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.
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <linux/reboot.h>
31 #include <linux/debugfs.h>
32 #include <linux/highmem.h>
33 #include <linux/file.h>
34 #include <linux/sysdev.h>
35 #include <linux/cpu.h>
36 #include <linux/sched.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
41 #include <asm/processor.h>
44 #include <asm/uaccess.h>
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static DEFINE_SPINLOCK(kvm_lock
);
51 static LIST_HEAD(vm_list
);
53 static cpumask_t cpus_hardware_enabled
;
55 struct kvm_arch_ops
*kvm_arch_ops
;
57 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
59 static struct kvm_stats_debugfs_item
{
62 struct dentry
*dentry
;
63 } debugfs_entries
[] = {
64 { "pf_fixed", STAT_OFFSET(pf_fixed
) },
65 { "pf_guest", STAT_OFFSET(pf_guest
) },
66 { "tlb_flush", STAT_OFFSET(tlb_flush
) },
67 { "invlpg", STAT_OFFSET(invlpg
) },
68 { "exits", STAT_OFFSET(exits
) },
69 { "io_exits", STAT_OFFSET(io_exits
) },
70 { "mmio_exits", STAT_OFFSET(mmio_exits
) },
71 { "signal_exits", STAT_OFFSET(signal_exits
) },
72 { "irq_window", STAT_OFFSET(irq_window_exits
) },
73 { "halt_exits", STAT_OFFSET(halt_exits
) },
74 { "request_irq", STAT_OFFSET(request_irq_exits
) },
75 { "irq_exits", STAT_OFFSET(irq_exits
) },
76 { "light_exits", STAT_OFFSET(light_exits
) },
77 { "efer_reload", STAT_OFFSET(efer_reload
) },
81 static struct dentry
*debugfs_dir
;
83 #define MAX_IO_MSRS 256
85 #define CR0_RESERVED_BITS \
86 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
87 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
88 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
89 #define LMSW_GUEST_MASK 0x0eULL
90 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
91 #define CR8_RESEVED_BITS (~0x0fULL)
92 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
95 // LDT or TSS descriptor in the GDT. 16 bytes.
96 struct segment_descriptor_64
{
97 struct segment_descriptor s
;
104 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
107 unsigned long segment_base(u16 selector
)
109 struct descriptor_table gdt
;
110 struct segment_descriptor
*d
;
111 unsigned long table_base
;
112 typedef unsigned long ul
;
118 asm ("sgdt %0" : "=m"(gdt
));
119 table_base
= gdt
.base
;
121 if (selector
& 4) { /* from ldt */
124 asm ("sldt %0" : "=g"(ldt_selector
));
125 table_base
= segment_base(ldt_selector
);
127 d
= (struct segment_descriptor
*)(table_base
+ (selector
& ~7));
128 v
= d
->base_low
| ((ul
)d
->base_mid
<< 16) | ((ul
)d
->base_high
<< 24);
131 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
132 v
|= ((ul
)((struct segment_descriptor_64
*)d
)->base_higher
) << 32;
136 EXPORT_SYMBOL_GPL(segment_base
);
138 static inline int valid_vcpu(int n
)
140 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
143 int kvm_read_guest(struct kvm_vcpu
*vcpu
, gva_t addr
, unsigned long size
,
146 unsigned char *host_buf
= dest
;
147 unsigned long req_size
= size
;
155 paddr
= gva_to_hpa(vcpu
, addr
);
157 if (is_error_hpa(paddr
))
160 guest_buf
= (hva_t
)kmap_atomic(
161 pfn_to_page(paddr
>> PAGE_SHIFT
),
163 offset
= addr
& ~PAGE_MASK
;
165 now
= min(size
, PAGE_SIZE
- offset
);
166 memcpy(host_buf
, (void*)guest_buf
, now
);
170 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
172 return req_size
- size
;
174 EXPORT_SYMBOL_GPL(kvm_read_guest
);
176 int kvm_write_guest(struct kvm_vcpu
*vcpu
, gva_t addr
, unsigned long size
,
179 unsigned char *host_buf
= data
;
180 unsigned long req_size
= size
;
189 paddr
= gva_to_hpa(vcpu
, addr
);
191 if (is_error_hpa(paddr
))
194 gfn
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
) >> PAGE_SHIFT
;
195 mark_page_dirty(vcpu
->kvm
, gfn
);
196 guest_buf
= (hva_t
)kmap_atomic(
197 pfn_to_page(paddr
>> PAGE_SHIFT
), KM_USER0
);
198 offset
= addr
& ~PAGE_MASK
;
200 now
= min(size
, PAGE_SIZE
- offset
);
201 memcpy((void*)guest_buf
, host_buf
, now
);
205 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
207 return req_size
- size
;
209 EXPORT_SYMBOL_GPL(kvm_write_guest
);
211 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
213 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
216 vcpu
->guest_fpu_loaded
= 1;
217 fx_save(vcpu
->host_fx_image
);
218 fx_restore(vcpu
->guest_fx_image
);
220 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
222 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
224 if (!vcpu
->guest_fpu_loaded
)
227 vcpu
->guest_fpu_loaded
= 0;
228 fx_save(vcpu
->guest_fx_image
);
229 fx_restore(vcpu
->host_fx_image
);
231 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
234 * Switches to specified vcpu, until a matching vcpu_put()
236 static void vcpu_load(struct kvm_vcpu
*vcpu
)
238 mutex_lock(&vcpu
->mutex
);
239 kvm_arch_ops
->vcpu_load(vcpu
);
242 static void vcpu_put(struct kvm_vcpu
*vcpu
)
244 kvm_arch_ops
->vcpu_put(vcpu
);
245 mutex_unlock(&vcpu
->mutex
);
248 static void ack_flush(void *_completed
)
250 atomic_t
*completed
= _completed
;
252 atomic_inc(completed
);
255 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
259 struct kvm_vcpu
*vcpu
;
262 atomic_set(&completed
, 0);
265 for (i
= 0; i
< kvm
->nvcpus
; ++i
) {
266 vcpu
= &kvm
->vcpus
[i
];
267 if (test_and_set_bit(KVM_TLB_FLUSH
, &vcpu
->requests
))
270 if (cpu
!= -1 && cpu
!= raw_smp_processor_id())
271 if (!cpu_isset(cpu
, cpus
)) {
278 * We really want smp_call_function_mask() here. But that's not
279 * available, so ipi all cpus in parallel and wait for them
282 for (cpu
= first_cpu(cpus
); cpu
!= NR_CPUS
; cpu
= next_cpu(cpu
, cpus
))
283 smp_call_function_single(cpu
, ack_flush
, &completed
, 1, 0);
284 while (atomic_read(&completed
) != needed
) {
290 static struct kvm
*kvm_create_vm(void)
292 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
296 return ERR_PTR(-ENOMEM
);
298 kvm_io_bus_init(&kvm
->pio_bus
);
299 spin_lock_init(&kvm
->lock
);
300 INIT_LIST_HEAD(&kvm
->active_mmu_pages
);
301 kvm_io_bus_init(&kvm
->mmio_bus
);
302 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
303 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[i
];
305 mutex_init(&vcpu
->mutex
);
308 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
310 spin_lock(&kvm_lock
);
311 list_add(&kvm
->vm_list
, &vm_list
);
312 spin_unlock(&kvm_lock
);
316 static int kvm_dev_open(struct inode
*inode
, struct file
*filp
)
322 * Free any memory in @free but not in @dont.
324 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
325 struct kvm_memory_slot
*dont
)
329 if (!dont
|| free
->phys_mem
!= dont
->phys_mem
)
330 if (free
->phys_mem
) {
331 for (i
= 0; i
< free
->npages
; ++i
)
332 if (free
->phys_mem
[i
])
333 __free_page(free
->phys_mem
[i
]);
334 vfree(free
->phys_mem
);
337 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
338 vfree(free
->dirty_bitmap
);
340 free
->phys_mem
= NULL
;
342 free
->dirty_bitmap
= NULL
;
345 static void kvm_free_physmem(struct kvm
*kvm
)
349 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
350 kvm_free_physmem_slot(&kvm
->memslots
[i
], NULL
);
353 static void free_pio_guest_pages(struct kvm_vcpu
*vcpu
)
357 for (i
= 0; i
< 2; ++i
)
358 if (vcpu
->pio
.guest_pages
[i
]) {
359 __free_page(vcpu
->pio
.guest_pages
[i
]);
360 vcpu
->pio
.guest_pages
[i
] = NULL
;
364 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
370 kvm_mmu_unload(vcpu
);
374 static void kvm_free_vcpu(struct kvm_vcpu
*vcpu
)
380 kvm_mmu_destroy(vcpu
);
382 kvm_arch_ops
->vcpu_free(vcpu
);
383 free_page((unsigned long)vcpu
->run
);
385 free_page((unsigned long)vcpu
->pio_data
);
386 vcpu
->pio_data
= NULL
;
387 free_pio_guest_pages(vcpu
);
390 static void kvm_free_vcpus(struct kvm
*kvm
)
395 * Unpin any mmu pages first.
