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;
423 struct kvm_memory_slot
*memslot
;
425 spin_lock(&vcpu
->kvm
->lock
);
426 memslot
= gfn_to_memslot(vcpu
->kvm
, pdpt_gfn
);
427 /* FIXME: !memslot - emulate? 0xff? */
428 pdpt
= kmap_atomic(gfn_to_page(memslot
, pdpt_gfn
), 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
);
849 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
853 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
854 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
856 if (gfn
>= memslot
->base_gfn
857 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
862 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
864 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
867 struct kvm_memory_slot
*memslot
= NULL
;
868 unsigned long rel_gfn
;
870 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
871 memslot
= &kvm
->memslots
[i
];
873 if (gfn
>= memslot
->base_gfn
874 && gfn
< memslot
->base_gfn
+ memslot
->npages
) {
876 if (!memslot
|| !memslot
->dirty_bitmap
)
879 rel_gfn
= gfn
- memslot
->base_gfn
;
882 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
883 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
889 static int emulator_read_std(unsigned long addr
,
892 struct x86_emulate_ctxt
*ctxt
)
894 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
898 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
899 unsigned offset
= addr
& (PAGE_SIZE
-1);
900 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
902 struct kvm_memory_slot
*memslot
;
905 if (gpa
== UNMAPPED_GVA
)
906 return X86EMUL_PROPAGATE_FAULT
;
907 pfn
= gpa
>> PAGE_SHIFT
;
908 memslot
= gfn_to_memslot(vcpu
->kvm
, pfn
);
910 return X86EMUL_UNHANDLEABLE
;
911 page
= kmap_atomic(gfn_to_page(memslot
, pfn
), KM_USER0
);
913 memcpy(data
, page
+ offset
, tocopy
);
915 kunmap_atomic(page
, KM_USER0
);
922 return X86EMUL_CONTINUE
;
925 static int emulator_write_std(unsigned long addr
,
928 struct x86_emulate_ctxt
*ctxt
)
930 printk(KERN_ERR
"emulator_write_std: addr %lx n %d\n",
932 return X86EMUL_UNHANDLEABLE
;
935 static int emulator_read_emulated(unsigned long addr
,
938 struct x86_emulate_ctxt
*ctxt
)
940 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
942 if (vcpu
->mmio_read_completed
) {
943 memcpy(val
, vcpu
->mmio_data
, bytes
);
944 vcpu
->mmio_read_completed
= 0;
945 return X86EMUL_CONTINUE
;
946 } else if (emulator_read_std(addr
, val
, bytes
, ctxt
)
948 return X86EMUL_CONTINUE
;
950 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
952 if (gpa
== UNMAPPED_GVA
)
953 return X86EMUL_PROPAGATE_FAULT
;
954 vcpu
->mmio_needed
= 1;
955 vcpu
->mmio_phys_addr
= gpa
;
956 vcpu
->mmio_size
= bytes
;
957 vcpu
->mmio_is_write
= 0;
959 return X86EMUL_UNHANDLEABLE
;
963 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
964 unsigned long val
, int bytes
)
966 struct kvm_memory_slot
*m
;
970 if (((gpa
+ bytes
- 1) >> PAGE_SHIFT
) != (gpa
>> PAGE_SHIFT
))
972 m
= gfn_to_memslot(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
975 page
= gfn_to_page(m
, gpa
>> PAGE_SHIFT
);
976 kvm_mmu_pre_write(vcpu
, gpa
, bytes
);
977 mark_page_dirty(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
978 virt
= kmap_atomic(page
, KM_USER0
);
979 memcpy(virt
+ offset_in_page(gpa
), &val
, bytes
);
980 kunmap_atomic(virt
, KM_USER0
);
981 kvm_mmu_post_write(vcpu
, gpa
, bytes
);
985 static int emulator_write_emulated(unsigned long addr
,
988 struct x86_emulate_ctxt
*ctxt
)
990 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
991 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
993 if (gpa
== UNMAPPED_GVA
)
994 return X86EMUL_PROPAGATE_FAULT
;
996 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
997 return X86EMUL_CONTINUE
;
999 vcpu
->mmio_needed
= 1;
1000 vcpu
->mmio_phys_addr
= gpa
;
1001 vcpu
->mmio_size
= bytes
;
1002 vcpu
->mmio_is_write
= 1;
1003 memcpy(vcpu
->mmio_data
, &val
, bytes
);
1005 return X86EMUL_CONTINUE
;
1008 static int emulator_cmpxchg_emulated(unsigned long addr
,
1012 struct x86_emulate_ctxt
*ctxt
)
1014 static int reported
;
1018 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1020 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
1023 #ifdef CONFIG_X86_32
1025 static int emulator_cmpxchg8b_emulated(unsigned long addr
,
1026 unsigned long old_lo
,
1027 unsigned long old_hi
,
1028 unsigned long new_lo
,
1029 unsigned long new_hi
,
1030 struct x86_emulate_ctxt
*ctxt
)
1032 static int reported
;
1037 printk(KERN_WARNING
"kvm: emulating exchange8b as write\n");
1039 r
= emulator_write_emulated(addr
, new_lo
, 4, ctxt
);
1040 if (r
!= X86EMUL_CONTINUE
)
1042 return emulator_write_emulated(addr
+4, new_hi
, 4, ctxt
);
1047 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1049 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
1052 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1054 return X86EMUL_CONTINUE
;
1057 int emulate_clts(struct kvm_vcpu
*vcpu
)
1061 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1062 cr0
= vcpu
->cr0
& ~CR0_TS_MASK
;
1063 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
1064 return X86EMUL_CONTINUE
;
