projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Use a shared page for kernel/user communication when runing a vcpu
[deliverable/linux.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
349static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
350{
bccf2150 351 if (!vcpu->vmcs)
1e8ba6fb
IM		 352 return;
353
bccf2150 354 vcpu_load(vcpu);
6aa8b732 355 kvm_mmu_destroy(vcpu);
08438475 356 vcpu_put(vcpu);
9ede74e0 357 kvm_arch_ops->vcpu_free(vcpu);
9a2bb7f4
AK		 358 free_page((unsigned long)vcpu->run);
359 vcpu->run = NULL;
6aa8b732
AK		 360}
361
362static void kvm_free_vcpus(struct kvm *kvm)
363{
364 unsigned int i;
365
366 for (i = 0; i < KVM_MAX_VCPUS; ++i)
367 kvm_free_vcpu(&kvm->vcpus[i]);
368}
369
370static int kvm_dev_release(struct inode *inode, struct file *filp)
371{
f17abe9a
AK		 372 return 0;
373}
6aa8b732 374
f17abe9a
AK		 375static void kvm_destroy_vm(struct kvm *kvm)
376{
133de902
AK		 377 spin_lock(&kvm_lock);
378 list_del(&kvm->vm_list);
379 spin_unlock(&kvm_lock);
6aa8b732
AK		 380 kvm_free_vcpus(kvm);
381 kvm_free_physmem(kvm);
382 kfree(kvm);
f17abe9a
AK		 383}
384
385static int kvm_vm_release(struct inode *inode, struct file *filp)
386{
387 struct kvm *kvm = filp->private_data;
388
389 kvm_destroy_vm(kvm);
6aa8b732
AK		 390 return 0;
391}
392
393static void inject_gp(struct kvm_vcpu *vcpu)
394{
395 kvm_arch_ops->inject_gp(vcpu, 0);
396}
397
1342d353
AK		 398/*
399 * Load the pae pdptrs. Return true is they are all valid.
400 */
401static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
6aa8b732
AK		 402{
403 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
1342d353 404 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
6aa8b732
AK		 405 int i;
406 u64 pdpte;
407 u64 *pdpt;
1342d353 408 int ret;
6aa8b732
AK		 409 struct kvm_memory_slot *memslot;
410
411 spin_lock(&vcpu->kvm->lock);
412 memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
413 /* FIXME: !memslot - emulate? 0xff? */
414 pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
415
1342d353 416 ret = 1;
6aa8b732
AK		 417 for (i = 0; i < 4; ++i) {
418 pdpte = pdpt[offset + i];
1342d353
AK		 419 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
420 ret = 0;
421 goto out;
422 }
6aa8b732
AK		 423 }
424
1342d353
AK		 425 for (i = 0; i < 4; ++i)
426 vcpu->pdptrs[i] = pdpt[offset + i];
427
428out:
6aa8b732
AK		 429 kunmap_atomic(pdpt, KM_USER0);
430 spin_unlock(&vcpu->kvm->lock);
431
1342d353 432 return ret;
6aa8b732
AK		 433}
434
435void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
436{
437 if (cr0 & CR0_RESEVED_BITS) {
438 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
439 cr0, vcpu->cr0);
440 inject_gp(vcpu);
441 return;
442 }
443
444 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
445 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
446 inject_gp(vcpu);
447 return;
448 }
449
450 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
451 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
452 "and a clear PE flag\n");
453 inject_gp(vcpu);
454 return;
455 }
456
457 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
05b3e0c2 458#ifdef CONFIG_X86_64
6aa8b732
AK		 459 if ((vcpu->shadow_efer & EFER_LME)) {
460 int cs_db, cs_l;
461
462 if (!is_pae(vcpu)) {
463 printk(KERN_DEBUG "set_cr0: #GP, start paging "
464 "in long mode while PAE is disabled\n");
465 inject_gp(vcpu);
466 return;
467 }
468 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
469 if (cs_l) {
470 printk(KERN_DEBUG "set_cr0: #GP, start paging "
471 "in long mode while CS.L == 1\n");
472 inject_gp(vcpu);
473 return;
474
475 }
476 } else
477#endif
1342d353 478 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 479 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
480 "reserved bits\n");
481 inject_gp(vcpu);
482 return;
483 }
484
485 }
486
487 kvm_arch_ops->set_cr0(vcpu, cr0);
488 vcpu->cr0 = cr0;
489
490 spin_lock(&vcpu->kvm->lock);
491 kvm_mmu_reset_context(vcpu);
492 spin_unlock(&vcpu->kvm->lock);
493 return;
494}
495EXPORT_SYMBOL_GPL(set_cr0);
496
497void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
498{
399badf3 499 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
6aa8b732
AK		 500 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
501}
502EXPORT_SYMBOL_GPL(lmsw);
503
504void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
505{
506 if (cr4 & CR4_RESEVED_BITS) {
507 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
508 inject_gp(vcpu);
509 return;
510 }
511
a9058ecd 512 if (is_long_mode(vcpu)) {
6aa8b732
AK		 513 if (!(cr4 & CR4_PAE_MASK)) {
514 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
515 "in long mode\n");
516 inject_gp(vcpu);
517 return;
518 }
519 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
1342d353 520 && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 521 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
522 inject_gp(vcpu);
523 }
524
525 if (cr4 & CR4_VMXE_MASK) {
526 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
527 inject_gp(vcpu);
528 return;
529 }
530 kvm_arch_ops->set_cr4(vcpu, cr4);
531 spin_lock(&vcpu->kvm->lock);
532 kvm_mmu_reset_context(vcpu);
533 spin_unlock(&vcpu->kvm->lock);
534}
535EXPORT_SYMBOL_GPL(set_cr4);
536
537void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
538{
a9058ecd 539 if (is_long_mode(vcpu)) {
d27d4aca 540 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
6aa8b732
AK		 541 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
542 inject_gp(vcpu);
543 return;
544 }
545 } else {
546 if (cr3 & CR3_RESEVED_BITS) {
547 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
548 inject_gp(vcpu);
549 return;
550 }
551 if (is_paging(vcpu) && is_pae(vcpu) &&
1342d353 552 !load_pdptrs(vcpu, cr3)) {
6aa8b732
AK		 553 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
554 "reserved bits\n");
555 inject_gp(vcpu);
556 return;
557 }
558 }
559
560 vcpu->cr3 = cr3;
561 spin_lock(&vcpu->kvm->lock);
d21225ee
IM		 562 /*
563 * Does the new cr3 value map to physical memory? (Note, we
564 * catch an invalid cr3 even in real-mode, because it would
565 * cause trouble later on when we turn on paging anyway.)
566 *
567 * A real CPU would silently accept an invalid cr3 and would
568 * attempt to use it - with largely undefined (and often hard
569 * to debug) behavior on the guest side.
570 */
571 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
572 inject_gp(vcpu);
573 else
574 vcpu->mmu.new_cr3(vcpu);
6aa8b732
AK		 575 spin_unlock(&vcpu->kvm->lock);
576}
577EXPORT_SYMBOL_GPL(set_cr3);
578
579void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
580{
581 if ( cr8 & CR8_RESEVED_BITS) {
582 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
583 inject_gp(vcpu);
584 return;
585 }
586 vcpu->cr8 = cr8;
587}
588EXPORT_SYMBOL_GPL(set_cr8);
589
590void fx_init(struct kvm_vcpu *vcpu)
591{
592 struct __attribute__ ((__packed__)) fx_image_s {
593 u16 control; //fcw
594 u16 status; //fsw
595 u16 tag; // ftw
596 u16 opcode; //fop
597 u64 ip; // fpu ip
598 u64 operand;// fpu dp
599 u32 mxcsr;
600 u32 mxcsr_mask;
601
602 } *fx_image;
603
604 fx_save(vcpu->host_fx_image);
605 fpu_init();
606 fx_save(vcpu->guest_fx_image);
607 fx_restore(vcpu->host_fx_image);
608
609 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
610 fx_image->mxcsr = 0x1f80;
611 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
612 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
613}
614EXPORT_SYMBOL_GPL(fx_init);
615
02b27c1f
UL		 616static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
617{
618 spin_lock(&vcpu->kvm->lock);
619 kvm_mmu_slot_remove_write_access(vcpu, slot);
620 spin_unlock(&vcpu->kvm->lock);
621}
622
6aa8b732
AK		 623/*
624 * Allocate some memory and give it an address in the guest physical address
625 * space.
626 *
627 * Discontiguous memory is allowed, mostly for framebuffers.
