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 <linux/magic.h>
24 #include <asm/processor.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
29 #include <linux/miscdevice.h>
30 #include <linux/vmalloc.h>
31 #include <asm/uaccess.h>
32 #include <linux/reboot.h>
34 #include <linux/debugfs.h>
35 #include <linux/highmem.h>
36 #include <linux/file.h>
38 #include <linux/sysdev.h>
39 #include <linux/cpu.h>
40 #include <linux/file.h>
42 #include <linux/mount.h>
44 #include "x86_emulate.h"
45 #include "segment_descriptor.h"
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static DEFINE_SPINLOCK(kvm_lock
);
51 static LIST_HEAD(vm_list
);
53 struct kvm_arch_ops
*kvm_arch_ops
;
54 struct kvm_stat kvm_stat
;
55 EXPORT_SYMBOL_GPL(kvm_stat
);
57 static struct kvm_stats_debugfs_item
{
60 struct dentry
*dentry
;
61 } debugfs_entries
[] = {
62 { "pf_fixed", &kvm_stat
.pf_fixed
},
63 { "pf_guest", &kvm_stat
.pf_guest
},
64 { "tlb_flush", &kvm_stat
.tlb_flush
},
65 { "invlpg", &kvm_stat
.invlpg
},
66 { "exits", &kvm_stat
.exits
},
67 { "io_exits", &kvm_stat
.io_exits
},
68 { "mmio_exits", &kvm_stat
.mmio_exits
},
69 { "signal_exits", &kvm_stat
.signal_exits
},
70 { "irq_window", &kvm_stat
.irq_window_exits
},
71 { "halt_exits", &kvm_stat
.halt_exits
},
72 { "request_irq", &kvm_stat
.request_irq_exits
},
73 { "irq_exits", &kvm_stat
.irq_exits
},
77 static struct dentry
*debugfs_dir
;
79 struct vfsmount
*kvmfs_mnt
;
81 #define MAX_IO_MSRS 256
83 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
84 #define LMSW_GUEST_MASK 0x0eULL
85 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
86 #define CR8_RESEVED_BITS (~0x0fULL)
87 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
90 // LDT or TSS descriptor in the GDT. 16 bytes.
91 struct segment_descriptor_64
{
92 struct segment_descriptor s
;
99 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
102 static struct inode
*kvmfs_inode(struct file_operations
*fops
)
105 struct inode
*inode
= new_inode(kvmfs_mnt
->mnt_sb
);
113 * Mark the inode dirty from the very beginning,
114 * that way it will never be moved to the dirty
115 * list because mark_inode_dirty() will think
116 * that it already _is_ on the dirty list.
118 inode
->i_state
= I_DIRTY
;
119 inode
->i_mode
= S_IRUSR
| S_IWUSR
;
120 inode
->i_uid
= current
->fsuid
;
121 inode
->i_gid
= current
->fsgid
;
122 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
126 return ERR_PTR(error
);
129 static struct file
*kvmfs_file(struct inode
*inode
, void *private_data
)
131 struct file
*file
= get_empty_filp();
134 return ERR_PTR(-ENFILE
);
136 file
->f_path
.mnt
= mntget(kvmfs_mnt
);
137 file
->f_path
.dentry
= d_alloc_anon(inode
);
138 if (!file
->f_path
.dentry
)
139 return ERR_PTR(-ENOMEM
);
140 file
->f_mapping
= inode
->i_mapping
;
143 file
->f_flags
= O_RDWR
;
144 file
->f_op
= inode
->i_fop
;
145 file
->f_mode
= FMODE_READ
| FMODE_WRITE
;
147 file
->private_data
= private_data
;
151 unsigned long segment_base(u16 selector
)
153 struct descriptor_table gdt
;
154 struct segment_descriptor
*d
;
155 unsigned long table_base
;
156 typedef unsigned long ul
;
162 asm ("sgdt %0" : "=m"(gdt
));
163 table_base
= gdt
.base
;
165 if (selector
& 4) { /* from ldt */
168 asm ("sldt %0" : "=g"(ldt_selector
));
169 table_base
= segment_base(ldt_selector
);
171 d
= (struct segment_descriptor
*)(table_base
+ (selector
& ~7));
172 v
= d
->base_low
| ((ul
)d
->base_mid
<< 16) | ((ul
)d
->base_high
<< 24);
175 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
176 v
|= ((ul
)((struct segment_descriptor_64
*)d
)->base_higher
) << 32;
180 EXPORT_SYMBOL_GPL(segment_base
);
182 static inline int valid_vcpu(int n
)
184 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
187 int kvm_read_guest(struct kvm_vcpu
*vcpu
, gva_t addr
, unsigned long size
,
190 unsigned char *host_buf
= dest
;
191 unsigned long req_size
= size
;
199 paddr
= gva_to_hpa(vcpu
, addr
);
201 if (is_error_hpa(paddr
))
204 guest_buf
= (hva_t
)kmap_atomic(
205 pfn_to_page(paddr
>> PAGE_SHIFT
),
207 offset
= addr
& ~PAGE_MASK
;
209 now
= min(size
, PAGE_SIZE
- offset
);
210 memcpy(host_buf
, (void*)guest_buf
, now
);
214 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
216 return req_size
- size
;
218 EXPORT_SYMBOL_GPL(kvm_read_guest
);
220 int kvm_write_guest(struct kvm_vcpu
*vcpu
, gva_t addr
, unsigned long size
,
223 unsigned char *host_buf
= data
;
224 unsigned long req_size
= size
;
233 paddr
= gva_to_hpa(vcpu
, addr
);
235 if (is_error_hpa(paddr
))
238 gfn
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
) >> PAGE_SHIFT
;
239 mark_page_dirty(vcpu
->kvm
, gfn
);
240 guest_buf
= (hva_t
)kmap_atomic(
241 pfn_to_page(paddr
>> PAGE_SHIFT
), KM_USER0
);
242 offset
= addr
& ~PAGE_MASK
;
244 now
= min(size
, PAGE_SIZE
- offset
);
245 memcpy((void*)guest_buf
, host_buf
, now
);
249 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
251 return req_size
- size
;
253 EXPORT_SYMBOL_GPL(kvm_write_guest
);
256 * Switches to specified vcpu, until a matching vcpu_put()
258 static void vcpu_load(struct kvm_vcpu
*vcpu
)
260 mutex_lock(&vcpu
->mutex
);
261 kvm_arch_ops
->vcpu_load(vcpu
);
265 * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
266 * if the slot is not populated.
268 static struct kvm_vcpu
*vcpu_load_slot(struct kvm
*kvm
, int slot
)
270 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[slot
];
272 mutex_lock(&vcpu
->mutex
);
274 mutex_unlock(&vcpu
->mutex
);
277 kvm_arch_ops
->vcpu_load(vcpu
);
281 static void vcpu_put(struct kvm_vcpu
*vcpu
)
283 kvm_arch_ops
->vcpu_put(vcpu
);
284 mutex_unlock(&vcpu
->mutex
);
287 static struct kvm
*kvm_create_vm(void)
289 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
293 return ERR_PTR(-ENOMEM
);
295 spin_lock_init(&kvm
->lock
);
296 INIT_LIST_HEAD(&kvm
->active_mmu_pages
);
297 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
298 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[i
];
300 mutex_init(&vcpu
->mutex
);
303 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
304 INIT_LIST_HEAD(&vcpu
->free_pages
);
305 spin_lock(&kvm_lock
);
306 list_add(&kvm
->vm_list
, &vm_list
);
307 spin_unlock(&kvm_lock
);
312 static int kvm_dev_open(struct inode
*inode
, struct file
*filp
)
318 * Free any memory in @free but not in @dont.
