[PATCH] KVM: MMU: Destroy mmu while we still have a vcpu left
[linux-2.6] / drivers / kvm / kvm_main.c
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>
23 #include <asm/processor.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
26 #include <asm/msr.h>
27 #include <linux/mm.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <asm/uaccess.h>
31 #include <linux/reboot.h>
32 #include <asm/io.h>
33 #include <linux/debugfs.h>
34 #include <linux/highmem.h>
35 #include <linux/file.h>
36 #include <asm/desc.h>
37
38 #include "x86_emulate.h"
39 #include "segment_descriptor.h"
40
41 MODULE_AUTHOR("Qumranet");
42 MODULE_LICENSE("GPL");
43
44 struct kvm_arch_ops *kvm_arch_ops;
45 struct kvm_stat kvm_stat;
46 EXPORT_SYMBOL_GPL(kvm_stat);
47
48 static struct kvm_stats_debugfs_item {
49         const char *name;
50         u32 *data;
51         struct dentry *dentry;
52 } debugfs_entries[] = {
53         { "pf_fixed", &kvm_stat.pf_fixed },
54         { "pf_guest", &kvm_stat.pf_guest },
55         { "tlb_flush", &kvm_stat.tlb_flush },
56         { "invlpg", &kvm_stat.invlpg },
57         { "exits", &kvm_stat.exits },
58         { "io_exits", &kvm_stat.io_exits },
59         { "mmio_exits", &kvm_stat.mmio_exits },
60         { "signal_exits", &kvm_stat.signal_exits },
61         { "irq_window", &kvm_stat.irq_window_exits },
62         { "halt_exits", &kvm_stat.halt_exits },
63         { "request_irq", &kvm_stat.request_irq_exits },
64         { "irq_exits", &kvm_stat.irq_exits },
65         { 0, 0 }
66 };
67
68 static struct dentry *debugfs_dir;
69
70 #define MAX_IO_MSRS 256
71
72 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
73 #define LMSW_GUEST_MASK 0x0eULL
74 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
75 #define CR8_RESEVED_BITS (~0x0fULL)
76 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
77
78 #ifdef CONFIG_X86_64
79 // LDT or TSS descriptor in the GDT. 16 bytes.
80 struct segment_descriptor_64 {
81         struct segment_descriptor s;
82         u32 base_higher;
83         u32 pad_zero;
84 };
85
86 #endif
87
88 unsigned long segment_base(u16 selector)
89 {
90         struct descriptor_table gdt;
91         struct segment_descriptor *d;
92         unsigned long table_base;
93         typedef unsigned long ul;
94         unsigned long v;
95
96         if (selector == 0)
97                 return 0;
98
99         asm ("sgdt %0" : "=m"(gdt));
100         table_base = gdt.base;
101
102         if (selector & 4) {           /* from ldt */
103                 u16 ldt_selector;
104
105                 asm ("sldt %0" : "=g"(ldt_selector));
106                 table_base = segment_base(ldt_selector);
107         }
108         d = (struct segment_descriptor *)(table_base + (selector & ~7));
109         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
110 #ifdef CONFIG_X86_64
111         if (d->system == 0
112             && (d->type == 2 || d->type == 9 || d->type == 11))
113                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
114 #endif
115         return v;
116 }
117 EXPORT_SYMBOL_GPL(segment_base);
118
119 static inline int valid_vcpu(int n)
120 {
121         return likely(n >= 0 && n < KVM_MAX_VCPUS);
122 }
123
124 int kvm_read_guest(struct kvm_vcpu *vcpu,
125                              gva_t addr,
126                              unsigned long size,
127                              void *dest)
128 {
129         unsigned char *host_buf = dest;
130         unsigned long req_size = size;
131
132         while (size) {
133                 hpa_t paddr;
134                 unsigned now;
135                 unsigned offset;
136                 hva_t guest_buf;
137
138                 paddr = gva_to_hpa(vcpu, addr);
139
140                 if (is_error_hpa(paddr))
141                         break;
142
143                 guest_buf = (hva_t)kmap_atomic(
144                                         pfn_to_page(paddr >> PAGE_SHIFT),
145                                         KM_USER0);
146                 offset = addr & ~PAGE_MASK;
147                 guest_buf |= offset;
148                 now = min(size, PAGE_SIZE - offset);
149                 memcpy(host_buf, (void*)guest_buf, now);
150                 host_buf += now;
151                 addr += now;
152                 size -= now;
153                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
154         }
155         return req_size - size;
156 }
157 EXPORT_SYMBOL_GPL(kvm_read_guest);
158
159 int kvm_write_guest(struct kvm_vcpu *vcpu,
160                              gva_t addr,
161                              unsigned long size,
162                              void *data)
163 {
164         unsigned char *host_buf = data;
165         unsigned long req_size = size;
166
167         while (size) {
168                 hpa_t paddr;
169                 unsigned now;
170                 unsigned offset;
171                 hva_t guest_buf;
172
173                 paddr = gva_to_hpa(vcpu, addr);
174
175                 if (is_error_hpa(paddr))
176                         break;
177
178                 guest_buf = (hva_t)kmap_atomic(
179                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
180                 offset = addr & ~PAGE_MASK;
181                 guest_buf |= offset;
182                 now = min(size, PAGE_SIZE - offset);
183                 memcpy((void*)guest_buf, host_buf, now);
184                 host_buf += now;
185                 addr += now;
186                 size -= now;
187                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
188         }
189         return req_size - size;
190 }
191 EXPORT_SYMBOL_GPL(kvm_write_guest);
192
193 static int vcpu_slot(struct kvm_vcpu *vcpu)
194 {
195         return vcpu - vcpu->kvm->vcpus;
196 }
197
198 /*
199  * Switches to specified vcpu, until a matching vcpu_put()
200  */
201 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
202 {
203         struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
204
205         mutex_lock(&vcpu->mutex);
206         if (unlikely(!vcpu->vmcs)) {
207                 mutex_unlock(&vcpu->mutex);
208                 return 0;
209         }
210         return kvm_arch_ops->vcpu_load(vcpu);
211 }
212
213 static void vcpu_put(struct kvm_vcpu *vcpu)
214 {
215         kvm_arch_ops->vcpu_put(vcpu);
216         mutex_unlock(&vcpu->mutex);
217 }
218
219 static int kvm_dev_open(struct inode *inode, struct file *filp)
220 {
221         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
222         int i;
223
224         if (!kvm)
225                 return -ENOMEM;
226
227         spin_lock_init(&kvm->lock);
228         INIT_LIST_HEAD(&kvm->active_mmu_pages);
229         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
230                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
231
232                 mutex_init(&vcpu->mutex);
233                 vcpu->mmu.root_hpa = INVALID_PAGE;
234                 INIT_LIST_HEAD(&vcpu->free_pages);
235         }
236         filp->private_data = kvm;
237         return 0;
238 }
239
240 /*
241  * Free any memory in @free but not in @dont.