628 */
2c6f5df9
AK		 629static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
630 struct kvm_memory_region *mem)
6aa8b732
AK		 631{
632 int r;
633 gfn_t base_gfn;
634 unsigned long npages;
635 unsigned long i;
636 struct kvm_memory_slot *memslot;
637 struct kvm_memory_slot old, new;
638 int memory_config_version;
639
640 r = -EINVAL;
641 /* General sanity checks */
642 if (mem->memory_size & (PAGE_SIZE - 1))
643 goto out;
644 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
645 goto out;
646 if (mem->slot >= KVM_MEMORY_SLOTS)
647 goto out;
648 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
649 goto out;
650
651 memslot = &kvm->memslots[mem->slot];
652 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
653 npages = mem->memory_size >> PAGE_SHIFT;
654
655 if (!npages)
656 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
657
658raced:
659 spin_lock(&kvm->lock);
660
661 memory_config_version = kvm->memory_config_version;
662 new = old = *memslot;
663
664 new.base_gfn = base_gfn;
665 new.npages = npages;
666 new.flags = mem->flags;
667
668 /* Disallow changing a memory slot's size. */
669 r = -EINVAL;
670 if (npages && old.npages && npages != old.npages)
671 goto out_unlock;
672
673 /* Check for overlaps */
674 r = -EEXIST;
675 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
676 struct kvm_memory_slot *s = &kvm->memslots[i];
677
678 if (s == memslot)
679 continue;
680 if (!((base_gfn + npages <= s->base_gfn) ||
681 (base_gfn >= s->base_gfn + s->npages)))
682 goto out_unlock;
683 }
684 /*
685 * Do memory allocations outside lock. memory_config_version will
686 * detect any races.
687 */
688 spin_unlock(&kvm->lock);
689
690 /* Deallocate if slot is being removed */
691 if (!npages)
8b6d44c7 692 new.phys_mem = NULL;
6aa8b732
AK		 693
694 /* Free page dirty bitmap if unneeded */
695 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
8b6d44c7 696 new.dirty_bitmap = NULL;
6aa8b732
AK		 697
698 r = -ENOMEM;
699
700 /* Allocate if a slot is being created */
701 if (npages && !new.phys_mem) {
702 new.phys_mem = vmalloc(npages * sizeof(struct page *));
703
704 if (!new.phys_mem)
705 goto out_free;
706
707 memset(new.phys_mem, 0, npages * sizeof(struct page *));
708 for (i = 0; i < npages; ++i) {
709 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
710 | __GFP_ZERO);
711 if (!new.phys_mem[i])
712 goto out_free;
5972e953 713 set_page_private(new.phys_mem[i],0);
6aa8b732
AK		 714 }
715 }
716
717 /* Allocate page dirty bitmap if needed */
718 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
719 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
720
721 new.dirty_bitmap = vmalloc(dirty_bytes);
722 if (!new.dirty_bitmap)
723 goto out_free;
724 memset(new.dirty_bitmap, 0, dirty_bytes);
725 }
726
727 spin_lock(&kvm->lock);
728
729 if (memory_config_version != kvm->memory_config_version) {
730 spin_unlock(&kvm->lock);
731 kvm_free_physmem_slot(&new, &old);
732 goto raced;
733 }
734
735 r = -EAGAIN;
736 if (kvm->busy)
737 goto out_unlock;
738
739 if (mem->slot >= kvm->nmemslots)
740 kvm->nmemslots = mem->slot + 1;
741
742 *memslot = new;
743 ++kvm->memory_config_version;
744
745 spin_unlock(&kvm->lock);
746
747 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
748 struct kvm_vcpu *vcpu;
749
bccf2150 750 vcpu = vcpu_load_slot(kvm, i);
6aa8b732
AK		 751 if (!vcpu)
752 continue;
ff990d59
UL		 753 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
754 do_remove_write_access(vcpu, mem->slot);
6aa8b732
AK		 755 kvm_mmu_reset_context(vcpu);
756 vcpu_put(vcpu);
757 }
758
759 kvm_free_physmem_slot(&old, &new);
760 return 0;
761
762out_unlock:
763 spin_unlock(&kvm->lock);
764out_free:
765 kvm_free_physmem_slot(&new, &old);
766out:
767 return r;
768}
769
770/*
771 * Get (and clear) the dirty memory log for a memory slot.
772 */
2c6f5df9
AK		 773static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
774 struct kvm_dirty_log *log)
6aa8b732
AK		 775{
776 struct kvm_memory_slot *memslot;
777 int r, i;
778 int n;
714b93da 779 int cleared;
6aa8b732
AK		 780 unsigned long any = 0;
781
782 spin_lock(&kvm->lock);
783
784 /*
785 * Prevent changes to guest memory configuration even while the lock
786 * is not taken.
787 */
788 ++kvm->busy;
789 spin_unlock(&kvm->lock);
790 r = -EINVAL;
791 if (log->slot >= KVM_MEMORY_SLOTS)
792 goto out;
793
794 memslot = &kvm->memslots[log->slot];
795 r = -ENOENT;
796 if (!memslot->dirty_bitmap)
797 goto out;
798
cd1a4a98 799 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
6aa8b732 800
cd1a4a98 801 for (i = 0; !any && i < n/sizeof(long); ++i)
6aa8b732
AK		 802 any = memslot->dirty_bitmap[i];
803
804 r = -EFAULT;
805 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
806 goto out;
807
6aa8b732 808 if (any) {
714b93da 809 cleared = 0;
6aa8b732 810 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
bccf2150 811 struct kvm_vcpu *vcpu;
6aa8b732 812
bccf2150 813 vcpu = vcpu_load_slot(kvm, i);
6aa8b732
AK		 814 if (!vcpu)
815 continue;
714b93da
AK		 816 if (!cleared) {
817 do_remove_write_access(vcpu, log->slot);
818 memset(memslot->dirty_bitmap, 0, n);
819 cleared = 1;
820 }
6aa8b732
AK		 821 kvm_arch_ops->tlb_flush(vcpu);
822 vcpu_put(vcpu);
823 }
824 }
825
826 r = 0;
827
828out:
829 spin_lock(&kvm->lock);
830 --kvm->busy;
831 spin_unlock(&kvm->lock);
832 return r;
833}
834
835struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
836{
837 int i;
838
839 for (i = 0; i < kvm->nmemslots; ++i) {
840 struct kvm_memory_slot *memslot = &kvm->memslots[i];
841
842 if (gfn >= memslot->base_gfn
843 && gfn < memslot->base_gfn + memslot->npages)
844 return memslot;
845 }
8b6d44c7 846 return NULL;
6aa8b732
AK		 847}
848EXPORT_SYMBOL_GPL(gfn_to_memslot);
849
850void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
851{
852 int i;
8b6d44c7 853 struct kvm_memory_slot *memslot = NULL;
6aa8b732
AK		 854 unsigned long rel_gfn;
855
856 for (i = 0; i < kvm->nmemslots; ++i) {
857 memslot = &kvm->memslots[i];
858
859 if (gfn >= memslot->base_gfn
860 && gfn < memslot->base_gfn + memslot->npages) {
861
862 if (!memslot || !memslot->dirty_bitmap)
863 return;
864
865 rel_gfn = gfn - memslot->base_gfn;
866
867 /* avoid RMW */
868 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
869 set_bit(rel_gfn, memslot->dirty_bitmap);
870 return;
871 }
872 }
873}
874
875static int emulator_read_std(unsigned long addr,
876 unsigned long *val,
877 unsigned int bytes,
878 struct x86_emulate_ctxt *ctxt)
879{
880 struct kvm_vcpu *vcpu = ctxt->vcpu;
881 void *data = val;
882
883 while (bytes) {
884 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
885 unsigned offset = addr & (PAGE_SIZE-1);
886 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
887 unsigned long pfn;
888 struct kvm_memory_slot *memslot;
889 void *page;
890
891 if (gpa == UNMAPPED_GVA)
892 return X86EMUL_PROPAGATE_FAULT;
893 pfn = gpa >> PAGE_SHIFT;
894 memslot = gfn_to_memslot(vcpu->kvm, pfn);
895 if (!memslot)
896 return X86EMUL_UNHANDLEABLE;
897 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
898
899 memcpy(data, page + offset, tocopy);
900
901 kunmap_atomic(page, KM_USER0);
902
903 bytes -= tocopy;
904 data += tocopy;
905 addr += tocopy;
906 }
907
908 return X86EMUL_CONTINUE;
909}
910
911static int emulator_write_std(unsigned long addr,
912 unsigned long val,
913 unsigned int bytes,
914 struct x86_emulate_ctxt *ctxt)
915{
916 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
917 addr, bytes);
918 return X86EMUL_UNHANDLEABLE;
919}
920
921static int emulator_read_emulated(unsigned long addr,
922 unsigned long *val,
923 unsigned int bytes,
924 struct x86_emulate_ctxt *ctxt)
925{
926 struct kvm_vcpu *vcpu = ctxt->vcpu;
927
928 if (vcpu->mmio_read_completed) {
929 memcpy(val, vcpu->mmio_data, bytes);
930 vcpu->mmio_read_completed = 0;
931 return X86EMUL_CONTINUE;