320 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
321 struct kvm_memory_slot
*dont
)
325 if (!dont
|| free
->phys_mem
!= dont
->phys_mem
)
326 if (free
->phys_mem
) {
327 for (i
= 0; i
< free
->npages
; ++i
)
328 if (free
->phys_mem
[i
])
329 __free_page(free
->phys_mem
[i
]);
330 vfree(free
->phys_mem
);
333 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
334 vfree(free
->dirty_bitmap
);
336 free
->phys_mem
= NULL
;
338 free
->dirty_bitmap
= NULL
;
341 static void kvm_free_physmem(struct kvm
*kvm
)
345 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
346 kvm_free_physmem_slot(&kvm
->memslots
[i
], NULL
);
349 static void kvm_free_vcpu(struct kvm_vcpu
*vcpu
)
355 kvm_mmu_destroy(vcpu
);
357 kvm_arch_ops
->vcpu_free(vcpu
);
358 free_page((unsigned long)vcpu
->run
);
362 static void kvm_free_vcpus(struct kvm
*kvm
)
366 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
367 kvm_free_vcpu(&kvm
->vcpus
[i
]);
370 static int kvm_dev_release(struct inode
*inode
, struct file
*filp
)
375 static void kvm_destroy_vm(struct kvm
*kvm
)
377 spin_lock(&kvm_lock
);
378 list_del(&kvm
->vm_list
);
379 spin_unlock(&kvm_lock
);
381 kvm_free_physmem(kvm
);
385 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
387 struct kvm
*kvm
= filp
->private_data
;
393 static void inject_gp(struct kvm_vcpu
*vcpu
)
395 kvm_arch_ops
->inject_gp(vcpu
, 0);
399 * Load the pae pdptrs. Return true is they are all valid.
401 static int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
403 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
404 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
409 struct kvm_memory_slot
*memslot
;
411 spin_lock(&vcpu
->kvm
->lock
);
412 memslot
= gfn_to_memslot(vcpu
->kvm
, pdpt_gfn
);
413 /* FIXME: !memslot - emulate? 0xff? */
414 pdpt
= kmap_atomic(gfn_to_page(memslot
, pdpt_gfn
), KM_USER0
);
417 for (i
= 0; i
< 4; ++i
) {
418 pdpte
= pdpt
[offset
+ i
];
419 if ((pdpte
& 1) && (pdpte
& 0xfffffff0000001e6ull
)) {
425 for (i
= 0; i
< 4; ++i
)
426 vcpu
->pdptrs
[i
] = pdpt
[offset
+ i
];
429 kunmap_atomic(pdpt
, KM_USER0
);
430 spin_unlock(&vcpu
->kvm
->lock
);
435 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
437 if (cr0
& CR0_RESEVED_BITS
) {
438 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
444 if ((cr0
& CR0_NW_MASK
) && !(cr0
& CR0_CD_MASK
)) {
445 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
450 if ((cr0
& CR0_PG_MASK
) && !(cr0
& CR0_PE_MASK
)) {
451 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
452 "and a clear PE flag\n");
457 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
459 if ((vcpu
->shadow_efer
& EFER_LME
)) {
463 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
464 "in long mode while PAE is disabled\n");
468 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
470 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
471 "in long mode while CS.L == 1\n");
478 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
479 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
487 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
490 spin_lock(&vcpu
->kvm
->lock
);
491 kvm_mmu_reset_context(vcpu
);
492 spin_unlock(&vcpu
->kvm
->lock
);
495 EXPORT_SYMBOL_GPL(set_cr0
);
497 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
499 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
500 set_cr0(vcpu
, (vcpu
->cr0
& ~0x0ful
) | (msw
& 0x0f));
502 EXPORT_SYMBOL_GPL(lmsw
);
504 void set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
506 if (cr4
& CR4_RESEVED_BITS
) {
507 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
512 if (is_long_mode(vcpu
)) {
513 if (!(cr4
& CR4_PAE_MASK
)) {
514 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
519 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& CR4_PAE_MASK
)
520 && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
521 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
525 if (cr4
& CR4_VMXE_MASK
) {
526 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
530 kvm_arch_ops
->set_cr4(vcpu
, cr4
);
531 spin_lock(&vcpu
->kvm
->lock
);
532 kvm_mmu_reset_context(vcpu
);
533 spin_unlock(&vcpu
->kvm
->lock
);
535 EXPORT_SYMBOL_GPL(set_cr4
);
537 void set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
539 if (is_long_mode(vcpu
)) {
540 if (cr3
& CR3_L_MODE_RESEVED_BITS
) {
541 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
546 if (cr3
& CR3_RESEVED_BITS
) {
547 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
551 if (is_paging(vcpu
) && is_pae(vcpu
) &&
552 !load_pdptrs(vcpu
, cr3
)) {
553 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
561 spin_lock(&vcpu
->kvm
->lock
);
563 * Does the new cr3 value map to physical memory? (Note, we
564 * catch an invalid cr3 even in real-mode, because it would
565 * cause trouble later on when we turn on paging anyway.)
567 * A real CPU would silently accept an invalid cr3 and would
568 * attempt to use it - with largely undefined (and often hard
569 * to debug) behavior on the guest side.
571 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
574 vcpu
->mmu
.new_cr3(vcpu
);
575 spin_unlock(&vcpu
->kvm
->lock
);
577 EXPORT_SYMBOL_GPL(set_cr3
);
579 void set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
581 if ( cr8
& CR8_RESEVED_BITS
) {
582 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
588 EXPORT_SYMBOL_GPL(set_cr8
);
590 void fx_init(struct kvm_vcpu
*vcpu
)
592 struct __attribute__ ((__packed__
)) fx_image_s
{
598 u64 operand
;// fpu dp
604 fx_save(vcpu
->host_fx_image
);
606 fx_save(vcpu
->guest_fx_image
);
607 fx_restore(vcpu
->host_fx_image
);
609 fx_image
= (struct fx_image_s
*)vcpu
->guest_fx_image
;
610 fx_image
->mxcsr
= 0x1f80;
611 memset(vcpu
->guest_fx_image
+ sizeof(struct fx_image_s
),
612 0, FX_IMAGE_SIZE
- sizeof(struct fx_image_s
));
614 EXPORT_SYMBOL_GPL(fx_init
);
616 static void do_remove_write_access(struct kvm_vcpu
*vcpu
, int slot
)
618 spin_lock(&vcpu
->kvm
->lock
);
619 kvm_mmu_slot_remove_write_access(vcpu
, slot
);
620 spin_unlock(&vcpu
->kvm
->lock
);
624 * Allocate some memory and give it an address in the guest physical address
627 * Discontiguous memory is allowed, mostly for framebuffers.
