projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: VMX: Avoid unnecessary vcpu_load()/vcpu_put() cycles
[GitHub/exynos8895/android_kernel_samsung_universal8895.git] / drivers / kvm / kvm_main.c
CommitLineData
6aa8b732
AK		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18#include "kvm.h"
19
20#include <linux/kvm.h>
21#include <linux/module.h>
22#include <linux/errno.h>
e9cdb1e3 23#include <linux/magic.h>
6aa8b732
AK		 24#include <asm/processor.h>
25#include <linux/percpu.h>
26#include <linux/gfp.h>
27#include <asm/msr.h>
28#include <linux/mm.h>
29#include <linux/miscdevice.h>
30#include <linux/vmalloc.h>
31#include <asm/uaccess.h>
32#include <linux/reboot.h>
33#include <asm/io.h>
34#include <linux/debugfs.h>
35#include <linux/highmem.h>
36#include <linux/file.h>
37#include <asm/desc.h>
59ae6c6b 38#include <linux/sysdev.h>
774c47f1 39#include <linux/cpu.h>
f17abe9a 40#include <linux/file.h>
37e29d90
AK		 41#include <linux/fs.h>
42#include <linux/mount.h>
6aa8b732
AK		 43
44#include "x86_emulate.h"
45#include "segment_descriptor.h"
46
47MODULE_AUTHOR("Qumranet");
48MODULE_LICENSE("GPL");
49
133de902
AK		 50static DEFINE_SPINLOCK(kvm_lock);
51static LIST_HEAD(vm_list);
52
6aa8b732
AK		 53struct kvm_arch_ops *kvm_arch_ops;
54struct kvm_stat kvm_stat;
55EXPORT_SYMBOL_GPL(kvm_stat);
56
57static struct kvm_stats_debugfs_item {
58 const char *name;
59 u32 *data;
60 struct dentry *dentry;
61} debugfs_entries[] = {
62 { "pf_fixed", &kvm_stat.pf_fixed },
63 { "pf_guest", &kvm_stat.pf_guest },
64 { "tlb_flush", &kvm_stat.tlb_flush },
65 { "invlpg", &kvm_stat.invlpg },
66 { "exits", &kvm_stat.exits },
67 { "io_exits", &kvm_stat.io_exits },
68 { "mmio_exits", &kvm_stat.mmio_exits },
69 { "signal_exits", &kvm_stat.signal_exits },
c1150d8c
DL		 70 { "irq_window", &kvm_stat.irq_window_exits },
71 { "halt_exits", &kvm_stat.halt_exits },
72 { "request_irq", &kvm_stat.request_irq_exits },
6aa8b732 73 { "irq_exits", &kvm_stat.irq_exits },
8b6d44c7 74 { NULL, NULL }
6aa8b732
AK		 75};
76
77static struct dentry *debugfs_dir;
78
37e29d90
AK		 79struct vfsmount *kvmfs_mnt;
80
6aa8b732
AK		 81#define MAX_IO_MSRS 256
82
83#define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
84#define LMSW_GUEST_MASK 0x0eULL
85#define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
86#define CR8_RESEVED_BITS (~0x0fULL)
87#define EFER_RESERVED_BITS 0xfffffffffffff2fe
88
05b3e0c2 89#ifdef CONFIG_X86_64
6aa8b732
AK		 90// LDT or TSS descriptor in the GDT. 16 bytes.
91struct segment_descriptor_64 {
92 struct segment_descriptor s;
93 u32 base_higher;
94 u32 pad_zero;
95};
96
97#endif
98
bccf2150
AK		 99static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
100 unsigned long arg);
101
f17abe9a
AK		 102static struct inode *kvmfs_inode(struct file_operations *fops)
103{
104 int error = -ENOMEM;
105 struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
106
107 if (!inode)
108 goto eexit_1;
109
110 inode->i_fop = fops;
111
112 /*
113 * Mark the inode dirty from the very beginning,
114 * that way it will never be moved to the dirty
115 * list because mark_inode_dirty() will think
116 * that it already _is_ on the dirty list.
117 */
118 inode->i_state = I_DIRTY;
119 inode->i_mode = S_IRUSR | S_IWUSR;
120 inode->i_uid = current->fsuid;
121 inode->i_gid = current->fsgid;
122 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
123 return inode;
124
125eexit_1:
126 return ERR_PTR(error);
127}
128
129static struct file *kvmfs_file(struct inode *inode, void *private_data)
130{
131 struct file *file = get_empty_filp();
132
133 if (!file)
134 return ERR_PTR(-ENFILE);
135
136 file->f_path.mnt = mntget(kvmfs_mnt);
137 file->f_path.dentry = d_alloc_anon(inode);
138 if (!file->f_path.dentry)
139 return ERR_PTR(-ENOMEM);
140 file->f_mapping = inode->i_mapping;
141
142 file->f_pos = 0;
143 file->f_flags = O_RDWR;
144 file->f_op = inode->i_fop;
145 file->f_mode = FMODE_READ | FMODE_WRITE;
146 file->f_version = 0;
147 file->private_data = private_data;
148 return file;
149}
150
6aa8b732
AK		 151unsigned long segment_base(u16 selector)
152{
153 struct descriptor_table gdt;
154 struct segment_descriptor *d;
155 unsigned long table_base;
156 typedef unsigned long ul;
157 unsigned long v;
158
159 if (selector == 0)
160 return 0;
161
162 asm ("sgdt %0" : "=m"(gdt));
163 table_base = gdt.base;
164
165 if (selector & 4) { /* from ldt */
166 u16 ldt_selector;
167
168 asm ("sldt %0" : "=g"(ldt_selector));
169 table_base = segment_base(ldt_selector);
170 }
171 d = (struct segment_descriptor *)(table_base + (selector & ~7));
172 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
05b3e0c2 173#ifdef CONFIG_X86_64
6aa8b732
AK		 174 if (d->system == 0
175 && (d->type == 2 || d->type == 9 || d->type == 11))
176 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
177#endif
178 return v;
179}
180EXPORT_SYMBOL_GPL(segment_base);
181
5aacf0ca
JM		 182static inline int valid_vcpu(int n)
183{
184 return likely(n >= 0 && n < KVM_MAX_VCPUS);
185}
186
d27d4aca
AK		 187int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
188 void *dest)
6aa8b732
AK		 189{
190 unsigned char *host_buf = dest;
191 unsigned long req_size = size;
192
193 while (size) {
194 hpa_t paddr;
195 unsigned now;
196 unsigned offset;
197 hva_t guest_buf;
198
199 paddr = gva_to_hpa(vcpu, addr);
200
201 if (is_error_hpa(paddr))
202 break;
203
204 guest_buf = (hva_t)kmap_atomic(
205 pfn_to_page(paddr >> PAGE_SHIFT),
206 KM_USER0);
207 offset = addr & ~PAGE_MASK;
208 guest_buf |= offset;
209 now = min(size, PAGE_SIZE - offset);
210 memcpy(host_buf, (void*)guest_buf, now);
211 host_buf += now;
212 addr += now;
213 size -= now;
214 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
215 }
216 return req_size - size;
217}
218EXPORT_SYMBOL_GPL(kvm_read_guest);
219
d27d4aca
AK		 220int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
221 void *data)
6aa8b732
AK		 222{
223 unsigned char *host_buf = data;
224 unsigned long req_size = size;
225
226 while (size) {
227 hpa_t paddr;
228 unsigned now;
229 unsigned offset;
230 hva_t guest_buf;
ab51a434 231 gfn_t gfn;
6aa8b732
AK		 232
233 paddr = gva_to_hpa(vcpu, addr);
234
235 if (is_error_hpa(paddr))
236 break;
237
ab51a434
UL		 238 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
239 mark_page_dirty(vcpu->kvm, gfn);
6aa8b732
AK		 240 guest_buf = (hva_t)kmap_atomic(
241 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
242 offset = addr & ~PAGE_MASK;
243 guest_buf |= offset;
244 now = min(size, PAGE_SIZE - offset);
245 memcpy((void*)guest_buf, host_buf, now);
246 host_buf += now;
247 addr += now;
248 size -= now;
249 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
250 }
251 return req_size - size;
252}
253EXPORT_SYMBOL_GPL(kvm_write_guest);
254
bccf2150
AK		 255/*
256 * Switches to specified vcpu, until a matching vcpu_put()
257 */
258static void vcpu_load(struct kvm_vcpu *vcpu)
6aa8b732 259{
bccf2150
AK		 260 mutex_lock(&vcpu->mutex);
261 kvm_arch_ops->vcpu_load(vcpu);
6aa8b732
AK		 262}
263
264/*
bccf2150
AK		 265 * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
266 * if the slot is not populated.
6aa8b732 267 */
bccf2150 268static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
6aa8b732 269{
bccf2150 270 struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
6aa8b732
AK		 271
272 mutex_lock(&vcpu->mutex);
bccf2150 273 if (!vcpu->vmcs) {
6aa8b732 274 mutex_unlock(&vcpu->mutex);
8b6d44c7 275 return NULL;
6aa8b732 276 }
bccf2150
AK		 277 kvm_arch_ops->vcpu_load(vcpu);
278 return vcpu;
6aa8b732
AK		 279}
280
281static void vcpu_put(struct kvm_vcpu *vcpu)
282{
283 kvm_arch_ops->vcpu_put(vcpu);
6aa8b732
AK		 284 mutex_unlock(&vcpu->mutex);
285}
286
f17abe9a 287static struct kvm *kvm_create_vm(void)
6aa8b732
AK		 288{
289 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
290 int i;
291
292 if (!kvm)
f17abe9a 293 return ERR_PTR(-ENOMEM);
6aa8b732
AK		 294
295 spin_lock_init(&kvm->lock);
296 INIT_LIST_HEAD(&kvm->active_mmu_pages);
297 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
298 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
299
300 mutex_init(&vcpu->mutex);
133de902 301 vcpu->cpu = -1;
86a2b42e 302 vcpu->kvm = kvm;
6aa8b732
AK		 303 vcpu->mmu.root_hpa = INVALID_PAGE;
304 INIT_LIST_HEAD(&vcpu->free_pages);
133de902
AK		 305 spin_lock(&kvm_lock);
306 list_add(&kvm->vm_list, &vm_list);
307 spin_unlock(&kvm_lock);
6aa8b732 308 }
f17abe9a
AK		 309 return kvm;
310}
311
312static int kvm_dev_open(struct inode *inode, struct file *filp)
313{
6aa8b732
AK		 314 return 0;
315}
316
317/*
318 * Free any memory in @free but not in @dont.