242  */
243 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
244                                   struct kvm_memory_slot *dont)
245 {
246         int i;
247
248         if (!dont || free->phys_mem != dont->phys_mem)
249                 if (free->phys_mem) {
250                         for (i = 0; i < free->npages; ++i)
251                                 if (free->phys_mem[i])
252                                         __free_page(free->phys_mem[i]);
253                         vfree(free->phys_mem);
254                 }
255
256         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
257                 vfree(free->dirty_bitmap);
258
259         free->phys_mem = 0;
260         free->npages = 0;
261         free->dirty_bitmap = 0;
262 }
263
264 static void kvm_free_physmem(struct kvm *kvm)
265 {
266         int i;
267
268         for (i = 0; i < kvm->nmemslots; ++i)
269                 kvm_free_physmem_slot(&kvm->memslots[i], 0);
270 }
271
272 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
273 {
274         kvm_mmu_destroy(vcpu);
275         kvm_arch_ops->vcpu_free(vcpu);
276 }
277
278 static void kvm_free_vcpus(struct kvm *kvm)
279 {
280         unsigned int i;
281
282         for (i = 0; i < KVM_MAX_VCPUS; ++i)
283                 kvm_free_vcpu(&kvm->vcpus[i]);
284 }
285
286 static int kvm_dev_release(struct inode *inode, struct file *filp)
287 {
288         struct kvm *kvm = filp->private_data;
289
290         kvm_free_vcpus(kvm);
291         kvm_free_physmem(kvm);
292         kfree(kvm);
293         return 0;
294 }
295
296 static void inject_gp(struct kvm_vcpu *vcpu)
297 {
298         kvm_arch_ops->inject_gp(vcpu, 0);
299 }
300
301 /*
302  * Load the pae pdptrs.  Return true is they are all valid.
303  */
304 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
305 {
306         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
307         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
308         int i;
309         u64 pdpte;
310         u64 *pdpt;
311         int ret;
312         struct kvm_memory_slot *memslot;
313
314         spin_lock(&vcpu->kvm->lock);
315         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
316         /* FIXME: !memslot - emulate? 0xff? */
317         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
318
319         ret = 1;
320         for (i = 0; i < 4; ++i) {
321                 pdpte = pdpt[offset + i];
322                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
323                         ret = 0;
324                         goto out;
325                 }
326         }
327
328         for (i = 0; i < 4; ++i)
329                 vcpu->pdptrs[i] = pdpt[offset + i];
330
331 out:
332         kunmap_atomic(pdpt, KM_USER0);
333         spin_unlock(&vcpu->kvm->lock);
334
335         return ret;
336 }
337
338 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
339 {
340         if (cr0 & CR0_RESEVED_BITS) {
341                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
342                        cr0, vcpu->cr0);
343                 inject_gp(vcpu);
344                 return;
345         }
346
347         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
348                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
349                 inject_gp(vcpu);
350                 return;
351         }
352
353         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
354                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
355                        "and a clear PE flag\n");
356                 inject_gp(vcpu);
357                 return;
358         }
359
360         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
361 #ifdef CONFIG_X86_64
362                 if ((vcpu->shadow_efer & EFER_LME)) {
363                         int cs_db, cs_l;
364
365                         if (!is_pae(vcpu)) {
366                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
367                                        "in long mode while PAE is disabled\n");
368                                 inject_gp(vcpu);
369                                 return;
370                         }
371                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
372                         if (cs_l) {
373                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
374                                        "in long mode while CS.L == 1\n");
375                                 inject_gp(vcpu);
376                                 return;
377
378                         }
379                 } else
380 #endif
381                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
382                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
383                                "reserved bits\n");
384                         inject_gp(vcpu);
385                         return;
386                 }
387
388         }
389
390         kvm_arch_ops->set_cr0(vcpu, cr0);
391         vcpu->cr0 = cr0;
392
393         spin_lock(&vcpu->kvm->lock);
394         kvm_mmu_reset_context(vcpu);
395         spin_unlock(&vcpu->kvm->lock);
396         return;
397 }
398 EXPORT_SYMBOL_GPL(set_cr0);
399
400 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
401 {
402         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
403         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
404 }
405 EXPORT_SYMBOL_GPL(lmsw);
406
407 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
408 {
409         if (cr4 & CR4_RESEVED_BITS) {
410                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
411                 inject_gp(vcpu);
412                 return;
413         }
414
415         if (is_long_mode(vcpu)) {
416                 if (!(cr4 & CR4_PAE_MASK)) {
417                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
418                                "in long mode\n");
419                         inject_gp(vcpu);
420                         return;
421                 }
422         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
423                    && !load_pdptrs(vcpu, vcpu->cr3)) {
424                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
425                 inject_gp(vcpu);
426         }
427
428         if (cr4 & CR4_VMXE_MASK) {
429                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
430                 inject_gp(vcpu);
431                 return;
432         }
433         kvm_arch_ops->set_cr4(vcpu, cr4);
434         spin_lock(&vcpu->kvm->lock);
435         kvm_mmu_reset_context(vcpu);
436         spin_unlock(&vcpu->kvm->lock);
437 }
438 EXPORT_SYMBOL_GPL(set_cr4);
439
440 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
441 {
442         if (is_long_mode(vcpu)) {
443                 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
444                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
445                         inject_gp(vcpu);
446                         return;
447                 }
448         } else {
449                 if (cr3 & CR3_RESEVED_BITS) {
450                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
451                         inject_gp(vcpu);
452                         return;
453                 }
454                 if (is_paging(vcpu) && is_pae(vcpu) &&
455                     !load_pdptrs(vcpu, cr3)) {
456                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
457                                "reserved bits\n");
458                         inject_gp(vcpu);
459                         return;
460                 }
461         }
462
463         vcpu->cr3 = cr3;
464         spin_lock(&vcpu->kvm->lock);
465         vcpu->mmu.new_cr3(vcpu);
466         spin_unlock(&vcpu->kvm->lock);
467 }
468 EXPORT_SYMBOL_GPL(set_cr3);
469
470 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
471 {
472         if ( cr8 & CR8_RESEVED_BITS) {
473                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
474                 inject_gp(vcpu);
475                 return;
476         }
477         vcpu->cr8 = cr8;
478 }
479 EXPORT_SYMBOL_GPL(set_cr8);
480
481 void fx_init(struct kvm_vcpu *vcpu)
482 {
483         struct __attribute__ ((__packed__)) fx_image_s {
484                 u16 control; //fcw
485                 u16 status; //fsw
486                 u16 tag; // ftw
487                 u16 opcode; //fop
488                 u64 ip; // fpu ip
489                 u64 operand;// fpu dp
490                 u32 mxcsr;
491                 u32 mxcsr_mask;
492
493         } *fx_image;
494
495         fx_save(vcpu->host_fx_image);
496         fpu_init();
497         fx_save(vcpu->guest_fx_image);
498         fx_restore(vcpu->host_fx_image);
499
500         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
501         fx_image->mxcsr = 0x1f80;
502         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
503                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
504 }
505 EXPORT_SYMBOL_GPL(fx_init);
506
507 /*
508  * Creates some virtual cpus.  Good luck creating more than one.
