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>
43 #include <linux/sched.h>
45 #include "x86_emulate.h"
46 #include "segment_descriptor.h"
48 MODULE_AUTHOR("Qumranet");
49 MODULE_LICENSE("GPL");
51 static DEFINE_SPINLOCK(kvm_lock
);
52 static LIST_HEAD(vm_list
);
54 struct kvm_arch_ops
*kvm_arch_ops
;
56 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
58 static struct kvm_stats_debugfs_item
{
61 struct dentry
*dentry
;
62 } debugfs_entries
[] = {
63 { "pf_fixed", STAT_OFFSET(pf_fixed
) },
64 { "pf_guest", STAT_OFFSET(pf_guest
) },
65 { "tlb_flush", STAT_OFFSET(tlb_flush
) },
66 { "invlpg", STAT_OFFSET(invlpg
) },
67 { "exits", STAT_OFFSET(exits
) },
68 { "io_exits", STAT_OFFSET(io_exits
) },
69 { "mmio_exits", STAT_OFFSET(mmio_exits
) },
70 { "signal_exits", STAT_OFFSET(signal_exits
) },
71 { "irq_window", STAT_OFFSET(irq_window_exits
) },
72 { "halt_exits", STAT_OFFSET(halt_exits
) },
73 { "request_irq", STAT_OFFSET(request_irq_exits
) },
74 { "irq_exits", STAT_OFFSET(irq_exits
) },
78 static struct dentry
*debugfs_dir
;
80 struct vfsmount
*kvmfs_mnt
;
82 #define MAX_IO_MSRS 256
84 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
85 #define LMSW_GUEST_MASK 0x0eULL
86 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
87 #define CR8_RESEVED_BITS (~0x0fULL)
88 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
91 // LDT or TSS descriptor in the GDT. 16 bytes.
92 struct segment_descriptor_64
{
93 struct segment_descriptor s
;
100 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
103 static struct inode
*kvmfs_inode(struct file_operations
*fops
)
106 struct inode
*inode
= new_inode(kvmfs_mnt
->mnt_sb
);
114 * Mark the inode dirty from the very beginning,
115 * that way it will never be moved to the dirty
116 * list because mark_inode_dirty() will think
117 * that it already _is_ on the dirty list.
119 inode
->i_state
= I_DIRTY
;
120 inode
->i_mode
= S_IRUSR
| S_IWUSR
;
121 inode
->i_uid
= current
->fsuid
;
122 inode
->i_gid
= current
->fsgid
;
123 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
127 return ERR_PTR(error
);
130 static struct file
*kvmfs_file(struct inode
*inode
, void *private_data
)
132 struct file
*file
= get_empty_filp();
135 return ERR_PTR(-ENFILE
);
137 file
->f_path
.mnt
= mntget(kvmfs_mnt
);
138 file
->f_path
.dentry
= d_alloc_anon(inode
);
139 if (!file
->f_path
.dentry
)
140 return ERR_PTR(-ENOMEM
);
141 file
->f_mapping
= inode
->i_mapping
;
144 file
->f_flags
= O_RDWR
;
145 file
->f_op
= inode
->i_fop
;
146 file
->f_mode
= FMODE_READ
| FMODE_WRITE
;
148 file
->private_data
= private_data
;
152 unsigned long segment_base(u16 selector
)
154 struct descriptor_table gdt
;
155 struct segment_descriptor
*d
;
156 unsigned long table_base
;
157 typedef unsigned long ul
;
163 asm ("sgdt %0" : "=m"(gdt
));
164 table_base
= gdt
.base
;
166 if (selector
& 4) { /* from ldt */
169 asm ("sldt %0" : "=g"(ldt_selector
));
170 table_base
= segment_base(ldt_selector
);
172 d
= (struct segment_descriptor
*)(table_base
+ (selector
& ~7));
173 v
= d
->base_low
| ((ul
)d
->base_mid
<< 16) | ((ul
)d
->base_high
<< 24);
176 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
177 v
|= ((ul
)((struct segment_descriptor_64
*)d
)->base_higher
) << 32;
181 EXPORT_SYMBOL_GPL(segment_base
);
183 static inline int valid_vcpu(int n
)
185 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
188 int kvm_read_guest(struct kvm_vcpu
*vcpu
, gva_t addr
, unsigned long size
,
191 unsigned char *host_buf
= dest
;
192 unsigned long req_size
= size
;
200 paddr
= gva_to_hpa(vcpu
, addr
);
202 if (is_error_hpa(paddr
))
205 guest_buf
= (hva_t
)kmap_atomic(
206 pfn_to_page(paddr
>> PAGE_SHIFT
),
208 offset
= addr
& ~PAGE_MASK
;
210 now
= min(size
, PAGE_SIZE
- offset
);
211 memcpy(host_buf
, (void*)guest_buf
, now
);
215 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
217 return req_size
- size
;
219 EXPORT_SYMBOL_GPL(kvm_read_guest
);
221 int kvm_write_guest(struct kvm_vcpu
*vcpu
, gva_t addr
, unsigned long size
,
224 unsigned char *host_buf
= data
;
225 unsigned long req_size
= size
;
234 paddr
= gva_to_hpa(vcpu
, addr
);
236 if (is_error_hpa(paddr
))
239 gfn
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
) >> PAGE_SHIFT
;
240 mark_page_dirty(vcpu
->kvm
, gfn
);
241 guest_buf
= (hva_t
)kmap_atomic(
242 pfn_to_page(paddr
>> PAGE_SHIFT
), KM_USER0
);
243 offset
= addr
& ~PAGE_MASK
;
245 now
= min(size
, PAGE_SIZE
- offset
);
246 memcpy((void*)guest_buf
, host_buf
, now
);
250 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
252 return req_size
- size
;
254 EXPORT_SYMBOL_GPL(kvm_write_guest
);
257 * Switches to specified vcpu, until a matching vcpu_put()
259 static void vcpu_load(struct kvm_vcpu
*vcpu
)
261 mutex_lock(&vcpu
->mutex
);
262 kvm_arch_ops
->vcpu_load(vcpu
);
266 * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
267 * if the slot is not populated.
269 static struct kvm_vcpu
*vcpu_load_slot(struct kvm
*kvm
, int slot
)
271 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[slot
];
273 mutex_lock(&vcpu
->mutex
);
275 mutex_unlock(&vcpu
->mutex
);
278 kvm_arch_ops
->vcpu_load(vcpu
);
282 static void vcpu_put(struct kvm_vcpu
*vcpu
)
284 kvm_arch_ops
->vcpu_put(vcpu
);
285 mutex_unlock(&vcpu
->mutex
);
288 static struct kvm
*kvm_create_vm(void)
290 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
294 return ERR_PTR(-ENOMEM
);
296 spin_lock_init(&kvm
->lock
);
297 INIT_LIST_HEAD(&kvm
->active_mmu_pages
);
298 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
299 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[i
];
301 mutex_init(&vcpu
->mutex
);
304 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
305 INIT_LIST_HEAD(&vcpu
->free_pages
);
306 spin_lock(&kvm_lock
);
307 list_add(&kvm
->vm_list
, &vm_list
);
308 spin_unlock(&kvm_lock
);
313 static int kvm_dev_open(struct inode
*inode
, struct file
*filp
)
319 * Free any memory in @free but not in @dont.