319 */
320static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
321 struct kvm_memory_slot *dont)
322{
323 int i;
324
325 if (!dont || free->phys_mem != dont->phys_mem)
326 if (free->phys_mem) {
327 for (i = 0; i < free->npages; ++i)
55a54f79
AK		 328 if (free->phys_mem[i])
329 __free_page(free->phys_mem[i]);
6aa8b732
AK		 330 vfree(free->phys_mem);
331 }
332
333 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
334 vfree(free->dirty_bitmap);
335
8b6d44c7 336 free->phys_mem = NULL;
6aa8b732 337 free->npages = 0;
8b6d44c7 338 free->dirty_bitmap = NULL;
6aa8b732
AK		 339}
340
341static void kvm_free_physmem(struct kvm *kvm)
342{
343 int i;
344
345 for (i = 0; i < kvm->nmemslots; ++i)
8b6d44c7 346 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
6aa8b732
AK		 347}
348
039576c0
AK		 349static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
350{
351 int i;
352
353 for (i = 0; i < 2; ++i)
354 if (vcpu->pio.guest_pages[i]) {
355 __free_page(vcpu->pio.guest_pages[i]);
356 vcpu->pio.guest_pages[i] = NULL;
357 }
358}
359
6aa8b732
AK		 360static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
361{
bccf2150 362 if (!vcpu->vmcs)
1e8ba6fb
IM		 363 return;
364
bccf2150 365 vcpu_load(vcpu);
6aa8b732 366 kvm_mmu_destroy(vcpu);
08438475 367 vcpu_put(vcpu);
9ede74e0 368 kvm_arch_ops->vcpu_free(vcpu);
9a2bb7f4
AK		 369 free_page((unsigned long)vcpu->run);
370 vcpu->run = NULL;
039576c0
AK		 371 free_page((unsigned long)vcpu->pio_data);
372 vcpu->pio_data = NULL;
373 free_pio_guest_pages(vcpu);
6aa8b732
AK		 374}
375
376static void kvm_free_vcpus(struct kvm *kvm)
377{
378 unsigned int i;
379
380 for (i = 0; i < KVM_MAX_VCPUS; ++i)
381 kvm_free_vcpu(&kvm->vcpus[i]);
382}
383
384static int kvm_dev_release(struct inode *inode, struct file *filp)
385{
f17abe9a
AK		 386 return 0;
387}
6aa8b732 388
f17abe9a
AK		 389static void kvm_destroy_vm(struct kvm *kvm)
390{
133de902
AK		 391 spin_lock(&kvm_lock);
392 list_del(&kvm->vm_list);
393 spin_unlock(&kvm_lock);
6aa8b732
AK		 394 kvm_free_vcpus(kvm);
395 kvm_free_physmem(kvm);
396 kfree(kvm);
f17abe9a
AK		 397}
398
399static int kvm_vm_release(struct inode *inode, struct file *filp)
400{
401 struct kvm *kvm = filp->private_data;
402
403 kvm_destroy_vm(kvm);
6aa8b732
AK		 404 return 0;
405}
406
407static void inject_gp(struct kvm_vcpu *vcpu)
408{
409 kvm_arch_ops->inject_gp(vcpu, 0);
410}
411
1342d353
AK		 412/*
413 * Load the pae pdptrs. Return true is they are all valid.
414 */
415static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
6aa8b732
AK		 416{
417 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
1342d353 418 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
6aa8b732
AK		 419 int i;
420 u64 pdpte;
421 u64 *pdpt;
1342d353 422 int ret;
954bbbc2 423 struct page *page;
6aa8b732
AK		 424
425 spin_lock(&vcpu->kvm->lock);
954bbbc2
AK		 426 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
427 /* FIXME: !page - emulate? 0xff? */
428 pdpt = kmap_atomic(page, KM_USER0);
6aa8b732 429
1342d353 430 ret = 1;
6aa8b732
AK		 431 for (i = 0; i < 4; ++i) {
432 pdpte = pdpt[offset + i];
1342d353
AK		 433 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
434 ret = 0;
435 goto out;
436 }
6aa8b732
AK		 437 }
438
1342d353
AK		 439 for (i = 0; i < 4; ++i)
440 vcpu->pdptrs[i] = pdpt[offset + i];
441
442out:
6aa8b732
AK		 443 kunmap_atomic(pdpt, KM_USER0);
444 spin_unlock(&vcpu->kvm->lock);
445
1342d353 446 return ret;
6aa8b732
AK		 447}
448
449void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
450{
451 if (cr0 & CR0_RESEVED_BITS) {
452 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
453 cr0, vcpu->cr0);
454 inject_gp(vcpu);
455 return;
456 }
457
458 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
459 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
460 inject_gp(vcpu);
461 return;
462 }
463
464 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
465 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
466 "and a clear PE flag\n");
467 inject_gp(vcpu);
468 return;
469 }
470
471 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
05b3e0c2 472#ifdef CONFIG_X86_64
6aa8b732
AK		 473 if ((vcpu->shadow_efer & EFER_LME)) {
474 int cs_db, cs_l;
475
476 if (!is_pae(vcpu)) {
477 printk(KERN_DEBUG "set_cr0: #GP, start paging "
478 "in long mode while PAE is disabled\n");
479 inject_gp(vcpu);
480 return;
481 }
482 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
483 if (cs_l) {
484 printk(KERN_DEBUG "set_cr0: #GP, start paging "
485 "in long mode while CS.L == 1\n");
486 inject_gp(vcpu);
487 return;
488
489 }
490 } else
491#endif
1342d353 492 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 493 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
494 "reserved bits\n");
495 inject_gp(vcpu);
496 return;
497 }
498
499 }
500
501 kvm_arch_ops->set_cr0(vcpu, cr0);
502 vcpu->cr0 = cr0;
503
504 spin_lock(&vcpu->kvm->lock);
505 kvm_mmu_reset_context(vcpu);
506 spin_unlock(&vcpu->kvm->lock);
507 return;
508}
509EXPORT_SYMBOL_GPL(set_cr0);
510
511void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
512{
399badf3 513 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
6aa8b732
AK		 514 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
515}
516EXPORT_SYMBOL_GPL(lmsw);
517
518void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
519{
520 if (cr4 & CR4_RESEVED_BITS) {
521 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
522 inject_gp(vcpu);
523 return;
524 }
525
a9058ecd 526 if (is_long_mode(vcpu)) {
6aa8b732
AK		 527 if (!(cr4 & CR4_PAE_MASK)) {
528 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
529 "in long mode\n");
530 inject_gp(vcpu);
531 return;
532 }
533 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
1342d353 534 && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 535 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
536 inject_gp(vcpu);
537 }
538
539 if (cr4 & CR4_VMXE_MASK) {
540 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
541 inject_gp(vcpu);
542 return;
543 }
544 kvm_arch_ops->set_cr4(vcpu, cr4);
545 spin_lock(&vcpu->kvm->lock);
546 kvm_mmu_reset_context(vcpu);
547 spin_unlock(&vcpu->kvm->lock);
548}
549EXPORT_SYMBOL_GPL(set_cr4);
550
551void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
552{
a9058ecd 553 if (is_long_mode(vcpu)) {
d27d4aca 554 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
6aa8b732
AK		 555 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
556 inject_gp(vcpu);
557 return;
558 }
559 } else {
560 if (cr3 & CR3_RESEVED_BITS) {
561 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
562 inject_gp(vcpu);
563 return;
564 }
565 if (is_paging(vcpu) && is_pae(vcpu) &&
1342d353 566 !load_pdptrs(vcpu, cr3)) {
6aa8b732
AK		 567 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
568 "reserved bits\n");
569 inject_gp(vcpu);
570 return;
571 }
572 }
573
574 vcpu->cr3 = cr3;
575 spin_lock(&vcpu->kvm->lock);
d21225ee
IM		 576 /*
577 * Does the new cr3 value map to physical memory? (Note, we
578 * catch an invalid cr3 even in real-mode, because it would
579 * cause trouble later on when we turn on paging anyway.)
580 *
581 * A real CPU would silently accept an invalid cr3 and would
582 * attempt to use it - with largely undefined (and often hard
583 * to debug) behavior on the guest side.
584 */
585 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
586 inject_gp(vcpu);
587 else
588 vcpu->mmu.new_cr3(vcpu);
6aa8b732
AK		 589 spin_unlock(&vcpu->kvm->lock);
590}
591EXPORT_SYMBOL_GPL(set_cr3);
592
593void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
594{
595 if ( cr8 & CR8_RESEVED_BITS) {
596 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
597 inject_gp(vcpu);
598 return;
599 }
600 vcpu->cr8 = cr8;
601}
602EXPORT_SYMBOL_GPL(set_cr8);
603
604void fx_init(struct kvm_vcpu *vcpu)
605{
606 struct __attribute__ ((__packed__)) fx_image_s {
607 u16 control; //fcw
608 u16 status; //fsw
609 u16 tag; // ftw
610 u16 opcode; //fop
611 u64 ip; // fpu ip
612 u64 operand;// fpu dp
613 u32 mxcsr;
614 u32 mxcsr_mask;
615
616 } *fx_image;
617
618 fx_save(vcpu->host_fx_image);
619 fpu_init();
620 fx_save(vcpu->guest_fx_image);
621 fx_restore(vcpu->host_fx_image);
622
623 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
624 fx_image->mxcsr = 0x1f80;
625 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
626 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
627}
628EXPORT_SYMBOL_GPL(fx_init);
629
02b27c1f
UL		 630static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
631{
632 spin_lock(&vcpu->kvm->lock);
633 kvm_mmu_slot_remove_write_access(vcpu, slot);
634 spin_unlock(&vcpu->kvm->lock);
635}
636
6aa8b732
AK		 637/*
638 * Allocate some memory and give it an address in the guest physical address
639 * space.
640 *
641 * Discontiguous memory is allowed, mostly for framebuffers.
642 */
2c6f5df9
AK		 643static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
644 struct kvm_memory_region *mem)
6aa8b732
AK		 645{
646 int r;
647 gfn_t base_gfn;
648 unsigned long npages;
649 unsigned long i;
650 struct kvm_memory_slot *memslot;
651 struct kvm_memory_slot old, new;
652 int memory_config_version;
653
654 r = -EINVAL;
655 /* General sanity checks */
656 if (mem->memory_size & (PAGE_SIZE - 1))
657 goto out;
658 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
659 goto out;
660 if (mem->slot >= KVM_MEMORY_SLOTS)
661 goto out;
662 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
663 goto out;
664
665 memslot = &kvm->memslots[mem->slot];
666 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
667 npages = mem->memory_size >> PAGE_SHIFT;
668
669 if (!npages)
670 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
671
672raced:
673 spin_lock(&kvm->lock);
674
675 memory_config_version = kvm->memory_config_version;
676 new = old = *memslot;
677
678 new.base_gfn = base_gfn;
679 new.npages = npages;
680 new.flags = mem->flags;
681
682 /* Disallow changing a memory slot's size. */
683 r = -EINVAL;
684 if (npages && old.npages && npages != old.npages)
685 goto out_unlock;
686
687 /* Check for overlaps */
688 r = -EEXIST;
689 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
690 struct kvm_memory_slot *s = &kvm->memslots[i];
691
692 if (s == memslot)
693 continue;
694 if (!((base_gfn + npages <= s->base_gfn) ||
695 (base_gfn >= s->base_gfn + s->npages)))
696 goto out_unlock;
697 }
698 /*
699 * Do memory allocations outside lock. memory_config_version will
700 * detect any races.
701 */
702 spin_unlock(&kvm->lock);
703
704 /* Deallocate if slot is being removed */
705 if (!npages)
8b6d44c7 706 new.phys_mem = NULL;
6aa8b732
AK		 707
708 /* Free page dirty bitmap if unneeded */
709 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
8b6d44c7 710 new.dirty_bitmap = NULL;
6aa8b732
AK		 711
712 r = -ENOMEM;
713
714 /* Allocate if a slot is being created */
715 if (npages && !new.phys_mem) {
716 new.phys_mem = vmalloc(npages * sizeof(struct page *));
717
718 if (!new.phys_mem)
719 goto out_free;
720
721 memset(new.phys_mem, 0, npages * sizeof(struct page *));
722 for (i = 0; i < npages; ++i) {
723 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
724 | __GFP_ZERO);
725 if (!new.phys_mem[i])
726 goto out_free;
5972e953 727 set_page_private(new.phys_mem[i],0);
6aa8b732
AK		 728 }
729 }
730
731 /* Allocate page dirty bitmap if needed */
732 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
733 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
734
735 new.dirty_bitmap = vmalloc(dirty_bytes);
736 if (!new.dirty_bitmap)
737 goto out_free;
738 memset(new.dirty_bitmap, 0, dirty_bytes);
739 }
740
741 spin_lock(&kvm->lock);
742
743 if (memory_config_version != kvm->memory_config_version) {
744 spin_unlock(&kvm->lock);
745 kvm_free_physmem_slot(&new, &old);
746 goto raced;
747 }
748
749 r = -EAGAIN;
750 if (kvm->busy)
751 goto out_unlock;
752
753 if (mem->slot >= kvm->nmemslots)
754 kvm->nmemslots = mem->slot + 1;
755
756 *memslot = new;
757 ++kvm->memory_config_version;
758
759 spin_unlock(&kvm->lock);
760
761 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
762 struct kvm_vcpu *vcpu;
763
bccf2150 764 vcpu = vcpu_load_slot(kvm, i);
6aa8b732
AK		 765 if (!vcpu)
766 continue;
ff990d59
UL		 767 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
768 do_remove_write_access(vcpu, mem->slot);
6aa8b732
AK		 769 kvm_mmu_reset_context(vcpu);
770 vcpu_put(vcpu);
771 }
772
773 kvm_free_physmem_slot(&old, &new);
774 return 0;
775
776out_unlock:
777 spin_unlock(&kvm->lock);
778out_free:
779 kvm_free_physmem_slot(&new, &old);
780out:
781 return r;
782}
783
784/*
785 * Get (and clear) the dirty memory log for a memory slot.