509  */
510 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
511 {
512         int r;
513         struct kvm_vcpu *vcpu;
514
515         r = -EINVAL;
516         if (!valid_vcpu(n))
517                 goto out;
518
519         vcpu = &kvm->vcpus[n];
520
521         mutex_lock(&vcpu->mutex);
522
523         if (vcpu->vmcs) {
524                 mutex_unlock(&vcpu->mutex);
525                 return -EEXIST;
526         }
527
528         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
529                                            FX_IMAGE_ALIGN);
530         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
531
532         vcpu->cpu = -1;  /* First load will set up TR */
533         vcpu->kvm = kvm;
534         r = kvm_arch_ops->vcpu_create(vcpu);
535         if (r < 0)
536                 goto out_free_vcpus;
537
538         r = kvm_mmu_create(vcpu);
539         if (r < 0)
540                 goto out_free_vcpus;
541
542         kvm_arch_ops->vcpu_load(vcpu);
543         r = kvm_mmu_setup(vcpu);
544         if (r >= 0)
545                 r = kvm_arch_ops->vcpu_setup(vcpu);
546         vcpu_put(vcpu);
547
548         if (r < 0)
549                 goto out_free_vcpus;
550
551         return 0;
552
553 out_free_vcpus:
554         kvm_free_vcpu(vcpu);
555         mutex_unlock(&vcpu->mutex);
556 out:
557         return r;
558 }
559
560 /*
561  * Allocate some memory and give it an address in the guest physical address
562  * space.
563  *
564  * Discontiguous memory is allowed, mostly for framebuffers.
565  */
566 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
567                                            struct kvm_memory_region *mem)
568 {
569         int r;
570         gfn_t base_gfn;
571         unsigned long npages;
572         unsigned long i;
573         struct kvm_memory_slot *memslot;
574         struct kvm_memory_slot old, new;
575         int memory_config_version;
576
577         r = -EINVAL;
578         /* General sanity checks */
579         if (mem->memory_size & (PAGE_SIZE - 1))
580                 goto out;
581         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
582                 goto out;
583         if (mem->slot >= KVM_MEMORY_SLOTS)
584                 goto out;
585         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
586                 goto out;
587
588         memslot = &kvm->memslots[mem->slot];
589         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
590         npages = mem->memory_size >> PAGE_SHIFT;
591
592         if (!npages)
593                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
594
595 raced:
596         spin_lock(&kvm->lock);
597
598         memory_config_version = kvm->memory_config_version;
599         new = old = *memslot;
600
601         new.base_gfn = base_gfn;
602         new.npages = npages;
603         new.flags = mem->flags;
604
605         /* Disallow changing a memory slot's size. */
606         r = -EINVAL;
607         if (npages && old.npages && npages != old.npages)
608                 goto out_unlock;
609
610         /* Check for overlaps */
611         r = -EEXIST;
612         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
613                 struct kvm_memory_slot *s = &kvm->memslots[i];
614
615                 if (s == memslot)
616                         continue;
617                 if (!((base_gfn + npages <= s->base_gfn) ||
618                       (base_gfn >= s->base_gfn + s->npages)))
619                         goto out_unlock;
620         }
621         /*
622          * Do memory allocations outside lock.  memory_config_version will
623          * detect any races.
624          */
625         spin_unlock(&kvm->lock);
626
627         /* Deallocate if slot is being removed */
628         if (!npages)
629                 new.phys_mem = 0;
630
631         /* Free page dirty bitmap if unneeded */
632         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
633                 new.dirty_bitmap = 0;
634
635         r = -ENOMEM;
636
637         /* Allocate if a slot is being created */
638         if (npages && !new.phys_mem) {
639                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
640
641                 if (!new.phys_mem)
642                         goto out_free;
643
644                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
645                 for (i = 0; i < npages; ++i) {
646                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
647                                                      | __GFP_ZERO);
648                         if (!new.phys_mem[i])
649                                 goto out_free;
650                         new.phys_mem[i]->private = 0;
651                 }
652         }
653
654         /* Allocate page dirty bitmap if needed */
655         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
656                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
657
658                 new.dirty_bitmap = vmalloc(dirty_bytes);
659                 if (!new.dirty_bitmap)
660                         goto out_free;
661                 memset(new.dirty_bitmap, 0, dirty_bytes);
662         }
663
664         spin_lock(&kvm->lock);
665
666         if (memory_config_version != kvm->memory_config_version) {
667                 spin_unlock(&kvm->lock);
668                 kvm_free_physmem_slot(&new, &old);
669                 goto raced;
670         }
671
672         r = -EAGAIN;
673         if (kvm->busy)
674                 goto out_unlock;
675
676         if (mem->slot >= kvm->nmemslots)
677                 kvm->nmemslots = mem->slot + 1;
678
679         *memslot = new;
680         ++kvm->memory_config_version;
681
682         spin_unlock(&kvm->lock);
683
684         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
685                 struct kvm_vcpu *vcpu;
686
687                 vcpu = vcpu_load(kvm, i);
688                 if (!vcpu)
689                         continue;
690                 kvm_mmu_reset_context(vcpu);
691                 vcpu_put(vcpu);
692         }
693
694         kvm_free_physmem_slot(&old, &new);
695         return 0;
696
697 out_unlock:
698         spin_unlock(&kvm->lock);
699 out_free:
700         kvm_free_physmem_slot(&new, &old);
701 out:
702         return r;
703 }
704
705 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
706 {
707         spin_lock(&vcpu->kvm->lock);
708         kvm_mmu_slot_remove_write_access(vcpu, slot);
709         spin_unlock(&vcpu->kvm->lock);
710 }
711
712 /*
713  * Get (and clear) the dirty memory log for a memory slot.
714  */
715 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
716                                        struct kvm_dirty_log *log)
717 {
718         struct kvm_memory_slot *memslot;
719         int r, i;
720         int n;
721         int cleared;
722         unsigned long any = 0;
723
724         spin_lock(&kvm->lock);
725
726         /*
727          * Prevent changes to guest memory configuration even while the lock
728          * is not taken.
