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 <asm/processor.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <asm/uaccess.h>
31 #include <linux/reboot.h>
33 #include <linux/debugfs.h>
34 #include <linux/highmem.h>
35 #include <linux/file.h>
38 #include "x86_emulate.h"
39 #include "segment_descriptor.h"
41 MODULE_AUTHOR("Qumranet");
42 MODULE_LICENSE("GPL");
44 struct kvm_arch_ops
*kvm_arch_ops
;
45 struct kvm_stat kvm_stat
;
46 EXPORT_SYMBOL_GPL(kvm_stat
);
48 static struct kvm_stats_debugfs_item
{
51 struct dentry
*dentry
;
52 } debugfs_entries
[] = {
53 { "pf_fixed", &kvm_stat
.pf_fixed
},
54 { "pf_guest", &kvm_stat
.pf_guest
},
55 { "tlb_flush", &kvm_stat
.tlb_flush
},
56 { "invlpg", &kvm_stat
.invlpg
},
57 { "exits", &kvm_stat
.exits
},
58 { "io_exits", &kvm_stat
.io_exits
},
59 { "mmio_exits", &kvm_stat
.mmio_exits
},
60 { "signal_exits", &kvm_stat
.signal_exits
},
61 { "irq_window", &kvm_stat
.irq_window_exits
},
62 { "halt_exits", &kvm_stat
.halt_exits
},
63 { "request_irq", &kvm_stat
.request_irq_exits
},
64 { "irq_exits", &kvm_stat
.irq_exits
},
68 static struct dentry
*debugfs_dir
;
70 #define MAX_IO_MSRS 256
72 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
73 #define LMSW_GUEST_MASK 0x0eULL
74 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
75 #define CR8_RESEVED_BITS (~0x0fULL)
76 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
79 // LDT or TSS descriptor in the GDT. 16 bytes.
80 struct segment_descriptor_64
{
81 struct segment_descriptor s
;
88 unsigned long segment_base(u16 selector
)
90 struct descriptor_table gdt
;
91 struct segment_descriptor
*d
;
92 unsigned long table_base
;
93 typedef unsigned long ul
;
99 asm ("sgdt %0" : "=m"(gdt
));
100 table_base
= gdt
.base
;
102 if (selector
& 4) { /* from ldt */
105 asm ("sldt %0" : "=g"(ldt_selector
));
106 table_base
= segment_base(ldt_selector
);
108 d
= (struct segment_descriptor
*)(table_base
+ (selector
& ~7));
109 v
= d
->base_low
| ((ul
)d
->base_mid
<< 16) | ((ul
)d
->base_high
<< 24);
112 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
113 v
|= ((ul
)((struct segment_descriptor_64
*)d
)->base_higher
) << 32;
117 EXPORT_SYMBOL_GPL(segment_base
);
119 static inline int valid_vcpu(int n
)
121 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
124 int kvm_read_guest(struct kvm_vcpu
*vcpu
,
129 unsigned char *host_buf
= dest
;
130 unsigned long req_size
= size
;
138 paddr
= gva_to_hpa(vcpu
, addr
);
140 if (is_error_hpa(paddr
))
143 guest_buf
= (hva_t
)kmap_atomic(
144 pfn_to_page(paddr
>> PAGE_SHIFT
),
146 offset
= addr
& ~PAGE_MASK
;
148 now
= min(size
, PAGE_SIZE
- offset
);
149 memcpy(host_buf
, (void*)guest_buf
, now
);
153 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
155 return req_size
- size
;
157 EXPORT_SYMBOL_GPL(kvm_read_guest
);
159 int kvm_write_guest(struct kvm_vcpu
*vcpu
,
164 unsigned char *host_buf
= data
;
165 unsigned long req_size
= size
;
173 paddr
= gva_to_hpa(vcpu
, addr
);
175 if (is_error_hpa(paddr
))
178 guest_buf
= (hva_t
)kmap_atomic(
179 pfn_to_page(paddr
>> PAGE_SHIFT
), KM_USER0
);
180 offset
= addr
& ~PAGE_MASK
;
182 now
= min(size
, PAGE_SIZE
- offset
);
183 memcpy((void*)guest_buf
, host_buf
, now
);
187 kunmap_atomic((void *)(guest_buf
& PAGE_MASK
), KM_USER0
);
189 return req_size
- size
;
191 EXPORT_SYMBOL_GPL(kvm_write_guest
);
193 static int vcpu_slot(struct kvm_vcpu
*vcpu
)
195 return vcpu
- vcpu
->kvm
->vcpus
;
199 * Switches to specified vcpu, until a matching vcpu_put()
201 static struct kvm_vcpu
*vcpu_load(struct kvm
*kvm
, int vcpu_slot
)
203 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[vcpu_slot
];
205 mutex_lock(&vcpu
->mutex
);
206 if (unlikely(!vcpu
->vmcs
)) {
207 mutex_unlock(&vcpu
->mutex
);
210 return kvm_arch_ops
->vcpu_load(vcpu
);
213 static void vcpu_put(struct kvm_vcpu
*vcpu
)
215 kvm_arch_ops
->vcpu_put(vcpu
);
216 mutex_unlock(&vcpu
->mutex
);
219 static int kvm_dev_open(struct inode
*inode
, struct file
*filp
)
221 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
227 spin_lock_init(&kvm
->lock
);
228 INIT_LIST_HEAD(&kvm
->active_mmu_pages
);
229 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
230 struct kvm_vcpu
*vcpu
= &kvm
->vcpus
[i
];
232 mutex_init(&vcpu
->mutex
);
234 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
235 INIT_LIST_HEAD(&vcpu
->free_pages
);
237 filp
->private_data
= kvm
;
242 * Free any memory in @free but not in @dont.