932 } else if (emulator_read_std(addr, val, bytes, ctxt)
933 == X86EMUL_CONTINUE)
934 return X86EMUL_CONTINUE;
935 else {
936 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
d27d4aca 937
6aa8b732 938 if (gpa == UNMAPPED_GVA)
d27d4aca 939 return X86EMUL_PROPAGATE_FAULT;
6aa8b732
AK		 940 vcpu->mmio_needed = 1;
941 vcpu->mmio_phys_addr = gpa;
942 vcpu->mmio_size = bytes;
943 vcpu->mmio_is_write = 0;
944
945 return X86EMUL_UNHANDLEABLE;
946 }
947}
948
da4a00f0
AK		 949static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
950 unsigned long val, int bytes)
951{
952 struct kvm_memory_slot *m;
953 struct page *page;
954 void *virt;
955
956 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
957 return 0;
958 m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
959 if (!m)
960 return 0;
961 page = gfn_to_page(m, gpa >> PAGE_SHIFT);
962 kvm_mmu_pre_write(vcpu, gpa, bytes);
ab51a434 963 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
da4a00f0
AK		 964 virt = kmap_atomic(page, KM_USER0);
965 memcpy(virt + offset_in_page(gpa), &val, bytes);
966 kunmap_atomic(virt, KM_USER0);
967 kvm_mmu_post_write(vcpu, gpa, bytes);
968 return 1;
969}
970
6aa8b732
AK		 971static int emulator_write_emulated(unsigned long addr,
972 unsigned long val,
973 unsigned int bytes,
974 struct x86_emulate_ctxt *ctxt)
975{
976 struct kvm_vcpu *vcpu = ctxt->vcpu;
977 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
978
979 if (gpa == UNMAPPED_GVA)
980 return X86EMUL_PROPAGATE_FAULT;
981
da4a00f0
AK		 982 if (emulator_write_phys(vcpu, gpa, val, bytes))
983 return X86EMUL_CONTINUE;
984
6aa8b732
AK		 985 vcpu->mmio_needed = 1;
986 vcpu->mmio_phys_addr = gpa;
987 vcpu->mmio_size = bytes;
988 vcpu->mmio_is_write = 1;
989 memcpy(vcpu->mmio_data, &val, bytes);
990
991 return X86EMUL_CONTINUE;
992}
993
994static int emulator_cmpxchg_emulated(unsigned long addr,
995 unsigned long old,
996 unsigned long new,
997 unsigned int bytes,
998 struct x86_emulate_ctxt *ctxt)
999{
1000 static int reported;
1001
1002 if (!reported) {
1003 reported = 1;
1004 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1005 }
1006 return emulator_write_emulated(addr, new, bytes, ctxt);
1007}
1008
32b35627
AK		 1009#ifdef CONFIG_X86_32
1010
1011static int emulator_cmpxchg8b_emulated(unsigned long addr,
1012 unsigned long old_lo,
1013 unsigned long old_hi,
1014 unsigned long new_lo,
1015 unsigned long new_hi,
1016 struct x86_emulate_ctxt *ctxt)
1017{
1018 static int reported;
1019 int r;
1020
1021 if (!reported) {
1022 reported = 1;
1023 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1024 }
1025 r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1026 if (r != X86EMUL_CONTINUE)
1027 return r;
1028 return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1029}
1030
1031#endif
1032
6aa8b732
AK		 1033static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1034{
1035 return kvm_arch_ops->get_segment_base(vcpu, seg);
1036}
1037
1038int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1039{
6aa8b732
AK		 1040 return X86EMUL_CONTINUE;
1041}
1042
1043int emulate_clts(struct kvm_vcpu *vcpu)
1044{
399badf3 1045 unsigned long cr0;
6aa8b732 1046
399badf3
AK		 1047 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1048 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
6aa8b732
AK		 1049 kvm_arch_ops->set_cr0(vcpu, cr0);
1050 return X86EMUL_CONTINUE;
1051}
1052
1053int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1054{
1055 struct kvm_vcpu *vcpu = ctxt->vcpu;
1056
1057 switch (dr) {
1058 case 0 ... 3:
1059 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1060 return X86EMUL_CONTINUE;
1061 default:
1062 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1063 __FUNCTION__, dr);
1064 return X86EMUL_UNHANDLEABLE;
1065 }
1066}
1067
1068int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1069{
1070 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1071 int exception;
1072
1073 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1074 if (exception) {
1075 /* FIXME: better handling */
1076 return X86EMUL_UNHANDLEABLE;
1077 }
1078 return X86EMUL_CONTINUE;
1079}
1080
1081static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1082{
1083 static int reported;
1084 u8 opcodes[4];
1085 unsigned long rip = ctxt->vcpu->rip;
1086 unsigned long rip_linear;
1087
1088 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1089
1090 if (reported)
1091 return;
1092
1093 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1094
1095 printk(KERN_ERR "emulation failed but !mmio_needed?"
1096 " rip %lx %02x %02x %02x %02x\n",
1097 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1098 reported = 1;
1099}
1100
1101struct x86_emulate_ops emulate_ops = {
1102 .read_std = emulator_read_std,
1103 .write_std = emulator_write_std,
1104 .read_emulated = emulator_read_emulated,
1105 .write_emulated = emulator_write_emulated,
1106 .cmpxchg_emulated = emulator_cmpxchg_emulated,
32b35627
AK		 1107#ifdef CONFIG_X86_32
1108 .cmpxchg8b_emulated = emulator_cmpxchg8b_emulated,
1109#endif
6aa8b732
AK		 1110};
1111
1112int emulate_instruction(struct kvm_vcpu *vcpu,
1113 struct kvm_run *run,
1114 unsigned long cr2,
1115 u16 error_code)
1116{
1117 struct x86_emulate_ctxt emulate_ctxt;
1118 int r;
1119 int cs_db, cs_l;
1120
1121 kvm_arch_ops->cache_regs(vcpu);
1122
1123 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1124
1125 emulate_ctxt.vcpu = vcpu;
1126 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1127 emulate_ctxt.cr2 = cr2;
1128 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1129 ? X86EMUL_MODE_REAL : cs_l
1130 ? X86EMUL_MODE_PROT64 : cs_db
1131 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1132
1133 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1134 emulate_ctxt.cs_base = 0;
1135 emulate_ctxt.ds_base = 0;
1136 emulate_ctxt.es_base = 0;
1137 emulate_ctxt.ss_base = 0;
1138 } else {
1139 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1140 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1141 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1142 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1143 }
1144
1145 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1146 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1147
1148 vcpu->mmio_is_write = 0;
1149 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1150
1151 if ((r || vcpu->mmio_is_write) && run) {
1152 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1153 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1154 run->mmio.len = vcpu->mmio_size;
1155 run->mmio.is_write = vcpu->mmio_is_write;
1156 }
1157
1158 if (r) {
a436036b
AK		 1159 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1160 return EMULATE_DONE;
6aa8b732
AK		 1161 if (!vcpu->mmio_needed) {
1162 report_emulation_failure(&emulate_ctxt);
1163 return EMULATE_FAIL;
1164 }
1165 return EMULATE_DO_MMIO;
1166 }
1167
1168 kvm_arch_ops->decache_regs(vcpu);
1169 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1170
1171 if (vcpu->mmio_is_write)
1172 return EMULATE_DO_MMIO;
1173
1174 return EMULATE_DONE;
1175}
1176EXPORT_SYMBOL_GPL(emulate_instruction);
1177
270fd9b9
AK		 1178int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1179{
1180 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1181
9b22bf57 1182 kvm_arch_ops->cache_regs(vcpu);
270fd9b9
AK		 1183 ret = -KVM_EINVAL;
1184#ifdef CONFIG_X86_64
1185 if (is_long_mode(vcpu)) {
1186 nr = vcpu->regs[VCPU_REGS_RAX];
1187 a0 = vcpu->regs[VCPU_REGS_RDI];
1188 a1 = vcpu->regs[VCPU_REGS_RSI];
1189 a2 = vcpu->regs[VCPU_REGS_RDX];
1190 a3 = vcpu->regs[VCPU_REGS_RCX];
1191 a4 = vcpu->regs[VCPU_REGS_R8];
1192 a5 = vcpu->regs[VCPU_REGS_R9];
1193 } else
1194#endif
1195 {