629 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
630 struct kvm_memory_region
*mem
)
634 unsigned long npages
;
636 struct kvm_memory_slot
*memslot
;
637 struct kvm_memory_slot old
, new;
638 int memory_config_version
;
641 /* General sanity checks */
642 if (mem
->memory_size
& (PAGE_SIZE
- 1))
644 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
646 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
648 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
651 memslot
= &kvm
->memslots
[mem
->slot
];
652 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
653 npages
= mem
->memory_size
>> PAGE_SHIFT
;
656 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
659 spin_lock(&kvm
->lock
);
661 memory_config_version
= kvm
->memory_config_version
;
662 new = old
= *memslot
;
664 new.base_gfn
= base_gfn
;
666 new.flags
= mem
->flags
;
668 /* Disallow changing a memory slot's size. */
670 if (npages
&& old
.npages
&& npages
!= old
.npages
)
673 /* Check for overlaps */
675 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
676 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
680 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
681 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
685 * Do memory allocations outside lock. memory_config_version will
688 spin_unlock(&kvm
->lock
);
690 /* Deallocate if slot is being removed */
694 /* Free page dirty bitmap if unneeded */
695 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
696 new.dirty_bitmap
= NULL
;
700 /* Allocate if a slot is being created */
701 if (npages
&& !new.phys_mem
) {
702 new.phys_mem
= vmalloc(npages
* sizeof(struct page
*));
707 memset(new.phys_mem
, 0, npages
* sizeof(struct page
*));
708 for (i
= 0; i
< npages
; ++i
) {
709 new.phys_mem
[i
] = alloc_page(GFP_HIGHUSER
711 if (!new.phys_mem
[i
])
713 set_page_private(new.phys_mem
[i
],0);
717 /* Allocate page dirty bitmap if needed */
718 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
719 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
721 new.dirty_bitmap
= vmalloc(dirty_bytes
);
722 if (!new.dirty_bitmap
)
724 memset(new.dirty_bitmap
, 0, dirty_bytes
);
727 spin_lock(&kvm
->lock
);
729 if (memory_config_version
!= kvm
->memory_config_version
) {
730 spin_unlock(&kvm
->lock
);
731 kvm_free_physmem_slot(&new, &old
);
739 if (mem
->slot
>= kvm
->nmemslots
)
740 kvm
->nmemslots
= mem
->slot
+ 1;
743 ++kvm
->memory_config_version
;
745 spin_unlock(&kvm
->lock
);
747 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
748 struct kvm_vcpu
*vcpu
;
750 vcpu
= vcpu_load_slot(kvm
, i
);
753 if (new.flags
& KVM_MEM_LOG_DIRTY_PAGES
)
754 do_remove_write_access(vcpu
, mem
->slot
);
755 kvm_mmu_reset_context(vcpu
);
759 kvm_free_physmem_slot(&old
, &new);
763 spin_unlock(&kvm
->lock
);
765 kvm_free_physmem_slot(&new, &old
);
771 * Get (and clear) the dirty memory log for a memory slot.
773 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
774 struct kvm_dirty_log
*log
)
776 struct kvm_memory_slot
*memslot
;
780 unsigned long any
= 0;
782 spin_lock(&kvm
->lock
);
785 * Prevent changes to guest memory configuration even while the lock
789 spin_unlock(&kvm
->lock
);
791 if (log
->slot
>= KVM_MEMORY_SLOTS
)
794 memslot
= &kvm
->memslots
[log
->slot
];
796 if (!memslot
->dirty_bitmap
)
799 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
801 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
802 any
= memslot
->dirty_bitmap
[i
];
805 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
810 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
811 struct kvm_vcpu
*vcpu
;
813 vcpu
= vcpu_load_slot(kvm
, i
);
817 do_remove_write_access(vcpu
, log
->slot
);
818 memset(memslot
->dirty_bitmap
, 0, n
);
821 kvm_arch_ops
->tlb_flush(vcpu
);
829 spin_lock(&kvm
->lock
);
831 spin_unlock(&kvm
->lock
);
835 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
839 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
840 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
842 if (gfn
>= memslot
->base_gfn
843 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
848 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
850 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
853 struct kvm_memory_slot
*memslot
= NULL
;
854 unsigned long rel_gfn
;
856 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
857 memslot
= &kvm
->memslots
[i
];
859 if (gfn
>= memslot
->base_gfn
860 && gfn
< memslot
->base_gfn
+ memslot
->npages
) {
862 if (!memslot
|| !memslot
->dirty_bitmap
)
865 rel_gfn
= gfn
- memslot
->base_gfn
;
868 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
869 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
875 static int emulator_read_std(unsigned long addr
,
878 struct x86_emulate_ctxt
*ctxt
)
880 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
884 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
885 unsigned offset
= addr
& (PAGE_SIZE
-1);
886 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
888 struct kvm_memory_slot
*memslot
;
891 if (gpa
== UNMAPPED_GVA
)
892 return X86EMUL_PROPAGATE_FAULT
;
893 pfn
= gpa
>> PAGE_SHIFT
;
894 memslot
= gfn_to_memslot(vcpu
->kvm
, pfn
);
896 return X86EMUL_UNHANDLEABLE
;
897 page
= kmap_atomic(gfn_to_page(memslot
, pfn
), KM_USER0
);
899 memcpy(data
, page
+ offset
, tocopy
);
901 kunmap_atomic(page
, KM_USER0
);
908 return X86EMUL_CONTINUE
;
911 static int emulator_write_std(unsigned long addr
,
914 struct x86_emulate_ctxt
*ctxt
)
916 printk(KERN_ERR
"emulator_write_std: addr %lx n %d\n",
918 return X86EMUL_UNHANDLEABLE
;
921 static int emulator_read_emulated(unsigned long addr
,
924 struct x86_emulate_ctxt
*ctxt
)
926 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
928 if (vcpu
->mmio_read_completed
) {
929 memcpy(val
, vcpu
->mmio_data
, bytes
);
930 vcpu
->mmio_read_completed
= 0;
931 return X86EMUL_CONTINUE
;
932 } else if (emulator_read_std(addr
, val
, bytes
, ctxt
)
934 return X86EMUL_CONTINUE
;
936 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
938 if (gpa
== UNMAPPED_GVA
)
939 return X86EMUL_PROPAGATE_FAULT
;
940 vcpu
->mmio_needed
= 1;
941 vcpu
->mmio_phys_addr
= gpa
;
942 vcpu
->mmio_size
= bytes
;
943 vcpu
->mmio_is_write
= 0;
945 return X86EMUL_UNHANDLEABLE
;
949 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
950 unsigned long val
, int bytes
)
952 struct kvm_memory_slot
*m
;
956 if (((gpa
+ bytes
- 1) >> PAGE_SHIFT
) != (gpa
>> PAGE_SHIFT
))
958 m
= gfn_to_memslot(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
961 page
= gfn_to_page(m
, gpa
>> PAGE_SHIFT
);
962 kvm_mmu_pre_write(vcpu
, gpa
, bytes
);
963 mark_page_dirty(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
964 virt
= kmap_atomic(page
, KM_USER0
);
965 memcpy(virt
+ offset_in_page(gpa
), &val
, bytes
);
966 kunmap_atomic(virt
, KM_USER0
);
967 kvm_mmu_post_write(vcpu
, gpa
, bytes
);
971 static int emulator_write_emulated(unsigned long addr
,
974 struct x86_emulate_ctxt