244 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
245 struct kvm_memory_slot
*dont
)
249 if (!dont
|| free
->phys_mem
!= dont
->phys_mem
)
250 if (free
->phys_mem
) {
251 for (i
= 0; i
< free
->npages
; ++i
)
252 if (free
->phys_mem
[i
])
253 __free_page(free
->phys_mem
[i
]);
254 vfree(free
->phys_mem
);
257 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
258 vfree(free
->dirty_bitmap
);
262 free
->dirty_bitmap
= 0;
265 static void kvm_free_physmem(struct kvm
*kvm
)
269 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
270 kvm_free_physmem_slot(&kvm
->memslots
[i
], 0);
273 static void kvm_free_vcpu(struct kvm_vcpu
*vcpu
)
275 kvm_mmu_destroy(vcpu
);
276 kvm_arch_ops
->vcpu_free(vcpu
);
279 static void kvm_free_vcpus(struct kvm
*kvm
)
283 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
284 kvm_free_vcpu(&kvm
->vcpus
[i
]);
287 static int kvm_dev_release(struct inode
*inode
, struct file
*filp
)
289 struct kvm
*kvm
= filp
->private_data
;
292 kvm_free_physmem(kvm
);
297 static void inject_gp(struct kvm_vcpu
*vcpu
)
299 kvm_arch_ops
->inject_gp(vcpu
, 0);
303 * Load the pae pdptrs. Return true is they are all valid.
305 static int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
307 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
308 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
313 struct kvm_memory_slot
*memslot
;
315 spin_lock(&vcpu
->kvm
->lock
);
316 memslot
= gfn_to_memslot(vcpu
->kvm
, pdpt_gfn
);
317 /* FIXME: !memslot - emulate? 0xff? */
318 pdpt
= kmap_atomic(gfn_to_page(memslot
, pdpt_gfn
), KM_USER0
);
321 for (i
= 0; i
< 4; ++i
) {
322 pdpte
= pdpt
[offset
+ i
];
323 if ((pdpte
& 1) && (pdpte
& 0xfffffff0000001e6ull
)) {
329 for (i
= 0; i
< 4; ++i
)
330 vcpu
->pdptrs
[i
] = pdpt
[offset
+ i
];
333 kunmap_atomic(pdpt
, KM_USER0
);
334 spin_unlock(&vcpu
->kvm
->lock
);
339 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
341 if (cr0
& CR0_RESEVED_BITS
) {
342 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
348 if ((cr0
& CR0_NW_MASK
) && !(cr0
& CR0_CD_MASK
)) {
349 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
354 if ((cr0
& CR0_PG_MASK
) && !(cr0
& CR0_PE_MASK
)) {
355 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
356 "and a clear PE flag\n");
361 if (!is_paging(vcpu
) && (cr0
& CR0_PG_MASK
)) {
363 if ((vcpu
->shadow_efer
& EFER_LME
)) {
367 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
368 "in long mode while PAE is disabled\n");
372 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
374 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
375 "in long mode while CS.L == 1\n");
382 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
383 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
391 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
394 spin_lock(&vcpu
->kvm
->lock
);
395 kvm_mmu_reset_context(vcpu
);
396 spin_unlock(&vcpu
->kvm
->lock
);
399 EXPORT_SYMBOL_GPL(set_cr0
);
401 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
403 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
404 set_cr0(vcpu
, (vcpu
->cr0
& ~0x0ful
) | (msw
& 0x0f));
406 EXPORT_SYMBOL_GPL(lmsw
);
408 void set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
410 if (cr4
& CR4_RESEVED_BITS
) {
411 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
416 if (is_long_mode(vcpu
)) {
417 if (!(cr4
& CR4_PAE_MASK
)) {
418 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
423 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& CR4_PAE_MASK
)
424 && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
425 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
429 if (cr4
& CR4_VMXE_MASK
) {
430 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
434 kvm_arch_ops
->set_cr4(vcpu
, cr4
);
435 spin_lock(&vcpu
->kvm
->lock
);
436 kvm_mmu_reset_context(vcpu
);
437 spin_unlock(&vcpu
->kvm
->lock
);
439 EXPORT_SYMBOL_GPL(set_cr4
);
441 void set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
443 if (is_long_mode(vcpu
)) {
444 if ( cr3
& CR3_L_MODE_RESEVED_BITS
) {
445 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
450 if (cr3
& CR3_RESEVED_BITS
) {
451 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
455 if (is_paging(vcpu
) && is_pae(vcpu
) &&
456 !load_pdptrs(vcpu
, cr3
)) {
457 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
465 spin_lock(&vcpu
->kvm
->lock
);
467 * Does the new cr3 value map to physical memory? (Note, we
468 * catch an invalid cr3 even in real-mode, because it would
469 * cause trouble later on when we turn on paging anyway.)
471 * A real CPU would silently accept an invalid cr3 and would
472 * attempt to use it - with largely undefined (and often hard
473 * to debug) behavior on the guest side.
475 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
478 vcpu
->mmu
.new_cr3(vcpu
);
479 spin_unlock(&vcpu
->kvm
->lock
);
481 EXPORT_SYMBOL_GPL(set_cr3
);
483 void set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
485 if ( cr8
& CR8_RESEVED_BITS
) {
486 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
492 EXPORT_SYMBOL_GPL(set_cr8
);
494 void fx_init(struct kvm_vcpu
*vcpu
)
496 struct __attribute__ ((__packed__
)) fx_image_s
{
502 u64 operand
;// fpu dp
508 fx_save(vcpu
->host_fx_image
);
510 fx_save(vcpu
->guest_fx_image
);
511 fx_restore(vcpu
->host_fx_image
);
513 fx_image
= (struct fx_image_s
*)vcpu
->guest_fx_image
;
514 fx_image
->mxcsr
= 0x1f80;
515 memset(vcpu
->guest_fx_image
+ sizeof(struct fx_image_s
),
516 0, FX_IMAGE_SIZE
- sizeof(struct fx_image_s
));
518 EXPORT_SYMBOL_GPL(fx_init
);
521 * Creates some virtual cpus. Good luck creating more than one.
