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
;
55 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
57 static struct kvm_stats_debugfs_item
{
60 struct dentry
*dentry
;
61 } debugfs_entries
[] = {
62 { "pf_fixed", STAT_OFFSET(pf_fixed
) },
63 { "pf_guest", STAT_OFFSET(pf_guest
) },
64 { "tlb_flush", STAT_OFFSET(tlb_flush
) },
65 { "invlpg", STAT_OFFSET(invlpg
) },
66 { "exits", STAT_OFFSET(exits
) },
67 { "io_exits", STAT_OFFSET(io_exits
) },
68 { "mmio_exits", STAT_OFFSET(mmio_exits
) },
69 { "signal_exits", STAT_OFFSET(signal_exits
) },
70 { "irq_window", STAT_OFFSET(irq_window_exits
) },
71 { "halt_exits", STAT_OFFSET(halt_exits
) },
72 { "request_irq", STAT_OFFSET(request_irq_exits
) },
73 { "irq_exits", STAT_OFFSET(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 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 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1121 cr0
= vcpu
->cr0
& ~CR0_TS_MASK
;
1122 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
1123 return X86EMUL_CONTINUE
;
1126 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
1128 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1132 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
1133 return X86EMUL_CONTINUE
;
1135 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1137 return X86EMUL_UNHANDLEABLE
;
1141 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1143 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1146 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1148 /* FIXME: better handling */
1149 return X86EMUL_UNHANDLEABLE
;
1151 return X86EMUL_CONTINUE
;
1154 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1156 static int reported
;
1158 unsigned long rip
= ctxt
->vcpu
->rip
;
1159 unsigned long rip_linear
;
1161 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1166 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1168 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1169 " rip %lx %02x %02x %02x %02x\n",
1170 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1174 struct x86_emulate_ops emulate_ops
= {
1175 .read_std
= emulator_read_std
,
1176 .write_std
= emulator_write_std
,
1177 .read_emulated
= emulator_read_emulated
,
1178 .write_emulated
= emulator_write_emulated
,
1179 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1182 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1183 struct kvm_run
*run
,
1187 struct x86_emulate_ctxt emulate_ctxt
;
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 return EMULATE_DO_MMIO
;
1244 return EMULATE_DONE
;
1246 EXPORT_SYMBOL_GPL(emulate_instruction
);
1248 int kvm_hypercall(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1250 unsigned long nr
, a0
, a1
, a2
, a3
, a4
, a5
, ret
;
1252 kvm_arch_ops
->cache_regs(vcpu
);
1254 #ifdef CONFIG_X86_64
1255 if (is_long_mode(vcpu
)) {
1256 nr
= vcpu
->regs
[VCPU_REGS_RAX
];
1257 a0
= vcpu
->regs
[VCPU_REGS_RDI
];
1258 a1
= vcpu
->regs
[VCPU_REGS_RSI
];
1259 a2
= vcpu
->regs
[VCPU_REGS_RDX
];
1260 a3
= vcpu
->regs
[VCPU_REGS_RCX
];
1261 a4
= vcpu
->regs
[VCPU_REGS_R8
];
1262 a5
= vcpu
->regs
[VCPU_REGS_R9
];
1266 nr
= vcpu
->regs
[VCPU_REGS_RBX
] & -1u;
1267 a0
= vcpu
->regs
[VCPU_REGS_RAX
] & -1u;
1268 a1
= vcpu
->regs
[VCPU_REGS_RCX
] & -1u;
1269 a2
= vcpu
->regs
[VCPU_REGS_RDX
] & -1u;
1270 a3
= vcpu
->regs
[VCPU_REGS_RSI
] & -1u;
1271 a4
= vcpu
->regs
[VCPU_REGS_RDI
] & -1u;
1272 a5
= vcpu
->regs
[VCPU_REGS_RBP
] & -1u;
1276 run
->hypercall
.args
[0] = a0
;
1277 run
->hypercall
.args
[1] = a1
;
1278 run
->hypercall
.args
[2] = a2
;
1279 run
->hypercall
.args
[3] = a3
;
1280 run
->hypercall
.args
[4] = a4
;
1281 run
->hypercall
.args
[5] = a5
;
1282 run
->hypercall
.ret
= ret
;
1283 run
->hypercall
.longmode
= is_long_mode(vcpu
);
1284 kvm_arch_ops
->decache_regs(vcpu
);
1287 vcpu
->regs
[VCPU_REGS_RAX
] = ret
;
1288 kvm_arch_ops
->decache_regs(vcpu
);
1291 EXPORT_SYMBOL_GPL(kvm_hypercall
);
1293 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1295 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1298 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1300 struct descriptor_table dt