729          */
730         ++kvm->busy;
731         spin_unlock(&kvm->lock);
732         r = -EINVAL;
733         if (log->slot >= KVM_MEMORY_SLOTS)
734                 goto out;
735
736         memslot = &kvm->memslots[log->slot];
737         r = -ENOENT;
738         if (!memslot->dirty_bitmap)
739                 goto out;
740
741         n = ALIGN(memslot->npages, 8) / 8;
742
743         for (i = 0; !any && i < n; ++i)
744                 any = memslot->dirty_bitmap[i];
745
746         r = -EFAULT;
747         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
748                 goto out;
749
750
751         if (any) {
752                 cleared = 0;
753                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
754                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
755
756                         if (!vcpu)
757                                 continue;
758                         if (!cleared) {
759                                 do_remove_write_access(vcpu, log->slot);
760                                 memset(memslot->dirty_bitmap, 0, n);
761                                 cleared = 1;
762                         }
763                         kvm_arch_ops->tlb_flush(vcpu);
764                         vcpu_put(vcpu);
765                 }
766         }
767
768         r = 0;
769
770 out:
771         spin_lock(&kvm->lock);
772         --kvm->busy;
773         spin_unlock(&kvm->lock);
774         return r;
775 }
776
777 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
778 {
779         int i;
780
781         for (i = 0; i < kvm->nmemslots; ++i) {
782                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
783
784                 if (gfn >= memslot->base_gfn
785                     && gfn < memslot->base_gfn + memslot->npages)
786                         return memslot;
787         }
788         return 0;
789 }
790 EXPORT_SYMBOL_GPL(gfn_to_memslot);
791
792 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
793 {
794         int i;
795         struct kvm_memory_slot *memslot = 0;
796         unsigned long rel_gfn;
797
798         for (i = 0; i < kvm->nmemslots; ++i) {
799                 memslot = &kvm->memslots[i];
800
801                 if (gfn >= memslot->base_gfn
802                     && gfn < memslot->base_gfn + memslot->npages) {
803
804                         if (!memslot || !memslot->dirty_bitmap)
805                                 return;
806
807                         rel_gfn = gfn - memslot->base_gfn;
808
809                         /* avoid RMW */
810                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
811                                 set_bit(rel_gfn, memslot->dirty_bitmap);
812                         return;
813                 }
814         }
815 }
816
817 static int emulator_read_std(unsigned long addr,
818                              unsigned long *val,
819                              unsigned int bytes,
820                              struct x86_emulate_ctxt *ctxt)
821 {
822         struct kvm_vcpu *vcpu = ctxt->vcpu;
823         void *data = val;
824
825         while (bytes) {
826                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
827                 unsigned offset = addr & (PAGE_SIZE-1);
828                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
829                 unsigned long pfn;
830                 struct kvm_memory_slot *memslot;
831                 void *page;
832
833                 if (gpa == UNMAPPED_GVA)
834                         return X86EMUL_PROPAGATE_FAULT;
835                 pfn = gpa >> PAGE_SHIFT;
836                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
837                 if (!memslot)
838                         return X86EMUL_UNHANDLEABLE;
839                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
840
841                 memcpy(data, page + offset, tocopy);
842
843                 kunmap_atomic(page, KM_USER0);
844
845                 bytes -= tocopy;
846                 data += tocopy;
847                 addr += tocopy;
848         }
849
850         return X86EMUL_CONTINUE;
851 }
852
853 static int emulator_write_std(unsigned long addr,
854                               unsigned long val,
855                               unsigned int bytes,
856                               struct x86_emulate_ctxt *ctxt)
857 {
858         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
859                addr, bytes);
860         return X86EMUL_UNHANDLEABLE;
861 }
862
863 static int emulator_read_emulated(unsigned long addr,
864                                   unsigned long *val,
865                                   unsigned int bytes,
866                                   struct x86_emulate_ctxt *ctxt)
867 {
868         struct kvm_vcpu *vcpu = ctxt->vcpu;
869
870         if (vcpu->mmio_read_completed) {
871                 memcpy(val, vcpu->mmio_data, bytes);
872                 vcpu->mmio_read_completed = 0;
873                 return X86EMUL_CONTINUE;
874         } else if (emulator_read_std(addr, val, bytes, ctxt)
875                    == X86EMUL_CONTINUE)
876                 return X86EMUL_CONTINUE;
877         else {
878                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
879                 if (gpa == UNMAPPED_GVA)
880                         return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
881                 vcpu->mmio_needed = 1;
882                 vcpu->mmio_phys_addr = gpa;
883                 vcpu->mmio_size = bytes;
884                 vcpu->mmio_is_write = 0;
885
886                 return X86EMUL_UNHANDLEABLE;
887         }
888 }
889
890 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
891                                unsigned long val, int bytes)
892 {
893         struct kvm_memory_slot *m;
894         struct page *page;
895         void *virt;
896
897         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
898                 return 0;
899         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
900         if (!m)
901                 return 0;
902         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
903         kvm_mmu_pre_write(vcpu, gpa, bytes);
904         virt = kmap_atomic(page, KM_USER0);
905         memcpy(virt + offset_in_page(gpa), &val, bytes);
906         kunmap_atomic(virt, KM_USER0);
907         kvm_mmu_post_write(vcpu, gpa, bytes);
908         return 1;
909 }
910
911 static int emulator_write_emulated(unsigned long addr,
912                                    unsigned long val,
913                                    unsigned int bytes,
914                                    struct x86_emulate_ctxt *ctxt)
915 {
916         struct kvm_vcpu *vcpu = ctxt->vcpu;
917         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
918
919         if (gpa == UNMAPPED_GVA)
920                 return X86EMUL_PROPAGATE_FAULT;
921
922         if (emulator_write_phys(vcpu, gpa, val, bytes))
923                 return X86EMUL_CONTINUE;
924
925         vcpu->mmio_needed = 1;
926         vcpu->mmio_phys_addr = gpa;
927         vcpu->mmio_size = bytes;
928         vcpu->mmio_is_write = 1;
929         memcpy(vcpu->mmio_data, &val, bytes);
930
931         return X86EMUL_CONTINUE;
932 }
933
934 static int emulator_cmpxchg_emulated(unsigned long addr,
935                                      unsigned long old,
936                                      unsigned long new,
937                                      unsigned int bytes,
938                                      struct x86_emulate_ctxt *ctxt)
939 {
940         static int reported;
941
942         if (!reported) {
943                 reported = 1;
944                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
945         }
946         return emulator_write_emulated(addr, new, bytes, ctxt);
947 }
948
949 #ifdef CONFIG_X86_32
950
951 static int emulator_cmpxchg8b_emulated(unsigned long addr,
952                                        unsigned long old_lo,
953                                        unsigned long old_hi,
954                                        unsigned long new_lo,
955                                        unsigned long new_hi,
956                                        struct x86_emulate_ctxt *ctxt)
957 {
958         static int reported;
959         int r;
960
961         if (!reported) {
962                 reported = 1;
963                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
964         }
965         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
966         if (r != X86EMUL_CONTINUE)
967                 return r;
968         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
969 }
970
971 #endif
972
973 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
974 {
975         return kvm_arch_ops->get_segment_base(vcpu, seg);
976 }
977
978 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
979 {
980         return X86EMUL_CONTINUE;
981 }
982
983 int emulate_clts(struct kvm_vcpu *vcpu)
984 {
985         unsigned long cr0;
986
987         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
988         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
989         kvm_arch_ops->set_cr0(vcpu, cr0);
990         return X86EMUL_CONTINUE;
991 }
992
993 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
994 {
995         struct kvm_vcpu *vcpu = ctxt->vcpu;
996
997         switch (dr) {
998         case 0 ... 3:
999                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1000                 return X86EMUL_CONTINUE;
1001         default:
1002                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1003                        __FUNCTION__, dr);
1004                 return X86EMUL_UNHANDLEABLE;
1005         }
1006 }
1007
1008 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1009 {
1010         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1011         int exception;
1012
1013         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1014         if (exception) {
1015                 /* FIXME: better handling */
1016                 return X86EMUL_UNHANDLEABLE;
1017         }
1018         return X86EMUL_CONTINUE;
1019 }
1020
1021 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1022 {
1023         static int reported;
1024         u8 opcodes[4];
1025         unsigned long rip = ctxt->vcpu->rip;
1026         unsigned long rip_linear;
1027
1028         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1029
1030         if (reported)
1031                 return;
1032
1033         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1034
1035         printk(KERN_ERR "emulation failed but !mmio_needed?"