1196 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1197 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1198 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1199 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1200 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1201 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1202 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1203 }
1204 switch (nr) {
1205 default:
1206 ;
1207 }
1208 vcpu->regs[VCPU_REGS_RAX] = ret;
9b22bf57 1209 kvm_arch_ops->decache_regs(vcpu);
270fd9b9
AK		 1210 return 1;
1211}
1212EXPORT_SYMBOL_GPL(kvm_hypercall);
1213
6aa8b732
AK		 1214static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1215{
1216 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1217}
1218
1219void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1220{
1221 struct descriptor_table dt = { limit, base };
1222
1223 kvm_arch_ops->set_gdt(vcpu, &dt);
1224}
1225
1226void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1227{
1228 struct descriptor_table dt = { limit, base };
1229
1230 kvm_arch_ops->set_idt(vcpu, &dt);
1231}
1232
1233void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1234 unsigned long *rflags)
1235{
1236 lmsw(vcpu, msw);
1237 *rflags = kvm_arch_ops->get_rflags(vcpu);
1238}
1239
1240unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1241{
399badf3 1242 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
6aa8b732
AK		 1243 switch (cr) {
1244 case 0:
1245 return vcpu->cr0;
1246 case 2:
1247 return vcpu->cr2;
1248 case 3:
1249 return vcpu->cr3;
1250 case 4:
1251 return vcpu->cr4;
1252 default:
1253 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1254 return 0;
1255 }
1256}
1257
1258void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1259 unsigned long *rflags)
1260{
1261 switch (cr) {
1262 case 0:
1263 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1264 *rflags = kvm_arch_ops->get_rflags(vcpu);
1265 break;
1266 case 2:
1267 vcpu->cr2 = val;
1268 break;
1269 case 3:
1270 set_cr3(vcpu, val);
1271 break;
1272 case 4:
1273 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1274 break;
1275 default:
1276 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1277 }
1278}
1279
102d8325
IM		 1280/*
1281 * Register the para guest with the host:
1282 */
1283static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1284{
1285 struct kvm_vcpu_para_state *para_state;
1286 hpa_t para_state_hpa, hypercall_hpa;
1287 struct page *para_state_page;
1288 unsigned char *hypercall;
1289 gpa_t hypercall_gpa;
1290
1291 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1292 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1293
1294 /*
1295 * Needs to be page aligned:
1296 */
1297 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1298 goto err_gp;
1299
1300 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1301 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1302 if (is_error_hpa(para_state_hpa))
1303 goto err_gp;
1304
ab51a434 1305 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
102d8325
IM		 1306 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1307 para_state = kmap_atomic(para_state_page, KM_USER0);
1308
1309 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1310 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1311
1312 para_state->host_version = KVM_PARA_API_VERSION;
1313 /*
1314 * We cannot support guests that try to register themselves
1315 * with a newer API version than the host supports:
1316 */
1317 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1318 para_state->ret = -KVM_EINVAL;
1319 goto err_kunmap_skip;
1320 }
1321
1322 hypercall_gpa = para_state->hypercall_gpa;
1323 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1324 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1325 if (is_error_hpa(hypercall_hpa)) {
1326 para_state->ret = -KVM_EINVAL;
1327 goto err_kunmap_skip;
1328 }
1329
1330 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1331 vcpu->para_state_page = para_state_page;
1332 vcpu->para_state_gpa = para_state_gpa;
1333 vcpu->hypercall_gpa = hypercall_gpa;
1334
ab51a434 1335 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
102d8325
IM		 1336 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1337 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1338 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1339 kunmap_atomic(hypercall, KM_USER1);
1340
1341 para_state->ret = 0;
1342err_kunmap_skip:
1343 kunmap_atomic(para_state, KM_USER0);
1344 return 0;
1345err_gp:
1346 return 1;
1347}
1348
3bab1f5d
AK		 1349int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1350{
1351 u64 data;
1352
1353 switch (msr) {
1354 case 0xc0010010: /* SYSCFG */
1355 case 0xc0010015: /* HWCR */
1356 case MSR_IA32_PLATFORM_ID:
1357 case MSR_IA32_P5_MC_ADDR:
1358 case MSR_IA32_P5_MC_TYPE:
1359 case MSR_IA32_MC0_CTL:
1360 case MSR_IA32_MCG_STATUS:
1361 case MSR_IA32_MCG_CAP:
1362 case MSR_IA32_MC0_MISC:
1363 case MSR_IA32_MC0_MISC+4:
1364 case MSR_IA32_MC0_MISC+8:
1365 case MSR_IA32_MC0_MISC+12:
1366 case MSR_IA32_MC0_MISC+16:
1367 case MSR_IA32_UCODE_REV:
a8d13ea2 1368 case MSR_IA32_PERF_STATUS:
3bab1f5d
AK		 1369 /* MTRR registers */
1370 case 0xfe:
1371 case 0x200 ... 0x2ff:
1372 data = 0;
1373 break;
a8d13ea2
AK		 1374 case 0xcd: /* fsb frequency */
1375 data = 3;
1376 break;
3bab1f5d
AK		 1377 case MSR_IA32_APICBASE:
1378 data = vcpu->apic_base;
1379 break;
6f00e68f
AK		 1380 case MSR_IA32_MISC_ENABLE:
1381 data = vcpu->ia32_misc_enable_msr;
1382 break;
3bab1f5d
AK		 1383#ifdef CONFIG_X86_64
1384 case MSR_EFER:
1385 data = vcpu->shadow_efer;
1386 break;
1387#endif
1388 default:
1389 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1390 return 1;
1391 }
1392 *pdata = data;
1393 return 0;
1394}
1395EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1396
6aa8b732
AK		 1397/*
1398 * Reads an msr value (of 'msr_index') into 'pdata'.
1399 * Returns 0 on success, non-0 otherwise.
1400 * Assumes vcpu_load() was already called.
1401 */
1402static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1403{
1404 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1405}
1406
05b3e0c2 1407#ifdef CONFIG_X86_64
6aa8b732 1408
3bab1f5d 1409static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
6aa8b732 1410{
6aa8b732
AK		 1411 if (efer & EFER_RESERVED_BITS) {
1412 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1413 efer);
1414 inject_gp(vcpu);
1415 return;
1416 }
1417
1418 if (is_paging(vcpu)
1419 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1420 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1421 inject_gp(vcpu);
1422 return;
1423 }
1424
7725f0ba
AK		 1425 kvm_arch_ops->set_efer(vcpu, efer);
1426
6aa8b732
AK		 1427 efer &= ~EFER_LMA;
1428 efer |= vcpu->shadow_efer & EFER_LMA;
1429
1430 vcpu->shadow_efer = efer;
6aa8b732 1431}
6aa8b732
AK		 1432
1433#endif
1434
3bab1f5d
AK		 1435int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1436{
1437 switch (msr) {
1438#ifdef CONFIG_X86_64
1439 case MSR_EFER:
1440 set_efer(vcpu, data);
1441 break;
1442#endif
1443 case MSR_IA32_MC0_STATUS:
1444 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1445 __FUNCTION__, data);
1446 break;
1447 case MSR_IA32_UCODE_REV:
1448 case MSR_IA32_UCODE_WRITE:
1449 case 0x200 ... 0x2ff: /* MTRRs */
1450 break;
1451 case MSR_IA32_APICBASE:
1452 vcpu->apic_base = data;
1453 break;
6f00e68f
AK		 1454 case MSR_IA32_MISC_ENABLE:
1455 vcpu->ia32_misc_enable_msr = data;
1456 break;
102d8325
IM		 1457 /*
1458 * This is the 'probe whether the host is KVM' logic:
1459 */
1460 case MSR_KVM_API_MAGIC:
1461 return vcpu_register_para(vcpu, data);
1462
3bab1f5d
AK		 1463 default:
1464 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1465 return 1;
1466 }
1467 return 0;
1468}
1469EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1470
6aa8b732
AK		 1471/*
1472 * Writes msr value into into the appropriate "register".
1473 * Returns 0 on success, non-0 otherwise.
1474 * Assumes vcpu_load() was already called.