*ctxt
)
976 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
977 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
979 if (gpa
== UNMAPPED_GVA
)
980 return X86EMUL_PROPAGATE_FAULT
;
982 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
983 return X86EMUL_CONTINUE
;
985 vcpu
->mmio_needed
= 1;
986 vcpu
->mmio_phys_addr
= gpa
;
987 vcpu
->mmio_size
= bytes
;
988 vcpu
->mmio_is_write
= 1;
989 memcpy(vcpu
->mmio_data
, &val
, bytes
);
991 return X86EMUL_CONTINUE
;
994 static int emulator_cmpxchg_emulated(unsigned long addr
,
998 struct x86_emulate_ctxt
*ctxt
)
1000 static int reported
;
1004 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1006 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
1009 #ifdef CONFIG_X86_32
1011 static int emulator_cmpxchg8b_emulated(unsigned long addr
,
1012 unsigned long old_lo
,
1013 unsigned long old_hi
,
1014 unsigned long new_lo
,
1015 unsigned long new_hi
,
1016 struct x86_emulate_ctxt
*ctxt
)
1018 static int reported
;
1023 printk(KERN_WARNING
"kvm: emulating exchange8b as write\n");
1025 r
= emulator_write_emulated(addr
, new_lo
, 4, ctxt
);
1026 if (r
!= X86EMUL_CONTINUE
)
1028 return emulator_write_emulated(addr
+4, new_hi
, 4, ctxt
);
1033 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1035 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
1038 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1040 return X86EMUL_CONTINUE
;
1043 int emulate_clts(struct kvm_vcpu
*vcpu
)
1047 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1048 cr0
= vcpu
->cr0
& ~CR0_TS_MASK
;
1049 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
1050 return X86EMUL_CONTINUE
;
1053 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
1055 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1059 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
1060 return X86EMUL_CONTINUE
;
1062 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1064 return X86EMUL_UNHANDLEABLE
;
1068 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1070 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1073 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1075 /* FIXME: better handling */
1076 return X86EMUL_UNHANDLEABLE
;
1078 return X86EMUL_CONTINUE
;
1081 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1083 static int reported
;
1085 unsigned long rip
= ctxt
->vcpu
->rip
;
1086 unsigned long rip_linear
;
1088 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1093 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1095 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1096 " rip %lx %02x %02x %02x %02x\n",
1097 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1101 struct x86_emulate_ops emulate_ops
= {
1102 .read_std
= emulator_read_std
,
1103 .write_std
= emulator_write_std
,
1104 .read_emulated
= emulator_read_emulated
,
1105 .write_emulated
= emulator_write_emulated
,
1106 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1107 #ifdef CONFIG_X86_32
1108 .cmpxchg8b_emulated
= emulator_cmpxchg8b_emulated
,
1112 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1113 struct kvm_run
*run
,
1117 struct x86_emulate_ctxt emulate_ctxt
;
1121 kvm_arch_ops
->cache_regs(vcpu
);
1123 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1125 emulate_ctxt
.vcpu
= vcpu
;
1126 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
1127 emulate_ctxt
.cr2
= cr2
;
1128 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1129 ? X86EMUL_MODE_REAL
: cs_l
1130 ? X86EMUL_MODE_PROT64
: cs_db
1131 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1133 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1134 emulate_ctxt
.cs_base
= 0;
1135 emulate_ctxt
.ds_base
= 0;
1136 emulate_ctxt
.es_base
= 0;
1137 emulate_ctxt
.ss_base
= 0;
1139 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1140 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1141 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1142 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1145 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1146 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1148 vcpu
->mmio_is_write
= 0;
1149 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1151 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1152 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1153 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1154 run
->mmio
.len
= vcpu
->mmio_size
;
1155 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1159 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1160 return EMULATE_DONE
;
1161 if (!vcpu
->mmio_needed
) {
1162 report_emulation_failure(&emulate_ctxt
);
1163 return EMULATE_FAIL
;
1165 return EMULATE_DO_MMIO
;
1168 kvm_arch_ops
->decache_regs(vcpu
);
1169 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1171 if (vcpu
->mmio_is_write
)
1172 return EMULATE_DO_MMIO
;
1174 return EMULATE_DONE
;
1176 EXPORT_SYMBOL_GPL(emulate_instruction
);
1178 int kvm_hypercall(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1180 unsigned long nr
, a0
, a1
, a2
, a3
, a4
, a5
, ret
;
1182 kvm_arch_ops
->cache_regs(vcpu
);
1184 #ifdef CONFIG_X86_64
1185 if (is_long_mode(vcpu
)) {
1186 nr
= vcpu
->regs
[VCPU_REGS_RAX
];
1187 a0
= vcpu
->regs
[VCPU_REGS_RDI
];
1188 a1
= vcpu
->regs
[VCPU_REGS_RSI
];
1189 a2
= vcpu
->regs
[VCPU_REGS_RDX
];
1190 a3
= vcpu
->regs
[VCPU_REGS_RCX
];
1191 a4
= vcpu
->regs
[VCPU_REGS_R8
];
1192 a5
= vcpu
->regs
[VCPU_REGS_R9
];
1196 nr
= vcpu
->regs
[VCPU_REGS_RBX
] & -1u;
1197 a0
= vcpu
->regs
[VCPU_REGS_RAX
] & -1u;
1198 a1
= vcpu
->regs
[VCPU_REGS_RCX
] & -1u;
1199 a2
= vcpu
->regs
[VCPU_REGS_RDX
] & -1u;
1200 a3
= vcpu
->regs
[VCPU_REGS_RSI
] & -1u;
1201 a4
= vcpu
->regs
[VCPU_REGS_RDI
] & -1u;
1202 a5
= vcpu
->regs
[VCPU_REGS_RBP
] & -1u;
1206 run
->hypercall
.args
[0] = a0
;
1207 run
->hypercall
.args
[1] = a1
;
1208 run
->hypercall
.args
[2] = a2
;
1209 run
->hypercall
.args
[3] = a3
;
1210 run
->hypercall
.args
[4] = a4
;
1211 run
->hypercall
.args
[5] = a5
;
1212 run
->hypercall
.ret
= ret
;
1213 run
->hypercall
.longmode
= is_long_mode(vcpu
);
1214 kvm_arch_ops
->decache_regs(vcpu
);
1217 vcpu
->regs
[VCPU_REGS_RAX
] = ret
;
1218 kvm_arch_ops
->decache_regs(vcpu
);
1221 EXPORT_SYMBOL_GPL(kvm_hypercall
);
1223 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1225 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1228 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1230 struct descriptor_table dt
= { limit
, base
};
1232 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1235 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1237 struct descriptor_table dt
= { limit
, base
};
1239 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1242 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1243 unsigned long *rflags
)
1246 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1249 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1251 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1262 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1267 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1268 unsigned long *rflags