= { limit
, base
};
1302 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1305 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1307 struct descriptor_table dt
= { limit
, base
};
1309 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1312 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1313 unsigned long *rflags
)
1316 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1319 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1321 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1332 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1337 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1338 unsigned long *rflags
)
1342 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1343 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1352 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1355 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1360 * Register the para guest with the host:
1362 static int vcpu_register_para(struct kvm_vcpu
*vcpu
, gpa_t para_state_gpa
)
1364 struct kvm_vcpu_para_state
*para_state
;
1365 hpa_t para_state_hpa
, hypercall_hpa
;
1366 struct page
*para_state_page
;
1367 unsigned char *hypercall
;
1368 gpa_t hypercall_gpa
;
1370 printk(KERN_DEBUG
"kvm: guest trying to enter paravirtual mode\n");
1371 printk(KERN_DEBUG
".... para_state_gpa: %08Lx\n", para_state_gpa
);
1374 * Needs to be page aligned:
1376 if (para_state_gpa
!= PAGE_ALIGN(para_state_gpa
))
1379 para_state_hpa
= gpa_to_hpa(vcpu
, para_state_gpa
);
1380 printk(KERN_DEBUG
".... para_state_hpa: %08Lx\n", para_state_hpa
);
1381 if (is_error_hpa(para_state_hpa
))
1384 mark_page_dirty(vcpu
->kvm
, para_state_gpa
>> PAGE_SHIFT
);
1385 para_state_page
= pfn_to_page(para_state_hpa
>> PAGE_SHIFT
);
1386 para_state
= kmap_atomic(para_state_page
, KM_USER0
);
1388 printk(KERN_DEBUG
".... guest version: %d\n", para_state
->guest_version
);
1389 printk(KERN_DEBUG
".... size: %d\n", para_state
->size
);
1391 para_state
->host_version
= KVM_PARA_API_VERSION
;
1393 * We cannot support guests that try to register themselves
1394 * with a newer API version than the host supports:
1396 if (para_state
->guest_version
> KVM_PARA_API_VERSION
) {
1397 para_state
->ret
= -KVM_EINVAL
;
1398 goto err_kunmap_skip
;
1401 hypercall_gpa
= para_state
->hypercall_gpa
;
1402 hypercall_hpa
= gpa_to_hpa(vcpu
, hypercall_gpa
);
1403 printk(KERN_DEBUG
".... hypercall_hpa: %08Lx\n", hypercall_hpa
);
1404 if (is_error_hpa(hypercall_hpa
)) {
1405 para_state
->ret
= -KVM_EINVAL
;
1406 goto err_kunmap_skip
;
1409 printk(KERN_DEBUG
"kvm: para guest successfully registered.\n");
1410 vcpu
->para_state_page
= para_state_page
;
1411 vcpu
->para_state_gpa
= para_state_gpa
;
1412 vcpu
->hypercall_gpa
= hypercall_gpa
;
1414 mark_page_dirty(vcpu
->kvm
, hypercall_gpa
>> PAGE_SHIFT
);
1415 hypercall
= kmap_atomic(pfn_to_page(hypercall_hpa
>> PAGE_SHIFT
),
1416 KM_USER1
) + (hypercall_hpa
& ~PAGE_MASK
);
1417 kvm_arch_ops
->patch_hypercall(vcpu
, hypercall
);
1418 kunmap_atomic(hypercall
, KM_USER1
);
1420 para_state
->ret
= 0;
1422 kunmap_atomic(para_state
, KM_USER0
);
1428 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1433 case 0xc0010010: /* SYSCFG */
1434 case 0xc0010015: /* HWCR */
1435 case MSR_IA32_PLATFORM_ID
:
1436 case MSR_IA32_P5_MC_ADDR
:
1437 case MSR_IA32_P5_MC_TYPE
:
1438 case MSR_IA32_MC0_CTL
:
1439 case MSR_IA32_MCG_STATUS
:
1440 case MSR_IA32_MCG_CAP
:
1441 case MSR_IA32_MC0_MISC
:
1442 case MSR_IA32_MC0_MISC
+4:
1443 case MSR_IA32_MC0_MISC
+8:
1444 case MSR_IA32_MC0_MISC
+12:
1445 case MSR_IA32_MC0_MISC
+16:
1446 case MSR_IA32_UCODE_REV
:
1447 case MSR_IA32_PERF_STATUS
:
1448 /* MTRR registers */
1450 case 0x200 ... 0x2ff:
1453 case 0xcd: /* fsb frequency */
1456 case MSR_IA32_APICBASE
:
1457 data
= vcpu
->apic_base
;
1459 case MSR_IA32_MISC_ENABLE
:
1460 data
= vcpu
->ia32_misc_enable_msr
;
1462 #ifdef CONFIG_X86_64
1464 data
= vcpu
->shadow_efer
;
1468 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1474 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1477 * Reads an msr value (of 'msr_index') into 'pdata'.
