2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
20 #include <linux/kvm.h>
21 #include <linux/module.h>
22 #include <linux/errno.h>
23 #include <asm/processor.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <asm/uaccess.h>
31 #include <linux/reboot.h>
33 #include <linux/debugfs.h>
34 #include <linux/highmem.h>
35 #include <linux/file.h>
38 #include "x86_emulate.h"
39 #include "segment_descriptor.h"
41 MODULE_AUTHOR("Qumranet");
42 MODULE_LICENSE("GPL");
44 struct kvm_arch_ops
*kvm_arch_ops
;
45 struct kvm_stat kvm_stat
;
46 EXPORT_SYMBOL_GPL(kvm_stat
);
48 static struct kvm_stats_debugfs_item
{
51 struct dentry
*dentry
;
52 } debugfs_entries
[] = {
53 { "pf_fixed", &kvm_stat
.pf_fixed
},
54 { "pf_guest", &kvm_stat
.pf_guest
},
55 { "tlb_flush", &kvm_stat
.tlb_flush
},
56 { "invlpg", &kvm_stat
.invlpg
},
57 { "exits", &kvm_stat
.exits
},
58 { "io_exits", &kvm_stat
.io_exits
},
59 { "mmio_exits", &kvm_stat
.mmio_exits
},
60 { "signal_exits", &kvm_stat
.signal_exits
},
61 { "irq_exits", &kvm_stat
.irq_exits
},
65 static struct dentry
*debugfs_dir
;
67 #define MAX_IO_MSRS 256
69 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
70 #define LMSW_GUEST_MASK 0x0eULL
71 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
72 #define CR8_RESEVED_BITS (~0x0fULL)
73 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
76 // LDT or TSS descriptor in the GDT. 16 bytes.
77 struct segment_descriptor_64
{
78 struct segment_descriptor s
;
85 unsigned long segment_base(u16 selector
)
87 struct descriptor_table gdt
;
88 struct segment_descriptor
*d
;
89 unsigned long table_base
;
90 typedef unsigned long ul
;
96 asm ("sgdt %0" : "=m"(gdt
));
97 table_base
= gdt
.base
;
99 if (selector
& 4) { /* from ldt */
102 asm ("sldt %0" : "=g"(ldt_selector
));
103 table_base
= segment_base(ldt_selector
);
105 d
= (struct segment_descriptor
*)(table_base
+ (selector
& ~7));
106 v
= d
->base_low
| ((ul
)d
->base_mid
<< 16) | ((ul
)d
->base_high
<< 24);
109 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
110 v
|= ((ul
)((struct segment_descriptor_64
*)d
)->base_higher
) << 32;
114 EXPORT_SYMBOL_GPL(segment_base
);
116 static inline int valid_vcpu(int n
)
118 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
121 int kvm_read_guest(struct kvm_vcpu
*vcpu
,
126 unsigned char *host_buf
= dest
;
127 unsigned long req_size
= size
;
135 paddr
= gva_to_hpa(vcpu
, addr
);
137 if (is_error_hpa(paddr
))
140 guest_buf
= (hva_t
)kmap_atomic(
141 pfn_to_page(paddr
>> PAGE_SHIFT
),
143 offset
= addr
& ~PAGE_MASK
;
145 now
= min(size
, PAGE_SIZE
- offset
);
146 memcpy(host_buf
, (void*)guest_buf
, now
);
150 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
152 return req_size
- size
;
154 EXPORT_SYMBOL_GPL(kvm_read_guest
);
156 int kvm_write_guest(struct kvm_vcpu
*vcpu
,
161 unsigned char *host_buf
= data
;
162 unsigned long req_size
= size
;
170 paddr
= gva_to_hpa(vcpu
, addr
);
172 if (is_error_hpa(paddr
))
175 guest_buf
= (hva_t
)kmap_atomic(
176 pfn_to_page(paddr
>> PAGE_SHIFT
), KM_USER0
);
177 offset
= addr
& ~PAGE_MASK
;
179 now
= min(size
, PAGE_SIZE
- offset
);
180 memcpy((void*)guest_buf
, host_buf
, now
);
184 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
186 return req_size
- size
;
188 EXPORT_SYMBOL_GPL(kvm_write_guest
);
190 static int vcpu_slot(struct kvm_vcpu
*vcpu
)
192 return vcpu
- vcpu
->kvm
->vcpus
;
196 * Switches to specified vcpu, until a matching vcpu_put()
198 static struct kvm_vcpu
*vcpu_load(struct kvm
*kvm
, int vcpu_slot
)
200 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[vcpu_slot
];
202 mutex_lock(&vcpu
->mutex
);
203 if (unlikely(!vcpu
->vmcs
)) {
204 mutex_unlock(&vcpu
->mutex
);
207 return kvm_arch_ops
->vcpu_load(vcpu
);
210 static void vcpu_put(struct kvm_vcpu
*vcpu
)
212 kvm_arch_ops
->vcpu_put(vcpu
);
213 mutex_unlock(&vcpu
->mutex
);
216 static int kvm_dev_open(struct inode
*inode
, struct file
*filp
)
218 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
224 spin_lock_init(&kvm
->lock
);
225 INIT_LIST_HEAD(&kvm
->active_mmu_pages
);
226 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
227 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[i
];
229 mutex_init(&vcpu
->mutex
);
230 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
231 INIT_LIST_HEAD(&vcpu
->free_pages
);
233 filp
->private_data
= kvm
;
238 * Free any memory in @free but not in @dont.