)
1272 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1273 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1282 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1285 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1290 * Register the para guest with the host:
1292 static int vcpu_register_para(struct kvm_vcpu
*vcpu
, gpa_t para_state_gpa
)
1294 struct kvm_vcpu_para_state
*para_state
;
1295 hpa_t para_state_hpa
, hypercall_hpa
;
1296 struct page
*para_state_page
;
1297 unsigned char *hypercall
;
1298 gpa_t hypercall_gpa
;
1300 printk(KERN_DEBUG
"kvm: guest trying to enter paravirtual mode\n");
1301 printk(KERN_DEBUG
".... para_state_gpa: %08Lx\n", para_state_gpa
);
1304 * Needs to be page aligned:
1306 if (para_state_gpa
!= PAGE_ALIGN(para_state_gpa
))
1309 para_state_hpa
= gpa_to_hpa(vcpu
, para_state_gpa
);
1310 printk(KERN_DEBUG
".... para_state_hpa: %08Lx\n", para_state_hpa
);
1311 if (is_error_hpa(para_state_hpa
))
1314 mark_page_dirty(vcpu
->kvm
, para_state_gpa
>> PAGE_SHIFT
);
1315 para_state_page
= pfn_to_page(para_state_hpa
>> PAGE_SHIFT
);
1316 para_state
= kmap_atomic(para_state_page
, KM_USER0
);
1318 printk(KERN_DEBUG
".... guest version: %d\n", para_state
->guest_version
);
1319 printk(KERN_DEBUG
".... size: %d\n", para_state
->size
);
1321 para_state
->host_version
= KVM_PARA_API_VERSION
;
1323 * We cannot support guests that try to register themselves
1324 * with a newer API version than the host supports:
1326 if (para_state
->guest_version
> KVM_PARA_API_VERSION
) {
1327 para_state
->ret
= -KVM_EINVAL
;
1328 goto err_kunmap_skip
;
1331 hypercall_gpa
= para_state
->hypercall_gpa
;
1332 hypercall_hpa
= gpa_to_hpa(vcpu
, hypercall_gpa
);
1333 printk(KERN_DEBUG
".... hypercall_hpa: %08Lx\n", hypercall_hpa
);
1334 if (is_error_hpa(hypercall_hpa
)) {
1335 para_state
->ret
= -KVM_EINVAL
;
1336 goto err_kunmap_skip
;
1339 printk(KERN_DEBUG
"kvm: para guest successfully registered.\n");
1340 vcpu
->para_state_page
= para_state_page
;
1341 vcpu
->para_state_gpa
= para_state_gpa
;
1342 vcpu
->hypercall_gpa
= hypercall_gpa
;
1344 mark_page_dirty(vcpu
->kvm
, hypercall_gpa
>> PAGE_SHIFT
);
1345 hypercall
= kmap_atomic(pfn_to_page(hypercall_hpa
>> PAGE_SHIFT
),
1346 KM_USER1
) + (hypercall_hpa
& ~PAGE_MASK
);
1347 kvm_arch_ops
->patch_hypercall(vcpu
, hypercall
);
1348 kunmap_atomic(hypercall
, KM_USER1
);
1350 para_state
->ret
= 0;
1352 kunmap_atomic(para_state
, KM_USER0
);
1358 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1363 case 0xc0010010: /* SYSCFG */
1364 case 0xc0010015: /* HWCR */
1365 case MSR_IA32_PLATFORM_ID
:
1366 case MSR_IA32_P5_MC_ADDR
:
1367 case MSR_IA32_P5_MC_TYPE
:
1368 case MSR_IA32_MC0_CTL
:
1369 case MSR_IA32_MCG_STATUS
:
1370 case MSR_IA32_MCG_CAP
:
1371 case MSR_IA32_MC0_MISC
:
1372 case MSR_IA32_MC0_MISC
+4:
1373 case MSR_IA32_MC0_MISC
+8:
1374 case MSR_IA32_MC0_MISC
+12:
1375 case MSR_IA32_MC0_MISC
+16:
1376 case MSR_IA32_UCODE_REV
:
1377 case MSR_IA32_PERF_STATUS
:
1378 /* MTRR registers */
1380 case 0x200 ... 0x2ff:
1383 case 0xcd: /* fsb frequency */
1386 case MSR_IA32_APICBASE
:
1387 data
= vcpu
->apic_base
;
1389 case MSR_IA32_MISC_ENABLE
:
1390 data
= vcpu
->ia32_misc_enable_msr
;
1392 #ifdef CONFIG_X86_64
1394 data
= vcpu
->shadow_efer
;
1398 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1404 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1407 * Reads an msr value (of 'msr_index') into 'pdata'.
1408 * Returns 0 on success, non-0 otherwise.
1409 * Assumes vcpu_load() was already called.
1411 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1413 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1416 #ifdef CONFIG_X86_64
1418 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1420 if (efer
& EFER_RESERVED_BITS
) {
1421 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1428 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1429 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1434 kvm_arch_ops
->set_efer(vcpu
, efer
);
1437 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1439 vcpu
->shadow_efer
= efer
;
1444 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1447 #ifdef CONFIG_X86_64
1449 set_efer(vcpu
, data
);
1452 case MSR_IA32_MC0_STATUS
:
1453 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1454 __FUNCTION__
, data
);
1456 case MSR_IA32_UCODE_REV
:
1457 case MSR_IA32_UCODE_WRITE
:
1458 case 0x200 ... 0x2ff: /* MTRRs */
1460 case MSR_IA32_APICBASE
:
1461 vcpu
->apic_base
= data
;
1463 case MSR_IA32_MISC_ENABLE
:
1464 vcpu
->ia32_misc_enable_msr
= data
;
1467 * This is the 'probe whether the host is KVM' logic:
1469 case MSR_KVM_API_MAGIC
:
1470 return vcpu_register_para(vcpu
, data
);
1473 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1478 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1481 * Writes msr value into into the appropriate "register".
1482 * Returns 0 on success, non-0 otherwise.
1483 * Assumes vcpu_load() was already called.
1485 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1487 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1490 void kvm_resched(struct kvm_vcpu
*vcpu
)
1496 EXPORT_SYMBOL_GPL(kvm_resched
);
1498 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1502 for (i
= 0; i
< n
; ++i
)
1503 wrmsrl(e
[i
].index
, e
[i
].data
);
1505 EXPORT_SYMBOL_GPL(load_msrs
);
1507 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1511 for (i
= 0; i
< n
; ++i
)
1512 rdmsrl(e
[i
].index
, e
[i
].data
);
1514 EXPORT_SYMBOL_GPL(save_msrs
);
1516 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
1520 struct kvm_cpuid_entry
*e
, *best
;
1522 kvm_arch_ops
->cache_regs(vcpu
);
1523 function
= vcpu
->regs
[VCPU_REGS_RAX
];
1524 vcpu
->regs
[VCPU_REGS_RAX
] = 0;
1525 vcpu
->regs
[VCPU_REGS_RBX
] = 0;
1526 vcpu
->regs
[VCPU_REGS_RCX
] = 0;
1527 vcpu
->regs
[VCPU_REGS_RDX
] = 0;
1529 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
1530 e
= &vcpu
->cpuid_entries
[i
];
1531 if (e
->function
== function
) {
1536 * Both basic or both extended?
1538 if (((e
->function
^ function
) & 0x80000000) == 0)
1539 if (!best
|| e
->function
> best
->function
)
1543 vcpu
->regs
[VCPU_REGS_RAX
] = best
->eax
;
1544 vcpu
->regs
[VCPU_REGS_RBX
] = best
->ebx
;
1545 vcpu
->regs
[VCPU_REGS_RCX
] = best
->ecx
;
1546 vcpu
->regs
[VCPU_REGS_RDX
] = best
->edx
;
1548 kvm_arch_ops
->decache_regs(vcpu
);
1549 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1551 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
1553 static void complete_pio(struct kvm_vcpu
*vcpu
)
1555 struct kvm_io
*io
= &vcpu
->run
->io
;
1558 kvm_arch_ops
->cache_regs(vcpu
);
1561 if (io
->direction
== KVM_EXIT_IO_IN
)
1562 memcpy(&vcpu
->regs
[VCPU_REGS_RAX
], &io
->value
,
1569 * The size of the register should really depend on
1570 * current address size.