1478 * Returns 0 on success, non-0 otherwise.
1479 * Assumes vcpu_load() was already called.
1481 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1483 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1486 #ifdef CONFIG_X86_64
1488 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1490 if (efer
& EFER_RESERVED_BITS
) {
1491 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1498 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1499 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1504 kvm_arch_ops
->set_efer(vcpu
, efer
);
1507 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1509 vcpu
->shadow_efer
= efer
;
1514 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1517 #ifdef CONFIG_X86_64
1519 set_efer(vcpu
, data
);
1522 case MSR_IA32_MC0_STATUS
:
1523 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1524 __FUNCTION__
, data
);
1526 case MSR_IA32_MCG_STATUS
:
1527 printk(KERN_WARNING
"%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1528 __FUNCTION__
, data
);
1530 case MSR_IA32_UCODE_REV
:
1531 case MSR_IA32_UCODE_WRITE
:
1532 case 0x200 ... 0x2ff: /* MTRRs */
1534 case MSR_IA32_APICBASE
:
1535 vcpu
->apic_base
= data
;
1537 case MSR_IA32_MISC_ENABLE
:
1538 vcpu
->ia32_misc_enable_msr
= data
;
1541 * This is the 'probe whether the host is KVM' logic:
1543 case MSR_KVM_API_MAGIC
:
1544 return vcpu_register_para(vcpu
, data
);
1547 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1552 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1555 * Writes msr value into into the appropriate "register".
1556 * Returns 0 on success, non-0 otherwise.
1557 * Assumes vcpu_load() was already called.
1559 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1561 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1564 void kvm_resched(struct kvm_vcpu
*vcpu
)
1566 if (!need_resched())
1572 EXPORT_SYMBOL_GPL(kvm_resched
);
1574 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1578 for (i
= 0; i
< n
; ++i
)
1579 wrmsrl(e
[i
].index
, e
[i
].data
);
1581 EXPORT_SYMBOL_GPL(load_msrs
);
1583 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1587 for (i
= 0; i
< n
; ++i
)
1588 rdmsrl(e
[i
].index
, e
[i
].data
);
1590 EXPORT_SYMBOL_GPL(save_msrs
);
1592 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
1596 struct kvm_cpuid_entry
*e
, *best
;
1598 kvm_arch_ops
->cache_regs(vcpu
);
1599 function
= vcpu
->regs
[VCPU_REGS_RAX
];
1600 vcpu
->regs
[VCPU_REGS_RAX
] = 0;
1601 vcpu
->regs
[VCPU_REGS_RBX
] = 0;
1602 vcpu
->regs
[VCPU_REGS_RCX
] = 0;
1603 vcpu
->regs
[VCPU_REGS_RDX
] = 0;
1605 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
1606 e
= &vcpu
->cpuid_entries
[i
];
1607 if (e
->function
== function
) {
1612 * Both basic or both extended?
1614 if (((e
->function
^ function
) & 0x80000000) == 0)
1615 if (!best
|| e
->function
> best
->function
)
1619 vcpu
->regs
[VCPU_REGS_RAX
] = best
->eax
;
1620 vcpu
->regs
[VCPU_REGS_RBX
] = best
->ebx
;
1621 vcpu
->regs
[VCPU_REGS_RCX
] = best
->ecx
;
1622 vcpu
->regs
[VCPU_REGS_RDX
] = best
->edx
;
1624 kvm_arch_ops
->decache_regs(vcpu
);
1625 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1627 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
1629 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
1631 void *p
= vcpu
->pio_data
;
1634 int nr_pages
= vcpu
->pio
.guest_pages
[1] ? 2 : 1;
1636 kvm_arch_ops
->vcpu_put(vcpu
);
1637 q
= vmap(vcpu
->pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
1640 kvm_arch_ops
->vcpu_load(vcpu
);
1641 free_pio_guest_pages(vcpu
);
1644 q
+= vcpu
->pio
.guest_page_offset
;
1645 bytes
= vcpu
->pio
.size
* vcpu
->pio
.cur_count
;
1647 memcpy(q
, p
, bytes
);
1649 memcpy(p
, q
, bytes
);
1650 q
-= vcpu
->pio
.guest_page_offset
;
1652 kvm_arch_ops
->vcpu_load(vcpu
);
1653 free_pio_guest_pages(vcpu
);
1657 static int complete_pio(struct kvm_vcpu
*vcpu
)
1659 struct kvm_pio_request
*io
= &vcpu
->pio
;
1663 kvm_arch_ops
->cache_regs(vcpu
);
1667 memcpy(&vcpu
->regs
[VCPU_REGS_RAX
], vcpu
->pio_data
,
1671 r
= pio_copy_data(vcpu
);
1673 kvm_arch_ops
->cache_regs(vcpu
);
1680 delta
*= io
->cur_count
;
1682 * The size of the register should really depend on
1683 * current address size.
