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_window", &kvm_stat
.irq_window_exits
},
62 { "halt_exits", &kvm_stat
.halt_exits
},
63 { "request_irq", &kvm_stat
.request_irq_exits
},
64 { "irq_exits", &kvm_stat
.irq_exits
},
68 static struct dentry
*debugfs_dir
;
70 #define MAX_IO_MSRS 256
72 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
73 #define LMSW_GUEST_MASK 0x0eULL
74 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
75 #define CR8_RESEVED_BITS (~0x0fULL)
76 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
79 // LDT or TSS descriptor in the GDT. 16 bytes.
80 struct segment_descriptor_64
{
81 struct segment_descriptor s
;
88 unsigned long segment_base(u16 selector
)
90 struct descriptor_table gdt
;
91 struct segment_descriptor
*d
;
92 unsigned long table_base
;
93 typedef unsigned long ul
;
99 asm ("sgdt %0" : "=m"(gdt
));
100 table_base
= gdt
.base
;
102 if (selector
& 4) { /* from ldt */
105 asm ("sldt %0" : "=g"(ldt_selector
));
106 table_base
= segment_base(ldt_selector
);
108 d
= (struct segment_descriptor
*)(table_base
+ (selector
& ~7));
109 v
= d
->base_low
| ((ul
)d
->base_mid
<< 16) | ((ul
)d
->base_high
<< 24);
112 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
113 v
|= ((ul
)((struct segment_descriptor_64
*)d
)->base_higher
) << 32;
117 EXPORT_SYMBOL_GPL(segment_base
);
119 static inline int valid_vcpu(int n
)
121 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
124 int kvm_read_guest(struct kvm_vcpu
*vcpu
,
129 unsigned char *host_buf
= dest
;
130 unsigned long req_size
= size
;
138 paddr
= gva_to_hpa(vcpu
, addr
);
140 if (is_error_hpa(paddr
))
143 guest_buf
= (hva_t
)kmap_atomic(
144 pfn_to_page(paddr
>> PAGE_SHIFT
),
146 offset
= addr
& ~PAGE_MASK
;
148 now
= min(size
, PAGE_SIZE
- offset
);
149 memcpy(host_buf
, (void*)guest_buf
, now
);
153 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
155 return req_size
- size
;
157 EXPORT_SYMBOL_GPL(kvm_read_guest
);
159 int kvm_write_guest(struct kvm_vcpu
*vcpu
,
164 unsigned char *host_buf
= data
;
165 unsigned long req_size
= size
;
173 paddr
= gva_to_hpa(vcpu
, addr
);
175 if (is_error_hpa(paddr
))
178 guest_buf
= (hva_t
)kmap_atomic(
179 pfn_to_page(paddr
>> PAGE_SHIFT
), KM_USER0
);
180 offset
= addr
& ~PAGE_MASK
;
182 now
= min(size
, PAGE_SIZE
- offset
);
183 memcpy((void*)guest_buf
, host_buf
, now
);
187 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
189 return req_size
- size
;
191 EXPORT_SYMBOL_GPL(kvm_write_guest
);
193 static int vcpu_slot(struct kvm_vcpu
*vcpu
)
195 return vcpu
- vcpu
->kvm
->vcpus
;
199 * Switches to specified vcpu, until a matching vcpu_put()
201 static struct kvm_vcpu
*vcpu_load(struct kvm
*kvm
, int vcpu_slot
)
203 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[vcpu_slot
];
205 mutex_lock(&vcpu
->mutex
);
206 if (unlikely(!vcpu
->vmcs
)) {
207 mutex_unlock(&vcpu
->mutex
);
210 return kvm_arch_ops
->vcpu_load(vcpu
);
213 static void vcpu_put(struct kvm_vcpu
*vcpu
)
215 kvm_arch_ops
->vcpu_put(vcpu
);
216 mutex_unlock(&vcpu
->mutex
);
219 static int kvm_dev_open(struct inode
*inode
, struct file
*filp
)
221 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
227 spin_lock_init(&kvm
->lock
);
228 INIT_LIST_HEAD(&kvm
->active_mmu_pages
);
229 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
230 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[i
];
232 mutex_init(&vcpu
->mutex
);
233 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
234 INIT_LIST_HEAD(&vcpu
->free_pages
);
236 filp
->private_data
= kvm
;
241 * Free any memory in @free but not in @dont.
243 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
244 struct kvm_memory_slot
*dont
)
248 if (!dont
|| free
->phys_mem
!= dont
->phys_mem
)
249 if (free
->phys_mem
) {
250 for (i
= 0; i
< free
->npages
; ++i
)
251 if (free
->phys_mem
[i
])
252 __free_page(free
->phys_mem
[i
]);
253 vfree(free
->phys_mem
);
256 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
257 vfree(free
->dirty_bitmap
);
261 free
->dirty_bitmap
= 0;
264 static void kvm_free_physmem(struct kvm
*kvm
)
268 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
269 kvm_free_physmem_slot(&kvm
->memslots
[i
], 0);
272 static void kvm_free_vcpu(struct kvm_vcpu
*vcpu
)
274 kvm_arch_ops
->vcpu_free(vcpu
);
275 kvm_mmu_destroy(vcpu
);
278 static void kvm_free_vcpus(struct kvm
*kvm
)
282 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
283 kvm_free_vcpu(&kvm
->vcpus
[i
]);
286 static int kvm_dev_release(struct inode
*inode
, struct file
*filp
)
288 struct kvm
*kvm
= filp
->private_data
;
291 kvm_free_physmem(kvm
);
296 static void inject_gp(struct kvm_vcpu
*vcpu
)
298 kvm_arch_ops
->inject_gp(vcpu
, 0);
302 * Load the pae pdptrs. Return true is they are all valid.