523 static int kvm_dev_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
526 struct kvm_vcpu
*vcpu
;
532 vcpu
= &kvm
->vcpus
[n
];
534 mutex_lock(&vcpu
->mutex
);
537 mutex_unlock(&vcpu
->mutex
);
541 vcpu
->host_fx_image
= (char*)ALIGN((hva_t
)vcpu
->fx_buf
,
543 vcpu
->guest_fx_image
= vcpu
->host_fx_image
+ FX_IMAGE_SIZE
;
545 vcpu
->cpu
= -1; /* First load will set up TR */
546 r
= kvm_arch_ops
->vcpu_create(vcpu
);
550 r
= kvm_mmu_create(vcpu
);
554 kvm_arch_ops
->vcpu_load(vcpu
);
555 r
= kvm_mmu_setup(vcpu
);
557 r
= kvm_arch_ops
->vcpu_setup(vcpu
);
567 mutex_unlock(&vcpu
->mutex
);
573 * Allocate some memory and give it an address in the guest physical address
576 * Discontiguous memory is allowed, mostly for framebuffers.
578 static int kvm_dev_ioctl_set_memory_region(struct kvm
*kvm
,
579 struct kvm_memory_region
*mem
)
583 unsigned long npages
;
585 struct kvm_memory_slot
*memslot
;
586 struct kvm_memory_slot old
, new;
587 int memory_config_version
;
590 /* General sanity checks */
591 if (mem
->memory_size
& (PAGE_SIZE
- 1))
593 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
595 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
597 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
600 memslot
= &kvm
->memslots
[mem
->slot
];
601 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
602 npages
= mem
->memory_size
>> PAGE_SHIFT
;
605 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
608 spin_lock(&kvm
->lock
);
610 memory_config_version
= kvm
->memory_config_version
;
611 new = old
= *memslot
;
613 new.base_gfn
= base_gfn
;
615 new.flags
= mem
->flags
;
617 /* Disallow changing a memory slot's size. */
619 if (npages
&& old
.npages
&& npages
!= old
.npages
)
622 /* Check for overlaps */
624 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
625 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
629 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
630 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
634 * Do memory allocations outside lock. memory_config_version will
637 spin_unlock(&kvm
->lock
);
639 /* Deallocate if slot is being removed */
643 /* Free page dirty bitmap if unneeded */
644 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
645 new.dirty_bitmap
= 0;
649 /* Allocate if a slot is being created */
650 if (npages
&& !new.phys_mem
) {
651 new.phys_mem
= vmalloc(npages
* sizeof(struct page
*));
656 memset(new.phys_mem
, 0, npages
* sizeof(struct page
*));
657 for (i
= 0; i
< npages
; ++i
) {
658 new.phys_mem
[i
] = alloc_page(GFP_HIGHUSER
660 if (!new.phys_mem
[i
])
662 new.phys_mem
[i
]->private = 0;
666 /* Allocate page dirty bitmap if needed */
667 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
668 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
670 new.dirty_bitmap
= vmalloc(dirty_bytes
);
671 if (!new.dirty_bitmap
)
673 memset(new.dirty_bitmap
, 0, dirty_bytes
);
676 spin_lock(&kvm
->lock
);
678 if (memory_config_version
!= kvm
->memory_config_version
) {
679 spin_unlock(&kvm
->lock
);
680 kvm_free_physmem_slot(&new, &old
);
688 if (mem
->slot
>= kvm
->nmemslots
)
689 kvm
->nmemslots
= mem
->slot
+ 1;
692 ++kvm
->memory_config_version
;
694 spin_unlock(&kvm
->lock
);
696 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
697 struct kvm_vcpu
*vcpu
;
699 vcpu
= vcpu_load(kvm
, i
);
702 kvm_mmu_reset_context(vcpu
);
706 kvm_free_physmem_slot(&old
, &new);
710 spin_unlock(&kvm
->lock
);
712 kvm_free_physmem_slot(&new, &old
);
717 static void do_remove_write_access(struct kvm_vcpu
*vcpu
, int slot
)
719 spin_lock(&vcpu
->kvm
->lock
);
720 kvm_mmu_slot_remove_write_access(vcpu
, slot
);
721 spin_unlock(&vcpu
->kvm
->lock
);
725 * Get (and clear) the dirty memory log for a memory slot.
727 static int kvm_dev_ioctl_get_dirty_log(struct kvm
*kvm
,
728 struct kvm_dirty_log
*log
)
730 struct kvm_memory_slot
*memslot
;
734 unsigned long any
= 0;
736 spin_lock(&kvm
->lock
);
739 * Prevent changes to guest memory configuration even while the lock
743 spin_unlock(&kvm
->lock
);
745 if (log
->slot
>= KVM_MEMORY_SLOTS
)
748 memslot
= &kvm
->memslots
[log
->slot
];
750 if (!memslot
->dirty_bitmap
)
753 n
= ALIGN(memslot
->npages
, 8) / 8;
755 for (i
= 0; !any
&& i
< n
; ++i
)
756 any
= memslot
->dirty_bitmap
[i
];
759 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
765 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
766 struct kvm_vcpu
*vcpu
= vcpu_load(kvm
, i
);
771 do_remove_write_access(vcpu
, log
->slot
);
772 memset(memslot
->dirty_bitmap
, 0, n
);
775 kvm_arch_ops
->tlb_flush(vcpu
);
783 spin_lock(&kvm
->lock
);
785 spin_unlock(&kvm
->lock
);
789 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
793 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
794 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
796 if (gfn
>= memslot
->base_gfn
797 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
802 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
804 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
807 struct kvm_memory_slot
*memslot
= 0;
808 unsigned long rel_gfn
;
810 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