786 */
2c6f5df9
AK		 787static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
788 struct kvm_dirty_log *log)
6aa8b732
AK		 789{
790 struct kvm_memory_slot *memslot;
791 int r, i;
792 int n;
714b93da 793 int cleared;
6aa8b732
AK		 794 unsigned long any = 0;
795
796 spin_lock(&kvm->lock);
797
798 /*
799 * Prevent changes to guest memory configuration even while the lock
800 * is not taken.
801 */
802 ++kvm->busy;
803 spin_unlock(&kvm->lock);
804 r = -EINVAL;
805 if (log->slot >= KVM_MEMORY_SLOTS)
806 goto out;
807
808 memslot = &kvm->memslots[log->slot];
809 r = -ENOENT;
810 if (!memslot->dirty_bitmap)
811 goto out;
812
cd1a4a98 813 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
6aa8b732 814
cd1a4a98 815 for (i = 0; !any && i < n/sizeof(long); ++i)
6aa8b732
AK		 816 any = memslot->dirty_bitmap[i];
817
818 r = -EFAULT;
819 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
820 goto out;
821
6aa8b732 822 if (any) {
714b93da 823 cleared = 0;
6aa8b732 824 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
bccf2150 825 struct kvm_vcpu *vcpu;
6aa8b732 826
bccf2150 827 vcpu = vcpu_load_slot(kvm, i);
6aa8b732
AK		 828 if (!vcpu)
829 continue;
714b93da
AK		 830 if (!cleared) {
831 do_remove_write_access(vcpu, log->slot);
832 memset(memslot->dirty_bitmap, 0, n);
833 cleared = 1;
834 }
6aa8b732
AK		 835 kvm_arch_ops->tlb_flush(vcpu);
836 vcpu_put(vcpu);
837 }
838 }
839
840 r = 0;
841
842out:
843 spin_lock(&kvm->lock);
844 --kvm->busy;
845 spin_unlock(&kvm->lock);
846 return r;
847}
848
e8207547
AK		 849/*
850 * Set a new alias region. Aliases map a portion of physical memory into
851 * another portion. This is useful for memory windows, for example the PC
852 * VGA region.
853 */
854static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
855 struct kvm_memory_alias *alias)
856{
857 int r, n;
858 struct kvm_mem_alias *p;
859
860 r = -EINVAL;
861 /* General sanity checks */
862 if (alias->memory_size & (PAGE_SIZE - 1))
863 goto out;
864 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
865 goto out;
866 if (alias->slot >= KVM_ALIAS_SLOTS)
867 goto out;
868 if (alias->guest_phys_addr + alias->memory_size
869 < alias->guest_phys_addr)
870 goto out;
871 if (alias->target_phys_addr + alias->memory_size
872 < alias->target_phys_addr)
873 goto out;
874
875 spin_lock(&kvm->lock);
876
877 p = &kvm->aliases[alias->slot];
878 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
879 p->npages = alias->memory_size >> PAGE_SHIFT;
880 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
881
882 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
883 if (kvm->aliases[n - 1].npages)
884 break;
885 kvm->naliases = n;
886
887 spin_unlock(&kvm->lock);
888
889 vcpu_load(&kvm->vcpus[0]);
890 spin_lock(&kvm->lock);
891 kvm_mmu_zap_all(&kvm->vcpus[0]);
892 spin_unlock(&kvm->lock);
893 vcpu_put(&kvm->vcpus[0]);
894
895 return 0;
896
897out:
898 return r;
899}
900
901static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
902{
903 int i;
904 struct kvm_mem_alias *alias;
905
906 for (i = 0; i < kvm->naliases; ++i) {
907 alias = &kvm->aliases[i];
908 if (gfn >= alias->base_gfn
909 && gfn < alias->base_gfn + alias->npages)
910 return alias->target_gfn + gfn - alias->base_gfn;
911 }
912 return gfn;
913}
914
915static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
6aa8b732
AK		 916{
917 int i;
918
919 for (i = 0; i < kvm->nmemslots; ++i) {
920 struct kvm_memory_slot *memslot = &kvm->memslots[i];
921
922 if (gfn >= memslot->base_gfn
923 && gfn < memslot->base_gfn + memslot->npages)
924 return memslot;
925 }
8b6d44c7 926 return NULL;
6aa8b732 927}
e8207547
AK		 928
929struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
930{
931 gfn = unalias_gfn(kvm, gfn);
932 return __gfn_to_memslot(kvm, gfn);
933}
6aa8b732 934
954bbbc2
AK		 935struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
936{
937 struct kvm_memory_slot *slot;
938
e8207547
AK		 939 gfn = unalias_gfn(kvm, gfn);
940 slot = __gfn_to_memslot(kvm, gfn);
954bbbc2
AK		 941 if (!slot)
942 return NULL;
943 return slot->phys_mem[gfn - slot->base_gfn];
944}
945EXPORT_SYMBOL_GPL(gfn_to_page);
946
6aa8b732
AK		 947void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
948{
949 int i;
8b6d44c7 950 struct kvm_memory_slot *memslot = NULL;
6aa8b732
AK		 951 unsigned long rel_gfn;
952
953 for (i = 0; i < kvm->nmemslots; ++i) {
954 memslot = &kvm->memslots[i];
955
956 if (gfn >= memslot->base_gfn
957 && gfn < memslot->base_gfn + memslot->npages) {
958
959 if (!memslot || !memslot->dirty_bitmap)
960 return;
961
962 rel_gfn = gfn - memslot->base_gfn;
963
964 /* avoid RMW */
965 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
966 set_bit(rel_gfn, memslot->dirty_bitmap);
967 return;
968 }
969 }
970}
971
972static int emulator_read_std(unsigned long addr,
973 unsigned long *val,
974 unsigned int bytes,
975 struct x86_emulate_ctxt *ctxt)
976{
977 struct kvm_vcpu *vcpu = ctxt->vcpu;
978 void *data = val;
979
980 while (bytes) {
981 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
982 unsigned offset = addr & (PAGE_SIZE-1);
983 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
984 unsigned long pfn;
954bbbc2
AK		 985 struct page *page;
986 void *page_virt;
6aa8b732
AK		 987
988 if (gpa == UNMAPPED_GVA)
989 return X86EMUL_PROPAGATE_FAULT;
990 pfn = gpa >> PAGE_SHIFT;
954bbbc2
AK		 991 page = gfn_to_page(vcpu->kvm, pfn);
992 if (!page)
6aa8b732 993 return X86EMUL_UNHANDLEABLE;
954bbbc2 994 page_virt = kmap_atomic(page, KM_USER0);
6aa8b732 995
954bbbc2 996 memcpy(data, page_virt + offset, tocopy);
6aa8b732 997
954bbbc2 998 kunmap_atomic(page_virt, KM_USER0);
6aa8b732
AK		 999
1000 bytes -= tocopy;
1001 data += tocopy;
1002 addr += tocopy;
1003 }
1004
1005 return X86EMUL_CONTINUE;
1006}
1007
1008static int emulator_write_std(unsigned long addr,
1009 unsigned long val,
1010 unsigned int bytes,
1011 struct x86_emulate_ctxt *ctxt)
1012{
1013 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1014 addr, bytes);
1015 return X86EMUL_UNHANDLEABLE;
1016}
1017
1018static int emulator_read_emulated(unsigned long addr,
1019 unsigned long *val,
1020 unsigned int bytes,
1021 struct x86_emulate_ctxt *ctxt)
1022{
1023 struct kvm_vcpu *vcpu = ctxt->vcpu;
1024
1025 if (vcpu->mmio_read_completed) {
1026 memcpy(val, vcpu->mmio_data, bytes);
1027 vcpu->mmio_read_completed = 0;
1028 return X86EMUL_CONTINUE;
1029 } else if (emulator_read_std(addr, val, bytes, ctxt)
1030 == X86EMUL_CONTINUE)
1031 return X86EMUL_CONTINUE;
1032 else {
1033 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
d27d4aca 1034
6aa8b732 1035 if (gpa == UNMAPPED_GVA)
d27d4aca 1036 return X86EMUL_PROPAGATE_FAULT;
6aa8b732
AK		 1037 vcpu->mmio_needed = 1;
1038 vcpu->mmio_phys_addr = gpa;
1039 vcpu->mmio_size = bytes;
1040 vcpu->mmio_is_write = 0;
1041
1042 return X86EMUL_UNHANDLEABLE;
1043 }
1044}
1045
da4a00f0
AK		 1046static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1047 unsigned long val, int bytes)
1048{
da4a00f0
AK		 1049 struct page *page;
1050 void *virt;
1051
1052 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1053 return 0;
954bbbc2
AK		 1054 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1055 if (!page)
da4a00f0 1056 return 0;
da4a00f0 1057 kvm_mmu_pre_write(vcpu, gpa, bytes);
ab51a434 1058 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
da4a00f0
AK		 1059 virt = kmap_atomic(page, KM_USER0);
1060 memcpy(virt + offset_in_page(gpa), &val, bytes);
1061 kunmap_atomic(virt, KM_USER0);
1062 kvm_mmu_post_write(vcpu, gpa, bytes);
1063 return 1;
1064}
1065
6aa8b732
AK		 1066static int emulator_write_emulated(unsigned long addr,
1067 unsigned long val,
1068 unsigned int bytes,
1069 struct x86_emulate_ctxt *ctxt)
1070{
1071 struct kvm_vcpu *vcpu = ctxt->vcpu;
1072 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1073
c9047f53
AK		 1074 if (gpa == UNMAPPED_GVA) {
1075 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
6aa8b732 1076 return X86EMUL_PROPAGATE_FAULT;
c9047f53 1077 }
6aa8b732 1078
da4a00f0
AK		 1079 if (emulator_write_phys(vcpu, gpa, val, bytes))
1080 return X86EMUL_CONTINUE;
1081
6aa8b732
AK		 1082 vcpu->mmio_needed = 1;
1083 vcpu->mmio_phys_addr = gpa;
1084 vcpu->mmio_size = bytes;
1085 vcpu->mmio_is_write = 1;
1086 memcpy(vcpu->mmio_data, &val, bytes);
1087
1088 return X86EMUL_CONTINUE;
1089}
1090
1091static int emulator_cmpxchg_emulated(unsigned long addr,
1092 unsigned long old,
1093 unsigned long new,
1094 unsigned int bytes,
1095 struct x86_emulate_ctxt *ctxt)
1096{
1097 static int reported;
1098
1099 if (!reported) {
1100 reported = 1;
1101 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1102 }
1103 return emulator_write_emulated(addr, new, bytes, ctxt);
1104}
1105
32b35627
AK		 1106#ifdef CONFIG_X86_32
1107
1108static int emulator_cmpxchg8b_emulated(unsigned long addr,
1109 unsigned long old_lo,
1110 unsigned long old_hi,
1111 unsigned long new_lo,
1112 unsigned long new_hi,
1113 struct x86_emulate_ctxt *ctxt)
1114{
1115 static int reported;
1116 int r;
1117
1118 if (!reported) {
1119 reported = 1;
1120 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1121 }
1122 r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1123 if (r != X86EMUL_CONTINUE)
1124 return r;
1125 return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1126}
1127
1128#endif
1129
6aa8b732
AK		 1130static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1131{
1132 return kvm_arch_ops->get_segment_base(vcpu, seg);
1133}
1134
1135int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1136{
6aa8b732
AK		 1137 return X86EMUL_CONTINUE;
1138}
1139
1140int emulate_clts(struct kvm_vcpu *vcpu)
1141{
399badf3 1142 unsigned long cr0;
6aa8b732 1143
399badf3
AK		 1144 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1145 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
6aa8b732
AK		 1146 kvm_arch_ops->set_cr0(vcpu, cr0);
1147 return X86EMUL_CONTINUE;
1148}
1149
1150int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1151{
1152 struct kvm_vcpu *vcpu = ctxt->vcpu;
1153
1154 switch (dr) {
1155 case 0 ... 3:
1156 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1157 return X86EMUL_CONTINUE;
1158 default:
1159 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1160 __FUNCTION__, dr);
1161 return X86EMUL_UNHANDLEABLE;
1162 }
1163}
1164
1165int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1166{
1167 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1168 int exception;
1169
1170 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1171 if (exception) {
1172 /* FIXME: better handling */
1173 return X86EMUL_UNHANDLEABLE;
1174 }
1175 return X86EMUL_CONTINUE;
1176}
1177
1178static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1179{
1180 static int reported;
1181 u8 opcodes[4];
1182 unsigned long rip = ctxt->vcpu->rip;
1183 unsigned long rip_linear;
1184
1185 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1186
1187 if (reported)
1188 return;
1189
1190 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1191
1192 printk(KERN_ERR "emulation failed but !mmio_needed?"
