7151 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
7154 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7156 u32 key
= kvm_async_pf_hash_fn(gfn
);
7158 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
7159 key
= kvm_async_pf_next_probe(key
);
7161 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
7164 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7167 u32 key
= kvm_async_pf_hash_fn(gfn
);
7169 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
7170 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
7171 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
7172 key
= kvm_async_pf_next_probe(key
);
7177 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7179 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
7182 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
7186 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
7188 vcpu
->arch
.apf
.gfns
[i
] = ~0;
7190 j
= kvm_async_pf_next_probe(j
);
7191 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
7193 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
7195 * k lies cyclically in ]i,j]
7197 * |....j i.k.| or |.k..j i...|
7199 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
7200 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
7205 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
7208 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
7212 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
7213 struct kvm_async_pf
*work
)
7215 struct x86_exception fault
;
7217 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
7218 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
7220 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
7221 (vcpu
->arch
.apf
.send_user_only
&&
7222 kvm_x86_ops
->get_cpl(vcpu
) == 0))
7223 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
7224 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
7225 fault
.vector
= PF_VECTOR
;
7226 fault
.error_code_valid
= true;
7227 fault
.error_code
= 0;
7228 fault
.nested_page_fault
= false;
7229 fault
.address
= work
->arch
.token
;
7230 kvm_inject_page_fault(vcpu
, &fault
);
7234 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
7235 struct kvm_async_pf
*work
)
7237 struct x86_exception fault
;
7239 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
7240 if (is_error_page(work
->page
))
7241 work
->arch
.token
= ~0; /* broadcast wakeup */
7243 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
7245 if ((vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) &&
7246 !apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
7247 fault
.vector
= PF_VECTOR
;
7248 fault
.error_code_valid
= true;
7249 fault
.error_code
= 0;
7250 fault
.nested_page_fault
= false;
7251 fault
.address
= work
->arch
.token
;
7252 kvm_inject_page_fault(vcpu
, &fault
);
7254 vcpu
->arch
.apf
.halted
= false;
7255 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7258 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
7260 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
7263 return !kvm_event_needs_reinjection(vcpu
) &&
7264 kvm_x86_ops
->interrupt_allowed(vcpu
);
7267 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
7268 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
7269 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
7270 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
7271 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
7272 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
7273 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
7274 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
7275 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
7276 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
7277 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
7278 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);