321 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
322 struct kvm_memory_slot
*dont
)
326 if (!dont
|| free
->phys_mem
!= dont
->phys_mem
)
327 if (free
->phys_mem
) {
328 for (i
= 0; i
< free
->npages
; ++i
)
329 if (free
->phys_mem
[i
])
330 __free_page(free
->phys_mem
[i
]);
331 vfree(free
->phys_mem
);
334 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
335 vfree(free
->dirty_bitmap
);
337 free
->phys_mem
= NULL
;
339 free
->dirty_bitmap
= NULL
;
342 static void kvm_free_physmem(struct kvm
*kvm
)
346 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
347 kvm_free_physmem_slot(&kvm
->memslots
[i
], NULL
);
350 static void free_pio_guest_pages(struct kvm_vcpu
*vcpu
)
354 for (i
= 0; i
< 2; ++i
)
355 if (vcpu
->pio
.guest_pages
[i
]) {
356 __free_page(vcpu
->pio
.guest_pages
[i
]);
357 vcpu
->pio
.guest_pages
[i
] = NULL
;
361 static void kvm_free_vcpu(struct kvm_vcpu
*vcpu
)
367 kvm_mmu_destroy(vcpu
);
369 kvm_arch_ops
->vcpu_free(vcpu
);
370 free_page((unsigned long)vcpu
->run
);
372 free_page((unsigned long)vcpu
->pio_data
);
373 vcpu
->pio_data
= NULL
;
374 free_pio_guest_pages(vcpu
);
377 static void kvm_free_vcpus(struct kvm
*kvm
)
381 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
382 kvm_free_vcpu(&kvm
->vcpus
[i
]);
385 static int kvm_dev_release(struct inode
*inode
, struct file
*filp
)
390 static void kvm_destroy_vm(struct kvm
*kvm
)
392 spin_lock(&kvm_lock
);
393 list_del(&kvm
->vm_list
);
394 spin_unlock(&kvm_lock
);
396 kvm_free_physmem(kvm
);
400 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
402 struct kvm
*kvm
= filp
->private_data
;
408 static void inject_gp(struct kvm_vcpu
*vcpu
)
410 kvm_arch_ops
->inject_gp(vcpu
, 0);
414 * Load the pae pdptrs. Return true is they are all valid.
416 static int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
418 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
419 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
426 spin_lock(&vcpu
->kvm
->lock
);
427 page
= gfn_to_page(vcpu
->kvm
, pdpt_gfn
);
428 /* FIXME: !page - emulate? 0xff? */
429 pdpt
= kmap_atomic(page
, KM_USER0
);
432 for (i
= 0; i
< 4; ++i
) {
433 pdpte
= pdpt
[offset
+ i
];
434 if ((pdpte
& 1) && (pdpte
& 0xfffffff0000001e6ull
)) {
440 for (i
= 0; i
< 4; ++i
)
441 vcpu
->pdptrs
[i
] = pdpt
[offset
+ i
];
444 kunmap_atomic(pdpt
, KM_USER0
);
445 spin_unlock(&vcpu
->kvm
->lock
);
450 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
452 if (cr0
& CR0_RESEVED_BITS
) {
453 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
459 if ((cr0
& CR0_NW_MASK
) && !(cr0
& CR0_CD_MASK
)) {
460 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
465 if ((cr0
& CR0_PG_MASK
) && !(cr0
& CR0_PE_MASK
)) {
466 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
467 "and a clear PE flag\n");
472 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
474 if ((vcpu
->shadow_efer
& EFER_LME
)) {
478 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
479 "in long mode while PAE is disabled\n");
483 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
485 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
486 "in long mode while CS.L == 1\n");
493 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
494 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
502 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
505 spin_lock(&vcpu
->kvm
->lock
);
506 kvm_mmu_reset_context(vcpu
);
507 spin_unlock(&vcpu
->kvm
->lock
);
510 EXPORT_SYMBOL_GPL(set_cr0
);
512 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
514 set_cr0(vcpu
, (vcpu
->cr0
& ~0x0ful
) | (msw
& 0x0f));
516 EXPORT_SYMBOL_GPL(lmsw
);
518 void set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
520 if (cr4
& CR4_RESEVED_BITS
) {
521 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
526 if (is_long_mode(vcpu
)) {
527 if (!(cr4
& CR4_PAE_MASK
)) {
528 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
533 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& CR4_PAE_MASK
)
534 && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
535 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
539 if (cr4
& CR4_VMXE_MASK
) {
540 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
544 kvm_arch_ops
->set_cr4(vcpu
, cr4
);
545 spin_lock(&vcpu
->kvm
->lock
);
546 kvm_mmu_reset_context(vcpu
);
547 spin_unlock(&vcpu
->kvm
->lock
);
549 EXPORT_SYMBOL_GPL(set_cr4
);
551 void set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
553 if (is_long_mode(vcpu
)) {
554 if (cr3
& CR3_L_MODE_RESEVED_BITS
) {
555 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
560 if (cr3
& CR3_RESEVED_BITS
) {
561 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
565 if (is_paging(vcpu
) && is_pae(vcpu
) &&
566 !load_pdptrs(vcpu
, cr3
)) {
567 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
575 spin_lock(&vcpu
->kvm
->lock
);
577 * Does the new cr3 value map to physical memory? (Note, we
578 * catch an invalid cr3 even in real-mode, because it would
579 * cause trouble later on when we turn on paging anyway.)
581 * A real CPU would silently accept an invalid cr3 and would
582 * attempt to use it - with largely undefined (and often hard
583 * to debug) behavior on the guest side.
585 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
588 vcpu
->mmu
.new_cr3(vcpu
);
589 spin_unlock(&vcpu
->kvm
->lock
);
591 EXPORT_SYMBOL_GPL(set_cr3
);
593 void set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
595 if ( cr8
& CR8_RESEVED_BITS
) {
596 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
602 EXPORT_SYMBOL_GPL(set_cr8
);
604 void fx_init(struct kvm_vcpu
*vcpu
)
606 struct __attribute__ ((__packed__
)) fx_image_s
{
612 u64 operand
;// fpu dp
618 fx_save(vcpu
->host_fx_image
);
620 fx_save(vcpu
->guest_fx_image
);
621 fx_restore(vcpu
->host_fx_image
);
623 fx_image
= (struct fx_image_s
*)vcpu
->guest_fx_image
;
624 fx_image
->mxcsr
= 0x1f80;
625 memset(vcpu
->guest_fx_image
+ sizeof(struct fx_image_s
),
626 0, FX_IMAGE_SIZE
- sizeof(struct fx_image_s
));
628 EXPORT_SYMBOL_GPL(fx_init
);
630 static void do_remove_write_access(struct kvm_vcpu
*vcpu
, int slot
)
632 spin_lock(&vcpu
->kvm
->lock
);
633 kvm_mmu_slot_remove_write_access(vcpu
, slot
);
634 spin_unlock(&vcpu
->kvm
->lock
);
638 * Allocate some memory and give it an address in the guest physical address
641 * Discontiguous memory is allowed, mostly for framebuffers.
643 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
644 struct kvm_memory_region
*mem
)
648 unsigned long npages
;
650 struct kvm_memory_slot
*memslot
;
651 struct kvm_memory_slot old
, new;
652 int memory_config_version
;
655 /* General sanity checks */
656 if (mem
->memory_size
& (PAGE_SIZE
- 1))
658 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
660 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
662 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
665 memslot
= &kvm
->memslots
[mem
->slot
];
666 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
667 npages
= mem
->memory_size
>> PAGE_SHIFT
;
670 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
673 spin_lock(&kvm
->lock
);
675 memory_config_version
= kvm
->memory_config_version
;
676 new = old
= *memslot
;
678 new.base_gfn
= base_gfn
;
680 new.flags
= mem
->flags
;
682 /* Disallow changing a memory slot's size. */
684 if (npages
&& old
.npages
&& npages
!= old
.npages
)
687 /* Check for overlaps */
689 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
690 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
694 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
695 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
699 * Do memory allocations outside lock. memory_config_version will
702 spin_unlock(&kvm
->lock
);
704 /* Deallocate if slot is being removed */
708 /* Free page dirty bitmap if unneeded */
709 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
710 new.dirty_bitmap
= NULL
;
714 /* Allocate if a slot is being created */
715 if (npages
&& !new.phys_mem
) {
716 new.phys_mem
= vmalloc(npages
* sizeof(struct page
*));
721 memset(new.phys_mem
, 0, npages
* sizeof(struct page
*));
722 for (i
= 0; i
< npages
; ++i
) {
723 new.phys_mem
[i
] = alloc_page(GFP_HIGHUSER
725 if (!new.phys_mem
[i
])
727 set_page_private(new.phys_mem
[i
],0);
731 /* Allocate page dirty bitmap if needed */
732 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
733 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
735 new.dirty_bitmap
= vmalloc(dirty_bytes
);
736 if (!new.dirty_bitmap
)
738 memset(new.dirty_bitmap
, 0, dirty_bytes
);
741 spin_lock(&kvm
->lock
);
743 if (memory_config_version
!= kvm
->memory_config_version
) {
744 spin_unlock(&kvm
->lock
);
745 kvm_free_physmem_slot(&new, &old
);
753 if (mem
->slot
>= kvm
->nmemslots
)
754 kvm
->nmemslots
= mem
->slot
+ 1;
757 ++kvm
->memory_config_version
;
759 spin_unlock(&kvm
->lock
);
761 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
762 struct kvm_vcpu
*vcpu
;
764 vcpu
= vcpu_load_slot(kvm
, i
);
767 if (new.flags
& KVM_MEM_LOG_DIRTY_PAGES
)
768 do_remove_write_access(vcpu
, mem
->slot
);
769 kvm_mmu_reset_context(vcpu
);
773 kvm_free_physmem_slot(&old
, &new);
777 spin_unlock(&kvm
->lock
);
779 kvm_free_physmem_slot(&new, &old
);
785 * Get (and clear) the dirty memory log for a memory slot.
