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 free_pio_guest_pages(struct kvm_vcpu
*vcpu
)
353 for (i
= 0; i
< 2; ++i
)
354 if (vcpu
->pio
.guest_pages
[i
]) {
355 __free_page(vcpu
->pio
.guest_pages
[i
]);
356 vcpu
->pio
.guest_pages
[i
] = NULL
;
360 static void kvm_free_vcpu(struct kvm_vcpu
*vcpu
)
366 kvm_mmu_destroy(vcpu
);
368 kvm_arch_ops
->vcpu_free(vcpu
);
369 free_page((unsigned long)vcpu
->run
);
371 free_page((unsigned long)vcpu
->pio_data
);
372 vcpu
->pio_data
= NULL
;
373 free_pio_guest_pages(vcpu
);
376 static void kvm_free_vcpus(struct kvm
*kvm
)
380 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
381 kvm_free_vcpu(&kvm
->vcpus
[i
]);
384 static int kvm_dev_release(struct inode
*inode
, struct file
*filp
)
389 static void kvm_destroy_vm(struct kvm
*kvm
)
391 spin_lock(&kvm_lock
);
392 list_del(&kvm
->vm_list
);
393 spin_unlock(&kvm_lock
);
395 kvm_free_physmem(kvm
);
399 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
401 struct kvm
*kvm
= filp
->private_data
;
407 static void inject_gp(struct kvm_vcpu
*vcpu
)
409 kvm_arch_ops
->inject_gp(vcpu
, 0);
413 * Load the pae pdptrs. Return true is they are all valid.
415 static int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
417 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
418 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
425 spin_lock(&vcpu
->kvm
->lock
);
426 page
= gfn_to_page(vcpu
->kvm
, pdpt_gfn
);
427 /* FIXME: !page - emulate? 0xff? */
428 pdpt
= kmap_atomic(page
, KM_USER0
);
431 for (i
= 0; i
< 4; ++i
) {
432 pdpte
= pdpt
[offset
+ i
];
433 if ((pdpte
& 1) && (pdpte
& 0xfffffff0000001e6ull
)) {
439 for (i
= 0; i
< 4; ++i
)
440 vcpu
->pdptrs
[i
] = pdpt
[offset
+ i
];
443 kunmap_atomic(pdpt
, KM_USER0
);
444 spin_unlock(&vcpu
->kvm
->lock
);
449 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
451 if (cr0
& CR0_RESEVED_BITS
) {
452 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
458 if ((cr0
& CR0_NW_MASK
) && !(cr0
& CR0_CD_MASK
)) {
459 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
464 if ((cr0
& CR0_PG_MASK
) && !(cr0
& CR0_PE_MASK
)) {
465 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
466 "and a clear PE flag\n");
471 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
473 if ((vcpu
->shadow_efer
& EFER_LME
)) {
477 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
478 "in long mode while PAE is disabled\n");
482 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
484 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
485 "in long mode while CS.L == 1\n");
492 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
493 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
501 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
504 spin_lock(&vcpu
->kvm
->lock
);
505 kvm_mmu_reset_context(vcpu
);
506 spin_unlock(&vcpu
->kvm
->lock
);
509 EXPORT_SYMBOL_GPL(set_cr0
);
511 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
513 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
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 unsigned long 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 return X86EMUL_PROPAGATE_FAULT
;
1077 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
1078 return X86EMUL_CONTINUE
;
1080 vcpu
->mmio_needed
= 1;
1081 vcpu
->mmio_phys_addr
= gpa
;
1082 vcpu
->mmio_size
= bytes
;
1083 vcpu
->mmio_is_write
= 1;
1084 memcpy(vcpu
->mmio_data
, &val
, bytes
);
1086 return X86EMUL_CONTINUE
;
1089 static int emulator_cmpxchg_emulated(unsigned long addr
,
1093 struct x86_emulate_ctxt
*ctxt
)
1095 static int reported
;
1099 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1101 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
1104 #ifdef CONFIG_X86_32
1106 static int emulator_cmpxchg8b_emulated(unsigned long addr
,
1107 unsigned long old_lo
,
1108 unsigned long old_hi
,
1109 unsigned long new_lo
,
1110 unsigned long new_hi
,
1111 struct x86_emulate_ctxt
*ctxt
)
1113 static int reported
;
1118 printk(KERN_WARNING
"kvm: emulating exchange8b as write\n");
1120 r
= emulator_write_emulated(addr
, new_lo
, 4, ctxt
);
1121 if (r
!= X86EMUL_CONTINUE
)
1123 return emulator_write_emulated(addr
+4, new_hi
, 4, ctxt
);
1128 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1130 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
1133 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1135 return X86EMUL_CONTINUE
;
1138 int emulate_clts(struct kvm_vcpu
*vcpu
)
1142 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1143 cr0
= vcpu
->cr0
& ~CR0_TS_MASK
;
1144 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
1145 return X86EMUL_CONTINUE
;
1148 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
1150 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1154 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
1155 return X86EMUL_CONTINUE
;
1157 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1159 return X86EMUL_UNHANDLEABLE
;
1163 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1165 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1168 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1170 /* FIXME: better handling */
1171 return X86EMUL_UNHANDLEABLE
;
1173 return X86EMUL_CONTINUE
;
1176 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1178 static int reported
;
1180 unsigned long rip
= ctxt
->vcpu
->rip
;
1181 unsigned long rip_linear
;