397 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
398 kvm_unload_vcpu_mmu(&kvm
->vcpus
[i
]);
399 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
400 kvm_free_vcpu(&kvm
->vcpus
[i
]);
403 static int kvm_dev_release(struct inode
*inode
, struct file
*filp
)
408 static void kvm_destroy_vm(struct kvm
*kvm
)
410 spin_lock(&kvm_lock
);
411 list_del(&kvm
->vm_list
);
412 spin_unlock(&kvm_lock
);
413 kvm_io_bus_destroy(&kvm
->pio_bus
);
414 kvm_io_bus_destroy(&kvm
->mmio_bus
);
416 kvm_free_physmem(kvm
);
420 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
422 struct kvm
*kvm
= filp
->private_data
;
428 static void inject_gp(struct kvm_vcpu
*vcpu
)
430 kvm_arch_ops
->inject_gp(vcpu
, 0);
434 * Load the pae pdptrs. Return true is they are all valid.
436 static int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
438 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
439 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
446 spin_lock(&vcpu
->kvm
->lock
);
447 page
= gfn_to_page(vcpu
->kvm
, pdpt_gfn
);
448 /* FIXME: !page - emulate? 0xff? */
449 pdpt
= kmap_atomic(page
, KM_USER0
);
452 for (i
= 0; i
< 4; ++i
) {
453 pdpte
= pdpt
[offset
+ i
];
454 if ((pdpte
& 1) && (pdpte
& 0xfffffff0000001e6ull
)) {
460 for (i
= 0; i
< 4; ++i
)
461 vcpu
->pdptrs
[i
] = pdpt
[offset
+ i
];
464 kunmap_atomic(pdpt
, KM_USER0
);
465 spin_unlock(&vcpu
->kvm
->lock
);
470 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
472 if (cr0
& CR0_RESERVED_BITS
) {
473 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
479 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
480 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
485 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
486 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
487 "and a clear PE flag\n");
492 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
494 if ((vcpu
->shadow_efer
& EFER_LME
)) {
498 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
499 "in long mode while PAE is disabled\n");
503 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
505 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
506 "in long mode while CS.L == 1\n");
513 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
514 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
522 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
525 spin_lock(&vcpu
->kvm
->lock
);
526 kvm_mmu_reset_context(vcpu
);
527 spin_unlock(&vcpu
->kvm
->lock
);
530 EXPORT_SYMBOL_GPL(set_cr0
);
532 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
534 set_cr0(vcpu
, (vcpu
->cr0
& ~0x0ful
) | (msw
& 0x0f));
536 EXPORT_SYMBOL_GPL(lmsw
);
538 void set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
540 if (cr4
& CR4_RESEVED_BITS
) {
541 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
546 if (is_long_mode(vcpu
)) {
547 if (!(cr4
& CR4_PAE_MASK
)) {
548 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
553 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& CR4_PAE_MASK
)
554 && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
555 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
559 if (cr4
& CR4_VMXE_MASK
) {
560 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
564 kvm_arch_ops
->set_cr4(vcpu
, cr4
);
565 spin_lock(&vcpu
->kvm
->lock
);
566 kvm_mmu_reset_context(vcpu
);
567 spin_unlock(&vcpu
->kvm
->lock
);
569 EXPORT_SYMBOL_GPL(set_cr4
);
571 void set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
573 if (is_long_mode(vcpu
)) {
574 if (cr3
& CR3_L_MODE_RESEVED_BITS
) {
575 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
580 if (cr3
& CR3_RESEVED_BITS
) {
581 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
585 if (is_paging(vcpu
) && is_pae(vcpu
) &&
586 !load_pdptrs(vcpu
, cr3
)) {
587 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
595 spin_lock(&vcpu
->kvm
->lock
);
597 * Does the new cr3 value map to physical memory? (Note, we
598 * catch an invalid cr3 even in real-mode, because it would
599 * cause trouble later on when we turn on paging anyway.)
601 * A real CPU would silently accept an invalid cr3 and would
602 * attempt to use it - with largely undefined (and often hard
603 * to debug) behavior on the guest side.
605 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
608 vcpu
->mmu
.new_cr3(vcpu
);
609 spin_unlock(&vcpu
->kvm
->lock
);
611 EXPORT_SYMBOL_GPL(set_cr3
);
613 void set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
615 if ( cr8
& CR8_RESEVED_BITS
) {
616 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
622 EXPORT_SYMBOL_GPL(set_cr8
);
624 void fx_init(struct kvm_vcpu
*vcpu
)
626 struct __attribute__ ((__packed__
)) fx_image_s
{
632 u64 operand
;// fpu dp
638 fx_save(vcpu
->host_fx_image
);
640 fx_save(vcpu
->guest_fx_image
);
641 fx_restore(vcpu
->host_fx_image
);
643 fx_image
= (struct fx_image_s
*)vcpu
->guest_fx_image
;
644 fx_image
->mxcsr
= 0x1f80;
645 memset(vcpu
->guest_fx_image
+ sizeof(struct fx_image_s
),
646 0, FX_IMAGE_SIZE
- sizeof(struct fx_image_s
));
648 EXPORT_SYMBOL_GPL(fx_init
);
651 * Allocate some memory and give it an address in the guest physical address
654 * Discontiguous memory is allowed, mostly for framebuffers.
656 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
657 struct kvm_memory_region
*mem
)
661 unsigned long npages
;
663 struct kvm_memory_slot
*memslot
;
664 struct kvm_memory_slot old
, new;
665 int memory_config_version
;
668 /* General sanity checks */
669 if (mem
->memory_size
& (PAGE_SIZE
- 1))
671 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
673 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
675 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
678 memslot
= &kvm
->memslots
[mem
->slot
];
679 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
680 npages
= mem
->memory_size
>> PAGE_SHIFT
;
683 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
686 spin_lock(&kvm
->lock
);
688 memory_config_version
= kvm
->memory_config_version
;
689 new = old
= *memslot
;
691 new.base_gfn
= base_gfn
;
693 new.flags
= mem
->flags
;
695 /* Disallow changing a memory slot's size. */
697 if (npages
&& old
.npages
&& npages
!= old
.npages
)
700 /* Check for overlaps */
702 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
703 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
707 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
708 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
712 * Do memory allocations outside lock. memory_config_version will
715 spin_unlock(&kvm
->lock
);
717 /* Deallocate if slot is being removed */
721 /* Free page dirty bitmap if unneeded */
722 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
723 new.dirty_bitmap
= NULL
;
727 /* Allocate if a slot is being created */
728 if (npages
&& !new.phys_mem
) {
729 new.phys_mem
= vmalloc(npages
* sizeof(struct page
*));
734 memset(new.phys_mem
, 0, npages
* sizeof(struct page
*));
735 for (i
= 0; i
< npages
; ++i
) {
736 new.phys_mem
[i
] = alloc_page(GFP_HIGHUSER
738 if (!new.phys_mem
[i
])
740 set_page_private(new.phys_mem
[i
],0);
744 /* Allocate page dirty bitmap if needed */
745 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
746 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
748 new.dirty_bitmap
= vmalloc(dirty_bytes
);
749 if (!new.dirty_bitmap
)
751 memset(new.dirty_bitmap
, 0, dirty_bytes
);
754 spin_lock(&kvm
->lock
);
756 if (memory_config_version
!= kvm
->memory_config_version
) {
757 spin_unlock(&kvm
->lock
);
758 kvm_free_physmem_slot(&new, &old
);
766 if (mem
->slot
>= kvm
->nmemslots
)
767 kvm
->nmemslots
= mem
->slot
+ 1;
770 ++kvm
->memory_config_version
;
772 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
773 kvm_flush_remote_tlbs(kvm
);
775 spin_unlock(&kvm
->lock
);
777 kvm_free_physmem_slot(&old
, &new);
781 spin_unlock(&kvm
->lock
);
783 kvm_free_physmem_slot(&new, &old
);
789 * Get (and clear) the dirty memory log for a memory slot.