787 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
788 struct kvm_dirty_log
*log
)
790 struct kvm_memory_slot
*memslot
;
794 unsigned long any
= 0;
796 spin_lock(&kvm
->lock
);
799 * Prevent changes to guest memory configuration even while the lock
803 spin_unlock(&kvm
->lock
);
805 if (log
->slot
>= KVM_MEMORY_SLOTS
)
808 memslot
= &kvm
->memslots
[log
->slot
];
810 if (!memslot
->dirty_bitmap
)
813 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
815 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
816 any
= memslot
->dirty_bitmap
[i
];
819 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
824 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
825 struct kvm_vcpu
*vcpu
;
827 vcpu
= vcpu_load_slot(kvm
, i
);
831 do_remove_write_access(vcpu
, log
->slot
);
832 memset(memslot
->dirty_bitmap
, 0, n
);
835 kvm_arch_ops
->tlb_flush(vcpu
);
843 spin_lock(&kvm
->lock
);
845 spin_unlock(&kvm
->lock
);
850 * Set a new alias region. Aliases map a portion of physical memory into
851 * another portion. This is useful for memory windows, for example the PC
854 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
855 struct kvm_memory_alias
*alias
)
858 struct kvm_mem_alias
*p
;
861 /* General sanity checks */
862 if (alias
->memory_size
& (PAGE_SIZE
- 1))
864 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
866 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
868 if (alias
->guest_phys_addr
+ alias
->memory_size
869 < alias
->guest_phys_addr
)
871 if (alias
->target_phys_addr
+ alias
->memory_size
872 < alias
->target_phys_addr
)
875 spin_lock(&kvm
->lock
);
877 p
= &kvm
->aliases
[alias
->slot
];
878 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
879 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
880 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
882 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
883 if (kvm
->aliases
[n
- 1].npages
)
887 spin_unlock(&kvm
->lock
);
889 vcpu_load(&kvm
->vcpus
[0]);
890 spin_lock(&kvm
->lock
);
891 kvm_mmu_zap_all(&kvm
->vcpus
[0]);
892 spin_unlock(&kvm
->lock
);
893 vcpu_put(&kvm
->vcpus
[0]);
901 static gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
904 struct kvm_mem_alias
*alias
;
906 for (i
= 0; i
< kvm
->naliases
; ++i
) {
907 alias
= &kvm
->aliases
[i
];
908 if (gfn
>= alias
->base_gfn
909 && gfn
< alias
->base_gfn
+ alias
->npages
)
910 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
915 static struct kvm_memory_slot
*__gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
919 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
920 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
922 if (gfn
>= memslot
->base_gfn
923 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
929 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
931 gfn
= unalias_gfn(kvm
, gfn
);
932 return __gfn_to_memslot(kvm
, gfn
);
935 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
937 struct kvm_memory_slot
*slot
;
939 gfn
= unalias_gfn(kvm
, gfn
);
940 slot
= __gfn_to_memslot(kvm
, gfn
);
943 return slot
->phys_mem
[gfn
- slot
->base_gfn
];
945 EXPORT_SYMBOL_GPL(gfn_to_page
);
947 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
950 struct kvm_memory_slot
*memslot
= NULL
;
951 unsigned long rel_gfn
;
953 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
954 memslot
= &kvm
->memslots
[i
];
956 if (gfn
>= memslot
->base_gfn
957 && gfn
< memslot
->base_gfn
+ memslot
->npages
) {
959 if (!memslot
|| !memslot
->dirty_bitmap
)
962 rel_gfn
= gfn
- memslot
->base_gfn
;
965 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
966 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
972 static int emulator_read_std(unsigned long addr
,
975 struct x86_emulate_ctxt
*ctxt
)
977 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
981 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
982 unsigned offset
= addr
& (PAGE_SIZE
-1);
983 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
988 if (gpa
== UNMAPPED_GVA
)
989 return X86EMUL_PROPAGATE_FAULT
;
990 pfn
= gpa
>> PAGE_SHIFT
;
991 page
= gfn_to_page(vcpu
->kvm
, pfn
);
993 return X86EMUL_UNHANDLEABLE
;
994 page_virt
= kmap_atomic(page
, KM_USER0
);
996 memcpy(data
, page_virt
+ offset
, tocopy
);
998 kunmap_atomic(page_virt
, KM_USER0
);
1005 return X86EMUL_CONTINUE
;
1008 static int emulator_write_std(unsigned long addr
,
1011 struct x86_emulate_ctxt
*ctxt
)
1013 printk(KERN_ERR
"emulator_write_std: addr %lx n %d\n",
1015 return X86EMUL_UNHANDLEABLE
;
1018 static int emulator_read_emulated(unsigned long addr
,
1021 struct x86_emulate_ctxt
*ctxt
)
1023 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1025 if (vcpu
->mmio_read_completed
) {
1026 memcpy(val
, vcpu
->mmio_data
, bytes
);
1027 vcpu
->mmio_read_completed
= 0;
1028 return X86EMUL_CONTINUE
;
1029 } else if (emulator_read_std(addr
, val
, bytes
, ctxt
)
1030 == X86EMUL_CONTINUE
)
1031 return X86EMUL_CONTINUE
;
1033 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
1035 if (gpa
== UNMAPPED_GVA
)
1036 return X86EMUL_PROPAGATE_FAULT
;
1037 vcpu
->mmio_needed
= 1;
1038 vcpu
->mmio_phys_addr
= gpa
;
1039 vcpu
->mmio_size
= bytes
;
1040 vcpu
->mmio_is_write
= 0;
1042 return X86EMUL_UNHANDLEABLE
;
1046 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1047 const void *val
, int bytes
)
1052 if (((gpa
+ bytes
- 1) >> PAGE_SHIFT
) != (gpa
>> PAGE_SHIFT
))
1054 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1057 kvm_mmu_pre_write(vcpu
, gpa
, bytes
);
1058 mark_page_dirty(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1059 virt
= kmap_atomic(page
, KM_USER0
);
1060 memcpy(virt
+ offset_in_page(gpa
), val
, bytes
);
1061 kunmap_atomic(virt
, KM_USER0
);
1062 kvm_mmu_post_write(vcpu
, gpa
, bytes
);
1066 static int emulator_write_emulated(unsigned long addr
,
1069 struct x86_emulate_ctxt
*ctxt
)
1071 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1072 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
1074 if (gpa
== UNMAPPED_GVA
) {
1075 kvm_arch_ops
->inject_page_fault(vcpu
, addr
, 2);
1076 return X86EMUL_PROPAGATE_FAULT
;
1079 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
1080 return X86EMUL_CONTINUE
;
1082 vcpu
->mmio_needed
= 1;
1083 vcpu
->mmio_phys_addr
= gpa
;
1084 vcpu
->mmio_size
= bytes
;
1085 vcpu
->mmio_is_write
= 1;
1086 memcpy(vcpu
->mmio_data
, val
, bytes
);
1088 return X86EMUL_CONTINUE
;
1091 static int emulator_cmpxchg_emulated(unsigned long addr
,
1095 struct x86_emulate_ctxt
*ctxt
)
1097 static int reported
;
1101 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1103 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
1106 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1108 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
1111 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1113 return X86EMUL_CONTINUE
;
1116 int emulate_clts(struct kvm_vcpu
*vcpu
)
1120 cr0
= vcpu
->cr0
& ~CR0_TS_MASK
;
1121 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
1122 return X86EMUL_CONTINUE
;
1125 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
1127 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1131 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
1132 return X86EMUL_CONTINUE
;
1134 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1136 return X86EMUL_UNHANDLEABLE
;
1140 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1142 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1145 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1147 /* FIXME: better handling */
1148 return X86EMUL_UNHANDLEABLE
;
1150 return X86EMUL_CONTINUE
;
1153 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1155 static int reported
;
1157 unsigned long rip
= ctxt
->vcpu
->rip
;
1158 unsigned long rip_linear
;
1160 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1165 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1167 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1168 " rip %lx %02x %02x %02x %02x\n",