811 memslot
= &kvm
->memslots
[i
];
813 if (gfn
>= memslot
->base_gfn
814 && gfn
< memslot
->base_gfn
+ memslot
->npages
) {
816 if (!memslot
|| !memslot
->dirty_bitmap
)
819 rel_gfn
= gfn
- memslot
->base_gfn
;
822 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
823 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
829 static int emulator_read_std(unsigned long addr
,
832 struct x86_emulate_ctxt
*ctxt
)
834 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
838 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
839 unsigned offset
= addr
& (PAGE_SIZE
-1);
840 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
842 struct kvm_memory_slot
*memslot
;
845 if (gpa
== UNMAPPED_GVA
)
846 return X86EMUL_PROPAGATE_FAULT
;
847 pfn
= gpa
>> PAGE_SHIFT
;
848 memslot
= gfn_to_memslot(vcpu
->kvm
, pfn
);
850 return X86EMUL_UNHANDLEABLE
;
851 page
= kmap_atomic(gfn_to_page(memslot
, pfn
), KM_USER0
);
853 memcpy(data
, page
+ offset
, tocopy
);
855 kunmap_atomic(page
, KM_USER0
);
862 return X86EMUL_CONTINUE
;
865 static int emulator_write_std(unsigned long addr
,
868 struct x86_emulate_ctxt
*ctxt
)
870 printk(KERN_ERR
"emulator_write_std: addr %lx n %d\n",
872 return X86EMUL_UNHANDLEABLE
;
875 static int emulator_read_emulated(unsigned long addr
,
878 struct x86_emulate_ctxt
*ctxt
)
880 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
882 if (vcpu
->mmio_read_completed
) {
883 memcpy(val
, vcpu
->mmio_data
, bytes
);
884 vcpu
->mmio_read_completed
= 0;
885 return X86EMUL_CONTINUE
;
886 } else if (emulator_read_std(addr
, val
, bytes
, ctxt
)
888 return X86EMUL_CONTINUE
;
890 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
891 if (gpa
== UNMAPPED_GVA
)
892 return vcpu_printf(vcpu
, "not present\n"), X86EMUL_PROPAGATE_FAULT
;
893 vcpu
->mmio_needed
= 1;
894 vcpu
->mmio_phys_addr
= gpa
;
895 vcpu
->mmio_size
= bytes
;
896 vcpu
->mmio_is_write
= 0;
898 return X86EMUL_UNHANDLEABLE
;
902 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
903 unsigned long val
, int bytes
)
905 struct kvm_memory_slot
*m
;
909 if (((gpa
+ bytes
- 1) >> PAGE_SHIFT
) != (gpa
>> PAGE_SHIFT
))
911 m
= gfn_to_memslot(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
914 page
= gfn_to_page(m
, gpa
>> PAGE_SHIFT
);
915 kvm_mmu_pre_write(vcpu
, gpa
, bytes
);
916 virt
= kmap_atomic(page
, KM_USER0
);
917 memcpy(virt
+ offset_in_page(gpa
), &val
, bytes
);
918 kunmap_atomic(virt
, KM_USER0
);
919 kvm_mmu_post_write(vcpu
, gpa
, bytes
);
923 static int emulator_write_emulated(unsigned long addr
,
926 struct x86_emulate_ctxt
*ctxt
)
928 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
929 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
931 if (gpa
== UNMAPPED_GVA
)
932 return X86EMUL_PROPAGATE_FAULT
;
934 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
935 return X86EMUL_CONTINUE
;
937 vcpu
->mmio_needed
= 1;
938 vcpu
->mmio_phys_addr
= gpa
;
939 vcpu
->mmio_size
= bytes
;
940 vcpu
->mmio_is_write
= 1;
941 memcpy(vcpu
->mmio_data
, &val
, bytes
);
943 return X86EMUL_CONTINUE
;
946 static int emulator_cmpxchg_emulated(unsigned long addr
,
950 struct x86_emulate_ctxt
*ctxt
)
956 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
958 return emulator_write_emulated(addr
, new, bytes
, ctxt
);
963 static int emulator_cmpxchg8b_emulated(unsigned long addr
,
964 unsigned long old_lo
,
965 unsigned long old_hi
,
966 unsigned long new_lo
,
967 unsigned long new_hi
,
968 struct x86_emulate_ctxt
*ctxt
)
975 printk(KERN_WARNING
"kvm: emulating exchange8b as write\n");
977 r
= emulator_write_emulated(addr
, new_lo
, 4, ctxt
);
978 if (r
!= X86EMUL_CONTINUE
)
980 return emulator_write_emulated(addr
+4, new_hi
, 4, ctxt
);
985 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
987 return kvm_arch_ops
->get_segment_base(vcpu
, seg
);
990 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
992 return X86EMUL_CONTINUE
;
995 int emulate_clts(struct kvm_vcpu
*vcpu
)
999 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1000 cr0
= vcpu
->cr0
& ~CR0_TS_MASK
;
1001 kvm_arch_ops
->set_cr0(vcpu
, cr0
);
1002 return X86EMUL_CONTINUE
;
1005 int emulator_get_dr(struct x86_emulate_ctxt
* ctxt
, int dr
, unsigned long *dest
)
1007 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1011 *dest
= kvm_arch_ops
->get_dr(vcpu
, dr
);
1012 return X86EMUL_CONTINUE
;
1014 printk(KERN_DEBUG
"%s: unexpected dr %u\n",
1016 return X86EMUL_UNHANDLEABLE
;
1020 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1022 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1025 kvm_arch_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1027 /* FIXME: better handling */
1028 return X86EMUL_UNHANDLEABLE
;
1030 return X86EMUL_CONTINUE
;
1033 static void report_emulation_failure(struct x86_emulate_ctxt
*ctxt
)
1035 static int reported
;
1037 unsigned long rip
= ctxt
->vcpu
->rip
;
1038 unsigned long rip_linear
;
1040 rip_linear
= rip
+ get_segment_base(ctxt
->vcpu
, VCPU_SREG_CS
);
1045 emulator_read_std(rip_linear
, (void *)opcodes
, 4, ctxt
);
1047 printk(KERN_ERR
"emulation failed but !mmio_needed?"