1685 vcpu
->regs
[VCPU_REGS_RCX
] -= delta
;
1691 vcpu
->regs
[VCPU_REGS_RDI
] += delta
;
1693 vcpu
->regs
[VCPU_REGS_RSI
] += delta
;
1696 vcpu
->run
->io_completed
= 0;
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 (kvm_run
->io_completed
) {
1805 if (vcpu
->pio
.cur_count
) {
1806 r
= complete_pio(vcpu
);
1810 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1811 vcpu
->mmio_read_completed
= 1;
1815 vcpu
->mmio_needed
= 0;
1817 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
1818 kvm_arch_ops
->cache_regs(vcpu
);
1819 vcpu
->regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
1820 kvm_arch_ops
->decache_regs(vcpu
);
1823 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1826 if (vcpu
->sigset_active
)
1827 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1833 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
,
1834 struct kvm_regs
*regs
)
1838 kvm_arch_ops
->cache_regs(vcpu
);
1840 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1841 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1842 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1843 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1844 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1845 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1846 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1847 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1848 #ifdef CONFIG_X86_64
1849 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1850 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1851 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1852 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1853 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1854 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1855 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1856 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1859 regs
->rip
= vcpu
->rip
;
1860 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1863 * Don't leak debug flags in case they were set for guest debugging
1865 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1866 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1873 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
,
1874 struct kvm_regs
*regs
)
1878 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1879 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1880 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1881 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1882 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1883 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1884 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1885 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1886 #ifdef CONFIG_X86_64
1887 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1888 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1889 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1890 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1891 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1892 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1893 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1894 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1897 vcpu
->rip
= regs
->rip
;
1898 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1900 kvm_arch_ops
->decache_regs(vcpu
);
1907 static void get_segment(struct kvm_vcpu
*vcpu
,
1908 struct kvm_segment
*var
, int seg
)
1910 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1913 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
1914 struct kvm_sregs
*sregs
)
1916 struct descriptor_table dt
;
1920 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1921 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1922 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1923 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1924 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1925 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1927 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1928 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1930 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1931 sregs
->idt
.limit
= dt
.limit
;
1932 sregs
->idt
.base
= dt
.base
;
1933 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1934 sregs
->gdt
.limit
= dt
.limit
;
1935 sregs
->gdt
.base
= dt
.base
;
1937 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1938 sregs
->cr0
= vcpu
->cr0
;
1939 sregs
->cr2
= vcpu
->cr2
;
1940 sregs
->cr3
= vcpu
->cr3
;
1941 sregs
->cr4
= vcpu
->cr4
;
1942 sregs
->cr8
= vcpu
->cr8
;
1943 sregs
->efer
= vcpu
->shadow_efer
;
1944 sregs
->apic_base
= vcpu
->apic_base
;
1946 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1947 sizeof sregs
->interrupt_bitmap
);
1954 static void set_segment(struct kvm_vcpu
*vcpu
,
1955 struct kvm_segment
*var
, int seg
)
1957 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1960 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
1961 struct kvm_sregs
*sregs
)
1963 int mmu_reset_needed
= 0;
1965 struct descriptor_table dt
;
1969 dt
.limit
= sregs
->idt
.limit
;
1970 dt
.base
= sregs
->idt
.base
;
1971 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1972 dt
.limit
= sregs
->gdt
.limit
;
1973 dt
.base
= sregs
->gdt
.base
;
1974 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1976 vcpu
->cr2
= sregs
->cr2
;
1977 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
1978 vcpu
->cr3
= sregs
->cr3
;
1980 vcpu
->cr8
= sregs
->cr8
;
1982 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
1983 #ifdef CONFIG_X86_64
1984 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
1986 vcpu
->apic_base
= sregs
->apic_base
;
1988 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1990 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
1991 kvm_arch_ops
->set_cr0(vcpu
, sregs
->cr0
);
1993 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
1994 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
1995 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
1996 load_pdptrs(vcpu
, vcpu
->cr3
);
1998 if (mmu_reset_needed
)
1999 kvm_mmu_reset_context(vcpu
);
2001 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
2002 sizeof vcpu
->irq_pending
);
2003 vcpu
->irq_summary
= 0;
2004 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
2005 if (vcpu
->irq_pending
[i
])
2006 __set_bit(i
, &vcpu
->irq_summary
);
2008 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2009 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2010 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2011 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2012 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2013 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2015 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2016 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2024 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2025 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2027 * This list is modified at module load time to reflect the
2028 * capabilities of the host cpu.