1572 vcpu
->regs
[VCPU_REGS_RCX
] -= delta
;
1574 if (io
->string_down
)
1577 if (io
->direction
== KVM_EXIT_IO_IN
)
1578 vcpu
->regs
[VCPU_REGS_RDI
] += delta
;
1580 vcpu
->regs
[VCPU_REGS_RSI
] += delta
;
1583 vcpu
->pio_pending
= 0;
1584 vcpu
->run
->io_completed
= 0;
1586 kvm_arch_ops
->decache_regs(vcpu
);
1588 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1591 static int kvm_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1598 if (vcpu
->sigset_active
)
1599 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1601 /* re-sync apic's tpr */
1602 vcpu
->cr8
= kvm_run
->cr8
;
1604 if (kvm_run
->io_completed
) {
1605 if (vcpu
->pio_pending
)
1608 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1609 vcpu
->mmio_read_completed
= 1;
1613 vcpu
->mmio_needed
= 0;
1615 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
1616 kvm_arch_ops
->cache_regs(vcpu
);
1617 vcpu
->regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
1618 kvm_arch_ops
->decache_regs(vcpu
);
1621 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1623 if (vcpu
->sigset_active
)
1624 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1630 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
,
1631 struct kvm_regs
*regs
)
1635 kvm_arch_ops
->cache_regs(vcpu
);
1637 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1638 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1639 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1640 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1641 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1642 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1643 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1644 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1645 #ifdef CONFIG_X86_64
1646 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1647 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1648 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1649 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1650 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1651 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1652 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1653 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1656 regs
->rip
= vcpu
->rip
;
1657 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1660 * Don't leak debug flags in case they were set for guest debugging
1662 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1663 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1670 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
,
1671 struct kvm_regs
*regs
)
1675 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1676 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1677 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1678 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1679 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1680 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1681 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1682 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1683 #ifdef CONFIG_X86_64
1684 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1685 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1686 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1687 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1688 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1689 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1690 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1691 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1694 vcpu
->rip
= regs
->rip
;
1695 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1697 kvm_arch_ops
->decache_regs(vcpu
);
1704 static void get_segment(struct kvm_vcpu
*vcpu
,
1705 struct kvm_segment
*var
, int seg
)
1707 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1710 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
1711 struct kvm_sregs
*sregs
)
1713 struct descriptor_table dt
;
1717 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1718 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1719 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1720 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1721 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1722 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1724 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1725 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1727 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1728 sregs
->idt
.limit
= dt
.limit
;
1729 sregs
->idt
.base
= dt
.base
;
1730 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1731 sregs
->gdt
.limit
= dt
.limit
;
1732 sregs
->gdt
.base
= dt
.base
;
1734 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1735 sregs
->cr0
= vcpu
->cr0
;
1736 sregs
->cr2
= vcpu
->cr2
;
1737 sregs
->cr3
= vcpu
->cr3
;
1738 sregs
->cr4
= vcpu
->cr4
;
1739 sregs
->cr8
= vcpu
->cr8
;
1740 sregs
->efer
= vcpu
->shadow_efer
;
1741 sregs
->apic_base
= vcpu
->apic_base
;
1743 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1744 sizeof sregs
->interrupt_bitmap
);
1751 static void set_segment(struct kvm_vcpu
*vcpu
,
1752 struct kvm_segment
*var
, int seg
)
1754 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1757 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
1758 struct kvm_sregs
*sregs
)
1760 int mmu_reset_needed
= 0;
1762 struct descriptor_table dt
;
1766 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1767 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1768 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1769 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1770 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1771 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1773 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1774 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1776 dt
.limit
= sregs
->idt
.limit
;
1777 dt
.base
= sregs
->idt
.base
;
1778 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1779 dt
.limit
= sregs
->gdt
.limit
;
1780 dt
.base
= sregs
->gdt
.base
;
1781 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1783 vcpu
->cr2
= sregs
->cr2
;
1784 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
1785 vcpu
->cr3
= sregs
->cr3
;
1787 vcpu
->cr8
= sregs
->cr8
;
1789 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
1790 #ifdef CONFIG_X86_64
1791 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
1793 vcpu
->apic_base
= sregs
->apic_base
;
1795 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1797 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
1798 kvm_arch_ops
->set_cr0_no_modeswitch(vcpu
, sregs
->cr0
);
1800 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
1801 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
1802 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
1803 load_pdptrs(vcpu
, vcpu
->cr3
);
1805 if (mmu_reset_needed
)
1806 kvm_mmu_reset_context(vcpu
);
1808 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
1809 sizeof vcpu
->irq_pending
);
1810 vcpu
->irq_summary
= 0;
1811 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
1812 if (vcpu
->irq_pending
[i
])
1813 __set_bit(i
, &vcpu
->irq_summary
);
1821 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1822 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1824 * This list is modified at module load time to reflect the
1825 * capabilities of the host cpu.
1827 static u32 msrs_to_save
[] = {
1828 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1830 #ifdef CONFIG_X86_64
1831 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1833 MSR_IA32_TIME_STAMP_COUNTER
,
1836 static unsigned num_msrs_to_save
;
1838 static u32 emulated_msrs
[] = {
1839 MSR_IA32_MISC_ENABLE
,
1842 static __init
void kvm_init_msr_list(void)
1847 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
1848 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
1851 msrs_to_save
[j
] = msrs_to_save
[i
];
1854 num_msrs_to_save
= j
;
1858 * Adapt set_msr() to msr_io()'s calling convention
1860 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1862 return set_msr(vcpu
, index
, *data
);
1866 * Read or write a bunch of msrs. All parameters are kernel addresses.
1868 * @return number of msrs set successfully.
1870 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
1871 struct kvm_msr_entry
*entries
,
1872 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1873 unsigned index
, u64
*data
))
1879 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1880 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1889 * Read or write a bunch of msrs. Parameters are user addresses.
1891 * @return number of msrs set successfully.
1893 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
1894 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1895 unsigned index
, u64
*data
),
1898 struct kvm_msrs msrs
;
1899 struct kvm_msr_entry
*entries
;
1904 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1908 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1912 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1913 entries
= vmalloc(size
);
1918 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1921 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
1926 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1938 * Translate a guest virtual address to a guest physical address.