1183 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1188 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1190 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1191 " rip %lx %02x %02x %02x %02x\n",
1192 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1196 struct x86_emulate_ops emulate_ops
= {
1197 .read_std
= emulator_read_std
,
1198 .write_std
= emulator_write_std
,
1199 .read_emulated
= emulator_read_emulated
,
1200 .write_emulated
= emulator_write_emulated
,
1201 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1202 #ifdef CONFIG_X86_32
1203 .cmpxchg8b_emulated
= emulator_cmpxchg8b_emulated
,
1207 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1208 struct kvm_run
*run
,
1212 struct x86_emulate_ctxt emulate_ctxt
;
1216 kvm_arch_ops
->cache_regs(vcpu
);
1218 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1220 emulate_ctxt
.vcpu
= vcpu
;
1221 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
1222 emulate_ctxt
.cr2
= cr2
;
1223 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1224 ? X86EMUL_MODE_REAL
: cs_l
1225 ? X86EMUL_MODE_PROT64
: cs_db
1226 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1228 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1229 emulate_ctxt
.cs_base
= 0;
1230 emulate_ctxt
.ds_base
= 0;
1231 emulate_ctxt
.es_base
= 0;
1232 emulate_ctxt
.ss_base
= 0;
1234 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1235 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1236 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1237 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1240 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1241 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1243 vcpu
->mmio_is_write
= 0;
1244 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1246 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1247 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1248 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1249 run
->mmio
.len
= vcpu
->mmio_size
;
1250 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1254 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1255 return EMULATE_DONE
;
1256 if (!vcpu
->mmio_needed
) {
1257 report_emulation_failure(&emulate_ctxt
);
1258 return EMULATE_FAIL
;
1260 return EMULATE_DO_MMIO
;
1263 kvm_arch_ops
->decache_regs(vcpu
);
1264 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1266 if (vcpu
->mmio_is_write
)
1267 return EMULATE_DO_MMIO
;
1269 return EMULATE_DONE
;
1271 EXPORT_SYMBOL_GPL(emulate_instruction
);
1273 int kvm_hypercall(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1275 unsigned long nr
, a0
, a1
, a2
, a3
, a4
, a5
, ret
;
1277 kvm_arch_ops
->cache_regs(vcpu
);
1279 #ifdef CONFIG_X86_64
1280 if (is_long_mode(vcpu
)) {
1281 nr
= vcpu
->regs
[VCPU_REGS_RAX
];
1282 a0
= vcpu
->regs
[VCPU_REGS_RDI
];
1283 a1
= vcpu
->regs
[VCPU_REGS_RSI
];
1284 a2
= vcpu
->regs
[VCPU_REGS_RDX
];
1285 a3
= vcpu
->regs
[VCPU_REGS_RCX
];
1286 a4
= vcpu
->regs
[VCPU_REGS_R8
];
1287 a5
= vcpu
->regs
[VCPU_REGS_R9
];
1291 nr
= vcpu
->regs
[VCPU_REGS_RBX
] & -1u;
1292 a0
= vcpu
->regs
[VCPU_REGS_RAX
] & -1u;
1293 a1
= vcpu
->regs
[VCPU_REGS_RCX
] & -1u;
1294 a2
= vcpu
->regs
[VCPU_REGS_RDX
] & -1u;
1295 a3
= vcpu
->regs
[VCPU_REGS_RSI
] & -1u;
1296 a4
= vcpu
->regs
[VCPU_REGS_RDI
] & -1u;
1297 a5
= vcpu
->regs
[VCPU_REGS_RBP
] & -1u;
1301 run
->hypercall
.args
[0] = a0
;
1302 run
->hypercall
.args
[1] = a1
;
1303 run
->hypercall
.args
[2] = a2
;
1304 run
->hypercall
.args
[3] = a3
;
1305 run
->hypercall
.args
[4] = a4
;
1306 run
->hypercall
.args
[5] = a5
;
1307 run
->hypercall
.ret
= ret
;
1308 run
->hypercall
.longmode
= is_long_mode(vcpu
);
1309 kvm_arch_ops
->decache_regs(vcpu
);
1312 vcpu
->regs
[VCPU_REGS_RAX
] = ret
;
1313 kvm_arch_ops
->decache_regs(vcpu
);
1316 EXPORT_SYMBOL_GPL(kvm_hypercall
);
1318 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1320 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1323 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1325 struct descriptor_table dt
= { limit
, base
};
1327 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1330 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1332 struct descriptor_table dt
= { limit
, base
};
1334 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1337 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1338 unsigned long *rflags
)
1341 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1344 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1346 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1357 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1362 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1363 unsigned long *rflags
)
1367 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1368 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1377 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1380 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1385 * Register the para guest with the host:
1387 static int vcpu_register_para(struct kvm_vcpu
*vcpu
, gpa_t para_state_gpa
)
1389 struct kvm_vcpu_para_state
*para_state
;
1390 hpa_t para_state_hpa
, hypercall_hpa
;
1391 struct page
*para_state_page
;
1392 unsigned char *hypercall
;
1393 gpa_t hypercall_gpa
;
1395 printk(KERN_DEBUG
"kvm: guest trying to enter paravirtual mode\n");
1396 printk(KERN_DEBUG
".... para_state_gpa: %08Lx\n", para_state_gpa
);
1399 * Needs to be page aligned:
1401 if (para_state_gpa
