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
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40 #include <linux/user-return-notifier.h>
41 #include <linux/srcu.h>
42 #include <trace/events/kvm.h>
43 #undef TRACE_INCLUDE_FILE
44 #define CREATE_TRACE_POINTS
47 #include <asm/debugreg.h>
48 #include <asm/uaccess.h>
54 #define MAX_IO_MSRS 256
55 #define CR0_RESERVED_BITS \
56 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
57 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
58 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
59 #define CR4_RESERVED_BITS \
60 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
61 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
62 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
63 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
65 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
67 #define KVM_MAX_MCE_BANKS 32
68 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
71 * - enable syscall per default because its emulated by KVM
72 * - enable LME and LMA per default on 64 bit KVM
75 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffafeULL
;
77 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffffeULL
;
80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
83 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
);
84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
85 struct kvm_cpuid_entry2 __user
*entries
);
87 struct kvm_x86_ops
*kvm_x86_ops
;
88 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
91 module_param_named(ignore_msrs
, ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
93 #define KVM_NR_SHARED_MSRS 16
95 struct kvm_shared_msrs_global
{
97 u32 msrs
[KVM_NR_SHARED_MSRS
];
100 struct kvm_shared_msrs
{
101 struct user_return_notifier urn
;
103 struct kvm_shared_msr_values
{
106 } values
[KVM_NR_SHARED_MSRS
];
109 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
110 static DEFINE_PER_CPU(struct kvm_shared_msrs
, shared_msrs
);
112 struct kvm_stats_debugfs_item debugfs_entries
[] = {
113 { "pf_fixed", VCPU_STAT(pf_fixed
) },
114 { "pf_guest", VCPU_STAT(pf_guest
) },
115 { "tlb_flush", VCPU_STAT(tlb_flush
) },
116 { "invlpg", VCPU_STAT(invlpg
) },
117 { "exits", VCPU_STAT(exits
) },
118 { "io_exits", VCPU_STAT(io_exits
) },
119 { "mmio_exits", VCPU_STAT(mmio_exits
) },
120 { "signal_exits", VCPU_STAT(signal_exits
) },
121 { "irq_window", VCPU_STAT(irq_window_exits
) },
122 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
123 { "halt_exits", VCPU_STAT(halt_exits
) },
124 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
125 { "hypercalls", VCPU_STAT(hypercalls
) },
126 { "request_irq", VCPU_STAT(request_irq_exits
) },
127 { "irq_exits", VCPU_STAT(irq_exits
) },
128 { "host_state_reload", VCPU_STAT(host_state_reload
) },
129 { "efer_reload", VCPU_STAT(efer_reload
) },
130 { "fpu_reload", VCPU_STAT(fpu_reload
) },
131 { "insn_emulation", VCPU_STAT(insn_emulation
) },
132 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
133 { "irq_injections", VCPU_STAT(irq_injections
) },
134 { "nmi_injections", VCPU_STAT(nmi_injections
) },
135 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
136 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
137 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
138 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
139 { "mmu_flooded", VM_STAT(mmu_flooded
) },
140 { "mmu_recycled", VM_STAT(mmu_recycled
) },
141 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
142 { "mmu_unsync", VM_STAT(mmu_unsync
) },
143 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
144 { "largepages", VM_STAT(lpages
) },
148 static void kvm_on_user_return(struct user_return_notifier
*urn
)
151 struct kvm_shared_msrs
*locals
152 = container_of(urn
, struct kvm_shared_msrs
, urn
);
153 struct kvm_shared_msr_values
*values
;
155 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
156 values
= &locals
->values
[slot
];
157 if (values
->host
!= values
->curr
) {
158 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
159 values
->curr
= values
->host
;
162 locals
->registered
= false;
163 user_return_notifier_unregister(urn
);
166 static void shared_msr_update(unsigned slot
, u32 msr
)
168 struct kvm_shared_msrs
*smsr
;
171 smsr
= &__get_cpu_var(shared_msrs
);
172 /* only read, and nobody should modify it at this time,
173 * so don't need lock */
174 if (slot
>= shared_msrs_global
.nr
) {
175 printk(KERN_ERR
"kvm: invalid MSR slot!");
178 rdmsrl_safe(msr
, &value
);
179 smsr
->values
[slot
].host
= value
;
180 smsr
->values
[slot
].curr
= value
;
183 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
185 if (slot
>= shared_msrs_global
.nr
)
186 shared_msrs_global
.nr
= slot
+ 1;
187 shared_msrs_global
.msrs
[slot
] = msr
;
188 /* we need ensured the shared_msr_global have been updated */
191 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
193 static void kvm_shared_msr_cpu_online(void)
197 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
198 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
201 void kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
203 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
205 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
207 smsr
->values
[slot
].curr
= value
;
208 wrmsrl(shared_msrs_global
.msrs
[slot
], value
);
209 if (!smsr
->registered
) {
210 smsr
->urn
.on_user_return
= kvm_on_user_return
;
211 user_return_notifier_register(&smsr
->urn
);
212 smsr
->registered
= true;
215 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
217 static void drop_user_return_notifiers(void *ignore
)
219 struct kvm_shared_msrs
*smsr
= &__get_cpu_var(shared_msrs
);
221 if (smsr
->registered
)
222 kvm_on_user_return(&smsr
->urn
);
225 unsigned long segment_base(u16 selector
)
227 struct descriptor_table gdt
;
228 struct desc_struct
*d
;
229 unsigned long table_base
;
236 table_base
= gdt
.base
;
238 if (selector
& 4) { /* from ldt */
239 u16 ldt_selector
= kvm_read_ldt();
241 table_base
= segment_base(ldt_selector
);
243 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
244 v
= get_desc_base(d
);
246 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
247 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
251 EXPORT_SYMBOL_GPL(segment_base
);
253 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
255 if (irqchip_in_kernel(vcpu
->kvm
))
256 return vcpu
->arch
.apic_base
;
258 return vcpu
->arch
.apic_base
;
260 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
262 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
264 /* TODO: reserve bits check */
265 if (irqchip_in_kernel(vcpu
->kvm
))
266 kvm_lapic_set_base(vcpu
, data
);
268 vcpu
->arch
.apic_base
= data
;
270 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
272 #define EXCPT_BENIGN 0
273 #define EXCPT_CONTRIBUTORY 1
276 static int exception_class(int vector
)
286 return EXCPT_CONTRIBUTORY
;
293 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
294 unsigned nr
, bool has_error
, u32 error_code
)
299 if (!vcpu
->arch
.exception
.pending
) {
301 vcpu
->arch
.exception
.pending
= true;
302 vcpu
->arch
.exception
.has_error_code
= has_error
;
303 vcpu
->arch
.exception
.nr
= nr
;
304 vcpu
->arch
.exception
.error_code
= error_code
;
308 /* to check exception */
309 prev_nr
= vcpu
->arch
.exception
.nr
;
310 if (prev_nr
== DF_VECTOR
) {
311 /* triple fault -> shutdown */
312 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
315 class1
= exception_class(prev_nr
);
316 class2
= exception_class(nr
);
317 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
318 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
319 /* generate double fault per SDM Table 5-5 */
320 vcpu
->arch
.exception
.pending
= true;
321 vcpu
->arch
.exception
.has_error_code
= true;
322 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
323 vcpu
->arch
.exception
.error_code
= 0;
325 /* replace previous exception with a new one in a hope
326 that instruction re-execution will regenerate lost
331 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
333 kvm_multiple_exception(vcpu
, nr
, false, 0);
335 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
337 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
340 ++vcpu
->stat
.pf_guest
;
341 vcpu
->arch
.cr2
= addr
;
342 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
345 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
347 vcpu
->arch
.nmi_pending
= 1;
349 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
351 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
353 kvm_multiple_exception(vcpu
, nr
, true, error_code
);
355 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
358 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
359 * a #GP and return false.
361 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
363 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
365 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
368 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
371 * Load the pae pdptrs. Return true is they are all valid.
373 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
375 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
376 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
379 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
381 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
382 offset
* sizeof(u64
), sizeof(pdpte
));
387 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
388 if (is_present_gpte(pdpte
[i
]) &&
389 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
396 memcpy(vcpu
->arch
.pdptrs
, pdpte
, sizeof(vcpu
->arch
.pdptrs
));
397 __set_bit(VCPU_EXREG_PDPTR
,
398 (unsigned long *)&vcpu
->arch
.regs_avail
);
399 __set_bit(VCPU_EXREG_PDPTR
,
400 (unsigned long *)&vcpu
->arch
.regs_dirty
);
405 EXPORT_SYMBOL_GPL(load_pdptrs
);
407 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
409 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
413 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
416 if (!test_bit(VCPU_EXREG_PDPTR
,
417 (unsigned long *)&vcpu
->arch
.regs_avail
))
420 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->arch
.cr3
& ~31u, pdpte
, sizeof(pdpte
));
423 changed
= memcmp(pdpte
, vcpu
->arch
.pdptrs
, sizeof(pdpte
)) != 0;
429 void kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
431 if (cr0
& CR0_RESERVED_BITS
) {
432 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
433 cr0
, vcpu
->arch
.cr0
);
434 kvm_inject_gp(vcpu
, 0);
438 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
439 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
440 kvm_inject_gp(vcpu
, 0);
444 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
445 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
446 "and a clear PE flag\n");
447 kvm_inject_gp(vcpu
, 0);
451 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
453 if ((vcpu
->arch
.shadow_efer
& EFER_LME
)) {
457 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
458 "in long mode while PAE is disabled\n");
459 kvm_inject_gp(vcpu
, 0);
462 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
464 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
465 "in long mode while CS.L == 1\n");
466 kvm_inject_gp(vcpu
, 0);
472 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
473 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
475 kvm_inject_gp(vcpu
, 0);
481 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
482 vcpu
->arch
.cr0
= cr0
;
484 kvm_mmu_reset_context(vcpu
);
487 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
489 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
491 kvm_set_cr0(vcpu
, (vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f));
493 EXPORT_SYMBOL_GPL(kvm_lmsw
);
495 void kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
497 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
498 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
;
500 if (cr4
& CR4_RESERVED_BITS
) {
501 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
502 kvm_inject_gp(vcpu
, 0);
506 if (is_long_mode(vcpu
)) {
507 if (!(cr4
& X86_CR4_PAE
)) {
508 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
510 kvm_inject_gp(vcpu
, 0);
513 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
514 && ((cr4
^ old_cr4
) & pdptr_bits
)
515 && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
516 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
517 kvm_inject_gp(vcpu
, 0);
521 if (cr4
& X86_CR4_VMXE
) {
522 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
523 kvm_inject_gp(vcpu
, 0);
526 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
527 vcpu
->arch
.cr4
= cr4
;
528 vcpu
->arch
.mmu
.base_role
.cr4_pge
= (cr4
& X86_CR4_PGE
) && !tdp_enabled
;
529 kvm_mmu_reset_context(vcpu
);
531 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
533 void kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
535 if (cr3
== vcpu
->arch
.cr3
&& !pdptrs_changed(vcpu
)) {
536 kvm_mmu_sync_roots(vcpu
);
537 kvm_mmu_flush_tlb(vcpu
);
541 if (is_long_mode(vcpu
)) {
542 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
543 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
544 kvm_inject_gp(vcpu
, 0);
549 if (cr3
& CR3_PAE_RESERVED_BITS
) {
551 "set_cr3: #GP, reserved bits\n");
552 kvm_inject_gp(vcpu
, 0);
555 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
556 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
558 kvm_inject_gp(vcpu
, 0);
563 * We don't check reserved bits in nonpae mode, because
564 * this isn't enforced, and VMware depends on this.
569 * Does the new cr3 value map to physical memory? (Note, we
570 * catch an invalid cr3 even in real-mode, because it would
571 * cause trouble later on when we turn on paging anyway.)
573 * A real CPU would silently accept an invalid cr3 and would
574 * attempt to use it - with largely undefined (and often hard
575 * to debug) behavior on the guest side.
577 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
578 kvm_inject_gp(vcpu
, 0);
580 vcpu
->arch
.cr3
= cr3
;
581 vcpu
->arch
.mmu
.new_cr3(vcpu
);
584 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
586 void kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
588 if (cr8
& CR8_RESERVED_BITS
) {
589 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
590 kvm_inject_gp(vcpu
, 0);
593 if (irqchip_in_kernel(vcpu
->kvm
))
594 kvm_lapic_set_tpr(vcpu
, cr8
);
596 vcpu
->arch
.cr8
= cr8
;
598 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
600 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
602 if (irqchip_in_kernel(vcpu
->kvm
))
603 return kvm_lapic_get_cr8(vcpu
);
605 return vcpu
->arch
.cr8
;
607 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
609 static inline u32
bit(int bitno
)
611 return 1 << (bitno
& 31);
615 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
616 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
618 * This list is modified at module load time to reflect the
619 * capabilities of the host cpu. This capabilities test skips MSRs that are
620 * kvm-specific. Those are put in the beginning of the list.
623 #define KVM_SAVE_MSRS_BEGIN 2
624 static u32 msrs_to_save
[] = {
625 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
626 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
629 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
631 MSR_IA32_TSC
, MSR_IA32_PERF_STATUS
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
634 static unsigned num_msrs_to_save
;
636 static u32 emulated_msrs
[] = {
637 MSR_IA32_MISC_ENABLE
,
640 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
642 if (efer
& efer_reserved_bits
) {
643 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
645 kvm_inject_gp(vcpu
, 0);
650 && (vcpu
->arch
.shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
651 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
652 kvm_inject_gp(vcpu
, 0);
656 if (efer
& EFER_FFXSR
) {
657 struct kvm_cpuid_entry2
*feat
;
659 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
660 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
))) {
661 printk(KERN_DEBUG
"set_efer: #GP, enable FFXSR w/o CPUID capability\n");
662 kvm_inject_gp(vcpu
, 0);
667 if (efer
& EFER_SVME
) {
668 struct kvm_cpuid_entry2
*feat
;
670 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
671 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
))) {
672 printk(KERN_DEBUG
"set_efer: #GP, enable SVM w/o SVM\n");
673 kvm_inject_gp(vcpu
, 0);
678 kvm_x86_ops
->set_efer(vcpu
, efer
);
681 efer
|= vcpu
->arch
.shadow_efer
& EFER_LMA
;
683 vcpu
->arch
.shadow_efer
= efer
;
685 vcpu
->arch
.mmu
.base_role
.nxe
= (efer
& EFER_NX
) && !tdp_enabled
;
686 kvm_mmu_reset_context(vcpu
);
689 void kvm_enable_efer_bits(u64 mask
)
691 efer_reserved_bits
&= ~mask
;
693 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
697 * Writes msr value into into the appropriate "register".
698 * Returns 0 on success, non-0 otherwise.
699 * Assumes vcpu_load() was already called.
701 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
703 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
707 * Adapt set_msr() to msr_io()'s calling convention
709 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
711 return kvm_set_msr(vcpu
, index
, *data
);
714 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
717 struct pvclock_wall_clock wc
;
718 struct timespec boot
;
725 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
728 * The guest calculates current wall clock time by adding
729 * system time (updated by kvm_write_guest_time below) to the
730 * wall clock specified here. guest system time equals host
731 * system time for us, thus we must fill in host boot time here.