1169 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1173 struct x86_emulate_ops emulate_ops
= {
1174 .read_std
= emulator_read_std
,
1175 .write_std
= emulator_write_std
,
1176 .read_emulated
= emulator_read_emulated
,
1177 .write_emulated
= emulator_write_emulated
,
1178 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1181 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1182 struct kvm_run
*run
,
1186 struct x86_emulate_ctxt emulate_ctxt
;
1190 vcpu
->mmio_fault_cr2
= cr2
;
1191 kvm_arch_ops
->cache_regs(vcpu
);
1193 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1195 emulate_ctxt
.vcpu
= vcpu
;
1196 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
1197 emulate_ctxt
.cr2
= cr2
;
1198 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1199 ? X86EMUL_MODE_REAL
: cs_l
1200 ? X86EMUL_MODE_PROT64
: cs_db
1201 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1203 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1204 emulate_ctxt
.cs_base
= 0;
1205 emulate_ctxt
.ds_base
= 0;
1206 emulate_ctxt
.es_base
= 0;
1207 emulate_ctxt
.ss_base
= 0;
1209 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1210 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1211 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1212 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1215 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1216 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1218 vcpu
->mmio_is_write
= 0;
1219 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1221 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1222 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1223 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1224 run
->mmio
.len
= vcpu
->mmio_size
;
1225 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1229 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1230 return EMULATE_DONE
;
1231 if (!vcpu
->mmio_needed
) {
1232 report_emulation_failure(&emulate_ctxt
);
1233 return EMULATE_FAIL
;
1235 return EMULATE_DO_MMIO
;
1238 kvm_arch_ops
->decache_regs(vcpu
);
1239 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1241 if (vcpu
->mmio_is_write
) {
1242 vcpu
->mmio_needed
= 0;
1243 return EMULATE_DO_MMIO
;
1246 return EMULATE_DONE
;
1248 EXPORT_SYMBOL_GPL(emulate_instruction
);
1250 int kvm_hypercall(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1252 unsigned long nr
, a0
, a1
, a2
, a3
, a4
, a5
, ret
;
1254 kvm_arch_ops
->cache_regs(vcpu
);
1256 #ifdef CONFIG_X86_64
1257 if (is_long_mode(vcpu
)) {
1258 nr
= vcpu
->regs
[VCPU_REGS_RAX
];
1259 a0
= vcpu
->regs
[VCPU_REGS_RDI
];
1260 a1
= vcpu
->regs
[VCPU_REGS_RSI
];
1261 a2
= vcpu
->regs
[VCPU_REGS_RDX
];
1262 a3
= vcpu
->regs
[VCPU_REGS_RCX
];
1263 a4
= vcpu
->regs
[VCPU_REGS_R8
];
1264 a5
= vcpu
->regs
[VCPU_REGS_R9
];
1268 nr
= vcpu
->regs
[VCPU_REGS_RBX
] & -1u;
1269 a0
= vcpu
->regs
[VCPU_REGS_RAX
] & -1u;
1270 a1
= vcpu
->regs
[VCPU_REGS_RCX
] & -1u;
1271 a2
= vcpu
->regs
[VCPU_REGS_RDX
] & -1u;
1272 a3
= vcpu
->regs
[VCPU_REGS_RSI
] & -1u;
1273 a4
= vcpu
->regs
[VCPU_REGS_RDI
] & -1u;
1274 a5
= vcpu
->regs
[VCPU_REGS_RBP
] & -1u;
1278 run
->hypercall
.args
[0] = a0
;
1279 run
->hypercall
.args
[1] = a1
;
1280 run
->hypercall
.args
[2] = a2
;
1281 run
->hypercall
.args
[3] = a3
;
1282 run
->hypercall
.args
[4] = a4
;
1283 run
->hypercall
.args
[5] = a5
;
1284 run
->hypercall
.ret
= ret
;
1285 run
->hypercall
.longmode
= is_long_mode(vcpu
);
1286 kvm_arch_ops
->decache_regs(vcpu
);
1289 vcpu
->regs
[VCPU_REGS_RAX
] = ret
;
1290 kvm_arch_ops
->decache_regs(vcpu
);
1293 EXPORT_SYMBOL_GPL(kvm_hypercall
);
1295 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1297 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1300 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1302 struct descriptor_table dt
= { limit
, base
};
1304 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1307 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1309 struct descriptor_table dt
= { limit
, base
};
1311 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1314 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1315 unsigned long *rflags
)
1318 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1321 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1323 kvm_arch_ops
->decache_cr4_guest_bits(vcpu
);
1334 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1339 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1340 unsigned long *rflags
)
1344 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1345 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1354 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1357 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1362 * Register the para guest with the host:
1364 static int vcpu_register_para(struct kvm_vcpu
*vcpu
, gpa_t para_state_gpa
)
1366 struct kvm_vcpu_para_state
*para_state
;
1367 hpa_t para_state_hpa
, hypercall_hpa
;
1368 struct page
*para_state_page
;
1369 unsigned char *hypercall
;
1370 gpa_t hypercall_gpa
;
1372 printk(KERN_DEBUG
"kvm: guest trying to enter paravirtual mode\n");
1373 printk(KERN_DEBUG
".... para_state_gpa: %08Lx\n", para_state_gpa
);
1376 * Needs to be page aligned:
1378 if (para_state_gpa
!= PAGE_ALIGN(para_state_gpa
))
1381 para_state_hpa
= gpa_to_hpa(vcpu
, para_state_gpa
);
1382 printk(KERN_DEBUG
".... para_state_hpa: %08Lx\n", para_state_hpa
);
1383 if (is_error_hpa(para_state_hpa
))
1386 mark_page_dirty(vcpu
->kvm
, para_state_gpa
>> PAGE_SHIFT
);
1387 para_state_page
= pfn_to_page(para_state_hpa
>> PAGE_SHIFT
);
1388 para_state
= kmap_atomic(para_state_page
, KM_USER0
);
1390 printk(KERN_DEBUG
".... guest version: %d\n", para_state
->guest_version
);
1391 printk(KERN_DEBUG
".... size: %d\n", para_state
->size
);
1393 para_state
->host_version
= KVM_PARA_API_VERSION
;
1395 * We cannot support guests that try to register themselves
1396 * with a newer API version than the host supports:
1398 if (para_state
->guest_version
> KVM_PARA_API_VERSION
) {
1399 para_state
->ret
= -KVM_EINVAL
;
1400 goto err_kunmap_skip
;
1403 hypercall_gpa
= para_state
->hypercall_gpa
;
1404 hypercall_hpa
= gpa_to_hpa(vcpu
, hypercall_gpa
);
1405 printk(KERN_DEBUG
".... hypercall_hpa: %08Lx\n", hypercall_hpa
);
1406 if (is_error_hpa(hypercall_hpa
)) {
1407 para_state
->ret
= -KVM_EINVAL
;
1408 goto err_kunmap_skip
;
1411 printk(KERN_DEBUG
"kvm: para guest successfully registered.\n");
1412 vcpu
->para_state_page
= para_state_page
;
1413 vcpu
->para_state_gpa
= para_state_gpa
;
1414 vcpu
->hypercall_gpa
= hypercall_gpa
;
1416 mark_page_dirty(vcpu
->kvm
, hypercall_gpa
>> PAGE_SHIFT
);
1417 hypercall
= kmap_atomic(pfn_to_page(hypercall_hpa
>> PAGE_SHIFT
),
1418 KM_USER1
) + (hypercall_hpa
& ~PAGE_MASK
);
1419 kvm_arch_ops
->patch_hypercall(vcpu
, hypercall
);
1420 kunmap_atomic(hypercall
, KM_USER1
);
1422 para_state
->ret
= 0;
1424 kunmap_atomic(para_state
, KM_USER0
);
1430 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1435 case 0xc0010010: /* SYSCFG */
1436 case 0xc0010015: /* HWCR */
1437 case MSR_IA32_PLATFORM_ID
:
1438 case MSR_IA32_P5_MC_ADDR
:
1439 case MSR_IA32_P5_MC_TYPE
:
1440 case MSR_IA32_MC0_CTL
:
1441 case MSR_IA32_MCG_STATUS
:
1442 case MSR_IA32_MCG_CAP
:
1443 case MSR_IA32_MC0_MISC
:
1444 case MSR_IA32_MC0_MISC
+4:
1445 case MSR_IA32_MC0_MISC
+8:
1446 case MSR_IA32_MC0_MISC
+12:
1447 case MSR_IA32_MC0_MISC
+16:
1448 case MSR_IA32_UCODE_REV
:
1449 case MSR_IA32_PERF_STATUS
:
1450 /* MTRR registers */
1452 case 0x200 ... 0x2ff:
1455 case 0xcd: /* fsb frequency */
1458 case MSR_IA32_APICBASE
:
1459 data
= vcpu
->apic_base
;
1461 case MSR_IA32_MISC_ENABLE
:
1462 data
= vcpu
->ia32_misc_enable_msr
;
1464 #ifdef CONFIG_X86_64
1466 data
= vcpu
->shadow_efer
;
1470 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1476 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1479 * Reads an msr value (of 'msr_index') into 'pdata'.