791 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
792 struct kvm_dirty_log
*log
)
794 struct kvm_memory_slot
*memslot
;
797 unsigned long any
= 0;
799 spin_lock(&kvm
->lock
);
802 * Prevent changes to guest memory configuration even while the lock
806 spin_unlock(&kvm
->lock
);
808 if (log
->slot
>= KVM_MEMORY_SLOTS
)
811 memslot
= &kvm
->memslots
[log
->slot
];
813 if (!memslot
->dirty_bitmap
)
816 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
818 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
819 any
= memslot
->dirty_bitmap
[i
];
822 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
825 spin_lock(&kvm
->lock
);
826 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
827 kvm_flush_remote_tlbs(kvm
);
828 memset(memslot
->dirty_bitmap
, 0, n
);
829 spin_unlock(&kvm
->lock
);
834 spin_lock(&kvm
->lock
);
836 spin_unlock(&kvm
->lock
);
841 * Set a new alias region. Aliases map a portion of physical memory into
842 * another portion. This is useful for memory windows, for example the PC
845 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
846 struct kvm_memory_alias
*alias
)
849 struct kvm_mem_alias
*p
;
852 /* General sanity checks */
853 if (alias
->memory_size
& (PAGE_SIZE
- 1))
855 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
857 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
859 if (alias
->guest_phys_addr
+ alias
->memory_size
860 < alias
->guest_phys_addr
)
862 if (alias
->target_phys_addr
+ alias
->memory_size
863 < alias
->target_phys_addr
)
866 spin_lock(&kvm
->lock
);
868 p
= &kvm
->aliases
[alias
->slot
];
869 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
870 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
871 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
873 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
874 if (kvm
->aliases
[n
- 1].npages
)
878 kvm_mmu_zap_all(kvm
);
880 spin_unlock(&kvm
->lock
);
888 static gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
891 struct kvm_mem_alias
*alias
;
893 for (i
= 0; i
< kvm
->naliases
; ++i
) {
894 alias
= &kvm
->aliases
[i
];
895 if (gfn
>= alias
->base_gfn
896 && gfn
< alias
->base_gfn
+ alias
->npages
)
897 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
902 static struct kvm_memory_slot
*__gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
906 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
907 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
909 if (gfn
>= memslot
->base_gfn
910 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
916 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
918 gfn
= unalias_gfn(kvm
, gfn
);
919 return __gfn_to_memslot(kvm
, gfn
);
922 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
924 struct kvm_memory_slot
*slot
;
926 gfn
= unalias_gfn(kvm
, gfn
);
927 slot
= __gfn_to_memslot(kvm
, gfn
);
930 return slot
->phys_mem
[gfn
- slot
->base_gfn
];
932 EXPORT_SYMBOL_GPL(gfn_to_page
);
934 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
937 struct kvm_memory_slot
*memslot
;
938 unsigned long rel_gfn
;
940 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
941 memslot
= &kvm
->memslots
[i
];
943 if (gfn
>= memslot
->base_gfn
944 && gfn
< memslot
->base_gfn
+ memslot
->npages
) {
946 if (!memslot
->dirty_bitmap
)
949 rel_gfn
= gfn
- memslot
->base_gfn
;
952 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
953 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
959 static int emulator_read_std(unsigned long addr
,
962 struct x86_emulate_ctxt
*ctxt
)
964 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
968 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
969 unsigned offset
= addr
& (PAGE_SIZE
-1);
970 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
975 if (gpa
== UNMAPPED_GVA
)
976 return X86EMUL_PROPAGATE_FAULT
;
977 pfn
= gpa
>> PAGE_SHIFT
;
978 page
= gfn_to_page(vcpu
->kvm
, pfn
);
980 return X86EMUL_UNHANDLEABLE
;
981 page_virt
= kmap_atomic(page
, KM_USER0
);
983 memcpy(data
, page_virt
+ offset
, tocopy
);
985 kunmap_atomic(page_virt
, KM_USER0
);
992 return X86EMUL_CONTINUE
;
995 static int emulator_write_std(unsigned long addr
,
998 struct x86_emulate_ctxt
*ctxt
)
1000 printk(KERN_ERR
"emulator_write_std: addr %lx n %d\n",
1002 return X86EMUL_UNHANDLEABLE
;
1005 static struct kvm_io_device
*vcpu_find_mmio_dev(struct kvm_vcpu
*vcpu
,
1009 * Note that its important to have this wrapper function because
1010 * in the very near future we will be checking for MMIOs against
1011 * the LAPIC as well as the general MMIO bus
1013 return kvm_io_bus_find_dev(&vcpu
->kvm
->mmio_bus
, addr
);
1016 static struct kvm_io_device
*vcpu_find_pio_dev(struct kvm_vcpu
*vcpu
,
1019 return kvm_io_bus_find_dev(&vcpu
->kvm
->pio_bus
, addr
);
1022 static int emulator_read_emulated(unsigned long addr
,
1025 struct x86_emulate_ctxt
*ctxt
)
1027 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1028 struct kvm_io_device
*mmio_dev
;
1031 if (vcpu
->mmio_read_completed
) {
1032 memcpy(val
, vcpu
->mmio_data
, bytes
);
1033 vcpu
->mmio_read_completed
= 0;
1034 return X86EMUL_CONTINUE
;
1035 } else if (emulator_read_std(addr
, val
, bytes
, ctxt
)
1036 == X86EMUL_CONTINUE
)
1037 return X86EMUL_CONTINUE
;
1039 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
1040 if (gpa
== UNMAPPED_GVA
)
1041 return X86EMUL_PROPAGATE_FAULT
;
1044 * Is this MMIO handled locally?
1046 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1048 kvm_iodevice_read(mmio_dev
, gpa
, bytes
, val
);
1049 return X86EMUL_CONTINUE
;
1052 vcpu
->mmio_needed
= 1;
1053 vcpu
->mmio_phys_addr
= gpa
;
1054 vcpu
->mmio_size
= bytes
;
1055 vcpu
->mmio_is_write
= 0;
1057 return X86EMUL_UNHANDLEABLE
;
1060 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1061 const void *val
, int bytes
)
1065 unsigned offset
= offset_in_page(gpa
);
1067 if (((gpa
+ bytes
- 1) >> PAGE_SHIFT
) != (gpa
>> PAGE_SHIFT
))
1069 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1072 mark_page_dirty(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1073 virt
= kmap_atomic(page
, KM_USER0
);
1074 kvm_mmu_pte_write(vcpu
, gpa
, virt
+ offset
, val
, bytes
);
1075 memcpy(virt
+ offset_in_page(gpa
), val
, bytes
);
1076 kunmap_atomic(virt
, KM_USER0
);
1080 static int emulator_write_emulated_onepage(unsigned long addr
,
1083 struct x86_emulate_ctxt
*ctxt
)
1085 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1086 struct kvm_io_device
*mmio_dev
;
1087 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
1089 if (gpa
== UNMAPPED_GVA
) {
1090 kvm_arch_ops
->inject_page_fault(vcpu
, addr
, 2);
1091 return X86EMUL_PROPAGATE_FAULT
;
1094 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
1095 return X86EMUL_CONTINUE
;
1098 * Is this MMIO handled locally?