1067 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
1069 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1073 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
1074 return X86EMUL_CONTINUE
;
1076 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1078 return X86EMUL_UNHANDLEABLE
;
1082 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1084 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1087 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1089 /* FIXME: better handling */
1090 return X86EMUL_UNHANDLEABLE
;
1092 return X86EMUL_CONTINUE
;
1095 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1097 static int reported
;
1099 unsigned long rip
= ctxt
->vcpu
->rip
;
1100 unsigned long rip_linear
;
1102 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1107 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1109 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1110 " rip %lx %02x %02x %02x %02x\n",
1111 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1115 struct x86_emulate_ops emulate_ops
= {
1116 .read_std
= emulator_read_std
,
1117 .write_std
= emulator_write_std
,
1118 .read_emulated
= emulator_read_emulated
,
1119 .write_emulated
= emulator_write_emulated
,
1120 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1121 #ifdef CONFIG_X86_32
1122 .cmpxchg8b_emulated
= emulator_cmpxchg8b_emulated
,
1126 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1127 struct kvm_run
*run
,
1131 struct x86_emulate_ctxt emulate_ctxt
;
1135 kvm_arch_ops
->cache_regs(vcpu
);
1137 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1139 emulate_ctxt
.vcpu
= vcpu
;
1140 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
1141 emulate_ctxt
.cr2
= cr2
;
1142 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1143 ? X86EMUL_MODE_REAL
: cs_l
1144 ? X86EMUL_MODE_PROT64
: cs_db
1145 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1147 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1148 emulate_ctxt
.cs_base
= 0;
1149 emulate_ctxt
.ds_base
= 0;
1150 emulate_ctxt
.es_base
= 0;
1151 emulate_ctxt
.ss_base
= 0;
1153 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1154 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1155 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1156 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1159 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1160 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1162 vcpu
->mmio_is_write
= 0;
1163 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1165 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1166 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1167 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1168 run
->mmio
.len
= vcpu
->mmio_size
;
1169 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1173 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1174 return EMULATE_DONE
;
1175 if (!vcpu
->mmio_needed
) {
1176 report_emulation_failure(&emulate_ctxt
);
1177 return EMULATE_FAIL
;
1179 return EMULATE_DO_MMIO
;
1182 kvm_arch_ops
->decache_regs(vcpu
);
1183 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1185 if (vcpu
->mmio_is_write
)
1186 return EMULATE_DO_MMIO
;
1188 return EMULATE_DONE
;
1190 EXPORT_SYMBOL_GPL(emulate_instruction
);
1192 int kvm_hypercall(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1194 unsigned long nr
, a0
, a1
, a2
, a3
, a4
, a5
, ret
;
1196 kvm_arch_ops
->cache_regs(vcpu
);
1198 #ifdef CONFIG_X86_64
1199 if (is_long_mode(vcpu
)) {
1200 nr
= vcpu
->regs
[VCPU_REGS_RAX
];
1201 a0
= vcpu
->regs
[VCPU_REGS_RDI
];
1202 a1
= vcpu
->regs
[VCPU_REGS_RSI
];
1203 a2
= vcpu
->regs
[VCPU_REGS_RDX
];
1204 a3
= vcpu
->regs
[VCPU_REGS_RCX
];
1205 a4
= vcpu
->regs
[VCPU_REGS_R8
];
1206 a5
= vcpu
->regs
[VCPU_REGS_R9
];
1210 nr
= vcpu
->regs
[VCPU_REGS_RBX
] & -1u;
1211 a0
= vcpu
->regs
[VCPU_REGS_RAX
] & -1u;
1212 a1
= vcpu
->regs
[VCPU_REGS_RCX
] & -1u;
1213 a2
= vcpu
->regs
[VCPU_REGS_RDX
] & -1u;
1214 a3
= vcpu
->regs
[VCPU_REGS_RSI
] & -1u;
1215 a4
= vcpu
->regs
[VCPU_REGS_RDI
] & -1u;
1216 a5
= vcpu
->regs
[VCPU_REGS_RBP
] & -1u;
1220 run
->hypercall
.args
[0] = a0
;
1221 run
->hypercall
.args
[1] = a1
;
1222 run
->hypercall
.args
[2] = a2
;
1223 run
->hypercall
.args
[3] = a3
;
1224 run
->hypercall
.args
[4] = a4
;
1225 run
->hypercall
.args
[5] = a5
;
1226 run
->hypercall
.ret
= ret
;
1227 run
->hypercall
.longmode
= is_long_mode(vcpu
);
1228 kvm_arch_ops
->decache_regs(vcpu
);
1231 vcpu
->regs
[VCPU_REGS_RAX
] = ret
;
1232 kvm_arch_ops
->decache_regs(vcpu
);
1235 EXPORT_SYMBOL_GPL(kvm_hypercall
);
1237 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1239 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1242 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1244 struct descriptor_table dt
= { limit
, base
};
1246 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1249 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1251 struct descriptor_table dt
= { limit
, base
};
1253 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1256 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1257 unsigned long *rflags
)