2030 static u32 msrs_to_save
[] = {
2031 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
2033 #ifdef CONFIG_X86_64
2034 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
2036 MSR_IA32_TIME_STAMP_COUNTER
,
2039 static unsigned num_msrs_to_save
;
2041 static u32 emulated_msrs
[] = {
2042 MSR_IA32_MISC_ENABLE
,
2045 static __init
void kvm_init_msr_list(void)
2050 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2051 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2054 msrs_to_save
[j
] = msrs_to_save
[i
];
2057 num_msrs_to_save
= j
;
2061 * Adapt set_msr() to msr_io()'s calling convention
2063 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
2065 return set_msr(vcpu
, index
, *data
);
2069 * Read or write a bunch of msrs. All parameters are kernel addresses.
2071 * @return number of msrs set successfully.
2073 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2074 struct kvm_msr_entry
*entries
,
2075 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2076 unsigned index
, u64
*data
))
2082 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2083 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2092 * Read or write a bunch of msrs. Parameters are user addresses.
2094 * @return number of msrs set successfully.
2096 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2097 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2098 unsigned index
, u64
*data
),
2101 struct kvm_msrs msrs
;
2102 struct kvm_msr_entry
*entries
;
2107 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2111 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2115 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2116 entries
= vmalloc(size
);
2121 if (copy_from_user(entries
, user_msrs
->entries
, size
))
2124 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2129 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2141 * Translate a guest virtual address to a guest physical address.
2143 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
2144 struct kvm_translation
*tr
)
2146 unsigned long vaddr
= tr
->linear_address
;
2150 spin_lock(&vcpu
->kvm
->lock
);
2151 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
2152 tr
->physical_address
= gpa
;
2153 tr
->valid
= gpa
!= UNMAPPED_GVA
;
2156 spin_unlock(&vcpu
->kvm
->lock
);
2162 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2163 struct kvm_interrupt
*irq
)
2165 if (irq
->irq
< 0 || irq
->irq
>= 256)
2169 set_bit(irq
->irq
, vcpu
->irq_pending
);
2170 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
2177 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
2178 struct kvm_debug_guest
*dbg
)
2184 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
2191 static struct page
*kvm_vcpu_nopage(struct vm_area_struct
*vma
,
2192 unsigned long address
,
2195 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2196 unsigned long pgoff
;
2199 *type
= VM_FAULT_MINOR
;
2200 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2202 page
= virt_to_page(vcpu
->run
);
2203 else if (pgoff
== KVM_PIO_PAGE_OFFSET
)
2204 page
= virt_to_page(vcpu
->pio_data
);
2206 return NOPAGE_SIGBUS
;
2211 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
2212 .nopage
= kvm_vcpu_nopage
,
2215 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2217 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2221 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2223 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2225 fput(vcpu
->kvm
->filp
);
2229 static struct file_operations kvm_vcpu_fops
= {
2230 .release
= kvm_vcpu_release
,
2231 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2232 .compat_ioctl
= kvm_vcpu_ioctl
,
2233 .mmap
= kvm_vcpu_mmap
,
2237 * Allocates an inode for the vcpu.
2239 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2242 struct inode
*inode
;
2245 atomic_inc(&vcpu
->kvm
->filp
->f_count
);
2246 inode
= kvmfs_inode(&kvm_vcpu_fops
);
2247 if (IS_ERR(inode
)) {
2252 file
= kvmfs_file(inode
, vcpu
);
2258 r
= get_unused_fd();
2262 fd_install(fd
, file
);
2271 fput(vcpu
->kvm
->filp
);
2276 * Creates some virtual cpus. Good luck creating more than one.