1100 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1102 kvm_iodevice_write(mmio_dev
, gpa
, bytes
, val
);
1103 return X86EMUL_CONTINUE
;
1106 vcpu
->mmio_needed
= 1;
1107 vcpu
->mmio_phys_addr
= gpa
;
1108 vcpu
->mmio_size
= bytes
;
1109 vcpu
->mmio_is_write
= 1;
1110 memcpy(vcpu
->mmio_data
, val
, bytes
);
1112 return X86EMUL_CONTINUE
;
1115 static int emulator_write_emulated(unsigned long addr
,
1118 struct x86_emulate_ctxt
*ctxt
)
1120 /* Crossing a page boundary? */
1121 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
1124 now
= -addr
& ~PAGE_MASK
;
1125 rc
= emulator_write_emulated_onepage(addr
, val
, now
, ctxt
);
1126 if (rc
!= X86EMUL_CONTINUE
)
1132 return emulator_write_emulated_onepage(addr
, val
, bytes
, ctxt
);
1135 static int emulator_cmpxchg_emulated(unsigned long addr
,
1139 struct x86_emulate_ctxt
*ctxt
)
1141 static int reported
;
1145 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1147 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
1150 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1152 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
1155 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1157 return X86EMUL_CONTINUE
;
1160 int emulate_clts(struct kvm_vcpu
*vcpu
)
1164 cr0
= vcpu
->cr0
& ~X86_CR0_TS
;
1165 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
1166 return X86EMUL_CONTINUE
;
1169 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
1171 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1175 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
1176 return X86EMUL_CONTINUE
;
1178 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1180 return X86EMUL_UNHANDLEABLE
;
1184 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1186 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1189 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1191 /* FIXME: better handling */
1192 return X86EMUL_UNHANDLEABLE
;
1194 return X86EMUL_CONTINUE
;
1197 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1199 static int reported
;
1201 unsigned long rip
= ctxt
->vcpu
->rip
;
1202 unsigned long rip_linear
;
1204 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1209 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1211 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1212 " rip %lx %02x %02x %02x %02x\n",
1213 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1217 struct x86_emulate_ops emulate_ops
= {
1218 .read_std
= emulator_read_std
,
1219 .write_std
= emulator_write_std
,
1220 .read_emulated
= emulator_read_emulated
,
1221 .write_emulated
= emulator_write_emulated
,
1222 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1225 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1226 struct kvm_run
*run
,
1230 struct x86_emulate_ctxt emulate_ctxt
;
1234 vcpu
->mmio_fault_cr2
= cr2
;
1235 kvm_arch_ops
->cache_regs(vcpu
);
1237 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1239 emulate_ctxt
.vcpu
= vcpu
;
1240 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
1241 emulate_ctxt
.cr2
= cr2
;
1242 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1243 ? X86EMUL_MODE_REAL
: cs_l
1244 ? X86EMUL_MODE_PROT64
: cs_db
1245 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1247 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1248 emulate_ctxt
.cs_base
= 0;
1249 emulate_ctxt
.ds_base
= 0;
1250 emulate_ctxt
.es_base
= 0;
1251 emulate_ctxt
.ss_base
= 0;
1253 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1254 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1255 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1256 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1259 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1260 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1262 vcpu
->mmio_is_write
= 0;
1263 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1265 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1266 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1267 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1268 run
->mmio
.len
= vcpu
->mmio_size
;
1269 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1273 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1274 return EMULATE_DONE
;
1275 if (!vcpu
->mmio_needed
) {
1276 report_emulation_failure(&emulate_ctxt
);
1277 return EMULATE_FAIL
;
1279 return EMULATE_DO_MMIO
;
1282 kvm_arch_ops
->decache_regs(vcpu
);
1283 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1285 if (vcpu
->mmio_is_write
) {
1286 vcpu
->mmio_needed
= 0;
1287 return EMULATE_DO_MMIO
;
1290 return EMULATE_DONE
;
1292 EXPORT_SYMBOL_GPL(emulate_instruction
);
1294 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
1296 if (vcpu
->irq_summary
)
1299 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
1300 ++vcpu
->stat
.halt_exits
;
1303 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
1305 int kvm_hypercall(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1307 unsigned long nr
, a0
, a1
, a2
, a3
, a4
, a5
, ret
;
1309 kvm_arch_ops
->cache_regs(vcpu
);
1311 #ifdef CONFIG_X86_64
1312 if (is_long_mode(vcpu
)) {
1313 nr
= vcpu
->regs
[VCPU_REGS_RAX
];
1314 a0
= vcpu
->regs
[VCPU_REGS_RDI
];
1315 a1
= vcpu
->regs
[VCPU_REGS_RSI
];
1316 a2
= vcpu
->regs
[VCPU_REGS_RDX
];
1317 a3
= vcpu
->regs
[VCPU_REGS_RCX
];
1318 a4
= vcpu
->regs
[VCPU_REGS_R8
];
1319 a5
= vcpu
->regs
[VCPU_REGS_R9
];
1323 nr
= vcpu
->regs
[VCPU_REGS_RBX
] & -1u;
1324 a0
= vcpu
->regs
[VCPU_REGS_RAX
] & -1u;
1325 a1
= vcpu
->regs
[VCPU_REGS_RCX
] & -1u;
1326 a2
= vcpu
->regs
[VCPU_REGS_RDX
] & -1u;
1327 a3
= vcpu
->regs
[VCPU_REGS_RSI
] & -1u;
1328 a4
= vcpu
->regs
[VCPU_REGS_RDI
] & -1u;
1329 a5
= vcpu
->regs
[VCPU_REGS_RBP
] & -1u;
1333 run
->hypercall
.args
[0] = a0
;
1334 run
->hypercall
.args
[1] = a1
;
1335 run
->hypercall
.args
[2] = a2
;
1336 run
->hypercall
.args
[3] = a3
;
1337 run
->hypercall
.args
[4] = a4
;
1338 run
->hypercall
.args
[5] = a5
;
1339 run
->hypercall
.ret
= ret
;
1340 run
->hypercall
.longmode
= is_long_mode(vcpu
);
1341 kvm_arch_ops
->decache_regs(vcpu
);
1344 vcpu
->regs
[VCPU_REGS_RAX
] = ret
;
1345 kvm_arch_ops
->decache_regs(vcpu
);
1348 EXPORT_SYMBOL_GPL(kvm_hypercall
);
1350 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1352 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1355 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1357 struct descriptor_table dt
= { limit
, base
};
1359 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1362 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1364 struct descriptor_table dt
= { limit
, base
};
1366 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1369 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1370 unsigned long *rflags
)
1373 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1376 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1378 kvm_arch_ops
->decache_cr4_guest_bits(vcpu
);
1389 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1394 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1395 unsigned long *rflags
)
1399 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1400 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1409 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1412 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1417 * Register the para guest with the host:
1419 static int vcpu_register_para(struct kvm_vcpu
*vcpu
, gpa_t para_state_gpa
)
1421 struct kvm_vcpu_para_state
*para_state
;
1422 hpa_t para_state_hpa
, hypercall_hpa
;
1423 struct page
*para_state_page
;
1424 unsigned char *hypercall
;
1425 gpa_t hypercall_gpa
;
1427 printk(KERN_DEBUG
"kvm: guest trying to enter paravirtual mode\n");
1428 printk(KERN_DEBUG
".... para_state_gpa: %08Lx\n", para_state_gpa
);
1431 * Needs to be page aligned:
1433 if (para_state_gpa
!= PAGE_ALIGN(para_state_gpa
))
1436 para_state_hpa
= gpa_to_hpa(vcpu
, para_state_gpa
);
1437 printk(KERN_DEBUG
".... para_state_hpa: %08Lx\n", para_state_hpa
);
1438 if (is_error_hpa(para_state_hpa
))
1441 mark_page_dirty(vcpu
->kvm
, para_state_gpa
>> PAGE_SHIFT
);
1442 para_state_page
= pfn_to_page(para_state_hpa
>> PAGE_SHIFT
);
1443 para_state
= kmap_atomic(para_state_page
, KM_USER0
);
1445 printk(KERN_DEBUG
".... guest version: %d\n", para_state
->guest_version
);
1446 printk(KERN_DEBUG
".... size: %d\n", para_state
->size
);
1448 para_state
->host_version
= KVM_PARA_API_VERSION
;
1450 * We cannot support guests that try to register themselves
1451 * with a newer API version than the host supports:
1453 if (para_state
->guest_version
> KVM_PARA_API_VERSION
) {
1454 para_state
->ret
= -KVM_EINVAL
;
1455 goto err_kunmap_skip
;
1458 hypercall_gpa
= para_state
->hypercall_gpa
;
1459 hypercall_hpa
= gpa_to_hpa(vcpu
, hypercall_gpa
);
1460 printk(KERN_DEBUG
".... hypercall_hpa: %08Lx\n", hypercall_hpa
);
1461 if (is_error_hpa(hypercall_hpa
)) {
1462 para_state
->ret
= -KVM_EINVAL
;
1463 goto err_kunmap_skip
;
1466 printk(KERN_DEBUG
"kvm: para guest successfully registered.\n");
1467 vcpu
->para_state_page
= para_state_page
;
1468 vcpu
->para_state_gpa
= para_state_gpa
;
1469 vcpu
->hypercall_gpa
= hypercall_gpa
;
1471 mark_page_dirty(vcpu
->kvm
, hypercall_gpa
>> PAGE_SHIFT
);
1472 hypercall
= kmap_atomic(pfn_to_page(hypercall_hpa
>> PAGE_SHIFT
),
1473 KM_USER1
) + (hypercall_hpa
& ~PAGE_MASK
);
1474 kvm_arch_ops
->patch_hypercall(vcpu
, hypercall
);
1475 kunmap_atomic(hypercall
, KM_USER1
);
1477 para_state
->ret
= 0;
1479 kunmap_atomic(para_state
, KM_USER0
);
1485 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1490 case 0xc0010010: /* SYSCFG */
1491 case 0xc0010015: /* HWCR */
1492 case MSR_IA32_PLATFORM_ID
:
1493 case MSR_IA32_P5_MC_ADDR
:
1494 case MSR_IA32_P5_MC_TYPE
:
1495 case MSR_IA32_MC0_CTL
:
1496 case MSR_IA32_MCG_STATUS
:
1497 case MSR_IA32_MCG_CAP
:
1498 case MSR_IA32_MC0_MISC
:
1499 case MSR_IA32_MC0_MISC
+4:
1500 case MSR_IA32_MC0_MISC
+8:
1501 case MSR_IA32_MC0_MISC
+12:
1502 case MSR_IA32_MC0_MISC
+16:
1503 case MSR_IA32_UCODE_REV
:
1504 case MSR_IA32_PERF_STATUS
:
1505 case MSR_IA32_EBL_CR_POWERON
:
1506 /* MTRR registers */
1508 case 0x200 ... 0x2ff:
1511 case 0xcd: /* fsb frequency */
1514 case MSR_IA32_APICBASE
:
1515 data
= vcpu
->apic_base
;
1517 case MSR_IA32_MISC_ENABLE
:
1518 data
= vcpu
->ia32_misc_enable_msr
;
1520 #ifdef CONFIG_X86_64
1522 data
= vcpu
->shadow_efer
;
1526 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1532 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1535 * Reads an msr value (of 'msr_index') into 'pdata'.