1260 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1263 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1265 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1276 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1281 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1282 unsigned long *rflags
)
1286 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1287 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1296 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1299 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1304 * Register the para guest with the host:
1306 static int vcpu_register_para(struct kvm_vcpu
*vcpu
, gpa_t para_state_gpa
)
1308 struct kvm_vcpu_para_state
*para_state
;
1309 hpa_t para_state_hpa
, hypercall_hpa
;
1310 struct page
*para_state_page
;
1311 unsigned char *hypercall
;
1312 gpa_t hypercall_gpa
;
1314 printk(KERN_DEBUG
"kvm: guest trying to enter paravirtual mode\n");
1315 printk(KERN_DEBUG
".... para_state_gpa: %08Lx\n", para_state_gpa
);
1318 * Needs to be page aligned:
1320 if (para_state_gpa
!= PAGE_ALIGN(para_state_gpa
))
1323 para_state_hpa
= gpa_to_hpa(vcpu
, para_state_gpa
);
1324 printk(KERN_DEBUG
".... para_state_hpa: %08Lx\n", para_state_hpa
);
1325 if (is_error_hpa(para_state_hpa
))
1328 mark_page_dirty(vcpu
->kvm
, para_state_gpa
>> PAGE_SHIFT
);
1329 para_state_page
= pfn_to_page(para_state_hpa
>> PAGE_SHIFT
);
1330 para_state
= kmap_atomic(para_state_page
, KM_USER0
);
1332 printk(KERN_DEBUG
".... guest version: %d\n", para_state
->guest_version
);
1333 printk(KERN_DEBUG
".... size: %d\n", para_state
->size
);
1335 para_state
->host_version
= KVM_PARA_API_VERSION
;
1337 * We cannot support guests that try to register themselves
1338 * with a newer API version than the host supports:
1340 if (para_state
->guest_version
> KVM_PARA_API_VERSION
) {
1341 para_state
->ret
= -KVM_EINVAL
;
1342 goto err_kunmap_skip
;
1345 hypercall_gpa
= para_state
->hypercall_gpa
;
1346 hypercall_hpa
= gpa_to_hpa(vcpu
, hypercall_gpa
);
1347 printk(KERN_DEBUG
".... hypercall_hpa: %08Lx\n", hypercall_hpa
);
1348 if (is_error_hpa(hypercall_hpa
)) {
1349 para_state
->ret
= -KVM_EINVAL
;
1350 goto err_kunmap_skip
;
1353 printk(KERN_DEBUG
"kvm: para guest successfully registered.\n");
1354 vcpu
->para_state_page
= para_state_page
;
1355 vcpu
->para_state_gpa
= para_state_gpa
;
1356 vcpu
->hypercall_gpa
= hypercall_gpa
;
1358 mark_page_dirty(vcpu
->kvm
, hypercall_gpa
>> PAGE_SHIFT
);
1359 hypercall
= kmap_atomic(pfn_to_page(hypercall_hpa
>> PAGE_SHIFT
),
1360 KM_USER1
) + (hypercall_hpa
& ~PAGE_MASK
);
1361 kvm_arch_ops
->patch_hypercall(vcpu
, hypercall
);
1362 kunmap_atomic(hypercall
, KM_USER1
);
1364 para_state
->ret
= 0;
1366 kunmap_atomic(para_state
, KM_USER0
);
1372 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1377 case 0xc0010010: /* SYSCFG */
1378 case 0xc0010015: /* HWCR */
1379 case MSR_IA32_PLATFORM_ID
:
1380 case MSR_IA32_P5_MC_ADDR
:
1381 case MSR_IA32_P5_MC_TYPE
:
1382 case MSR_IA32_MC0_CTL
:
1383 case MSR_IA32_MCG_STATUS
:
1384 case MSR_IA32_MCG_CAP
:
1385 case MSR_IA32_MC0_MISC
:
1386 case MSR_IA32_MC0_MISC
+4:
1387 case MSR_IA32_MC0_MISC
+8:
1388 case MSR_IA32_MC0_MISC
+12:
1389 case MSR_IA32_MC0_MISC
+16:
1390 case MSR_IA32_UCODE_REV
:
1391 case MSR_IA32_PERF_STATUS
:
1392 /* MTRR registers */
1394 case 0x200 ... 0x2ff:
1397 case 0xcd: /* fsb frequency */
1400 case MSR_IA32_APICBASE
:
1401 data
= vcpu
->apic_base
;
1403 case MSR_IA32_MISC_ENABLE
:
1404 data
= vcpu
->ia32_misc_enable_msr
;
1406 #ifdef CONFIG_X86_64
1408 data
= vcpu
->shadow_efer
;
1412 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1418 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1421 * Reads an msr value (of 'msr_index') into 'pdata'.
1422 * Returns 0 on success, non-0 otherwise.
1423 * Assumes vcpu_load() was already called.
1425 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1427 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1430 #ifdef CONFIG_X86_64
1432 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1434 if (efer
& EFER_RESERVED_BITS
) {
1435 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1442 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1443 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1448 kvm_arch_ops
->set_efer(vcpu
, efer
);
1451 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1453 vcpu
->shadow_efer
= efer
;
1458 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1461 #ifdef CONFIG_X86_64
1463 set_efer(vcpu
, data
);
1466 case MSR_IA32_MC0_STATUS
:
1467 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1468 __FUNCTION__
, data
);
1470 case MSR_IA32_UCODE_REV
:
1471 case MSR_IA32_UCODE_WRITE
:
1472 case 0x200 ... 0x2ff: /* MTRRs */
1474 case MSR_IA32_APICBASE
:
1475 vcpu
->apic_base
= data
;
1477 case MSR_IA32_MISC_ENABLE
:
1478 vcpu
->ia32_misc_enable_msr
= data
;
1481 * This is the 'probe whether the host is KVM' logic:
1483 case MSR_KVM_API_MAGIC
:
1484 return vcpu_register_para(vcpu
, data
);
1487 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1492 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1495 * Writes msr value into into the appropriate "register".