304 static int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
306 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
307 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
312 struct kvm_memory_slot
*memslot
;
314 spin_lock(&vcpu
->kvm
->lock
);
315 memslot
= gfn_to_memslot(vcpu
->kvm
, pdpt_gfn
);
316 /* FIXME: !memslot - emulate? 0xff? */
317 pdpt
= kmap_atomic(gfn_to_page(memslot
, pdpt_gfn
), KM_USER0
);
320 for (i
= 0; i
< 4; ++i
) {
321 pdpte
= pdpt
[offset
+ i
];
322 if ((pdpte
& 1) && (pdpte
& 0xfffffff0000001e6ull
)) {
328 for (i
= 0; i
< 4; ++i
)
329 vcpu
->pdptrs
[i
] = pdpt
[offset
+ i
];
332 kunmap_atomic(pdpt
, KM_USER0
);
333 spin_unlock(&vcpu
->kvm
->lock
);
338 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
340 if (cr0
& CR0_RESEVED_BITS
) {
341 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
347 if ((cr0
& CR0_NW_MASK
) && !(cr0
& CR0_CD_MASK
)) {
348 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
353 if ((cr0
& CR0_PG_MASK
) && !(cr0
& CR0_PE_MASK
)) {
354 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
355 "and a clear PE flag\n");
360 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
362 if ((vcpu
->shadow_efer
& EFER_LME
)) {
366 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
367 "in long mode while PAE is disabled\n");
371 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
373 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
374 "in long mode while CS.L == 1\n");
381 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
382 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
390 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
393 spin_lock(&vcpu
->kvm
->lock
);
394 kvm_mmu_reset_context(vcpu
);
395 spin_unlock(&vcpu
->kvm
->lock
);
398 EXPORT_SYMBOL_GPL(set_cr0
);
400 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
402 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
403 set_cr0(vcpu
, (vcpu
->cr0
& ~0x0ful
) | (msw
& 0x0f));
405 EXPORT_SYMBOL_GPL(lmsw
);
407 void set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
409 if (cr4
& CR4_RESEVED_BITS
) {
410 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
415 if (is_long_mode(vcpu
)) {
416 if (!(cr4
& CR4_PAE_MASK
)) {
417 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
422 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& CR4_PAE_MASK
)
423 && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
424 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
428 if (cr4
& CR4_VMXE_MASK
) {
429 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
433 kvm_arch_ops
->set_cr4(vcpu
, cr4
);
434 spin_lock(&vcpu
->kvm
->lock
);
435 kvm_mmu_reset_context(vcpu
);
436 spin_unlock(&vcpu
->kvm
->lock
);
438 EXPORT_SYMBOL_GPL(set_cr4
);
440 void set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
442 if (is_long_mode(vcpu
)) {
443 if ( cr3
& CR3_L_MODE_RESEVED_BITS
) {
444 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
449 if (cr3
& CR3_RESEVED_BITS
) {
450 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
454 if (is_paging(vcpu
) && is_pae(vcpu
) &&
455 !load_pdptrs(vcpu
, cr3
)) {
456 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
464 spin_lock(&vcpu
->kvm
->lock
);
465 vcpu
->mmu
.new_cr3(vcpu
);
466 spin_unlock(&vcpu
->kvm
->lock
);
468 EXPORT_SYMBOL_GPL(set_cr3
);
470 void set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
472 if ( cr8
& CR8_RESEVED_BITS
) {
473 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
479 EXPORT_SYMBOL_GPL(set_cr8
);
481 void fx_init(struct kvm_vcpu
*vcpu
)
483 struct __attribute__ ((__packed__
)) fx_image_s
{
489 u64 operand
;// fpu dp
495 fx_save(vcpu
->host_fx_image
);
497 fx_save(vcpu
->guest_fx_image
);
498 fx_restore(vcpu
->host_fx_image
);
500 fx_image
= (struct fx_image_s
*)vcpu
->guest_fx_image
;
501 fx_image
->mxcsr
= 0x1f80;
502 memset(vcpu
->guest_fx_image
+ sizeof(struct fx_image_s
),
503 0, FX_IMAGE_SIZE
- sizeof(struct fx_image_s
));
505 EXPORT_SYMBOL_GPL(fx_init
);
508 * Creates some virtual cpus. Good luck creating more than one.
510 static int kvm_dev_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
513 struct kvm_vcpu
*vcpu
;
519 vcpu
= &kvm
->vcpus
[n
];
521 mutex_lock(&vcpu
->mutex
);
524 mutex_unlock(&vcpu
->mutex
);
528 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
530 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
532 vcpu
->cpu
= -1; /* First load will set up TR */
534 r
= kvm_arch_ops
->vcpu_create(vcpu
);
538 r
= kvm_mmu_create(vcpu
);
542 kvm_arch_ops
->vcpu_load(vcpu
);
543 r
= kvm_mmu_setup(vcpu
);
545 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
555 mutex_unlock(&vcpu
->mutex
);
561 * Allocate some memory and give it an address in the guest physical address
564 * Discontiguous memory is allowed, mostly for framebuffers.