1940 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
1941 struct kvm_translation
*tr
)
1943 unsigned long vaddr
= tr
->linear_address
;
1947 spin_lock(&vcpu
->kvm
->lock
);
1948 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
1949 tr
->physical_address
= gpa
;
1950 tr
->valid
= gpa
!= UNMAPPED_GVA
;
1953 spin_unlock(&vcpu
->kvm
->lock
);
1959 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1960 struct kvm_interrupt
*irq
)
1962 if (irq
->irq
< 0 || irq
->irq
>= 256)
1966 set_bit(irq
->irq
, vcpu
->irq_pending
);
1967 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
1974 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
1975 struct kvm_debug_guest
*dbg
)
1981 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
1988 static struct page
*kvm_vcpu_nopage(struct vm_area_struct
*vma
,
1989 unsigned long address
,
1992 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1993 unsigned long pgoff
;
1996 *type
= VM_FAULT_MINOR
;
1997 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1999 return NOPAGE_SIGBUS
;
2000 page
= virt_to_page(vcpu
->run
);
2005 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
2006 .nopage
= kvm_vcpu_nopage
,
2009 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2011 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2015 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2017 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2019 fput(vcpu
->kvm
->filp
);
2023 static struct file_operations kvm_vcpu_fops
= {
2024 .release
= kvm_vcpu_release
,
2025 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2026 .compat_ioctl
= kvm_vcpu_ioctl
,
2027 .mmap
= kvm_vcpu_mmap
,
2031 * Allocates an inode for the vcpu.
2033 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2036 struct inode
*inode
;
2039 atomic_inc(&vcpu
->kvm
->filp
->f_count
);
2040 inode
= kvmfs_inode(&kvm_vcpu_fops
);
2041 if (IS_ERR(inode
)) {
2046 file
= kvmfs_file(inode
, vcpu
);
2052 r
= get_unused_fd();
2056 fd_install(fd
, file
);
2065 fput(vcpu
->kvm
->filp
);
2070 * Creates some virtual cpus. Good luck creating more than one.
2072 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
2075 struct kvm_vcpu
*vcpu
;
2082 vcpu
= &kvm
->vcpus
[n
];
2084 mutex_lock(&vcpu
->mutex
);
2087 mutex_unlock(&vcpu
->mutex
);
2091 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2095 vcpu
->run
= page_address(page
);
2097 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
2099 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
2101 r
= kvm_arch_ops
->vcpu_create(vcpu
);
2103 goto out_free_vcpus
;
2105 r
= kvm_mmu_create(vcpu
);
2107 goto out_free_vcpus
;
2109 kvm_arch_ops
->vcpu_load(vcpu
);
2110 r
= kvm_mmu_setup(vcpu
);
2112 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
2116 goto out_free_vcpus
;
2118 r
= create_vcpu_fd(vcpu
);
2120 goto out_free_vcpus
;
2125 kvm_free_vcpu(vcpu
);
2127 mutex_unlock(&vcpu
->mutex
);
2132 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
2133 struct kvm_cpuid
*cpuid
,
2134 struct kvm_cpuid_entry __user
*entries
)
2139 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2142 if (copy_from_user(&vcpu
->cpuid_entries
, entries
,
2143 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
2145 vcpu
->cpuid_nent
= cpuid
->nent
;
2152 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2155 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2156 vcpu
->sigset_active
= 1;
2157 vcpu
->sigset
= *sigset
;
2159 vcpu
->sigset_active
= 0;
2163 static long kvm_vcpu_ioctl(struct file
*filp
,
2164 unsigned int ioctl
, unsigned long arg
)
2166 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2167 void __user
*argp
= (void __user
*)arg
;
2175 r
= kvm_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2177 case KVM_GET_REGS
: {
2178 struct kvm_regs kvm_regs
;
2180 memset(&kvm_regs
, 0, sizeof kvm_regs
);
2181 r
= kvm_vcpu_ioctl_get_regs(vcpu
, &kvm_regs
);
2185 if (copy_to_user(argp
, &kvm_regs
, sizeof kvm_regs
))
2190 case KVM_SET_REGS
: {
2191 struct kvm_regs kvm_regs
;
2194 if (copy_from_user(&kvm_regs
, argp
, sizeof kvm_regs
))
2196 r
= kvm_vcpu_ioctl_set_regs(vcpu
, &kvm_regs
);
2202 case KVM_GET_SREGS
: {
2203 struct kvm_sregs kvm_sregs
;
2205 memset(&kvm_sregs
, 0, sizeof kvm_sregs
);
2206 r
= kvm_vcpu_ioctl_get_sregs(vcpu
, &kvm_sregs
);
2210 if (copy_to_user(argp
, &kvm_sregs
, sizeof kvm_sregs
))
2215 case KVM_SET_SREGS
: {
2216 struct kvm_sregs kvm_sregs
;
2219 if (copy_from_user(&kvm_sregs
, argp
, sizeof kvm_sregs
))
2221 r
= kvm_vcpu_ioctl_set_sregs(vcpu
, &kvm_sregs
);
2227 case KVM_TRANSLATE
: {
2228 struct kvm_translation tr
;
2231 if (copy_from_user(&tr
, argp
, sizeof tr
))
2233 r
= kvm_vcpu_ioctl_translate(vcpu
, &tr
);
2237 if (copy_to_user(argp
, &tr
, sizeof tr
))
2242 case KVM_INTERRUPT
: {
2243 struct kvm_interrupt irq
;
2246 if (copy_from_user(&irq
, argp
, sizeof irq
))
2248 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2254 case KVM_DEBUG_GUEST
: {
2255 struct kvm_debug_guest dbg
;
2258 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2260 r
= kvm_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
2267 r
= msr_io(vcpu
, argp
, get_msr
, 1);
2270 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2272 case KVM_SET_CPUID
: {
2273 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2274 struct kvm_cpuid cpuid
;
2277 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2279 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2284 case KVM_SET_SIGNAL_MASK
: {
2285 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2286 struct kvm_signal_mask kvm_sigmask
;
2287 sigset_t sigset
, *p
;
2292 if (copy_from_user(&kvm_sigmask
, argp
,
2293 sizeof kvm_sigmask
))
2296 if (kvm_sigmask
.len
!= sizeof sigset
)
2299 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2304 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2314 static long kvm_vm_ioctl(struct file
*filp
,
2315 unsigned int ioctl
, unsigned long arg
)
2317 struct kvm
*kvm
= filp
->private_data
;
2318 void __user
*argp
= (void __user
*)arg
;
2322 case KVM_CREATE_VCPU
:
2323 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2327 case KVM_SET_MEMORY_REGION
: {
2328 struct kvm_memory_region kvm_mem
;
2331 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2333 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_mem
);
2338 case KVM_GET_DIRTY_LOG
: {
2339 struct kvm_dirty_log log
;
2342 if (copy_from_user(&log
, argp
, sizeof log
))
2344 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2356 static struct page
*kvm_vm_nopage(struct vm_area_struct
*vma
,
2357 unsigned long address
,
2360 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2361 unsigned long pgoff
;
2362 struct kvm_memory_slot
*slot
;
2365 *type
= VM_FAULT_MINOR
;
2366 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2367 slot
= gfn_to_memslot(kvm
, pgoff
);
2369 return NOPAGE_SIGBUS
;
2370 page
= gfn_to_page(slot
, pgoff
);
2372 return NOPAGE_SIGBUS
;
2377 static struct vm_operations_struct kvm_vm_vm_ops
= {
2378 .nopage
= kvm_vm_nopage
,
2381 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2383 vma
->vm_ops
= &kvm_vm_vm_ops
;
2387 static struct file_operations kvm_vm_fops
= {
2388 .release
= kvm_vm_release
,
2389 .unlocked_ioctl
= kvm_vm_ioctl
,
2390 .compat_ioctl
= kvm_vm_ioctl
,
2391 .mmap
= kvm_vm_mmap
,
2394 static int kvm_dev_ioctl_create_vm(void)
2397 struct inode
*inode
;
2401 inode
= kvmfs_inode(&kvm_vm_fops
);
2402 if (IS_ERR(inode
)) {
2407 kvm
= kvm_create_vm();
2413 file
= kvmfs_file(inode
, kvm
);
2420 r
= get_unused_fd();
2424 fd_install(fd
, file
);
2431 kvm_destroy_vm(kvm
);
2438 static long kvm_dev_ioctl(struct file
*filp
,
2439 unsigned int ioctl
, unsigned long arg
)
2441 void __user
*argp
= (void __user
*)arg
;
2445 case KVM_GET_API_VERSION
:
2449 r
= KVM_API_VERSION
;
2455 r
= kvm_dev_ioctl_create_vm();
2457 case KVM_GET_MSR_INDEX_LIST
: {
2458 struct kvm_msr_list __user
*user_msr_list
= argp
;
2459 struct kvm_msr_list msr_list
;
2463 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2466 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2467 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2470 if (n
< num_msrs_to_save
)
2473 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2474 num_msrs_to_save
* sizeof(u32
)))
2476 if (copy_to_user(user_msr_list
->indices
2477 + num_msrs_to_save
* sizeof(u32
),
2479 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2484 case KVM_CHECK_EXTENSION
:
2486 * No extensions defined at present.