1193 " rip %lx %02x %02x %02x %02x\n",
1194 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1195 reported = 1;
1196}
1197
1198struct x86_emulate_ops emulate_ops = {
1199 .read_std = emulator_read_std,
1200 .write_std = emulator_write_std,
1201 .read_emulated = emulator_read_emulated,
1202 .write_emulated = emulator_write_emulated,
1203 .cmpxchg_emulated = emulator_cmpxchg_emulated,
32b35627
AK		 1204#ifdef CONFIG_X86_32
1205 .cmpxchg8b_emulated = emulator_cmpxchg8b_emulated,
1206#endif
6aa8b732
AK		 1207};
1208
1209int emulate_instruction(struct kvm_vcpu *vcpu,
1210 struct kvm_run *run,
1211 unsigned long cr2,
1212 u16 error_code)
1213{
1214 struct x86_emulate_ctxt emulate_ctxt;
1215 int r;
1216 int cs_db, cs_l;
1217
1218 kvm_arch_ops->cache_regs(vcpu);
1219
1220 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1221
1222 emulate_ctxt.vcpu = vcpu;
1223 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1224 emulate_ctxt.cr2 = cr2;
1225 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1226 ? X86EMUL_MODE_REAL : cs_l
1227 ? X86EMUL_MODE_PROT64 : cs_db
1228 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1229
1230 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1231 emulate_ctxt.cs_base = 0;
1232 emulate_ctxt.ds_base = 0;
1233 emulate_ctxt.es_base = 0;
1234 emulate_ctxt.ss_base = 0;
1235 } else {
1236 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1237 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1238 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1239 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1240 }
1241
1242 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1243 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1244
1245 vcpu->mmio_is_write = 0;
1246 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1247
1248 if ((r || vcpu->mmio_is_write) && run) {
1249 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1250 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1251 run->mmio.len = vcpu->mmio_size;
1252 run->mmio.is_write = vcpu->mmio_is_write;
1253 }
1254
1255 if (r) {
a436036b
AK		 1256 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1257 return EMULATE_DONE;
6aa8b732
AK		 1258 if (!vcpu->mmio_needed) {
1259 report_emulation_failure(&emulate_ctxt);
1260 return EMULATE_FAIL;
1261 }
1262 return EMULATE_DO_MMIO;
1263 }
1264
1265 kvm_arch_ops->decache_regs(vcpu);
1266 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1267
1268 if (vcpu->mmio_is_write)
1269 return EMULATE_DO_MMIO;
1270
1271 return EMULATE_DONE;
1272}
1273EXPORT_SYMBOL_GPL(emulate_instruction);
1274
270fd9b9
AK		 1275int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1276{
1277 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1278
9b22bf57 1279 kvm_arch_ops->cache_regs(vcpu);
270fd9b9
AK		 1280 ret = -KVM_EINVAL;
1281#ifdef CONFIG_X86_64
1282 if (is_long_mode(vcpu)) {
1283 nr = vcpu->regs[VCPU_REGS_RAX];
1284 a0 = vcpu->regs[VCPU_REGS_RDI];
1285 a1 = vcpu->regs[VCPU_REGS_RSI];
1286 a2 = vcpu->regs[VCPU_REGS_RDX];
1287 a3 = vcpu->regs[VCPU_REGS_RCX];
1288 a4 = vcpu->regs[VCPU_REGS_R8];
1289 a5 = vcpu->regs[VCPU_REGS_R9];
1290 } else
1291#endif
1292 {
1293 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1294 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1295 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1296 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1297 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1298 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1299 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1300 }
1301 switch (nr) {
1302 default:
b4e63f56
AK		 1303 run->hypercall.args[0] = a0;
1304 run->hypercall.args[1] = a1;
1305 run->hypercall.args[2] = a2;
1306 run->hypercall.args[3] = a3;
1307 run->hypercall.args[4] = a4;
1308 run->hypercall.args[5] = a5;
1309 run->hypercall.ret = ret;
1310 run->hypercall.longmode = is_long_mode(vcpu);
1311 kvm_arch_ops->decache_regs(vcpu);
1312 return 0;
270fd9b9
AK		 1313 }
1314 vcpu->regs[VCPU_REGS_RAX] = ret;
9b22bf57 1315 kvm_arch_ops->decache_regs(vcpu);
270fd9b9
AK		 1316 return 1;
1317}
1318EXPORT_SYMBOL_GPL(kvm_hypercall);
1319
6aa8b732
AK		 1320static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1321{
1322 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1323}
1324
1325void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1326{
1327 struct descriptor_table dt = { limit, base };
1328
1329 kvm_arch_ops->set_gdt(vcpu, &dt);
1330}
1331
1332void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1333{
1334 struct descriptor_table dt = { limit, base };
1335
1336 kvm_arch_ops->set_idt(vcpu, &dt);
1337}
1338
1339void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1340 unsigned long *rflags)
1341{
1342 lmsw(vcpu, msw);
1343 *rflags = kvm_arch_ops->get_rflags(vcpu);
1344}
1345
1346unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1347{
399badf3 1348 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
6aa8b732
AK		 1349 switch (cr) {
1350 case 0:
1351 return vcpu->cr0;
1352 case 2:
1353 return vcpu->cr2;
1354 case 3:
1355 return vcpu->cr3;
1356 case 4:
1357 return vcpu->cr4;
1358 default:
1359 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1360 return 0;
1361 }
1362}
1363
1364void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1365 unsigned long *rflags)
1366{
1367 switch (cr) {
1368 case 0:
1369 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1370 *rflags = kvm_arch_ops->get_rflags(vcpu);
1371 break;
1372 case 2:
1373 vcpu->cr2 = val;
1374 break;
1375 case 3:
1376 set_cr3(vcpu, val);
1377 break;
1378 case 4:
1379 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1380 break;
1381 default:
1382 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1383 }
1384}
1385
102d8325
IM		 1386/*
1387 * Register the para guest with the host:
1388 */
1389static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1390{
1391 struct kvm_vcpu_para_state *para_state;
1392 hpa_t para_state_hpa, hypercall_hpa;
1393 struct page *para_state_page;
1394 unsigned char *hypercall;
1395 gpa_t hypercall_gpa;
1396
1397 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1398 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1399
1400 /*
1401 * Needs to be page aligned:
1402 */
1403 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1404 goto err_gp;
1405
1406 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1407 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1408 if (is_error_hpa(para_state_hpa))
1409 goto err_gp;
1410
ab51a434 1411 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
102d8325
IM		 1412 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1413 para_state = kmap_atomic(para_state_page, KM_USER0);
1414
1415 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1416 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1417
1418 para_state->host_version = KVM_PARA_API_VERSION;
1419 /*
1420 * We cannot support guests that try to register themselves
1421 * with a newer API version than the host supports:
1422 */
1423 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1424 para_state->ret = -KVM_EINVAL;
1425 goto err_kunmap_skip;
1426 }
1427
1428 hypercall_gpa = para_state->hypercall_gpa;
1429 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1430 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1431 if (is_error_hpa(hypercall_hpa)) {
1432 para_state->ret = -KVM_EINVAL;
1433 goto err_kunmap_skip;
1434 }
1435
1436 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1437 vcpu->para_state_page = para_state_page;
1438 vcpu->para_state_gpa = para_state_gpa;
1439 vcpu->hypercall_gpa = hypercall_gpa;
1440
ab51a434 1441 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
102d8325
IM		 1442 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1443 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1444 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1445 kunmap_atomic(hypercall, KM_USER1);
1446
1447 para_state->ret = 0;
1448err_kunmap_skip:
1449 kunmap_atomic(para_state, KM_USER0);
1450 return 0;
1451err_gp:
1452 return 1;
1453}
1454
3bab1f5d
AK		 1455int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1456{
1457 u64 data;
1458
1459 switch (msr) {
1460 case 0xc0010010: /* SYSCFG */
1461 case 0xc0010015: /* HWCR */
1462 case MSR_IA32_PLATFORM_ID:
1463 case MSR_IA32_P5_MC_ADDR:
1464 case MSR_IA32_P5_MC_TYPE:
1465 case MSR_IA32_MC0_CTL:
1466 case MSR_IA32_MCG_STATUS:
1467 case MSR_IA32_MCG_CAP:
1468 case MSR_IA32_MC0_MISC:
1469 case MSR_IA32_MC0_MISC+4:
1470 case MSR_IA32_MC0_MISC+8:
1471 case MSR_IA32_MC0_MISC+12:
1472 case MSR_IA32_MC0_MISC+16:
1473 case MSR_IA32_UCODE_REV:
a8d13ea2 1474 case MSR_IA32_PERF_STATUS:
3bab1f5d
AK		 1475 /* MTRR registers */
1476 case 0xfe:
1477 case 0x200 ... 0x2ff:
1478 data = 0;
1479 break;
a8d13ea2
AK		 1480 case 0xcd: /* fsb frequency */
1481 data = 3;
1482 break;
3bab1f5d
AK		 1483 case MSR_IA32_APICBASE:
1484 data = vcpu->apic_base;
1485 break;
6f00e68f
AK		 1486 case MSR_IA32_MISC_ENABLE:
1487 data = vcpu->ia32_misc_enable_msr;
1488 break;
3bab1f5d
AK		 1489#ifdef CONFIG_X86_64
1490 case MSR_EFER:
1491 data = vcpu->shadow_efer;
1492 break;
1493#endif
1494 default:
1495 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1496 return 1;
1497 }
1498 *pdata = data;
1499 return 0;
1500}
1501EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1502
6aa8b732
AK		 1503/*
1504 * Reads an msr value (of 'msr_index') into 'pdata'.
1505 * Returns 0 on success, non-0 otherwise.
1506 * Assumes vcpu_load() was already called.
1507 */
1508static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1509{
1510 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1511}
1512
05b3e0c2 1513#ifdef CONFIG_X86_64
6aa8b732 1514
3bab1f5d 1515static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
6aa8b732 1516{
6aa8b732
AK		 1517 if (efer & EFER_RESERVED_BITS) {
1518 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1519 efer);
1520 inject_gp(vcpu);
1521 return;
1522 }
1523
1524 if (is_paging(vcpu)
1525 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1526 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1527 inject_gp(vcpu);
1528 return;
1529 }
1530
7725f0ba
AK		 1531 kvm_arch_ops->set_efer(vcpu, efer);
1532
6aa8b732
AK		 1533 efer &= ~EFER_LMA;
1534 efer |= vcpu->shadow_efer & EFER_LMA;
1535
1536 vcpu->shadow_efer = efer;
6aa8b732 1537}
6aa8b732
AK		 1538
1539#endif
1540
3bab1f5d
AK		 1541int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1542{
1543 switch (msr) {
1544#ifdef CONFIG_X86_64
1545 case MSR_EFER:
1546 set_efer(vcpu, data);
1547 break;
1548#endif
1549 case MSR_IA32_MC0_STATUS:
1550 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1551 __FUNCTION__, data);
1552 break;
0e5bf0d0
SK		 1553 case MSR_IA32_MCG_STATUS:
1554 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1555 __FUNCTION__, data);
1556 break;
3bab1f5d
AK		 1557 case MSR_IA32_UCODE_REV:
1558 case MSR_IA32_UCODE_WRITE:
1559 case 0x200 ... 0x2ff: /* MTRRs */
1560 break;
1561 case MSR_IA32_APICBASE:
1562 vcpu->apic_base = data;
1563 break;
6f00e68f
AK		 1564 case MSR_IA32_MISC_ENABLE:
1565 vcpu->ia32_misc_enable_msr = data;
1566 break;
102d8325
IM		 1567 /*
1568 * This is the 'probe whether the host is KVM' logic:
1569 */
1570 case MSR_KVM_API_MAGIC:
1571 return vcpu_register_para(vcpu, data);
1572
3bab1f5d
AK		 1573 default:
1574 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1575 return 1;
1576 }
1577 return 0;
1578}
1579EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1580
6aa8b732
AK		 1581/*
1582 * Writes msr value into into the appropriate "register".