735 wc
.sec
= boot
.tv_sec
;
736 wc
.nsec
= boot
.tv_nsec
;
737 wc
.version
= version
;
739 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
742 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
745 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
747 uint32_t quotient
, remainder
;
749 /* Don't try to replace with do_div(), this one calculates
750 * "(dividend << 32) / divisor" */
752 : "=a" (quotient
), "=d" (remainder
)
753 : "0" (0), "1" (dividend
), "r" (divisor
) );
757 static void kvm_set_time_scale(uint32_t tsc_khz
, struct pvclock_vcpu_time_info
*hv_clock
)
759 uint64_t nsecs
= 1000000000LL;
764 tps64
= tsc_khz
* 1000LL;
765 while (tps64
> nsecs
*2) {
770 tps32
= (uint32_t)tps64
;
771 while (tps32
<= (uint32_t)nsecs
) {
776 hv_clock
->tsc_shift
= shift
;
777 hv_clock
->tsc_to_system_mul
= div_frac(nsecs
, tps32
);
779 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
780 __func__
, tsc_khz
, hv_clock
->tsc_shift
,
781 hv_clock
->tsc_to_system_mul
);
784 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
786 static void kvm_write_guest_time(struct kvm_vcpu
*v
)
790 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
792 unsigned long this_tsc_khz
;
794 if ((!vcpu
->time_page
))
797 this_tsc_khz
= get_cpu_var(cpu_tsc_khz
);
798 if (unlikely(vcpu
->hv_clock_tsc_khz
!= this_tsc_khz
)) {
799 kvm_set_time_scale(this_tsc_khz
, &vcpu
->hv_clock
);
800 vcpu
->hv_clock_tsc_khz
= this_tsc_khz
;
802 put_cpu_var(cpu_tsc_khz
);
804 /* Keep irq disabled to prevent changes to the clock */
805 local_irq_save(flags
);
806 kvm_get_msr(v
, MSR_IA32_TSC
, &vcpu
->hv_clock
.tsc_timestamp
);
808 monotonic_to_bootbased(&ts
);
809 local_irq_restore(flags
);
811 /* With all the info we got, fill in the values */
813 vcpu
->hv_clock
.system_time
= ts
.tv_nsec
+
814 (NSEC_PER_SEC
* (u64
)ts
.tv_sec
) + v
->kvm
->arch
.kvmclock_offset
;
817 * The interface expects us to write an even number signaling that the
818 * update is finished. Since the guest won't see the intermediate
819 * state, we just increase by 2 at the end.
821 vcpu
->hv_clock
.version
+= 2;
823 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
825 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
826 sizeof(vcpu
->hv_clock
));
828 kunmap_atomic(shared_kaddr
, KM_USER0
);
830 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
833 static int kvm_request_guest_time_update(struct kvm_vcpu
*v
)
835 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
837 if (!vcpu
->time_page
)
839 set_bit(KVM_REQ_KVMCLOCK_UPDATE
, &v
->requests
);
843 static bool msr_mtrr_valid(unsigned msr
)
846 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
847 case MSR_MTRRfix64K_00000
:
848 case MSR_MTRRfix16K_80000
:
849 case MSR_MTRRfix16K_A0000
:
850 case MSR_MTRRfix4K_C0000
:
851 case MSR_MTRRfix4K_C8000
:
852 case MSR_MTRRfix4K_D0000
:
853 case MSR_MTRRfix4K_D8000
:
854 case MSR_MTRRfix4K_E0000
:
855 case MSR_MTRRfix4K_E8000
:
856 case MSR_MTRRfix4K_F0000
:
857 case MSR_MTRRfix4K_F8000
:
858 case MSR_MTRRdefType
:
859 case MSR_IA32_CR_PAT
:
867 static bool valid_pat_type(unsigned t
)
869 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
872 static bool valid_mtrr_type(unsigned t
)
874 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
877 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
881 if (!msr_mtrr_valid(msr
))
884 if (msr
== MSR_IA32_CR_PAT
) {
885 for (i
= 0; i
< 8; i
++)
886 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
889 } else if (msr
== MSR_MTRRdefType
) {
892 return valid_mtrr_type(data
& 0xff);
893 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
894 for (i
= 0; i
< 8 ; i
++)
895 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
901 return valid_mtrr_type(data
& 0xff);
904 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
906 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
908 if (!mtrr_valid(vcpu
, msr
, data
))
911 if (msr
== MSR_MTRRdefType
) {
912 vcpu
->arch
.mtrr_state
.def_type
= data
;
913 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
914 } else if (msr
== MSR_MTRRfix64K_00000
)
916 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
917 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
918 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
919 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
920 else if (msr
== MSR_IA32_CR_PAT
)
921 vcpu
->arch
.pat
= data
;
922 else { /* Variable MTRRs */
923 int idx
, is_mtrr_mask
;
926 idx
= (msr
- 0x200) / 2;
927 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
930 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
933 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
937 kvm_mmu_reset_context(vcpu
);
941 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
943 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
944 unsigned bank_num
= mcg_cap
& 0xff;
947 case MSR_IA32_MCG_STATUS
:
948 vcpu
->arch
.mcg_status
= data
;
950 case MSR_IA32_MCG_CTL
:
951 if (!(mcg_cap
& MCG_CTL_P
))
953 if (data
!= 0 && data
!= ~(u64
)0)
955 vcpu
->arch
.mcg_ctl
= data
;
958 if (msr
>= MSR_IA32_MC0_CTL
&&
959 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
960 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
961 /* only 0 or all 1s can be written to IA32_MCi_CTL */
962 if ((offset
& 0x3) == 0 &&
963 data
!= 0 && data
!= ~(u64
)0)
965 vcpu
->arch
.mce_banks
[offset
] = data
;
973 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
975 struct kvm
*kvm
= vcpu
->kvm
;
976 int lm
= is_long_mode(vcpu
);
977 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
978 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
979 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
980 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
981 u32 page_num
= data
& ~PAGE_MASK
;
982 u64 page_addr
= data
& PAGE_MASK
;
987 if (page_num
>= blob_size
)
990 page
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
994 if (copy_from_user(page
, blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
))
996 if (kvm_write_guest(kvm
, page_addr
, page
, PAGE_SIZE
))
1005 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1009 set_efer(vcpu
, data
);
1012 data
&= ~(u64
)0x40; /* ignore flush filter disable */
1014 pr_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
1019 case MSR_FAM10H_MMIO_CONF_BASE
:
1021 pr_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
1026 case MSR_AMD64_NB_CFG
:
1028 case MSR_IA32_DEBUGCTLMSR
:
1030 /* We support the non-activated case already */
1032 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
1033 /* Values other than LBR and BTF are vendor-specific,
1034 thus reserved and should throw a #GP */
1037 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1040 case MSR_IA32_UCODE_REV
:
1041 case MSR_IA32_UCODE_WRITE
:
1042 case MSR_VM_HSAVE_PA
:
1043 case MSR_AMD64_PATCH_LOADER
:
1045 case 0x200 ... 0x2ff:
1046 return set_msr_mtrr(vcpu
, msr
, data
);
1047 case MSR_IA32_APICBASE
:
1048 kvm_set_apic_base(vcpu
, data
);
1050 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1051 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
1052 case MSR_IA32_MISC_ENABLE
:
1053 vcpu
->arch
.ia32_misc_enable_msr
= data
;
1055 case MSR_KVM_WALL_CLOCK
:
1056 vcpu
->kvm
->arch
.wall_clock
= data
;
1057 kvm_write_wall_clock(vcpu
->kvm
, data
);
1059 case MSR_KVM_SYSTEM_TIME
: {
1060 if (vcpu
->arch
.time_page
) {
1061 kvm_release_page_dirty(vcpu
->arch
.time_page
);
1062 vcpu
->arch
.time_page
= NULL
;
1065 vcpu
->arch
.time
= data
;
1067 /* we verify if the enable bit is set... */
1071 /* ...but clean it before doing the actual write */
1072 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
1074 vcpu
->arch
.time_page
=
1075 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
1077 if (is_error_page(vcpu
->arch
.time_page
)) {
1078 kvm_release_page_clean(vcpu
->arch
.time_page
);
1079 vcpu
->arch
.time_page
= NULL
;
1082 kvm_request_guest_time_update(vcpu
);
1085 case MSR_IA32_MCG_CTL
:
1086 case MSR_IA32_MCG_STATUS
:
1087 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1088 return set_msr_mce(vcpu
, msr
, data
);
1090 /* Performance counters are not protected by a CPUID bit,
1091 * so we should check all of them in the generic path for the sake of
1092 * cross vendor migration.
1093 * Writing a zero into the event select MSRs disables them,
1094 * which we perfectly emulate ;-). Any other value should be at least
1095 * reported, some guests depend on them.
1097 case MSR_P6_EVNTSEL0
:
1098 case MSR_P6_EVNTSEL1
:
1099 case MSR_K7_EVNTSEL0
:
1100 case MSR_K7_EVNTSEL1
:
1101 case MSR_K7_EVNTSEL2
:
1102 case MSR_K7_EVNTSEL3
:
1104 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
1105 "0x%x data 0x%llx\n", msr
, data
);
1107 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1108 * so we ignore writes to make it happy.
1110 case MSR_P6_PERFCTR0
:
1111 case MSR_P6_PERFCTR1
:
1112 case MSR_K7_PERFCTR0
:
1113 case MSR_K7_PERFCTR1
:
1114 case MSR_K7_PERFCTR2
:
1115 case MSR_K7_PERFCTR3
:
1116 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
1117 "0x%x data 0x%llx\n", msr
, data
);
1120 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
1121 return xen_hvm_config(vcpu
, data
);
1123 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
1127 pr_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
1134 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1138 * Reads an msr value (of 'msr_index') into 'pdata'.
1139 * Returns 0 on success, non-0 otherwise.
1140 * Assumes vcpu_load() was already called.
1142 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1144 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
1147 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1149 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
1151 if (!msr_mtrr_valid(msr
))
1154 if (msr
== MSR_MTRRdefType
)
1155 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
1156 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
1157 else if (msr
== MSR_MTRRfix64K_00000
)
1159 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
1160 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
1161 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
1162 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
1163 else if (msr
== MSR_IA32_CR_PAT
)
1164 *pdata
= vcpu
->arch
.pat
;
1165 else { /* Variable MTRRs */
1166 int idx
, is_mtrr_mask
;
1169 idx
= (msr
- 0x200) / 2;
1170 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
1173 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
1176 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1183 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1186 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1187 unsigned bank_num
= mcg_cap
& 0xff;
1190 case MSR_IA32_P5_MC_ADDR
:
1191 case MSR_IA32_P5_MC_TYPE
:
1194 case MSR_IA32_MCG_CAP
:
1195 data
= vcpu
->arch
.mcg_cap
;
1197 case MSR_IA32_MCG_CTL
:
1198 if (!(mcg_cap
& MCG_CTL_P
))
1200 data
= vcpu
->arch
.mcg_ctl
;
1202 case MSR_IA32_MCG_STATUS
:
1203 data
= vcpu
->arch
.mcg_status
;
1206 if (msr
>= MSR_IA32_MC0_CTL
&&
1207 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1208 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1209 data
= vcpu
->arch
.mce_banks
[offset
];
1218 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1223 case MSR_IA32_PLATFORM_ID
:
1224 case MSR_IA32_UCODE_REV
:
1225 case MSR_IA32_EBL_CR_POWERON
:
1226 case MSR_IA32_DEBUGCTLMSR
:
1227 case MSR_IA32_LASTBRANCHFROMIP
:
1228 case MSR_IA32_LASTBRANCHTOIP
:
1229 case MSR_IA32_LASTINTFROMIP
:
1230 case MSR_IA32_LASTINTTOIP
:
1233 case MSR_VM_HSAVE_PA
:
1234 case MSR_P6_PERFCTR0
:
1235 case MSR_P6_PERFCTR1
:
1236 case MSR_P6_EVNTSEL0
:
1237 case MSR_P6_EVNTSEL1
:
1238 case MSR_K7_EVNTSEL0
:
1239 case MSR_K7_PERFCTR0
:
1240 case MSR_K8_INT_PENDING_MSG
:
1241 case MSR_AMD64_NB_CFG
:
1242 case MSR_FAM10H_MMIO_CONF_BASE
:
1246 data
= 0x500 | KVM_NR_VAR_MTRR
;
1248 case 0x200 ... 0x2ff:
1249 return get_msr_mtrr(vcpu
, msr
, pdata
);
1250 case 0xcd: /* fsb frequency */
1253 case MSR_IA32_APICBASE
:
1254 data
= kvm_get_apic_base(vcpu
);
1256 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1257 return kvm_x2apic_msr_read(vcpu
, msr
, pdata
);
1259 case MSR_IA32_MISC_ENABLE
:
1260 data
= vcpu
->arch
.ia32_misc_enable_msr
;
1262 case MSR_IA32_PERF_STATUS
:
1263 /* TSC increment by tick */
1265 /* CPU multiplier */
1266 data
|= (((uint64_t)4ULL) << 40);
1269 data
= vcpu
->arch
.shadow_efer
;
1271 case MSR_KVM_WALL_CLOCK
:
1272 data
= vcpu
->kvm
->arch
.wall_clock
;
1274 case MSR_KVM_SYSTEM_TIME
:
1275 data
= vcpu
->arch
.time
;
1277 case MSR_IA32_P5_MC_ADDR
:
1278 case MSR_IA32_P5_MC_TYPE
:
1279 case MSR_IA32_MCG_CAP
:
1280 case MSR_IA32_MCG_CTL
:
1281 case MSR_IA32_MCG_STATUS
:
1282 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1283 return get_msr_mce(vcpu
, msr
, pdata
);
1286 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
1289 pr_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr
);
1297 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1300 * Read or write a bunch of msrs. All parameters are kernel addresses.
1302 * @return number of msrs set successfully.
1304 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
1305 struct kvm_msr_entry
*entries
,
1306 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1307 unsigned index
, u64
*data
))
1313 down_read(&vcpu
->kvm
->slots_lock
);
1314 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1315 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1317 up_read(&vcpu
->kvm
->slots_lock
);
1325 * Read or write a bunch of msrs. Parameters are user addresses.
1327 * @return number of msrs set successfully.