1475 */
1476static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1477{
1478 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1479}
1480
1481void kvm_resched(struct kvm_vcpu *vcpu)
1482{
1483 vcpu_put(vcpu);
1484 cond_resched();
bccf2150 1485 vcpu_load(vcpu);
6aa8b732
AK		 1486}
1487EXPORT_SYMBOL_GPL(kvm_resched);
1488
1489void load_msrs(struct vmx_msr_entry *e, int n)
1490{
1491 int i;
1492
1493 for (i = 0; i < n; ++i)
1494 wrmsrl(e[i].index, e[i].data);
1495}
1496EXPORT_SYMBOL_GPL(load_msrs);
1497
1498void save_msrs(struct vmx_msr_entry *e, int n)
1499{
1500 int i;
1501
1502 for (i = 0; i < n; ++i)
1503 rdmsrl(e[i].index, e[i].data);
1504}
1505EXPORT_SYMBOL_GPL(save_msrs);
1506
bccf2150 1507static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 1508{
6aa8b732
AK		 1509 int r;
1510
bccf2150 1511 vcpu_load(vcpu);
6aa8b732 1512
54810342
DL		 1513 /* re-sync apic's tpr */
1514 vcpu->cr8 = kvm_run->cr8;
1515
6aa8b732
AK		 1516 if (kvm_run->emulated) {
1517 kvm_arch_ops->skip_emulated_instruction(vcpu);
1518 kvm_run->emulated = 0;
1519 }
1520
1521 if (kvm_run->mmio_completed) {
1522 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1523 vcpu->mmio_read_completed = 1;
1524 }
1525
1526 vcpu->mmio_needed = 0;
1527
1528 r = kvm_arch_ops->run(vcpu, kvm_run);
1529
1530 vcpu_put(vcpu);
1531 return r;
1532}
1533
bccf2150
AK		 1534static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1535 struct kvm_regs *regs)
6aa8b732 1536{
bccf2150 1537 vcpu_load(vcpu);
6aa8b732
AK		 1538
1539 kvm_arch_ops->cache_regs(vcpu);
1540
1541 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1542 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1543 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1544 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1545 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1546 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1547 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1548 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
05b3e0c2 1549#ifdef CONFIG_X86_64
6aa8b732
AK		 1550 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1551 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1552 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1553 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1554 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1555 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1556 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1557 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1558#endif
1559
1560 regs->rip = vcpu->rip;
1561 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1562
1563 /*
1564 * Don't leak debug flags in case they were set for guest debugging
1565 */
1566 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1567 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1568
1569 vcpu_put(vcpu);
1570
1571 return 0;
1572}
1573
bccf2150
AK		 1574static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1575 struct kvm_regs *regs)
6aa8b732 1576{
bccf2150 1577 vcpu_load(vcpu);
6aa8b732
AK		 1578
1579 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1580 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1581 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1582 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1583 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1584 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1585 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1586 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
05b3e0c2 1587#ifdef CONFIG_X86_64
6aa8b732
AK		 1588 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1589 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1590 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1591 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1592 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1593 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1594 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1595 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1596#endif
1597
1598 vcpu->rip = regs->rip;
1599 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1600
1601 kvm_arch_ops->decache_regs(vcpu);
1602
1603 vcpu_put(vcpu);
1604
1605 return 0;
1606}
1607
1608static void get_segment(struct kvm_vcpu *vcpu,
1609 struct kvm_segment *var, int seg)
1610{
1611 return kvm_arch_ops->get_segment(vcpu, var, seg);
1612}
1613
bccf2150
AK		 1614static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1615 struct kvm_sregs *sregs)
6aa8b732 1616{
6aa8b732
AK		 1617 struct descriptor_table dt;
1618
bccf2150 1619 vcpu_load(vcpu);
6aa8b732
AK		 1620
1621 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1622 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1623 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1624 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1625 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1626 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1627
1628 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1629 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1630
1631 kvm_arch_ops->get_idt(vcpu, &dt);
1632 sregs->idt.limit = dt.limit;
1633 sregs->idt.base = dt.base;
1634 kvm_arch_ops->get_gdt(vcpu, &dt);
1635 sregs->gdt.limit = dt.limit;
1636 sregs->gdt.base = dt.base;
1637
399badf3 1638 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
6aa8b732
AK		 1639 sregs->cr0 = vcpu->cr0;
1640 sregs->cr2 = vcpu->cr2;
1641 sregs->cr3 = vcpu->cr3;
1642 sregs->cr4 = vcpu->cr4;
1643 sregs->cr8 = vcpu->cr8;
1644 sregs->efer = vcpu->shadow_efer;
1645 sregs->apic_base = vcpu->apic_base;
1646
1647 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1648 sizeof sregs->interrupt_bitmap);
1649
1650 vcpu_put(vcpu);
1651
1652 return 0;
1653}
1654
1655static void set_segment(struct kvm_vcpu *vcpu,
1656 struct kvm_segment *var, int seg)
1657{
1658 return kvm_arch_ops->set_segment(vcpu, var, seg);
1659}
1660
bccf2150
AK		 1661static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1662 struct kvm_sregs *sregs)
6aa8b732 1663{
6aa8b732
AK		 1664 int mmu_reset_needed = 0;
1665 int i;
1666 struct descriptor_table dt;
1667
bccf2150 1668 vcpu_load(vcpu);
6aa8b732
AK		 1669
1670 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1671 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1672 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1673 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1674 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1675 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1676
1677 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1678 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1679
1680 dt.limit = sregs->idt.limit;
1681 dt.base = sregs->idt.base;
1682 kvm_arch_ops->set_idt(vcpu, &dt);
1683 dt.limit = sregs->gdt.limit;
1684 dt.base = sregs->gdt.base;
1685 kvm_arch_ops->set_gdt(vcpu, &dt);
1686
1687 vcpu->cr2 = sregs->cr2;
1688 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1689 vcpu->cr3 = sregs->cr3;
1690
1691 vcpu->cr8 = sregs->cr8;
1692
1693 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
05b3e0c2 1694#ifdef CONFIG_X86_64
6aa8b732
AK		 1695 kvm_arch_ops->set_efer(vcpu, sregs->efer);
1696#endif
1697 vcpu->apic_base = sregs->apic_base;
1698
399badf3
AK		 1699 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1700
6aa8b732
AK		 1701 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1702 kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1703
1704 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1705 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1b0973bd
AK		 1706 if (!is_long_mode(vcpu) && is_pae(vcpu))
1707 load_pdptrs(vcpu, vcpu->cr3);
6aa8b732
AK		 1708
1709 if (mmu_reset_needed)
1710 kvm_mmu_reset_context(vcpu);
1711
1712 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1713 sizeof vcpu->irq_pending);
1714 vcpu->irq_summary = 0;
1715 for (i = 0; i < NR_IRQ_WORDS; ++i)
1716 if (vcpu->irq_pending[i])
1717 __set_bit(i, &vcpu->irq_summary);
1718
1719 vcpu_put(vcpu);
1720
1721 return 0;
1722}
1723
1724/*
1725 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1726 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
bf591b24
MR		 1727 *
1728 * This list is modified at module load time to reflect the
1729 * capabilities of the host cpu.
6aa8b732
AK		 1730 */
1731static u32 msrs_to_save[] = {
1732 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1733 MSR_K6_STAR,
05b3e0c2 1734#ifdef CONFIG_X86_64
6aa8b732
AK		 1735 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1736#endif
1737 MSR_IA32_TIME_STAMP_COUNTER,
1738};
1739
bf591b24
MR		 1740static unsigned num_msrs_to_save;
1741
6f00e68f
AK		 1742static u32 emulated_msrs[] = {
1743 MSR_IA32_MISC_ENABLE,
1744};
1745
bf591b24
MR		 1746static __init void kvm_init_msr_list(void)
1747{
1748 u32 dummy[2];
1749 unsigned i, j;
1750
1751 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1752 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1753 continue;
1754 if (j < i)
1755 msrs_to_save[j] = msrs_to_save[i];
1756 j++;
1757 }
1758 num_msrs_to_save = j;
1759}
6aa8b732
AK		 1760
1761/*
1762 * Adapt set_msr() to msr_io()'s calling convention
1763 */
1764static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1765{
1766 return set_msr(vcpu, index, *data);
1767}
1768
1769/*
1770 * Read or write a bunch of msrs. All parameters are kernel addresses.
1771 *
1772 * @return number of msrs set successfully.
1773 */
bccf2150 1774static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
6aa8b732
AK		 1775 struct kvm_msr_entry *entries,
1776 int (*do_msr)(struct kvm_vcpu *vcpu,
1777 unsigned index, u64 *data))
1778{
6aa8b732
AK		 1779 int i;
1780
bccf2150 1781 vcpu_load(vcpu);
6aa8b732
AK		 1782
1783 for (i = 0; i < msrs->nmsrs; ++i)
1784 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1785 break;
1786
1787 vcpu_put(vcpu);
1788
1789 return i;
1790}
1791
1792/*
1793 * Read or write a bunch of msrs. Parameters are user addresses.