1536 * Returns 0 on success, non-0 otherwise.
1537 * Assumes vcpu_load() was already called.
1539 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1541 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1544 #ifdef CONFIG_X86_64
1546 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1548 if (efer
& EFER_RESERVED_BITS
) {
1549 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1556 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1557 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1562 kvm_arch_ops
->set_efer(vcpu
, efer
);
1565 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1567 vcpu
->shadow_efer
= efer
;
1572 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1575 #ifdef CONFIG_X86_64
1577 set_efer(vcpu
, data
);
1580 case MSR_IA32_MC0_STATUS
:
1581 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1582 __FUNCTION__
, data
);
1584 case MSR_IA32_MCG_STATUS
:
1585 printk(KERN_WARNING
"%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1586 __FUNCTION__
, data
);
1588 case MSR_IA32_UCODE_REV
:
1589 case MSR_IA32_UCODE_WRITE
:
1590 case 0x200 ... 0x2ff: /* MTRRs */
1592 case MSR_IA32_APICBASE
:
1593 vcpu
->apic_base
= data
;
1595 case MSR_IA32_MISC_ENABLE
:
1596 vcpu
->ia32_misc_enable_msr
= data
;
1599 * This is the 'probe whether the host is KVM' logic:
1601 case MSR_KVM_API_MAGIC
:
1602 return vcpu_register_para(vcpu
, data
);
1605 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1610 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1613 * Writes msr value into into the appropriate "register".
1614 * Returns 0 on success, non-0 otherwise.
1615 * Assumes vcpu_load() was already called.
1617 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1619 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1622 void kvm_resched(struct kvm_vcpu
*vcpu
)
1624 if (!need_resched())
1630 EXPORT_SYMBOL_GPL(kvm_resched
);
1632 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1636 for (i
= 0; i
< n
; ++i
)
1637 wrmsrl(e
[i
].index
, e
[i
].data
);
1639 EXPORT_SYMBOL_GPL(load_msrs
);
1641 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1645 for (i
= 0; i
< n
; ++i
)
1646 rdmsrl(e
[i
].index
, e
[i
].data
);
1648 EXPORT_SYMBOL_GPL(save_msrs
);
1650 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
1654 struct kvm_cpuid_entry
*e
, *best
;
1656 kvm_arch_ops
->cache_regs(vcpu
);
1657 function
= vcpu
->regs
[VCPU_REGS_RAX
];
1658 vcpu
->regs
[VCPU_REGS_RAX
] = 0;
1659 vcpu
->regs
[VCPU_REGS_RBX
] = 0;
1660 vcpu
->regs
[VCPU_REGS_RCX
] = 0;
1661 vcpu
->regs
[VCPU_REGS_RDX
] = 0;
1663 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
1664 e
= &vcpu
->cpuid_entries
[i
];
1665 if (e
->function
== function
) {
1670 * Both basic or both extended?
1672 if (((e
->function
^ function
) & 0x80000000) == 0)
1673 if (!best
|| e
->function
> best
->function
)
1677 vcpu
->regs
[VCPU_REGS_RAX
] = best
->eax
;
1678 vcpu
->regs
[VCPU_REGS_RBX
] = best
->ebx
;
1679 vcpu
->regs
[VCPU_REGS_RCX
] = best
->ecx
;
1680 vcpu
->regs
[VCPU_REGS_RDX
] = best
->edx
;
1682 kvm_arch_ops
->decache_regs(vcpu
);
1683 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1685 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
1687 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
1689 void *p
= vcpu
->pio_data
;
1692 int nr_pages
= vcpu
->pio
.guest_pages
[1] ? 2 : 1;
1694 kvm_arch_ops
->vcpu_put(vcpu
);
1695 q
= vmap(vcpu
->pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
1698 kvm_arch_ops
->vcpu_load(vcpu
);
1699 free_pio_guest_pages(vcpu
);
1702 q
+= vcpu
->pio
.guest_page_offset
;
1703 bytes
= vcpu
->pio
.size
* vcpu
->pio
.cur_count
;
1705 memcpy(q
, p
, bytes
);
1707 memcpy(p
, q
, bytes
);
1708 q
-= vcpu
->pio
.guest_page_offset
;
1710 kvm_arch_ops
->vcpu_load(vcpu
);
1711 free_pio_guest_pages(vcpu
);
1715 static int complete_pio(struct kvm_vcpu
*vcpu
)
1717 struct kvm_pio_request
*io
= &vcpu
->pio
;
1721 kvm_arch_ops
->cache_regs(vcpu
);
1725 memcpy(&vcpu
->regs
[VCPU_REGS_RAX
], vcpu
->pio_data
,
1729 r
= pio_copy_data(vcpu
);
1731 kvm_arch_ops
->cache_regs(vcpu
);
1738 delta
*= io
->cur_count
;
1740 * The size of the register should really depend on
1741 * current address size.
1743 vcpu
->regs
[VCPU_REGS_RCX
] -= delta
;
1749 vcpu
->regs
[VCPU_REGS_RDI
] += delta
;
1751 vcpu
->regs
[VCPU_REGS_RSI
] += delta
;
1754 kvm_arch_ops
->decache_regs(vcpu
);
1756 io
->count
-= io
->cur_count
;
1760 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1764 static void kernel_pio(struct kvm_io_device
*pio_dev
,
1765 struct kvm_vcpu
*vcpu
,
1768 /* TODO: String I/O for in kernel device */
1771 kvm_iodevice_read(pio_dev
, vcpu
->pio
.port
,
1775 kvm_iodevice_write(pio_dev
, vcpu
->pio
.port
,
1780 static void pio_string_write(struct kvm_io_device
*pio_dev
,
1781 struct kvm_vcpu
*vcpu
)
1783 struct kvm_pio_request
*io
= &vcpu
->pio
;
1784 void *pd
= vcpu
->pio_data
;
1787 for (i
= 0; i
< io
->cur_count
; i
++) {
1788 kvm_iodevice_write(pio_dev
, io
->port
,
1795 int kvm_setup_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
1796 int size
, unsigned long count
, int string
, int down
,
1797 gva_t address
, int rep
, unsigned port
)
1799 unsigned now
, in_page
;
1803 struct kvm_io_device
*pio_dev
;
1805 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
1806 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
1807 vcpu
->run
->io
.size
= size
;
1808 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
1809 vcpu
->run
->io
.count
= count
;
1810 vcpu
->run
->io
.port
= port
;
1811 vcpu
->pio
.count
= count
;
1812 vcpu
->pio
.cur_count
= count
;
1813 vcpu
->pio
.size
= size
;
1815 vcpu
->pio
.port
= port
;
1816 vcpu
->pio
.string
= string
;
1817 vcpu
->pio
.down
= down
;
1818 vcpu
->pio
.guest_page_offset
= offset_in_page(address
);
1819 vcpu
->pio
.rep
= rep
;
1821 pio_dev
= vcpu_find_pio_dev(vcpu
, port
);
1823 kvm_arch_ops
->cache_regs(vcpu
);
1824 memcpy(vcpu
->pio_data
, &vcpu
->regs
[VCPU_REGS_RAX
], 4);
1825 kvm_arch_ops
->decache_regs(vcpu
);
1827 kernel_pio(pio_dev
, vcpu
, vcpu
->pio_data
);
1835 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1839 now
= min(count
, PAGE_SIZE
/ size
);
1842 in_page
= PAGE_SIZE
- offset_in_page(address
);
1844 in_page
= offset_in_page(address
) + size
;
1845 now
= min(count
, (unsigned long)in_page
/ size
);
1848 * String I/O straddles page boundary. Pin two guest pages
1849 * so that we satisfy atomicity constraints. Do just one
1850 * transaction to avoid complexity.
1857 * String I/O in reverse. Yuck. Kill the guest, fix later.