1036                " rip %lx %02x %02x %02x %02x\n",
1037                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1038         reported = 1;
1039 }
1040
1041 struct x86_emulate_ops emulate_ops = {
1042         .read_std            = emulator_read_std,
1043         .write_std           = emulator_write_std,
1044         .read_emulated       = emulator_read_emulated,
1045         .write_emulated      = emulator_write_emulated,
1046         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1047 #ifdef CONFIG_X86_32
1048         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1049 #endif
1050 };
1051
1052 int emulate_instruction(struct kvm_vcpu *vcpu,
1053                         struct kvm_run *run,
1054                         unsigned long cr2,
1055                         u16 error_code)
1056 {
1057         struct x86_emulate_ctxt emulate_ctxt;
1058         int r;
1059         int cs_db, cs_l;
1060
1061         kvm_arch_ops->cache_regs(vcpu);
1062
1063         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1064
1065         emulate_ctxt.vcpu = vcpu;
1066         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1067         emulate_ctxt.cr2 = cr2;
1068         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1069                 ? X86EMUL_MODE_REAL : cs_l
1070                 ? X86EMUL_MODE_PROT64 : cs_db
1071                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1072
1073         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1074                 emulate_ctxt.cs_base = 0;
1075                 emulate_ctxt.ds_base = 0;
1076                 emulate_ctxt.es_base = 0;
1077                 emulate_ctxt.ss_base = 0;
1078         } else {
1079                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1080                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1081                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1082                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1083         }
1084
1085         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1086         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1087
1088         vcpu->mmio_is_write = 0;
1089         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1090
1091         if ((r || vcpu->mmio_is_write) && run) {
1092                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1093                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1094                 run->mmio.len = vcpu->mmio_size;
1095                 run->mmio.is_write = vcpu->mmio_is_write;
1096         }
1097
1098         if (r) {
1099                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1100                         return EMULATE_DONE;
1101                 if (!vcpu->mmio_needed) {
1102                         report_emulation_failure(&emulate_ctxt);
1103                         return EMULATE_FAIL;
1104                 }
1105                 return EMULATE_DO_MMIO;
1106         }
1107
1108         kvm_arch_ops->decache_regs(vcpu);
1109         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1110
1111         if (vcpu->mmio_is_write)
1112                 return EMULATE_DO_MMIO;
1113
1114         return EMULATE_DONE;
1115 }
1116 EXPORT_SYMBOL_GPL(emulate_instruction);
1117
1118 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1119 {
1120         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1121 }
1122
1123 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1124 {
1125         struct descriptor_table dt = { limit, base };
1126
1127         kvm_arch_ops->set_gdt(vcpu, &dt);
1128 }
1129
1130 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1131 {
1132         struct descriptor_table dt = { limit, base };
1133
1134         kvm_arch_ops->set_idt(vcpu, &dt);
1135 }
1136
1137 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1138                    unsigned long *rflags)
1139 {
1140         lmsw(vcpu, msw);
1141         *rflags = kvm_arch_ops->get_rflags(vcpu);
1142 }
1143
1144 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1145 {
1146         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1147         switch (cr) {
1148         case 0:
1149                 return vcpu->cr0;
1150         case 2:
1151                 return vcpu->cr2;
1152         case 3:
1153                 return vcpu->cr3;
1154         case 4:
1155                 return vcpu->cr4;
1156         default:
1157                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1158                 return 0;
1159         }
1160 }
1161
1162 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1163                      unsigned long *rflags)
1164 {
1165         switch (cr) {
1166         case 0:
1167                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1168                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1169                 break;
1170         case 2:
1171                 vcpu->cr2 = val;
1172                 break;
1173         case 3:
1174                 set_cr3(vcpu, val);
1175                 break;
1176         case 4:
1177                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1178                 break;
1179         default:
1180                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1181         }
1182 }
1183
1184 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1185 {
1186         u64 data;
1187
1188         switch (msr) {
1189         case 0xc0010010: /* SYSCFG */
1190         case 0xc0010015: /* HWCR */
1191         case MSR_IA32_PLATFORM_ID:
1192         case MSR_IA32_P5_MC_ADDR:
1193         case MSR_IA32_P5_MC_TYPE:
1194         case MSR_IA32_MC0_CTL:
1195         case MSR_IA32_MCG_STATUS:
1196         case MSR_IA32_MCG_CAP:
1197         case MSR_IA32_MC0_MISC:
1198         case MSR_IA32_MC0_MISC+4:
1199         case MSR_IA32_MC0_MISC+8:
1200         case MSR_IA32_MC0_MISC+12:
1201         case MSR_IA32_MC0_MISC+16:
1202         case MSR_IA32_UCODE_REV:
1203         case MSR_IA32_PERF_STATUS:
1204                 /* MTRR registers */
1205         case 0xfe:
1206         case 0x200 ... 0x2ff:
1207                 data = 0;
1208                 break;
1209         case 0xcd: /* fsb frequency */
1210                 data = 3;
1211                 break;
1212         case MSR_IA32_APICBASE:
1213                 data = vcpu->apic_base;
1214                 break;
1215 #ifdef CONFIG_X86_64
1216         case MSR_EFER:
1217                 data = vcpu->shadow_efer;
1218                 break;
1219 #endif
1220         default:
1221                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1222                 return 1;
1223         }
1224         *pdata = data;
1225         return 0;
1226 }
1227 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1228
1229 /*
1230  * Reads an msr value (of 'msr_index') into 'pdata'.
1231  * Returns 0 on success, non-0 otherwise.
1232  * Assumes vcpu_load() was already called.
1233  */
1234 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1235 {
1236         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1237 }
1238
1239 #ifdef CONFIG_X86_64
1240
1241 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1242 {
1243         if (efer & EFER_RESERVED_BITS) {
1244                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1245                        efer);
1246                 inject_gp(vcpu);
1247                 return;
1248         }
1249
1250         if (is_paging(vcpu)
1251             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1252                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1253                 inject_gp(vcpu);
1254                 return;
1255         }
1256
1257         kvm_arch_ops->set_efer(vcpu, efer);
1258
1259         efer &= ~EFER_LMA;
1260         efer |= vcpu->shadow_efer & EFER_LMA;
1261
1262         vcpu->shadow_efer = efer;
1263 }
1264
1265 #endif
1266
1267 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1268 {
1269         switch (msr) {
1270 #ifdef CONFIG_X86_64
1271         case MSR_EFER:
1272                 set_efer(vcpu, data);
1273                 break;
1274 #endif
1275         case MSR_IA32_MC0_STATUS:
1276                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1277                        __FUNCTION__, data);
1278                 break;
1279         case MSR_IA32_UCODE_REV:
1280         case MSR_IA32_UCODE_WRITE:
1281         case 0x200 ... 0x2ff: /* MTRRs */
1282                 break;
1283         case MSR_IA32_APICBASE:
1284                 vcpu->apic_base = data;
1285                 break;
1286         default:
1287                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1288                 return 1;
1289         }
1290         return 0;
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1293
1294 /*
1295  * Writes msr value into into the appropriate "register".
1296  * Returns 0 on success, non-0 otherwise.
1297  * Assumes vcpu_load() was already called.