!= PAGE_ALIGN(para_state_gpa
))
1404 para_state_hpa
= gpa_to_hpa(vcpu
, para_state_gpa
);
1405 printk(KERN_DEBUG
".... para_state_hpa: %08Lx\n", para_state_hpa
);
1406 if (is_error_hpa(para_state_hpa
))
1409 mark_page_dirty(vcpu
->kvm
, para_state_gpa
>> PAGE_SHIFT
);
1410 para_state_page
= pfn_to_page(para_state_hpa
>> PAGE_SHIFT
);
1411 para_state
= kmap_atomic(para_state_page
, KM_USER0
);
1413 printk(KERN_DEBUG
".... guest version: %d\n", para_state
->guest_version
);
1414 printk(KERN_DEBUG
".... size: %d\n", para_state
->size
);
1416 para_state
->host_version
= KVM_PARA_API_VERSION
;
1418 * We cannot support guests that try to register themselves
1419 * with a newer API version than the host supports:
1421 if (para_state
->guest_version
> KVM_PARA_API_VERSION
) {
1422 para_state
->ret
= -KVM_EINVAL
;
1423 goto err_kunmap_skip
;
1426 hypercall_gpa
= para_state
->hypercall_gpa
;
1427 hypercall_hpa
= gpa_to_hpa(vcpu
, hypercall_gpa
);
1428 printk(KERN_DEBUG
".... hypercall_hpa: %08Lx\n", hypercall_hpa
);
1429 if (is_error_hpa(hypercall_hpa
)) {
1430 para_state
->ret
= -KVM_EINVAL
;
1431 goto err_kunmap_skip
;
1434 printk(KERN_DEBUG
"kvm: para guest successfully registered.\n");
1435 vcpu
->para_state_page
= para_state_page
;
1436 vcpu
->para_state_gpa
= para_state_gpa
;
1437 vcpu
->hypercall_gpa
= hypercall_gpa
;
1439 mark_page_dirty(vcpu
->kvm
, hypercall_gpa
>> PAGE_SHIFT
);
1440 hypercall
= kmap_atomic(pfn_to_page(hypercall_hpa
>> PAGE_SHIFT
),
1441 KM_USER1
) + (hypercall_hpa
& ~PAGE_MASK
);
1442 kvm_arch_ops
->patch_hypercall(vcpu
, hypercall
);
1443 kunmap_atomic(hypercall
, KM_USER1
);
1445 para_state
->ret
= 0;
1447 kunmap_atomic(para_state
, KM_USER0
);
1453 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1458 case 0xc0010010: /* SYSCFG */
1459 case 0xc0010015: /* HWCR */
1460 case MSR_IA32_PLATFORM_ID
:
1461 case MSR_IA32_P5_MC_ADDR
:
1462 case MSR_IA32_P5_MC_TYPE
:
1463 case MSR_IA32_MC0_CTL
:
1464 case MSR_IA32_MCG_STATUS
:
1465 case MSR_IA32_MCG_CAP
:
1466 case MSR_IA32_MC0_MISC
:
1467 case MSR_IA32_MC0_MISC
+4:
1468 case MSR_IA32_MC0_MISC
+8:
1469 case MSR_IA32_MC0_MISC
+12:
1470 case MSR_IA32_MC0_MISC
+16:
1471 case MSR_IA32_UCODE_REV
:
1472 case MSR_IA32_PERF_STATUS
:
1473 /* MTRR registers */
1475 case 0x200 ... 0x2ff:
1478 case 0xcd: /* fsb frequency */
1481 case MSR_IA32_APICBASE
:
1482 data
= vcpu
->apic_base
;
1484 case MSR_IA32_MISC_ENABLE
:
1485 data
= vcpu
->ia32_misc_enable_msr
;
1487 #ifdef CONFIG_X86_64
1489 data
= vcpu
->shadow_efer
;
1493 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1499 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1502 * Reads an msr value (of 'msr_index') into 'pdata'.
1503 * Returns 0 on success, non-0 otherwise.
1504 * Assumes vcpu_load() was already called.
1506 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1508 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1511 #ifdef CONFIG_X86_64
1513 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1515 if (efer
& EFER_RESERVED_BITS
) {
1516 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1523 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1524 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1529 kvm_arch_ops
->set_efer(vcpu
, efer
);
1532 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1534 vcpu
->shadow_efer
= efer
;
1539 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1542 #ifdef CONFIG_X86_64
1544 set_efer(vcpu
, data
);
1547 case MSR_IA32_MC0_STATUS
:
1548 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1549 __FUNCTION__
, data
);
1551 case MSR_IA32_MCG_STATUS
:
1552 printk(KERN_WARNING
"%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1553 __FUNCTION__
, data
);
1555 case MSR_IA32_UCODE_REV
:
1556 case MSR_IA32_UCODE_WRITE
:
1557 case 0x200 ... 0x2ff: /* MTRRs */
1559 case MSR_IA32_APICBASE
:
1560 vcpu
->apic_base
= data
;
1562 case MSR_IA32_MISC_ENABLE
:
1563 vcpu
->ia32_misc_enable_msr
= data
;
1566 * This is the 'probe whether the host is KVM' logic:
1568 case MSR_KVM_API_MAGIC
:
1569 return vcpu_register_para(vcpu
, data
);
1572 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1577 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1580 * Writes msr value into into the appropriate "register".
1581 * Returns 0 on success, non-0 otherwise.
1582 * Assumes vcpu_load() was already called.
1584 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1586 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1589 void kvm_resched(struct kvm_vcpu
*vcpu
)
1595 EXPORT_SYMBOL_GPL(kvm_resched
);
1597 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1601 for (i
= 0; i
< n
; ++i
)
1602 wrmsrl(e
[i
].index
, e
[i
].data
);
1604 EXPORT_SYMBOL_GPL(load_msrs
);
1606 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1610 for (i
= 0; i
< n
; ++i
)
1611 rdmsrl(e
[i
].index
, e
[i
].data
);
1613 EXPORT_SYMBOL_GPL(save_msrs
);
1615 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
1619 struct kvm_cpuid_entry
*e
, *best
;
1621 kvm_arch_ops
->cache_regs(vcpu
);
1622 function
= vcpu
->regs
[VCPU_REGS_RAX
];
1623 vcpu
->regs
[VCPU_REGS_RAX
] = 0;
1624 vcpu
->regs
[VCPU_REGS_RBX
] = 0;
1625 vcpu
->regs
[VCPU_REGS_RCX
] = 0;
1626 vcpu
->regs
[VCPU_REGS_RDX
] = 0;
1628 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
1629 e
= &vcpu
->cpuid_entries
[i
];
1630 if (e
->function
== function
) {
1635 * Both basic or both extended?