1859 printk(KERN_ERR
"kvm: guest string pio down\n");
1863 vcpu
->run
->io
.count
= now
;
1864 vcpu
->pio
.cur_count
= now
;
1866 for (i
= 0; i
< nr_pages
; ++i
) {
1867 spin_lock(&vcpu
->kvm
->lock
);
1868 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
1871 vcpu
->pio
.guest_pages
[i
] = page
;
1872 spin_unlock(&vcpu
->kvm
->lock
);
1875 free_pio_guest_pages(vcpu
);
1880 if (!vcpu
->pio
.in
) {
1881 /* string PIO write */
1882 ret
= pio_copy_data(vcpu
);
1883 if (ret
>= 0 && pio_dev
) {
1884 pio_string_write(pio_dev
, vcpu
);
1886 if (vcpu
->pio
.count
== 0)
1890 printk(KERN_ERR
"no string pio read support yet, "
1891 "port %x size %d count %ld\n",
1896 EXPORT_SYMBOL_GPL(kvm_setup_pio
);
1898 static int kvm_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1905 if (vcpu
->sigset_active
)
1906 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1908 /* re-sync apic's tpr */
1909 vcpu
->cr8
= kvm_run
->cr8
;
1911 if (vcpu
->pio
.cur_count
) {
1912 r
= complete_pio(vcpu
);
1917 if (vcpu
->mmio_needed
) {
1918 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1919 vcpu
->mmio_read_completed
= 1;
1920 vcpu
->mmio_needed
= 0;
1921 r
= emulate_instruction(vcpu
, kvm_run
,
1922 vcpu
->mmio_fault_cr2
, 0);
1923 if (r
== EMULATE_DO_MMIO
) {
1925 * Read-modify-write. Back to userspace.
1927 kvm_run
->exit_reason
= KVM_EXIT_MMIO
;
1933 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
1934 kvm_arch_ops
->cache_regs(vcpu
);
1935 vcpu
->regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
1936 kvm_arch_ops
->decache_regs(vcpu
);
1939 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1942 if (vcpu
->sigset_active
)
1943 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1949 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
,
1950 struct kvm_regs
*regs
)
1954 kvm_arch_ops
->cache_regs(vcpu
);
1956 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1957 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1958 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1959 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1960 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1961 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1962 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1963 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1964 #ifdef CONFIG_X86_64
1965 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1966 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1967 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1968 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1969 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1970 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1971 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1972 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1975 regs
->rip
= vcpu
->rip
;
1976 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1979 * Don't leak debug flags in case they were set for guest debugging
1981 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1982 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1989 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
,
1990 struct kvm_regs
*regs
)
1994 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1995 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1996 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1997 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1998 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1999 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
2000 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
2001 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
2002 #ifdef CONFIG_X86_64
2003 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
2004 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
2005 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
2006 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
2007 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
2008 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
2009 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
2010 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
2013 vcpu
->rip
= regs
->rip
;
2014 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
2016 kvm_arch_ops
->decache_regs(vcpu
);
2023 static void get_segment(struct kvm_vcpu
*vcpu
,
2024 struct kvm_segment
*var
, int seg
)
2026 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
2029 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
2030 struct kvm_sregs
*sregs
)
2032 struct descriptor_table dt
;
2036 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2037 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2038 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2039 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2040 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2041 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2043 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2044 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2046 kvm_arch_ops
->get_idt(vcpu
, &dt
);
2047 sregs
->idt
.limit
= dt
.limit
;
2048 sregs
->idt
.base
= dt
.base
;
2049 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
2050 sregs
->gdt
.limit
= dt
.limit
;
2051 sregs
->gdt
.base
= dt
.base
;
2053 kvm_arch_ops
->decache_cr4_guest_bits(vcpu
);
2054 sregs
->cr0
= vcpu
->cr0
;
2055 sregs
->cr2
= vcpu
->cr2
;
2056 sregs
->cr3
= vcpu
->cr3
;
2057 sregs
->cr4
= vcpu
->cr4
;
2058 sregs
->cr8
= vcpu
->cr8
;
2059 sregs
->efer
= vcpu
->shadow_efer
;
2060 sregs
->apic_base
= vcpu
->apic_base
;
2062 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
2063 sizeof sregs
->interrupt_bitmap
);
2070 static void set_segment(struct kvm_vcpu
*vcpu
,
2071 struct kvm_segment
*var
, int seg
)
2073 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
2076 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
2077 struct kvm_sregs
*sregs
)
2079 int mmu_reset_needed
= 0;
2081 struct descriptor_table dt
;
2085 dt
.limit
= sregs
->idt
.limit
;
2086 dt
.base
= sregs
->idt
.base
;
2087 kvm_arch_ops
->set_idt(vcpu
, &dt
);
2088 dt
.limit
= sregs
->gdt
.limit
;
2089 dt
.base
= sregs
->gdt
.base
;
2090 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
2092 vcpu
->cr2
= sregs
->cr2
;
2093 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
2094 vcpu
->cr3
= sregs
->cr3
;
2096 vcpu
->cr8
= sregs
->cr8
;
2098 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
2099 #ifdef CONFIG_X86_64
2100 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
2102 vcpu
->apic_base
= sregs
->apic_base
;
2104 kvm_arch_ops
->decache_cr4_guest_bits(vcpu
);
2106 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
2107 kvm_arch_ops
->set_cr0(vcpu
, sregs
->cr0
);
2109 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
2110 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
2111 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
2112 load_pdptrs(vcpu
, vcpu
->cr3
);
2114 if (mmu_reset_needed
)
2115 kvm_mmu_reset_context(vcpu
);
2117 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
2118 sizeof vcpu
->irq_pending
);
2119 vcpu
->irq_summary
= 0;
2120 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
2121 if (vcpu
->irq_pending
[i
])
2122 __set_bit(i
, &vcpu
->irq_summary
);
2124 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2125 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2126 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2127 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2128 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2129 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2131 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2132 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2140 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2141 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2143 * This list is modified at module load time to reflect the
2144 * capabilities of the host cpu.
2146 static u32 msrs_to_save
[] = {
2147 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
2149 #ifdef CONFIG_X86_64
2150 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
2152 MSR_IA32_TIME_STAMP_COUNTER
,
2155 static unsigned num_msrs_to_save
;
2157 static u32 emulated_msrs
[] = {
2158 MSR_IA32_MISC_ENABLE
,
2161 static __init
void kvm_init_msr_list(void)
2166 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2167 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2170 msrs_to_save
[j
] = msrs_to_save
[i
];
2173 num_msrs_to_save
= j
;
2177 * Adapt set_msr() to msr_io()'s calling convention
2179 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
2181 return kvm_set_msr(vcpu
, index
, *data
);
2185 * Read or write a bunch of msrs. All parameters are kernel addresses.
2187 * @return number of msrs set successfully.
2189 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2190 struct kvm_msr_entry
*entries
,
2191 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2192 unsigned index
, u64
*data
))
2198 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2199 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2208 * Read or write a bunch of msrs. Parameters are user addresses.
2210 * @return number of msrs set successfully.
2212 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2213 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2214 unsigned index
, u64
*data
),
2217 struct kvm_msrs msrs
;
2218 struct kvm_msr_entry
*entries
;
2223 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2227 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2231 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2232 entries
= vmalloc(size
);
2237 if (copy_from_user(entries
, user_msrs
->entries
, size
))
2240 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2245 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2257 * Translate a guest virtual address to a guest physical address.
2259 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
2260 struct kvm_translation
*tr
)
2262 unsigned long vaddr
= tr
->linear_address
;
2266 spin_lock(&vcpu
->kvm
->lock
);
2267 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
2268 tr
->physical_address
= gpa
;
2269 tr
->valid
= gpa
!= UNMAPPED_GVA
;
2272 spin_unlock(&vcpu
->kvm
->lock
);
2278 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
2279 struct kvm_interrupt
*irq
)
2281 if (irq
->irq
< 0 || irq
->irq
>= 256)
2285 set_bit(irq
->irq
, vcpu
->irq_pending
);
2286 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
2293 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
2294 struct kvm_debug_guest
*dbg
)
2300 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
2307 static struct page
*kvm_vcpu_nopage(struct vm_area_struct
*vma
,
2308 unsigned long address
,
2311 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2312 unsigned long pgoff
;
2315 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2317 page
= virt_to_page(vcpu
->run
);
2318 else if (pgoff
== KVM_PIO_PAGE_OFFSET
)
2319 page
= virt_to_page(vcpu
->pio_data
);
2321 return NOPAGE_SIGBUS
;
2324 *type
= VM_FAULT_MINOR
;
2329 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
2330 .nopage
= kvm_vcpu_nopage
,
2333 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2335 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2339 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2341 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2343 fput(vcpu
->kvm
->filp
);
2347 static struct file_operations kvm_vcpu_fops
= {
2348 .release
= kvm_vcpu_release
,
2349 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2350 .compat_ioctl
= kvm_vcpu_ioctl
,
2351 .mmap
= kvm_vcpu_mmap
,
2355 * Allocates an inode for the vcpu.
2357 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2360 struct inode
*inode
;
2363 r
= anon_inode_getfd(&fd
, &inode
, &file
,
2364 "kvm-vcpu", &kvm_vcpu_fops
, vcpu
);
2367 atomic_inc(&vcpu
->kvm
->filp
->f_count
);
2372 * Creates some virtual cpus. Good luck creating more than one.