566 static int kvm_dev_ioctl_set_memory_region(struct kvm
*kvm
,
567 struct kvm_memory_region
*mem
)
571 unsigned long npages
;
573 struct kvm_memory_slot
*memslot
;
574 struct kvm_memory_slot old
, new;
575 int memory_config_version
;
578 /* General sanity checks */
579 if (mem
->memory_size
& (PAGE_SIZE
- 1))
581 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
583 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
585 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
588 memslot
= &kvm
->memslots
[mem
->slot
];
589 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
590 npages
= mem
->memory_size
>> PAGE_SHIFT
;
593 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
596 spin_lock(&kvm
->lock
);
598 memory_config_version
= kvm
->memory_config_version
;
599 new = old
= *memslot
;
601 new.base_gfn
= base_gfn
;
603 new.flags
= mem
->flags
;
605 /* Disallow changing a memory slot's size. */
607 if (npages
&& old
.npages
&& npages
!= old
.npages
)
610 /* Check for overlaps */
612 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
613 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
617 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
618 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
622 * Do memory allocations outside lock. memory_config_version will
625 spin_unlock(&kvm
->lock
);
627 /* Deallocate if slot is being removed */
631 /* Free page dirty bitmap if unneeded */
632 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
633 new.dirty_bitmap
= 0;
637 /* Allocate if a slot is being created */
638 if (npages
&& !new.phys_mem
) {
639 new.phys_mem
= vmalloc(npages
* sizeof(struct page
*));
644 memset(new.phys_mem
, 0, npages
* sizeof(struct page
*));
645 for (i
= 0; i
< npages
; ++i
) {
646 new.phys_mem
[i
] = alloc_page(GFP_HIGHUSER
648 if (!new.phys_mem
[i
])
650 new.phys_mem
[i
]->private = 0;
654 /* Allocate page dirty bitmap if needed */
655 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
656 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
658 new.dirty_bitmap
= vmalloc(dirty_bytes
);
659 if (!new.dirty_bitmap
)
661 memset(new.dirty_bitmap
, 0, dirty_bytes
);
664 spin_lock(&kvm
->lock
);
666 if (memory_config_version
!= kvm
->memory_config_version
) {
667 spin_unlock(&kvm
->lock
);
668 kvm_free_physmem_slot(&new, &old
);
676 if (mem
->slot
>= kvm
->nmemslots
)
677 kvm
->nmemslots
= mem
->slot
+ 1;
680 ++kvm
->memory_config_version
;
682 spin_unlock(&kvm
->lock
);
684 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
685 struct kvm_vcpu
*vcpu
;
687 vcpu
= vcpu_load(kvm
, i
);
690 kvm_mmu_reset_context(vcpu
);
694 kvm_free_physmem_slot(&old
, &new);
698 spin_unlock(&kvm
->lock
);
700 kvm_free_physmem_slot(&new, &old
);
706 * Get (and clear) the dirty memory log for a memory slot.
708 static int kvm_dev_ioctl_get_dirty_log(struct kvm
*kvm
,
709 struct kvm_dirty_log
*log
)
711 struct kvm_memory_slot
*memslot
;
714 unsigned long any
= 0;
716 spin_lock(&kvm
->lock
);
719 * Prevent changes to guest memory configuration even while the lock
723 spin_unlock(&kvm
->lock
);
725 if (log
->slot
>= KVM_MEMORY_SLOTS
)
728 memslot
= &kvm
->memslots
[log
->slot
];
730 if (!memslot
->dirty_bitmap
)
733 n
= ALIGN(memslot
->npages
, 8) / 8;
735 for (i
= 0; !any
&& i
< n
; ++i
)
736 any
= memslot
->dirty_bitmap
[i
];
739 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
744 spin_lock(&kvm
->lock
);
745 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
746 spin_unlock(&kvm
->lock
);
747 memset(memslot
->dirty_bitmap
, 0, n
);
748 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
749 struct kvm_vcpu
*vcpu
= vcpu_load(kvm
, i
);
753 kvm_arch_ops
->tlb_flush(vcpu
);
761 spin_lock(&kvm
->lock
);
763 spin_unlock(&kvm
->lock
);
767 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
771 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
772 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
774 if (gfn
>= memslot
->base_gfn
775 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
780 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
782 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
785 struct kvm_memory_slot
*memslot
= 0;
786 unsigned long rel_gfn
;
788 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
789 memslot
= &kvm
->memslots
[i
];
791 if (gfn
>= memslot
->base_gfn
792 && gfn
< memslot
->base_gfn
+ memslot
->npages
) {
794 if (!memslot
|| !memslot
->dirty_bitmap
)
797 rel_gfn
= gfn
- memslot
->base_gfn
;
800 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
801 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
807 static int emulator_read_std(unsigned long addr
,
810 struct x86_emulate_ctxt
*ctxt
)
812 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
816 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
817 unsigned offset
= addr
& (PAGE_SIZE
-1);
818 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
820 struct kvm_memory_slot
*memslot
;
823 if (gpa
== UNMAPPED_GVA
)
824 return X86EMUL_PROPAGATE_FAULT