1496 * Returns 0 on success, non-0 otherwise.
1497 * Assumes vcpu_load() was already called.
1499 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1501 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1504 void kvm_resched(struct kvm_vcpu
*vcpu
)
1510 EXPORT_SYMBOL_GPL(kvm_resched
);
1512 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1516 for (i
= 0; i
< n
; ++i
)
1517 wrmsrl(e
[i
].index
, e
[i
].data
);
1519 EXPORT_SYMBOL_GPL(load_msrs
);
1521 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1525 for (i
= 0; i
< n
; ++i
)
1526 rdmsrl(e
[i
].index
, e
[i
].data
);
1528 EXPORT_SYMBOL_GPL(save_msrs
);
1530 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
1534 struct kvm_cpuid_entry
*e
, *best
;
1536 kvm_arch_ops
->cache_regs(vcpu
);
1537 function
= vcpu
->regs
[VCPU_REGS_RAX
];
1538 vcpu
->regs
[VCPU_REGS_RAX
] = 0;
1539 vcpu
->regs
[VCPU_REGS_RBX
] = 0;
1540 vcpu
->regs
[VCPU_REGS_RCX
] = 0;
1541 vcpu
->regs
[VCPU_REGS_RDX
] = 0;
1543 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
1544 e
= &vcpu
->cpuid_entries
[i
];
1545 if (e
->function
== function
) {
1550 * Both basic or both extended?
1552 if (((e
->function
^ function
) & 0x80000000) == 0)
1553 if (!best
|| e
->function
> best
->function
)
1557 vcpu
->regs
[VCPU_REGS_RAX
] = best
->eax
;
1558 vcpu
->regs
[VCPU_REGS_RBX
] = best
->ebx
;
1559 vcpu
->regs
[VCPU_REGS_RCX
] = best
->ecx
;
1560 vcpu
->regs
[VCPU_REGS_RDX
] = best
->edx
;
1562 kvm_arch_ops
->decache_regs(vcpu
);
1563 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1565 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
1567 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
1569 void *p
= vcpu
->pio_data
;
1572 int nr_pages
= vcpu
->pio
.guest_pages
[1] ? 2 : 1;
1574 kvm_arch_ops
->vcpu_put(vcpu
);
1575 q
= vmap(vcpu
->pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
1578 kvm_arch_ops
->vcpu_load(vcpu
);
1579 free_pio_guest_pages(vcpu
);
1582 q
+= vcpu
->pio
.guest_page_offset
;
1583 bytes
= vcpu
->pio
.size
* vcpu
->pio
.cur_count
;
1585 memcpy(q
, p
, bytes
);
1587 memcpy(p
, q
, bytes
);
1588 q
-= vcpu
->pio
.guest_page_offset
;
1590 kvm_arch_ops
->vcpu_load(vcpu
);
1591 free_pio_guest_pages(vcpu
);
1595 static int complete_pio(struct kvm_vcpu
*vcpu
)
1597 struct kvm_pio_request
*io
= &vcpu
->pio
;
1601 kvm_arch_ops
->cache_regs(vcpu
);
1605 memcpy(&vcpu
->regs
[VCPU_REGS_RAX
], vcpu
->pio_data
,
1609 r
= pio_copy_data(vcpu
);
1611 kvm_arch_ops
->cache_regs(vcpu
);
1618 delta
*= io
->cur_count
;
1620 * The size of the register should really depend on
1621 * current address size.
1623 vcpu
->regs
[VCPU_REGS_RCX
] -= delta
;
1629 vcpu
->regs
[VCPU_REGS_RDI
] += delta
;
1631 vcpu
->regs
[VCPU_REGS_RSI
] += delta
;
1634 vcpu
->run
->io_completed
= 0;
1636 kvm_arch_ops
->decache_regs(vcpu
);
1638 io
->count
-= io
->cur_count
;
1642 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1646 int kvm_setup_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
1647 int size
, unsigned long count
, int string
, int down
,
1648 gva_t address
, int rep
, unsigned port
)
1650 unsigned now
, in_page
;
1655 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
1656 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
1657 vcpu
->run
->io
.size
= size
;
1658 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
1659 vcpu
->run
->io
.count
= count
;
1660 vcpu
->run
->io
.port
= port
;
1661 vcpu
->pio
.count
= count
;
1662 vcpu
->pio
.cur_count
= count
;
1663 vcpu
->pio
.size
= size
;
1665 vcpu
->pio
.string
= string
;
1666 vcpu
->pio
.down
= down
;
1667 vcpu
->pio
.guest_page_offset
= offset_in_page(address
);
1668 vcpu
->pio
.rep
= rep
;
1671 kvm_arch_ops
->cache_regs(vcpu
);
1672 memcpy(vcpu
->pio_data
, &vcpu
->regs
[VCPU_REGS_RAX
], 4);
1673 kvm_arch_ops
->decache_regs(vcpu
);
1678 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1682 now
= min(count
, PAGE_SIZE
/ size
);
1685 in_page
= PAGE_SIZE
- offset_in_page(address
);
1687 in_page
= offset_in_page(address
) + size
;
1688 now
= min(count
, (unsigned long)in_page
/ size
);
1691 * String I/O straddles page boundary. Pin two guest pages
1692 * so that we satisfy atomicity constraints. Do just one
1693 * transaction to avoid complexity.
1700 * String I/O in reverse. Yuck. Kill the guest, fix later.