1480 * Returns 0 on success, non-0 otherwise.
1481 * Assumes vcpu_load() was already called.
1483 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1485 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1488 #ifdef CONFIG_X86_64
1490 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1492 if (efer
& EFER_RESERVED_BITS
) {
1493 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1500 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1501 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1506 kvm_arch_ops
->set_efer(vcpu
, efer
);
1509 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1511 vcpu
->shadow_efer
= efer
;
1516 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1519 #ifdef CONFIG_X86_64
1521 set_efer(vcpu
, data
);
1524 case MSR_IA32_MC0_STATUS
:
1525 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1526 __FUNCTION__
, data
);
1528 case MSR_IA32_MCG_STATUS
:
1529 printk(KERN_WARNING
"%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1530 __FUNCTION__
, data
);
1532 case MSR_IA32_UCODE_REV
:
1533 case MSR_IA32_UCODE_WRITE
:
1534 case 0x200 ... 0x2ff: /* MTRRs */
1536 case MSR_IA32_APICBASE
:
1537 vcpu
->apic_base
= data
;
1539 case MSR_IA32_MISC_ENABLE
:
1540 vcpu
->ia32_misc_enable_msr
= data
;
1543 * This is the 'probe whether the host is KVM' logic:
1545 case MSR_KVM_API_MAGIC
:
1546 return vcpu_register_para(vcpu
, data
);
1549 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1554 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1557 * Writes msr value into into the appropriate "register".
1558 * Returns 0 on success, non-0 otherwise.
1559 * Assumes vcpu_load() was already called.
1561 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1563 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1566 void kvm_resched(struct kvm_vcpu
*vcpu
)
1568 if (!need_resched())
1574 EXPORT_SYMBOL_GPL(kvm_resched
);
1576 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1580 for (i
= 0; i
< n
; ++i
)
1581 wrmsrl(e
[i
].index
, e
[i
].data
);
1583 EXPORT_SYMBOL_GPL(load_msrs
);
1585 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1589 for (i
= 0; i
< n
; ++i
)
1590 rdmsrl(e
[i
].index
, e
[i
].data
);
1592 EXPORT_SYMBOL_GPL(save_msrs
);
1594 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
1598 struct kvm_cpuid_entry
*e
, *best
;
1600 kvm_arch_ops
->cache_regs(vcpu
);
1601 function
= vcpu
->regs
[VCPU_REGS_RAX
];
1602 vcpu
->regs
[VCPU_REGS_RAX
] = 0;
1603 vcpu
->regs
[VCPU_REGS_RBX
] = 0;
1604 vcpu
->regs
[VCPU_REGS_RCX
] = 0;
1605 vcpu
->regs
[VCPU_REGS_RDX
] = 0;
1607 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
1608 e
= &vcpu
->cpuid_entries
[i
];
1609 if (e
->function
== function
) {
1614 * Both basic or both extended?
1616 if (((e
->function
^ function
) & 0x80000000) == 0)
1617 if (!best
|| e
->function
> best
->function
)
1621 vcpu
->regs
[VCPU_REGS_RAX
] = best
->eax
;
1622 vcpu
->regs
[VCPU_REGS_RBX
] = best
->ebx
;
1623 vcpu
->regs
[VCPU_REGS_RCX
] = best
->ecx
;
1624 vcpu
->regs
[VCPU_REGS_RDX
] = best
->edx
;
1626 kvm_arch_ops
->decache_regs(vcpu
);
1627 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1629 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
1631 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
1633 void *p
= vcpu
->pio_data
;
1636 int nr_pages
= vcpu
->pio
.guest_pages
[1] ? 2 : 1;
1638 kvm_arch_ops
->vcpu_put(vcpu
);
1639 q
= vmap(vcpu
->pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
1642 kvm_arch_ops
->vcpu_load(vcpu
);
1643 free_pio_guest_pages(vcpu
);
1646 q
+= vcpu
->pio
.guest_page_offset
;
1647 bytes
= vcpu
->pio
.size
* vcpu
->pio
.cur_count
;
1649 memcpy(q
, p
, bytes
);
1651 memcpy(p
, q
, bytes
);
1652 q
-= vcpu
->pio
.guest_page_offset
;
1654 kvm_arch_ops
->vcpu_load(vcpu
);
1655 free_pio_guest_pages(vcpu
);
1659 static int complete_pio(struct kvm_vcpu
*vcpu
)
1661 struct kvm_pio_request
*io
= &vcpu
->pio
;
1665 kvm_arch_ops
->cache_regs(vcpu
);
1669 memcpy(&vcpu
->regs
[VCPU_REGS_RAX
], vcpu
->pio_data
,
1673 r
= pio_copy_data(vcpu
);
1675 kvm_arch_ops
->cache_regs(vcpu
);
1682 delta
*= io
->cur_count
;
1684 * The size of the register should really depend on
1685 * current address size.
1687 vcpu
->regs
[VCPU_REGS_RCX
] -= delta
;
1693 vcpu
->regs
[VCPU_REGS_RDI
] += delta
;
1695 vcpu
->regs
[VCPU_REGS_RSI
] += delta
;
1698 kvm_arch_ops
->decache_regs(vcpu
);
1700 io
->count
-= io
->cur_count
;
1704 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1708 int kvm_setup_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
1709 int size
, unsigned long count
, int string
, int down
,
1710 gva_t address
, int rep
, unsigned port
)
1712 unsigned now
, in_page
;
1717 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
1718 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
1719 vcpu
->run
->io
.size
= size
;
1720 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
1721 vcpu
->run
->io
.count
= count
;
1722 vcpu
->run
->io
.port
= port
;
1723 vcpu
->pio
.count
= count
;
1724 vcpu
->pio
.cur_count
= count
;
1725 vcpu
->pio
.size
= size
;
1727 vcpu
->pio
.string
= string
;
1728 vcpu
->pio
.down
= down
;
1729 vcpu
->pio
.guest_page_offset
= offset_in_page(address
);
1730 vcpu
->pio
.rep
= rep
;
1733 kvm_arch_ops
->cache_regs(vcpu
);
1734 memcpy(vcpu
->pio_data
, &vcpu
->regs
[VCPU_REGS_RAX
], 4);
1735 kvm_arch_ops
->decache_regs(vcpu
);
1740 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1744 now
= min(count
, PAGE_SIZE
/ size
);
1747 in_page
= PAGE_SIZE
- offset_in_page(address
);
1749 in_page
= offset_in_page(address
) + size
;
1750 now
= min(count
, (unsigned long)in_page
/ size
);
1753 * String I/O straddles page boundary. Pin two guest pages
1754 * so that we satisfy atomicity constraints. Do just one
1755 * transaction to avoid complexity.
1762 * String I/O in reverse. Yuck. Kill the guest, fix later.