1329 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
1330 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1331 unsigned index
, u64
*data
),
1334 struct kvm_msrs msrs
;
1335 struct kvm_msr_entry
*entries
;
1340 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1344 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1348 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1349 entries
= vmalloc(size
);
1354 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1357 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
1362 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1373 int kvm_dev_ioctl_check_extension(long ext
)
1378 case KVM_CAP_IRQCHIP
:
1380 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
1381 case KVM_CAP_SET_TSS_ADDR
:
1382 case KVM_CAP_EXT_CPUID
:
1383 case KVM_CAP_CLOCKSOURCE
:
1385 case KVM_CAP_NOP_IO_DELAY
:
1386 case KVM_CAP_MP_STATE
:
1387 case KVM_CAP_SYNC_MMU
:
1388 case KVM_CAP_REINJECT_CONTROL
:
1389 case KVM_CAP_IRQ_INJECT_STATUS
:
1390 case KVM_CAP_ASSIGN_DEV_IRQ
:
1392 case KVM_CAP_IOEVENTFD
:
1394 case KVM_CAP_PIT_STATE2
:
1395 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
1396 case KVM_CAP_XEN_HVM
:
1397 case KVM_CAP_ADJUST_CLOCK
:
1398 case KVM_CAP_VCPU_EVENTS
:
1401 case KVM_CAP_COALESCED_MMIO
:
1402 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
1405 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
1407 case KVM_CAP_NR_VCPUS
:
1410 case KVM_CAP_NR_MEMSLOTS
:
1411 r
= KVM_MEMORY_SLOTS
;
1413 case KVM_CAP_PV_MMU
: /* obsolete */
1420 r
= KVM_MAX_MCE_BANKS
;
1430 long kvm_arch_dev_ioctl(struct file
*filp
,
1431 unsigned int ioctl
, unsigned long arg
)
1433 void __user
*argp
= (void __user
*)arg
;
1437 case KVM_GET_MSR_INDEX_LIST
: {
1438 struct kvm_msr_list __user
*user_msr_list
= argp
;
1439 struct kvm_msr_list msr_list
;
1443 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1446 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
1447 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1450 if (n
< msr_list
.nmsrs
)
1453 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1454 num_msrs_to_save
* sizeof(u32
)))
1456 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
1458 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
1463 case KVM_GET_SUPPORTED_CPUID
: {
1464 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1465 struct kvm_cpuid2 cpuid
;
1468 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1470 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
1471 cpuid_arg
->entries
);
1476 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1481 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
1484 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
1486 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
1498 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1500 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
1501 if (unlikely(per_cpu(cpu_tsc_khz
, cpu
) == 0)) {
1502 unsigned long khz
= cpufreq_quick_get(cpu
);
1505 per_cpu(cpu_tsc_khz
, cpu
) = khz
;
1507 kvm_request_guest_time_update(vcpu
);
1510 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
1512 kvm_x86_ops
->vcpu_put(vcpu
);
1513 kvm_put_guest_fpu(vcpu
);
1516 static int is_efer_nx(void)
1518 unsigned long long efer
= 0;
1520 rdmsrl_safe(MSR_EFER
, &efer
);
1521 return efer
& EFER_NX
;
1524 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
1527 struct kvm_cpuid_entry2
*e
, *entry
;
1530 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
1531 e
= &vcpu
->arch
.cpuid_entries
[i
];
1532 if (e
->function
== 0x80000001) {
1537 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
1538 entry
->edx
&= ~(1 << 20);
1539 printk(KERN_INFO
"kvm: guest NX capability removed\n");
1543 /* when an old userspace process fills a new kernel module */
1544 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
1545 struct kvm_cpuid
*cpuid
,
1546 struct kvm_cpuid_entry __user
*entries
)
1549 struct kvm_cpuid_entry
*cpuid_entries
;
1552 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1555 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
1559 if (copy_from_user(cpuid_entries
, entries
,
1560 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
1562 for (i
= 0; i
< cpuid
->nent
; i
++) {
1563 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
1564 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
1565 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
1566 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
1567 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
1568 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
1569 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
1570 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
1571 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
1572 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
1574 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1575 cpuid_fix_nx_cap(vcpu
);
1577 kvm_apic_set_version(vcpu
);
1578 kvm_x86_ops
->cpuid_update(vcpu
);
1581 vfree(cpuid_entries
);
1586 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
1587 struct kvm_cpuid2
*cpuid
,
1588 struct kvm_cpuid_entry2 __user
*entries
)
1593 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1596 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
1597 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
1599 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1600 kvm_apic_set_version(vcpu
);
1601 kvm_x86_ops
->cpuid_update(vcpu
);
1608 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
1609 struct kvm_cpuid2
*cpuid
,
1610 struct kvm_cpuid_entry2 __user
*entries
)
1615 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
1618 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
1619 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
1624 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
1628 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1631 entry
->function
= function
;
1632 entry
->index
= index
;
1633 cpuid_count(entry
->function
, entry
->index
,
1634 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
1638 #define F(x) bit(X86_FEATURE_##x)
1640 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1641 u32 index
, int *nent
, int maxnent
)
1643 unsigned f_nx
= is_efer_nx() ? F(NX
) : 0;
1644 unsigned f_gbpages
= kvm_x86_ops
->gb_page_enable() ? F(GBPAGES
) : 0;
1645 #ifdef CONFIG_X86_64
1646 unsigned f_lm
= F(LM
);
1650 unsigned f_rdtscp
= kvm_x86_ops
->rdtscp_supported() ? F(RDTSCP
) : 0;
1653 const u32 kvm_supported_word0_x86_features
=
1654 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
1655 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
1656 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SEP
) |
1657 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
1658 F(PAT
) | F(PSE36
) | 0 /* PSN */ | F(CLFLSH
) |
1659 0 /* Reserved, DS, ACPI */ | F(MMX
) |
1660 F(FXSR
) | F(XMM
) | F(XMM2
) | F(SELFSNOOP
) |
1661 0 /* HTT, TM, Reserved, PBE */;
1662 /* cpuid 0x80000001.edx */
1663 const u32 kvm_supported_word1_x86_features
=
1664 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
1665 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
1666 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SYSCALL
) |
1667 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
1668 F(PAT
) | F(PSE36
) | 0 /* Reserved */ |
1669 f_nx
| 0 /* Reserved */ | F(MMXEXT
) | F(MMX
) |
1670 F(FXSR
) | F(FXSR_OPT
) | f_gbpages
| f_rdtscp
|
1671 0 /* Reserved */ | f_lm
| F(3DNOWEXT
) | F(3DNOW
);
1673 const u32 kvm_supported_word4_x86_features
=
1674 F(XMM3
) | 0 /* Reserved, DTES64, MONITOR */ |
1675 0 /* DS-CPL, VMX, SMX, EST */ |
1676 0 /* TM2 */ | F(SSSE3
) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1677 0 /* Reserved */ | F(CX16
) | 0 /* xTPR Update, PDCM */ |
1678 0 /* Reserved, DCA */ | F(XMM4_1
) |
1679 F(XMM4_2
) | F(X2APIC
) | F(MOVBE
) | F(POPCNT
) |
1680 0 /* Reserved, XSAVE, OSXSAVE */;
1681 /* cpuid 0x80000001.ecx */
1682 const u32 kvm_supported_word6_x86_features
=
1683 F(LAHF_LM
) | F(CMP_LEGACY
) | F(SVM
) | 0 /* ExtApicSpace */ |
1684 F(CR8_LEGACY
) | F(ABM
) | F(SSE4A
) | F(MISALIGNSSE
) |
1685 F(3DNOWPREFETCH
) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5
) |
1686 0 /* SKINIT */ | 0 /* WDT */;
1688 /* all calls to cpuid_count() should be made on the same cpu */
1690 do_cpuid_1_ent(entry
, function
, index
);
1695 entry
->eax
= min(entry
->eax
, (u32
)0xb);
1698 entry
->edx
&= kvm_supported_word0_x86_features
;
1699 entry
->ecx
&= kvm_supported_word4_x86_features
;
1700 /* we support x2apic emulation even if host does not support
1701 * it since we emulate x2apic in software */
1702 entry
->ecx
|= F(X2APIC
);
1704 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1705 * may return different values. This forces us to get_cpu() before
1706 * issuing the first command, and also to emulate this annoying behavior
1707 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1709 int t
, times
= entry
->eax
& 0xff;
1711 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1712 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
1713 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
1714 do_cpuid_1_ent(&entry
[t
], function
, 0);
1715 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1720 /* function 4 and 0xb have additional index. */
1724 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1725 /* read more entries until cache_type is zero */
1726 for (i
= 1; *nent
< maxnent
; ++i
) {
1727 cache_type
= entry
[i
- 1].eax
& 0x1f;
1730 do_cpuid_1_ent(&entry
[i
], function
, i
);
1732 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1740 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1741 /* read more entries until level_type is zero */
1742 for (i
= 1; *nent
< maxnent
; ++i
) {
1743 level_type
= entry
[i
- 1].ecx
& 0xff00;
1746 do_cpuid_1_ent(&entry
[i
], function
, i
);
1748 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1754 entry
->eax
= min(entry
->eax
, 0x8000001a);
1757 entry
->edx
&= kvm_supported_word1_x86_features
;
1758 entry
->ecx
&= kvm_supported_word6_x86_features
;
1766 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
1767 struct kvm_cpuid_entry2 __user
*entries
)
1769 struct kvm_cpuid_entry2
*cpuid_entries
;
1770 int limit
, nent
= 0, r
= -E2BIG
;
1773 if (cpuid
->nent
< 1)
1775 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1776 cpuid
->nent
= KVM_MAX_CPUID_ENTRIES
;
1778 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
1782 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
1783 limit
= cpuid_entries
[0].eax
;
1784 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1785 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1786 &nent
, cpuid
->nent
);
1788 if (nent
>= cpuid
->nent
)
1791 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1792 limit
= cpuid_entries
[nent
- 1].eax
;
1793 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1794 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1795 &nent
, cpuid
->nent
);
1797 if (nent
>= cpuid
->nent
)
1801 if (copy_to_user(entries
, cpuid_entries
,
1802 nent
* sizeof(struct kvm_cpuid_entry2
)))
1808 vfree(cpuid_entries
);
1813 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1814 struct kvm_lapic_state
*s
)
1817 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
1823 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1824 struct kvm_lapic_state
*s
)
1827 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1828 kvm_apic_post_state_restore(vcpu
);
1829 update_cr8_intercept(vcpu
);
1835 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1836 struct kvm_interrupt
*irq
)
1838 if (irq
->irq
< 0 || irq
->irq
>= 256)
1840 if (irqchip_in_kernel(vcpu
->kvm
))
1844 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
1851 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
1854 kvm_inject_nmi(vcpu
);
1860 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
1861 struct kvm_tpr_access_ctl
*tac
)
1865 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
1869 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
1873 unsigned bank_num
= mcg_cap
& 0xff, bank
;
1876 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
1878 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
1881 vcpu
->arch
.mcg_cap
= mcg_cap
;
1882 /* Init IA32_MCG_CTL to all 1s */
1883 if (mcg_cap
& MCG_CTL_P
)
1884 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
1885 /* Init IA32_MCi_CTL to all 1s */
1886 for (bank
= 0; bank
< bank_num
; bank
++)
1887 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
1892 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
1893 struct kvm_x86_mce
*mce
)
1895 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1896 unsigned bank_num
= mcg_cap
& 0xff;
1897 u64
*banks
= vcpu
->arch
.mce_banks
;
1899 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
1902 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1903 * reporting is disabled
1905 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
1906 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
1908 banks
+= 4 * mce
->bank
;
1910 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1911 * reporting is disabled for the bank
1913 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
1915 if (mce
->status
& MCI_STATUS_UC
) {
1916 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
1917 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
1918 printk(KERN_DEBUG
"kvm: set_mce: "
1919 "injects mce exception while "
1920 "previous one is in progress!\n");
1921 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
1924 if (banks
[1] & MCI_STATUS_VAL
)
1925 mce
->status
|= MCI_STATUS_OVER
;
1926 banks
[2] = mce
->addr
;
1927 banks
[3] = mce
->misc
;
1928 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
1929 banks
[1] = mce
->status
;
1930 kvm_queue_exception(vcpu
, MC_VECTOR
);
1931 } else if (!(banks
[1] & MCI_STATUS_VAL
)
1932 || !(banks
[1] & MCI_STATUS_UC
)) {
1933 if (banks
[1] & MCI_STATUS_VAL
)
1934 mce
->status
|= MCI_STATUS_OVER
;
1935 banks
[2] = mce
->addr
;
1936 banks
[3] = mce
->misc
;
1937 banks
[1] = mce
->status
;
1939 banks
[1] |= MCI_STATUS_OVER
;
1943 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
1944 struct kvm_vcpu_events
*events
)
1948 events
->exception
.injected
= vcpu
->arch
.exception
.pending
;
1949 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
1950 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
1951 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
1953 events
->interrupt
.injected
= vcpu
->arch
.interrupt
.pending
;
1954 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
1955 events
->interrupt
.soft
= vcpu
->arch
.interrupt
.soft
;
1957 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
1958 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
;
1959 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
1961 events
->sipi_vector
= vcpu
->arch
.sipi_vector
;
1963 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
1964 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
);
1969 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
1970 struct kvm_vcpu_events
*events
)
1972 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
1973 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
))
1978 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
1979 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
1980 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
1981 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
1983 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
1984 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
1985 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
1986 if (vcpu
->arch
.interrupt
.pending
&& irqchip_in_kernel(vcpu
->kvm
))
1987 kvm_pic_clear_isr_ack(vcpu
->kvm
);
1989 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
1990 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
1991 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
1992 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
1994 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
)
1995 vcpu
->arch
.sipi_vector
= events
->sipi_vector
;
2002 long kvm_arch_vcpu_ioctl(struct file
*filp
,
2003 unsigned int ioctl
, unsigned long arg
)
2005 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2006 void __user
*argp
= (void __user
*)arg
;
2008 struct kvm_lapic_state
*lapic
= NULL
;
2011 case KVM_GET_LAPIC
: {
2013 if (!vcpu
->arch
.apic
)
2015 lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
2020 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, lapic
);
2024 if (copy_to_user(argp
, lapic
, sizeof(struct kvm_lapic_state
)))
2029 case KVM_SET_LAPIC
: {
2031 if (!vcpu
->arch
.apic
)
2033 lapic
= kmalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
2038 if (copy_from_user(lapic
, argp
, sizeof(struct kvm_lapic_state
)))
2040 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, lapic
);
2046 case KVM_INTERRUPT
: {
2047 struct kvm_interrupt irq
;
2050 if (copy_from_user(&irq
, argp
, sizeof irq
))
2052 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2059 r
= kvm_vcpu_ioctl_nmi(vcpu
);
2065 case KVM_SET_CPUID
: {
2066 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2067 struct kvm_cpuid cpuid
;
2070 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2072 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2077 case KVM_SET_CPUID2
: {
2078 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2079 struct kvm_cpuid2 cpuid
;
2082 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2084 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