1794 *
1795 * @return number of msrs set successfully.
1796 */
bccf2150 1797static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
6aa8b732
AK		 1798 int (*do_msr)(struct kvm_vcpu *vcpu,
1799 unsigned index, u64 *data),
1800 int writeback)
1801{
1802 struct kvm_msrs msrs;
1803 struct kvm_msr_entry *entries;
1804 int r, n;
1805 unsigned size;
1806
1807 r = -EFAULT;
1808 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1809 goto out;
1810
1811 r = -E2BIG;
1812 if (msrs.nmsrs >= MAX_IO_MSRS)
1813 goto out;
1814
1815 r = -ENOMEM;
1816 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1817 entries = vmalloc(size);
1818 if (!entries)
1819 goto out;
1820
1821 r = -EFAULT;
1822 if (copy_from_user(entries, user_msrs->entries, size))
1823 goto out_free;
1824
bccf2150 1825 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
6aa8b732
AK		 1826 if (r < 0)
1827 goto out_free;
1828
1829 r = -EFAULT;
1830 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1831 goto out_free;
1832
1833 r = n;
1834
1835out_free:
1836 vfree(entries);
1837out:
1838 return r;
1839}
1840
1841/*
1842 * Translate a guest virtual address to a guest physical address.
1843 */
bccf2150
AK		 1844static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1845 struct kvm_translation *tr)
6aa8b732
AK		 1846{
1847 unsigned long vaddr = tr->linear_address;
6aa8b732
AK		 1848 gpa_t gpa;
1849
bccf2150
AK		 1850 vcpu_load(vcpu);
1851 spin_lock(&vcpu->kvm->lock);
6aa8b732
AK		 1852 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1853 tr->physical_address = gpa;
1854 tr->valid = gpa != UNMAPPED_GVA;
1855 tr->writeable = 1;
1856 tr->usermode = 0;
bccf2150 1857 spin_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 1858 vcpu_put(vcpu);
1859
1860 return 0;
1861}
1862
bccf2150
AK		 1863static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1864 struct kvm_interrupt *irq)
6aa8b732 1865{
6aa8b732
AK		 1866 if (irq->irq < 0 || irq->irq >= 256)
1867 return -EINVAL;
bccf2150 1868 vcpu_load(vcpu);
6aa8b732
AK		 1869
1870 set_bit(irq->irq, vcpu->irq_pending);
1871 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1872
1873 vcpu_put(vcpu);
1874
1875 return 0;
1876}
1877
bccf2150
AK		 1878static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1879 struct kvm_debug_guest *dbg)
6aa8b732 1880{
6aa8b732
AK		 1881 int r;
1882
bccf2150 1883 vcpu_load(vcpu);
6aa8b732
AK		 1884
1885 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1886
1887 vcpu_put(vcpu);
1888
1889 return r;
1890}
1891
9a2bb7f4
AK		 1892static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1893 unsigned long address,
1894 int *type)
1895{
1896 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1897 unsigned long pgoff;
1898 struct page *page;
1899
1900 *type = VM_FAULT_MINOR;
1901 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1902 if (pgoff != 0)
1903 return NOPAGE_SIGBUS;
1904 page = virt_to_page(vcpu->run);
1905 get_page(page);
1906 return page;
1907}
1908
1909static struct vm_operations_struct kvm_vcpu_vm_ops = {
1910 .nopage = kvm_vcpu_nopage,
1911};
1912
1913static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1914{
1915 vma->vm_ops = &kvm_vcpu_vm_ops;
1916 return 0;
1917}
1918
bccf2150
AK		 1919static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1920{
1921 struct kvm_vcpu *vcpu = filp->private_data;
1922
1923 fput(vcpu->kvm->filp);
1924 return 0;
1925}
1926
1927static struct file_operations kvm_vcpu_fops = {
1928 .release = kvm_vcpu_release,
1929 .unlocked_ioctl = kvm_vcpu_ioctl,
1930 .compat_ioctl = kvm_vcpu_ioctl,
9a2bb7f4 1931 .mmap = kvm_vcpu_mmap,
bccf2150
AK		 1932};
1933
1934/*
1935 * Allocates an inode for the vcpu.
1936 */
1937static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1938{
1939 int fd, r;
1940 struct inode *inode;
1941 struct file *file;
1942
1943 atomic_inc(&vcpu->kvm->filp->f_count);
1944 inode = kvmfs_inode(&kvm_vcpu_fops);
1945 if (IS_ERR(inode)) {
1946 r = PTR_ERR(inode);
1947 goto out1;
1948 }
1949
1950 file = kvmfs_file(inode, vcpu);
1951 if (IS_ERR(file)) {
1952 r = PTR_ERR(file);
1953 goto out2;
1954 }
1955
1956 r = get_unused_fd();
1957 if (r < 0)
1958 goto out3;
1959 fd = r;
1960 fd_install(fd, file);
1961
1962 return fd;
1963
1964out3:
1965 fput(file);
1966out2:
1967 iput(inode);
1968out1:
1969 fput(vcpu->kvm->filp);
1970 return r;
1971}
1972
c5ea7660
AK		 1973/*
1974 * Creates some virtual cpus. Good luck creating more than one.
1975 */
1976static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1977{
1978 int r;
1979 struct kvm_vcpu *vcpu;
9a2bb7f4 1980 struct page *page;
c5ea7660
AK		 1981
1982 r = -EINVAL;
1983 if (!valid_vcpu(n))
1984 goto out;
1985
1986 vcpu = &kvm->vcpus[n];
1987
1988 mutex_lock(&vcpu->mutex);
1989
1990 if (vcpu->vmcs) {
1991 mutex_unlock(&vcpu->mutex);
1992 return -EEXIST;
1993 }
1994
9a2bb7f4
AK		 1995 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1996 r = -ENOMEM;
1997 if (!page)
1998 goto out_unlock;
1999 vcpu->run = page_address(page);
2000
c5ea7660
AK		 2001 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2002 FX_IMAGE_ALIGN);
2003 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2004
2005 r = kvm_arch_ops->vcpu_create(vcpu);
2006 if (r < 0)
2007 goto out_free_vcpus;
2008
2009 r = kvm_mmu_create(vcpu);
2010 if (r < 0)
2011 goto out_free_vcpus;
2012
2013 kvm_arch_ops->vcpu_load(vcpu);
2014 r = kvm_mmu_setup(vcpu);
2015 if (r >= 0)
2016 r = kvm_arch_ops->vcpu_setup(vcpu);
2017 vcpu_put(vcpu);
2018
2019 if (r < 0)
2020 goto out_free_vcpus;
2021
bccf2150
AK		 2022 r = create_vcpu_fd(vcpu);
2023 if (r < 0)
2024 goto out_free_vcpus;
2025
2026 return r;
c5ea7660
AK		 2027
2028out_free_vcpus:
2029 kvm_free_vcpu(vcpu);
9a2bb7f4 2030out_unlock:
c5ea7660
AK		 2031 mutex_unlock(&vcpu->mutex);
2032out:
2033 return r;
2034}
2035
bccf2150
AK		 2036static long kvm_vcpu_ioctl(struct file *filp,
2037 unsigned int ioctl, unsigned long arg)
6aa8b732 2038{
bccf2150 2039 struct kvm_vcpu *vcpu = filp->private_data;
2f366987 2040 void __user *argp = (void __user *)arg;
6aa8b732
AK		 2041 int r = -EINVAL;
2042
2043 switch (ioctl) {
9a2bb7f4
AK		 2044 case KVM_RUN:
2045 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
6aa8b732 2046 break;
6aa8b732
AK		 2047 case KVM_GET_REGS: {
2048 struct kvm_regs kvm_regs;
2049
bccf2150
AK		 2050 memset(&kvm_regs, 0, sizeof kvm_regs);
2051 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2052 if (r)
2053 goto out;
2054 r = -EFAULT;
2f366987 2055 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
6aa8b732
AK		 2056 goto out;
2057 r = 0;
2058 break;
2059 }
2060 case KVM_SET_REGS: {
2061 struct kvm_regs kvm_regs;
2062
2063 r = -EFAULT;
2f366987 2064 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
6aa8b732 2065 goto out;
bccf2150 2066 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2067 if (r)
2068 goto out;
2069 r = 0;
2070 break;
2071 }
2072 case KVM_GET_SREGS: {
2073 struct kvm_sregs kvm_sregs;
2074
bccf2150