1298  */
1299 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1300 {
1301         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1302 }
1303
1304 void kvm_resched(struct kvm_vcpu *vcpu)
1305 {
1306         vcpu_put(vcpu);
1307         cond_resched();
1308         /* Cannot fail -  no vcpu unplug yet. */
1309         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1310 }
1311 EXPORT_SYMBOL_GPL(kvm_resched);
1312
1313 void load_msrs(struct vmx_msr_entry *e, int n)
1314 {
1315         int i;
1316
1317         for (i = 0; i < n; ++i)
1318                 wrmsrl(e[i].index, e[i].data);
1319 }
1320 EXPORT_SYMBOL_GPL(load_msrs);
1321
1322 void save_msrs(struct vmx_msr_entry *e, int n)
1323 {
1324         int i;
1325
1326         for (i = 0; i < n; ++i)
1327                 rdmsrl(e[i].index, e[i].data);
1328 }
1329 EXPORT_SYMBOL_GPL(save_msrs);
1330
1331 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1332 {
1333         struct kvm_vcpu *vcpu;
1334         int r;
1335
1336         if (!valid_vcpu(kvm_run->vcpu))
1337                 return -EINVAL;
1338
1339         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1340         if (!vcpu)
1341                 return -ENOENT;
1342
1343         if (kvm_run->emulated) {
1344                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1345                 kvm_run->emulated = 0;
1346         }
1347
1348         if (kvm_run->mmio_completed) {
1349                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1350                 vcpu->mmio_read_completed = 1;
1351         }
1352
1353         vcpu->mmio_needed = 0;
1354
1355         r = kvm_arch_ops->run(vcpu, kvm_run);
1356
1357         vcpu_put(vcpu);
1358         return r;
1359 }
1360
1361 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1362 {
1363         struct kvm_vcpu *vcpu;
1364
1365         if (!valid_vcpu(regs->vcpu))
1366                 return -EINVAL;
1367
1368         vcpu = vcpu_load(kvm, regs->vcpu);
1369         if (!vcpu)
1370                 return -ENOENT;
1371
1372         kvm_arch_ops->cache_regs(vcpu);
1373
1374         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1375         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1376         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1377         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1378         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1379         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1380         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1381         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1382 #ifdef CONFIG_X86_64
1383         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1384         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1385         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1386         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1387         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1388         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1389         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1390         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1391 #endif
1392
1393         regs->rip = vcpu->rip;
1394         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1395
1396         /*
1397          * Don't leak debug flags in case they were set for guest debugging
1398          */
1399         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1400                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1401
1402         vcpu_put(vcpu);
1403
1404         return 0;
1405 }
1406
1407 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1408 {
1409         struct kvm_vcpu *vcpu;
1410
1411         if (!valid_vcpu(regs->vcpu))
1412                 return -EINVAL;
1413
1414         vcpu = vcpu_load(kvm, regs->vcpu);
1415         if (!vcpu)
1416                 return -ENOENT;
1417
1418         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1419         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1420         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1421         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1422         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1423         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1424         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1425         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1426 #ifdef CONFIG_X86_64
1427         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1428         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1429         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1430         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1431         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1432         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1433         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1434         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1435 #endif
1436
1437         vcpu->rip = regs->rip;
1438         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1439
1440         kvm_arch_ops->decache_regs(vcpu);
1441
1442         vcpu_put(vcpu);
1443
1444         return 0;
1445 }
1446
1447 static void get_segment(struct kvm_vcpu *vcpu,
1448                         struct kvm_segment *var, int seg)
1449 {
1450         return kvm_arch_ops->get_segment(vcpu, var, seg);
1451 }
1452
1453 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1454 {
1455         struct kvm_vcpu *vcpu;
1456         struct descriptor_table dt;
1457
1458         if (!valid_vcpu(sregs->vcpu))
1459                 return -EINVAL;
1460         vcpu = vcpu_load(kvm, sregs->vcpu);
1461         if (!vcpu)
1462                 return -ENOENT;
1463
1464         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1465         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1466         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1467         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1468         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1469         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1470
1471         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1472         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1473
1474         kvm_arch_ops->get_idt(vcpu, &dt);
1475         sregs->idt.limit = dt.limit;
1476         sregs->idt.base = dt.base;
1477         kvm_arch_ops->get_gdt(vcpu, &dt);
1478         sregs->gdt.limit = dt.limit;
1479         sregs->gdt.base = dt.base;
1480
1481         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1482         sregs->cr0 = vcpu->cr0;
1483         sregs->cr2 = vcpu->cr2;
1484         sregs->cr3 = vcpu->cr3;
1485         sregs->cr4 = vcpu->cr4;
1486         sregs->cr8 = vcpu->cr8;
1487         sregs->efer = vcpu->shadow_efer;
1488         sregs->apic_base = vcpu->apic_base;
1489
1490         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1491                sizeof sregs->interrupt_bitmap);
1492
1493         vcpu_put(vcpu);
1494
1495         return 0;
1496 }
1497
1498 static void set_segment(struct kvm_vcpu *vcpu,
1499                         struct kvm_segment *var, int seg)
1500 {
1501         return kvm_arch_ops->set_segment(vcpu, var, seg);
1502 }
1503
1504 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1505 {
1506         struct kvm_vcpu *vcpu;
1507         int mmu_reset_needed = 0;
1508         int i;
1509         struct descriptor_table dt;
1510
1511         if (!valid_vcpu(sregs->vcpu))
1512                 return -EINVAL;
1513         vcpu = vcpu_load(kvm, sregs->vcpu);
1514         if (!vcpu)
1515                 return -ENOENT;
1516
1517         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1518         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1519         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1520         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1521         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1522         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1523
1524         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1525         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1526
1527         dt.limit = sregs->idt.limit;
1528         dt.base = sregs->idt.base;
1529         kvm_arch_ops->set_idt(vcpu, &dt);
1530         dt.limit = sregs->gdt.limit;
1531         dt.base = sregs->gdt.base;
1532         kvm_arch_ops->set_gdt(vcpu, &dt);
1533
1534         vcpu->cr2 = sregs->cr2;
1535         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1536         vcpu->cr3 = sregs->cr3;
1537
1538         vcpu->cr8 = sregs->cr8;
1539
1540         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1541 #ifdef CONFIG_X86_64
1542         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1543 #endif
1544         vcpu->apic_base = sregs->apic_base;
1545
1546         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1547
1548         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1549         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1550
1551         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1552         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1553         if (!is_long_mode(vcpu) && is_pae(vcpu))
1554                 load_pdptrs(vcpu, vcpu->cr3);
1555
1556         if (mmu_reset_needed)
1557                 kvm_mmu_reset_context(vcpu);
1558
1559         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1560                sizeof vcpu->irq_pending);
1561         vcpu->irq_summary = 0;
1562         for (i = 0; i < NR_IRQ_WORDS; ++i)
1563                 if (vcpu->irq_pending[i])
1564                         __set_bit(i, &vcpu->irq_summary);
1565
1566         vcpu_put(vcpu);
1567
1568         return 0;
1569 }
1570
1571 /*
1572  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1573  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1574  *
1575  * This list is modified at module load time to reflect the
1576  * capabilities of the host cpu.