1583 * Returns 0 on success, non-0 otherwise.
1584 * Assumes vcpu_load() was already called.
1585 */
1586static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1587{
1588 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1589}
1590
1591void kvm_resched(struct kvm_vcpu *vcpu)
1592{
3fca0365
YD		 1593 if (!need_resched())
1594 return;
6aa8b732
AK		 1595 vcpu_put(vcpu);
1596 cond_resched();
bccf2150 1597 vcpu_load(vcpu);
6aa8b732
AK		 1598}
1599EXPORT_SYMBOL_GPL(kvm_resched);
1600
1601void load_msrs(struct vmx_msr_entry *e, int n)
1602{
1603 int i;
1604
1605 for (i = 0; i < n; ++i)
1606 wrmsrl(e[i].index, e[i].data);
1607}
1608EXPORT_SYMBOL_GPL(load_msrs);
1609
1610void save_msrs(struct vmx_msr_entry *e, int n)
1611{
1612 int i;
1613
1614 for (i = 0; i < n; ++i)
1615 rdmsrl(e[i].index, e[i].data);
1616}
1617EXPORT_SYMBOL_GPL(save_msrs);
1618
06465c5a
AK		 1619void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1620{
1621 int i;
1622 u32 function;
1623 struct kvm_cpuid_entry *e, *best;
1624
1625 kvm_arch_ops->cache_regs(vcpu);
1626 function = vcpu->regs[VCPU_REGS_RAX];
1627 vcpu->regs[VCPU_REGS_RAX] = 0;
1628 vcpu->regs[VCPU_REGS_RBX] = 0;
1629 vcpu->regs[VCPU_REGS_RCX] = 0;
1630 vcpu->regs[VCPU_REGS_RDX] = 0;
1631 best = NULL;
1632 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1633 e = &vcpu->cpuid_entries[i];
1634 if (e->function == function) {
1635 best = e;
1636 break;
1637 }
1638 /*
1639 * Both basic or both extended?
1640 */
1641 if (((e->function ^ function) & 0x80000000) == 0)
1642 if (!best || e->function > best->function)
1643 best = e;
1644 }
1645 if (best) {
1646 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1647 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1648 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1649 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1650 }
1651 kvm_arch_ops->decache_regs(vcpu);
1652 kvm_arch_ops->skip_emulated_instruction(vcpu);
1653}
1654EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1655
039576c0 1656static int pio_copy_data(struct kvm_vcpu *vcpu)
46fc1477 1657{
039576c0
AK		 1658 void *p = vcpu->pio_data;
1659 void *q;
1660 unsigned bytes;
1661 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1662
1663 kvm_arch_ops->vcpu_put(vcpu);
1664 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1665 PAGE_KERNEL);
1666 if (!q) {
1667 kvm_arch_ops->vcpu_load(vcpu);
1668 free_pio_guest_pages(vcpu);
1669 return -ENOMEM;
1670 }
1671 q += vcpu->pio.guest_page_offset;
1672 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1673 if (vcpu->pio.in)
1674 memcpy(q, p, bytes);
1675 else
1676 memcpy(p, q, bytes);
1677 q -= vcpu->pio.guest_page_offset;
1678 vunmap(q);
1679 kvm_arch_ops->vcpu_load(vcpu);
1680 free_pio_guest_pages(vcpu);
1681 return 0;
1682}
1683
1684static int complete_pio(struct kvm_vcpu *vcpu)
1685{
1686 struct kvm_pio_request *io = &vcpu->pio;
46fc1477 1687 long delta;
039576c0 1688 int r;
46fc1477
AK		 1689
1690 kvm_arch_ops->cache_regs(vcpu);
1691
1692 if (!io->string) {
039576c0
AK		 1693 if (io->in)
1694 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
46fc1477
AK		 1695 io->size);
1696 } else {
039576c0
AK		 1697 if (io->in) {
1698 r = pio_copy_data(vcpu);
1699 if (r) {
1700 kvm_arch_ops->cache_regs(vcpu);
1701 return r;
1702 }
1703 }
1704
46fc1477
AK		 1705 delta = 1;
1706 if (io->rep) {
039576c0 1707 delta *= io->cur_count;
46fc1477
AK		 1708 /*
1709 * The size of the register should really depend on
1710 * current address size.
1711 */
1712 vcpu->regs[VCPU_REGS_RCX] -= delta;
1713 }
039576c0 1714 if (io->down)
46fc1477
AK		 1715 delta = -delta;
1716 delta *= io->size;
039576c0 1717 if (io->in)
46fc1477
AK		 1718 vcpu->regs[VCPU_REGS_RDI] += delta;
1719 else
1720 vcpu->regs[VCPU_REGS_RSI] += delta;
1721 }
1722
46fc1477
AK		 1723 vcpu->run->io_completed = 0;
1724
1725 kvm_arch_ops->decache_regs(vcpu);
1726
039576c0
AK		 1727 io->count -= io->cur_count;
1728 io->cur_count = 0;
1729
1730 if (!io->count)
1731 kvm_arch_ops->skip_emulated_instruction(vcpu);
1732 return 0;
46fc1477
AK		 1733}
1734
039576c0
AK		 1735int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1736 int size, unsigned long count, int string, int down,
1737 gva_t address, int rep, unsigned port)
1738{
1739 unsigned now, in_page;
1740 int i;
1741 int nr_pages = 1;
1742 struct page *page;
1743
1744 vcpu->run->exit_reason = KVM_EXIT_IO;
1745 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1746 vcpu->run->io.size = size;
1747 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1748 vcpu->run->io.count = count;
1749 vcpu->run->io.port = port;
1750 vcpu->pio.count = count;
1751 vcpu->pio.cur_count = count;
1752 vcpu->pio.size = size;
1753 vcpu->pio.in = in;
1754 vcpu->pio.string = string;
1755 vcpu->pio.down = down;
1756 vcpu->pio.guest_page_offset = offset_in_page(address);
1757 vcpu->pio.rep = rep;
1758
1759 if (!string) {
1760 kvm_arch_ops->cache_regs(vcpu);
1761 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1762 kvm_arch_ops->decache_regs(vcpu);
1763 return 0;
1764 }
1765
1766 if (!count) {
1767 kvm_arch_ops->skip_emulated_instruction(vcpu);
1768 return 1;
1769 }
1770
1771 now = min(count, PAGE_SIZE / size);
1772
1773 if (!down)
1774 in_page = PAGE_SIZE - offset_in_page(address);
1775 else
1776 in_page = offset_in_page(address) + size;
1777 now = min(count, (unsigned long)in_page / size);
1778 if (!now) {
1779 /*
1780 * String I/O straddles page boundary. Pin two guest pages
1781 * so that we satisfy atomicity constraints. Do just one
1782 * transaction to avoid complexity.
1783 */
1784 nr_pages = 2;
1785 now = 1;
1786 }
1787 if (down) {
1788 /*
1789 * String I/O in reverse. Yuck. Kill the guest, fix later.
1790 */
1791 printk(KERN_ERR "kvm: guest string pio down\n");
1792 inject_gp(vcpu);
1793 return 1;
1794 }
1795 vcpu->run->io.count = now;
1796 vcpu->pio.cur_count = now;
1797
1798 for (i = 0; i < nr_pages; ++i) {
1799 spin_lock(&vcpu->kvm->lock);
1800 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1801 if (page)
1802 get_page(page);
1803 vcpu->pio.guest_pages[i] = page;
1804 spin_unlock(&vcpu->kvm->lock);
1805 if (!page) {
1806 inject_gp(vcpu);
1807 free_pio_guest_pages(vcpu);
1808 return 1;
1809 }
1810 }
1811
1812 if (!vcpu->pio.in)
1813 return pio_copy_data(vcpu);
1814 return 0;
1815}
1816EXPORT_SYMBOL_GPL(kvm_setup_pio);
1817
bccf2150 1818static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 1819{
6aa8b732 1820 int r;
1961d276 1821 sigset_t sigsaved;
6aa8b732 1822
bccf2150 1823 vcpu_load(vcpu);
6aa8b732 1824
1961d276
AK		 1825 if (vcpu->sigset_active)
1826 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1827
54810342
DL		 1828 /* re-sync apic's tpr */
1829 vcpu->cr8 = kvm_run->cr8;
1830
46fc1477 1831 if (kvm_run->io_completed) {
039576c0
AK		 1832 if (vcpu->pio.cur_count) {
1833 r = complete_pio(vcpu);
1834 if (r)
1835 goto out;
1836 } else {
46fc1477
AK		 1837 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1838 vcpu->mmio_read_completed = 1;
1839 }
6aa8b732
AK		 1840 }
1841
1842 vcpu->mmio_needed = 0;
1843
8eb7d334 1844 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
b4e63f56
AK		 1845 kvm_arch_ops->cache_regs(vcpu);
1846 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1847 kvm_arch_ops->decache_regs(vcpu);
1848 }
1849
6aa8b732
AK		 1850 r = kvm_arch_ops->run(vcpu, kvm_run);
1851
039576c0 1852out:
1961d276
AK		 1853 if (vcpu->sigset_active)
1854 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1855
6aa8b732
AK		 1856 vcpu_put(vcpu);
1857 return r;
1858}
1859
bccf2150
AK		 1860static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1861 struct kvm_regs *regs)
6aa8b732 1862{
bccf2150 1863 vcpu_load(vcpu);
6aa8b732
AK		 1864
1865 kvm_arch_ops->cache_regs(vcpu);
1866
1867 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1868 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1869 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1870 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1871 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1872 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1873 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1874 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
05b3e0c2 1875#ifdef CONFIG_X86_64
6aa8b732
AK		 1876 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1877 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1878 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1879 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1880 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1881 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1882 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1883 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1884#endif
1885
1886 regs->rip = vcpu->rip;
1887 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1888
1889 /*
1890 * Don't leak debug flags in case they were set for guest debugging
1891 */
1892 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1893 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1894
1895 vcpu_put(vcpu);
1896
1897 return 0;
1898}
1899
bccf2150
AK		 1900static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1901 struct kvm_regs *regs)
6aa8b732 1902{
bccf2150 1903 vcpu_load(vcpu);
6aa8b732
AK		 1904
1905 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1906 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1907 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1908 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1909 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1910 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1911 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1912 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
05b3e0c2 1913#ifdef CONFIG_X86_64
6aa8b732
AK		 1914 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1915 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1916 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1917 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1918 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1919 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1920 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1921 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1922#endif
1923
1924 vcpu->rip = regs->rip;
1925 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1926
1927 kvm_arch_ops->decache_regs(vcpu);
1928
1929 vcpu_put(vcpu);
1930
1931 return 0;
1932}
1933
1934static void get_segment(struct kvm_vcpu *vcpu,
1935 struct kvm_segment *var, int seg)
1936{
1937 return kvm_arch_ops->get_segment(vcpu, var, seg);
1938}
1939
bccf2150
AK		 1940static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1941 struct kvm_sregs *sregs)
6aa8b732 1942{
6aa8b732
AK		 1943 struct descriptor_table dt;
1944
bccf2150 1945 vcpu_load(vcpu);
6aa8b732
AK		 1946
1947 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1948 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1949 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1950 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1951 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1952 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1953
1954 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1955 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1956
1957 kvm_arch_ops->get_idt(vcpu, &dt);
1958 sregs->idt.limit = dt.limit;
1959 sregs->idt.base = dt.base;
1960 kvm_arch_ops->get_gdt(vcpu, &dt);
1961 sregs->gdt.limit = dt.limit;
1962 sregs->gdt.base = dt.base;
1963
399badf3 1964 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
6aa8b732
AK		 1965 sregs->cr0 = vcpu->cr0;
1966 sregs->cr2 = vcpu->cr2;
1967 sregs->cr3 = vcpu->cr3;
1968 sregs->cr4 = vcpu->cr4;
1969 sregs->cr8 = vcpu->cr8;
1970 sregs->efer = vcpu->shadow_efer;
1971 sregs->apic_base = vcpu->apic_base;
1972
1973 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1974 sizeof sregs->interrupt_bitmap);
1975
1976 vcpu_put(vcpu);
1977
1978 return 0;
1979}
1980
1981static void set_segment(struct kvm_vcpu *vcpu,
1982 struct kvm_segment *var, int seg)
1983{
1984 return kvm_arch_ops->set_segment(vcpu, var, seg);
1985}
1986
bccf2150
AK		 1987static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1988 struct kvm_sregs *sregs)
6aa8b732 1989{
6aa8b732
AK		 1990 int mmu_reset_needed = 0;
1991 int i;
1992 struct descriptor_table dt;
1993
bccf2150 1994 vcpu_load(vcpu);
6aa8b732 1995
6aa8b732
AK		 1996 dt.limit = sregs->idt.limit;
1997 dt.base = sregs->idt.base;
1998 kvm_arch_ops->set_idt(vcpu, &dt);
1999 dt.limit = sregs->gdt.limit;
2000 dt.base = sregs->gdt.base;
2001 kvm_arch_ops->set_gdt(vcpu, &dt);
2002
2003 vcpu->cr2 = sregs->cr2;
2004 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2005 vcpu->cr3 = sregs->cr3;
2006
2007 vcpu->cr8 = sregs->cr8;
2008
2009 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
05b3e0c2 2010#ifdef CONFIG_X86_64
6aa8b732
AK		 2011 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2012#endif
2013 vcpu->apic_base = sregs->apic_base;
2014
399badf3
AK		 2015 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
2016
6aa8b732 2017 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
f6528b03 2018 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
6aa8b732
AK		 2019
2020 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2021 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1b0973bd
AK		 2022 if (!is_long_mode(vcpu) && is_pae(vcpu))
2023 load_pdptrs(vcpu, vcpu->cr3);
6aa8b732
AK		 2024
2025 if (mmu_reset_needed)
2026 kvm_mmu_reset_context(vcpu);
2027
2028 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2029 sizeof vcpu->irq_pending);
2030 vcpu->irq_summary = 0;
2031 for (i = 0; i < NR_IRQ_WORDS; ++i)
2032 if (vcpu->irq_pending[i])
2033 __set_bit(i, &vcpu->irq_summary);
2034
024aa1c0
AK		 2035 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2036 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2037 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2038 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2039 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2040 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2041
2042 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2043 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2044
6aa8b732
AK		 2045 vcpu_put(vcpu);
2046
2047 return 0;
2048}
2049
2050/*
2051 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2052 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
bf591b24
MR		 2053 *
2054 * This list is modified at module load time to reflect the
2055 * capabilities of the host cpu.