1637 if (((e
->function
^ function
) & 0x80000000) == 0)
1638 if (!best
|| e
->function
> best
->function
)
1642 vcpu
->regs
[VCPU_REGS_RAX
] = best
->eax
;
1643 vcpu
->regs
[VCPU_REGS_RBX
] = best
->ebx
;
1644 vcpu
->regs
[VCPU_REGS_RCX
] = best
->ecx
;
1645 vcpu
->regs
[VCPU_REGS_RDX
] = best
->edx
;
1647 kvm_arch_ops
->decache_regs(vcpu
);
1648 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1650 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
1652 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
1654 void *p
= vcpu
->pio_data
;
1657 int nr_pages
= vcpu
->pio
.guest_pages
[1] ? 2 : 1;
1659 kvm_arch_ops
->vcpu_put(vcpu
);
1660 q
= vmap(vcpu
->pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
1663 kvm_arch_ops
->vcpu_load(vcpu
);
1664 free_pio_guest_pages(vcpu
);
1667 q
+= vcpu
->pio
.guest_page_offset
;
1668 bytes
= vcpu
->pio
.size
* vcpu
->pio
.cur_count
;
1670 memcpy(q
, p
, bytes
);
1672 memcpy(p
, q
, bytes
);
1673 q
-= vcpu
->pio
.guest_page_offset
;
1675 kvm_arch_ops
->vcpu_load(vcpu
);
1676 free_pio_guest_pages(vcpu
);
1680 static int complete_pio(struct kvm_vcpu
*vcpu
)
1682 struct kvm_pio_request
*io
= &vcpu
->pio
;
1686 kvm_arch_ops
->cache_regs(vcpu
);
1690 memcpy(&vcpu
->regs
[VCPU_REGS_RAX
], vcpu
->pio_data
,
1694 r
= pio_copy_data(vcpu
);
1696 kvm_arch_ops
->cache_regs(vcpu
);
1703 delta
*= io
->cur_count
;
1705 * The size of the register should really depend on
1706 * current address size.
1708 vcpu
->regs
[VCPU_REGS_RCX
] -= delta
;
1714 vcpu
->regs
[VCPU_REGS_RDI
] += delta
;
1716 vcpu
->regs
[VCPU_REGS_RSI
] += delta
;
1719 vcpu
->run
->io_completed
= 0;
1721 kvm_arch_ops
->decache_regs(vcpu
);
1723 io
->count
-= io
->cur_count
;
1727 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1731 int kvm_setup_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
1732 int size
, unsigned long count
, int string
, int down
,
1733 gva_t address
, int rep
, unsigned port
)
1735 unsigned now
, in_page
;
1740 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
1741 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
1742 vcpu
->run
->io
.size
= size
;
1743 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
1744 vcpu
->run
->io
.count
= count
;
1745 vcpu
->run
->io
.port
= port
;
1746 vcpu
->pio
.count
= count
;
1747 vcpu
->pio
.cur_count
= count
;
1748 vcpu
->pio
.size
= size
;
1750 vcpu
->pio
.string
= string
;
1751 vcpu
->pio
.down
= down
;
1752 vcpu
->pio
.guest_page_offset
= offset_in_page(address
);
1753 vcpu
->pio
.rep
= rep
;
1756 kvm_arch_ops
->cache_regs(vcpu
);
1757 memcpy(vcpu
->pio_data
, &vcpu
->regs
[VCPU_REGS_RAX
], 4);
1758 kvm_arch_ops
->decache_regs(vcpu
);
1763 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1767 now
= min(count
, PAGE_SIZE
/ size
);
1770 in_page
= PAGE_SIZE
- offset_in_page(address
);
1772 in_page
= offset_in_page(address
) + size
;
1773 now
= min(count
, (unsigned long)in_page
/ size
);
1776 * String I/O straddles page boundary. Pin two guest pages
1777 * so that we satisfy atomicity constraints. Do just one
1778 * transaction to avoid complexity.
1785 * String I/O in reverse. Yuck. Kill the guest, fix later.