2085 cpuid_arg
->entries
);
2090 case KVM_GET_CPUID2
: {
2091 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2092 struct kvm_cpuid2 cpuid
;
2095 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2097 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
2098 cpuid_arg
->entries
);
2102 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2108 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
2111 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2113 case KVM_TPR_ACCESS_REPORTING
: {
2114 struct kvm_tpr_access_ctl tac
;
2117 if (copy_from_user(&tac
, argp
, sizeof tac
))
2119 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
2123 if (copy_to_user(argp
, &tac
, sizeof tac
))
2128 case KVM_SET_VAPIC_ADDR
: {
2129 struct kvm_vapic_addr va
;
2132 if (!irqchip_in_kernel(vcpu
->kvm
))
2135 if (copy_from_user(&va
, argp
, sizeof va
))
2138 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
2141 case KVM_X86_SETUP_MCE
: {
2145 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
2147 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
2150 case KVM_X86_SET_MCE
: {
2151 struct kvm_x86_mce mce
;
2154 if (copy_from_user(&mce
, argp
, sizeof mce
))
2156 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
2159 case KVM_GET_VCPU_EVENTS
: {
2160 struct kvm_vcpu_events events
;
2162 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
2165 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
2170 case KVM_SET_VCPU_EVENTS
: {
2171 struct kvm_vcpu_events events
;
2174 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
2177 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
2188 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
2192 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
2194 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
2198 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
2201 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
2205 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
2206 u32 kvm_nr_mmu_pages
)
2208 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
2211 down_write(&kvm
->slots_lock
);
2212 spin_lock(&kvm
->mmu_lock
);
2214 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
2215 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
2217 spin_unlock(&kvm
->mmu_lock
);
2218 up_write(&kvm
->slots_lock
);
2222 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
2224 return kvm
->arch
.n_alloc_mmu_pages
;
2227 gfn_t
unalias_gfn_instantiation(struct kvm
*kvm
, gfn_t gfn
)
2230 struct kvm_mem_alias
*alias
;
2231 struct kvm_mem_aliases
*aliases
;
2233 aliases
= rcu_dereference(kvm
->arch
.aliases
);
2235 for (i
= 0; i
< aliases
->naliases
; ++i
) {
2236 alias
= &aliases
->aliases
[i
];
2237 if (alias
->flags
& KVM_ALIAS_INVALID
)
2239 if (gfn
>= alias
->base_gfn
2240 && gfn
< alias
->base_gfn
+ alias
->npages
)
2241 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
2246 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
2249 struct kvm_mem_alias
*alias
;
2250 struct kvm_mem_aliases
*aliases
;
2252 aliases
= rcu_dereference(kvm
->arch
.aliases
);
2254 for (i
= 0; i
< aliases
->naliases
; ++i
) {
2255 alias
= &aliases
->aliases
[i
];
2256 if (gfn
>= alias
->base_gfn
2257 && gfn
< alias
->base_gfn
+ alias
->npages
)
2258 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
2264 * Set a new alias region. Aliases map a portion of physical memory into
2265 * another portion. This is useful for memory windows, for example the PC
2268 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
2269 struct kvm_memory_alias
*alias
)
2272 struct kvm_mem_alias
*p
;
2273 struct kvm_mem_aliases
*aliases
, *old_aliases
;
2276 /* General sanity checks */
2277 if (alias
->memory_size
& (PAGE_SIZE
- 1))
2279 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
2281 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
2283 if (alias
->guest_phys_addr
+ alias
->memory_size
2284 < alias
->guest_phys_addr
)
2286 if (alias
->target_phys_addr
+ alias
->memory_size
2287 < alias
->target_phys_addr
)
2291 aliases
= kzalloc(sizeof(struct kvm_mem_aliases
), GFP_KERNEL
);
2295 down_write(&kvm
->slots_lock
);
2297 /* invalidate any gfn reference in case of deletion/shrinking */
2298 memcpy(aliases
, kvm
->arch
.aliases
, sizeof(struct kvm_mem_aliases
));
2299 aliases
->aliases
[alias
->slot
].flags
|= KVM_ALIAS_INVALID
;
2300 old_aliases
= kvm
->arch
.aliases
;
2301 rcu_assign_pointer(kvm
->arch
.aliases
, aliases
);
2302 synchronize_srcu_expedited(&kvm
->srcu
);
2303 kvm_mmu_zap_all(kvm
);
2307 aliases
= kzalloc(sizeof(struct kvm_mem_aliases
), GFP_KERNEL
);
2311 memcpy(aliases
, kvm
->arch
.aliases
, sizeof(struct kvm_mem_aliases
));
2313 p
= &aliases
->aliases
[alias
->slot
];
2314 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
2315 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
2316 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
2317 p
->flags
&= ~(KVM_ALIAS_INVALID
);
2319 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
2320 if (aliases
->aliases
[n
- 1].npages
)
2322 aliases
->naliases
= n
;
2324 old_aliases
= kvm
->arch
.aliases
;
2325 rcu_assign_pointer(kvm
->arch
.aliases
, aliases
);
2326 synchronize_srcu_expedited(&kvm
->srcu
);
2331 up_write(&kvm
->slots_lock
);
2336 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
2341 switch (chip
->chip_id
) {
2342 case KVM_IRQCHIP_PIC_MASTER
:
2343 memcpy(&chip
->chip
.pic
,
2344 &pic_irqchip(kvm
)->pics
[0],
2345 sizeof(struct kvm_pic_state
));
2347 case KVM_IRQCHIP_PIC_SLAVE
:
2348 memcpy(&chip
->chip
.pic
,
2349 &pic_irqchip(kvm
)->pics
[1],
2350 sizeof(struct kvm_pic_state
));
2352 case KVM_IRQCHIP_IOAPIC
:
2353 r
= kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
2362 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
2367 switch (chip
->chip_id
) {
2368 case KVM_IRQCHIP_PIC_MASTER
:
2369 spin_lock(&pic_irqchip(kvm
)->lock
);
2370 memcpy(&pic_irqchip(kvm
)->pics
[0],
2372 sizeof(struct kvm_pic_state
));
2373 spin_unlock(&pic_irqchip(kvm
)->lock
);
2375 case KVM_IRQCHIP_PIC_SLAVE
:
2376 spin_lock(&pic_irqchip(kvm
)->lock
);
2377 memcpy(&pic_irqchip(kvm
)->pics
[1],
2379 sizeof(struct kvm_pic_state
));
2380 spin_unlock(&pic_irqchip(kvm
)->lock
);
2382 case KVM_IRQCHIP_IOAPIC
:
2383 r
= kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
2389 kvm_pic_update_irq(pic_irqchip(kvm
));
2393 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
2397 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2398 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
2399 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2403 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
2407 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2408 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
2409 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
, 0);
2410 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2414 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
2418 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2419 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
2420 sizeof(ps
->channels
));
2421 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
2422 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2426 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
2428 int r
= 0, start
= 0;
2429 u32 prev_legacy
, cur_legacy
;
2430 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2431 prev_legacy
= kvm
->arch
.vpit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
2432 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
2433 if (!prev_legacy
&& cur_legacy
)
2435 memcpy(&kvm
->arch
.vpit
->pit_state
.channels
, &ps
->channels
,
2436 sizeof(kvm
->arch
.vpit
->pit_state
.channels
));
2437 kvm
->arch
.vpit
->pit_state
.flags
= ps
->flags
;
2438 kvm_pit_load_count(kvm
, 0, kvm
->arch
.vpit
->pit_state
.channels
[0].count
, start
);
2439 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2443 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
2444 struct kvm_reinject_control
*control
)
2446 if (!kvm
->arch
.vpit
)
2448 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2449 kvm
->arch
.vpit
->pit_state
.pit_timer
.reinject
= control
->pit_reinject
;
2450 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2455 * Get (and clear) the dirty memory log for a memory slot.
2457 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2458 struct kvm_dirty_log
*log
)
2461 struct kvm_memory_slot
*memslot
;
2462 unsigned long is_dirty
= 0;
2463 unsigned long *dirty_bitmap
= NULL
;
2465 down_write(&kvm
->slots_lock
);
2468 if (log
->slot
>= KVM_MEMORY_SLOTS
)
2471 memslot
= &kvm
->memslots
->memslots
[log
->slot
];
2473 if (!memslot
->dirty_bitmap
)
2476 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
2479 dirty_bitmap
= vmalloc(n
);
2482 memset(dirty_bitmap
, 0, n
);
2484 for (i
= 0; !is_dirty
&& i
< n
/sizeof(long); i
++)
2485 is_dirty
= memslot
->dirty_bitmap
[i
];
2487 /* If nothing is dirty, don't bother messing with page tables. */
2489 struct kvm_memslots
*slots
, *old_slots
;
2491 spin_lock(&kvm
->mmu_lock
);
2492 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
2493 spin_unlock(&kvm
->mmu_lock
);
2495 slots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
2499 memcpy(slots
, kvm
->memslots
, sizeof(struct kvm_memslots
));
2500 slots
->memslots
[log
->slot
].dirty_bitmap
= dirty_bitmap
;
2502 old_slots
= kvm
->memslots
;
2503 rcu_assign_pointer(kvm
->memslots
, slots
);
2504 synchronize_srcu_expedited(&kvm
->srcu
);
2505 dirty_bitmap
= old_slots
->memslots
[log
->slot
].dirty_bitmap
;
2510 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap
, n
))
2513 vfree(dirty_bitmap
);
2515 up_write(&kvm
->slots_lock
);
2519 long kvm_arch_vm_ioctl(struct file
*filp
,
2520 unsigned int ioctl
, unsigned long arg
)
2522 struct kvm
*kvm
= filp
->private_data
;
2523 void __user
*argp
= (void __user
*)arg
;
2526 * This union makes it completely explicit to gcc-3.x
2527 * that these two variables' stack usage should be
2528 * combined, not added together.
2531 struct kvm_pit_state ps
;
2532 struct kvm_pit_state2 ps2
;
2533 struct kvm_memory_alias alias
;
2534 struct kvm_pit_config pit_config
;
2538 case KVM_SET_TSS_ADDR
:
2539 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
2543 case KVM_SET_IDENTITY_MAP_ADDR
: {
2547 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
2549 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
2554 case KVM_SET_MEMORY_REGION
: {
2555 struct kvm_memory_region kvm_mem
;
2556 struct kvm_userspace_memory_region kvm_userspace_mem
;
2559 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2561 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
2562 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
2563 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
2564 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
2565 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
2570 case KVM_SET_NR_MMU_PAGES
:
2571 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
2575 case KVM_GET_NR_MMU_PAGES
:
2576 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
2578 case KVM_SET_MEMORY_ALIAS
:
2580 if (copy_from_user(&u
.alias
, argp
, sizeof(struct kvm_memory_alias
)))
2582 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &u
.alias
);
2586 case KVM_CREATE_IRQCHIP
: {
2587 struct kvm_pic
*vpic
;
2589 mutex_lock(&kvm
->lock
);
2592 goto create_irqchip_unlock
;
2594 vpic
= kvm_create_pic(kvm
);
2596 r
= kvm_ioapic_init(kvm
);
2599 goto create_irqchip_unlock
;
2602 goto create_irqchip_unlock
;
2604 kvm
->arch
.vpic
= vpic
;
2606 r
= kvm_setup_default_irq_routing(kvm
);
2608 mutex_lock(&kvm
->irq_lock
);
2609 kfree(kvm
->arch
.vpic
);
2610 kfree(kvm
->arch
.vioapic
);
2611 kvm
->arch
.vpic
= NULL
;
2612 kvm
->arch
.vioapic
= NULL
;
2613 mutex_unlock(&kvm
->irq_lock
);
2615 create_irqchip_unlock
:
2616 mutex_unlock(&kvm
->lock
);
2619 case KVM_CREATE_PIT
:
2620 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
2622 case KVM_CREATE_PIT2
:
2624 if (copy_from_user(&u
.pit_config
, argp
,
2625 sizeof(struct kvm_pit_config
)))
2628 down_write(&kvm
->slots_lock
);
2631 goto create_pit_unlock
;
2633 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
2637 up_write(&kvm
->slots_lock
);
2639 case KVM_IRQ_LINE_STATUS
:
2640 case KVM_IRQ_LINE
: {
2641 struct kvm_irq_level irq_event
;
2644 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2646 if (irqchip_in_kernel(kvm
)) {
2648 status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
2649 irq_event
.irq
, irq_event
.level
);
2650 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2651 irq_event
.status
= status
;
2652 if (copy_to_user(argp
, &irq_event
,
2660 case KVM_GET_IRQCHIP
: {
2661 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2662 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
2668 if (copy_from_user(chip
, argp
, sizeof *chip
))
2669 goto get_irqchip_out
;
2671 if (!irqchip_in_kernel(kvm
))
2672 goto get_irqchip_out
;
2673 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
2675 goto get_irqchip_out
;
2677 if (copy_to_user(argp
, chip
, sizeof *chip
))
2678 goto get_irqchip_out
;
2686 case KVM_SET_IRQCHIP
: {
2687 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2688 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
2694 if (copy_from_user(chip
, argp
, sizeof *chip
))
2695 goto set_irqchip_out
;
2697 if (!irqchip_in_kernel(kvm
))
2698 goto set_irqchip_out
;
2699 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
2701 goto set_irqchip_out
;
2711 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
2714 if (!kvm
->arch
.vpit
)
2716 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
2720 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
2727 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
2730 if (!kvm
->arch
.vpit
)
2732 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
2738 case KVM_GET_PIT2
: {
2740 if (!kvm
->arch
.vpit
)
2742 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
2746 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
2751 case KVM_SET_PIT2
: {
2753 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
2756 if (!kvm
->arch
.vpit
)
2758 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
2764 case KVM_REINJECT_CONTROL
: {
2765 struct kvm_reinject_control control
;
2767 if (copy_from_user(&control
, argp
, sizeof(control
)))
2769 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
2775 case KVM_XEN_HVM_CONFIG
: {
2777 if (copy_from_user(&kvm
->arch
.xen_hvm_config
, argp
,
2778 sizeof(struct kvm_xen_hvm_config
)))
2781 if (kvm
->arch
.xen_hvm_config
.flags
)
2786 case KVM_SET_CLOCK
: {
2787 struct timespec now
;
2788 struct kvm_clock_data user_ns
;
2793 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
2802 now_ns
= timespec_to_ns(&now
);
2803 delta
= user_ns
.clock
- now_ns
;
2804 kvm
->arch
.kvmclock_offset
= delta
;
2807 case KVM_GET_CLOCK
: {
2808 struct timespec now
;
2809 struct kvm_clock_data user_ns
;
2813 now_ns
= timespec_to_ns(&now
);
2814 user_ns
.clock
= kvm
->arch
.kvmclock_offset
+ now_ns
;
2818 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
2831 static void kvm_init_msr_list(void)
2836 /* skip the first msrs in the list. KVM-specific */
2837 for (i
= j
= KVM_SAVE_MSRS_BEGIN
; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2838 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2841 msrs_to_save
[j
] = msrs_to_save
[i
];
2844 num_msrs_to_save
= j
;
2847 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
2850 if (vcpu
->arch
.apic
&&
2851 !kvm_iodevice_write(&vcpu
->arch
.apic
->dev
, addr
, len
, v
))
2854 return kvm_io_bus_write(&vcpu
->kvm
->mmio_bus
, addr
, len
, v
);
2857 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
2859 if (vcpu
->arch
.apic
&&
2860 !kvm_iodevice_read(&vcpu
->arch
.apic
->dev
, addr
, len
, v
))
2863 return kvm_io_bus_read(&vcpu
->kvm
->mmio_bus
, addr
, len
, v
);
2866 static int kvm_read_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2867 struct kvm_vcpu
*vcpu
)
2870 int r
= X86EMUL_CONTINUE
;
2873 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2874 unsigned offset
= addr
& (PAGE_SIZE
-1);
2875 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2878 if (gpa
== UNMAPPED_GVA
) {
2879 r
= X86EMUL_PROPAGATE_FAULT
;
2882 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
2884 r
= X86EMUL_UNHANDLEABLE
;
2896 static int kvm_write_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2897 struct kvm_vcpu
*vcpu
)
2900 int r
= X86EMUL_CONTINUE
;
2903 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2904 unsigned offset
= addr
& (PAGE_SIZE
-1);
2905 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2908 if (gpa
== UNMAPPED_GVA
) {
2909 r
= X86EMUL_PROPAGATE_FAULT
;
2912 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
2914 r
= X86EMUL_UNHANDLEABLE
;
2927 static int emulator_read_emulated(unsigned long addr
,
2930 struct kvm_vcpu
*vcpu
)
2934 if (vcpu
->mmio_read_completed
) {
2935 memcpy(val
, vcpu
->mmio_data
, bytes
);
2936 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
2937 vcpu
->mmio_phys_addr
, *(u64
*)val
);
2938 vcpu
->mmio_read_completed
= 0;
2939 return X86EMUL_CONTINUE
;
2942 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2944 /* For APIC access vmexit */
2945 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2948 if (kvm_read_guest_virt(addr
, val
, bytes
, vcpu
)
2949 == X86EMUL_CONTINUE
)
2950 return X86EMUL_CONTINUE
;
2951 if (gpa
== UNMAPPED_GVA
)
2952 return X86EMUL_PROPAGATE_FAULT
;
2956 * Is this MMIO handled locally?