240 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
241 struct kvm_memory_slot
*dont
)
245 if (!dont
|| free
->phys_mem
!= dont
->phys_mem
)
246 if (free
->phys_mem
) {
247 for (i
= 0; i
< free
->npages
; ++i
)
248 __free_page(free
->phys_mem
[i
]);
249 vfree(free
->phys_mem
);
252 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
253 vfree(free
->dirty_bitmap
);
257 free
->dirty_bitmap
= 0;
260 static void kvm_free_physmem(struct kvm
*kvm
)
264 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
265 kvm_free_physmem_slot(&kvm
->memslots
[i
], 0);
268 static void kvm_free_vcpu(struct kvm_vcpu
*vcpu
)
270 kvm_arch_ops
->vcpu_free(vcpu
);
271 kvm_mmu_destroy(vcpu
);
274 static void kvm_free_vcpus(struct kvm
*kvm
)
278 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
279 kvm_free_vcpu(&kvm
->vcpus
[i
]);
282 static int kvm_dev_release(struct inode
*inode
, struct file
*filp
)
284 struct kvm
*kvm
= filp
->private_data
;
287 kvm_free_physmem(kvm
);
292 static void inject_gp(struct kvm_vcpu
*vcpu
)
294 kvm_arch_ops
->inject_gp(vcpu
, 0);
297 static int pdptrs_have_reserved_bits_set(struct kvm_vcpu
*vcpu
,
300 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
301 unsigned offset
= (cr3
& (PAGE_SIZE
-1)) >> 5;
305 struct kvm_memory_slot
*memslot
;
307 spin_lock(&vcpu
->kvm
->lock
);
308 memslot
= gfn_to_memslot(vcpu
->kvm
, pdpt_gfn
);
309 /* FIXME: !memslot - emulate? 0xff? */
310 pdpt
= kmap_atomic(gfn_to_page(memslot
, pdpt_gfn
), KM_USER0
);
312 for (i
= 0; i
< 4; ++i
) {
313 pdpte
= pdpt
[offset
+ i
];
314 if ((pdpte
& 1) && (pdpte
& 0xfffffff0000001e6ull
))
318 kunmap_atomic(pdpt
, KM_USER0
);
319 spin_unlock(&vcpu
->kvm
->lock
);
324 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
326 if (cr0
& CR0_RESEVED_BITS
) {
327 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
333 if ((cr0
& CR0_NW_MASK
) && !(cr0
& CR0_CD_MASK
)) {
334 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
339 if ((cr0
& CR0_PG_MASK
) && !(cr0
& CR0_PE_MASK
)) {
340 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
341 "and a clear PE flag\n");
346 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
348 if ((vcpu
->shadow_efer
& EFER_LME
)) {
352 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
353 "in long mode while PAE is disabled\n");
357 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
359 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
360 "in long mode while CS.L == 1\n");
368 pdptrs_have_reserved_bits_set(vcpu
, vcpu
->cr3
)) {
369 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
377 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
380 spin_lock(&vcpu
->kvm
->lock
);
381 kvm_mmu_reset_context(vcpu
);
382 spin_unlock(&vcpu
->kvm
->lock
);
385 EXPORT_SYMBOL_GPL(set_cr0
);
387 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
389 set_cr0(vcpu
, (vcpu
->cr0
& ~0x0ful
) | (msw
& 0x0f));
391 EXPORT_SYMBOL_GPL(lmsw
);
393 void set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
395 if (cr4
& CR4_RESEVED_BITS
) {
396 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
401 if (kvm_arch_ops
->is_long_mode(vcpu
)) {
402 if (!(cr4
& CR4_PAE_MASK
)) {
403 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
408 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& CR4_PAE_MASK
)
409 && pdptrs_have_reserved_bits_set(vcpu
, vcpu
->cr3
)) {
410 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
414 if (cr4
& CR4_VMXE_MASK
) {
415 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
419 kvm_arch_ops
->set_cr4(vcpu
, cr4
);
420 spin_lock(&vcpu
->kvm
->lock
);
421 kvm_mmu_reset_context(vcpu
);
422 spin_unlock(&vcpu
->kvm
->lock
);
424 EXPORT_SYMBOL_GPL(set_cr4
);
426 void set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
428 if (kvm_arch_ops
->is_long_mode(vcpu
)) {
429 if ( cr3
& CR3_L_MODE_RESEVED_BITS
) {
430 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
435 if (cr3
& CR3_RESEVED_BITS
) {
436 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
440 if (is_paging(vcpu
) && is_pae(vcpu
) &&
441 pdptrs_have_reserved_bits_set(vcpu
, cr3
)) {
442 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
450 spin_lock(&vcpu
->kvm
->lock
);
451 vcpu
->mmu
.new_cr3(vcpu
);
452 spin_unlock(&vcpu
->kvm
->lock
);
454 EXPORT_SYMBOL_GPL(set_cr3
);
456 void set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
458 if ( cr8
& CR8_RESEVED_BITS
) {
459 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
465 EXPORT_SYMBOL_GPL(set_cr8
);
467 void fx_init(struct kvm_vcpu
*vcpu
)
469 struct __attribute__ ((__packed__
)) fx_image_s
{
475 u64 operand
;// fpu dp
481 fx_save(vcpu
->host_fx_image
);
483 fx_save(vcpu
->guest_fx_image
);
484 fx_restore(vcpu
->host_fx_image
);
486 fx_image
= (struct fx_image_s
*)vcpu
->guest_fx_image
;
487 fx_image
->mxcsr
= 0x1f80;
488 memset(vcpu
->guest_fx_image
+ sizeof(struct fx_image_s
),
489 0, FX_IMAGE_SIZE
- sizeof(struct fx_image_s
));
491 EXPORT_SYMBOL_GPL(fx_init
);
494 * Creates some virtual cpus. Good luck creating more than one.