AK		 2075 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2076 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2077 if (r)
2078 goto out;
2079 r = -EFAULT;
2f366987 2080 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
6aa8b732
AK		 2081 goto out;
2082 r = 0;
2083 break;
2084 }
2085 case KVM_SET_SREGS: {
2086 struct kvm_sregs kvm_sregs;
2087
2088 r = -EFAULT;
2f366987 2089 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
6aa8b732 2090 goto out;
bccf2150 2091 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2092 if (r)
2093 goto out;
2094 r = 0;
2095 break;
2096 }
2097 case KVM_TRANSLATE: {
2098 struct kvm_translation tr;
2099
2100 r = -EFAULT;
2f366987 2101 if (copy_from_user(&tr, argp, sizeof tr))
6aa8b732 2102 goto out;
bccf2150 2103 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
6aa8b732
AK		 2104 if (r)
2105 goto out;
2106 r = -EFAULT;
2f366987 2107 if (copy_to_user(argp, &tr, sizeof tr))
6aa8b732
AK		 2108 goto out;
2109 r = 0;
2110 break;
2111 }
2112 case KVM_INTERRUPT: {
2113 struct kvm_interrupt irq;
2114
2115 r = -EFAULT;
2f366987 2116 if (copy_from_user(&irq, argp, sizeof irq))
6aa8b732 2117 goto out;
bccf2150 2118 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
6aa8b732
AK		 2119 if (r)
2120 goto out;
2121 r = 0;
2122 break;
2123 }
2124 case KVM_DEBUG_GUEST: {
2125 struct kvm_debug_guest dbg;
2126
2127 r = -EFAULT;
2f366987 2128 if (copy_from_user(&dbg, argp, sizeof dbg))
6aa8b732 2129 goto out;
bccf2150 2130 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
6aa8b732
AK		 2131 if (r)
2132 goto out;
2133 r = 0;
2134 break;
2135 }
bccf2150
AK		 2136 case KVM_GET_MSRS:
2137 r = msr_io(vcpu, argp, get_msr, 1);
2138 break;
2139 case KVM_SET_MSRS:
2140 r = msr_io(vcpu, argp, do_set_msr, 0);
2141 break;
2142 default:
2143 ;
2144 }
2145out:
2146 return r;
2147}
2148
2149static long kvm_vm_ioctl(struct file *filp,
2150 unsigned int ioctl, unsigned long arg)
2151{
2152 struct kvm *kvm = filp->private_data;
2153 void __user *argp = (void __user *)arg;
2154 int r = -EINVAL;
2155
2156 switch (ioctl) {
2157 case KVM_CREATE_VCPU:
2158 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2159 if (r < 0)
2160 goto out;
2161 break;
6aa8b732
AK		 2162 case KVM_SET_MEMORY_REGION: {
2163 struct kvm_memory_region kvm_mem;
2164
2165 r = -EFAULT;
2f366987 2166 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
6aa8b732 2167 goto out;
2c6f5df9 2168 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
6aa8b732
AK		 2169 if (r)
2170 goto out;
2171 break;
2172 }
2173 case KVM_GET_DIRTY_LOG: {
2174 struct kvm_dirty_log log;
2175
2176 r = -EFAULT;
2f366987 2177 if (copy_from_user(&log, argp, sizeof log))
6aa8b732 2178 goto out;
2c6f5df9 2179 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
6aa8b732
AK		 2180 if (r)
2181 goto out;
2182 break;
2183 }
f17abe9a
AK		 2184 default:
2185 ;
2186 }
2187out:
2188 return r;
2189}
2190
2191static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2192 unsigned long address,
2193 int *type)
2194{
2195 struct kvm *kvm = vma->vm_file->private_data;
2196 unsigned long pgoff;
2197 struct kvm_memory_slot *slot;
2198 struct page *page;
2199
2200 *type = VM_FAULT_MINOR;
2201 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2202 slot = gfn_to_memslot(kvm, pgoff);
2203 if (!slot)
2204 return NOPAGE_SIGBUS;
2205 page = gfn_to_page(slot, pgoff);
2206 if (!page)
2207 return NOPAGE_SIGBUS;
2208 get_page(page);
2209 return page;
2210}
2211
2212static struct vm_operations_struct kvm_vm_vm_ops = {
2213 .nopage = kvm_vm_nopage,
2214};
2215
2216static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2217{
2218 vma->vm_ops = &kvm_vm_vm_ops;
2219 return 0;
2220}
2221
2222static struct file_operations kvm_vm_fops = {
2223 .release = kvm_vm_release,
2224 .unlocked_ioctl = kvm_vm_ioctl,
2225 .compat_ioctl = kvm_vm_ioctl,
2226 .mmap = kvm_vm_mmap,
2227};
2228
2229static int kvm_dev_ioctl_create_vm(void)
2230{
2231 int fd, r;
2232 struct inode *inode;
2233 struct file *file;
2234 struct kvm *kvm;
2235
2236 inode = kvmfs_inode(&kvm_vm_fops);
2237 if (IS_ERR(inode)) {
2238 r = PTR_ERR(inode);
2239 goto out1;
2240 }
2241
2242 kvm = kvm_create_vm();
2243 if (IS_ERR(kvm)) {
2244 r = PTR_ERR(kvm);
2245 goto out2;
2246 }
2247
2248 file = kvmfs_file(inode, kvm);
2249 if (IS_ERR(file)) {
2250 r = PTR_ERR(file);
2251 goto out3;
2252 }
bccf2150 2253 kvm->filp = file;
f17abe9a
AK		 2254
2255 r = get_unused_fd();
2256 if (r < 0)
2257 goto out4;
2258 fd = r;
2259 fd_install(fd, file);
2260
2261 return fd;
2262
2263out4:
2264 fput(file);
2265out3:
2266 kvm_destroy_vm(kvm);
2267out2:
2268 iput(inode);
2269out1:
2270 return r;
2271}
2272
2273static long kvm_dev_ioctl(struct file *filp,
2274 unsigned int ioctl, unsigned long arg)
2275{
2276 void __user *argp = (void __user *)arg;
2277 int r = -EINVAL;
2278
2279 switch (ioctl) {
2280 case KVM_GET_API_VERSION:
2281 r = KVM_API_VERSION;
2282 break;
2283 case KVM_CREATE_VM:
2284 r = kvm_dev_ioctl_create_vm();
2285 break;
6aa8b732 2286 case KVM_GET_MSR_INDEX_LIST: {
2f366987 2287 struct kvm_msr_list __user *user_msr_list = argp;
6aa8b732
AK		 2288 struct kvm_msr_list msr_list;
2289 unsigned n;
2290
2291 r = -EFAULT;
2292 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2293 goto out;
2294 n = msr_list.nmsrs;
6f00e68f 2295 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
6aa8b732
AK		 2296 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2297 goto out;
2298 r = -E2BIG;
bf591b24 2299 if (n < num_msrs_to_save)
6aa8b732
AK		 2300 goto out;
2301 r = -EFAULT;
2302 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
bf591b24 2303 num_msrs_to_save * sizeof(u32)))
6aa8b732 2304 goto out;
6f00e68f
AK		 2305 if (copy_to_user(user_msr_list->indices
2306 + num_msrs_to_save * sizeof(u32),
2307 &emulated_msrs,
2308 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2309 goto out;
6aa8b732 2310 r = 0;
cc1d8955 2311 break;
6aa8b732
AK		 2312 }
2313 default:
2314 ;
2315 }
2316out:
2317 return r;
2318}
2319
6aa8b732
AK		 2320static struct file_operations kvm_chardev_ops = {
2321 .open = kvm_dev_open,
2322 .release = kvm_dev_release,
2323 .unlocked_ioctl = kvm_dev_ioctl,
2324 .compat_ioctl = kvm_dev_ioctl,
6aa8b732
AK		 2325};
2326
2327static struct miscdevice kvm_dev = {
bbe4432e 2328 KVM_MINOR,
6aa8b732
AK		 2329 "kvm",
2330 &kvm_chardev_ops,
2331};
2332
2333static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2334 void *v)
2335{
2336 if (val == SYS_RESTART) {
2337 /*
2338 * Some (well, at least mine) BIOSes hang on reboot if
2339 * in vmx root mode.
2340 */
2341 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
8b6d44c7 2342 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
6aa8b732
AK		 2343 }
2344 return NOTIFY_OK;
2345}
2346
2347static struct notifier_block kvm_reboot_notifier = {
2348 .notifier_call = kvm_reboot,
2349 .priority = 0,
2350};
2351
774c47f1
AK		 2352/*
2353 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2354 * cached on it.