2958 if (!vcpu_mmio_read(vcpu
, gpa
, bytes
, val
)) {
2959 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
, gpa
, *(u64
*)val
);
2960 return X86EMUL_CONTINUE
;
2963 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
2965 vcpu
->mmio_needed
= 1;
2966 vcpu
->mmio_phys_addr
= gpa
;
2967 vcpu
->mmio_size
= bytes
;
2968 vcpu
->mmio_is_write
= 0;
2970 return X86EMUL_UNHANDLEABLE
;
2973 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2974 const void *val
, int bytes
)
2978 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
2981 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
, 1);
2985 static int emulator_write_emulated_onepage(unsigned long addr
,
2988 struct kvm_vcpu
*vcpu
)
2992 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2994 if (gpa
== UNMAPPED_GVA
) {
2995 kvm_inject_page_fault(vcpu
, addr
, 2);
2996 return X86EMUL_PROPAGATE_FAULT
;
2999 /* For APIC access vmexit */
3000 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
3003 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
3004 return X86EMUL_CONTINUE
;
3007 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
3009 * Is this MMIO handled locally?
3011 if (!vcpu_mmio_write(vcpu
, gpa
, bytes
, val
))
3012 return X86EMUL_CONTINUE
;
3014 vcpu
->mmio_needed
= 1;
3015 vcpu
->mmio_phys_addr
= gpa
;
3016 vcpu
->mmio_size
= bytes
;
3017 vcpu
->mmio_is_write
= 1;
3018 memcpy(vcpu
->mmio_data
, val
, bytes
);
3020 return X86EMUL_CONTINUE
;
3023 int emulator_write_emulated(unsigned long addr
,
3026 struct kvm_vcpu
*vcpu
)
3028 /* Crossing a page boundary? */
3029 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
3032 now
= -addr
& ~PAGE_MASK
;
3033 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
3034 if (rc
!= X86EMUL_CONTINUE
)
3040 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
3042 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
3044 static int emulator_cmpxchg_emulated(unsigned long addr
,
3048 struct kvm_vcpu
*vcpu
)
3050 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
3051 #ifndef CONFIG_X86_64
3052 /* guests cmpxchg8b have to be emulated atomically */
3059 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
3061 if (gpa
== UNMAPPED_GVA
||
3062 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
3065 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
3070 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
3072 kaddr
= kmap_atomic(page
, KM_USER0
);
3073 set_64bit((u64
*)(kaddr
+ offset_in_page(gpa
)), val
);
3074 kunmap_atomic(kaddr
, KM_USER0
);
3075 kvm_release_page_dirty(page
);
3080 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
3083 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
3085 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
3088 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
3090 kvm_mmu_invlpg(vcpu
, address
);
3091 return X86EMUL_CONTINUE
;
3094 int emulate_clts(struct kvm_vcpu
*vcpu
)
3096 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
& ~X86_CR0_TS
);
3097 return X86EMUL_CONTINUE
;
3100 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
3102 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
3106 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
3107 return X86EMUL_CONTINUE
;
3109 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __func__
, dr
);
3110 return X86EMUL_UNHANDLEABLE
;
3114 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
3116 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
3119 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
3121 /* FIXME: better handling */
3122 return X86EMUL_UNHANDLEABLE
;
3124 return X86EMUL_CONTINUE
;
3127 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
3130 unsigned long rip
= kvm_rip_read(vcpu
);
3131 unsigned long rip_linear
;
3133 if (!printk_ratelimit())
3136 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
3138 kvm_read_guest_virt(rip_linear
, (void *)opcodes
, 4, vcpu
);
3140 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3141 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
3143 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
3145 static struct x86_emulate_ops emulate_ops
= {
3146 .read_std
= kvm_read_guest_virt
,
3147 .read_emulated
= emulator_read_emulated
,
3148 .write_emulated
= emulator_write_emulated
,
3149 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
3152 static void cache_all_regs(struct kvm_vcpu
*vcpu
)
3154 kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3155 kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3156 kvm_register_read(vcpu
, VCPU_REGS_RIP
);
3157 vcpu
->arch
.regs_dirty
= ~0;
3160 int emulate_instruction(struct kvm_vcpu
*vcpu
,
3166 struct decode_cache
*c
;
3167 struct kvm_run
*run
= vcpu
->run
;
3169 kvm_clear_exception_queue(vcpu
);
3170 vcpu
->arch
.mmio_fault_cr2
= cr2
;
3172 * TODO: fix emulate.c to use guest_read/write_register
3173 * instead of direct ->regs accesses, can save hundred cycles
3174 * on Intel for instructions that don't read/change RSP, for
3177 cache_all_regs(vcpu
);
3179 vcpu
->mmio_is_write
= 0;
3180 vcpu
->arch
.pio
.string
= 0;
3182 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
3184 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
3186 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
3187 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_get_rflags(vcpu
);
3188 vcpu
->arch
.emulate_ctxt
.mode
=
3189 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
3190 ? X86EMUL_MODE_REAL
: cs_l
3191 ? X86EMUL_MODE_PROT64
: cs_db
3192 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
3194 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
3196 /* Only allow emulation of specific instructions on #UD
3197 * (namely VMMCALL, sysenter, sysexit, syscall)*/
3198 c
= &vcpu
->arch
.emulate_ctxt
.decode
;
3199 if (emulation_type
& EMULTYPE_TRAP_UD
) {
3201 return EMULATE_FAIL
;
3203 case 0x01: /* VMMCALL */
3204 if (c
->modrm_mod
!= 3 || c
->modrm_rm
!= 1)
3205 return EMULATE_FAIL
;
3207 case 0x34: /* sysenter */
3208 case 0x35: /* sysexit */
3209 if (c
->modrm_mod
!= 0 || c
->modrm_rm
!= 0)
3210 return EMULATE_FAIL
;
3212 case 0x05: /* syscall */
3213 if (c
->modrm_mod
!= 0 || c
->modrm_rm
!= 0)
3214 return EMULATE_FAIL
;
3217 return EMULATE_FAIL
;
3220 if (!(c
->modrm_reg
== 0 || c
->modrm_reg
== 3))
3221 return EMULATE_FAIL
;
3224 ++vcpu
->stat
.insn_emulation
;
3226 ++vcpu
->stat
.insn_emulation_fail
;
3227 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
3228 return EMULATE_DONE
;
3229 return EMULATE_FAIL
;
3233 if (emulation_type
& EMULTYPE_SKIP
) {
3234 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.decode
.eip
);
3235 return EMULATE_DONE
;
3238 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
3239 shadow_mask
= vcpu
->arch
.emulate_ctxt
.interruptibility
;
3242 kvm_x86_ops
->set_interrupt_shadow(vcpu
, shadow_mask
);
3244 if (vcpu
->arch
.pio
.string
)
3245 return EMULATE_DO_MMIO
;
3247 if ((r
|| vcpu
->mmio_is_write
) && run
) {
3248 run
->exit_reason
= KVM_EXIT_MMIO
;
3249 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
3250 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
3251 run
->mmio
.len
= vcpu
->mmio_size
;
3252 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
3256 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
3257 return EMULATE_DONE
;
3258 if (!vcpu
->mmio_needed
) {
3259 kvm_report_emulation_failure(vcpu
, "mmio");
3260 return EMULATE_FAIL
;
3262 return EMULATE_DO_MMIO
;
3265 kvm_set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
3267 if (vcpu
->mmio_is_write
) {
3268 vcpu
->mmio_needed
= 0;
3269 return EMULATE_DO_MMIO
;
3272 return EMULATE_DONE
;
3274 EXPORT_SYMBOL_GPL(emulate_instruction
);
3276 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
3278 void *p
= vcpu
->arch
.pio_data
;
3279 gva_t q
= vcpu
->arch
.pio
.guest_gva
;
3283 bytes
= vcpu
->arch
.pio
.size
* vcpu
->arch
.pio
.cur_count
;
3284 if (vcpu
->arch
.pio
.in
)
3285 ret
= kvm_write_guest_virt(q
, p
, bytes
, vcpu
);
3287 ret
= kvm_read_guest_virt(q
, p
, bytes
, vcpu
);
3291 int complete_pio(struct kvm_vcpu
*vcpu
)
3293 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
3300 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3301 memcpy(&val
, vcpu
->arch
.pio_data
, io
->size
);
3302 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
3306 r
= pio_copy_data(vcpu
);
3313 delta
*= io
->cur_count
;
3315 * The size of the register should really depend on
3316 * current address size.
3318 val
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3320 kvm_register_write(vcpu
, VCPU_REGS_RCX
, val
);
3326 val
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3328 kvm_register_write(vcpu
, VCPU_REGS_RDI
, val
);
3330 val
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3332 kvm_register_write(vcpu
, VCPU_REGS_RSI
, val
);
3336 io
->count
-= io
->cur_count
;
3342 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
3344 /* TODO: String I/O for in kernel device */
3347 if (vcpu
->arch
.pio
.in
)
3348 r
= kvm_io_bus_read(&vcpu
->kvm
->pio_bus
, vcpu
->arch
.pio
.port
,
3349 vcpu
->arch
.pio
.size
, pd
);
3351 r
= kvm_io_bus_write(&vcpu
->kvm
->pio_bus
, vcpu
->arch
.pio
.port
,
3352 vcpu
->arch
.pio
.size
, pd
);
3356 static int pio_string_write(struct kvm_vcpu
*vcpu
)
3358 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
3359 void *pd
= vcpu
->arch
.pio_data
;
3362 for (i
= 0; i
< io
->cur_count
; i
++) {
3363 if (kvm_io_bus_write(&vcpu
->kvm
->pio_bus
,
3364 io
->port
, io
->size
, pd
)) {
3373 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, int in
, int size
, unsigned port
)
3377 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
3378 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
3379 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
3380 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
3381 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= 1;
3382 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
3383 vcpu
->arch
.pio
.in
= in
;
3384 vcpu
->arch
.pio
.string
= 0;
3385 vcpu
->arch
.pio
.down
= 0;
3386 vcpu
->arch
.pio
.rep
= 0;
3388 trace_kvm_pio(vcpu
->run
->io
.direction
== KVM_EXIT_IO_OUT
, port
,
3391 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3392 memcpy(vcpu
->arch
.pio_data
, &val
, 4);
3394 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
3400 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
3402 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, int in
,
3403 int size
, unsigned long count
, int down
,
3404 gva_t address
, int rep
, unsigned port
)
3406 unsigned now
, in_page
;
3409 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
3410 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
3411 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
3412 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
3413 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= count
;
3414 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
3415 vcpu
->arch
.pio
.in
= in
;
3416 vcpu
->arch
.pio
.string
= 1;
3417 vcpu
->arch
.pio
.down
= down
;
3418 vcpu
->arch
.pio
.rep
= rep
;
3420 trace_kvm_pio(vcpu
->run
->io
.direction
== KVM_EXIT_IO_OUT
, port
,
3424 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3429 in_page
= PAGE_SIZE
- offset_in_page(address
);
3431 in_page
= offset_in_page(address
) + size
;
3432 now
= min(count
, (unsigned long)in_page
/ size
);
3437 * String I/O in reverse. Yuck. Kill the guest, fix later.
3439 pr_unimpl(vcpu
, "guest string pio down\n");
3440 kvm_inject_gp(vcpu
, 0);
3443 vcpu
->run
->io
.count
= now
;
3444 vcpu
->arch
.pio
.cur_count
= now
;
3446 if (vcpu
->arch
.pio
.cur_count
== vcpu
->arch
.pio
.count
)
3447 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3449 vcpu
->arch
.pio
.guest_gva
= address
;
3451 if (!vcpu
->arch
.pio
.in
) {
3452 /* string PIO write */
3453 ret
= pio_copy_data(vcpu
);
3454 if (ret
== X86EMUL_PROPAGATE_FAULT
) {
3455 kvm_inject_gp(vcpu
, 0);
3458 if (ret
== 0 && !pio_string_write(vcpu
)) {
3460 if (vcpu
->arch
.pio
.count
== 0)
3464 /* no string PIO read support yet */
3468 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
3470 static void bounce_off(void *info
)
3475 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
3478 struct cpufreq_freqs
*freq
= data
;
3480 struct kvm_vcpu
*vcpu
;
3481 int i
, send_ipi
= 0;
3483 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
3485 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
3487 per_cpu(cpu_tsc_khz
, freq
->cpu
) = freq
->new;
3489 spin_lock(&kvm_lock
);
3490 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3491 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3492 if (vcpu
->cpu
!= freq
->cpu
)
3494 if (!kvm_request_guest_time_update(vcpu
))
3496 if (vcpu
->cpu
!= smp_processor_id())
3500 spin_unlock(&kvm_lock
);
3502 if (freq
->old
< freq
->new && send_ipi
) {
3504 * We upscale the frequency. Must make the guest
3505 * doesn't see old kvmclock values while running with
3506 * the new frequency, otherwise we risk the guest sees
3507 * time go backwards.
3509 * In case we update the frequency for another cpu
3510 * (which might be in guest context) send an interrupt
3511 * to kick the cpu out of guest context. Next time
3512 * guest context is entered kvmclock will be updated,
3513 * so the guest will not see stale values.