496 static int kvm_dev_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
499 struct kvm_vcpu
*vcpu
;
505 vcpu
= &kvm
->vcpus
[n
];
507 mutex_lock(&vcpu
->mutex
);
510 mutex_unlock(&vcpu
->mutex
);
514 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
516 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
518 vcpu
->cpu
= -1; /* First load will set up TR */
520 r
= kvm_arch_ops
->vcpu_create(vcpu
);
524 kvm_arch_ops
->vcpu_load(vcpu
);
526 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
528 r
= kvm_mmu_init(vcpu
);
539 mutex_unlock(&vcpu
->mutex
);
545 * Allocate some memory and give it an address in the guest physical address
548 * Discontiguous memory is allowed, mostly for framebuffers.
550 static int kvm_dev_ioctl_set_memory_region(struct kvm
*kvm
,
551 struct kvm_memory_region
*mem
)
555 unsigned long npages
;
557 struct kvm_memory_slot
*memslot
;
558 struct kvm_memory_slot old
, new;
559 int memory_config_version
;
562 /* General sanity checks */
563 if (mem
->memory_size
& (PAGE_SIZE
- 1))
565 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
567 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
569 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
572 memslot
= &kvm
->memslots
[mem
->slot
];
573 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
574 npages
= mem
->memory_size
>> PAGE_SHIFT
;
577 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
580 spin_lock(&kvm
->lock
);
582 memory_config_version
= kvm
->memory_config_version
;
583 new = old
= *memslot
;
585 new.base_gfn
= base_gfn
;
587 new.flags
= mem
->flags
;
589 /* Disallow changing a memory slot's size. */
591 if (npages
&& old
.npages
&& npages
!= old
.npages
)
594 /* Check for overlaps */
596 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
597 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
601 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
602 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
606 * Do memory allocations outside lock. memory_config_version will
609 spin_unlock(&kvm
->lock
);
611 /* Deallocate if slot is being removed */
615 /* Free page dirty bitmap if unneeded */
616 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
617 new.dirty_bitmap
= 0;
621 /* Allocate if a slot is being created */
622 if (npages
&& !new.phys_mem
) {
623 new.phys_mem
= vmalloc(npages
* sizeof(struct page
*));
628 memset(new.phys_mem
, 0, npages
* sizeof(struct page
*));
629 for (i
= 0; i
< npages
; ++i
) {
630 new.phys_mem
[i
] = alloc_page(GFP_HIGHUSER
632 if (!new.phys_mem
[i
])
637 /* Allocate page dirty bitmap if needed */
638 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
639 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
641 new.dirty_bitmap
= vmalloc(dirty_bytes
);
642 if (!new.dirty_bitmap
)
644 memset(new.dirty_bitmap
, 0, dirty_bytes
);
647 spin_lock(&kvm
->lock
);
649 if (memory_config_version
!= kvm
->memory_config_version
) {
650 spin_unlock(&kvm
->lock
);
651 kvm_free_physmem_slot(&new, &old
);
659 if (mem
->slot
>= kvm
->nmemslots
)
660 kvm
->nmemslots
= mem
->slot
+ 1;
663 ++kvm
->memory_config_version
;
665 spin_unlock(&kvm
->lock
);
667 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
668 struct kvm_vcpu
*vcpu
;
670 vcpu
= vcpu_load(kvm
, i
);
673 kvm_mmu_reset_context(vcpu
);
677 kvm_free_physmem_slot(&old
, &new);
681 spin_unlock(&kvm
->lock
);
683 kvm_free_physmem_slot(&new, &old
);
689 * Get (and clear) the dirty memory log for a memory slot.