1577  */
1578 static u32 msrs_to_save[] = {
1579         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1580         MSR_K6_STAR,
1581 #ifdef CONFIG_X86_64
1582         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1583 #endif
1584         MSR_IA32_TIME_STAMP_COUNTER,
1585 };
1586
1587 static unsigned num_msrs_to_save;
1588
1589 static __init void kvm_init_msr_list(void)
1590 {
1591         u32 dummy[2];
1592         unsigned i, j;
1593
1594         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1595                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1596                         continue;
1597                 if (j < i)
1598                         msrs_to_save[j] = msrs_to_save[i];
1599                 j++;
1600         }
1601         num_msrs_to_save = j;
1602 }
1603
1604 /*
1605  * Adapt set_msr() to msr_io()'s calling convention
1606  */
1607 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1608 {
1609         return set_msr(vcpu, index, *data);
1610 }
1611
1612 /*
1613  * Read or write a bunch of msrs. All parameters are kernel addresses.
1614  *
1615  * @return number of msrs set successfully.
1616  */
1617 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1618                     struct kvm_msr_entry *entries,
1619                     int (*do_msr)(struct kvm_vcpu *vcpu,
1620                                   unsigned index, u64 *data))
1621 {
1622         struct kvm_vcpu *vcpu;
1623         int i;
1624
1625         if (!valid_vcpu(msrs->vcpu))
1626                 return -EINVAL;
1627
1628         vcpu = vcpu_load(kvm, msrs->vcpu);
1629         if (!vcpu)
1630                 return -ENOENT;
1631
1632         for (i = 0; i < msrs->nmsrs; ++i)
1633                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1634                         break;
1635
1636         vcpu_put(vcpu);
1637
1638         return i;
1639 }
1640
1641 /*
1642  * Read or write a bunch of msrs. Parameters are user addresses.
1643  *
1644  * @return number of msrs set successfully.
1645  */
1646 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1647                   int (*do_msr)(struct kvm_vcpu *vcpu,
1648                                 unsigned index, u64 *data),
1649                   int writeback)
1650 {
1651         struct kvm_msrs msrs;
1652         struct kvm_msr_entry *entries;
1653         int r, n;
1654         unsigned size;
1655
1656         r = -EFAULT;
1657         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1658                 goto out;
1659
1660         r = -E2BIG;
1661         if (msrs.nmsrs >= MAX_IO_MSRS)
1662                 goto out;
1663
1664         r = -ENOMEM;
1665         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1666         entries = vmalloc(size);
1667         if (!entries)
1668                 goto out;
1669
1670         r = -EFAULT;
1671         if (copy_from_user(entries, user_msrs->entries, size))
1672                 goto out_free;
1673
1674         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1675         if (r < 0)
1676                 goto out_free;
1677
1678         r = -EFAULT;
1679         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1680                 goto out_free;
1681
1682         r = n;
1683
1684 out_free:
1685         vfree(entries);
1686 out:
1687         return r;
1688 }
1689
1690 /*
1691  * Translate a guest virtual address to a guest physical address.
1692  */
1693 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1694 {
1695         unsigned long vaddr = tr->linear_address;
1696         struct kvm_vcpu *vcpu;
1697         gpa_t gpa;
1698
1699         vcpu = vcpu_load(kvm, tr->vcpu);
1700         if (!vcpu)
1701                 return -ENOENT;
1702         spin_lock(&kvm->lock);
1703         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1704         tr->physical_address = gpa;
1705         tr->valid = gpa != UNMAPPED_GVA;
1706         tr->writeable = 1;
1707         tr->usermode = 0;
1708         spin_unlock(&kvm->lock);
1709         vcpu_put(vcpu);
1710
1711         return 0;
1712 }
1713
1714 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1715 {
1716         struct kvm_vcpu *vcpu;
1717
1718         if (!valid_vcpu(irq->vcpu))
1719                 return -EINVAL;
1720         if (irq->irq < 0 || irq->irq >= 256)
1721                 return -EINVAL;
1722         vcpu = vcpu_load(kvm, irq->vcpu);
1723         if (!vcpu)
1724                 return -ENOENT;
1725
1726         set_bit(irq->irq, vcpu->irq_pending);
1727         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1728
1729         vcpu_put(vcpu);
1730
1731         return 0;
1732 }
1733
1734 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1735                                      struct kvm_debug_guest *dbg)
1736 {
1737         struct kvm_vcpu *vcpu;
1738         int r;
1739
1740         if (!valid_vcpu(dbg->vcpu))
1741                 return -EINVAL;
1742         vcpu = vcpu_load(kvm, dbg->vcpu);
1743         if (!vcpu)
1744                 return -ENOENT;
1745
1746         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1747
1748         vcpu_put(vcpu);
1749
1750         return r;
1751 }
1752
1753 static long kvm_dev_ioctl(struct file *filp,
1754                           unsigned int ioctl, unsigned long arg)
1755 {
1756         struct kvm *kvm = filp->private_data;
1757         int r = -EINVAL;
1758
1759         switch (ioctl) {
1760         case KVM_GET_API_VERSION:
1761                 r = KVM_API_VERSION;
1762                 break;
1763         case KVM_CREATE_VCPU: {
1764                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1765                 if (r)
1766                         goto out;
1767                 break;
1768         }
1769         case KVM_RUN: {
1770                 struct kvm_run kvm_run;
1771
1772                 r = -EFAULT;
1773                 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1774                         goto out;
1775                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1776                 if (r < 0 &&  r != -EINTR)
1777                         goto out;
1778                 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run)) {
1779                         r = -EFAULT;
1780                         goto out;
1781                 }
1782                 break;
1783         }
1784         case KVM_GET_REGS: {
1785                 struct kvm_regs kvm_regs;
1786
1787                 r = -EFAULT;
1788                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1789                         goto out;
1790                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1791                 if (r)
1792                         goto out;
1793                 r = -EFAULT;
1794                 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1795                         goto out;
1796                 r = 0;
1797                 break;
1798         }
1799         case KVM_SET_REGS: {
1800                 struct kvm_regs kvm_regs;
1801
1802                 r = -EFAULT;
1803                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1804                         goto out;
1805                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1806                 if (r)
1807                         goto out;
1808                 r = 0;
1809                 break;
1810         }
1811         case KVM_GET_SREGS: {
1812                 struct kvm_sregs kvm_sregs;
1813
1814                 r = -EFAULT;
1815                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1816                         goto out;
1817                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1818                 if (r)
1819                         goto out;
1820                 r = -EFAULT;
1821                 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1822                         goto out;
1823                 r = 0;
1824                 break;
1825         }
1826         case KVM_SET_SREGS: {
1827                 struct kvm_sregs kvm_sregs;
1828
1829                 r = -EFAULT;
1830                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1831                         goto out;
1832                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1833                 if (r)
1834                         goto out;
1835                 r = 0;
1836                 break;
1837         }
1838         case KVM_TRANSLATE: {
1839                 struct kvm_translation tr;
1840
1841                 r = -EFAULT;