2490 case KVM_GET_VCPU_MMAP_SIZE
:
2503 static struct file_operations kvm_chardev_ops
= {
2504 .open
= kvm_dev_open
,
2505 .release
= kvm_dev_release
,
2506 .unlocked_ioctl
= kvm_dev_ioctl
,
2507 .compat_ioctl
= kvm_dev_ioctl
,
2510 static struct miscdevice kvm_dev
= {
2516 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2519 if (val
== SYS_RESTART
) {
2521 * Some (well, at least mine) BIOSes hang on reboot if
2524 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2525 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2530 static struct notifier_block kvm_reboot_notifier
= {
2531 .notifier_call
= kvm_reboot
,
2536 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2539 static void decache_vcpus_on_cpu(int cpu
)
2542 struct kvm_vcpu
*vcpu
;
2545 spin_lock(&kvm_lock
);
2546 list_for_each_entry(vm
, &vm_list
, vm_list
)
2547 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2548 vcpu
= &vm
->vcpus
[i
];
2550 * If the vcpu is locked, then it is running on some
2551 * other cpu and therefore it is not cached on the
2554 * If it's not locked, check the last cpu it executed
2557 if (mutex_trylock(&vcpu
->mutex
)) {
2558 if (vcpu
->cpu
== cpu
) {
2559 kvm_arch_ops
->vcpu_decache(vcpu
);
2562 mutex_unlock(&vcpu
->mutex
);
2565 spin_unlock(&kvm_lock
);
2568 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2574 case CPU_DOWN_PREPARE
:
2575 case CPU_UP_CANCELED
:
2576 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2578 decache_vcpus_on_cpu(cpu
);
2579 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_disable
,
2583 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2585 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_enable
,
2592 static struct notifier_block kvm_cpu_notifier
= {
2593 .notifier_call
= kvm_cpu_hotplug
,
2594 .priority
= 20, /* must be > scheduler priority */
2597 static __init
void kvm_init_debug(void)
2599 struct kvm_stats_debugfs_item
*p
;
2601 debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2602 for (p
= debugfs_entries
; p
->name
; ++p
)
2603 p
->dentry
= debugfs_create_u32(p
->name
, 0444, debugfs_dir
,
2607 static void kvm_exit_debug(void)
2609 struct kvm_stats_debugfs_item
*p
;
2611 for (p
= debugfs_entries
; p
->name
; ++p
)
2612 debugfs_remove(p
->dentry
);
2613 debugfs_remove(debugfs_dir
);
2616 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2618 decache_vcpus_on_cpu(raw_smp_processor_id());
2619 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2623 static int kvm_resume(struct sys_device
*dev
)
2625 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
2629 static struct sysdev_class kvm_sysdev_class
= {
2630 set_kset_name("kvm"),
2631 .suspend
= kvm_suspend
,
2632 .resume
= kvm_resume
,
2635 static struct sys_device kvm_sysdev
= {
2637 .cls
= &kvm_sysdev_class
,
2640 hpa_t bad_page_address
;
2642 static int kvmfs_get_sb(struct file_system_type
*fs_type
, int flags
,
2643 const char *dev_name
, void *data
, struct vfsmount
*mnt
)
2645 return get_sb_pseudo(fs_type
, "kvm:", NULL
, KVMFS_SUPER_MAGIC
, mnt
);
2648 static struct file_system_type kvm_fs_type
= {
2650 .get_sb
= kvmfs_get_sb
,
2651 .kill_sb
= kill_anon_super
,
2654 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
2659 printk(KERN_ERR
"kvm: already loaded the other module\n");
2663 if (!ops
->cpu_has_kvm_support()) {
2664 printk(KERN_ERR
"kvm: no hardware support\n");
2667 if (ops
->disabled_by_bios()) {
2668 printk(KERN_ERR
"kvm: disabled by bios\n");
2674 r
= kvm_arch_ops
->hardware_setup();
2678 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
2679 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2682 register_reboot_notifier(&kvm_reboot_notifier
);
2684 r
= sysdev_class_register(&kvm_sysdev_class
);
2688 r
= sysdev_register(&kvm_sysdev
);
2692 kvm_chardev_ops
.owner
= module
;
2694 r
= misc_register(&kvm_dev
);
2696 printk (KERN_ERR
"kvm: misc device register failed\n");
2703 sysdev_unregister(&kvm_sysdev
);
2705 sysdev_class_unregister(&kvm_sysdev_class
);
2707 unregister_reboot_notifier(&kvm_reboot_notifier
);
2708 unregister_cpu_notifier(&kvm_cpu_notifier
);
2710 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2711 kvm_arch_ops
->hardware_unsetup();
2713 kvm_arch_ops
= NULL
;
2717 void kvm_exit_arch(void)
2719 misc_deregister(&kvm_dev
);
2720 sysdev_unregister(&kvm_sysdev
);
2721 sysdev_class_unregister(&kvm_sysdev_class
);
2722 unregister_reboot_notifier(&kvm_reboot_notifier
);
2723 unregister_cpu_notifier(&kvm_cpu_notifier
);
2724 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2725 kvm_arch_ops
->hardware_unsetup();
2726 kvm_arch_ops
= NULL
;
2729 static __init
int kvm_init(void)
2731 static struct page
*bad_page
;
2734 r
= register_filesystem(&kvm_fs_type
);
2738 kvmfs_mnt
= kern_mount(&kvm_fs_type
);
2739 r
= PTR_ERR(kvmfs_mnt
);
2740 if (IS_ERR(kvmfs_mnt
))
2744 kvm_init_msr_list();
2746 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
2751 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
2752 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
2760 unregister_filesystem(&kvm_fs_type
);
2765 static __exit
void kvm_exit(void)
2768 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
2770 unregister_filesystem(&kvm_fs_type
);
2773 module_init(kvm_init
)
2774 module_exit(kvm_exit
)
2776 EXPORT_SYMBOL_GPL(kvm_init_arch
);
2777 EXPORT_SYMBOL_GPL(kvm_exit_arch
);