1702 printk(KERN_ERR
"kvm: guest string pio down\n");
1706 vcpu
->run
->io
.count
= now
;
1707 vcpu
->pio
.cur_count
= now
;
1709 for (i
= 0; i
< nr_pages
; ++i
) {
1710 spin_lock(&vcpu
->kvm
->lock
);
1711 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
1714 vcpu
->pio
.guest_pages
[i
] = page
;
1715 spin_unlock(&vcpu
->kvm
->lock
);
1718 free_pio_guest_pages(vcpu
);
1724 return pio_copy_data(vcpu
);
1727 EXPORT_SYMBOL_GPL(kvm_setup_pio
);
1729 static int kvm_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1736 if (vcpu
->sigset_active
)
1737 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1739 /* re-sync apic's tpr */
1740 vcpu
->cr8
= kvm_run
->cr8
;
1742 if (kvm_run
->io_completed
) {
1743 if (vcpu
->pio
.cur_count
) {
1744 r
= complete_pio(vcpu
);
1748 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1749 vcpu
->mmio_read_completed
= 1;
1753 vcpu
->mmio_needed
= 0;
1755 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
1756 kvm_arch_ops
->cache_regs(vcpu
);
1757 vcpu
->regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
1758 kvm_arch_ops
->decache_regs(vcpu
);
1761 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1764 if (vcpu
->sigset_active
)
1765 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1771 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
,
1772 struct kvm_regs
*regs
)
1776 kvm_arch_ops
->cache_regs(vcpu
);
1778 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1779 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1780 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1781 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1782 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1783 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1784 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1785 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1786 #ifdef CONFIG_X86_64
1787 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1788 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1789 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1790 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1791 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1792 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1793 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1794 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1797 regs
->rip
= vcpu
->rip
;
1798 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1801 * Don't leak debug flags in case they were set for guest debugging
1803 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1804 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1811 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
,
1812 struct kvm_regs
*regs
)
1816 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1817 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1818 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1819 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1820 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1821 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1822 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1823 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1824 #ifdef CONFIG_X86_64
1825 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1826 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1827 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1828 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1829 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1830 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1831 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1832 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1835 vcpu
->rip
= regs
->rip
;
1836 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1838 kvm_arch_ops
->decache_regs(vcpu
);
1845 static void get_segment(struct kvm_vcpu
*vcpu
,
1846 struct kvm_segment
*var
, int seg
)
1848 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1851 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
1852 struct kvm_sregs
*sregs
)
1854 struct descriptor_table dt
;
1858 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1859 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1860 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1861 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1862 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1863 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1865 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1866 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1868 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1869 sregs
->idt
.limit
= dt
.limit
;
1870 sregs
->idt
.base
= dt
.base
;
1871 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1872 sregs
->gdt
.limit
= dt
.limit
;
1873 sregs
->gdt
.base
= dt
.base
;
1875 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1876 sregs
->cr0
= vcpu
->cr0
;
1877 sregs
->cr2
= vcpu
->cr2
;
1878 sregs
->cr3
= vcpu
->cr3
;
1879 sregs
->cr4
= vcpu
->cr4
;
1880 sregs
->cr8
= vcpu
->cr8
;
1881 sregs
->efer
= vcpu
->shadow_efer
;
1882 sregs
->apic_base
= vcpu
->apic_base
;
1884 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1885 sizeof sregs
->interrupt_bitmap
);
1892 static void set_segment(struct kvm_vcpu
*vcpu
,
1893 struct kvm_segment
*var
, int seg
)
1895 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1898 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
1899 struct kvm_sregs
*sregs
)
1901 int mmu_reset_needed
= 0;
1903 struct descriptor_table dt
;
1907 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1908 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1909 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1910 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1911 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1912 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1914 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1915 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1917 dt
.limit
= sregs
->idt
.limit
;
1918 dt
.base
= sregs
->idt
.base
;
1919 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1920 dt
.limit
= sregs
->gdt
.limit
;
1921 dt
.base
= sregs
->gdt
.base
;
1922 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1924 vcpu
->cr2
= sregs
->cr2
;
1925 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
1926 vcpu
->cr3
= sregs
->cr3
;
1928 vcpu
->cr8
= sregs
->cr8
;
1930 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
1931 #ifdef CONFIG_X86_64
1932 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
1934 vcpu
->apic_base
= sregs
->apic_base
;
1936 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1938 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
1939 kvm_arch_ops
->set_cr0_no_modeswitch(vcpu
, sregs
->cr0
);
1941 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
1942 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
1943 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
1944 load_pdptrs(vcpu
, vcpu
->cr3
);
1946 if (mmu_reset_needed
)
1947 kvm_mmu_reset_context(vcpu
);
1949 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
1950 sizeof vcpu
->irq_pending
);
1951 vcpu
->irq_summary
= 0;
1952 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
1953 if (vcpu
->irq_pending
[i
])
1954 __set_bit(i
, &vcpu
->irq_summary
);
1962 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1963 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1965 * This list is modified at module load time to reflect the
1966 * capabilities of the host cpu.
1968 static u32 msrs_to_save
[] = {
1969 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1971 #ifdef CONFIG_X86_64
1972 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1974 MSR_IA32_TIME_STAMP_COUNTER
,
1977 static unsigned num_msrs_to_save
;
1979 static u32 emulated_msrs
[] = {
1980 MSR_IA32_MISC_ENABLE
,
1983 static __init
void kvm_init_msr_list(void)
1988 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
1989 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
1992 msrs_to_save
[j
] = msrs_to_save
[i
];
1995 num_msrs_to_save
= j
;
1999 * Adapt set_msr() to msr_io()'s calling convention
2001 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
2003 return set_msr(vcpu
, index
, *data
);
2007 * Read or write a bunch of msrs. All parameters are kernel addresses.