;
825 pfn
= gpa
>> PAGE_SHIFT
;
826 memslot
= gfn_to_memslot(vcpu
->kvm
, pfn
);
828 return X86EMUL_UNHANDLEABLE
;
829 page
= kmap_atomic(gfn_to_page(memslot
, pfn
), KM_USER0
);
831 memcpy(data
, page
+ offset
, tocopy
);
833 kunmap_atomic(page
, KM_USER0
);
840 return X86EMUL_CONTINUE
;
843 static int emulator_write_std(unsigned long addr
,
846 struct x86_emulate_ctxt
*ctxt
)
848 printk(KERN_ERR
"emulator_write_std: addr %lx n %d\n",
850 return X86EMUL_UNHANDLEABLE
;
853 static int emulator_read_emulated(unsigned long addr
,
856 struct x86_emulate_ctxt
*ctxt
)
858 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
860 if (vcpu
->mmio_read_completed
) {
861 memcpy(val
, vcpu
->mmio_data
, bytes
);
862 vcpu
->mmio_read_completed
= 0;
863 return X86EMUL_CONTINUE
;
864 } else if (emulator_read_std(addr
, val
, bytes
, ctxt
)
866 return X86EMUL_CONTINUE
;
868 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
869 if (gpa
== UNMAPPED_GVA
)
870 return vcpu_printf(vcpu
, "not present\n"), X86EMUL_PROPAGATE_FAULT
;
871 vcpu
->mmio_needed
= 1;
872 vcpu
->mmio_phys_addr
= gpa
;
873 vcpu
->mmio_size
= bytes
;
874 vcpu
->mmio_is_write
= 0;
876 return X86EMUL_UNHANDLEABLE
;
880 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
881 unsigned long val
, int bytes
)
883 struct kvm_memory_slot
*m
;
887 if (((gpa
+ bytes
- 1) >> PAGE_SHIFT
) != (gpa
>> PAGE_SHIFT
))
889 m
= gfn_to_memslot(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
892 page
= gfn_to_page(m
, gpa
>> PAGE_SHIFT
);
893 kvm_mmu_pre_write(vcpu
, gpa
, bytes
);
894 virt
= kmap_atomic(page
, KM_USER0
);
895 memcpy(virt
+ offset_in_page(gpa
), &val
, bytes
);
896 kunmap_atomic(virt
, KM_USER0
);
897 kvm_mmu_post_write(vcpu
, gpa
, bytes
);
901 static int emulator_write_emulated(unsigned long addr
,
904 struct x86_emulate_ctxt
*ctxt
)
906 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
907 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
909 if (gpa
== UNMAPPED_GVA
)
910 return X86EMUL_PROPAGATE_FAULT
;
912 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
913 return X86EMUL_CONTINUE
;
915 vcpu
->mmio_needed
= 1;
916 vcpu
->mmio_phys_addr
= gpa
;
917 vcpu
->mmio_size
= bytes
;
918 vcpu
->mmio_is_write
= 1;
919 memcpy(vcpu
->mmio_data
, &val
, bytes
);
921 return X86EMUL_CONTINUE
;
924 static int emulator_cmpxchg_emulated(unsigned long addr
,
928 struct x86_emulate_ctxt
*ctxt
)
934 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
936 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
941 static int emulator_cmpxchg8b_emulated(unsigned long addr
,
942 unsigned long old_lo
,
943 unsigned long old_hi
,
944 unsigned long new_lo
,
945 unsigned long new_hi
,
946 struct x86_emulate_ctxt
*ctxt
)
953 printk(KERN_WARNING
"kvm: emulating exchange8b as write\n");
955 r
= emulator_write_emulated(addr
, new_lo
, 4, ctxt
);
956 if (r
!= X86EMUL_CONTINUE
)
958 return emulator_write_emulated(addr
+4, new_hi
, 4, ctxt
);
963 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
965 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
968 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
970 return X86EMUL_CONTINUE
;
973 int emulate_clts(struct kvm_vcpu
*vcpu
)
977 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
978 cr0
= vcpu
->cr0
& ~CR0_TS_MASK
;
979 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
980 return X86EMUL_CONTINUE
;
983 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
985 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
989 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
990 return X86EMUL_CONTINUE
;
992 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
994 return X86EMUL_UNHANDLEABLE
;
998 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1000 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1003 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1005 /* FIXME: better handling */
1006 return X86EMUL_UNHANDLEABLE
;
1008 return X86EMUL_CONTINUE
;
1011 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1013 static int reported
;
1015 unsigned long rip
= ctxt
->vcpu
->rip
;
1016 unsigned long rip_linear
;
1018 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1023 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1025 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1026 " rip %lx %02x %02x %02x %02x\n",
1027 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1031 struct x86_emulate_ops emulate_ops
= {
1032 .read_std
= emulator_read_std
,
1033 .write_std
= emulator_write_std
,
1034 .read_emulated
= emulator_read_emulated
,
1035 .write_emulated
= emulator_write_emulated
,
1036 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1037 #ifdef CONFIG_X86_32
1038 .cmpxchg8b_emulated
= emulator_cmpxchg8b_emulated
,
1042 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1043 struct kvm_run
*run
,
1047 struct x86_emulate_ctxt emulate_ctxt
;
1051 kvm_arch_ops
->cache_regs(vcpu
);
1053 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1055 emulate_ctxt