6aa8b732
AK		 2056 */
2057static u32 msrs_to_save[] = {
2058 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2059 MSR_K6_STAR,
05b3e0c2 2060#ifdef CONFIG_X86_64
6aa8b732
AK		 2061 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2062#endif
2063 MSR_IA32_TIME_STAMP_COUNTER,
2064};
2065
bf591b24
MR		 2066static unsigned num_msrs_to_save;
2067
6f00e68f
AK		 2068static u32 emulated_msrs[] = {
2069 MSR_IA32_MISC_ENABLE,
2070};
2071
bf591b24
MR		 2072static __init void kvm_init_msr_list(void)
2073{
2074 u32 dummy[2];
2075 unsigned i, j;
2076
2077 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2078 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2079 continue;
2080 if (j < i)
2081 msrs_to_save[j] = msrs_to_save[i];
2082 j++;
2083 }
2084 num_msrs_to_save = j;
2085}
6aa8b732
AK		 2086
2087/*
2088 * Adapt set_msr() to msr_io()'s calling convention
2089 */
2090static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2091{
2092 return set_msr(vcpu, index, *data);
2093}
2094
2095/*
2096 * Read or write a bunch of msrs. All parameters are kernel addresses.
2097 *
2098 * @return number of msrs set successfully.
2099 */
bccf2150 2100static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
6aa8b732
AK		 2101 struct kvm_msr_entry *entries,
2102 int (*do_msr)(struct kvm_vcpu *vcpu,
2103 unsigned index, u64 *data))
2104{
6aa8b732
AK		 2105 int i;
2106
bccf2150 2107 vcpu_load(vcpu);
6aa8b732
AK		 2108
2109 for (i = 0; i < msrs->nmsrs; ++i)
2110 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2111 break;
2112
2113 vcpu_put(vcpu);
2114
2115 return i;
2116}
2117
2118/*
2119 * Read or write a bunch of msrs. Parameters are user addresses.
2120 *
2121 * @return number of msrs set successfully.
2122 */
bccf2150 2123static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
6aa8b732
AK		 2124 int (*do_msr)(struct kvm_vcpu *vcpu,
2125 unsigned index, u64 *data),
2126 int writeback)
2127{
2128 struct kvm_msrs msrs;
2129 struct kvm_msr_entry *entries;
2130 int r, n;
2131 unsigned size;
2132
2133 r = -EFAULT;
2134 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2135 goto out;
2136
2137 r = -E2BIG;
2138 if (msrs.nmsrs >= MAX_IO_MSRS)
2139 goto out;
2140
2141 r = -ENOMEM;
2142 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2143 entries = vmalloc(size);
2144 if (!entries)
2145 goto out;
2146
2147 r = -EFAULT;
2148 if (copy_from_user(entries, user_msrs->entries, size))
2149 goto out_free;
2150
bccf2150 2151 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
6aa8b732
AK		 2152 if (r < 0)
2153 goto out_free;
2154
2155 r = -EFAULT;
2156 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2157 goto out_free;
2158
2159 r = n;
2160
2161out_free:
2162 vfree(entries);
2163out:
2164 return r;
2165}
2166
2167/*
2168 * Translate a guest virtual address to a guest physical address.
2169 */
bccf2150
AK		 2170static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2171 struct kvm_translation *tr)
6aa8b732
AK		 2172{
2173 unsigned long vaddr = tr->linear_address;
6aa8b732
AK		 2174 gpa_t gpa;
2175
bccf2150
AK		 2176 vcpu_load(vcpu);
2177 spin_lock(&vcpu->kvm->lock);
6aa8b732
AK		 2178 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2179 tr->physical_address = gpa;
2180 tr->valid = gpa != UNMAPPED_GVA;
2181 tr->writeable = 1;
2182 tr->usermode = 0;
bccf2150 2183 spin_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 2184 vcpu_put(vcpu);
2185
2186 return 0;
2187}
2188
bccf2150
AK		 2189static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2190 struct kvm_interrupt *irq)
6aa8b732 2191{
6aa8b732
AK		 2192 if (irq->irq < 0 || irq->irq >= 256)
2193 return -EINVAL;
bccf2150 2194 vcpu_load(vcpu);
6aa8b732
AK		 2195
2196 set_bit(irq->irq, vcpu->irq_pending);
2197 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2198
2199 vcpu_put(vcpu);
2200
2201 return 0;
2202}
2203
bccf2150
AK		 2204static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2205 struct kvm_debug_guest *dbg)
6aa8b732 2206{
6aa8b732
AK		 2207 int r;
2208
bccf2150 2209 vcpu_load(vcpu);
6aa8b732
AK		 2210
2211 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2212
2213 vcpu_put(vcpu);
2214
2215 return r;
2216}
2217
9a2bb7f4
AK		 2218static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2219 unsigned long address,
2220 int *type)
2221{
2222 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2223 unsigned long pgoff;
2224 struct page *page;
2225
2226 *type = VM_FAULT_MINOR;
2227 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
039576c0
AK		 2228 if (pgoff == 0)
2229 page = virt_to_page(vcpu->run);
2230 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2231 page = virt_to_page(vcpu->pio_data);
2232 else
9a2bb7f4 2233 return NOPAGE_SIGBUS;
9a2bb7f4
AK		 2234 get_page(page);
2235 return page;
2236}
2237
2238static struct vm_operations_struct kvm_vcpu_vm_ops = {
2239 .nopage = kvm_vcpu_nopage,
2240};
2241
2242static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2243{
2244 vma->vm_ops = &kvm_vcpu_vm_ops;
2245 return 0;
2246}
2247
bccf2150
AK		 2248static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2249{
2250 struct kvm_vcpu *vcpu = filp->private_data;
2251
2252 fput(vcpu->kvm->filp);
2253 return 0;
2254}
2255
2256static struct file_operations kvm_vcpu_fops = {
2257 .release = kvm_vcpu_release,
2258 .unlocked_ioctl = kvm_vcpu_ioctl,
2259 .compat_ioctl = kvm_vcpu_ioctl,
9a2bb7f4 2260 .mmap = kvm_vcpu_mmap,
bccf2150
AK		 2261};
2262
2263/*
2264 * Allocates an inode for the vcpu.
2265 */
2266static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2267{
2268 int fd, r;
2269 struct inode *inode;
2270 struct file *file;
2271
2272 atomic_inc(&vcpu->kvm->filp->f_count);
2273 inode = kvmfs_inode(&kvm_vcpu_fops);
2274 if (IS_ERR(inode)) {
2275 r = PTR_ERR(inode);
2276 goto out1;
2277 }
2278
2279 file = kvmfs_file(inode, vcpu);
2280 if (IS_ERR(file)) {
2281 r = PTR_ERR(file);
2282 goto out2;
2283 }
2284
2285 r = get_unused_fd();
2286 if (r < 0)
2287 goto out3;
2288 fd = r;
2289 fd_install(fd, file);
2290
2291 return fd;
2292
2293out3:
2294 fput(file);
2295out2:
2296 iput(inode);
2297out1:
2298 fput(vcpu->kvm->filp);
2299 return r;
2300}
2301
c5ea7660
AK		 2302/*
2303 * Creates some virtual cpus. Good luck creating more than one.