3515 smp_call_function_single(freq
->cpu
, bounce_off
, NULL
, 1);
3520 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
3521 .notifier_call
= kvmclock_cpufreq_notifier
3524 static void kvm_timer_init(void)
3528 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
3529 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
3530 CPUFREQ_TRANSITION_NOTIFIER
);
3531 for_each_online_cpu(cpu
) {
3532 unsigned long khz
= cpufreq_get(cpu
);
3535 per_cpu(cpu_tsc_khz
, cpu
) = khz
;
3538 for_each_possible_cpu(cpu
)
3539 per_cpu(cpu_tsc_khz
, cpu
) = tsc_khz
;
3543 int kvm_arch_init(void *opaque
)
3546 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
3549 printk(KERN_ERR
"kvm: already loaded the other module\n");
3554 if (!ops
->cpu_has_kvm_support()) {
3555 printk(KERN_ERR
"kvm: no hardware support\n");
3559 if (ops
->disabled_by_bios()) {
3560 printk(KERN_ERR
"kvm: disabled by bios\n");
3565 r
= kvm_mmu_module_init();
3569 kvm_init_msr_list();
3572 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3573 kvm_mmu_set_base_ptes(PT_PRESENT_MASK
);
3574 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
3575 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
3585 void kvm_arch_exit(void)
3587 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
3588 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
3589 CPUFREQ_TRANSITION_NOTIFIER
);
3591 kvm_mmu_module_exit();
3594 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
3596 ++vcpu
->stat
.halt_exits
;
3597 if (irqchip_in_kernel(vcpu
->kvm
)) {
3598 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
3601 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
3605 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
3607 static inline gpa_t
hc_gpa(struct kvm_vcpu
*vcpu
, unsigned long a0
,
3610 if (is_long_mode(vcpu
))
3613 return a0
| ((gpa_t
)a1
<< 32);
3616 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
3618 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
3621 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3622 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3623 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3624 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3625 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3627 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
3629 if (!is_long_mode(vcpu
)) {
3637 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
3643 case KVM_HC_VAPIC_POLL_IRQ
:
3647 r
= kvm_pv_mmu_op(vcpu
, a0
, hc_gpa(vcpu
, a1
, a2
), &ret
);
3654 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
3655 ++vcpu
->stat
.hypercalls
;
3658 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
3660 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
3662 char instruction
[3];
3664 unsigned long rip
= kvm_rip_read(vcpu
);
3668 * Blow out the MMU to ensure that no other VCPU has an active mapping
3669 * to ensure that the updated hypercall appears atomically across all
3672 kvm_mmu_zap_all(vcpu
->kvm
);
3674 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
3675 if (emulator_write_emulated(rip
, instruction
, 3, vcpu
)
3676 != X86EMUL_CONTINUE
)
3682 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
3684 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
3687 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
3689 struct descriptor_table dt
= { limit
, base
};
3691 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
3694 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
3696 struct descriptor_table dt
= { limit
, base
};
3698 kvm_x86_ops
->set_idt(vcpu
, &dt
);
3701 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
3702 unsigned long *rflags
)
3704 kvm_lmsw(vcpu
, msw
);
3705 *rflags
= kvm_get_rflags(vcpu
);
3708 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
3710 unsigned long value
;
3714 value
= vcpu
->arch
.cr0
;
3717 value
= vcpu
->arch
.cr2
;
3720 value
= vcpu
->arch
.cr3
;
3723 value
= kvm_read_cr4(vcpu
);
3726 value
= kvm_get_cr8(vcpu
);
3729 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
3736 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
3737 unsigned long *rflags
)
3741 kvm_set_cr0(vcpu
, mk_cr_64(vcpu
->arch
.cr0
, val
));
3742 *rflags
= kvm_get_rflags(vcpu
);
3745 vcpu
->arch
.cr2
= val
;
3748 kvm_set_cr3(vcpu
, val
);
3751 kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
3754 kvm_set_cr8(vcpu
, val
& 0xfUL
);
3757 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
3761 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
3763 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
3764 int j
, nent
= vcpu
->arch
.cpuid_nent
;
3766 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
3767 /* when no next entry is found, the current entry[i] is reselected */
3768 for (j
= i
+ 1; ; j
= (j
+ 1) % nent
) {
3769 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
3770 if (ej
->function
== e
->function
) {
3771 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
3775 return 0; /* silence gcc, even though control never reaches here */
3778 /* find an entry with matching function, matching index (if needed), and that
3779 * should be read next (if it's stateful) */
3780 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
3781 u32 function
, u32 index
)
3783 if (e
->function
!= function
)
3785 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
3787 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
3788 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
3793 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
3794 u32 function
, u32 index
)
3797 struct kvm_cpuid_entry2
*best
= NULL
;
3799 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
3800 struct kvm_cpuid_entry2
*e
;
3802 e
= &vcpu
->arch
.cpuid_entries
[i
];
3803 if (is_matching_cpuid_entry(e
, function
, index
)) {
3804 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
3805 move_to_next_stateful_cpuid_entry(vcpu
, i
);
3810 * Both basic or both extended?
3812 if (((e
->function
^ function
) & 0x80000000) == 0)
3813 if (!best
|| e
->function
> best
->function
)
3818 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry
);
3820 int cpuid_maxphyaddr(struct kvm_vcpu
*vcpu
)
3822 struct kvm_cpuid_entry2
*best
;
3824 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
3826 return best
->eax
& 0xff;
3830 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
3832 u32 function
, index
;
3833 struct kvm_cpuid_entry2
*best
;
3835 function
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3836 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3837 kvm_register_write(vcpu
, VCPU_REGS_RAX
, 0);
3838 kvm_register_write(vcpu
, VCPU_REGS_RBX
, 0);
3839 kvm_register_write(vcpu
, VCPU_REGS_RCX
, 0);
3840 kvm_register_write(vcpu
, VCPU_REGS_RDX
, 0);
3841 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
3843 kvm_register_write(vcpu
, VCPU_REGS_RAX
, best
->eax
);
3844 kvm_register_write(vcpu
, VCPU_REGS_RBX
, best
->ebx
);
3845 kvm_register_write(vcpu
, VCPU_REGS_RCX
, best
->ecx
);
3846 kvm_register_write(vcpu
, VCPU_REGS_RDX
, best
->edx
);
3848 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3849 trace_kvm_cpuid(function
,
3850 kvm_register_read(vcpu
, VCPU_REGS_RAX
),
3851 kvm_register_read(vcpu
, VCPU_REGS_RBX
),
3852 kvm_register_read(vcpu
, VCPU_REGS_RCX
),
3853 kvm_register_read(vcpu
, VCPU_REGS_RDX
));
3855 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
3858 * Check if userspace requested an interrupt window, and that the
3859 * interrupt window is open.
3861 * No need to exit to userspace if we already have an interrupt queued.
3863 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
3865 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
3866 vcpu
->run
->request_interrupt_window
&&
3867 kvm_arch_interrupt_allowed(vcpu
));
3870 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
3872 struct kvm_run
*kvm_run
= vcpu
->run
;
3874 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
3875 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
3876 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
3877 if (irqchip_in_kernel(vcpu
->kvm
))
3878 kvm_run
->ready_for_interrupt_injection
= 1;
3880 kvm_run
->ready_for_interrupt_injection
=
3881 kvm_arch_interrupt_allowed(vcpu
) &&
3882 !kvm_cpu_has_interrupt(vcpu
) &&
3883 !kvm_event_needs_reinjection(vcpu
);
3886 static void vapic_enter(struct kvm_vcpu
*vcpu
)
3888 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3891 if (!apic
|| !apic
->vapic_addr
)
3894 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3896 vcpu
->arch
.apic
->vapic_page
= page
;
3899 static void vapic_exit(struct kvm_vcpu
*vcpu
)
3901 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3903 if (!apic
|| !apic
->vapic_addr
)
3906 down_read(&vcpu
->kvm
->slots_lock
);
3907 kvm_release_page_dirty(apic
->vapic_page
);
3908 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3909 up_read(&vcpu
->kvm
->slots_lock
);
3912 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
3916 if (!kvm_x86_ops
->update_cr8_intercept
)
3919 if (!vcpu
->arch
.apic
)
3922 if (!vcpu
->arch
.apic
->vapic_addr
)
3923 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
3930 tpr
= kvm_lapic_get_cr8(vcpu
);
3932 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
3935 static void inject_pending_event(struct kvm_vcpu
*vcpu
)
3937 /* try to reinject previous events if any */
3938 if (vcpu
->arch
.exception
.pending
) {
3939 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
3940 vcpu
->arch
.exception
.has_error_code
,
3941 vcpu
->arch
.exception
.error_code
);
3945 if (vcpu
->arch
.nmi_injected
) {
3946 kvm_x86_ops
->set_nmi(vcpu
);
3950 if (vcpu
->arch
.interrupt
.pending
) {
3951 kvm_x86_ops
->set_irq(vcpu
);
3955 /* try to inject new event if pending */
3956 if (vcpu
->arch
.nmi_pending
) {
3957 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
3958 vcpu
->arch
.nmi_pending
= false;
3959 vcpu
->arch
.nmi_injected
= true;
3960 kvm_x86_ops
->set_nmi(vcpu
);
3962 } else if (kvm_cpu_has_interrupt(vcpu
)) {
3963 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
3964 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
3966 kvm_x86_ops
->set_irq(vcpu
);
3971 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
3974 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
3975 vcpu
->run
->request_interrupt_window
;
3978 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
3979 kvm_mmu_unload(vcpu
);
3981 r
= kvm_mmu_reload(vcpu
);
3985 if (vcpu
->requests
) {
3986 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
))
3987 __kvm_migrate_timers(vcpu
);
3988 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE
, &vcpu
->requests
))
3989 kvm_write_guest_time(vcpu
);
3990 if (test_and_clear_bit(KVM_REQ_MMU_SYNC
, &vcpu
->requests
))
3991 kvm_mmu_sync_roots(vcpu
);
3992 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
3993 kvm_x86_ops
->tlb_flush(vcpu
);
3994 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS
,
3996 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
4000 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
)) {
4001 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
4009 kvm_x86_ops
->prepare_guest_switch(vcpu
);
4010 kvm_load_guest_fpu(vcpu
);
4012 local_irq_disable();
4014 clear_bit(KVM_REQ_KICK
, &vcpu
->requests
);
4015 smp_mb__after_clear_bit();
4017 if (vcpu
->requests
|| need_resched() || signal_pending(current
)) {
4018 set_bit(KVM_REQ_KICK
, &vcpu
->requests
);
4025 inject_pending_event(vcpu
);
4027 /* enable NMI/IRQ window open exits if needed */
4028 if (vcpu
->arch
.nmi_pending
)
4029 kvm_x86_ops
->enable_nmi_window(vcpu
);
4030 else if (kvm_cpu_has_interrupt(vcpu
) || req_int_win
)
4031 kvm_x86_ops
->enable_irq_window(vcpu
);
4033 if (kvm_lapic_enabled(vcpu
)) {
4034 update_cr8_intercept(vcpu
);
4035 kvm_lapic_sync_to_vapic(vcpu
);
4038 up_read(&vcpu
->kvm
->slots_lock
);
4042 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
4044 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
4045 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
4046 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
4047 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
4050 trace_kvm_entry(vcpu
->vcpu_id
);
4051 kvm_x86_ops
->run(vcpu
);
4054 * If the guest has used debug registers, at least dr7
4055 * will be disabled while returning to the host.
4056 * If we don't have active breakpoints in the host, we don't
4057 * care about the messed up debug address registers. But if
4058 * we have some of them active, restore the old state.
4060 if (hw_breakpoint_active())
4061 hw_breakpoint_restore();
4063 set_bit(KVM_REQ_KICK
, &vcpu
->requests
);
4069 * We must have an instruction between local_irq_enable() and
4070 * kvm_guest_exit(), so the timer interrupt isn't delayed by
4071 * the interrupt shadow. The stat.exits increment will do nicely.
4072 * But we need to prevent reordering, hence this barrier():
4080 down_read(&vcpu
->kvm
->slots_lock
);
4083 * Profile KVM exit RIPs:
4085 if (unlikely(prof_on
== KVM_PROFILING
)) {
4086 unsigned long rip
= kvm_rip_read(vcpu
);
4087 profile_hit(KVM_PROFILING
, (void *)rip
);
4091 kvm_lapic_sync_from_vapic(vcpu
);
4093 r
= kvm_x86_ops
->handle_exit(vcpu
);
4099 static int __vcpu_run(struct kvm_vcpu
*vcpu
)
4103 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
4104 pr_debug("vcpu %d received sipi with vector # %x\n",
4105 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
4106 kvm_lapic_reset(vcpu
);
4107 r
= kvm_arch_vcpu_reset(vcpu
);
4110 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4113 down_read(&vcpu
->kvm
->slots_lock
);
4118 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
)
4119 r
= vcpu_enter_guest(vcpu
);
4121 up_read(&vcpu
->kvm
->slots_lock
);
4122 kvm_vcpu_block(vcpu
);
4123 down_read(&vcpu
->kvm
->slots_lock
);
4124 if (test_and_clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
))
4126 switch(vcpu
->arch
.mp_state
) {
4127 case KVM_MP_STATE_HALTED
:
4128 vcpu
->arch
.mp_state
=
4129 KVM_MP_STATE_RUNNABLE
;
4130 case KVM_MP_STATE_RUNNABLE
:
4132 case KVM_MP_STATE_SIPI_RECEIVED
:
4143 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
4144 if (kvm_cpu_has_pending_timer(vcpu
))
4145 kvm_inject_pending_timer_irqs(vcpu
);
4147 if (dm_request_for_irq_injection(vcpu
)) {
4149 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
4150 ++vcpu
->stat
.request_irq_exits
;
4152 if (signal_pending(current
)) {
4154 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
4155 ++vcpu
->stat
.signal_exits
;
4157 if (need_resched()) {
4158 up_read(&vcpu
->kvm
->slots_lock
);
4160 down_read(&vcpu
->kvm
->slots_lock
);
4164 up_read(&vcpu
->kvm
->slots_lock
);
4165 post_kvm_run_save(vcpu
);
4172 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
4179 if (vcpu
->sigset_active
)
4180 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
4182 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
4183 kvm_vcpu_block(vcpu
);
4184 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
4189 /* re-sync apic's tpr */
4190 if (!irqchip_in_kernel(vcpu
->kvm
))
4191 kvm_set_cr8(vcpu
, kvm_run
->cr8
);
4193 if (vcpu
->arch
.pio
.cur_count
) {
4194 r
= complete_pio(vcpu
);
4198 if (vcpu
->mmio_needed
) {
4199 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
4200 vcpu
->mmio_read_completed
= 1;
4201 vcpu
->mmio_needed
= 0;
4203 down_read(&vcpu
->kvm
->slots_lock
);
4204 r
= emulate_instruction(vcpu
, vcpu
->arch
.mmio_fault_cr2
, 0,
4205 EMULTYPE_NO_DECODE
);
4206 up_read(&vcpu
->kvm
->slots_lock
);
4207 if (r
== EMULATE_DO_MMIO
) {
4209 * Read-modify-write. Back to userspace.