.vcpu
= vcpu
;
1056 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
1057 emulate_ctxt
.cr2
= cr2
;
1058 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1059 ? X86EMUL_MODE_REAL
: cs_l
1060 ? X86EMUL_MODE_PROT64
: cs_db
1061 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1063 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1064 emulate_ctxt
.cs_base
= 0;
1065 emulate_ctxt
.ds_base
= 0;
1066 emulate_ctxt
.es_base
= 0;
1067 emulate_ctxt
.ss_base
= 0;
1069 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1070 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1071 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1072 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1075 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1076 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1078 vcpu
->mmio_is_write
= 0;
1079 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1081 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1082 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1083 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1084 run
->mmio
.len
= vcpu
->mmio_size
;
1085 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1089 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1090 return EMULATE_DONE
;
1091 if (!vcpu
->mmio_needed
) {
1092 report_emulation_failure(&emulate_ctxt
);
1093 return EMULATE_FAIL
;
1095 return EMULATE_DO_MMIO
;
1098 kvm_arch_ops
->decache_regs(vcpu
);
1099 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1101 if (vcpu
->mmio_is_write
)
1102 return EMULATE_DO_MMIO
;
1104 return EMULATE_DONE
;
1106 EXPORT_SYMBOL_GPL(emulate_instruction
);
1108 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1110 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1113 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1115 struct descriptor_table dt
= { limit
, base
};
1117 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1120 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1122 struct descriptor_table dt
= { limit
, base
};
1124 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1127 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1128 unsigned long *rflags
)
1131 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1134 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1136 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1147 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1152 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1153 unsigned long *rflags
)
1157 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1158 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1167 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1170 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1174 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1179 case 0xc0010010: /* SYSCFG */
1180 case 0xc0010015: /* HWCR */
1181 case MSR_IA32_PLATFORM_ID
:
1182 case MSR_IA32_P5_MC_ADDR
:
1183 case MSR_IA32_P5_MC_TYPE
:
1184 case MSR_IA32_MC0_CTL
:
1185 case MSR_IA32_MCG_STATUS
:
1186 case MSR_IA32_MCG_CAP
:
1187 case MSR_IA32_MC0_MISC
:
1188 case MSR_IA32_MC0_MISC
+4:
1189 case MSR_IA32_MC0_MISC
+8:
1190 case MSR_IA32_MC0_MISC
+12:
1191 case MSR_IA32_MC0_MISC
+16:
1192 case MSR_IA32_UCODE_REV
:
1193 case MSR_IA32_PERF_STATUS
:
1194 /* MTRR registers */
1196 case 0x200 ... 0x2ff:
1199 case 0xcd: /* fsb frequency */
1202 case MSR_IA32_APICBASE
:
1203 data
= vcpu
->apic_base
;
1205 #ifdef CONFIG_X86_64
1207 data
= vcpu
->shadow_efer
;
1211 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1217 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1220 * Reads an msr value (of 'msr_index') into 'pdata'.
1221 * Returns 0 on success, non-0 otherwise.
1222 * Assumes vcpu_load() was already called.
1224 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1226 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1229 #ifdef CONFIG_X86_64
1231 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1233 if (efer
& EFER_RESERVED_BITS
) {
1234 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1241 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1242 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1247 kvm_arch_ops
->set_efer(vcpu
, efer
);
1250 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1252 vcpu
->shadow_efer
= efer
;
1257 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1260 #ifdef CONFIG_X86_64
1262 set_efer(vcpu
, data
);
1265 case MSR_IA32_MC0_STATUS
:
1266 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1267 __FUNCTION__
, data
);
1269 case MSR_IA32_UCODE_REV
:
1270 case MSR_IA32_UCODE_WRITE
:
1271 case 0x200 ... 0x2ff: /* MTRRs */
1273 case MSR_IA32_APICBASE
:
1274 vcpu
->apic_base
= data
;
1277 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1282 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1285 * Writes msr value into into the appropriate "register".
1286 * Returns 0 on success, non-0 otherwise.
1287 * Assumes vcpu_load() was already called.