2304 */
2305static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2306{
2307 int r;
2308 struct kvm_vcpu *vcpu;
9a2bb7f4 2309 struct page *page;
c5ea7660
AK		 2310
2311 r = -EINVAL;
2312 if (!valid_vcpu(n))
2313 goto out;
2314
2315 vcpu = &kvm->vcpus[n];
2316
2317 mutex_lock(&vcpu->mutex);
2318
2319 if (vcpu->vmcs) {
2320 mutex_unlock(&vcpu->mutex);
2321 return -EEXIST;
2322 }
2323
9a2bb7f4
AK		 2324 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2325 r = -ENOMEM;
2326 if (!page)
2327 goto out_unlock;
2328 vcpu->run = page_address(page);
2329
039576c0
AK		 2330 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2331 r = -ENOMEM;
2332 if (!page)
2333 goto out_free_run;
2334 vcpu->pio_data = page_address(page);
2335
c5ea7660
AK		 2336 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2337 FX_IMAGE_ALIGN);
2338 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
d917a6b9 2339 vcpu->cr0 = 0x10;
c5ea7660
AK		 2340
2341 r = kvm_arch_ops->vcpu_create(vcpu);
2342 if (r < 0)
2343 goto out_free_vcpus;
2344
2345 r = kvm_mmu_create(vcpu);
2346 if (r < 0)
2347 goto out_free_vcpus;
2348
2349 kvm_arch_ops->vcpu_load(vcpu);
2350 r = kvm_mmu_setup(vcpu);
2351 if (r >= 0)
2352 r = kvm_arch_ops->vcpu_setup(vcpu);
2353 vcpu_put(vcpu);
2354
2355 if (r < 0)
2356 goto out_free_vcpus;
2357
bccf2150
AK		 2358 r = create_vcpu_fd(vcpu);
2359 if (r < 0)
2360 goto out_free_vcpus;
2361
2362 return r;
c5ea7660
AK		 2363
2364out_free_vcpus:
2365 kvm_free_vcpu(vcpu);
039576c0
AK		 2366out_free_run:
2367 free_page((unsigned long)vcpu->run);
2368 vcpu->run = NULL;
9a2bb7f4 2369out_unlock:
c5ea7660
AK		 2370 mutex_unlock(&vcpu->mutex);
2371out:
2372 return r;
2373}
2374
06465c5a
AK		 2375static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2376 struct kvm_cpuid *cpuid,
2377 struct kvm_cpuid_entry __user *entries)
2378{
2379 int r;
2380
2381 r = -E2BIG;
2382 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2383 goto out;
2384 r = -EFAULT;
2385 if (copy_from_user(&vcpu->cpuid_entries, entries,
2386 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2387 goto out;
2388 vcpu->cpuid_nent = cpuid->nent;
2389 return 0;
2390
2391out:
2392 return r;
2393}
2394
1961d276
AK		 2395static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2396{
2397 if (sigset) {
2398 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2399 vcpu->sigset_active = 1;
2400 vcpu->sigset = *sigset;
2401 } else
2402 vcpu->sigset_active = 0;
2403 return 0;
2404}
2405
b8836737
AK		 2406/*
2407 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2408 * we have asm/x86/processor.h
2409 */
2410struct fxsave {
2411 u16 cwd;
2412 u16 swd;
2413 u16 twd;
2414 u16 fop;
2415 u64 rip;
2416 u64 rdp;
2417 u32 mxcsr;
2418 u32 mxcsr_mask;
2419 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2420#ifdef CONFIG_X86_64
2421 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2422#else
2423 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2424#endif
2425};
2426
2427static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2428{
2429 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2430
2431 vcpu_load(vcpu);
2432
2433 memcpy(fpu->fpr, fxsave->st_space, 128);
2434 fpu->fcw = fxsave->cwd;
2435 fpu->fsw = fxsave->swd;
2436 fpu->ftwx = fxsave->twd;
2437 fpu->last_opcode = fxsave->fop;
2438 fpu->last_ip = fxsave->rip;
2439 fpu->last_dp = fxsave->rdp;
2440 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2441
2442 vcpu_put(vcpu);
2443
2444 return 0;
2445}
2446
2447static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2448{
2449 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2450
2451 vcpu_load(vcpu);
2452
2453 memcpy(fxsave->st_space, fpu->fpr, 128);
2454 fxsave->cwd = fpu->fcw;
2455 fxsave->swd = fpu->fsw;
2456 fxsave->twd = fpu->ftwx;
2457 fxsave->fop = fpu->last_opcode;
2458 fxsave->rip = fpu->last_ip;
2459 fxsave->rdp = fpu->last_dp;
2460 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2461
2462 vcpu_put(vcpu);
2463
2464 return 0;
2465}
2466
bccf2150
AK		 2467static long kvm_vcpu_ioctl(struct file *filp,
2468 unsigned int ioctl, unsigned long arg)
6aa8b732 2469{
bccf2150 2470 struct kvm_vcpu *vcpu = filp->private_data;
2f366987 2471 void __user *argp = (void __user *)arg;
6aa8b732
AK		 2472 int r = -EINVAL;
2473
2474 switch (ioctl) {
9a2bb7f4 2475 case KVM_RUN:
f0fe5108
AK		 2476 r = -EINVAL;
2477 if (arg)
2478 goto out;
9a2bb7f4 2479 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
6aa8b732 2480 break;
6aa8b732
AK		 2481 case KVM_GET_REGS: {
2482 struct kvm_regs kvm_regs;
2483
bccf2150
AK		 2484 memset(&kvm_regs, 0, sizeof kvm_regs);
2485 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2486 if (r)
2487 goto out;
2488 r = -EFAULT;
2f366987 2489 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
6aa8b732
AK		 2490 goto out;
2491 r = 0;
2492 break;
2493 }
2494 case KVM_SET_REGS: {
2495 struct kvm_regs kvm_regs;
2496
2497 r = -EFAULT;
2f366987 2498 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
6aa8b732 2499 goto out;
bccf2150 2500 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2501 if (r)
2502 goto out;
2503 r = 0;
2504 break;
2505 }
2506 case KVM_GET_SREGS: {
2507 struct kvm_sregs kvm_sregs;
2508
bccf2150
AK		 2509 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2510 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2511 if (r)
2512 goto out;
2513 r = -EFAULT;
2f366987 2514 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
6aa8b732
AK		 2515 goto out;
2516 r = 0;
2517 break;
2518 }
2519 case KVM_SET_SREGS: {
2520 struct kvm_sregs kvm_sregs;
2521
2522 r = -EFAULT;
2f366987 2523 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
6aa8b732 2524 goto out;
bccf2150 2525 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2526 if (r)
2527 goto out;
2528 r = 0;
2529 break;
2530 }
2531 case KVM_TRANSLATE: {
2532 struct kvm_translation tr;
2533
2534 r = -EFAULT;
2f366987 2535 if (copy_from_user(&tr, argp, sizeof tr))
6aa8b732 2536 goto out;
bccf2150 2537 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
6aa8b732
AK		 2538 if (r)
2539 goto out;
2540 r = -EFAULT;
2f366987 2541 if (copy_to_user(argp, &tr, sizeof tr))
6aa8b732
AK		 2542 goto out;
2543 r = 0;
2544 break;
2545 }
2546 case KVM_INTERRUPT: {
2547 struct kvm_interrupt irq;
2548
2549 r = -EFAULT;
2f366987 2550 if (copy_from_user(&irq, argp, sizeof irq))
6aa8b732 2551 goto out;
bccf2150 2552 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
6aa8b732
AK		 2553 if (r)
2554 goto out;
2555 r = 0;
2556 break;
2557 }
2558 case KVM_DEBUG_GUEST: {
2559 struct kvm_debug_guest dbg;
2560
2561 r = -EFAULT;
2f366987 2562 if (copy_from_user(&dbg, argp, sizeof dbg))
6aa8b732 2563 goto out;
bccf2150 2564 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
6aa8b732
AK		 2565 if (r)
2566 goto out;
2567 r = 0;
2568 break;
2569 }
bccf2150
AK		 2570 case KVM_GET_MSRS:
2571 r = msr_io(vcpu, argp, get_msr, 1);
2572 break;
2573 case KVM_SET_MSRS:
2574 r = msr_io(vcpu, argp, do_set_msr, 0);
2575 break;
06465c5a
AK		 2576 case KVM_SET_CPUID: {
2577 struct kvm_cpuid __user *cpuid_arg = argp;
2578 struct kvm_cpuid cpuid;
2579
2580 r = -EFAULT;
2581 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2582 goto out;
2583 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2584 if (r)
2585 goto out;
2586 break;
2587 }
1961d276
AK		 2588 case KVM_SET_SIGNAL_MASK: {
2589 struct kvm_signal_mask __user *sigmask_arg = argp;
2590 struct kvm_signal_mask kvm_sigmask;
2591 sigset_t sigset, *p;
2592
2593 p = NULL;
2594 if (argp) {
2595 r = -EFAULT;
2596 if (copy_from_user(&kvm_sigmask, argp,
2597 sizeof kvm_sigmask))
2598 goto out;
2599 r = -EINVAL;
2600 if (kvm_sigmask.len != sizeof sigset)
2601 goto out;
2602 r = -EFAULT;
2603 if (copy_from_user(&sigset, sigmask_arg->sigset,
2604 sizeof sigset))
2605 goto out;
2606 p = &sigset;
2607 }
2608 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2609 break;
2610 }
b8836737
AK		 2611 case KVM_GET_FPU: {
2612 struct kvm_fpu fpu;
2613
2614 memset(&fpu, 0, sizeof fpu);
2615 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2616 if (r)
2617 goto out;
2618 r = -EFAULT;
2619 if (copy_to_user(argp, &fpu, sizeof fpu))
2620 goto out;
2621 r = 0;
2622 break;
2623 }
2624 case KVM_SET_FPU: {
2625 struct kvm_fpu fpu;
2626
2627 r = -EFAULT;
2628 if (copy_from_user(&fpu, argp, sizeof fpu))
2629 goto out;
2630 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2631 if (r)
2632 goto out;
2633 r = 0;
2634 break;
2635 }
bccf2150
AK		 2636 default:
2637 ;
2638 }
2639out:
2640 return r;
2641}
2642
2643static long kvm_vm_ioctl(struct file *filp,
2644 unsigned int ioctl, unsigned long arg)
2645{
2646 struct kvm *kvm = filp->private_data;
2647 void __user *argp = (void __user *)arg;
2648 int r = -EINVAL;
2649
2650 switch (ioctl) {
2651 case KVM_CREATE_VCPU:
2652 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2653 if (r < 0)
2654 goto out;
2655 break;
6aa8b732
AK		 2656 case KVM_SET_MEMORY_REGION: {
2657 struct kvm_memory_region kvm_mem;
2658
2659 r = -EFAULT;
2f366987 2660 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
6aa8b732 2661 goto out;
2c6f5df9 2662 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
6aa8b732
AK		 2663 if (r)
2664 goto out;
2665 break;
2666 }
2667 case KVM_GET_DIRTY_LOG: {
2668 struct kvm_dirty_log log;
2669
2670 r = -EFAULT;
2f366987 2671 if (copy_from_user(&log, argp, sizeof log))
6aa8b732 2672 goto out;
2c6f5df9 2673 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
6aa8b732
AK		 2674 if (r)
2675 goto out;
2676 break;
2677 }
e8207547
AK		 2678 case KVM_SET_MEMORY_ALIAS: {
2679 struct kvm_memory_alias alias;
2680
2681 r = -EFAULT;
2682 if (copy_from_user(&alias, argp, sizeof alias))
2683 goto out;
2684 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2685 if (r)
2686 goto out;
2687 break;
2688 }
f17abe9a
AK		 2689 default:
2690 ;
2691 }
2692out:
2693 return r;
2694}
2695
2696static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2697 unsigned long address,
2698 int *type)
2699{
2700 struct kvm *kvm = vma->vm_file->private_data;
2701 unsigned long pgoff;
f17abe9a
AK		 2702 struct page *page;
2703
2704 *type = VM_FAULT_MINOR;
2705 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
954bbbc2 2706 page = gfn_to_page(kvm, pgoff);
f17abe9a
AK		 2707 if (!page)
2708 return NOPAGE_SIGBUS;
2709 get_page(page);
2710 return page;
2711}
2712
2713static struct vm_operations_struct kvm_vm_vm_ops = {
2714 .nopage = kvm_vm_nopage,
2715};
2716
2717static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2718{
2719 vma->vm_ops = &kvm_vm_vm_ops;
2720 return 0;
2721}
2722
2723static struct file_operations kvm_vm_fops = {
2724 .release = kvm_vm_release,
2725 .unlocked_ioctl = kvm_vm_ioctl,
2726 .compat_ioctl = kvm_vm_ioctl,
2727 .mmap = kvm_vm_mmap,
2728};
2729
2730static int kvm_dev_ioctl_create_vm(void)
2731{
2732 int fd, r;
2733 struct inode *inode;
2734 struct file *file;
2735 struct kvm *kvm;
2736
2737 inode = kvmfs_inode(&kvm_vm_fops);
2738 if (IS_ERR(inode)) {
2739 r = PTR_ERR(inode);
2740 goto out1;
2741 }
2742
2743 kvm = kvm_create_vm();
2744 if (IS_ERR(kvm)) {
2745 r = PTR_ERR(kvm);
2746 goto out2;
2747 }
2748
2749 file = kvmfs_file(inode, kvm);
2750 if (IS_ERR(file)) {
2751 r = PTR_ERR(file);
2752 goto out3;
2753 }
bccf2150 2754 kvm->filp = file;
f17abe9a
AK		 2755
2756 r = get_unused_fd();
2757 if (r < 0)
2758 goto out4;
2759 fd = r;
2760 fd_install(fd, file);
2761
2762 return fd;
2763
2764out4:
2765 fput(file);
2766out3:
2767 kvm_destroy_vm(kvm);
2768out2:
2769 iput(inode);
2770out1:
2771 return r;
2772}
2773
2774static long kvm_dev_ioctl(struct file *filp,
2775 unsigned int ioctl, unsigned long arg)
2776{
2777 void __user *argp = (void __user *)arg;
07c45a36 2778 long r = -EINVAL;
f17abe9a
AK		 2779
2780 switch (ioctl) {
2781 case KVM_GET_API_VERSION:
f0fe5108
AK		 2782 r = -EINVAL;
2783 if (arg)
2784 goto out;
f17abe9a
AK		 2785 r = KVM_API_VERSION;
2786 break;
2787 case KVM_CREATE_VM:
f0fe5108
AK		 2788 r = -EINVAL;
2789 if (arg)
2790 goto out;
f17abe9a
AK		 2791 r = kvm_dev_ioctl_create_vm();
2792 break;
6aa8b732 2793 case KVM_GET_MSR_INDEX_LIST: {
2f366987 2794 struct kvm_msr_list __user *user_msr_list = argp;
6aa8b732
AK		 2795 struct kvm_msr_list msr_list;
2796 unsigned n;
2797
2798 r = -EFAULT;
2799 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2800 goto out;
2801 n = msr_list.nmsrs;
6f00e68f 2802 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
6aa8b732
AK		 2803 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2804 goto out;
2805 r = -E2BIG;
bf591b24 2806 if (n < num_msrs_to_save)
6aa8b732
AK		 2807 goto out;
2808 r = -EFAULT;
2809 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
bf591b24 2810 num_msrs_to_save * sizeof(u32)))
6aa8b732 2811 goto out;
6f00e68f
AK		 2812 if (copy_to_user(user_msr_list->indices
2813 + num_msrs_to_save * sizeof(u32),
2814 &emulated_msrs,
2815 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2816 goto out;
6aa8b732 2817 r = 0;
cc1d8955 2818 break;
6aa8b732 2819 }
5d308f45
AK		 2820 case KVM_CHECK_EXTENSION:
2821 /*
2822 * No extensions defined at present.