2278 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
2281 struct kvm_vcpu
*vcpu
;
2288 vcpu
= &kvm
->vcpus
[n
];
2290 mutex_lock(&vcpu
->mutex
);
2293 mutex_unlock(&vcpu
->mutex
);
2297 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2301 vcpu
->run
= page_address(page
);
2303 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2307 vcpu
->pio_data
= page_address(page
);
2309 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
2311 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
2314 r
= kvm_arch_ops
->vcpu_create(vcpu
);
2316 goto out_free_vcpus
;
2318 r
= kvm_mmu_create(vcpu
);
2320 goto out_free_vcpus
;
2322 kvm_arch_ops
->vcpu_load(vcpu
);
2323 r
= kvm_mmu_setup(vcpu
);
2325 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
2329 goto out_free_vcpus
;
2331 r
= create_vcpu_fd(vcpu
);
2333 goto out_free_vcpus
;
2338 kvm_free_vcpu(vcpu
);
2340 free_page((unsigned long)vcpu
->run
);
2343 mutex_unlock(&vcpu
->mutex
);
2348 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
2349 struct kvm_cpuid
*cpuid
,
2350 struct kvm_cpuid_entry __user
*entries
)
2355 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2358 if (copy_from_user(&vcpu
->cpuid_entries
, entries
,
2359 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
2361 vcpu
->cpuid_nent
= cpuid
->nent
;
2368 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2371 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2372 vcpu
->sigset_active
= 1;
2373 vcpu
->sigset
= *sigset
;
2375 vcpu
->sigset_active
= 0;
2380 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2381 * we have asm/x86/processor.h
2392 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2393 #ifdef CONFIG_X86_64
2394 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2396 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2400 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2402 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2406 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
2407 fpu
->fcw
= fxsave
->cwd
;
2408 fpu
->fsw
= fxsave
->swd
;
2409 fpu
->ftwx
= fxsave
->twd
;
2410 fpu
->last_opcode
= fxsave
->fop
;
2411 fpu
->last_ip
= fxsave
->rip
;
2412 fpu
->last_dp
= fxsave
->rdp
;
2413 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
2420 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2422 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2426 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
2427 fxsave
->cwd
= fpu
->fcw
;
2428 fxsave
->swd
= fpu
->fsw
;
2429 fxsave
->twd
= fpu
->ftwx
;
2430 fxsave
->fop
= fpu
->last_opcode
;
2431 fxsave
->rip
= fpu
->last_ip
;
2432 fxsave
->rdp
= fpu
->last_dp
;
2433 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
2440 static long kvm_vcpu_ioctl(struct file
*filp
,
2441 unsigned int ioctl
, unsigned long arg
)
2443 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2444 void __user
*argp
= (void __user
*)arg
;
2452 r
= kvm_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2454 case KVM_GET_REGS
: {
2455 struct kvm_regs kvm_regs
;
2457 memset(&kvm_regs
, 0, sizeof kvm_regs
);
2458 r
= kvm_vcpu_ioctl_get_regs(vcpu
, &kvm_regs
);
2462 if (copy_to_user(argp
, &kvm_regs
, sizeof kvm_regs
))
2467 case KVM_SET_REGS
: {
2468 struct kvm_regs kvm_regs
;
2471 if (copy_from_user(&kvm_regs
, argp
, sizeof kvm_regs
))
2473 r
= kvm_vcpu_ioctl_set_regs(vcpu
, &kvm_regs
);
2479 case KVM_GET_SREGS
: {
2480 struct kvm_sregs kvm_sregs
;
2482 memset(&kvm_sregs
, 0, sizeof kvm_sregs
);
2483 r
= kvm_vcpu_ioctl_get_sregs(vcpu
, &kvm_sregs
);
2487 if (copy_to_user(argp
, &kvm_sregs
, sizeof kvm_sregs
))
2492 case KVM_SET_SREGS
: {
2493 struct kvm_sregs kvm_sregs
;
2496 if (copy_from_user(&kvm_sregs
, argp
, sizeof kvm_sregs
))
2498 r
= kvm_vcpu_ioctl_set_sregs(vcpu
, &kvm_sregs
);
2504 case KVM_TRANSLATE
: {
2505 struct kvm_translation tr
;
2508 if (copy_from_user(&tr
, argp
, sizeof tr
))
2510 r
= kvm_vcpu_ioctl_translate(vcpu
, &tr
);
2514 if (copy_to_user(argp
, &tr
, sizeof tr
))
2519 case KVM_INTERRUPT
: {
2520 struct kvm_interrupt irq
;
2523 if (copy_from_user(&irq
, argp
, sizeof irq
))
2525 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2531 case KVM_DEBUG_GUEST
: {
2532 struct kvm_debug_guest dbg
;
2535 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2537 r
= kvm_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
2544 r
= msr_io(vcpu
, argp
, get_msr
, 1);
2547 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2549 case KVM_SET_CPUID
: {
2550 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2551 struct kvm_cpuid cpuid
;
2554 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2556 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2561 case KVM_SET_SIGNAL_MASK