1289 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1291 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1294 void kvm_resched(struct kvm_vcpu
*vcpu
)
1298 /* Cannot fail - no vcpu unplug yet. */
1299 vcpu_load(vcpu
->kvm
, vcpu_slot(vcpu
));
1301 EXPORT_SYMBOL_GPL(kvm_resched
);
1303 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1307 for (i
= 0; i
< n
; ++i
)
1308 wrmsrl(e
[i
].index
, e
[i
].data
);
1310 EXPORT_SYMBOL_GPL(load_msrs
);
1312 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1316 for (i
= 0; i
< n
; ++i
)
1317 rdmsrl(e
[i
].index
, e
[i
].data
);
1319 EXPORT_SYMBOL_GPL(save_msrs
);
1321 static int kvm_dev_ioctl_run(struct kvm
*kvm
, struct kvm_run
*kvm_run
)
1323 struct kvm_vcpu
*vcpu
;
1326 if (!valid_vcpu(kvm_run
->vcpu
))
1329 vcpu
= vcpu_load(kvm
, kvm_run
->vcpu
);
1333 if (kvm_run
->emulated
) {
1334 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1335 kvm_run
->emulated
= 0;
1338 if (kvm_run
->mmio_completed
) {
1339 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1340 vcpu
->mmio_read_completed
= 1;
1343 vcpu
->mmio_needed
= 0;
1345 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1351 static int kvm_dev_ioctl_get_regs(struct kvm
*kvm
, struct kvm_regs
*regs
)
1353 struct kvm_vcpu
*vcpu
;
1355 if (!valid_vcpu(regs
->vcpu
))
1358 vcpu
= vcpu_load(kvm
, regs
->vcpu
);
1362 kvm_arch_ops
->cache_regs(vcpu
);
1364 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1365 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1366 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1367 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1368 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1369 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1370 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1371 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1372 #ifdef CONFIG_X86_64
1373 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1374 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1375 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1376 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1377 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1378 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1379 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1380 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1383 regs
->rip
= vcpu
->rip
;
1384 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1387 * Don't leak debug flags in case they were set for guest debugging
1389 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1390 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1397 static int kvm_dev_ioctl_set_regs(struct kvm
*kvm
, struct kvm_regs
*regs
)
1399 struct kvm_vcpu
*vcpu
;
1401 if (!valid_vcpu(regs
->vcpu
))
1404 vcpu
= vcpu_load(kvm
, regs
->vcpu
);
1408 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1409 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1410 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1411 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1412 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1413 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1414 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1415 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1416 #ifdef CONFIG_X86_64
1417 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1418 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1419 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1420 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1421 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1422 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1423 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1424 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1427 vcpu
->rip
= regs
->rip
;
1428 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1430 kvm_arch_ops
->decache_regs(vcpu
);
1437 static void get_segment(struct kvm_vcpu
*vcpu
,
1438 struct kvm_segment
*var
, int seg
)
1440 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1443 static int kvm_dev_ioctl_get_sregs(struct kvm
*kvm
, struct kvm_sregs
*sregs
)
1445 struct kvm_vcpu
*vcpu
;
1446 struct descriptor_table dt
;
1448 if (!valid_vcpu(sregs
->vcpu
))
1450 vcpu
= vcpu_load(kvm
, sregs
->vcpu
);
1454 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1455 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1456 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1457 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1458 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1459 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1461 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1462 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1464 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1465 sregs
->idt
.limit
= dt
.limit
;
1466 sregs
->idt
.base
= dt
.base
;
1467 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1468 sregs
->gdt
.limit
= dt
.limit
;
1469 sregs
->gdt
.base
= dt
.base
;
1471 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1472 sregs
->cr0
= vcpu
->cr0
;
1473 sregs
->cr2
= vcpu
->cr2
;
1474 sregs
->cr3
= vcpu
->cr3
;
1475 sregs
->cr4
= vcpu
->cr4
;
1476 sregs
->cr8
= vcpu
->cr8
;
1477 sregs
->efer
= vcpu
->shadow_efer
;
1478 sregs
->apic_base
= vcpu
->apic_base
;
1480 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1481 sizeof sregs
->interrupt_bitmap
);
1488 static void set_segment(struct kvm_vcpu
*vcpu
,
1489 struct kvm_segment
*var
, int seg
)
1491 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1494 static int kvm_dev_ioctl_set_sregs(struct kvm
*kvm
, struct kvm_sregs
*sregs
)
1496 struct kvm_vcpu
*vcpu
;
1497 int mmu_reset_needed
= 0;
1499 struct descriptor_table dt
;
1501 if (!valid_vcpu(sregs
->vcpu
))
1503 vcpu
= vcpu_load(kvm
, sregs
->vcpu
);
1507 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1508 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1509 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1510 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1511 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1512 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1514 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1515 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1517 dt
.limit
= sregs
->idt
.limit
;
1518 dt
.base
= sregs
->idt
.base
;
1519 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1520 dt
.limit
= sregs
->gdt
.limit
;
1521 dt
.base
= sregs
->gdt
.base
;
1522 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1524 vcpu
->cr2
= sregs
->cr2
;
1525 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
1526 vcpu
->cr3
= sregs
->cr3
;
1528 vcpu
->cr8
= sregs
->cr8
;
1530 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
1531 #ifdef CONFIG_X86_64
1532 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
1534 vcpu
->apic_base
= sregs
->apic_base
;
1536 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1538 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
1539 kvm_arch_ops
->set_cr0_no_modeswitch(vcpu
, sregs
->cr0
);
1541 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
1542 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
1543 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
1544 load_pdptrs(vcpu
, vcpu
->cr3
);
1546 if (mmu_reset_needed
)
1547 kvm_mmu_reset_context(vcpu
);
1549 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
1550 sizeof vcpu
->irq_pending
);
1551 vcpu
->irq_summary
= 0;
1552 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
1553 if (vcpu
->irq_pending
[i
])
1554 __set_bit(i
, &vcpu
->irq_summary
);
1562 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1563 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1565 * This list is modified at module load time to reflect the
1566 * capabilities of the host cpu.
1568 static u32 msrs_to_save
[] = {
1569 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1571 #ifdef CONFIG_X86_64
1572 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1574 MSR_IA32_TIME_STAMP_COUNTER
,
1577 static unsigned num_msrs_to_save
;
1579 static __init
void kvm_init_msr_list(void)
1584 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
1585 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
1588 msrs_to_save
[j
] = msrs_to_save
[i
];
1591 num_msrs_to_save
= j
;
1595 * Adapt set_msr() to msr_io()'s calling convention
1597 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1599 return set_msr(vcpu
, index
, *data
);
1603 * Read or write a bunch of msrs. All parameters are kernel addresses.