2009 * @return number of msrs set successfully.
2011 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2012 struct kvm_msr_entry
*entries
,
2013 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2014 unsigned index
, u64
*data
))
2020 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2021 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2030 * Read or write a bunch of msrs. Parameters are user addresses.
2032 * @return number of msrs set successfully.
2034 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2035 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2036 unsigned index
, u64
*data
),
2039 struct kvm_msrs msrs
;
2040 struct kvm_msr_entry
*entries
;
2045 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2049 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2053 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2054 entries
= vmalloc(size
);
2059 if (copy_from_user(entries
, user_msrs
->entries
, size
))
2062 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2067 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2079 * Translate a guest virtual address to a guest physical address.
2081 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
2082 struct kvm_translation
*tr
)
2084 unsigned long vaddr
= tr
->linear_address
;
2088 spin_lock(&vcpu
->kvm
->lock
);
2089 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
2090 tr
->physical_address
= gpa
;
2091 tr
->valid
= gpa
!= UNMAPPED_GVA
;
2094 spin_unlock(&vcpu
->kvm
->lock
);
2100 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2101 struct kvm_interrupt
*irq
)
2103 if (irq
->irq
< 0 || irq
->irq
>= 256)
2107 set_bit(irq
->irq
, vcpu
->irq_pending
);
2108 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
2115 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
2116 struct kvm_debug_guest
*dbg
)
2122 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
2129 static struct page
*kvm_vcpu_nopage(struct vm_area_struct
*vma
,
2130 unsigned long address
,
2133 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2134 unsigned long pgoff
;
2137 *type
= VM_FAULT_MINOR
;
2138 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2140 page
= virt_to_page(vcpu
->run
);
2141 else if (pgoff
== KVM_PIO_PAGE_OFFSET
)
2142 page
= virt_to_page(vcpu
->pio_data
);
2144 return NOPAGE_SIGBUS
;
2149 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
2150 .nopage
= kvm_vcpu_nopage
,
2153 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2155 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2159 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2161 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2163 fput(vcpu
->kvm
->filp
);
2167 static struct file_operations kvm_vcpu_fops
= {
2168 .release
= kvm_vcpu_release
,
2169 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2170 .compat_ioctl
= kvm_vcpu_ioctl
,
2171 .mmap
= kvm_vcpu_mmap
,
2175 * Allocates an inode for the vcpu.
2177 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2180 struct inode
*inode
;
2183 atomic_inc(&vcpu
->kvm
->filp
->f_count
);
2184 inode
= kvmfs_inode(&kvm_vcpu_fops
);
2185 if (IS_ERR(inode
)) {
2190 file
= kvmfs_file(inode
, vcpu
);
2196 r
= get_unused_fd();
2200 fd_install(fd
, file
);
2209 fput(vcpu
->kvm
->filp
);
2214 * Creates some virtual cpus. Good luck creating more than one.
2216 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
2219 struct kvm_vcpu
*vcpu
;
2226 vcpu
= &kvm
->vcpus
[n
];
2228 mutex_lock(&vcpu
->mutex
);
2231 mutex_unlock(&vcpu
->mutex
);
2235 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2239 vcpu
->run
= page_address(page
);
2241 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2245 vcpu
->pio_data
= page_address(page
);
2247 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
2249 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
2251 r
= kvm_arch_ops
->vcpu_create(vcpu
);
2253 goto out_free_vcpus
;
2255 r
= kvm_mmu_create(vcpu
);
2257 goto out_free_vcpus
;
2259 kvm_arch_ops
->vcpu_load(vcpu
);
2260 r
= kvm_mmu_setup(vcpu
);
2262 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
2266 goto out_free_vcpus
;
2268 r
= create_vcpu_fd(vcpu
);
2270 goto out_free_vcpus
;
2275 kvm_free_vcpu(vcpu
);
2277 free_page((unsigned long)vcpu
->run
);
2280 mutex_unlock(&vcpu
->mutex
);
2285 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
2286 struct kvm_cpuid
*cpuid
,
2287 struct kvm_cpuid_entry __user
*entries
)
2292 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2295 if (copy_from_user(&vcpu
->cpuid_entries
, entries
,
2296 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
2298 vcpu
->cpuid_nent
= cpuid
->nent
;
2305 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2308 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2309 vcpu
->sigset_active
= 1;
2310 vcpu
->sigset
= *sigset
;
2312 vcpu
->sigset_active
= 0;
2316 static long kvm_vcpu_ioctl(struct file
*filp
,
2317 unsigned int ioctl