1764 printk(KERN_ERR
"kvm: guest string pio down\n");
1768 vcpu
->run
->io
.count
= now
;
1769 vcpu
->pio
.cur_count
= now
;
1771 for (i
= 0; i
< nr_pages
; ++i
) {
1772 spin_lock(&vcpu
->kvm
->lock
);
1773 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
1776 vcpu
->pio
.guest_pages
[i
] = page
;
1777 spin_unlock(&vcpu
->kvm
->lock
);
1780 free_pio_guest_pages(vcpu
);
1786 return pio_copy_data(vcpu
);
1789 EXPORT_SYMBOL_GPL(kvm_setup_pio
);
1791 static int kvm_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1798 if (vcpu
->sigset_active
)
1799 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1801 /* re-sync apic's tpr */
1802 vcpu
->cr8
= kvm_run
->cr8
;
1804 if (vcpu
->pio
.cur_count
) {
1805 r
= complete_pio(vcpu
);
1810 if (vcpu
->mmio_needed
) {
1811 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1812 vcpu
->mmio_read_completed
= 1;
1813 vcpu
->mmio_needed
= 0;
1814 r
= emulate_instruction(vcpu
, kvm_run
,
1815 vcpu
->mmio_fault_cr2
, 0);
1816 if (r
== EMULATE_DO_MMIO
) {
1818 * Read-modify-write. Back to userspace.
1820 kvm_run
->exit_reason
= KVM_EXIT_MMIO
;
1826 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
1827 kvm_arch_ops
->cache_regs(vcpu
);
1828 vcpu
->regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
1829 kvm_arch_ops
->decache_regs(vcpu
);
1832 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1835 if (vcpu
->sigset_active
)
1836 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1842 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
,
1843 struct kvm_regs
*regs
)
1847 kvm_arch_ops
->cache_regs(vcpu
);
1849 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1850 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1851 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1852 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1853 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1854 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1855 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1856 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1857 #ifdef CONFIG_X86_64
1858 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1859 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1860 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1861 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1862 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1863 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1864 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1865 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1868 regs
->rip
= vcpu
->rip
;
1869 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1872 * Don't leak debug flags in case they were set for guest debugging
1874 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1875 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1882 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
,
1883 struct kvm_regs
*regs
)
1887 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1888 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1889 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1890 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1891 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1892 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1893 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1894 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1895 #ifdef CONFIG_X86_64
1896 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1897 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1898 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1899 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1900 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1901 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1902 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1903 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1906 vcpu
->rip
= regs
->rip
;
1907 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1909 kvm_arch_ops
->decache_regs(vcpu
);
1916 static void get_segment(struct kvm_vcpu
*vcpu
,
1917 struct kvm_segment
*var
, int seg
)
1919 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1922 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
1923 struct kvm_sregs
*sregs
)
1925 struct descriptor_table dt
;
1929 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1930 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1931 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1932 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1933 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1934 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1936 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1937 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1939 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1940 sregs
->idt
.limit
= dt
.limit
;
1941 sregs
->idt
.base
= dt
.base
;
1942 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1943 sregs
->gdt
.limit
= dt
.limit
;
1944 sregs
->gdt
.base
= dt
.base
;
1946 kvm_arch_ops
->decache_cr4_guest_bits(vcpu
);
1947 sregs
->cr0
= vcpu
->cr0
;
1948 sregs
->cr2
= vcpu
->cr2
;
1949 sregs
->cr3
= vcpu
->cr3
;
1950 sregs
->cr4
= vcpu
->cr4
;
1951 sregs
->cr8
= vcpu
->cr8
;
1952 sregs
->efer
= vcpu
->shadow_efer
;
1953 sregs
->apic_base
= vcpu
->apic_base
;
1955 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1956 sizeof sregs
->interrupt_bitmap
);
1963 static void set_segment(struct kvm_vcpu
*vcpu
,
1964 struct kvm_segment
*var
, int seg
)
1966 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1969 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
1970 struct kvm_sregs
*sregs
)
1972 int mmu_reset_needed
= 0;
1974 struct descriptor_table dt
;
1978 dt
.limit
= sregs
->idt
.limit
;
1979 dt
.base
= sregs
->idt
.base
;
1980 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1981 dt
.limit
= sregs
->gdt
.limit
;
1982 dt
.base
= sregs
->gdt
.base
;
1983 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1985 vcpu
->cr2
= sregs
->cr2
;
1986 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
1987 vcpu
->cr3
= sregs
->cr3
;
1989 vcpu
->cr8
= sregs
->cr8
;
1991 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
1992 #ifdef CONFIG_X86_64
1993 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
1995 vcpu
->apic_base
= sregs
->apic_base
;
1997 kvm_arch_ops
->decache_cr4_guest_bits(vcpu
);
1999 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
2000 kvm_arch_ops
->set_cr0(vcpu
, sregs
->cr0
);
2002 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
2003 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
2004 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
2005 load_pdptrs(vcpu
, vcpu
->cr3
);
2007 if (mmu_reset_needed
)
2008 kvm_mmu_reset_context(vcpu
);
2010 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
2011 sizeof vcpu
->irq_pending
);
2012 vcpu
->irq_summary
= 0;
2013 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
2014 if (vcpu
->irq_pending
[i
])
2015 __set_bit(i
, &vcpu
->irq_summary
);
2017 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2018 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2019 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2020 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2021 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2022 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2024 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2025 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2033 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2034 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2036 * This list is modified at module load time to reflect the
2037 * capabilities of the host cpu.
2039 static u32 msrs_to_save
[] = {
2040 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
2042 #ifdef CONFIG_X86_64
2043 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
2045 MSR_IA32_TIME_STAMP_COUNTER
,
2048 static unsigned num_msrs_to_save
;
2050 static u32 emulated_msrs
[] = {
2051 MSR_IA32_MISC_ENABLE
,
2054 static __init
void kvm_init_msr_list(void)
2059 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2060 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2063 msrs_to_save
[j
] = msrs_to_save
[i
];
2066 num_msrs_to_save
= j
;
2070 * Adapt set_msr() to msr_io()'s calling convention
2072 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
2074 return set_msr(vcpu
, index
, *data
);
2078 * Read or write a bunch of msrs. All parameters are kernel addresses.
2080 * @return number of msrs set successfully.
2082 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2083 struct kvm_msr_entry
*entries
,
2084 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2085 unsigned index
, u64
*data
))
2091 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2092 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2101 * Read or write a bunch of msrs. Parameters are user addresses.
2103 * @return number of msrs set successfully.
2105 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2106 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2107 unsigned index
, u64
*data
),
2110 struct kvm_msrs msrs
;
2111 struct kvm_msr_entry
*entries
;
2116 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2120 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2124 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2125 entries
= vmalloc(size
);
2130 if (copy_from_user(entries
, user_msrs
->entries
, size
))
2133 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2138 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2150 * Translate a guest virtual address to a guest physical address.
2152 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
2153 struct kvm_translation
*tr
)
2155 unsigned long vaddr
= tr
->linear_address
;
2159 spin_lock(&vcpu
->kvm
->lock
);
2160 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
2161 tr
->physical_address
= gpa
;
2162 tr
->valid
= gpa
!= UNMAPPED_GVA
;
2165 spin_unlock(&vcpu
->kvm
->lock
);
2171 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2172 struct kvm_interrupt
*irq
)
2174 if (irq
->irq
< 0 || irq
->irq
>= 256)
2178 set_bit(irq
->irq
, vcpu
->irq_pending
);
2179 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
2186 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
2187 struct kvm_debug_guest
*dbg
)
2193 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
2200 static struct page
*kvm_vcpu_nopage(struct vm_area_struct
*vma
,
2201 unsigned long address
,
2204 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2205 unsigned long pgoff
;
2208 *type
= VM_FAULT_MINOR
;
2209 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2211 page
= virt_to_page(vcpu
->run
);
2212 else if (pgoff
== KVM_PIO_PAGE_OFFSET
)
2213 page
= virt_to_page(vcpu
->pio_data
);
2215 return NOPAGE_SIGBUS
;
2220 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
2221 .nopage
= kvm_vcpu_nopage
,
2224 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2226 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2230 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2232 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2234 fput(vcpu
->kvm
->filp
);
2238 static struct file_operations kvm_vcpu_fops
= {
2239 .release
= kvm_vcpu_release
,
2240 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2241 .compat_ioctl
= kvm_vcpu_ioctl
,
2242 .mmap
= kvm_vcpu_mmap
,
2246 * Allocates an inode for the vcpu.