691 static int kvm_dev_ioctl_get_dirty_log(struct kvm
*kvm
,
692 struct kvm_dirty_log
*log
)
694 struct kvm_memory_slot
*memslot
;
697 unsigned long any
= 0;
699 spin_lock(&kvm
->lock
);
702 * Prevent changes to guest memory configuration even while the lock
706 spin_unlock(&kvm
->lock
);
708 if (log
->slot
>= KVM_MEMORY_SLOTS
)
711 memslot
= &kvm
->memslots
[log
->slot
];
713 if (!memslot
->dirty_bitmap
)
716 n
= ALIGN(memslot
->npages
, 8) / 8;
718 for (i
= 0; !any
&& i
< n
; ++i
)
719 any
= memslot
->dirty_bitmap
[i
];
722 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
727 spin_lock(&kvm
->lock
);
728 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
729 spin_unlock(&kvm
->lock
);
730 memset(memslot
->dirty_bitmap
, 0, n
);
731 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
732 struct kvm_vcpu
*vcpu
= vcpu_load(kvm
, i
);
736 kvm_arch_ops
->tlb_flush(vcpu
);
744 spin_lock(&kvm
->lock
);
746 spin_unlock(&kvm
->lock
);
750 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
754 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
755 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
757 if (gfn
>= memslot
->base_gfn
758 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
763 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
765 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
768 struct kvm_memory_slot
*memslot
= 0;
769 unsigned long rel_gfn
;
771 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
772 memslot
= &kvm
->memslots
[i
];
774 if (gfn
>= memslot
->base_gfn
775 && gfn
< memslot
->base_gfn
+ memslot
->npages
) {
777 if (!memslot
|| !memslot
->dirty_bitmap
)
780 rel_gfn
= gfn
- memslot
->base_gfn
;
783 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
784 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
790 static int emulator_read_std(unsigned long addr
,
793 struct x86_emulate_ctxt
*ctxt
)
795 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
799 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
800 unsigned offset
= addr
& (PAGE_SIZE
-1);
801 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
803 struct kvm_memory_slot
*memslot
;
806 if (gpa
== UNMAPPED_GVA
)
807 return X86EMUL_PROPAGATE_FAULT
;
808 pfn
= gpa
>> PAGE_SHIFT
;
809 memslot
= gfn_to_memslot(vcpu
->kvm
, pfn
);
811 return X86EMUL_UNHANDLEABLE
;
812 page
= kmap_atomic(gfn_to_page(memslot
, pfn
), KM_USER0
);
814 memcpy(data
, page
+ offset
, tocopy
);
816 kunmap_atomic(page
, KM_USER0
);
823 return X86EMUL_CONTINUE
;
826 static int emulator_write_std(unsigned long addr
,
829 struct x86_emulate_ctxt
*ctxt
)
831 printk(KERN_ERR
"emulator_write_std: addr %lx n %d\n",
833 return X86EMUL_UNHANDLEABLE
;
836 static int emulator_read_emulated(unsigned long addr
,
839 struct x86_emulate_ctxt
*ctxt
)
841 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
843 if (vcpu
->mmio_read_completed
) {
844 memcpy(val
, vcpu
->mmio_data
, bytes
);
845 vcpu
->mmio_read_completed
= 0;
846 return X86EMUL_CONTINUE
;
847 } else if (emulator_read_std(addr
, val
, bytes
, ctxt
)
849 return X86EMUL_CONTINUE
;
851 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
852 if (gpa
== UNMAPPED_GVA
)
853 return vcpu_printf(vcpu
, "not present\n"), X86EMUL_PROPAGATE_FAULT
;
854 vcpu
->mmio_needed
= 1;
855 vcpu
->mmio_phys_addr
= gpa
;
856 vcpu
->mmio_size
= bytes
;
857 vcpu
->mmio_is_write
= 0;
859 return X86EMUL_UNHANDLEABLE
;
863 static int emulator_write_emulated(unsigned long addr
,
866 struct x86_emulate_ctxt
*ctxt
)
868 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
869 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
871 if (gpa
== UNMAPPED_GVA
)
872 return X86EMUL_PROPAGATE_FAULT
;
874 vcpu
->mmio_needed
= 1;
875 vcpu
->mmio_phys_addr
= gpa
;
876 vcpu
->mmio_size
= bytes
;
877 vcpu
->mmio_is_write
= 1;
878 memcpy(vcpu
->mmio_data
, &val
, bytes
);
880 return X86EMUL_CONTINUE
;
883 static int emulator_cmpxchg_emulated(unsigned long addr
,
887 struct x86_emulate_ctxt
*ctxt
)
893 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
895 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
898 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
900 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
903 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
905 spin_lock(&vcpu
->kvm
->lock
);
906 vcpu
->mmu
.inval_page(vcpu
, address
);
907 spin_unlock(&vcpu
->kvm
->lock
);
908 kvm_arch_ops
->invlpg(vcpu
, address
);
909 return X86EMUL_CONTINUE
;
912 int emulate_clts(struct kvm_vcpu
*vcpu
)
914 unsigned long cr0
= vcpu
->cr0
;
917 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
918 return X86EMUL_CONTINUE
;
921 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
923 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
927 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
928 return X86EMUL_CONTINUE
;
930 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
932 return X86EMUL_UNHANDLEABLE
;
936 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
938 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
941 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
943 /* FIXME: better handling */
944 return X86EMUL_UNHANDLEABLE
;
946 return X86EMUL_CONTINUE
;
949 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
953 unsigned long rip
= ctxt
->vcpu
->rip
;
954 unsigned long rip_linear
;
956 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
961 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
963 printk(KERN_ERR
"emulation failed but !mmio_needed?"