1787 printk(KERN_ERR
"kvm: guest string pio down\n");
1791 vcpu
->run
->io
.count
= now
;
1792 vcpu
->pio
.cur_count
= now
;
1794 for (i
= 0; i
< nr_pages
; ++i
) {
1795 spin_lock(&vcpu
->kvm
->lock
);
1796 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
1799 vcpu
->pio
.guest_pages
[i
] = page
;
1800 spin_unlock(&vcpu
->kvm
->lock
);
1803 free_pio_guest_pages(vcpu
);
1809 return pio_copy_data(vcpu
);
1812 EXPORT_SYMBOL_GPL(kvm_setup_pio
);
1814 static int kvm_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1821 if (vcpu
->sigset_active
)
1822 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1824 /* re-sync apic's tpr */
1825 vcpu
->cr8
= kvm_run
->cr8
;
1827 if (kvm_run
->io_completed
) {
1828 if (vcpu
->pio
.cur_count
) {
1829 r
= complete_pio(vcpu
);
1833 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1834 vcpu
->mmio_read_completed
= 1;
1838 vcpu
->mmio_needed
= 0;
1840 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
1841 kvm_arch_ops
->cache_regs(vcpu
);
1842 vcpu
->regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
1843 kvm_arch_ops
->decache_regs(vcpu
);
1846 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1849 if (vcpu
->sigset_active
)
1850 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1856 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
,
1857 struct kvm_regs
*regs
)
1861 kvm_arch_ops
->cache_regs(vcpu
);
1863 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1864 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1865 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1866 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1867 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1868 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1869 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1870 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1871 #ifdef CONFIG_X86_64
1872 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1873 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1874 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1875 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1876 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1877 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1878 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1879 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1882 regs
->rip
= vcpu
->rip
;
1883 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1886 * Don't leak debug flags in case they were set for guest debugging
1888 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1889 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1896 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
,
1897 struct kvm_regs
*regs
)
1901 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1902 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1903 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1904 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1905 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1906 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1907 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1908 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1909 #ifdef CONFIG_X86_64
1910 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1911 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1912 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1913 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1914 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1915 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1916 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1917 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1920 vcpu
->rip
= regs
->rip
;
1921 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1923 kvm_arch_ops
->decache_regs(vcpu
);
1930 static void get_segment(struct kvm_vcpu
*vcpu
,
1931 struct kvm_segment
*var
, int seg
)
1933 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1936 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
1937 struct kvm_sregs
*sregs
)
1939 struct descriptor_table dt
;
1943 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1944 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1945 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1946 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1947 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1948 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1950 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1951 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1953 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1954 sregs
->idt
.limit
= dt
.limit
;
1955 sregs
->idt
.base
= dt
.base
;
1956 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1957 sregs
->gdt
.limit
= dt
.limit
;
1958 sregs
->gdt
.base
= dt
.base
;
1960 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1961 sregs
->cr0
= vcpu
->cr0
;
1962 sregs
->cr2
= vcpu
->cr2
;
1963 sregs
->cr3
= vcpu
->cr3
;
1964 sregs
->cr4
= vcpu
->cr4
;
1965 sregs
->cr8
= vcpu
->cr8
;
1966 sregs
->efer
= vcpu
->shadow_efer
;
1967 sregs
->apic_base
= vcpu
->apic_base
;
1969 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1970 sizeof sregs
->interrupt_bitmap
);
1977 static void set_segment(struct kvm_vcpu
*vcpu
,
1978 struct kvm_segment
*var
, int seg
)
1980 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1983 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
1984 struct kvm_sregs
*sregs
)
1986 int mmu_reset_needed
= 0;
1988 struct descriptor_table dt
;
1992 dt
.limit
= sregs
->idt
.limit
;
1993 dt
.base
= sregs
->idt
.base
;
1994 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1995 dt
.limit
= sregs
->gdt
.limit
;
1996 dt
.base
= sregs
->gdt
.base
;
1997 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1999 vcpu
->cr2
= sregs
->cr2
;
2000 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
2001 vcpu
->cr3
= sregs
->cr3
;
2003 vcpu
->cr8
= sregs
->cr8
;
2005 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
2006 #ifdef CONFIG_X86_64
2007 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
2009 vcpu
->apic_base
= sregs
->apic_base
;
2011 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
2013 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
2014 kvm_arch_ops
->set_cr0(vcpu
, sregs
->cr0
);
2016 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
2017 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
2018 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
2019 load_pdptrs(vcpu
, vcpu
->cr3
);
2021 if (mmu_reset_needed
)
2022 kvm_mmu_reset_context(vcpu
);
2024 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
2025 sizeof vcpu
->irq_pending
);
2026 vcpu
->irq_summary
= 0;
2027 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
2028 if (vcpu
->irq_pending
[i
])
2029 __set_bit(i
, &vcpu
->irq_summary
);
2031 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2032 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2033 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2034 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2035 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2036 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2038 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2039 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2047 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2048 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2050 * This list is modified at module load time to reflect the
2051 * capabilities of the host cpu.
2053 static u32 msrs_to_save
[] = {
2054 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
2056 #ifdef CONFIG_X86_64
2057 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
2059 MSR_IA32_TIME_STAMP_COUNTER
,
2062 static unsigned num_msrs_to_save
;
2064 static u32 emulated_msrs
[] = {
2065 MSR_IA32_MISC_ENABLE
,
2068 static __init
void kvm_init_msr_list(void)
2073 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2074 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2077 msrs_to_save
[j
] = msrs_to_save
[i
];
2080 num_msrs_to_save
= j
;
2084 * Adapt set_msr() to msr_io()'s calling convention
2086 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
2088 return set_msr(vcpu
, index
, *data
);
2092 * Read or write a bunch of msrs. All parameters are kernel addresses.
2094 * @return number of msrs set successfully.
2096 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2097 struct kvm_msr_entry
*entries
,
2098 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2099 unsigned index
, u64
*data
))
2105 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2106 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2115 * Read or write a bunch of msrs. Parameters are user addresses.
2117 * @return number of msrs set successfully.
2119 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2120 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2121 unsigned index
, u64
*data
),
2124 struct kvm_msrs msrs
;
2125 struct kvm_msr_entry
*entries
;
2130 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2134 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2138 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2139 entries
= vmalloc(size
);
2144 if (copy_from_user(entries
, user_msrs
->entries
, size
))
2147 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2152 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2164 * Translate a guest virtual address to a guest physical address.