, unsigned long arg
)
2319 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2320 void __user
*argp
= (void __user
*)arg
;
2328 r
= kvm_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2330 case KVM_GET_REGS
: {
2331 struct kvm_regs kvm_regs
;
2333 memset(&kvm_regs
, 0, sizeof kvm_regs
);
2334 r
= kvm_vcpu_ioctl_get_regs(vcpu
, &kvm_regs
);
2338 if (copy_to_user(argp
, &kvm_regs
, sizeof kvm_regs
))
2343 case KVM_SET_REGS
: {
2344 struct kvm_regs kvm_regs
;
2347 if (copy_from_user(&kvm_regs
, argp
, sizeof kvm_regs
))
2349 r
= kvm_vcpu_ioctl_set_regs(vcpu
, &kvm_regs
);
2355 case KVM_GET_SREGS
: {
2356 struct kvm_sregs kvm_sregs
;
2358 memset(&kvm_sregs
, 0, sizeof kvm_sregs
);
2359 r
= kvm_vcpu_ioctl_get_sregs(vcpu
, &kvm_sregs
);
2363 if (copy_to_user(argp
, &kvm_sregs
, sizeof kvm_sregs
))
2368 case KVM_SET_SREGS
: {
2369 struct kvm_sregs kvm_sregs
;
2372 if (copy_from_user(&kvm_sregs
, argp
, sizeof kvm_sregs
))
2374 r
= kvm_vcpu_ioctl_set_sregs(vcpu
, &kvm_sregs
);
2380 case KVM_TRANSLATE
: {
2381 struct kvm_translation tr
;
2384 if (copy_from_user(&tr
, argp
, sizeof tr
))
2386 r
= kvm_vcpu_ioctl_translate(vcpu
, &tr
);
2390 if (copy_to_user(argp
, &tr
, sizeof tr
))
2395 case KVM_INTERRUPT
: {
2396 struct kvm_interrupt irq
;
2399 if (copy_from_user(&irq
, argp
, sizeof irq
))
2401 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2407 case KVM_DEBUG_GUEST
: {
2408 struct kvm_debug_guest dbg
;
2411 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2413 r
= kvm_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
2420 r
= msr_io(vcpu
, argp
, get_msr
, 1);
2423 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2425 case KVM_SET_CPUID
: {
2426 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2427 struct kvm_cpuid cpuid
;
2430 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2432 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2437 case KVM_SET_SIGNAL_MASK
: {
2438 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2439 struct kvm_signal_mask kvm_sigmask
;
2440 sigset_t sigset
, *p
;
2445 if (copy_from_user(&kvm_sigmask
, argp
,
2446 sizeof kvm_sigmask
))
2449 if (kvm_sigmask
.len
!= sizeof sigset
)
2452 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2457 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2467 static long kvm_vm_ioctl(struct file
*filp
,
2468 unsigned int ioctl
, unsigned long arg
)
2470 struct kvm
*kvm
= filp
->private_data
;
2471 void __user
*argp
= (void __user
*)arg
;
2475 case KVM_CREATE_VCPU
:
2476 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2480 case KVM_SET_MEMORY_REGION
: {
2481 struct kvm_memory_region kvm_mem
;
2484 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2486 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_mem
);
2491 case KVM_GET_DIRTY_LOG
: {
2492 struct kvm_dirty_log log
;
2495 if (copy_from_user(&log
, argp
, sizeof log
))
2497 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2509 static struct page
*kvm_vm_nopage(struct vm_area_struct
*vma
,
2510 unsigned long address
,
2513 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2514 unsigned long pgoff
;
2515 struct kvm_memory_slot
*slot
;
2518 *type
= VM_FAULT_MINOR
;
2519 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2520 slot
= gfn_to_memslot(kvm
, pgoff
);
2522 return NOPAGE_SIGBUS
;
2523 page
= gfn_to_page(slot
, pgoff
);
2525 return NOPAGE_SIGBUS
;
2530 static struct vm_operations_struct kvm_vm_vm_ops
= {
2531 .nopage
= kvm_vm_nopage
,
2534 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2536 vma
->vm_ops
= &kvm_vm_vm_ops
;
2540 static struct file_operations kvm_vm_fops
= {
2541 .release
= kvm_vm_release
,
2542 .unlocked_ioctl
= kvm_vm_ioctl
,
2543 .compat_ioctl
= kvm_vm_ioctl
,
2544 .mmap
= kvm_vm_mmap
,
2547 static int kvm_dev_ioctl_create_vm(void)
2550 struct inode
*inode
;
2554 inode
= kvmfs_inode(&kvm_vm_fops
);
2555 if (IS_ERR(inode
)) {
2560 kvm
= kvm_create_vm();
2566 file
= kvmfs_file(inode
, kvm
);
2573 r
= get_unused_fd();
2577 fd_install(fd
, file
);
2584 kvm_destroy_vm(kvm
);
2591 static long kvm_dev_ioctl(struct file
*filp
,
2592 unsigned int ioctl
, unsigned long arg
)
2594 void __user
*argp
= (void __user
*)arg
;
2598 case KVM_GET_API_VERSION
:
2602 r
= KVM_API_VERSION
;
2608 r
= kvm_dev_ioctl_create_vm();
2610 case KVM_GET_MSR_INDEX_LIST
: {
2611 struct kvm_msr_list __user
*user_msr_list
= argp
;
2612 struct kvm_msr_list msr_list
;
2616 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2619 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2620 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2623 if (n
< num_msrs_to_save
)
2626 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2627 num_msrs_to_save
* sizeof(u32
)))
2629 if (copy_to_user(user_msr_list
->indices
2630 + num_msrs_to_save
* sizeof(u32
),
2632 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2637 case KVM_CHECK_EXTENSION
:
2639 * No extensions defined at present.