964 " rip %lx %02x %02x %02x %02x\n",
965 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
969 struct x86_emulate_ops emulate_ops
= {
970 .read_std
= emulator_read_std
,
971 .write_std
= emulator_write_std
,
972 .read_emulated
= emulator_read_emulated
,
973 .write_emulated
= emulator_write_emulated
,
974 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
977 int emulate_instruction(struct kvm_vcpu
*vcpu
,
982 struct x86_emulate_ctxt emulate_ctxt
;
986 kvm_arch_ops
->cache_regs(vcpu
);
988 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
990 emulate_ctxt
.vcpu
= vcpu
;
991 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
992 emulate_ctxt
.cr2
= cr2
;
993 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
994 ? X86EMUL_MODE_REAL
: cs_l
995 ? X86EMUL_MODE_PROT64
: cs_db
996 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
998 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
999 emulate_ctxt
.cs_base
= 0;
1000 emulate_ctxt
.ds_base
= 0;
1001 emulate_ctxt
.es_base
= 0;
1002 emulate_ctxt
.ss_base
= 0;
1004 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1005 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1006 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1007 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1010 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1011 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1013 vcpu
->mmio_is_write
= 0;
1014 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1016 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1017 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1018 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1019 run
->mmio
.len
= vcpu
->mmio_size
;
1020 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1024 if (!vcpu
->mmio_needed
) {
1025 report_emulation_failure(&emulate_ctxt
);
1026 return EMULATE_FAIL
;
1028 return EMULATE_DO_MMIO
;
1031 kvm_arch_ops
->decache_regs(vcpu
);
1032 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1034 if (vcpu
->mmio_is_write
)
1035 return EMULATE_DO_MMIO
;
1037 return EMULATE_DONE
;
1039 EXPORT_SYMBOL_GPL(emulate_instruction
);
1041 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1043 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1046 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1048 struct descriptor_table dt
= { limit
, base
};
1050 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1053 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1055 struct descriptor_table dt
= { limit
, base
};
1057 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1060 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1061 unsigned long *rflags
)
1064 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1067 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1079 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1084 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1085 unsigned long *rflags
)
1089 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1090 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1099 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1102 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1107 * Reads an msr value (of 'msr_index') into 'pdata'.
1108 * Returns 0 on success, non-0 otherwise.
1109 * Assumes vcpu_load() was already called.
1111 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1113 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1116 #ifdef CONFIG_X86_64
1118 void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1120 if (efer
& EFER_RESERVED_BITS
) {
1121 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1128 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1129 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1134 kvm_arch_ops
->set_efer(vcpu
, efer
);
1137 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1139 vcpu
->shadow_efer
= efer
;
1141 EXPORT_SYMBOL_GPL(set_efer
);
1146 * Writes msr value into into the appropriate "register".
1147 * Returns 0 on success, non-0 otherwise.
1148 * Assumes vcpu_load() was already called.
1150 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1152 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1155 void kvm_resched(struct kvm_vcpu
*vcpu
)
1159 /* Cannot fail - no vcpu unplug yet. */
1160 vcpu_load(vcpu
->kvm
, vcpu_slot(vcpu
));
1162 EXPORT_SYMBOL_GPL(kvm_resched
);
1164 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1168 for (i
= 0; i
< n
; ++i
)
1169 wrmsrl(e
[i
].index
, e
[i
].data
);
1171 EXPORT_SYMBOL_GPL(load_msrs
);
1173 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1177 for (i
= 0; i
< n
; ++i
)
1178 rdmsrl(e
[i
].index
, e
[i
].data
);
1180 EXPORT_SYMBOL_GPL(save_msrs
);
1182 static int kvm_dev_ioctl_run(struct kvm
*kvm
, struct kvm_run
*kvm_run
)
1184 struct kvm_vcpu
*vcpu
;
1187 if (!valid_vcpu(kvm_run
->vcpu
))
1190 vcpu
= vcpu_load(kvm
, kvm_run
->vcpu
);
1194 if (kvm_run
->emulated
) {
1195 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1196 kvm_run
->emulated
= 0;
1199 if (kvm_run
->mmio_completed
) {
1200 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1201 vcpu
->mmio_read_completed
= 1;
1204 vcpu
->mmio_needed
= 0;
1206 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1212 static int kvm_dev_ioctl_get_regs(struct kvm
*kvm
, struct kvm_regs
*regs
)
1214 struct kvm_vcpu
*vcpu
;
1216 if (!valid_vcpu(regs
->vcpu
))
1219 vcpu
= vcpu_load(kvm
, regs
->vcpu
);
1223 kvm_arch_ops
->cache_regs(vcpu
);
1225 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1226 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1227 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1228 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1229 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1230 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1231 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1232 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1233 #ifdef CONFIG_X86_64
1234 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1235 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1236 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1237 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1238 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1239 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1240 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1241 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1244 regs