2166 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
2167 struct kvm_translation
*tr
)
2169 unsigned long vaddr
= tr
->linear_address
;
2173 spin_lock(&vcpu
->kvm
->lock
);
2174 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
2175 tr
->physical_address
= gpa
;
2176 tr
->valid
= gpa
!= UNMAPPED_GVA
;
2179 spin_unlock(&vcpu
->kvm
->lock
);
2185 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2186 struct kvm_interrupt
*irq
)
2188 if (irq
->irq
< 0 || irq
->irq
>= 256)
2192 set_bit(irq
->irq
, vcpu
->irq_pending
);
2193 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
2200 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
2201 struct kvm_debug_guest
*dbg
)
2207 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
2214 static struct page
*kvm_vcpu_nopage(struct vm_area_struct
*vma
,
2215 unsigned long address
,
2218 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2219 unsigned long pgoff
;
2222 *type
= VM_FAULT_MINOR
;
2223 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2225 page
= virt_to_page(vcpu
->run
);
2226 else if (pgoff
== KVM_PIO_PAGE_OFFSET
)
2227 page
= virt_to_page(vcpu
->pio_data
);
2229 return NOPAGE_SIGBUS
;
2234 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
2235 .nopage
= kvm_vcpu_nopage
,
2238 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2240 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2244 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2246 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2248 fput(vcpu
->kvm
->filp
);
2252 static struct file_operations kvm_vcpu_fops
= {
2253 .release
= kvm_vcpu_release
,
2254 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2255 .compat_ioctl
= kvm_vcpu_ioctl
,
2256 .mmap
= kvm_vcpu_mmap
,
2260 * Allocates an inode for the vcpu.
2262 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2265 struct inode
*inode
;
2268 atomic_inc(&vcpu
->kvm
->filp
->f_count
);
2269 inode
= kvmfs_inode(&kvm_vcpu_fops
);
2270 if (IS_ERR(inode
)) {
2275 file
= kvmfs_file(inode
, vcpu
);
2281 r
= get_unused_fd();
2285 fd_install(fd
, file
);
2294 fput(vcpu
->kvm
->filp
);
2299 * Creates some virtual cpus. Good luck creating more than one.
2301 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
2304 struct kvm_vcpu
*vcpu
;
2311 vcpu
= &kvm
->vcpus
[n
];
2313 mutex_lock(&vcpu
->mutex
);
2316 mutex_unlock(&vcpu
->mutex
);
2320 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2324 vcpu
->run
= page_address(page
);
2326 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2330 vcpu
->pio_data
= page_address(page
);
2332 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
2334 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
2337 r
= kvm_arch_ops
->vcpu_create(vcpu
);
2339 goto out_free_vcpus
;
2341 r
= kvm_mmu_create(vcpu
);
2343 goto out_free_vcpus
;
2345 kvm_arch_ops
->vcpu_load(vcpu
);
2346 r
= kvm_mmu_setup(vcpu
);
2348 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
2352 goto out_free_vcpus
;
2354 r
= create_vcpu_fd(vcpu
);
2356 goto out_free_vcpus
;
2361 kvm_free_vcpu(vcpu
);
2363 free_page((unsigned long)vcpu
->run
);
2366 mutex_unlock(&vcpu
->mutex
);
2371 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
2372 struct kvm_cpuid
*cpuid
,
2373 struct kvm_cpuid_entry __user
*entries
)
2378 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2381 if (copy_from_user(&vcpu
->cpuid_entries
, entries
,
2382 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
2384 vcpu
->cpuid_nent
= cpuid
->nent
;
2391 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2394 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2395 vcpu
->sigset_active
= 1;
2396 vcpu
->sigset
= *sigset
;
2398 vcpu
->sigset_active
= 0;
2403 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2404 * we have asm/x86/processor.h
2415 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2416 #ifdef CONFIG_X86_64
2417 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2419 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2423 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2425 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2429 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
2430 fpu
->fcw
= fxsave
->cwd
;
2431 fpu
->fsw
= fxsave
->swd
;
2432 fpu
->ftwx
= fxsave
->twd
;
2433 fpu
->last_opcode
= fxsave
->fop
;
2434 fpu
->last_ip
= fxsave
->rip
;
2435 fpu
->last_dp
= fxsave
->rdp
;
2436 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
2443 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2445 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2449 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
2450 fxsave
->cwd
= fpu
->fcw
;
2451 fxsave
->swd
= fpu
->fsw
;
2452 fxsave
->twd
= fpu
->ftwx
;
2453 fxsave
->fop
= fpu
->last_opcode
;
2454 fxsave
->rip
= fpu
->last_ip
;
2455 fxsave
->rdp
= fpu
->last_dp
;
2456 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
2463 static long kvm_vcpu_ioctl(struct file
*filp
,
2464 unsigned int ioctl
, unsigned long arg
)
2466 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2467 void __user
*argp
= (void __user
*)arg
;
2475 r