2823 */
2824 r = 0;
2825 break;
07c45a36
AK		 2826 case KVM_GET_VCPU_MMAP_SIZE:
2827 r = -EINVAL;
2828 if (arg)
2829 goto out;
039576c0 2830 r = 2 * PAGE_SIZE;
07c45a36 2831 break;
6aa8b732
AK		 2832 default:
2833 ;
2834 }
2835out:
2836 return r;
2837}
2838
6aa8b732
AK		 2839static struct file_operations kvm_chardev_ops = {
2840 .open = kvm_dev_open,
2841 .release = kvm_dev_release,
2842 .unlocked_ioctl = kvm_dev_ioctl,
2843 .compat_ioctl = kvm_dev_ioctl,
6aa8b732
AK		 2844};
2845
2846static struct miscdevice kvm_dev = {
bbe4432e 2847 KVM_MINOR,
6aa8b732
AK		 2848 "kvm",
2849 &kvm_chardev_ops,
2850};
2851
2852static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2853 void *v)
2854{
2855 if (val == SYS_RESTART) {
2856 /*
2857 * Some (well, at least mine) BIOSes hang on reboot if
2858 * in vmx root mode.
2859 */
2860 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
8b6d44c7 2861 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
6aa8b732
AK		 2862 }
2863 return NOTIFY_OK;
2864}
2865
2866static struct notifier_block kvm_reboot_notifier = {
2867 .notifier_call = kvm_reboot,
2868 .priority = 0,
2869};
2870
774c47f1
AK		 2871/*
2872 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2873 * cached on it.
2874 */
2875static void decache_vcpus_on_cpu(int cpu)
2876{
2877 struct kvm *vm;
2878 struct kvm_vcpu *vcpu;
2879 int i;
2880
2881 spin_lock(&kvm_lock);
2882 list_for_each_entry(vm, &vm_list, vm_list)
2883 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2884 vcpu = &vm->vcpus[i];
2885 /*
2886 * If the vcpu is locked, then it is running on some
2887 * other cpu and therefore it is not cached on the
2888 * cpu in question.
2889 *
2890 * If it's not locked, check the last cpu it executed
2891 * on.
2892 */
2893 if (mutex_trylock(&vcpu->mutex)) {
2894 if (vcpu->cpu == cpu) {
2895 kvm_arch_ops->vcpu_decache(vcpu);
2896 vcpu->cpu = -1;
2897 }
2898 mutex_unlock(&vcpu->mutex);
2899 }
2900 }
2901 spin_unlock(&kvm_lock);
2902}
2903
2904static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2905 void *v)
2906{
2907 int cpu = (long)v;
2908
2909 switch (val) {
43934a38 2910 case CPU_DOWN_PREPARE:
774c47f1 2911 case CPU_UP_CANCELED:
43934a38
JK		 2912 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2913 cpu);
774c47f1
AK		 2914 decache_vcpus_on_cpu(cpu);
2915 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2916 NULL, 0, 1);
2917 break;
43934a38
JK		 2918 case CPU_ONLINE:
2919 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2920 cpu);
774c47f1
AK		 2921 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2922 NULL, 0, 1);
2923 break;
2924 }
2925 return NOTIFY_OK;
2926}
2927
2928static struct notifier_block kvm_cpu_notifier = {
2929 .notifier_call = kvm_cpu_hotplug,
2930 .priority = 20, /* must be > scheduler priority */
2931};
2932
6aa8b732
AK		 2933static __init void kvm_init_debug(void)
2934{
2935 struct kvm_stats_debugfs_item *p;
2936
8b6d44c7 2937 debugfs_dir = debugfs_create_dir("kvm", NULL);
6aa8b732
AK		 2938 for (p = debugfs_entries; p->name; ++p)
2939 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2940 p->data);
2941}
2942
2943static void kvm_exit_debug(void)
2944{
2945 struct kvm_stats_debugfs_item *p;
2946
2947 for (p = debugfs_entries; p->name; ++p)
2948 debugfs_remove(p->dentry);
2949 debugfs_remove(debugfs_dir);
2950}
2951
59ae6c6b
AK		 2952static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2953{
2954 decache_vcpus_on_cpu(raw_smp_processor_id());
19d1408d 2955 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
59ae6c6b
AK		 2956 return 0;
2957}
2958
2959static int kvm_resume(struct sys_device *dev)
2960{
19d1408d 2961 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
59ae6c6b
AK		 2962 return 0;
2963}
2964
2965static struct sysdev_class kvm_sysdev_class = {
2966 set_kset_name("kvm"),
2967 .suspend = kvm_suspend,
2968 .resume = kvm_resume,
2969};
2970
2971static struct sys_device kvm_sysdev = {
2972 .id = 0,
2973 .cls = &kvm_sysdev_class,
2974};
2975
6aa8b732
AK		 2976hpa_t bad_page_address;
2977
37e29d90
AK		 2978static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2979 const char *dev_name, void *data, struct vfsmount *mnt)
2980{
e9cdb1e3 2981 return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
37e29d90
AK		 2982}
2983
2984static struct file_system_type kvm_fs_type = {
2985 .name = "kvmfs",
2986 .get_sb = kvmfs_get_sb,
2987 .kill_sb = kill_anon_super,
2988};
2989
6aa8b732
AK		 2990int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2991{
2992 int r;
2993
09db28b8
YI		 2994 if (kvm_arch_ops) {
2995 printk(KERN_ERR "kvm: already loaded the other module\n");
2996 return -EEXIST;
2997 }
2998
e097f35c 2999 if (!ops->cpu_has_kvm_support()) {
6aa8b732
AK		 3000 printk(KERN_ERR "kvm: no hardware support\n");
3001 return -EOPNOTSUPP;
3002 }
e097f35c 3003 if (ops->disabled_by_bios()) {
6aa8b732
AK		 3004 printk(KERN_ERR "kvm: disabled by bios\n");
3005 return -EOPNOTSUPP;
3006 }
3007
e097f35c
YI		 3008 kvm_arch_ops = ops;
3009
6aa8b732
AK		 3010 r = kvm_arch_ops->hardware_setup();
3011 if (r < 0)
ca45aaae 3012 goto out;
6aa8b732 3013
8b6d44c7 3014 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
774c47f1
AK		 3015 r = register_cpu_notifier(&kvm_cpu_notifier);
3016 if (r)
3017 goto out_free_1;
6aa8b732
AK		 3018 register_reboot_notifier(&kvm_reboot_notifier);
3019
59ae6c6b
AK		 3020 r = sysdev_class_register(&kvm_sysdev_class);
3021 if (r)
3022 goto out_free_2;
3023
3024 r = sysdev_register(&kvm_sysdev);
3025 if (r)
3026 goto out_free_3;
3027
6aa8b732
AK		 3028 kvm_chardev_ops.owner = module;
3029
3030 r = misc_register(&kvm_dev);
3031 if (r) {
3032 printk (KERN_ERR "kvm: misc device register failed\n");
3033 goto out_free;
3034 }
3035
3036 return r;
3037
3038out_free:
59ae6c6b
AK		 3039 sysdev_unregister(&kvm_sysdev);
3040out_free_3:
3041 sysdev_class_unregister(&kvm_sysdev_class);
3042out_free_2:
6aa8b732 3043 unregister_reboot_notifier(&kvm_reboot_notifier);
774c47f1
AK		 3044 unregister_cpu_notifier(&kvm_cpu_notifier);
3045out_free_1:
8b6d44c7 3046 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
6aa8b732 3047 kvm_arch_ops->hardware_unsetup();
ca45aaae
AK		 3048out:
3049 kvm_arch_ops = NULL;
6aa8b732
AK		 3050 return r;
3051}
3052
3053void kvm_exit_arch(void)
3054{
3055 misc_deregister(&kvm_dev);
59ae6c6b
AK		 3056 sysdev_unregister(&kvm_sysdev);
3057 sysdev_class_unregister(&kvm_sysdev_class);
6aa8b732 3058 unregister_reboot_notifier(&kvm_reboot_notifier);
59ae6c6b 3059 unregister_cpu_notifier(&kvm_cpu_notifier);
8b6d44c7 3060 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
6aa8b732 3061 kvm_arch_ops->hardware_unsetup();
09db28b8 3062 kvm_arch_ops = NULL;
6aa8b732
AK		 3063}
3064
3065static __init int kvm_init(void)
3066{
3067 static struct page *bad_page;
37e29d90
AK		 3068 int r;
3069
b5a33a75
AK		 3070 r = kvm_mmu_module_init();
3071 if (r)
3072 goto out4;
3073
37e29d90
AK		 3074 r = register_filesystem(&kvm_fs_type);
3075 if (r)
3076 goto out3;
6aa8b732 3077
37e29d90
AK		 3078 kvmfs_mnt = kern_mount(&kvm_fs_type);
3079 r = PTR_ERR(kvmfs_mnt);
3080 if (IS_ERR(kvmfs_mnt))
3081 goto out2;
6aa8b732
AK		 3082 kvm_init_debug();
3083
bf591b24
MR		 3084 kvm_init_msr_list();
3085
6aa8b732
AK		 3086 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3087 r = -ENOMEM;
3088 goto out;
3089 }
3090
3091 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3092 memset(__va(bad_page_address), 0, PAGE_SIZE);
3093
58e690e6 3094 return 0;
6aa8b732
AK		 3095
3096out:
3097 kvm_exit_debug();
37e29d90
AK		 3098 mntput(kvmfs_mnt);
3099out2:
3100 unregister_filesystem(&kvm_fs_type);
3101out3:
b5a33a75
AK		 3102 kvm_mmu_module_exit();
3103out4:
6aa8b732
AK		 3104 return r;
3105}
3106
3107static __exit void kvm_exit(void)
3108{
3109 kvm_exit_debug();
3110 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
37e29d90
AK		 3111 mntput(kvmfs_mnt);
3112 unregister_filesystem(&kvm_fs_type);
b5a33a75 3113 kvm_mmu_module_exit();
6aa8b732
AK		 3114}
3115
3116module_init(kvm_init)
3117module_exit(kvm_exit)
3118
3119EXPORT_SYMBOL_GPL(kvm_init_arch);
3120EXPORT_SYMBOL_GPL(kvm_exit_arch);