->rip
= vcpu
->rip
;
1245 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1248 * Don't leak debug flags in case they were set for guest debugging
1250 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1251 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1258 static int kvm_dev_ioctl_set_regs(struct kvm
*kvm
, struct kvm_regs
*regs
)
1260 struct kvm_vcpu
*vcpu
;
1262 if (!valid_vcpu(regs
->vcpu
))
1265 vcpu
= vcpu_load(kvm
, regs
->vcpu
);
1269 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1270 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1271 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1272 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1273 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1274 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1275 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1276 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1277 #ifdef CONFIG_X86_64
1278 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1279 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1280 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1281 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1282 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1283 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1284 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1285 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1288 vcpu
->rip
= regs
->rip
;
1289 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1291 kvm_arch_ops
->decache_regs(vcpu
);
1298 static void get_segment(struct kvm_vcpu
*vcpu
,
1299 struct kvm_segment
*var
, int seg
)
1301 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1304 static int kvm_dev_ioctl_get_sregs(struct kvm
*kvm
, struct kvm_sregs
*sregs
)
1306 struct kvm_vcpu
*vcpu
;
1307 struct descriptor_table dt
;
1309 if (!valid_vcpu(sregs
->vcpu
))
1311 vcpu
= vcpu_load(kvm
, sregs
->vcpu
);
1315 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1316 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1317 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1318 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1319 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1320 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1322 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1323 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1325 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1326 sregs
->idt
.limit
= dt
.limit
;
1327 sregs
->idt
.base
= dt
.base
;
1328 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1329 sregs
->gdt
.limit
= dt
.limit
;
1330 sregs
->gdt
.base
= dt
.base
;
1332 sregs
->cr0
= vcpu
->cr0
;
1333 sregs
->cr2
= vcpu
->cr2
;
1334 sregs
->cr3
= vcpu
->cr3
;
1335 sregs
->cr4
= vcpu
->cr4
;
1336 sregs
->cr8
= vcpu
->cr8
;
1337 sregs
->efer
= vcpu
->shadow_efer
;
1338 sregs
->apic_base
= vcpu
->apic_base
;
1340 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1341 sizeof sregs
->interrupt_bitmap
);
1348 static void set_segment(struct kvm_vcpu
*vcpu
,
1349 struct kvm_segment
*var
, int seg
)
1351 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1354 static int kvm_dev_ioctl_set_sregs(struct kvm
*kvm
, struct kvm_sregs
*sregs
)
1356 struct kvm_vcpu
*vcpu
;
1357 int mmu_reset_needed
= 0;
1359 struct descriptor_table dt
;
1361 if (!valid_vcpu(sregs
->vcpu
))
1363 vcpu
= vcpu_load(kvm
, sregs
->vcpu
);
1367 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1368 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1369 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1370 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1371 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1372 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1374 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1375 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1377 dt
.limit
= sregs
->idt
.limit
;
1378 dt
.base
= sregs
->idt
.base
;
1379 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1380 dt
.limit
= sregs
->gdt
.limit
;
1381 dt
.base
= sregs
->gdt
.base
;
1382 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1384 vcpu
->cr2
= sregs
->cr2
;
1385 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
1386 vcpu
->cr3
= sregs
->cr3
;
1388 vcpu
->cr8
= sregs
->cr8
;
1390 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
1391 #ifdef CONFIG_X86_64
1392 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
1394 vcpu
->apic_base
= sregs
->apic_base
;
1396 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
1397 kvm_arch_ops
->set_cr0_no_modeswitch(vcpu
, sregs
->cr0
);
1399 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
1400 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
1402 if (mmu_reset_needed
)
1403 kvm_mmu_reset_context(vcpu
);
1405 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
1406 sizeof vcpu
->irq_pending
);
1407 vcpu
->irq_summary
= 0;
1408 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
1409 if (vcpu
->irq_pending
[i
])
1410 __set_bit(i
, &vcpu
->irq_summary
);
1418 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1419 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1421 static u32 msrs_to_save
[] = {
1422 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1424 #ifdef CONFIG_X86_64
1425 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1427 MSR_IA32_TIME_STAMP_COUNTER
,
1432 * Adapt set_msr() to msr_io()'s calling convention
1434 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1436 return set_msr(vcpu
, index
, *data
);
1440 * Read or write a bunch of msrs. All parameters are kernel addresses.
1442 * @return number of msrs set successfully.
1444 static int __msr_io(struct kvm
*kvm
, struct kvm_msrs
*msrs
,
1445 struct kvm_msr_entry
*entries
,
1446 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1447 unsigned index
, u64
*data
))
1449 struct kvm_vcpu
*vcpu
;
1452 if (!valid_vcpu(msrs
->vcpu
))
1455 vcpu
= vcpu_load(kvm
, msrs
->vcpu
);
1459 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1460 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1469 * Read or write a bunch of msrs. Parameters are user addresses.