2643 case KVM_GET_VCPU_MMAP_SIZE
:
2656 static struct file_operations kvm_chardev_ops
= {
2657 .open
= kvm_dev_open
,
2658 .release
= kvm_dev_release
,
2659 .unlocked_ioctl
= kvm_dev_ioctl
,
2660 .compat_ioctl
= kvm_dev_ioctl
,
2663 static struct miscdevice kvm_dev
= {
2669 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2672 if (val
== SYS_RESTART
) {
2674 * Some (well, at least mine) BIOSes hang on reboot if
2677 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2678 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2683 static struct notifier_block kvm_reboot_notifier
= {
2684 .notifier_call
= kvm_reboot
,
2689 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2692 static void decache_vcpus_on_cpu(int cpu
)
2695 struct kvm_vcpu
*vcpu
;
2698 spin_lock(&kvm_lock
);
2699 list_for_each_entry(vm
, &vm_list
, vm_list
)
2700 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2701 vcpu
= &vm
->vcpus
[i
];
2703 * If the vcpu is locked, then it is running on some
2704 * other cpu and therefore it is not cached on the
2707 * If it's not locked, check the last cpu it executed
2710 if (mutex_trylock(&vcpu
->mutex
)) {
2711 if (vcpu
->cpu
== cpu
) {
2712 kvm_arch_ops
->vcpu_decache(vcpu
);
2715 mutex_unlock(&vcpu
->mutex
);
2718 spin_unlock(&kvm_lock
);
2721 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2727 case CPU_DOWN_PREPARE
:
2728 case CPU_UP_CANCELED
:
2729 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2731 decache_vcpus_on_cpu(cpu
);
2732 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_disable
,
2736 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2738 smp_call_function_single(cpu
, kvm_arch_ops
->hardware_enable
,
2745 static struct notifier_block kvm_cpu_notifier
= {
2746 .notifier_call
= kvm_cpu_hotplug
,
2747 .priority
= 20, /* must be > scheduler priority */
2750 static __init
void kvm_init_debug(void)
2752 struct kvm_stats_debugfs_item
*p
;
2754 debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2755 for (p
= debugfs_entries
; p
->name
; ++p
)
2756 p
->dentry
= debugfs_create_u32(p
->name
, 0444, debugfs_dir
,
2760 static void kvm_exit_debug(void)
2762 struct kvm_stats_debugfs_item
*p
;
2764 for (p
= debugfs_entries
; p
->name
; ++p
)
2765 debugfs_remove(p
->dentry
);
2766 debugfs_remove(debugfs_dir
);
2769 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2771 decache_vcpus_on_cpu(raw_smp_processor_id());
2772 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2776 static int kvm_resume(struct sys_device
*dev
)
2778 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
2782 static struct sysdev_class kvm_sysdev_class
= {
2783 set_kset_name("kvm"),
2784 .suspend
= kvm_suspend
,
2785 .resume
= kvm_resume
,
2788 static struct sys_device kvm_sysdev
= {
2790 .cls
= &kvm_sysdev_class
,
2793 hpa_t bad_page_address
;
2795 static int kvmfs_get_sb(struct file_system_type
*fs_type
, int flags
,
2796 const char *dev_name
, void *data
, struct vfsmount
*mnt
)
2798 return get_sb_pseudo(fs_type
, "kvm:", NULL
, KVMFS_SUPER_MAGIC
, mnt
);
2801 static struct file_system_type kvm_fs_type
= {
2803 .get_sb
= kvmfs_get_sb
,
2804 .kill_sb
= kill_anon_super
,
2807 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
2812 printk(KERN_ERR
"kvm: already loaded the other module\n");
2816 if (!ops
->cpu_has_kvm_support()) {
2817 printk(KERN_ERR
"kvm: no hardware support\n");
2820 if (ops
->disabled_by_bios()) {
2821 printk(KERN_ERR
"kvm: disabled by bios\n");
2827 r
= kvm_arch_ops
->hardware_setup();
2831 on_each_cpu(kvm_arch_ops
->hardware_enable
, NULL
, 0, 1);
2832 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2835 register_reboot_notifier(&kvm_reboot_notifier
);
2837 r
= sysdev_class_register(&kvm_sysdev_class
);
2841 r
= sysdev_register(&kvm_sysdev
);
2845 kvm_chardev_ops
.owner
= module
;
2847 r
= misc_register(&kvm_dev
);
2849 printk (KERN_ERR
"kvm: misc device register failed\n");
2856 sysdev_unregister(&kvm_sysdev
);
2858 sysdev_class_unregister(&kvm_sysdev_class
);
2860 unregister_reboot_notifier(&kvm_reboot_notifier
);
2861 unregister_cpu_notifier(&kvm_cpu_notifier
);
2863 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2864 kvm_arch_ops
->hardware_unsetup();
2866 kvm_arch_ops
= NULL
;
2870 void kvm_exit_arch(void)
2872 misc_deregister(&kvm_dev
);
2873 sysdev_unregister(&kvm_sysdev
);
2874 sysdev_class_unregister(&kvm_sysdev_class
);
2875 unregister_reboot_notifier(&kvm_reboot_notifier
);
2876 unregister_cpu_notifier(&kvm_cpu_notifier
);
2877 on_each_cpu(kvm_arch_ops
->hardware_disable
, NULL
, 0, 1);
2878 kvm_arch_ops
->hardware_unsetup();
2879 kvm_arch_ops
= NULL
;
2882 static __init
int kvm_init(void)
2884 static struct page
*bad_page
;
2887 r
= register_filesystem(&kvm_fs_type
);
2891 kvmfs_mnt
= kern_mount(&kvm_fs_type
);
2892 r
= PTR_ERR(kvmfs_mnt
);
2893 if (IS_ERR(kvmfs_mnt
))
2897 kvm_init_msr_list();
2899 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
2904 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
2905 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
2913 unregister_filesystem(&kvm_fs_type
);
2918 static __exit
void kvm_exit(void)
2921 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
2923 unregister_filesystem(&kvm_fs_type
);
2926 module_init(kvm_init
)
2927 module_exit(kvm_exit
)
2929 EXPORT_SYMBOL_GPL(kvm_init_arch
);
2930 EXPORT_SYMBOL_GPL(kvm_exit_arch
);