2374 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
2377 struct kvm_vcpu
*vcpu
;
2384 vcpu
= &kvm
->vcpus
[n
];
2387 mutex_lock(&vcpu
->mutex
);
2390 mutex_unlock(&vcpu
->mutex
);
2394 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2398 vcpu
->run
= page_address(page
);
2400 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2404 vcpu
->pio_data
= page_address(page
);
2406 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
2408 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
2411 r
= kvm_arch_ops
->vcpu_create(vcpu
);
2413 goto out_free_vcpus
;
2415 r
= kvm_mmu_create(vcpu
);
2417 goto out_free_vcpus
;
2419 kvm_arch_ops
->vcpu_load(vcpu
);
2420 r
= kvm_mmu_setup(vcpu
);
2422 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
2426 goto out_free_vcpus
;
2428 r
= create_vcpu_fd(vcpu
);
2430 goto out_free_vcpus
;
2432 spin_lock(&kvm_lock
);
2433 if (n
>= kvm
->nvcpus
)
2434 kvm
->nvcpus
= n
+ 1;
2435 spin_unlock(&kvm_lock
);
2440 kvm_free_vcpu(vcpu
);
2442 free_page((unsigned long)vcpu
->run
);
2445 mutex_unlock(&vcpu
->mutex
);
2450 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
2454 struct kvm_cpuid_entry
*e
, *entry
;
2456 rdmsrl(MSR_EFER
, efer
);
2458 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
2459 e
= &vcpu
->cpuid_entries
[i
];
2460 if (e
->function
== 0x80000001) {
2465 if (entry
&& (entry
->edx
& (1 << 20)) && !(efer
& EFER_NX
)) {
2466 entry
->edx
&= ~(1 << 20);
2467 printk(KERN_INFO
"kvm: guest NX capability removed\n");
2471 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
2472 struct kvm_cpuid
*cpuid
,
2473 struct kvm_cpuid_entry __user
*entries
)
2478 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
2481 if (copy_from_user(&vcpu
->cpuid_entries
, entries
,
2482 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
2484 vcpu
->cpuid_nent
= cpuid
->nent
;
2485 cpuid_fix_nx_cap(vcpu
);
2492 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2495 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2496 vcpu
->sigset_active
= 1;
2497 vcpu
->sigset
= *sigset
;
2499 vcpu
->sigset_active
= 0;
2504 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2505 * we have asm/x86/processor.h
2516 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2517 #ifdef CONFIG_X86_64
2518 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2520 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2524 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2526 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2530 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
2531 fpu
->fcw
= fxsave
->cwd
;
2532 fpu
->fsw
= fxsave
->swd
;
2533 fpu
->ftwx
= fxsave
->twd
;
2534 fpu
->last_opcode
= fxsave
->fop
;
2535 fpu
->last_ip
= fxsave
->rip
;
2536 fpu
->last_dp
= fxsave
->rdp
;
2537 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
2544 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2546 struct fxsave
*fxsave
= (struct fxsave
*)vcpu
->guest_fx_image
;
2550 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
2551 fxsave
->cwd
= fpu
->fcw
;
2552 fxsave
->swd
= fpu
->fsw
;
2553 fxsave
->twd
= fpu
->ftwx
;
2554 fxsave
->fop
= fpu
->last_opcode
;
2555 fxsave
->rip
= fpu
->last_ip
;
2556 fxsave
->rdp
= fpu
->last_dp
;
2557 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
2564 static long kvm_vcpu_ioctl(struct file
*filp
,
2565 unsigned int ioctl
, unsigned long arg
)
2567 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2568 void __user
*argp
= (void __user
*)arg
;
2576 r
= kvm_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2578 case KVM_GET_REGS
: {
2579 struct kvm_regs kvm_regs
;
2581 memset(&kvm_regs
, 0, sizeof kvm_regs
);
2582 r
= kvm_vcpu_ioctl_get_regs(vcpu
, &kvm_regs
);
2586 if (copy_to_user(argp
, &kvm_regs
, sizeof kvm_regs
))
2591 case KVM_SET_REGS
: {
2592 struct kvm_regs kvm_regs
;
2595 if (copy_from_user(&kvm_regs
, argp
, sizeof kvm_regs
))
2597 r
= kvm_vcpu_ioctl_set_regs(vcpu
, &kvm_regs
);
2603 case KVM_GET_SREGS
: {
2604 struct kvm_sregs kvm_sregs
;
2606 memset(&kvm_sregs
, 0, sizeof kvm_sregs
);
2607 r
= kvm_vcpu_ioctl_get_sregs(vcpu
, &kvm_sregs
);
2611 if (copy_to_user(argp
, &kvm_sregs
, sizeof kvm_sregs
))
2616 case KVM_SET_SREGS
: {
2617 struct kvm_sregs kvm_sregs
;
2620 if (copy_from_user(&kvm_sregs
, argp
, sizeof kvm_sregs
))
2622 r
= kvm_vcpu_ioctl_set_sregs(vcpu
, &kvm_sregs
);
2628 case KVM_TRANSLATE
: {
2629 struct kvm_translation tr
;
2632 if (copy_from_user(&tr
, argp
, sizeof tr
))
2634 r
= kvm_vcpu_ioctl_translate(vcpu
, &tr
);
2638 if (copy_to_user(argp
, &tr
, sizeof tr
))
2643 case KVM_INTERRUPT
: {
2644 struct kvm_interrupt irq
;
2647 if (copy_from_user(&irq
, argp
, sizeof irq
))
2649 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
2655 case KVM_DEBUG_GUEST
: {
2656 struct kvm_debug_guest dbg
;
2659 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2661 r
= kvm_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
2668 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
2671 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
2673 case KVM_SET_CPUID
: {
2674 struct kvm_cpuid __user
*cpuid_arg
= argp
;
2675 struct kvm_cpuid cpuid
;
2678 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2680 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
2685 case KVM_SET_SIGNAL_MASK
: {
2686 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2687 struct kvm_signal_mask kvm_sigmask
;
2688 sigset_t sigset
, *p
;
2693 if (copy_from_user(&kvm_sigmask
, argp
,
2694 sizeof kvm_sigmask
))
2697 if (kvm_sigmask
.len
!= sizeof sigset
)
2700 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2705 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2711 memset(&fpu
, 0, sizeof fpu
);
2712 r
= kvm_vcpu_ioctl_get_fpu(vcpu
, &fpu
);
2716 if (copy_to_user(argp
, &fpu
, sizeof fpu
))
2725 if (copy_from_user(&fpu
, argp
, sizeof fpu
))
2727 r
= kvm_vcpu_ioctl_set_fpu(vcpu
, &fpu
);
2740 static long kvm_vm_ioctl(struct file
*filp
,
2741 unsigned int ioctl
, unsigned long arg
)
2743 struct kvm
*kvm
= filp
->private_data
;
2744 void __user
*argp
= (void __user
*)arg
;
2748 case KVM_CREATE_VCPU
:
2749 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2753 case KVM_SET_MEMORY_REGION
: {
2754 struct kvm_memory_region kvm_mem
;
2757 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2759 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_mem
);
2764 case KVM_GET_DIRTY_LOG
: {
2765 struct kvm_dirty_log log
;
2768 if (copy_from_user(&log
, argp
, sizeof log
))
2770 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2775 case KVM_SET_MEMORY_ALIAS
: {
2776 struct kvm_memory_alias alias
;
2779 if (copy_from_user(&alias
, argp
, sizeof alias
))
2781 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &alias
);
2793 static struct page
*kvm_vm_nopage(struct vm_area_struct
*vma
,
2794 unsigned long address
,
2797 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2798 unsigned long pgoff
;
2801 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2802 page
= gfn_to_page(kvm
, pgoff
);
2804 return NOPAGE_SIGBUS
;
2807 *type
= VM_FAULT_MINOR
;
2812 static struct vm_operations_struct kvm_vm_vm_ops
= {
2813 .nopage
= kvm_vm_nopage
,
2816 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2818 vma
->vm_ops
= &kvm_vm_vm_ops
;
2822 static struct file_operations kvm_vm_fops
= {
2823 .release
= kvm_vm_release
,
2824 .unlocked_ioctl
= kvm_vm_ioctl
,
2825 .compat_ioctl
= kvm_vm_ioctl
,
2826 .mmap
= kvm_vm_mmap
,
2829 static int kvm_dev_ioctl_create_vm(void)
2832 struct inode
*inode
;
2836 kvm
= kvm_create_vm();
2838 return PTR_ERR(kvm
);
2839 r
= anon_inode_getfd(&fd
, &inode
, &file
, "kvm-vm", &kvm_vm_fops
, kvm
);
2841 kvm_destroy_vm(kvm
);
2850 static long kvm_dev_ioctl(struct file
*filp
,
2851 unsigned int ioctl
, unsigned long arg
)
2853 void __user
*argp
= (void __user
*)arg
;
2857 case KVM_GET_API_VERSION
:
2861 r
= KVM_API_VERSION
;
2867 r
= kvm_dev_ioctl_create_vm();
2869 case KVM_GET_MSR_INDEX_LIST
: {
2870 struct kvm_msr_list __user
*user_msr_list
= argp
;
2871 struct kvm_msr_list msr_list
;
2875 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2878 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
2879 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2882 if (n
< num_msrs_to_save
)
2885 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2886 num_msrs_to_save
* sizeof(u32
)))
2888 if (copy_to_user(user_msr_list
->indices
2889 + num_msrs_to_save
* sizeof(u32
),
2891 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
2896 case KVM_CHECK_EXTENSION
:
2898 * No extensions defined at present.