= kvm_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2477 case KVM_GET_REGS
: {
2478 struct kvm_regs kvm_regs
;
2480 memset(&kvm_regs
, 0, sizeof kvm_regs
);
2481 r
= kvm_vcpu_ioctl_get_regs(vcpu
, &kvm_regs
);
2485 if (copy_to_user(argp
, &kvm_regs
, sizeof kvm_regs
))
2490 case KVM_SET_REGS
: {
2491 struct kvm_regs kvm_regs
;
2494 if (copy_from_user(&kvm_regs
, argp
, sizeof kvm_regs
))
2496 r
= kvm_vcpu_ioctl_set_regs(vcpu
, &kvm_regs
);
2502 case KVM_GET_SREGS
: {
2503 struct kvm_sregs kvm_sregs
;
2505 memset(&kvm_sregs
, 0, sizeof kvm_sregs
);
2506 r
= kvm_vcpu_ioctl_get_sregs(vcpu
, &kvm_sregs
);
2510 if (copy_to_user(argp
, &kvm_sregs
, sizeof kvm_sregs
))
2515 case KVM_SET_SREGS
: {
2516 struct kvm_sregs kvm_sregs
;
2519 if (copy_from_user(&kvm_sregs
, argp
, sizeof kvm_sregs
))
2521 r
= kvm_vcpu_ioctl_set_sregs(vcpu
, &kvm_sregs
);
2527 case KVM_TRANSLATE
: {
2528 struct kvm_translation tr
;
2531 if (copy_from_user(&tr
, argp
, sizeof tr
))
2533 r
= kvm_vcpu_ioctl_translate(vcpu
, &tr
);
2537 if (copy_to_user(argp
, &tr
, sizeof tr
))
2542 case KVM_INTERRUPT
: {
2543 struct kvm_interrupt irq
;
2546 if (copy_from_user(&irq
, argp
, sizeof irq
))
2548 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2554 case KVM_DEBUG_GUEST
: {
2555 struct kvm_debug_guest dbg
;
2558 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2560 r
= kvm_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
2567 r
= msr_io(vcpu
, argp
, get_msr
, 1);
2570 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2572 case KVM_SET_CPUID
: {
2573 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2574 struct kvm_cpuid cpuid
;
2577 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2579 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2584 case KVM_SET_SIGNAL_MASK
: {
2585 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2586 struct kvm_signal_mask kvm_sigmask
;
2587 sigset_t sigset
, *p
;
2592 if (copy_from_user(&kvm_sigmask
, argp
,
2593 sizeof kvm_sigmask
))
2596 if (kvm_sigmask
.len
!= sizeof sigset
)
2599 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2604 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2610 memset(&fpu
, 0, sizeof fpu
);
2611 r
= kvm_vcpu_ioctl_get_fpu(vcpu
, &fpu
);
2615 if (copy_to_user(argp
, &fpu
, sizeof fpu
))
2624 if (copy_from_user(&fpu
, argp
, sizeof fpu
))
2626 r
= kvm_vcpu_ioctl_set_fpu(vcpu
, &fpu
);
2639 static long kvm_vm_ioctl(struct file
*filp
,
2640 unsigned int ioctl
, unsigned long arg
)
2642 struct kvm
*kvm
= filp
->private_data
;
2643 void __user
*argp
= (void __user
*)arg
;
2647 case KVM_CREATE_VCPU
:
2648 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2652 case KVM_SET_MEMORY_REGION
: {
2653 struct kvm_memory_region kvm_mem
;
2656 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2658 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_mem
);
2663 case KVM_GET_DIRTY_LOG
: {
2664 struct kvm_dirty_log log
;
2667 if (copy_from_user(&log
, argp
, sizeof log
))
2669 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2674 case KVM_SET_MEMORY_ALIAS
: {
2675 struct kvm_memory_alias alias
;
2678 if (copy_from_user(&alias
, argp
, sizeof alias
))
2680 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &alias
);
2692 static struct page
*kvm_vm_nopage(struct vm_area_struct
*vma
,
2693 unsigned long address
,
2696 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2697 unsigned long pgoff
;
2700 *type
= VM_FAULT_MINOR
;
2701 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2702 page
= gfn_to_page(kvm
, pgoff
);
2704 return NOPAGE_SIGBUS
;
2709 static struct vm_operations_struct kvm_vm_vm_ops
= {
2710 .nopage
= kvm_vm_nopage
,
2713 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2715 vma
->vm_ops
= &kvm_vm_vm_ops
;
2719 static struct file_operations kvm_vm_fops
= {
2720 .release
= kvm_vm_release
,
2721 .unlocked_ioctl
= kvm_vm_ioctl
,
2722 .compat_ioctl
= kvm_vm_ioctl
,
2723 .mmap
= kvm_vm_mmap
,
2726 static int kvm_dev_ioctl_create_vm(void)
2729 struct inode
*inode
;
2733 inode
= kvmfs_inode(&kvm_vm_fops
);
2734 if (IS_ERR(inode
)) {
2739 kvm
= kvm_create_vm();
2745 file
= kvmfs_file(inode
, kvm
);
2752 r
= get_unused_fd();
2756 fd_install(fd
, file
);
2763 kvm_destroy_vm(kvm
);
2770 static long kvm_dev_ioctl(struct file
*filp
,
2771 unsigned int ioctl
, unsigned long arg
)
2773 void __user
*argp
= (void __user
*)arg
;
2777 case KVM_GET_API_VERSION
:
2781 r
= KVM_API_VERSION
;
2787 r
= kvm_dev_ioctl_create_vm();
2789 case KVM_GET_MSR_INDEX_LIST
: {
2790 struct kvm_msr_list __user
*user_msr_list
= argp
;
2791 struct kvm_msr_list msr_list
;
2795 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2798 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2799 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2802 if (n
< num_msrs_to_save
)
2805 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2806 num_msrs_to_save
* sizeof(u32
)))
2808 if (copy_to_user(user_msr_list
->indices
2809 + num_msrs_to_save
* sizeof(u32
),
2811 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2816 case KVM_CHECK_EXTENSION
:
2818 * No extensions defined at present.