1605 * @return number of msrs set successfully.
1607 static int __msr_io(struct kvm
*kvm
, struct kvm_msrs
*msrs
,
1608 struct kvm_msr_entry
*entries
,
1609 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1610 unsigned index
, u64
*data
))
1612 struct kvm_vcpu
*vcpu
;
1615 if (!valid_vcpu(msrs
->vcpu
))
1618 vcpu
= vcpu_load(kvm
, msrs
->vcpu
);
1622 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1623 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1632 * Read or write a bunch of msrs. Parameters are user addresses.
1634 * @return number of msrs set successfully.
1636 static int msr_io(struct kvm
*kvm
, struct kvm_msrs __user
*user_msrs
,
1637 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1638 unsigned index
, u64
*data
),
1641 struct kvm_msrs msrs
;
1642 struct kvm_msr_entry
*entries
;
1647 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1651 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1655 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1656 entries
= vmalloc(size
);
1661 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1664 r
= n
= __msr_io(kvm
, &msrs
, entries
, do_msr
);
1669 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1681 * Translate a guest virtual address to a guest physical address.
1683 static int kvm_dev_ioctl_translate(struct kvm
*kvm
, struct kvm_translation
*tr
)
1685 unsigned long vaddr
= tr
->linear_address
;
1686 struct kvm_vcpu
*vcpu
;
1689 vcpu
= vcpu_load(kvm
, tr
->vcpu
);
1692 spin_lock(&kvm
->lock
);
1693 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
1694 tr
->physical_address
= gpa
;
1695 tr
->valid
= gpa
!= UNMAPPED_GVA
;
1698 spin_unlock(&kvm
->lock
);
1704 static int kvm_dev_ioctl_interrupt(struct kvm
*kvm
, struct kvm_interrupt
*irq
)
1706 struct kvm_vcpu
*vcpu
;
1708 if (!valid_vcpu(irq
->vcpu
))
1710 if (irq
->irq
< 0 || irq
->irq
>= 256)
1712 vcpu
= vcpu_load(kvm
, irq
->vcpu
);
1716 set_bit(irq
->irq
, vcpu
->irq_pending
);
1717 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
1724 static int kvm_dev_ioctl_debug_guest(struct kvm
*kvm
,
1725 struct kvm_debug_guest
*dbg
)
1727 struct kvm_vcpu
*vcpu
;
1730 if (!valid_vcpu(dbg
->vcpu
))
1732 vcpu
= vcpu_load(kvm
, dbg
->vcpu
);
1736 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
1743 static long kvm_dev_ioctl(struct file
*filp
,
1744 unsigned int ioctl
, unsigned long arg
)
1746 struct kvm
*kvm
= filp
->private_data
;
1750 case KVM_GET_API_VERSION
:
1751 r
= KVM_API_VERSION
;
1753 case KVM_CREATE_VCPU
: {
1754 r
= kvm_dev_ioctl_create_vcpu(kvm
, arg
);
1760 struct kvm_run kvm_run
;
1763 if (copy_from_user(&kvm_run
, (void *)arg
, sizeof kvm_run
))
1765 r
= kvm_dev_ioctl_run(kvm
, &kvm_run
);
1766 if (r
< 0 && r
!= -EINTR
)
1768 if (copy_to_user((void *)arg
, &kvm_run
, sizeof kvm_run
)) {
1774 case KVM_GET_REGS
: {
1775 struct kvm_regs kvm_regs
;
1778 if (copy_from_user(&kvm_regs
, (void *)arg
, sizeof kvm_regs
))
1780 r
= kvm_dev_ioctl_get_regs(kvm
, &kvm_regs
);
1784 if (copy_to_user((void *)arg
, &kvm_regs
, sizeof kvm_regs
))
1789 case KVM_SET_REGS
: {
1790 struct kvm_regs kvm_regs
;
1793 if (copy_from_user(&kvm_regs
, (void *)arg
, sizeof kvm_regs
))
1795 r
= kvm_dev_ioctl_set_regs(kvm
, &kvm_regs
);
1801 case KVM_GET_SREGS
: {
1802 struct kvm_sregs kvm_sregs
;
1805 if (copy_from_user(&kvm_sregs
, (void *)arg
, sizeof kvm_sregs
))
1807 r
= kvm_dev_ioctl_get_sregs(kvm
, &kvm_sregs
);
1811 if (copy_to_user((void *)arg
, &kvm_sregs
, sizeof kvm_sregs
))
1816 case KVM_SET_SREGS
: {
1817 struct kvm_sregs kvm_sregs
;
1820 if (copy_from_user(&kvm_sregs
, (void *)arg
, sizeof kvm_sregs
))
1822 r
= kvm_dev_ioctl_set_sregs(kvm
, &kvm_sregs
);
1828 case KVM_TRANSLATE
: {
1829 struct kvm_translation tr
;
1832 if (copy_from_user(&tr
, (void *)arg
, sizeof tr
))
1834 r
= kvm_dev_ioctl_translate(kvm
, &tr
);
1838 if (copy_to_user((void *)arg
, &tr
, sizeof tr
))
1843 case KVM_INTERRUPT
: {
1844 struct kvm_interrupt irq
;
1847 if (copy_from_user(&irq
, (void *)arg
, sizeof irq
))
1849 r
= kvm_dev_ioctl_interrupt(kvm
, &irq
);
1855 case KVM_DEBUG_GUEST
: {
1856 struct kvm_debug_guest dbg
;
1859 if (copy_from_user(&dbg
, (void *)arg
, sizeof dbg
))
1861 r
= kvm_dev_ioctl_debug_guest(kvm