1471 * @return number of msrs set successfully.
1473 static int msr_io(struct kvm
*kvm
, struct kvm_msrs __user
*user_msrs
,
1474 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1475 unsigned index
, u64
*data
),
1478 struct kvm_msrs msrs
;
1479 struct kvm_msr_entry
*entries
;
1484 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1488 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1492 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1493 entries
= vmalloc(size
);
1498 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1501 r
= n
= __msr_io(kvm
, &msrs
, entries
, do_msr
);
1506 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1518 * Translate a guest virtual address to a guest physical address.
1520 static int kvm_dev_ioctl_translate(struct kvm
*kvm
, struct kvm_translation
*tr
)
1522 unsigned long vaddr
= tr
->linear_address
;
1523 struct kvm_vcpu
*vcpu
;
1526 vcpu
= vcpu_load(kvm
, tr
->vcpu
);
1529 spin_lock(&kvm
->lock
);
1530 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
1531 tr
->physical_address
= gpa
;
1532 tr
->valid
= gpa
!= UNMAPPED_GVA
;
1535 spin_unlock(&kvm
->lock
);
1541 static int kvm_dev_ioctl_interrupt(struct kvm
*kvm
, struct kvm_interrupt
*irq
)
1543 struct kvm_vcpu
*vcpu
;
1545 if (!valid_vcpu(irq
->vcpu
))
1547 if (irq
->irq
< 0 || irq
->irq
>= 256)
1549 vcpu
= vcpu_load(kvm
, irq
->vcpu
);
1553 set_bit(irq
->irq
, vcpu
->irq_pending
);
1554 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
1561 static int kvm_dev_ioctl_debug_guest(struct kvm
*kvm
,
1562 struct kvm_debug_guest
*dbg
)
1564 struct kvm_vcpu
*vcpu
;
1567 if (!valid_vcpu(dbg
->vcpu
))
1569 vcpu
= vcpu_load(kvm
, dbg
->vcpu
);
1573 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
1580 static long kvm_dev_ioctl(struct file
*filp
,
1581 unsigned int ioctl
, unsigned long arg
)
1583 struct kvm
*kvm
= filp
->private_data
;
1587 case KVM_CREATE_VCPU
: {
1588 r
= kvm_dev_ioctl_create_vcpu(kvm
, arg
);
1594 struct kvm_run kvm_run
;
1597 if (copy_from_user(&kvm_run
, (void *)arg
, sizeof kvm_run
))
1599 r
= kvm_dev_ioctl_run(kvm
, &kvm_run
);
1603 if (copy_to_user((void *)arg
, &kvm_run
, sizeof kvm_run
))
1608 case KVM_GET_REGS
: {
1609 struct kvm_regs kvm_regs
;
1612 if (copy_from_user(&kvm_regs
, (void *)arg
, sizeof kvm_regs
))
1614 r
= kvm_dev_ioctl_get_regs(kvm
, &kvm_regs
);
1618 if (copy_to_user((void *)arg
, &kvm_regs
, sizeof kvm_regs
))
1623 case KVM_SET_REGS
: {
1624 struct kvm_regs kvm_regs
;
1627 if (copy_from_user(&kvm_regs
, (void *)arg
, sizeof kvm_regs
))
1629 r
= kvm_dev_ioctl_set_regs(kvm
, &kvm_regs
);
1635 case KVM_GET_SREGS
: {
1636 struct kvm_sregs kvm_sregs
;
1639 if (copy_from_user(&kvm_sregs
, (void *)arg
, sizeof kvm_sregs
))
1641 r
= kvm_dev_ioctl_get_sregs(kvm
, &kvm_sregs
);
1645 if (copy_to_user((void *)arg
, &kvm_sregs
, sizeof kvm_sregs
))
1650 case KVM_SET_SREGS
: {
1651 struct kvm_sregs kvm_sregs
;
1654 if (copy_from_user(&kvm_sregs
, (void *)arg
, sizeof kvm_sregs
))
1656 r
= kvm_dev_ioctl_set_sregs(kvm
, &kvm_sregs
);
1662 case KVM_TRANSLATE
: {
1663 struct kvm_translation tr
;
1666 if (copy_from_user(&tr
, (void *)arg
, sizeof tr
))
1668 r
= kvm_dev_ioctl_translate(kvm
, &tr
);
1672 if (copy_to_user((void *)arg
, &tr
, sizeof tr
))
1677 case KVM_INTERRUPT
: {
1678 struct kvm_interrupt irq
;
1681 if (copy_from_user(&irq
, (void *)arg
, sizeof irq
))
1683 r
= kvm_dev_ioctl_interrupt(kvm
, &irq
);
1689 case KVM_DEBUG_GUEST
: {
1690 struct kvm_debug_guest dbg
;
1693 if (copy_from_user(&dbg
, (void *)arg
, sizeof dbg
))
1695 r
= kvm_dev_ioctl_debug_guest(kvm
, &dbg
);
1701 case KVM_SET_MEMORY_REGION
: {
1702 struct kvm_memory_region kvm_mem
;
1705 if (copy_from_user(&kvm_mem
, (void *)arg
, sizeof kvm_mem
))
1707 r
= kvm_dev_ioctl_set_memory_region(kvm
, &kvm_mem
);
1712 case KVM_GET_DIRTY_LOG
: {