: {
2562 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2563 struct kvm_signal_mask kvm_sigmask
;
2564 sigset_t sigset
, *p
;
2569 if (copy_from_user(&kvm_sigmask
, argp
,
2570 sizeof kvm_sigmask
))
2573 if (kvm_sigmask
.len
!= sizeof sigset
)
2576 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2581 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2587 memset(&fpu
, 0, sizeof fpu
);
2588 r
= kvm_vcpu_ioctl_get_fpu(vcpu
, &fpu
);
2592 if (copy_to_user(argp
, &fpu
, sizeof fpu
))
2601 if (copy_from_user(&fpu
, argp
, sizeof fpu
))
2603 r
= kvm_vcpu_ioctl_set_fpu(vcpu
, &fpu
);
2616 static long kvm_vm_ioctl(struct file
*filp
,
2617 unsigned int ioctl
, unsigned long arg
)
2619 struct kvm
*kvm
= filp
->private_data
;
2620 void __user
*argp
= (void __user
*)arg
;
2624 case KVM_CREATE_VCPU
:
2625 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2629 case KVM_SET_MEMORY_REGION
: {
2630 struct kvm_memory_region kvm_mem
;
2633 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2635 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_mem
);
2640 case KVM_GET_DIRTY_LOG
: {
2641 struct kvm_dirty_log log
;
2644 if (copy_from_user(&log
, argp
, sizeof log
))
2646 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2651 case KVM_SET_MEMORY_ALIAS
: {
2652 struct kvm_memory_alias alias
;
2655 if (copy_from_user(&alias
, argp
, sizeof alias
))
2657 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &alias
);
2669 static struct page
*kvm_vm_nopage(struct vm_area_struct
*vma
,
2670 unsigned long address
,
2673 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2674 unsigned long pgoff
;
2677 *type
= VM_FAULT_MINOR
;
2678 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2679 page
= gfn_to_page(kvm
, pgoff
);
2681 return NOPAGE_SIGBUS
;
2686 static struct vm_operations_struct kvm_vm_vm_ops
= {
2687 .nopage
= kvm_vm_nopage
,
2690 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2692 vma
->vm_ops
= &kvm_vm_vm_ops
;
2696 static struct file_operations kvm_vm_fops
= {
2697 .release
= kvm_vm_release
,
2698 .unlocked_ioctl
= kvm_vm_ioctl
,
2699 .compat_ioctl
= kvm_vm_ioctl
,
2700 .mmap
= kvm_vm_mmap
,
2703 static int kvm_dev_ioctl_create_vm(void)
2706 struct inode
*inode
;
2710 inode
= kvmfs_inode(&kvm_vm_fops
);
2711 if (IS_ERR(inode
)) {
2716 kvm
= kvm_create_vm();
2722 file
= kvmfs_file(inode
, kvm
);
2729 r
= get_unused_fd();
2733 fd_install(fd
, file
);
2740 kvm_destroy_vm(kvm
);
2747 static long kvm_dev_ioctl(struct file
*filp
,
2748 unsigned int ioctl
, unsigned long arg
)
2750 void __user
*argp
= (void __user
*)arg
;
2754 case KVM_GET_API_VERSION
:
2758 r
= KVM_API_VERSION
;
2764 r
= kvm_dev_ioctl_create_vm();
2766 case KVM_GET_MSR_INDEX_LIST
: {
2767 struct kvm_msr_list __user
*user_msr_list
= argp
;
2768 struct kvm_msr_list msr_list
;
2772 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2775 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2776 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2779 if (n
< num_msrs_to_save
)
2782 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2783 num_msrs_to_save
* sizeof(u32
)))
2785 if (copy_to_user(user_msr_list
->indices
2786 + num_msrs_to_save
* sizeof(u32
),
2788 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2793 case KVM_CHECK_EXTENSION
:
2795 * No extensions defined at present.
2799 case KVM_GET_VCPU_MMAP_SIZE
:
2812 static struct file_operations kvm_chardev_ops
= {
2813 .open
= kvm_dev_open
,
2814 .release
= kvm_dev_release
,
2815 .unlocked_ioctl
= kvm_dev_ioctl
,
2816 .compat_ioctl
= kvm_dev_ioctl
,
2819 static struct miscdevice kvm_dev
= {
2825 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2828 if (val
== SYS_RESTART
) {
2830 * Some (well, at least mine) BIOSes hang on reboot if
2833 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2834 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2839 static struct notifier_block kvm_reboot_notifier
= {
2840 .notifier_call
= kvm_reboot
,
2845 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2848 static void decache_vcpus_on_cpu(int cpu
)
2851 struct kvm_vcpu
*vcpu
;
2854 spin_lock(&kvm_lock
);
2855 list_for_each_entry(vm
, &vm_list
, vm_list
)
2856 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2857 vcpu
= &vm
->vcpus
[i
];
2859 * If the vcpu is locked, then it is running on some
2860 * other cpu and therefore it is not cached on the
2863 * If it's not locked, check the last cpu it executed
2866 if (mutex_trylock(&vcpu
->mutex
)) {
2867 if (vcpu
->cpu
== cpu
) {
2868 kvm_arch_ops
->vcpu_decache(vcpu
);
2871 mutex_unlock(&vcpu
->mutex
);
2874 spin_unlock(&kvm_lock
);