1048 " rip %lx %02x %02x %02x %02x\n",
1049 rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1053 struct x86_emulate_ops emulate_ops
= {
1054 .read_std
= emulator_read_std
,
1055 .write_std
= emulator_write_std
,
1056 .read_emulated
= emulator_read_emulated
,
1057 .write_emulated
= emulator_write_emulated
,
1058 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1059 #ifdef CONFIG_X86_32
1060 .cmpxchg8b_emulated
= emulator_cmpxchg8b_emulated
,
1064 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1065 struct kvm_run
*run
,
1069 struct x86_emulate_ctxt emulate_ctxt
;
1073 kvm_arch_ops
->cache_regs(vcpu
);
1075 kvm_arch_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1077 emulate_ctxt
.vcpu
= vcpu
;
1078 emulate_ctxt
.eflags
= kvm_arch_ops
->get_rflags(vcpu
);
1079 emulate_ctxt
.cr2
= cr2
;
1080 emulate_ctxt
.mode
= (emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1081 ? X86EMUL_MODE_REAL
: cs_l
1082 ? X86EMUL_MODE_PROT64
: cs_db
1083 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1085 if (emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1086 emulate_ctxt
.cs_base
= 0;
1087 emulate_ctxt
.ds_base
= 0;
1088 emulate_ctxt
.es_base
= 0;
1089 emulate_ctxt
.ss_base
= 0;
1091 emulate_ctxt
.cs_base
= get_segment_base(vcpu
, VCPU_SREG_CS
);
1092 emulate_ctxt
.ds_base
= get_segment_base(vcpu
, VCPU_SREG_DS
);
1093 emulate_ctxt
.es_base
= get_segment_base(vcpu
, VCPU_SREG_ES
);
1094 emulate_ctxt
.ss_base
= get_segment_base(vcpu
, VCPU_SREG_SS
);
1097 emulate_ctxt
.gs_base
= get_segment_base(vcpu
, VCPU_SREG_GS
);
1098 emulate_ctxt
.fs_base
= get_segment_base(vcpu
, VCPU_SREG_FS
);
1100 vcpu
->mmio_is_write
= 0;
1101 r
= x86_emulate_memop(&emulate_ctxt
, &emulate_ops
);
1103 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1104 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1105 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1106 run
->mmio
.len
= vcpu
->mmio_size
;
1107 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1111 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1112 return EMULATE_DONE
;
1113 if (!vcpu
->mmio_needed
) {
1114 report_emulation_failure(&emulate_ctxt
);
1115 return EMULATE_FAIL
;
1117 return EMULATE_DO_MMIO
;
1120 kvm_arch_ops
->decache_regs(vcpu
);
1121 kvm_arch_ops
->set_rflags(vcpu
, emulate_ctxt
.eflags
);
1123 if (vcpu
->mmio_is_write
)
1124 return EMULATE_DO_MMIO
;
1126 return EMULATE_DONE
;
1128 EXPORT_SYMBOL_GPL(emulate_instruction
);
1130 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
1132 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
1135 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1137 struct descriptor_table dt
= { limit
, base
};
1139 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1142 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
1144 struct descriptor_table dt
= { limit
, base
};
1146 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1149 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
1150 unsigned long *rflags
)
1153 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1156 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
1158 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1169 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1174 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
1175 unsigned long *rflags
)
1179 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
1180 *rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1189 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
1192 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
1196 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1201 case 0xc0010010: /* SYSCFG */
1202 case 0xc0010015: /* HWCR */
1203 case MSR_IA32_PLATFORM_ID
:
1204 case MSR_IA32_P5_MC_ADDR
:
1205 case MSR_IA32_P5_MC_TYPE
:
1206 case MSR_IA32_MC0_CTL
:
1207 case MSR_IA32_MCG_STATUS
:
1208 case MSR_IA32_MCG_CAP
:
1209 case MSR_IA32_MC0_MISC
:
1210 case MSR_IA32_MC0_MISC
+4:
1211 case MSR_IA32_MC0_MISC
+8:
1212 case MSR_IA32_MC0_MISC
+12:
1213 case MSR_IA32_MC0_MISC
+16:
1214 case MSR_IA32_UCODE_REV
:
1215 case MSR_IA32_PERF_STATUS
:
1216 /* MTRR registers */
1218 case 0x200 ... 0x2ff:
1221 case 0xcd: /* fsb frequency */
1224 case MSR_IA32_APICBASE
:
1225 data
= vcpu
->apic_base
;
1227 #ifdef CONFIG_X86_64
1229 data
= vcpu
->shadow_efer
;
1233 printk(KERN_ERR
"kvm: unhandled rdmsr: 0x%x\n", msr
);
1239 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1242 * Reads an msr value (of 'msr_index') into 'pdata'.
1243 * Returns 0 on success, non-0 otherwise.
1244 * Assumes vcpu_load() was already called.
1246 static int get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1248 return kvm_arch_ops
->get_msr(vcpu
, msr_index
, pdata
);
1251 #ifdef CONFIG_X86_64
1253 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1255 if (efer
& EFER_RESERVED_BITS
) {
1256 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
1263 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
1264 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
1269 kvm_arch_ops
->set_efer(vcpu
, efer
);
1272 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
1274 vcpu
->shadow_efer
= efer
;
1279 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
1282 #ifdef CONFIG_X86_64
1284 set_efer(vcpu
, data
);
1287 case MSR_IA32_MC0_STATUS
:
1288 printk(KERN_WARNING
"%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1289 __FUNCTION__
, data
);
1291 case MSR_IA32_UCODE_REV
:
1292 case MSR_IA32_UCODE_WRITE
:
1293 case 0x200 ... 0x2ff: /* MTRRs */
1295 case MSR_IA32_APICBASE
:
1296 vcpu
->apic_base
= data
;
1299 printk(KERN_ERR
"kvm: unhandled wrmsr: 0x%x\n", msr
);
1304 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
1307 * Writes msr value into into the appropriate "register".
1308 * Returns 0 on success, non-0 otherwise.
1309 * Assumes vcpu_load() was already called.