1713 struct kvm_dirty_log log
;
1716 if (copy_from_user(&log
, (void *)arg
, sizeof log
))
1718 r
= kvm_dev_ioctl_get_dirty_log(kvm
, &log
);
1724 r
= msr_io(kvm
, (void __user
*)arg
, get_msr
, 1);
1727 r
= msr_io(kvm
, (void __user
*)arg
, do_set_msr
, 0);
1729 case KVM_GET_MSR_INDEX_LIST
: {
1730 struct kvm_msr_list __user
*user_msr_list
= (void __user
*)arg
;
1731 struct kvm_msr_list msr_list
;
1735 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1738 msr_list
.nmsrs
= ARRAY_SIZE(msrs_to_save
);
1739 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1742 if (n
< ARRAY_SIZE(msrs_to_save
))
1745 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1746 sizeof msrs_to_save
))
1757 static struct page
*kvm_dev_nopage(struct vm_area_struct
*vma
,
1758 unsigned long address
,
1761 struct kvm
*kvm
= vma
->vm_file
->private_data
;
1762 unsigned long pgoff
;
1763 struct kvm_memory_slot
*slot
;
1766 *type
= VM_FAULT_MINOR
;
1767 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1768 slot
= gfn_to_memslot(kvm
, pgoff
);
1770 return NOPAGE_SIGBUS
;
1771 page
= gfn_to_page(slot
, pgoff
);
1773 return NOPAGE_SIGBUS
;
1778 static struct vm_operations_struct kvm_dev_vm_ops
= {
1779 .nopage
= kvm_dev_nopage
,
1782 static int kvm_dev_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1784 vma
->vm_ops
= &kvm_dev_vm_ops
;
1788 static struct file_operations kvm_chardev_ops
= {
1789 .open
= kvm_dev_open
,
1790 .release
= kvm_dev_release
,
1791 .unlocked_ioctl
= kvm_dev_ioctl
,
1792 .compat_ioctl
= kvm_dev_ioctl
,
1793 .mmap
= kvm_dev_mmap
,
1796 static struct miscdevice kvm_dev
= {
1802 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
1805 if (val
== SYS_RESTART
) {
1807 * Some (well, at least mine) BIOSes hang on reboot if
1810 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
1811 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
1816 static struct notifier_block kvm_reboot_notifier
= {
1817 .notifier_call
= kvm_reboot
,
1821 static __init
void kvm_init_debug(void)
1823 struct kvm_stats_debugfs_item
*p
;
1825 debugfs_dir
= debugfs_create_dir("kvm", 0);
1826 for (p
= debugfs_entries
; p
->name
; ++p
)
1827 p
->dentry
= debugfs_create_u32(p
->name
, 0444, debugfs_dir
,
1831 static void kvm_exit_debug(void)
1833 struct kvm_stats_debugfs_item
*p
;
1835 for (p
= debugfs_entries
; p
->name
; ++p
)
1836 debugfs_remove(p
->dentry
);
1837 debugfs_remove(debugfs_dir
);
1840 hpa_t bad_page_address
;
1842 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
1848 if (!kvm_arch_ops
->cpu_has_kvm_support()) {
1849 printk(KERN_ERR
"kvm: no hardware support\n");
1852 if (kvm_arch_ops
->disabled_by_bios()) {
1853 printk(KERN_ERR
"kvm: disabled by bios\n");
1857 r
= kvm_arch_ops
->hardware_setup();
1861 on_each_cpu(kvm_arch_ops
->hardware_enable
, 0, 0, 1);
1862 register_reboot_notifier(&kvm_reboot_notifier
);
1864 kvm_chardev_ops
.owner
= module
;
1866 r
= misc_register(&kvm_dev
);
1868 printk (KERN_ERR
"kvm: misc device register failed\n");
1875 unregister_reboot_notifier(&kvm_reboot_notifier
);
1876 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
1877 kvm_arch_ops
->hardware_unsetup();
1881 void kvm_exit_arch(void)
1883 misc_deregister(&kvm_dev
);
1885 unregister_reboot_notifier(&kvm_reboot_notifier
);
1886 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
1887 kvm_arch_ops
->hardware_unsetup();
1890 static __init
int kvm_init(void)
1892 static struct page
*bad_page
;
1897 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
1902 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
1903 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
1912 static __exit
void kvm_exit(void)
1915 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
1918 module_init(kvm_init
)
1919 module_exit(kvm_exit
)
1921 EXPORT_SYMBOL_GPL(kvm_init_arch
);
1922 EXPORT_SYMBOL_GPL(kvm_exit_arch
);