2877 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2883 case CPU_DOWN_PREPARE
:
2884 case CPU_UP_CANCELED
:
2885 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2887 decache_vcpus_on_cpu(cpu
);
2888 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_disable
,
2892 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2894 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_enable
,
2901 static struct notifier_block kvm_cpu_notifier
= {
2902 .notifier_call
= kvm_cpu_hotplug
,
2903 .priority
= 20, /* must be > scheduler priority */
2906 static u64
stat_get(void *_offset
)
2908 unsigned offset
= (long)_offset
;
2911 struct kvm_vcpu
*vcpu
;
2914 spin_lock(&kvm_lock
);
2915 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2916 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2917 vcpu
= &kvm
->vcpus
[i
];
2918 total
+= *(u32
*)((void *)vcpu
+ offset
);
2920 spin_unlock(&kvm_lock
);
2924 static void stat_set(void *offset
, u64 val
)
2928 DEFINE_SIMPLE_ATTRIBUTE(stat_fops
, stat_get
, stat_set
, "%llu\n");
2930 static __init
void kvm_init_debug(void)
2932 struct kvm_stats_debugfs_item
*p
;
2934 debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2935 for (p
= debugfs_entries
; p
->name
; ++p
)
2936 p
->dentry
= debugfs_create_file(p
->name
, 0444, debugfs_dir
,
2937 (void *)(long)p
->offset
,
2941 static void kvm_exit_debug(void)
2943 struct kvm_stats_debugfs_item
*p
;
2945 for (p
= debugfs_entries
; p
->name
; ++p
)
2946 debugfs_remove(p
->dentry
);
2947 debugfs_remove(debugfs_dir
);
2950 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2952 decache_vcpus_on_cpu(raw_smp_processor_id());
2953 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2957 static int kvm_resume(struct sys_device
*dev
)
2959 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
2963 static struct sysdev_class kvm_sysdev_class
= {
2964 set_kset_name("kvm"),
2965 .suspend
= kvm_suspend
,
2966 .resume
= kvm_resume
,
2969 static struct sys_device kvm_sysdev
= {
2971 .cls
= &kvm_sysdev_class
,
2974 hpa_t bad_page_address
;
2976 static int kvmfs_get_sb(struct file_system_type
*fs_type
, int flags
,
2977 const char *dev_name
, void *data
, struct vfsmount
*mnt
)
2979 return get_sb_pseudo(fs_type
, "kvm:", NULL
, KVMFS_SUPER_MAGIC
, mnt
);
2982 static struct file_system_type kvm_fs_type
= {
2984 .get_sb
= kvmfs_get_sb
,
2985 .kill_sb
= kill_anon_super
,
2988 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
2993 printk(KERN_ERR
"kvm: already loaded the other module\n");
2997 if (!ops
->cpu_has_kvm_support()) {
2998 printk(KERN_ERR
"kvm: no hardware support\n");
3001 if (ops
->disabled_by_bios()) {
3002 printk(KERN_ERR
"kvm: disabled by bios\n");
3008 r
= kvm_arch_ops
->hardware_setup();
3012 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
3013 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3016 register_reboot_notifier(&kvm_reboot_notifier
);
3018 r
= sysdev_class_register(&kvm_sysdev_class
);
3022 r
= sysdev_register(&kvm_sysdev
);
3026 kvm_chardev_ops
.owner
= module
;
3028 r
= misc_register(&kvm_dev
);
3030 printk (KERN_ERR
"kvm: misc device register failed\n");
3037 sysdev_unregister(&kvm_sysdev
);
3039 sysdev_class_unregister(&kvm_sysdev_class
);
3041 unregister_reboot_notifier(&kvm_reboot_notifier
);
3042 unregister_cpu_notifier(&kvm_cpu_notifier
);
3044 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
3045 kvm_arch_ops
->hardware_unsetup();
3047 kvm_arch_ops
= NULL
;
3051 void kvm_exit_arch(void)
3053 misc_deregister(&kvm_dev
);
3054 sysdev_unregister(&kvm_sysdev
);
3055 sysdev_class_unregister(&kvm_sysdev_class
);
3056 unregister_reboot_notifier(&kvm_reboot_notifier
);
3057 unregister_cpu_notifier(&kvm_cpu_notifier
);
3058 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
3059 kvm_arch_ops
->hardware_unsetup();
3060 kvm_arch_ops
= NULL
;
3063 static __init
int kvm_init(void)
3065 static struct page
*bad_page
;
3068 r
= kvm_mmu_module_init();
3072 r
= register_filesystem(&kvm_fs_type
);
3076 kvmfs_mnt
= kern_mount(&kvm_fs_type
);
3077 r
= PTR_ERR(kvmfs_mnt
);
3078 if (IS_ERR(kvmfs_mnt
))
3082 kvm_init_msr_list();
3084 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
3089 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
3090 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
3098 unregister_filesystem(&kvm_fs_type
);
3100 kvm_mmu_module_exit();
3105 static __exit
void kvm_exit(void)
3108 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
3110 unregister_filesystem(&kvm_fs_type
);
3111 kvm_mmu_module_exit();
3114 module_init(kvm_init
)
3115 module_exit(kvm_exit
)
3117 EXPORT_SYMBOL_GPL(kvm_init_arch
);
3118 EXPORT_SYMBOL_GPL(kvm_exit_arch
);