, &dbg
);
1867 case KVM_SET_MEMORY_REGION
: {
1868 struct kvm_memory_region kvm_mem
;
1871 if (copy_from_user(&kvm_mem
, (void *)arg
, sizeof kvm_mem
))
1873 r
= kvm_dev_ioctl_set_memory_region(kvm
, &kvm_mem
);
1878 case KVM_GET_DIRTY_LOG
: {
1879 struct kvm_dirty_log log
;
1882 if (copy_from_user(&log
, (void *)arg
, sizeof log
))
1884 r
= kvm_dev_ioctl_get_dirty_log(kvm
, &log
);
1890 r
= msr_io(kvm
, (void __user
*)arg
, get_msr
, 1);
1893 r
= msr_io(kvm
, (void __user
*)arg
, do_set_msr
, 0);
1895 case KVM_GET_MSR_INDEX_LIST
: {
1896 struct kvm_msr_list __user
*user_msr_list
= (void __user
*)arg
;
1897 struct kvm_msr_list msr_list
;
1901 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1904 msr_list
.nmsrs
= num_msrs_to_save
;
1905 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1908 if (n
< num_msrs_to_save
)
1911 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1912 num_msrs_to_save
* sizeof(u32
)))
1923 static struct page
*kvm_dev_nopage(struct vm_area_struct
*vma
,
1924 unsigned long address
,
1927 struct kvm
*kvm
= vma
->vm_file
->private_data
;
1928 unsigned long pgoff
;
1929 struct kvm_memory_slot
*slot
;
1932 *type
= VM_FAULT_MINOR
;
1933 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1934 slot
= gfn_to_memslot(kvm
, pgoff
);
1936 return NOPAGE_SIGBUS
;
1937 page
= gfn_to_page(slot
, pgoff
);
1939 return NOPAGE_SIGBUS
;
1944 static struct vm_operations_struct kvm_dev_vm_ops
= {
1945 .nopage
= kvm_dev_nopage
,
1948 static int kvm_dev_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1950 vma
->vm_ops
= &kvm_dev_vm_ops
;
1954 static struct file_operations kvm_chardev_ops
= {
1955 .open
= kvm_dev_open
,
1956 .release
= kvm_dev_release
,
1957 .unlocked_ioctl
= kvm_dev_ioctl
,
1958 .compat_ioctl
= kvm_dev_ioctl
,
1959 .mmap
= kvm_dev_mmap
,
1962 static struct miscdevice kvm_dev
= {
1968 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
1971 if (val
== SYS_RESTART
) {
1973 * Some (well, at least mine) BIOSes hang on reboot if
1976 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
1977 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
1982 static struct notifier_block kvm_reboot_notifier
= {
1983 .notifier_call
= kvm_reboot
,
1987 static __init
void kvm_init_debug(void)
1989 struct kvm_stats_debugfs_item
*p
;
1991 debugfs_dir
= debugfs_create_dir("kvm", 0);
1992 for (p
= debugfs_entries
; p
->name
; ++p
)
1993 p
->dentry
= debugfs_create_u32(p
->name
, 0444, debugfs_dir
,
1997 static void kvm_exit_debug(void)
1999 struct kvm_stats_debugfs_item
*p
;
2001 for (p
= debugfs_entries
; p
->name
; ++p
)
2002 debugfs_remove(p
->dentry
);
2003 debugfs_remove(debugfs_dir
);
2006 hpa_t bad_page_address
;
2008 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
2013 printk(KERN_ERR
"kvm: already loaded the other module\n");
2017 if (!ops
->cpu_has_kvm_support()) {
2018 printk(KERN_ERR
"kvm: no hardware support\n");
2021 if (ops
->disabled_by_bios()) {
2022 printk(KERN_ERR
"kvm: disabled by bios\n");
2028 r
= kvm_arch_ops
->hardware_setup();
2032 on_each_cpu(kvm_arch_ops
->hardware_enable
, 0, 0, 1);
2033 register_reboot_notifier(&kvm_reboot_notifier
);
2035 kvm_chardev_ops
.owner
= module
;
2037 r
= misc_register(&kvm_dev
);
2039 printk (KERN_ERR
"kvm: misc device register failed\n");
2046 unregister_reboot_notifier(&kvm_reboot_notifier
);
2047 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
2048 kvm_arch_ops
->hardware_unsetup();
2052 void kvm_exit_arch(void)
2054 misc_deregister(&kvm_dev
);
2056 unregister_reboot_notifier(&kvm_reboot_notifier
);
2057 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
2058 kvm_arch_ops
->hardware_unsetup();
2059 kvm_arch_ops
= NULL
;
2062 static __init
int kvm_init(void)
2064 static struct page
*bad_page
;
2069 kvm_init_msr_list();
2071 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
2076 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
2077 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
2086 static __exit
void kvm_exit(void)
2089 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
2092 module_init(kvm_init
)
2093 module_exit(kvm_exit
)
2095 EXPORT_SYMBOL_GPL(kvm_init_arch
);
2096 EXPORT_SYMBOL_GPL(kvm_exit_arch
);