1311 static int set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1313 return kvm_arch_ops
->set_msr(vcpu
, msr_index
, data
);
1316 void kvm_resched(struct kvm_vcpu
*vcpu
)
1320 /* Cannot fail - no vcpu unplug yet. */
1321 vcpu_load(vcpu
->kvm
, vcpu_slot(vcpu
));
1323 EXPORT_SYMBOL_GPL(kvm_resched
);
1325 void load_msrs(struct vmx_msr_entry
*e
, int n
)
1329 for (i
= 0; i
< n
; ++i
)
1330 wrmsrl(e
[i
].index
, e
[i
].data
);
1332 EXPORT_SYMBOL_GPL(load_msrs
);
1334 void save_msrs(struct vmx_msr_entry
*e
, int n
)
1338 for (i
= 0; i
< n
; ++i
)
1339 rdmsrl(e
[i
].index
, e
[i
].data
);
1341 EXPORT_SYMBOL_GPL(save_msrs
);
1343 static int kvm_dev_ioctl_run(struct kvm
*kvm
, struct kvm_run
*kvm_run
)
1345 struct kvm_vcpu
*vcpu
;
1348 if (!valid_vcpu(kvm_run
->vcpu
))
1351 vcpu
= vcpu_load(kvm
, kvm_run
->vcpu
);
1355 if (kvm_run
->emulated
) {
1356 kvm_arch_ops
->skip_emulated_instruction(vcpu
);
1357 kvm_run
->emulated
= 0;
1360 if (kvm_run
->mmio_completed
) {
1361 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
1362 vcpu
->mmio_read_completed
= 1;
1365 vcpu
->mmio_needed
= 0;
1367 r
= kvm_arch_ops
->run(vcpu
, kvm_run
);
1373 static int kvm_dev_ioctl_get_regs(struct kvm
*kvm
, struct kvm_regs
*regs
)
1375 struct kvm_vcpu
*vcpu
;
1377 if (!valid_vcpu(regs
->vcpu
))
1380 vcpu
= vcpu_load(kvm
, regs
->vcpu
);
1384 kvm_arch_ops
->cache_regs(vcpu
);
1386 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
1387 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
1388 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
1389 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
1390 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
1391 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
1392 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
1393 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
1394 #ifdef CONFIG_X86_64
1395 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
1396 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
1397 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
1398 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
1399 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
1400 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
1401 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
1402 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
1405 regs
->rip
= vcpu
->rip
;
1406 regs
->rflags
= kvm_arch_ops
->get_rflags(vcpu
);
1409 * Don't leak debug flags in case they were set for guest debugging
1411 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
1412 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
1419 static int kvm_dev_ioctl_set_regs(struct kvm
*kvm
, struct kvm_regs
*regs
)
1421 struct kvm_vcpu
*vcpu
;
1423 if (!valid_vcpu(regs
->vcpu
))
1426 vcpu
= vcpu_load(kvm
, regs
->vcpu
);
1430 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
1431 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
1432 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
1433 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
1434 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
1435 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
1436 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
1437 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
1438 #ifdef CONFIG_X86_64
1439 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
1440 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
1441 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
1442 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
1443 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
1444 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
1445 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
1446 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
1449 vcpu
->rip
= regs
->rip
;
1450 kvm_arch_ops
->set_rflags(vcpu
, regs
->rflags
);
1452 kvm_arch_ops
->decache_regs(vcpu
);
1459 static void get_segment(struct kvm_vcpu
*vcpu
,
1460 struct kvm_segment
*var
, int seg
)
1462 return kvm_arch_ops
->get_segment(vcpu
, var
, seg
);
1465 static int kvm_dev_ioctl_get_sregs(struct kvm
*kvm
, struct kvm_sregs
*sregs
)
1467 struct kvm_vcpu
*vcpu
;
1468 struct descriptor_table dt
;
1470 if (!valid_vcpu(sregs
->vcpu
))
1472 vcpu
= vcpu_load(kvm
, sregs
->vcpu
);
1476 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1477 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1478 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1479 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1480 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1481 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1483 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1484 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1486 kvm_arch_ops
->get_idt(vcpu
, &dt
);
1487 sregs
->idt
.limit
= dt
.limit
;
1488 sregs
->idt
.base
= dt
.base
;
1489 kvm_arch_ops
->get_gdt(vcpu
, &dt
);
1490 sregs
->gdt
.limit
= dt
.limit
;
1491 sregs
->gdt
.base
= dt
.base
;
1493 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1494 sregs
->cr0
= vcpu
->cr0
;
1495 sregs
->cr2
= vcpu
->cr2
;
1496 sregs
->cr3
= vcpu
->cr3
;
1497 sregs
->cr4
= vcpu
->cr4
;
1498 sregs
->cr8
= vcpu
->cr8
;
1499 sregs
->efer
= vcpu
->shadow_efer
;
1500 sregs
->apic_base
= vcpu
->apic_base
;
1502 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
1503 sizeof sregs
->interrupt_bitmap
);
1510 static void set_segment(struct kvm_vcpu
*vcpu
,
1511 struct kvm_segment
*var
, int seg
)
1513 return kvm_arch_ops
->set_segment(vcpu
, var
, seg
);
1516 static int kvm_dev_ioctl_set_sregs(struct kvm
*kvm
, struct kvm_sregs
*sregs
)
1518 struct kvm_vcpu
*vcpu
;
1519 int mmu_reset_needed
= 0;
1521 struct descriptor_table dt
;
1523 if (!valid_vcpu(sregs
->vcpu
))
1525 vcpu
= vcpu_load(kvm
, sregs
->vcpu
);
1529 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
1530 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
1531 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
1532 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
1533 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
1534 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
1536 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
1537 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
1539 dt
.limit
= sregs
->idt
.limit
;
1540 dt
.base
= sregs
->idt
.base
;
1541 kvm_arch_ops
->set_idt(vcpu
, &dt
);
1542 dt
.limit
= sregs
->gdt
.limit
;
1543 dt
.base
= sregs
->gdt
.base
;
1544 kvm_arch_ops
->set_gdt(vcpu
, &dt
);
1546 vcpu
->cr2
= sregs
->cr2
;
1547 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
1548 vcpu
->cr3
= sregs
->cr3
;
1550 vcpu
->cr8
= sregs
->cr8
;
1552 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
1553 #ifdef CONFIG_X86_64
1554 kvm_arch_ops
->set_efer(vcpu
, sregs
->efer
);
1556 vcpu
->apic_base
= sregs
->apic_base
;
1558 kvm_arch_ops
->decache_cr0_cr4_guest_bits(vcpu
);
1560 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
1561 kvm_arch_ops
->set_cr0_no_modeswitch(vcpu
, sregs
->cr0
);
1563 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
1564 kvm_arch_ops
->set_cr4(vcpu
, sregs
->cr4
);
1565 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
1566 load_pdptrs(vcpu
, vcpu
->cr3
);
1568 if (mmu_reset_needed
)
1569 kvm_mmu_reset_context(vcpu
);
1571 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
1572 sizeof vcpu
->irq_pending
);
1573 vcpu
->irq_summary
= 0;
1574 for (i
= 0; i
< NR_IRQ_WORDS
; ++i
)
1575 if (vcpu
->irq_pending
[i
])
1576 __set_bit(i
, &vcpu
->irq_summary
);
1584 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1585 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1587 * This list is modified at module load time to reflect the
1588 * capabilities of the host cpu.