2355 */
2356static void decache_vcpus_on_cpu(int cpu)
2357{
2358 struct kvm *vm;
2359 struct kvm_vcpu *vcpu;
2360 int i;
2361
2362 spin_lock(&kvm_lock);
2363 list_for_each_entry(vm, &vm_list, vm_list)
2364 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2365 vcpu = &vm->vcpus[i];
2366 /*
2367 * If the vcpu is locked, then it is running on some
2368 * other cpu and therefore it is not cached on the
2369 * cpu in question.
2370 *
2371 * If it's not locked, check the last cpu it executed
2372 * on.
2373 */
2374 if (mutex_trylock(&vcpu->mutex)) {
2375 if (vcpu->cpu == cpu) {
2376 kvm_arch_ops->vcpu_decache(vcpu);
2377 vcpu->cpu = -1;
2378 }
2379 mutex_unlock(&vcpu->mutex);
2380 }
2381 }
2382 spin_unlock(&kvm_lock);
2383}
2384
2385static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2386 void *v)
2387{
2388 int cpu = (long)v;
2389
2390 switch (val) {
43934a38 2391 case CPU_DOWN_PREPARE:
774c47f1 2392 case CPU_UP_CANCELED:
43934a38
JK		 2393 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2394 cpu);
774c47f1
AK		 2395 decache_vcpus_on_cpu(cpu);
2396 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2397 NULL, 0, 1);
2398 break;
43934a38
JK		 2399 case CPU_ONLINE:
2400 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2401 cpu);
774c47f1
AK		 2402 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2403 NULL, 0, 1);
2404 break;
2405 }
2406 return NOTIFY_OK;
2407}
2408
2409static struct notifier_block kvm_cpu_notifier = {
2410 .notifier_call = kvm_cpu_hotplug,
2411 .priority = 20, /* must be > scheduler priority */
2412};
2413
6aa8b732
AK		 2414static __init void kvm_init_debug(void)
2415{
2416 struct kvm_stats_debugfs_item *p;
2417
8b6d44c7 2418 debugfs_dir = debugfs_create_dir("kvm", NULL);
6aa8b732
AK		 2419 for (p = debugfs_entries; p->name; ++p)
2420 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2421 p->data);
2422}
2423
2424static void kvm_exit_debug(void)
2425{
2426 struct kvm_stats_debugfs_item *p;
2427
2428 for (p = debugfs_entries; p->name; ++p)
2429 debugfs_remove(p->dentry);
2430 debugfs_remove(debugfs_dir);
2431}
2432
59ae6c6b
AK		 2433static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2434{
2435 decache_vcpus_on_cpu(raw_smp_processor_id());
19d1408d 2436 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
59ae6c6b
AK		 2437 return 0;
2438}
2439
2440static int kvm_resume(struct sys_device *dev)
2441{
19d1408d 2442 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
59ae6c6b
AK		 2443 return 0;
2444}
2445
2446static struct sysdev_class kvm_sysdev_class = {
2447 set_kset_name("kvm"),
2448 .suspend = kvm_suspend,
2449 .resume = kvm_resume,
2450};
2451
2452static struct sys_device kvm_sysdev = {
2453 .id = 0,
2454 .cls = &kvm_sysdev_class,
2455};
2456
6aa8b732
AK		 2457hpa_t bad_page_address;
2458
37e29d90
AK		 2459static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2460 const char *dev_name, void *data, struct vfsmount *mnt)
2461{
e9cdb1e3 2462 return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
37e29d90
AK		 2463}
2464
2465static struct file_system_type kvm_fs_type = {
2466 .name = "kvmfs",
2467 .get_sb = kvmfs_get_sb,
2468 .kill_sb = kill_anon_super,
2469};
2470
6aa8b732
AK		 2471int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2472{
2473 int r;
2474
09db28b8
YI		 2475 if (kvm_arch_ops) {
2476 printk(KERN_ERR "kvm: already loaded the other module\n");
2477 return -EEXIST;
2478 }
2479
e097f35c 2480 if (!ops->cpu_has_kvm_support()) {
6aa8b732
AK		 2481 printk(KERN_ERR "kvm: no hardware support\n");
2482 return -EOPNOTSUPP;
2483 }
e097f35c 2484 if (ops->disabled_by_bios()) {
6aa8b732
AK		 2485 printk(KERN_ERR "kvm: disabled by bios\n");
2486 return -EOPNOTSUPP;
2487 }
2488
e097f35c
YI		 2489 kvm_arch_ops = ops;
2490
6aa8b732
AK		 2491 r = kvm_arch_ops->hardware_setup();
2492 if (r < 0)
ca45aaae 2493 goto out;
6aa8b732 2494
8b6d44c7 2495 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
774c47f1
AK		 2496 r = register_cpu_notifier(&kvm_cpu_notifier);
2497 if (r)
2498 goto out_free_1;
6aa8b732
AK		 2499 register_reboot_notifier(&kvm_reboot_notifier);
2500
59ae6c6b
AK		 2501 r = sysdev_class_register(&kvm_sysdev_class);
2502 if (r)
2503 goto out_free_2;
2504
2505 r = sysdev_register(&kvm_sysdev);
2506 if (r)
2507 goto out_free_3;
2508
6aa8b732
AK		 2509 kvm_chardev_ops.owner = module;
2510
2511 r = misc_register(&kvm_dev);
2512 if (r) {
2513 printk (KERN_ERR "kvm: misc device register failed\n");
2514 goto out_free;
2515 }
2516
2517 return r;
2518
2519out_free:
59ae6c6b
AK		 2520 sysdev_unregister(&kvm_sysdev);
2521out_free_3:
2522 sysdev_class_unregister(&kvm_sysdev_class);
2523out_free_2:
6aa8b732 2524 unregister_reboot_notifier(&kvm_reboot_notifier);
774c47f1
AK		 2525 unregister_cpu_notifier(&kvm_cpu_notifier);
2526out_free_1:
8b6d44c7 2527 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
6aa8b732 2528 kvm_arch_ops->hardware_unsetup();
ca45aaae
AK		 2529out:
2530 kvm_arch_ops = NULL;
6aa8b732
AK		 2531 return r;
2532}
2533
2534void kvm_exit_arch(void)
2535{
2536 misc_deregister(&kvm_dev);
59ae6c6b
AK		 2537 sysdev_unregister(&kvm_sysdev);
2538 sysdev_class_unregister(&kvm_sysdev_class);
6aa8b732 2539 unregister_reboot_notifier(&kvm_reboot_notifier);
59ae6c6b 2540 unregister_cpu_notifier(&kvm_cpu_notifier);
8b6d44c7 2541 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
6aa8b732 2542 kvm_arch_ops->hardware_unsetup();
09db28b8 2543 kvm_arch_ops = NULL;
6aa8b732
AK		 2544}
2545
2546static __init int kvm_init(void)
2547{
2548 static struct page *bad_page;
37e29d90
AK		 2549 int r;
2550
2551 r = register_filesystem(&kvm_fs_type);
2552 if (r)
2553 goto out3;
6aa8b732 2554
37e29d90
AK		 2555 kvmfs_mnt = kern_mount(&kvm_fs_type);
2556 r = PTR_ERR(kvmfs_mnt);
2557 if (IS_ERR(kvmfs_mnt))
2558 goto out2;
6aa8b732
AK		 2559 kvm_init_debug();
2560
bf591b24
MR		 2561 kvm_init_msr_list();
2562
6aa8b732
AK		 2563 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2564 r = -ENOMEM;
2565 goto out;
2566 }
2567
2568 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2569 memset(__va(bad_page_address), 0, PAGE_SIZE);
2570
58e690e6 2571 return 0;
6aa8b732
AK		 2572
2573out:
2574 kvm_exit_debug();
37e29d90
AK		 2575 mntput(kvmfs_mnt);
2576out2:
2577 unregister_filesystem(&kvm_fs_type);
2578out3:
6aa8b732
AK		 2579 return r;
2580}
2581
2582static __exit void kvm_exit(void)
2583{
2584 kvm_exit_debug();
2585 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
37e29d90
AK		 2586 mntput(kvmfs_mnt);
2587 unregister_filesystem(&kvm_fs_type);
6aa8b732
AK		 2588}
2589
2590module_init(kvm_init)
2591module_exit(kvm_exit)
2592
2593EXPORT_SYMBOL_GPL(kvm_init_arch);
2594EXPORT_SYMBOL_GPL(kvm_exit_arch);
Linux kernel - Deliverables
This page took 0.18278 seconds and 5 git commands to generate.