2822 case KVM_GET_VCPU_MMAP_SIZE
:
2835 static struct file_operations kvm_chardev_ops
= {
2836 .open
= kvm_dev_open
,
2837 .release
= kvm_dev_release
,
2838 .unlocked_ioctl
= kvm_dev_ioctl
,
2839 .compat_ioctl
= kvm_dev_ioctl
,
2842 static struct miscdevice kvm_dev
= {
2848 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2851 if (val
== SYS_RESTART
) {
2853 * Some (well, at least mine) BIOSes hang on reboot if
2856 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2857 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2862 static struct notifier_block kvm_reboot_notifier
= {
2863 .notifier_call
= kvm_reboot
,
2868 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2871 static void decache_vcpus_on_cpu(int cpu
)
2874 struct kvm_vcpu
*vcpu
;
2877 spin_lock(&kvm_lock
);
2878 list_for_each_entry(vm
, &vm_list
, vm_list
)
2879 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2880 vcpu
= &vm
->vcpus
[i
];
2882 * If the vcpu is locked, then it is running on some
2883 * other cpu and therefore it is not cached on the
2886 * If it's not locked, check the last cpu it executed
2889 if (mutex_trylock(&vcpu
->mutex
)) {
2890 if (vcpu
->cpu
== cpu
) {
2891 kvm_arch_ops
->vcpu_decache(vcpu
);
2894 mutex_unlock(&vcpu
->mutex
);
2897 spin_unlock(&kvm_lock
);
2900 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2906 case CPU_DOWN_PREPARE
:
2907 case CPU_UP_CANCELED
:
2908 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2910 decache_vcpus_on_cpu(cpu
);
2911 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_disable
,
2915 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2917 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_enable
,
2924 static struct notifier_block kvm_cpu_notifier
= {
2925 .notifier_call
= kvm_cpu_hotplug
,
2926 .priority
= 20, /* must be > scheduler priority */
2929 static __init
void kvm_init_debug(void)
2931 struct kvm_stats_debugfs_item
*p
;
2933 debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2934 for (p
= debugfs_entries
; p
->name
; ++p
)
2935 p
->dentry
= debugfs_create_u32(p
->name
, 0444, debugfs_dir
,
2939 static void kvm_exit_debug(void)
2941 struct kvm_stats_debugfs_item
*p
;
2943 for (p
= debugfs_entries
; p
->name
; ++p
)
2944 debugfs_remove(p
->dentry
);
2945 debugfs_remove(debugfs_dir
);
2948 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2950 decache_vcpus_on_cpu(raw_smp_processor_id());
2951 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2955 static int kvm_resume(struct sys_device
*dev
)
2957 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
2961 static struct sysdev_class kvm_sysdev_class
= {
2962 set_kset_name("kvm"),
2963 .suspend
= kvm_suspend
,
2964 .resume
= kvm_resume
,
2967 static struct sys_device kvm_sysdev
= {
2969 .cls
= &kvm_sysdev_class
,
2972 hpa_t bad_page_address
;
2974 static int kvmfs_get_sb(struct file_system_type
*fs_type
, int flags
,
2975 const char *dev_name
, void *data
, struct vfsmount
*mnt
)
2977 return get_sb_pseudo(fs_type
, "kvm:", NULL
, KVMFS_SUPER_MAGIC
, mnt
);
2980 static struct file_system_type kvm_fs_type
= {
2982 .get_sb
= kvmfs_get_sb
,
2983 .kill_sb
= kill_anon_super
,
2986 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
2991 printk(KERN_ERR
"kvm: already loaded the other module\n");
2995 if (!ops
->cpu_has_kvm_support()) {
2996 printk(KERN_ERR
"kvm: no hardware support\n");
2999 if (ops
->disabled_by_bios()) {
3000 printk(KERN_ERR
"kvm: disabled by bios\n");
3006 r
= kvm_arch_ops
->hardware_setup();
3010 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
3011 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3014 register_reboot_notifier(&kvm_reboot_notifier
);
3016 r
= sysdev_class_register(&kvm_sysdev_class
);
3020 r
= sysdev_register(&kvm_sysdev
);
3024 kvm_chardev_ops
.owner
= module
;
3026 r
= misc_register(&kvm_dev
);
3028 printk (KERN_ERR
"kvm: misc device register failed\n");
3035 sysdev_unregister(&kvm_sysdev
);
3037 sysdev_class_unregister(&kvm_sysdev_class
);
3039 unregister_reboot_notifier(&kvm_reboot_notifier
);
3040 unregister_cpu_notifier(&kvm_cpu_notifier
);
3042 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
3043 kvm_arch_ops
->hardware_unsetup();
3045 kvm_arch_ops
= NULL
;
3049 void kvm_exit_arch(void)
3051 misc_deregister(&kvm_dev
);
3052 sysdev_unregister(&kvm_sysdev
);
3053 sysdev_class_unregister(&kvm_sysdev_class
);
3054 unregister_reboot_notifier(&kvm_reboot_notifier
);
3055 unregister_cpu_notifier(&kvm_cpu_notifier
);
3056 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
3057 kvm_arch_ops
->hardware_unsetup();
3058 kvm_arch_ops
= NULL
;
3061 static __init
int kvm_init(void)
3063 static struct page
*bad_page
;
3066 r
= register_filesystem(&kvm_fs_type
);
3070 kvmfs_mnt
= kern_mount(&kvm_fs_type
);
3071 r
= PTR_ERR(kvmfs_mnt
);
3072 if (IS_ERR(kvmfs_mnt
))
3076 kvm_init_msr_list();
3078 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
3083 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
3084 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
3092 unregister_filesystem(&kvm_fs_type
);
3097 static __exit
void kvm_exit(void)
3100 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
3102 unregister_filesystem(&kvm_fs_type
);
3105 module_init(kvm_init
)
3106 module_exit(kvm_exit
)
3108 EXPORT_SYMBOL_GPL(kvm_init_arch
);
3109 EXPORT_SYMBOL_GPL(kvm_exit_arch
);