1590 static u32 msrs_to_save
[] = {
1591 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1593 #ifdef CONFIG_X86_64
1594 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1596 MSR_IA32_TIME_STAMP_COUNTER
,
1599 static unsigned num_msrs_to_save
;
1601 static __init
void kvm_init_msr_list(void)
1606 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
1607 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
1610 msrs_to_save
[j
] = msrs_to_save
[i
];
1613 num_msrs_to_save
= j
;
1617 * Adapt set_msr() to msr_io()'s calling convention
1619 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1621 return set_msr(vcpu
, index
, *data
);
1625 * Read or write a bunch of msrs. All parameters are kernel addresses.
1627 * @return number of msrs set successfully.
1629 static int __msr_io(struct kvm
*kvm
, struct kvm_msrs
*msrs
,
1630 struct kvm_msr_entry
*entries
,
1631 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1632 unsigned index
, u64
*data
))
1634 struct kvm_vcpu
*vcpu
;
1637 if (!valid_vcpu(msrs
->vcpu
))
1640 vcpu
= vcpu_load(kvm
, msrs
->vcpu
);
1644 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1645 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1654 * Read or write a bunch of msrs. Parameters are user addresses.
1656 * @return number of msrs set successfully.
1658 static int msr_io(struct kvm
*kvm
, struct kvm_msrs __user
*user_msrs
,
1659 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1660 unsigned index
, u64
*data
),
1663 struct kvm_msrs msrs
;
1664 struct kvm_msr_entry
*entries
;
1669 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1673 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1677 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1678 entries
= vmalloc(size
);
1683 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1686 r
= n
= __msr_io(kvm
, &msrs
, entries
, do_msr
);
1691 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1703 * Translate a guest virtual address to a guest physical address.
1705 static int kvm_dev_ioctl_translate(struct kvm
*kvm
, struct kvm_translation
*tr
)
1707 unsigned long vaddr
= tr
->linear_address
;
1708 struct kvm_vcpu
*vcpu
;
1711 vcpu
= vcpu_load(kvm
, tr
->vcpu
);
1714 spin_lock(&kvm
->lock
);
1715 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
1716 tr
->physical_address
= gpa
;
1717 tr
->valid
= gpa
!= UNMAPPED_GVA
;
1720 spin_unlock(&kvm
->lock
);
1726 static int kvm_dev_ioctl_interrupt(struct kvm
*kvm
, struct kvm_interrupt
*irq
)
1728 struct kvm_vcpu
*vcpu
;
1730 if (!valid_vcpu(irq
->vcpu
))
1732 if (irq
->irq
< 0 || irq
->irq
>= 256)
1734 vcpu
= vcpu_load(kvm
, irq
->vcpu
);
1738 set_bit(irq
->irq
, vcpu
->irq_pending
);
1739 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
1746 static int kvm_dev_ioctl_debug_guest(struct kvm
*kvm
,
1747 struct kvm_debug_guest
*dbg
)
1749 struct kvm_vcpu
*vcpu
;
1752 if (!valid_vcpu(dbg
->vcpu
))
1754 vcpu
= vcpu_load(kvm
, dbg
->vcpu
);
1758 r
= kvm_arch_ops
->set_guest_debug(vcpu
, dbg
);
1765 static long kvm_dev_ioctl(struct file
*filp
,
1766 unsigned int ioctl
, unsigned long arg
)
1768 struct kvm
*kvm
= filp
->private_data
;
1772 case KVM_GET_API_VERSION
:
1773 r
= KVM_API_VERSION
;
1775 case KVM_CREATE_VCPU
: {
1776 r
= kvm_dev_ioctl_create_vcpu(kvm
, arg
);
1782 struct kvm_run kvm_run
;
1785 if (copy_from_user(&kvm_run
, (void *)arg
, sizeof kvm_run
))
1787 r
= kvm_dev_ioctl_run(kvm
, &kvm_run
);
1788 if (r
< 0 && r
!= -EINTR
)
1790 if (copy_to_user((void *)arg
, &kvm_run
, sizeof kvm_run
)) {
1796 case KVM_GET_REGS
: {
1797 struct kvm_regs kvm_regs
;
1800 if (copy_from_user(&kvm_regs
, (void *)arg
, sizeof kvm_regs
))
1802 r
= kvm_dev_ioctl_get_regs(kvm
, &kvm_regs
);
1806 if (copy_to_user((void *)arg
, &kvm_regs
, sizeof kvm_regs
))
1811 case KVM_SET_REGS
: {
1812 struct kvm_regs kvm_regs
;
1815 if (copy_from_user(&kvm_regs
, (void *)arg
, sizeof kvm_regs
))
1817 r
= kvm_dev_ioctl_set_regs(kvm
, &kvm_regs
);
1823 case KVM_GET_SREGS
: {
1824 struct kvm_sregs kvm_sregs
;
1827 if (copy_from_user(&kvm_sregs
, (void *)arg
, sizeof kvm_sregs
))
1829 r
= kvm_dev_ioctl_get_sregs(kvm
, &kvm_sregs
);
1833 if (copy_to_user((void *)arg
, &kvm_sregs
, sizeof kvm_sregs
))
1838 case KVM_SET_SREGS
: {
1839 struct kvm_sregs kvm_sregs
;
1842 if (copy_from_user(&kvm_sregs
, (void *)arg
, sizeof kvm_sregs
))
1844 r
= kvm_dev_ioctl_set_sregs(kvm
, &kvm_sregs
);
1850 case KVM_TRANSLATE
: {
1851 struct kvm_translation tr
;
1854 if (copy_from_user(&tr
, (void *)arg
, sizeof tr
))
1856 r
= kvm_dev_ioctl_translate(kvm
, &tr
);
1860 if (copy_to_user((void *)arg