1842                 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1843                         goto out;
1844                 r = kvm_dev_ioctl_translate(kvm, &tr);
1845                 if (r)
1846                         goto out;
1847                 r = -EFAULT;
1848                 if (copy_to_user((void *)arg, &tr, sizeof tr))
1849                         goto out;
1850                 r = 0;
1851                 break;
1852         }
1853         case KVM_INTERRUPT: {
1854                 struct kvm_interrupt irq;
1855
1856                 r = -EFAULT;
1857                 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1858                         goto out;
1859                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1860                 if (r)
1861                         goto out;
1862                 r = 0;
1863                 break;
1864         }
1865         case KVM_DEBUG_GUEST: {
1866                 struct kvm_debug_guest dbg;
1867
1868                 r = -EFAULT;
1869                 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1870                         goto out;
1871                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1872                 if (r)
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877         case KVM_SET_MEMORY_REGION: {
1878                 struct kvm_memory_region kvm_mem;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1882                         goto out;
1883                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1884                 if (r)
1885                         goto out;
1886                 break;
1887         }
1888         case KVM_GET_DIRTY_LOG: {
1889                 struct kvm_dirty_log log;
1890
1891                 r = -EFAULT;
1892                 if (copy_from_user(&log, (void *)arg, sizeof log))
1893                         goto out;
1894                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1895                 if (r)
1896                         goto out;
1897                 break;
1898         }
1899         case KVM_GET_MSRS:
1900                 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1901                 break;
1902         case KVM_SET_MSRS:
1903                 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1904                 break;
1905         case KVM_GET_MSR_INDEX_LIST: {
1906                 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1907                 struct kvm_msr_list msr_list;
1908                 unsigned n;
1909
1910                 r = -EFAULT;
1911                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1912                         goto out;
1913                 n = msr_list.nmsrs;
1914                 msr_list.nmsrs = num_msrs_to_save;
1915                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1916                         goto out;
1917                 r = -E2BIG;
1918                 if (n < num_msrs_to_save)
1919                         goto out;
1920                 r = -EFAULT;
1921                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1922                                  num_msrs_to_save * sizeof(u32)))
1923                         goto out;
1924                 r = 0;
1925         }
1926         default:
1927                 ;
1928         }
1929 out:
1930         return r;
1931 }
1932
1933 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1934                                    unsigned long address,
1935                                    int *type)
1936 {
1937         struct kvm *kvm = vma->vm_file->private_data;
1938         unsigned long pgoff;
1939         struct kvm_memory_slot *slot;
1940         struct page *page;
1941
1942         *type = VM_FAULT_MINOR;
1943         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1944         slot = gfn_to_memslot(kvm, pgoff);
1945         if (!slot)
1946                 return NOPAGE_SIGBUS;
1947         page = gfn_to_page(slot, pgoff);
1948         if (!page)
1949                 return NOPAGE_SIGBUS;
1950         get_page(page);
1951         return page;
1952 }
1953
1954 static struct vm_operations_struct kvm_dev_vm_ops = {
1955         .nopage = kvm_dev_nopage,
1956 };
1957
1958 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1959 {
1960         vma->vm_ops = &kvm_dev_vm_ops;
1961         return 0;
1962 }
1963
1964 static struct file_operations kvm_chardev_ops = {
1965         .open           = kvm_dev_open,
1966         .release        = kvm_dev_release,
1967         .unlocked_ioctl = kvm_dev_ioctl,
1968         .compat_ioctl   = kvm_dev_ioctl,
1969         .mmap           = kvm_dev_mmap,
1970 };
1971
1972 static struct miscdevice kvm_dev = {
1973         MISC_DYNAMIC_MINOR,
1974         "kvm",
1975         &kvm_chardev_ops,
1976 };
1977
1978 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1979                        void *v)
1980 {
1981         if (val == SYS_RESTART) {
1982                 /*
1983                  * Some (well, at least mine) BIOSes hang on reboot if
1984                  * in vmx root mode.
1985                  */
1986                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1987                 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1988         }
1989         return NOTIFY_OK;
1990 }
1991
1992 static struct notifier_block kvm_reboot_notifier = {
1993         .notifier_call = kvm_reboot,
1994         .priority = 0,
1995 };
1996
1997 static __init void kvm_init_debug(void)
1998 {
1999         struct kvm_stats_debugfs_item *p;
2000
2001         debugfs_dir = debugfs_create_dir("kvm", 0);
2002         for (p = debugfs_entries; p->name; ++p)
2003                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2004                                                p->data);
2005 }
2006
2007 static void kvm_exit_debug(void)
2008 {
2009         struct kvm_stats_debugfs_item *p;
2010
2011         for (p = debugfs_entries; p->name; ++p)
2012                 debugfs_remove(p->dentry);
2013         debugfs_remove(debugfs_dir);
2014 }
2015
2016 hpa_t bad_page_address;
2017
2018 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2019 {
2020         int r;
2021
2022         if (kvm_arch_ops) {
2023                 printk(KERN_ERR "kvm: already loaded the other module\n");
2024                 return -EEXIST;
2025         }
2026
2027         if (!ops->cpu_has_kvm_support()) {
2028                 printk(KERN_ERR "kvm: no hardware support\n");
2029                 return -EOPNOTSUPP;
2030         }
2031         if (ops->disabled_by_bios()) {
2032                 printk(KERN_ERR "kvm: disabled by bios\n");
2033                 return -EOPNOTSUPP;
2034         }
2035
2036         kvm_arch_ops = ops;
2037
2038         r = kvm_arch_ops->hardware_setup();
2039         if (r < 0)
2040             return r;
2041
2042         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2043         register_reboot_notifier(&kvm_reboot_notifier);
2044
2045         kvm_chardev_ops.owner = module;
2046
2047         r = misc_register(&kvm_dev);
2048         if (r) {
2049                 printk (KERN_ERR "kvm: misc device register failed\n");
2050                 goto out_free;
2051         }
2052
2053         return r;
2054
2055 out_free:
2056         unregister_reboot_notifier(&kvm_reboot_notifier);
2057         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2058         kvm_arch_ops->hardware_unsetup();
2059         return r;
2060 }
2061
2062 void kvm_exit_arch(void)
2063 {
2064         misc_deregister(&kvm_dev);
2065
2066         unregister_reboot_notifier(&kvm_reboot_notifier);
2067         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2068         kvm_arch_ops->hardware_unsetup();
2069         kvm_arch_ops = NULL;
2070 }
2071
2072 static __init int kvm_init(void)
2073 {
2074         static struct page *bad_page;
2075         int r = 0;
2076
2077         kvm_init_debug();
2078
2079         kvm_init_msr_list();
2080
2081         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2082                 r = -ENOMEM;
2083                 goto out;
2084         }
2085
2086         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2087         memset(__va(bad_page_address), 0, PAGE_SIZE);
2088
2089         return r;
2090
2091 out:
2092         kvm_exit_debug();
2093         return r;
2094 }
2095
2096 static __exit void kvm_exit(void)
2097 {
2098         kvm_exit_debug();
2099         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2100 }
2101
2102 module_init(kvm_init)
2103 module_exit(kvm_exit)
2104
2105 EXPORT_SYMBOL_GPL(kvm_init_arch);
2106 EXPORT_SYMBOL_GPL(kvm_exit_arch);