2902 case KVM_GET_VCPU_MMAP_SIZE
:
2915 static struct file_operations kvm_chardev_ops
= {
2916 .open
= kvm_dev_open
,
2917 .release
= kvm_dev_release
,
2918 .unlocked_ioctl
= kvm_dev_ioctl
,
2919 .compat_ioctl
= kvm_dev_ioctl
,
2922 static struct miscdevice kvm_dev
= {
2929 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2932 static void decache_vcpus_on_cpu(int cpu
)
2935 struct kvm_vcpu
*vcpu
;
2938 spin_lock(&kvm_lock
);
2939 list_for_each_entry(vm
, &vm_list
, vm_list
)
2940 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2941 vcpu
= &vm
->vcpus
[i
];
2943 * If the vcpu is locked, then it is running on some
2944 * other cpu and therefore it is not cached on the
2947 * If it's not locked, check the last cpu it executed
2950 if (mutex_trylock(&vcpu
->mutex
)) {
2951 if (vcpu
->cpu
== cpu
) {
2952 kvm_arch_ops
->vcpu_decache(vcpu
);
2955 mutex_unlock(&vcpu
->mutex
);
2958 spin_unlock(&kvm_lock
);
2961 static void hardware_enable(void *junk
)
2963 int cpu
= raw_smp_processor_id();
2965 if (cpu_isset(cpu
, cpus_hardware_enabled
))
2967 cpu_set(cpu
, cpus_hardware_enabled
);
2968 kvm_arch_ops
->hardware_enable(NULL
);
2971 static void hardware_disable(void *junk
)
2973 int cpu
= raw_smp_processor_id();
2975 if (!cpu_isset(cpu
, cpus_hardware_enabled
))
2977 cpu_clear(cpu
, cpus_hardware_enabled
);
2978 decache_vcpus_on_cpu(cpu
);
2979 kvm_arch_ops
->hardware_disable(NULL
);
2982 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2989 case CPU_DYING_FROZEN
:
2990 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2992 hardware_disable(NULL
);
2994 case CPU_UP_CANCELED
:
2995 case CPU_UP_CANCELED_FROZEN
:
2996 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2998 smp_call_function_single(cpu
, hardware_disable
, NULL
, 0, 1);
3001 case CPU_ONLINE_FROZEN
:
3002 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
3004 smp_call_function_single(cpu
, hardware_enable
, NULL
, 0, 1);
3010 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3013 if (val
== SYS_RESTART
) {
3015 * Some (well, at least mine) BIOSes hang on reboot if
3018 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
3019 on_each_cpu(hardware_disable
, NULL
, 0, 1);
3024 static struct notifier_block kvm_reboot_notifier
= {
3025 .notifier_call
= kvm_reboot
,
3029 void kvm_io_bus_init(struct kvm_io_bus
*bus
)
3031 memset(bus
, 0, sizeof(*bus
));
3034 void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3038 for (i
= 0; i
< bus
->dev_count
; i
++) {
3039 struct kvm_io_device
*pos
= bus
->devs
[i
];
3041 kvm_iodevice_destructor(pos
);
3045 struct kvm_io_device
*kvm_io_bus_find_dev(struct kvm_io_bus
*bus
, gpa_t addr
)
3049 for (i
= 0; i
< bus
->dev_count
; i
++) {
3050 struct kvm_io_device
*pos
= bus
->devs
[i
];
3052 if (pos
->in_range(pos
, addr
))
3059 void kvm_io_bus_register_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
)
3061 BUG_ON(bus
->dev_count
> (NR_IOBUS_DEVS
-1));
3063 bus
->devs
[bus
->dev_count
++] = dev
;
3066 static struct notifier_block kvm_cpu_notifier
= {
3067 .notifier_call
= kvm_cpu_hotplug
,
3068 .priority
= 20, /* must be > scheduler priority */
3071 static u64
stat_get(void *_offset
)
3073 unsigned offset
= (long)_offset
;
3076 struct kvm_vcpu
*vcpu
;
3079 spin_lock(&kvm_lock
);
3080 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3081 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
3082 vcpu
= &kvm
->vcpus
[i
];
3083 total
+= *(u32
*)((void *)vcpu
+ offset
);
3085 spin_unlock(&kvm_lock
);
3089 static void stat_set(void *offset
, u64 val
)
3093 DEFINE_SIMPLE_ATTRIBUTE(stat_fops
, stat_get
, stat_set
, "%llu\n");
3095 static __init
void kvm_init_debug(void)
3097 struct kvm_stats_debugfs_item
*p
;
3099 debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3100 for (p
= debugfs_entries
; p
->name
; ++p
)
3101 p
->dentry
= debugfs_create_file(p
->name
, 0444, debugfs_dir
,
3102 (void *)(long)p
->offset
,
3106 static void kvm_exit_debug(void)
3108 struct kvm_stats_debugfs_item
*p
;
3110 for (p
= debugfs_entries
; p
->name
; ++p
)
3111 debugfs_remove(p
->dentry
);
3112 debugfs_remove(debugfs_dir
);
3115 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
3117 hardware_disable(NULL
);
3121 static int kvm_resume(struct sys_device
*dev
)
3123 hardware_enable(NULL
);
3127 static struct sysdev_class kvm_sysdev_class
= {
3128 set_kset_name("kvm"),
3129 .suspend
= kvm_suspend
,
3130 .resume
= kvm_resume
,
3133 static struct sys_device kvm_sysdev
= {
3135 .cls
= &kvm_sysdev_class
,
3138 hpa_t bad_page_address
;
3140 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
3145 printk(KERN_ERR
"kvm: already loaded the other module\n");
3149 if (!ops
->cpu_has_kvm_support()) {
3150 printk(KERN_ERR
"kvm: no hardware support\n");
3153 if (ops
->disabled_by_bios()) {
3154 printk(KERN_ERR
"kvm: disabled by bios\n");
3160 r
= kvm_arch_ops
->hardware_setup();
3164 on_each_cpu(hardware_enable
, NULL
, 0, 1);
3165 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3168 register_reboot_notifier(&kvm_reboot_notifier
);
3170 r
= sysdev_class_register(&kvm_sysdev_class
);
3174 r
= sysdev_register(&kvm_sysdev
);
3178 kvm_chardev_ops
.owner
= module
;
3180 r
= misc_register(&kvm_dev
);
3182 printk (KERN_ERR
"kvm: misc device register failed\n");
3189 sysdev_unregister(&kvm_sysdev
);
3191 sysdev_class_unregister(&kvm_sysdev_class
);
3193 unregister_reboot_notifier(&kvm_reboot_notifier
);
3194 unregister_cpu_notifier(&kvm_cpu_notifier
);
3196 on_each_cpu(hardware_disable
, NULL
, 0, 1);
3197 kvm_arch_ops
->hardware_unsetup();
3199 kvm_arch_ops
= NULL
;
3203 void kvm_exit_arch(void)
3205 misc_deregister(&kvm_dev
);
3206 sysdev_unregister(&kvm_sysdev
);
3207 sysdev_class_unregister(&kvm_sysdev_class
);
3208 unregister_reboot_notifier(&kvm_reboot_notifier
);
3209 unregister_cpu_notifier(&kvm_cpu_notifier
);
3210 on_each_cpu(hardware_disable
, NULL
, 0, 1);
3211 kvm_arch_ops
->hardware_unsetup();
3212 kvm_arch_ops
= NULL
;
3215 static __init
int kvm_init(void)
3217 static struct page
*bad_page
;
3220 r
= kvm_mmu_module_init();
3226 kvm_init_msr_list();
3228 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
3233 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
3234 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
3240 kvm_mmu_module_exit();
3245 static __exit
void kvm_exit(void)
3248 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
3249 kvm_mmu_module_exit();
3252 module_init(kvm_init
)
3253 module_exit(kvm_exit
)
3255 EXPORT_SYMBOL_GPL(kvm_init_arch
);
3256 EXPORT_SYMBOL_GPL(kvm_exit_arch
);