4215 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
)
4216 kvm_register_write(vcpu
, VCPU_REGS_RAX
,
4217 kvm_run
->hypercall
.ret
);
4219 r
= __vcpu_run(vcpu
);
4222 if (vcpu
->sigset_active
)
4223 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
4229 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
4233 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4234 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4235 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4236 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4237 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
4238 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
4239 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4240 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
4241 #ifdef CONFIG_X86_64
4242 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
4243 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
4244 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
4245 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
4246 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
4247 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
4248 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
4249 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
4252 regs
->rip
= kvm_rip_read(vcpu
);
4253 regs
->rflags
= kvm_get_rflags(vcpu
);
4260 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
4264 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
4265 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
4266 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
4267 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
4268 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
4269 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
4270 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
4271 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
4272 #ifdef CONFIG_X86_64
4273 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
4274 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
4275 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
4276 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
4277 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
4278 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
4279 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
4280 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
4283 kvm_rip_write(vcpu
, regs
->rip
);
4284 kvm_set_rflags(vcpu
, regs
->rflags
);
4286 vcpu
->arch
.exception
.pending
= false;
4293 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
4294 struct kvm_segment
*var
, int seg
)
4296 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
4299 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
4301 struct kvm_segment cs
;
4303 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
4307 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
4309 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
4310 struct kvm_sregs
*sregs
)
4312 struct descriptor_table dt
;
4316 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
4317 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
4318 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
4319 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
4320 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
4321 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
4323 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
4324 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
4326 kvm_x86_ops
->get_idt(vcpu
, &dt
);
4327 sregs
->idt
.limit
= dt
.limit
;
4328 sregs
->idt
.base
= dt
.base
;
4329 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
4330 sregs
->gdt
.limit
= dt
.limit
;
4331 sregs
->gdt
.base
= dt
.base
;
4333 sregs
->cr0
= vcpu
->arch
.cr0
;
4334 sregs
->cr2
= vcpu
->arch
.cr2
;
4335 sregs
->cr3
= vcpu
->arch
.cr3
;
4336 sregs
->cr4
= kvm_read_cr4(vcpu
);
4337 sregs
->cr8
= kvm_get_cr8(vcpu
);
4338 sregs
->efer
= vcpu
->arch
.shadow_efer
;
4339 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
4341 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
4343 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
4344 set_bit(vcpu
->arch
.interrupt
.nr
,
4345 (unsigned long *)sregs
->interrupt_bitmap
);
4352 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
4353 struct kvm_mp_state
*mp_state
)
4356 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
4361 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
4362 struct kvm_mp_state
*mp_state
)
4365 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
4370 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
4371 struct kvm_segment
*var
, int seg
)
4373 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
4376 static void seg_desct_to_kvm_desct(struct desc_struct
*seg_desc
, u16 selector
,
4377 struct kvm_segment
*kvm_desct
)
4379 kvm_desct
->base
= get_desc_base(seg_desc
);
4380 kvm_desct
->limit
= get_desc_limit(seg_desc
);
4382 kvm_desct
->limit
<<= 12;
4383 kvm_desct
->limit
|= 0xfff;
4385 kvm_desct
->selector
= selector
;
4386 kvm_desct
->type
= seg_desc
->type
;
4387 kvm_desct
->present
= seg_desc
->p
;
4388 kvm_desct
->dpl
= seg_desc
->dpl
;
4389 kvm_desct
->db
= seg_desc
->d
;
4390 kvm_desct
->s
= seg_desc
->s
;
4391 kvm_desct
->l
= seg_desc
->l
;
4392 kvm_desct
->g
= seg_desc
->g
;
4393 kvm_desct
->avl
= seg_desc
->avl
;
4395 kvm_desct
->unusable
= 1;
4397 kvm_desct
->unusable
= 0;
4398 kvm_desct
->padding
= 0;
4401 static void get_segment_descriptor_dtable(struct kvm_vcpu
*vcpu
,
4403 struct descriptor_table
*dtable
)
4405 if (selector
& 1 << 2) {
4406 struct kvm_segment kvm_seg
;
4408 kvm_get_segment(vcpu
, &kvm_seg
, VCPU_SREG_LDTR
);
4410 if (kvm_seg
.unusable
)
4413 dtable
->limit
= kvm_seg
.limit
;
4414 dtable
->base
= kvm_seg
.base
;
4417 kvm_x86_ops
->get_gdt(vcpu
, dtable
);
4420 /* allowed just for 8 bytes segments */
4421 static int load_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
4422 struct desc_struct
*seg_desc
)
4424 struct descriptor_table dtable
;
4425 u16 index
= selector
>> 3;
4427 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
4429 if (dtable
.limit
< index
* 8 + 7) {
4430 kvm_queue_exception_e(vcpu
, GP_VECTOR
, selector
& 0xfffc);
4433 return kvm_read_guest_virt(dtable
.base
+ index
*8, seg_desc
, sizeof(*seg_desc
), vcpu
);
4436 /* allowed just for 8 bytes segments */
4437 static int save_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
4438 struct desc_struct
*seg_desc
)
4440 struct descriptor_table dtable
;
4441 u16 index
= selector
>> 3;
4443 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
4445 if (dtable
.limit
< index
* 8 + 7)
4447 return kvm_write_guest_virt(dtable
.base
+ index
*8, seg_desc
, sizeof(*seg_desc
), vcpu
);
4450 static gpa_t
get_tss_base_addr(struct kvm_vcpu
*vcpu
,
4451 struct desc_struct
*seg_desc
)
4453 u32 base_addr
= get_desc_base(seg_desc
);
4455 return vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, base_addr
);
4458 static u16
get_segment_selector(struct kvm_vcpu
*vcpu
, int seg
)
4460 struct kvm_segment kvm_seg
;
4462 kvm_get_segment(vcpu
, &kvm_seg
, seg
);
4463 return kvm_seg
.selector
;
4466 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu
*vcpu
,
4468 struct kvm_segment
*kvm_seg
)
4470 struct desc_struct seg_desc
;
4472 if (load_guest_segment_descriptor(vcpu
, selector
, &seg_desc
))
4474 seg_desct_to_kvm_desct(&seg_desc
, selector
, kvm_seg
);
4478 static int kvm_load_realmode_segment(struct kvm_vcpu
*vcpu
, u16 selector
, int seg
)
4480 struct kvm_segment segvar
= {
4481 .base
= selector
<< 4,
4483 .selector
= selector
,
4494 kvm_x86_ops
->set_segment(vcpu
, &segvar
, seg
);
4498 static int is_vm86_segment(struct kvm_vcpu
*vcpu
, int seg
)
4500 return (seg
!= VCPU_SREG_LDTR
) &&
4501 (seg
!= VCPU_SREG_TR
) &&
4502 (kvm_get_rflags(vcpu
) & X86_EFLAGS_VM
);
4505 static void kvm_check_segment_descriptor(struct kvm_vcpu
*vcpu
, int seg
,
4508 /* NULL selector is not valid for CS and SS */
4509 if (seg
== VCPU_SREG_CS
|| seg
== VCPU_SREG_SS
)
4511 kvm_queue_exception_e(vcpu
, TS_VECTOR
, selector
>> 3);
4514 int kvm_load_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
4515 int type_bits
, int seg
)
4517 struct kvm_segment kvm_seg
;
4519 if (is_vm86_segment(vcpu
, seg
) || !(vcpu
->arch
.cr0
& X86_CR0_PE
))
4520 return kvm_load_realmode_segment(vcpu
, selector
, seg
);
4521 if (load_segment_descriptor_to_kvm_desct(vcpu
, selector
, &kvm_seg
))
4524 kvm_check_segment_descriptor(vcpu
, seg
, selector
);
4525 kvm_seg
.type
|= type_bits
;
4527 if (seg
!= VCPU_SREG_SS
&& seg
!= VCPU_SREG_CS
&&
4528 seg
!= VCPU_SREG_LDTR
)
4530 kvm_seg
.unusable
= 1;
4532 kvm_set_segment(vcpu
, &kvm_seg
, seg
);
4536 static void save_state_to_tss32(struct kvm_vcpu
*vcpu
,
4537 struct tss_segment_32
*tss
)
4539 tss
->cr3
= vcpu
->arch
.cr3
;
4540 tss
->eip
= kvm_rip_read(vcpu
);
4541 tss
->eflags
= kvm_get_rflags(vcpu
);
4542 tss
->eax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4543 tss
->ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4544 tss
->edx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4545 tss
->ebx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4546 tss
->esp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4547 tss
->ebp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
4548 tss
->esi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
4549 tss
->edi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
4550 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
4551 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
4552 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
4553 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
4554 tss
->fs
= get_segment_selector(vcpu
, VCPU_SREG_FS
);
4555 tss
->gs
= get_segment_selector(vcpu
, VCPU_SREG_GS
);
4556 tss
->ldt_selector
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
4559 static int load_state_from_tss32(struct kvm_vcpu
*vcpu
,
4560 struct tss_segment_32
*tss
)
4562 kvm_set_cr3(vcpu
, tss
->cr3
);
4564 kvm_rip_write(vcpu
, tss
->eip
);
4565 kvm_set_rflags(vcpu
, tss
->eflags
| 2);
4567 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->eax
);
4568 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->ecx
);
4569 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->edx
);
4570 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->ebx
);
4571 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->esp
);
4572 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->ebp
);
4573 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->esi
);
4574 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->edi
);
4576 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt_selector
, 0, VCPU_SREG_LDTR
))
4579 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
4582 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
4585 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
4588 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
4591 if (kvm_load_segment_descriptor(vcpu
, tss
->fs
, 1, VCPU_SREG_FS
))
4594 if (kvm_load_segment_descriptor(vcpu
, tss
->gs
, 1, VCPU_SREG_GS
))
4599 static void save_state_to_tss16(struct kvm_vcpu
*vcpu
,
4600 struct tss_segment_16
*tss
)
4602 tss
->ip
= kvm_rip_read(vcpu
);
4603 tss
->flag
= kvm_get_rflags(vcpu
);
4604 tss
->ax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4605 tss
->cx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4606 tss
->dx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4607 tss
->bx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4608 tss
->sp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4609 tss
->bp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
4610 tss
->si
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
4611 tss
->di
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
4613 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
4614 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
4615 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
4616 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
4617 tss
->ldt
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
4620 static int load_state_from_tss16(struct kvm_vcpu
*vcpu
,
4621 struct tss_segment_16
*tss
)
4623 kvm_rip_write(vcpu
, tss
->ip
);
4624 kvm_set_rflags(vcpu
, tss
->flag
| 2);
4625 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->ax
);
4626 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->cx
);
4627 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->dx
);
4628 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->bx
);
4629 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->sp
);
4630 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->bp
);
4631 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->si
);
4632 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->di
);
4634 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt
, 0, VCPU_SREG_LDTR
))
4637 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
4640 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
4643 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
4646 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
4651 static int kvm_task_switch_16(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
4652 u16 old_tss_sel
, u32 old_tss_base
,
4653 struct desc_struct
*nseg_desc
)
4655 struct tss_segment_16 tss_segment_16
;
4658 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
4659 sizeof tss_segment_16
))
4662 save_state_to_tss16(vcpu
, &tss_segment_16
);
4664 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
4665 sizeof tss_segment_16
))
4668 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
4669 &tss_segment_16
, sizeof tss_segment_16
))
4672 if (old_tss_sel
!= 0xffff) {
4673 tss_segment_16
.prev_task_link
= old_tss_sel
;
4675 if (kvm_write_guest(vcpu
->kvm
,
4676 get_tss_base_addr(vcpu
, nseg_desc
),
4677 &tss_segment_16
.prev_task_link
,
4678 sizeof tss_segment_16
.prev_task_link
))
4682 if (load_state_from_tss16(vcpu
, &tss_segment_16
))
4690 static int kvm_task_switch_32(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
4691 u16 old_tss_sel
, u32 old_tss_base
,
4692 struct desc_struct
*nseg_desc
)
4694 struct tss_segment_32 tss_segment_32
;
4697 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
4698 sizeof tss_segment_32
))
4701 save_state_to_tss32(vcpu
, &tss_segment_32
);
4703 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
4704 sizeof tss_segment_32
))
4707 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
4708 &tss_segment_32
, sizeof tss_segment_32
))
4711 if (old_tss_sel
!= 0xffff) {
4712 tss_segment_32
.prev_task_link
= old_tss_sel
;
4714 if (kvm_write_guest(vcpu
->kvm
,
4715 get_tss_base_addr(vcpu
, nseg_desc
),
4716 &tss_segment_32
.prev_task_link
,
4717 sizeof tss_segment_32
.prev_task_link
))
4721 if (load_state_from_tss32(vcpu
, &tss_segment_32
))
4729 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int reason
)
4731 struct kvm_segment tr_seg
;
4732 struct desc_struct cseg_desc
;
4733 struct desc_struct nseg_desc
;
4735 u32 old_tss_base
= get_segment_base(vcpu
, VCPU_SREG_TR
);
4736 u16 old_tss_sel
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
4738 old_tss_base
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, old_tss_base
);
4740 /* FIXME: Handle errors. Failure to read either TSS or their
4741 * descriptors should generate a pagefault.