2248 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2251 struct inode
*inode
;
2254 atomic_inc(&vcpu
->kvm
->filp
->f_count
);
2255 inode
= kvmfs_inode(&kvm_vcpu_fops
);
2256 if (IS_ERR(inode
)) {
2261 file
= kvmfs_file(inode
, vcpu
);
2267 r
= get_unused_fd();
2271 fd_install(fd
, file
);
2280 fput(vcpu
->kvm
->filp
);
2285 * Creates some virtual cpus. Good luck creating more than one.
2287 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
2290 struct kvm_vcpu
*vcpu
;
2297 vcpu
= &kvm
->vcpus
[n
];
2299 mutex_lock(&vcpu
->mutex
);
2302 mutex_unlock(&vcpu
->mutex
);
2306 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2310 vcpu
->run
= page_address(page
);
2312 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2316 vcpu
->pio_data
= page_address(page
);
2318 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
2320 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
2323 r
= kvm_arch_ops
->vcpu_create(vcpu
);
2325 goto out_free_vcpus
;
2327 r
= kvm_mmu_create(vcpu
);
2329 goto out_free_vcpus
;
2331 kvm_arch_ops
->vcpu_load(vcpu
);
2332 r
= kvm_mmu_setup(vcpu
);
2334 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
2338 goto out_free_vcpus
;
2340 r
= create_vcpu_fd(vcpu
);
2342 goto out_free_vcpus
;
2347 kvm_free_vcpu(vcpu
);
2349 free_page((unsigned long)vcpu
->run
);
2352 mutex_unlock(&vcpu
->mutex
);
2357 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
2358 struct kvm_cpuid
*cpuid
,
2359 struct kvm_cpuid_entry __user
*entries
)
2364 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2367 if (copy_from_user(&vcpu
->cpuid_entries
, entries
,
2368 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
2370 vcpu
->cpuid_nent
= cpuid
->nent
;
2377 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2380 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2381 vcpu
->sigset_active
= 1;
2382 vcpu
->sigset
= *sigset
;
2384 vcpu
->sigset_active
= 0;
2389 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2390 * we have asm/x86/processor.h
2401 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2402 #ifdef CONFIG_X86_64
2403 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2405 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2409 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2411 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2415 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
2416 fpu
->fcw
= fxsave
->cwd
;
2417 fpu
->fsw
= fxsave
->swd
;
2418 fpu
->ftwx
= fxsave
->twd
;
2419 fpu
->last_opcode
= fxsave
->fop
;
2420 fpu
->last_ip
= fxsave
->rip
;
2421 fpu
->last_dp
= fxsave
->rdp
;
2422 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
2429 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2431 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2435 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
2436 fxsave
->cwd
= fpu
->fcw
;
2437 fxsave
->swd
= fpu
->fsw
;
2438 fxsave
->twd
= fpu
->ftwx
;
2439 fxsave
->fop
= fpu
->last_opcode
;
2440 fxsave
->rip
= fpu
->last_ip
;
2441 fxsave
->rdp
= fpu
->last_dp
;
2442 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
2449 static long kvm_vcpu_ioctl(struct file
*filp
,
2450 unsigned int ioctl
, unsigned long arg
)
2452 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2453 void __user
*argp
= (void __user
*)arg
;
2461 r
= kvm_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2463 case KVM_GET_REGS
: {
2464 struct kvm_regs kvm_regs
;
2466 memset(&kvm_regs
, 0, sizeof kvm_regs
);
2467 r
= kvm_vcpu_ioctl_get_regs(vcpu
, &kvm_regs
);
2471 if (copy_to_user(argp
, &kvm_regs
, sizeof kvm_regs
))
2476 case KVM_SET_REGS
: {
2477 struct kvm_regs kvm_regs
;
2480 if (copy_from_user(&kvm_regs
, argp
, sizeof kvm_regs
))
2482 r
= kvm_vcpu_ioctl_set_regs(vcpu
, &kvm_regs
);
2488 case KVM_GET_SREGS
: {
2489 struct kvm_sregs kvm_sregs
;
2491 memset(&kvm_sregs
, 0, sizeof kvm_sregs
);
2492 r
= kvm_vcpu_ioctl_get_sregs(vcpu
, &kvm_sregs
);
2496 if (copy_to_user(argp
, &kvm_sregs
, sizeof kvm_sregs
))
2501 case KVM_SET_SREGS
: {
2502 struct kvm_sregs kvm_sregs
;
2505 if (copy_from_user(&kvm_sregs
, argp
, sizeof kvm_sregs
))
2507 r
= kvm_vcpu_ioctl_set_sregs(vcpu
, &kvm_sregs
);
2513 case KVM_TRANSLATE
: {
2514 struct kvm_translation tr
;
2517 if (copy_from_user(&tr
, argp
, sizeof tr
))
2519 r
= kvm_vcpu_ioctl_translate(vcpu
, &tr
);
2523 if (copy_to_user(argp
, &tr
, sizeof tr
))
2528 case KVM_INTERRUPT
: {
2529 struct kvm_interrupt irq
;
2532 if (copy_from_user(&irq
, argp
, sizeof irq
))
2534 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2540 case KVM_DEBUG_GUEST
: {
2541 struct kvm_debug_guest dbg
;
2544 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2546 r
= kvm_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
2553 r
= msr_io(vcpu
, argp
, get_msr
, 1);
2556 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2558 case KVM_SET_CPUID
: {
2559 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2560 struct kvm_cpuid cpuid
;
2563 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2565 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2570 case KVM_SET_SIGNAL_MASK
: {
2571 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2572 struct kvm_signal_mask kvm_sigmask
;
2573 sigset_t sigset
, *p
;
2578 if (copy_from_user(&kvm_sigmask
, argp
,
2579 sizeof kvm_sigmask
))
2582 if (kvm_sigmask
.len
!= sizeof sigset
)
2585 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2590 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2596 memset(&fpu
, 0, sizeof fpu
);
2597 r
= kvm_vcpu_ioctl_get_fpu(vcpu
, &fpu
);
2601 if (copy_to_user(argp
, &fpu
, sizeof fpu
))
2610 if (copy_from_user(&fpu
, argp
, sizeof fpu
))
2612 r
= kvm_vcpu_ioctl_set_fpu(vcpu
, &fpu
);
2625 static long kvm_vm_ioctl(struct file
*filp
,
2626 unsigned int ioctl
, unsigned long arg
)
2628 struct kvm
*kvm
= filp
->private_data
;
2629 void __user
*argp
= (void __user
*)arg
;
2633 case KVM_CREATE_VCPU
:
2634 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2638 case KVM_SET_MEMORY_REGION
: {
2639 struct kvm_memory_region kvm_mem
;
2642 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2644 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_mem
);
2649 case KVM_GET_DIRTY_LOG
: {
2650 struct kvm_dirty_log log
;
2653 if (copy_from_user(&log
, argp
, sizeof log
))
2655 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2660 case KVM_SET_MEMORY_ALIAS
: {
2661 struct kvm_memory_alias alias
;
2664 if (copy_from_user(&alias
, argp
, sizeof alias
))
2666 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &alias
);
2678 static struct page
*kvm_vm_nopage(struct vm_area_struct
*vma
,
2679 unsigned long address
,
2682 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2683 unsigned long pgoff
;
2686 *type
= VM_FAULT_MINOR
;
2687 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2688 page
= gfn_to_page(kvm
, pgoff
);
2690 return NOPAGE_SIGBUS
;
2695 static struct vm_operations_struct kvm_vm_vm_ops
= {
2696 .nopage
= kvm_vm_nopage
,
2699 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2701 vma
->vm_ops
= &kvm_vm_vm_ops
;
2705 static struct file_operations kvm_vm_fops
= {
2706 .release
= kvm_vm_release
,
2707 .unlocked_ioctl
= kvm_vm_ioctl
,
2708 .compat_ioctl
= kvm_vm_ioctl
,
2709 .mmap
= kvm_vm_mmap
,
2712 static int kvm_dev_ioctl_create_vm(void)
2715 struct inode
*inode
;
2719 inode
= kvmfs_inode(&kvm_vm_fops
);
2720 if (IS_ERR(inode
)) {
2725 kvm
= kvm_create_vm();
2731 file
= kvmfs_file(inode
, kvm
);
2738 r
= get_unused_fd();
2742 fd_install(fd
, file
);
2749 kvm_destroy_vm(kvm
);
2756 static long kvm_dev_ioctl(struct file
*filp
,
2757 unsigned int ioctl
, unsigned long arg
)
2759 void __user
*argp
= (void __user
*)arg
;
2763 case KVM_GET_API_VERSION
:
2767 r
= KVM_API_VERSION
;
2773 r
= kvm_dev_ioctl_create_vm();
2775 case KVM_GET_MSR_INDEX_LIST
: {
2776 struct kvm_msr_list __user
*user_msr_list
= argp
;
2777 struct kvm_msr_list msr_list
;
2781 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2784 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2785 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2788 if (n
< num_msrs_to_save
)
2791 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2792 num_msrs_to_save
* sizeof(u32
)))
2794 if (copy_to_user(user_msr_list
->indices
2795 + num_msrs_to_save
* sizeof(u32
),
2797 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2802 case KVM_CHECK_EXTENSION
:
2804 * No extensions defined at present.