, &tr
, sizeof tr
))
1865 case KVM_INTERRUPT
: {
1866 struct kvm_interrupt irq
;
1869 if (copy_from_user(&irq
, (void *)arg
, sizeof irq
))
1871 r
= kvm_dev_ioctl_interrupt(kvm
, &irq
);
1877 case KVM_DEBUG_GUEST
: {
1878 struct kvm_debug_guest dbg
;
1881 if (copy_from_user(&dbg
, (void *)arg
, sizeof dbg
))
1883 r
= kvm_dev_ioctl_debug_guest(kvm
, &dbg
);
1889 case KVM_SET_MEMORY_REGION
: {
1890 struct kvm_memory_region kvm_mem
;
1893 if (copy_from_user(&kvm_mem
, (void *)arg
, sizeof kvm_mem
))
1895 r
= kvm_dev_ioctl_set_memory_region(kvm
, &kvm_mem
);
1900 case KVM_GET_DIRTY_LOG
: {
1901 struct kvm_dirty_log log
;
1904 if (copy_from_user(&log
, (void *)arg
, sizeof log
))
1906 r
= kvm_dev_ioctl_get_dirty_log(kvm
, &log
);
1912 r
= msr_io(kvm
, (void __user
*)arg
, get_msr
, 1);
1915 r
= msr_io(kvm
, (void __user
*)arg
, do_set_msr
, 0);
1917 case KVM_GET_MSR_INDEX_LIST
: {
1918 struct kvm_msr_list __user
*user_msr_list
= (void __user
*)arg
;
1919 struct kvm_msr_list msr_list
;
1923 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1926 msr_list
.nmsrs
= num_msrs_to_save
;
1927 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1930 if (n
< num_msrs_to_save
)
1933 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1934 num_msrs_to_save
* sizeof(u32
)))
1946 static struct page
*kvm_dev_nopage(struct vm_area_struct
*vma
,
1947 unsigned long address
,
1950 struct kvm
*kvm
= vma
->vm_file
->private_data
;
1951 unsigned long pgoff
;
1952 struct kvm_memory_slot
*slot
;
1955 *type
= VM_FAULT_MINOR
;
1956 pgoff
= ((address
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1957 slot
= gfn_to_memslot(kvm
, pgoff
);
1959 return NOPAGE_SIGBUS
;
1960 page
= gfn_to_page(slot
, pgoff
);
1962 return NOPAGE_SIGBUS
;
1967 static struct vm_operations_struct kvm_dev_vm_ops
= {
1968 .nopage
= kvm_dev_nopage
,
1971 static int kvm_dev_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1973 vma
->vm_ops
= &kvm_dev_vm_ops
;
1977 static struct file_operations kvm_chardev_ops
= {
1978 .open
= kvm_dev_open
,
1979 .release
= kvm_dev_release
,
1980 .unlocked_ioctl
= kvm_dev_ioctl
,
1981 .compat_ioctl
= kvm_dev_ioctl
,
1982 .mmap
= kvm_dev_mmap
,
1985 static struct miscdevice kvm_dev
= {
1991 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
1994 if (val
== SYS_RESTART
) {
1996 * Some (well, at least mine) BIOSes hang on reboot if
1999 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2000 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
2005 static struct notifier_block kvm_reboot_notifier
= {
2006 .notifier_call
= kvm_reboot
,
2010 static __init
void kvm_init_debug(void)
2012 struct kvm_stats_debugfs_item
*p
;
2014 debugfs_dir
= debugfs_create_dir("kvm", 0);
2015 for (p
= debugfs_entries
; p
->name
; ++p
)
2016 p
->dentry
= debugfs_create_u32(p
->name
, 0444, debugfs_dir
,
2020 static void kvm_exit_debug(void)
2022 struct kvm_stats_debugfs_item
*p
;
2024 for (p
= debugfs_entries
; p
->name
; ++p
)
2025 debugfs_remove(p
->dentry
);
2026 debugfs_remove(debugfs_dir
);
2029 hpa_t bad_page_address
;
2031 int kvm_init_arch(struct kvm_arch_ops
*ops
, struct module
*module
)
2036 printk(KERN_ERR
"kvm: already loaded the other module\n");
2040 if (!ops
->cpu_has_kvm_support()) {
2041 printk(KERN_ERR
"kvm: no hardware support\n");
2044 if (ops
->disabled_by_bios()) {
2045 printk(KERN_ERR
"kvm: disabled by bios\n");
2051 r
= kvm_arch_ops
->hardware_setup();
2055 on_each_cpu(kvm_arch_ops
->hardware_enable
, 0, 0, 1);
2056 register_reboot_notifier(&kvm_reboot_notifier
);
2058 kvm_chardev_ops
.owner
= module
;
2060 r
= misc_register(&kvm_dev
);
2062 printk (KERN_ERR
"kvm: misc device register failed\n");
2069 unregister_reboot_notifier(&kvm_reboot_notifier
);
2070 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
2071 kvm_arch_ops
->hardware_unsetup();
2075 void kvm_exit_arch(void)
2077 misc_deregister(&kvm_dev
);
2079 unregister_reboot_notifier(&kvm_reboot_notifier
);
2080 on_each_cpu(kvm_arch_ops
->hardware_disable
, 0, 0, 1);
2081 kvm_arch_ops
->hardware_unsetup();
2082 kvm_arch_ops
= NULL
;
2085 static __init
int kvm_init(void)
2087 static struct page
*bad_page
;
2092 kvm_init_msr_list();
2094 if ((bad_page
= alloc_page(GFP_KERNEL
)) == NULL
) {
2099 bad_page_address
= page_to_pfn(bad_page
) << PAGE_SHIFT
;
2100 memset(__va(bad_page_address
), 0, PAGE_SIZE
);
2109 static __exit
void kvm_exit(void)
2112 __free_page(pfn_to_page(bad_page_address
>> PAGE_SHIFT
));
2115 module_init(kvm_init
)
2116 module_exit(kvm_exit
)
2118 EXPORT_SYMBOL_GPL(kvm_init_arch
);
2119 EXPORT_SYMBOL_GPL(kvm_exit_arch
);