4743 if (load_guest_segment_descriptor(vcpu
, tss_selector
, &nseg_desc
))
4746 if (load_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
))
4749 if (reason
!= TASK_SWITCH_IRET
) {
4752 cpl
= kvm_x86_ops
->get_cpl(vcpu
);
4753 if ((tss_selector
& 3) > nseg_desc
.dpl
|| cpl
> nseg_desc
.dpl
) {
4754 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
4759 if (!nseg_desc
.p
|| get_desc_limit(&nseg_desc
) < 0x67) {
4760 kvm_queue_exception_e(vcpu
, TS_VECTOR
, tss_selector
& 0xfffc);
4764 if (reason
== TASK_SWITCH_IRET
|| reason
== TASK_SWITCH_JMP
) {
4765 cseg_desc
.type
&= ~(1 << 1); //clear the B flag
4766 save_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
);
4769 if (reason
== TASK_SWITCH_IRET
) {
4770 u32 eflags
= kvm_get_rflags(vcpu
);
4771 kvm_set_rflags(vcpu
, eflags
& ~X86_EFLAGS_NT
);
4774 /* set back link to prev task only if NT bit is set in eflags
4775 note that old_tss_sel is not used afetr this point */
4776 if (reason
!= TASK_SWITCH_CALL
&& reason
!= TASK_SWITCH_GATE
)
4777 old_tss_sel
= 0xffff;
4779 if (nseg_desc
.type
& 8)
4780 ret
= kvm_task_switch_32(vcpu
, tss_selector
, old_tss_sel
,
4781 old_tss_base
, &nseg_desc
);
4783 ret
= kvm_task_switch_16(vcpu
, tss_selector
, old_tss_sel
,
4784 old_tss_base
, &nseg_desc
);
4786 if (reason
== TASK_SWITCH_CALL
|| reason
== TASK_SWITCH_GATE
) {
4787 u32 eflags
= kvm_get_rflags(vcpu
);
4788 kvm_set_rflags(vcpu
, eflags
| X86_EFLAGS_NT
);
4791 if (reason
!= TASK_SWITCH_IRET
) {
4792 nseg_desc
.type
|= (1 << 1);
4793 save_guest_segment_descriptor(vcpu
, tss_selector
,
4797 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
| X86_CR0_TS
);
4798 seg_desct_to_kvm_desct(&nseg_desc
, tss_selector
, &tr_seg
);
4800 kvm_set_segment(vcpu
, &tr_seg
, VCPU_SREG_TR
);
4804 EXPORT_SYMBOL_GPL(kvm_task_switch
);
4806 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
4807 struct kvm_sregs
*sregs
)
4809 int mmu_reset_needed
= 0;
4810 int pending_vec
, max_bits
;
4811 struct descriptor_table dt
;
4815 dt
.limit
= sregs
->idt
.limit
;
4816 dt
.base
= sregs
->idt
.base
;
4817 kvm_x86_ops
->set_idt(vcpu
, &dt
);
4818 dt
.limit
= sregs
->gdt
.limit
;
4819 dt
.base
= sregs
->gdt
.base
;
4820 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
4822 vcpu
->arch
.cr2
= sregs
->cr2
;
4823 mmu_reset_needed
|= vcpu
->arch
.cr3
!= sregs
->cr3
;
4824 vcpu
->arch
.cr3
= sregs
->cr3
;
4826 kvm_set_cr8(vcpu
, sregs
->cr8
);
4828 mmu_reset_needed
|= vcpu
->arch
.shadow_efer
!= sregs
->efer
;
4829 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
4830 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
4832 mmu_reset_needed
|= vcpu
->arch
.cr0
!= sregs
->cr0
;
4833 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
4834 vcpu
->arch
.cr0
= sregs
->cr0
;
4836 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
4837 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
4838 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
4839 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
4840 mmu_reset_needed
= 1;
4843 if (mmu_reset_needed
)
4844 kvm_mmu_reset_context(vcpu
);
4846 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
4847 pending_vec
= find_first_bit(
4848 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
4849 if (pending_vec
< max_bits
) {
4850 kvm_queue_interrupt(vcpu
, pending_vec
, false);
4851 pr_debug("Set back pending irq %d\n", pending_vec
);
4852 if (irqchip_in_kernel(vcpu
->kvm
))
4853 kvm_pic_clear_isr_ack(vcpu
->kvm
);
4856 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
4857 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
4858 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
4859 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
4860 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
4861 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
4863 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
4864 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
4866 update_cr8_intercept(vcpu
);
4868 /* Older userspace won't unhalt the vcpu on reset. */
4869 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
4870 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
4871 !(vcpu
->arch
.cr0
& X86_CR0_PE
))
4872 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4879 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
4880 struct kvm_guest_debug
*dbg
)
4882 unsigned long rflags
;
4887 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
4889 if (vcpu
->arch
.exception
.pending
)
4891 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
4892 kvm_queue_exception(vcpu
, DB_VECTOR
);
4894 kvm_queue_exception(vcpu
, BP_VECTOR
);
4898 * Read rflags as long as potentially injected trace flags are still
4901 rflags
= kvm_get_rflags(vcpu
);
4903 vcpu
->guest_debug
= dbg
->control
;
4904 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
4905 vcpu
->guest_debug
= 0;
4907 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
4908 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
4909 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
4910 vcpu
->arch
.switch_db_regs
=
4911 (dbg
->arch
.debugreg
[7] & DR7_BP_EN_MASK
);
4913 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
4914 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
4915 vcpu
->arch
.switch_db_regs
= (vcpu
->arch
.dr7
& DR7_BP_EN_MASK
);
4918 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
) {
4919 vcpu
->arch
.singlestep_cs
=
4920 get_segment_selector(vcpu
, VCPU_SREG_CS
);
4921 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
);
4925 * Trigger an rflags update that will inject or remove the trace
4928 kvm_set_rflags(vcpu
, rflags
);
4930 kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
4941 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4942 * we have asm/x86/processor.h
4953 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4954 #ifdef CONFIG_X86_64
4955 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4957 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4962 * Translate a guest virtual address to a guest physical address.
4964 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
4965 struct kvm_translation
*tr
)
4967 unsigned long vaddr
= tr
->linear_address
;
4971 down_read(&vcpu
->kvm
->slots_lock
);
4972 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, vaddr
);
4973 up_read(&vcpu
->kvm
->slots_lock
);
4974 tr
->physical_address
= gpa
;
4975 tr
->valid
= gpa
!= UNMAPPED_GVA
;
4983 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4985 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4989 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
4990 fpu
->fcw
= fxsave
->cwd
;
4991 fpu
->fsw
= fxsave
->swd
;
4992 fpu
->ftwx
= fxsave
->twd
;
4993 fpu
->last_opcode
= fxsave
->fop
;
4994 fpu
->last_ip
= fxsave
->rip
;
4995 fpu
->last_dp
= fxsave
->rdp
;
4996 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
5003 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
5005 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
5009 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
5010 fxsave
->cwd
= fpu
->fcw
;
5011 fxsave
->swd
= fpu
->fsw
;
5012 fxsave
->twd
= fpu
->ftwx
;
5013 fxsave
->fop
= fpu
->last_opcode
;
5014 fxsave
->rip
= fpu
->last_ip
;
5015 fxsave
->rdp
= fpu
->last_dp
;
5016 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
5023 void fx_init(struct kvm_vcpu
*vcpu
)
5025 unsigned after_mxcsr_mask
;
5028 * Touch the fpu the first time in non atomic context as if
5029 * this is the first fpu instruction the exception handler
5030 * will fire before the instruction returns and it'll have to
5031 * allocate ram with GFP_KERNEL.
5034 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
5036 /* Initialize guest FPU by resetting ours and saving into guest's */
5038 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
5040 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
5041 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
5044 vcpu
->arch
.cr0
|= X86_CR0_ET
;
5045 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
5046 vcpu
->arch
.guest_fx_image
.mxcsr
= 0x1f80;
5047 memset((void *)&vcpu
->arch
.guest_fx_image
+ after_mxcsr_mask
,
5048 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
5050 EXPORT_SYMBOL_GPL(fx_init
);
5052 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
5054 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
5057 vcpu
->guest_fpu_loaded
= 1;
5058 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
5059 kvm_fx_restore(&vcpu
->arch
.guest_fx_image
);
5061 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
5063 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
5065 if (!vcpu
->guest_fpu_loaded
)
5068 vcpu
->guest_fpu_loaded
= 0;
5069 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
5070 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
5071 ++vcpu
->stat
.fpu_reload
;
5073 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
5075 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
5077 if (vcpu
->arch
.time_page
) {
5078 kvm_release_page_dirty(vcpu
->arch
.time_page
);
5079 vcpu
->arch
.time_page
= NULL
;
5082 kvm_x86_ops
->vcpu_free(vcpu
);
5085 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
5088 return kvm_x86_ops
->vcpu_create(kvm
, id
);
5091 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
5095 /* We do fxsave: this must be aligned. */
5096 BUG_ON((unsigned long)&vcpu
->arch
.host_fx_image
& 0xF);
5098 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
5100 r
= kvm_arch_vcpu_reset(vcpu
);
5102 r
= kvm_mmu_setup(vcpu
);
5109 kvm_x86_ops
->vcpu_free(vcpu
);
5113 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
5116 kvm_mmu_unload(vcpu
);
5119 kvm_x86_ops
->vcpu_free(vcpu
);
5122 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
5124 vcpu
->arch
.nmi_pending
= false;
5125 vcpu
->arch
.nmi_injected
= false;
5127 vcpu
->arch
.switch_db_regs
= 0;
5128 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
5129 vcpu
->arch
.dr6
= DR6_FIXED_1
;
5130 vcpu
->arch
.dr7
= DR7_FIXED_1
;
5132 return kvm_x86_ops
->vcpu_reset(vcpu
);
5135 int kvm_arch_hardware_enable(void *garbage
)
5138 * Since this may be called from a hotplug notifcation,
5139 * we can't get the CPU frequency directly.
5141 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
5142 int cpu
= raw_smp_processor_id();
5143 per_cpu(cpu_tsc_khz
, cpu
) = 0;
5146 kvm_shared_msr_cpu_online();
5148 return kvm_x86_ops
->hardware_enable(garbage
);
5151 void kvm_arch_hardware_disable(void *garbage
)
5153 kvm_x86_ops
->hardware_disable(garbage
);
5154 drop_user_return_notifiers(garbage
);
5157 int kvm_arch_hardware_setup(void)
5159 return kvm_x86_ops
->hardware_setup();
5162 void kvm_arch_hardware_unsetup(void)
5164 kvm_x86_ops
->hardware_unsetup();
5167 void kvm_arch_check_processor_compat(void *rtn
)
5169 kvm_x86_ops
->check_processor_compatibility(rtn
);
5172 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
5178 BUG_ON(vcpu
->kvm
== NULL
);
5181 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
5182 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
5183 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
5185 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
5187 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
5192 vcpu
->arch
.pio_data
= page_address(page
);
5194 r
= kvm_mmu_create(vcpu
);
5196 goto fail_free_pio_data
;
5198 if (irqchip_in_kernel(kvm
)) {
5199 r
= kvm_create_lapic(vcpu
);
5201 goto fail_mmu_destroy
;
5204 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
5206 if (!vcpu
->arch
.mce_banks
) {
5208 goto fail_free_lapic
;
5210 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
5214 kvm_free_lapic(vcpu
);
5216 kvm_mmu_destroy(vcpu
);
5218 free_page((unsigned long)vcpu
->arch
.pio_data
);
5223 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
5225 kfree(vcpu
->arch
.mce_banks
);
5226 kvm_free_lapic(vcpu
);
5227 down_read(&vcpu
->kvm
->slots_lock
);
5228 kvm_mmu_destroy(vcpu
);
5229 up_read(&vcpu
->kvm
->slots_lock
);
5230 free_page((unsigned long)vcpu
->arch
.pio_data
);
5233 struct kvm
*kvm_arch_create_vm(void)
5235 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
5238 return ERR_PTR(-ENOMEM
);
5240 kvm
->arch
.aliases
= kzalloc(sizeof(struct kvm_mem_aliases
), GFP_KERNEL
);
5241 if (!kvm
->arch
.aliases
) {
5243 return ERR_PTR(-ENOMEM
);
5246 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
5247 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
5249 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5250 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
5252 rdtscll(kvm
->arch
.vm_init_tsc
);
5257 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
5260 kvm_mmu_unload(vcpu
);
5264 static void kvm_free_vcpus(struct kvm
*kvm
)
5267 struct kvm_vcpu
*vcpu
;
5270 * Unpin any mmu pages first.
5272 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5273 kvm_unload_vcpu_mmu(vcpu
);
5274 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5275 kvm_arch_vcpu_free(vcpu
);
5277 mutex_lock(&kvm
->lock
);
5278 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
5279 kvm
->vcpus
[i
] = NULL
;
5281 atomic_set(&kvm
->online_vcpus
, 0);
5282 mutex_unlock(&kvm
->lock
);
5285 void kvm_arch_sync_events(struct kvm
*kvm
)
5287 kvm_free_all_assigned_devices(kvm
);
5290 void kvm_arch_destroy_vm(struct kvm
*kvm
)
5292 kvm_iommu_unmap_guest(kvm
);
5294 kfree(kvm
->arch
.vpic
);
5295 kfree(kvm
->arch
.vioapic
);
5296 kvm_free_vcpus(kvm
);
5297 kvm_free_physmem(kvm
);
5298 if (kvm
->arch
.apic_access_page
)
5299 put_page(kvm
->arch
.apic_access_page
);
5300 if (kvm
->arch
.ept_identity_pagetable
)
5301 put_page(kvm
->arch
.ept_identity_pagetable
);
5302 kfree(kvm
->arch
.aliases
);
5306 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
5307 struct kvm_memory_slot
*memslot
,
5308 struct kvm_memory_slot old
,
5309 struct kvm_userspace_memory_region
*mem
,
5312 int npages
= memslot
->npages
;
5314 /*To keep backward compatibility with older userspace,
5315 *x86 needs to hanlde !user_alloc case.
5318 if (npages
&& !old
.rmap
) {
5319 unsigned long userspace_addr
;
5321 down_write(¤t
->mm
->mmap_sem
);
5322 userspace_addr
= do_mmap(NULL
, 0,
5324 PROT_READ
| PROT_WRITE
,
5325 MAP_PRIVATE
| MAP_ANONYMOUS
,
5327 up_write(¤t
->mm
->mmap_sem
);
5329 if (IS_ERR((void *)userspace_addr
))
5330 return PTR_ERR((void *)userspace_addr
);
5332 memslot
->userspace_addr
= userspace_addr
;
5340 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
5341 struct kvm_userspace_memory_region
*mem
,
5342 struct kvm_memory_slot old
,
5346 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
5348 if (!user_alloc
&& !old
.user_alloc
&& old
.rmap
&& !npages
) {
5351 down_write(¤t
->mm
->mmap_sem
);
5352 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
5353 old
.npages
* PAGE_SIZE
);
5354 up_write(¤t
->mm
->mmap_sem
);
5357 "kvm_vm_ioctl_set_memory_region: "
5358 "failed to munmap memory\n");
5361 spin_lock(&kvm
->mmu_lock
);
5362 if (!kvm
->arch
.n_requested_mmu_pages
) {
5363 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
5364 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
5367 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
5368 spin_unlock(&kvm
->mmu_lock
);
5371 void kvm_arch_flush_shadow(struct kvm
*kvm
)
5373 kvm_mmu_zap_all(kvm
);
5374 kvm_reload_remote_mmus(kvm
);
5377 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
5379 return vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
5380 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
5381 || vcpu
->arch
.nmi_pending
||
5382 (kvm_arch_interrupt_allowed(vcpu
) &&
5383 kvm_cpu_has_interrupt(vcpu
));
5386 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
5389 int cpu
= vcpu
->cpu
;
5391 if (waitqueue_active(&vcpu
->wq
)) {
5392 wake_up_interruptible(&vcpu
->wq
);
5393 ++vcpu
->stat
.halt_wakeup
;
5397 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
5398 if (!test_and_set_bit(KVM_REQ_KICK
, &vcpu
->requests
))
5399 smp_send_reschedule(cpu
);
5403 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
5405 return kvm_x86_ops
->interrupt_allowed(vcpu
);
5408 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
5410 unsigned long rflags
;
5412 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5413 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
5414 rflags
&= ~(unsigned long)(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
5417 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
5419 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
5421 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
5422 vcpu
->arch
.singlestep_cs
==
5423 get_segment_selector(vcpu
, VCPU_SREG_CS
) &&
5424 vcpu
->arch
.singlestep_rip
== kvm_rip_read(vcpu
))
5425 rflags
|= X86_EFLAGS_TF
| X86_EFLAGS_RF
;
5426 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
5428 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
5430 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
5431 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
5432 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
5433 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
5434 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
5435 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
5436 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
5437 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
5438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
5439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
5440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);