2808 case KVM_GET_VCPU_MMAP_SIZE
:
2821 static struct file_operations kvm_chardev_ops
= {
2822 .open
= kvm_dev_open
,
2823 .release
= kvm_dev_release
,
2824 .unlocked_ioctl
= kvm_dev_ioctl
,
2825 .compat_ioctl
= kvm_dev_ioctl
,
2828 static struct miscdevice kvm_dev
= {
2834 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2837 if (val
== SYS_RESTART
) {
2839 * Some (well, at least mine) BIOSes hang on reboot if
2842 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2843 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2848 static struct notifier_block kvm_reboot_notifier
= {
2849 .notifier_call
= kvm_reboot
,
2854 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2857 static void decache_vcpus_on_cpu(int cpu
)
2860 struct kvm_vcpu
*vcpu
;
2863 spin_lock(&kvm_lock
);
2864 list_for_each_entry(vm
, &vm_list
, vm_list
)
2865 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2866 vcpu
= &vm
->vcpus
[i
];
2868 * If the vcpu is locked, then it is running on some
2869 * other cpu and therefore it is not cached on the
2872 * If it's not locked, check the last cpu it executed
2875 if (mutex_trylock(&vcpu
->mutex
)) {
2876 if (vcpu
->cpu
== cpu
) {
2877 kvm_arch_ops
->vcpu_decache(vcpu
);
2880 mutex_unlock(&vcpu
->mutex
);
2883 spin_unlock(&kvm_lock
);
2886 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2892 case CPU_DOWN_PREPARE
:
2893 case CPU_DOWN_PREPARE_FROZEN
:
2894 case CPU_UP_CANCELED
:
2895 case CPU_UP_CANCELED_FROZEN
:
2896 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2898 decache_vcpus_on_cpu(cpu
);
2899 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_disable
,
2903 case CPU_ONLINE_FROZEN
:
2904 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2906 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_enable
,
2913 static struct notifier_block kvm_cpu_notifier
= {
2914 .notifier_call
= kvm_cpu_hotplug
,
2915 .priority
= 20, /* must be > scheduler priority */
2918 static u64
stat_get(void *_offset
)
2920 unsigned offset
= (long)_offset
;
2923 struct kvm_vcpu
*vcpu
;
2926 spin_lock(&kvm_lock
);
2927 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2928 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2929 vcpu
= &kvm
->vcpus
[i
];
2930 total
+= *(u32
*)((void *)vcpu
+ offset
);
2932 spin_unlock(&kvm_lock
);
2936 static void stat_set(void *offset
, u64 val
)
2940 DEFINE_SIMPLE_ATTRIBUTE(stat_fops
, stat_get
, stat_set
, "%llu\n");
2942 static __init
void kvm_init_debug(void)
2944 struct kvm_stats_debugfs_item
*p
;
2946 debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2947 for (p
= debugfs_entries
; p
->name
; ++p
)
2948 p
->dentry
= debugfs_create_file(p
->name
, 0444, debugfs_dir
,
2949 (void *)(long)p
->offset
,
2953 static void kvm_exit_debug(void)
2955 struct kvm_stats_debugfs_item
*p
;
2957 for (p
= debugfs_entries
; p
->name
; ++p
)
2958 debugfs_remove(p
->dentry
);
2959 debugfs_remove(debugfs_dir
);
2962 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2964 decache_vcpus_on_cpu(raw_smp_processor_id());
2965 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2969 static int kvm_resume(struct sys_device
*dev
)
2971 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
2975 static struct sysdev_class kvm_sysdev_class
= {
2976 set_kset_name("kvm"),
2977 .suspend
= kvm_suspend
,
2978 .resume
= kvm_resume
,
2981 static struct sys_device kvm_sysdev
= {
2983 .cls
= &kvm_sysdev_class
,
2986 hpa_t bad_page_address
;
2988 static int kvmfs_get_sb(struct file_system_type
*fs_type
, int flags
,
2989 const char *dev_name
, void *data
, struct vfsmount
*mnt
)
2991 return get_sb_pseudo(fs_type
, "kvm:", NULL
, KVMFS_SUPER_MAGIC
, mnt
);
2994 static struct file_system_type kvm_fs_type
= {
2996 .get_sb
= kvmfs_get_sb
,
2997 .kill_sb
= kill_anon_super
,
3000 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
3005 printk(KERN_ERR
"kvm: already loaded the other module\n");
3009 if (!ops
->cpu_has_kvm_support()) {
3010 printk(KERN_ERR
"kvm: no hardware support\n");
3013 if (ops
->disabled_by_bios()) {
3014 printk(KERN_ERR
"kvm: disabled by bios\n");
3020 r
= kvm_arch_ops
->hardware_setup();
3024 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
3025 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3028 register_reboot_notifier(&kvm_reboot_notifier
);
3030 r
= sysdev_class_register(&kvm_sysdev_class
);
3034 r
= sysdev_register(&kvm_sysdev
);
3038 kvm_chardev_ops
.owner
= module
;
3040 r
= misc_register(&kvm_dev
);
3042 printk (KERN_ERR
"kvm: misc device register failed\n");
3049 sysdev_unregister(&kvm_sysdev
);
3051 sysdev_class_unregister(&kvm_sysdev_class
);
3053 unregister_reboot_notifier(&kvm_reboot_notifier
);
3054 unregister_cpu_notifier(&kvm_cpu_notifier
);
3056 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
3057 kvm_arch_ops
->hardware_unsetup();
3059 kvm_arch_ops
= NULL
;
3063 void kvm_exit_arch(void)
3065 misc_deregister(&kvm_dev
);
3066 sysdev_unregister(&kvm_sysdev
);
3067 sysdev_class_unregister(&kvm_sysdev_class
);
3068 unregister_reboot_notifier(&kvm_reboot_notifier
);
3069 unregister_cpu_notifier(&kvm_cpu_notifier
);
3070 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
3071 kvm_arch_ops
->hardware_unsetup();
3072 kvm_arch_ops
= NULL
;
3075 static __init
int kvm_init(void)
3077 static struct page
*bad_page
;
3080 r
= kvm_mmu_module_init();
3084 r
= register_filesystem(&kvm_fs_type
);
3088 kvmfs_mnt
= kern_mount(&kvm_fs_type
);
3089 r
= PTR_ERR(kvmfs_mnt
);
3090 if (IS_ERR(kvmfs_mnt
))
3094 kvm_init_msr_list();
3096 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
3101 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
3102 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
3110 unregister_filesystem(&kvm_fs_type
);
3112 kvm_mmu_module_exit();
3117 static __exit
void kvm_exit(void)
3120 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
3122 unregister_filesystem(&kvm_fs_type
);
3123 kvm_mmu_module_exit();
3126 module_init(kvm_init
)
3127 module_exit(kvm_exit
)
3129 EXPORT_SYMBOL_GPL(kvm_init_arch
);
3130 EXPORT_SYMBOL_GPL(kvm_exit_arch
);