31db9b4d3016e91843aa94abf631edf3ebe864dc
[linux-2.6.git] / virt / 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 "iodev.h"
19
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43
44 #include <asm/processor.h>
45 #include <asm/io.h>
46 #include <asm/uaccess.h>
47 #include <asm/pgtable.h>
48
49 MODULE_AUTHOR("Qumranet");
50 MODULE_LICENSE("GPL");
51
52 DEFINE_SPINLOCK(kvm_lock);
53 LIST_HEAD(vm_list);
54
55 static cpumask_t cpus_hardware_enabled;
56
57 struct kmem_cache *kvm_vcpu_cache;
58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
59
60 static __read_mostly struct preempt_ops kvm_preempt_ops;
61
62 static struct dentry *debugfs_dir;
63
64 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
65                            unsigned long arg);
66
67 static inline int valid_vcpu(int n)
68 {
69         return likely(n >= 0 && n < KVM_MAX_VCPUS);
70 }
71
72 /*
73  * Switches to specified vcpu, until a matching vcpu_put()
74  */
75 void vcpu_load(struct kvm_vcpu *vcpu)
76 {
77         int cpu;
78
79         mutex_lock(&vcpu->mutex);
80         cpu = get_cpu();
81         preempt_notifier_register(&vcpu->preempt_notifier);
82         kvm_arch_vcpu_load(vcpu, cpu);
83         put_cpu();
84 }
85
86 void vcpu_put(struct kvm_vcpu *vcpu)
87 {
88         preempt_disable();
89         kvm_arch_vcpu_put(vcpu);
90         preempt_notifier_unregister(&vcpu->preempt_notifier);
91         preempt_enable();
92         mutex_unlock(&vcpu->mutex);
93 }
94
95 static void ack_flush(void *_completed)
96 {
97 }
98
99 void kvm_flush_remote_tlbs(struct kvm *kvm)
100 {
101         int i, cpu;
102         cpumask_t cpus;
103         struct kvm_vcpu *vcpu;
104
105         cpus_clear(cpus);
106         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
107                 vcpu = kvm->vcpus[i];
108                 if (!vcpu)
109                         continue;
110                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
111                         continue;
112                 cpu = vcpu->cpu;
113                 if (cpu != -1 && cpu != raw_smp_processor_id())
114                         cpu_set(cpu, cpus);
115         }
116         if (cpus_empty(cpus))
117                 return;
118         ++kvm->stat.remote_tlb_flush;
119         smp_call_function_mask(cpus, ack_flush, NULL, 1);
120 }
121
122 void kvm_reload_remote_mmus(struct kvm *kvm)
123 {
124         int i, cpu;
125         cpumask_t cpus;
126         struct kvm_vcpu *vcpu;
127
128         cpus_clear(cpus);
129         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
130                 vcpu = kvm->vcpus[i];
131                 if (!vcpu)
132                         continue;
133                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
134                         continue;
135                 cpu = vcpu->cpu;
136                 if (cpu != -1 && cpu != raw_smp_processor_id())
137                         cpu_set(cpu, cpus);
138         }
139         if (cpus_empty(cpus))
140                 return;
141         smp_call_function_mask(cpus, ack_flush, NULL, 1);
142 }
143
144
145 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
146 {
147         struct page *page;
148         int r;
149
150         mutex_init(&vcpu->mutex);
151         vcpu->cpu = -1;
152         vcpu->kvm = kvm;
153         vcpu->vcpu_id = id;
154         init_waitqueue_head(&vcpu->wq);
155
156         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
157         if (!page) {
158                 r = -ENOMEM;
159                 goto fail;
160         }
161         vcpu->run = page_address(page);
162
163         r = kvm_arch_vcpu_init(vcpu);
164         if (r < 0)
165                 goto fail_free_run;
166         return 0;
167
168 fail_free_run:
169         free_page((unsigned long)vcpu->run);
170 fail:
171         return r;
172 }
173 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
174
175 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
176 {
177         kvm_arch_vcpu_uninit(vcpu);
178         free_page((unsigned long)vcpu->run);
179 }
180 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
181
182 static struct kvm *kvm_create_vm(void)
183 {
184         struct kvm *kvm = kvm_arch_create_vm();
185
186         if (IS_ERR(kvm))
187                 goto out;
188
189         kvm->mm = current->mm;
190         atomic_inc(&kvm->mm->mm_count);
191         spin_lock_init(&kvm->mmu_lock);
192         kvm_io_bus_init(&kvm->pio_bus);
193         mutex_init(&kvm->lock);
194         kvm_io_bus_init(&kvm->mmio_bus);
195         init_rwsem(&kvm->slots_lock);
196         spin_lock(&kvm_lock);
197         list_add(&kvm->vm_list, &vm_list);
198         spin_unlock(&kvm_lock);
199 out:
200         return kvm;
201 }
202
203 /*
204  * Free any memory in @free but not in @dont.
205  */
206 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
207                                   struct kvm_memory_slot *dont)
208 {
209         if (!dont || free->rmap != dont->rmap)
210                 vfree(free->rmap);
211
212         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
213                 vfree(free->dirty_bitmap);
214
215         if (!dont || free->lpage_info != dont->lpage_info)
216                 vfree(free->lpage_info);
217
218         free->npages = 0;
219         free->dirty_bitmap = NULL;
220         free->rmap = NULL;
221         free->lpage_info = NULL;
222 }
223
224 void kvm_free_physmem(struct kvm *kvm)
225 {
226         int i;
227
228         for (i = 0; i < kvm->nmemslots; ++i)
229                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
230 }
231
232 static void kvm_destroy_vm(struct kvm *kvm)
233 {
234         struct mm_struct *mm = kvm->mm;
235
236         spin_lock(&kvm_lock);
237         list_del(&kvm->vm_list);
238         spin_unlock(&kvm_lock);
239         kvm_io_bus_destroy(&kvm->pio_bus);
240         kvm_io_bus_destroy(&kvm->mmio_bus);
241         kvm_arch_destroy_vm(kvm);
242         mmdrop(mm);
243 }
244
245 static int kvm_vm_release(struct inode *inode, struct file *filp)
246 {
247         struct kvm *kvm = filp->private_data;
248
249         kvm_destroy_vm(kvm);
250         return 0;
251 }
252
253 /*
254  * Allocate some memory and give it an address in the guest physical address
255  * space.
256  *
257  * Discontiguous memory is allowed, mostly for framebuffers.
258  *
259  * Must be called holding mmap_sem for write.
260  */
261 int __kvm_set_memory_region(struct kvm *kvm,
262                             struct kvm_userspace_memory_region *mem,
263                             int user_alloc)
264 {
265         int r;
266         gfn_t base_gfn;
267         unsigned long npages;
268         unsigned long i;
269         struct kvm_memory_slot *memslot;
270         struct kvm_memory_slot old, new;
271
272         r = -EINVAL;
273         /* General sanity checks */
274         if (mem->memory_size & (PAGE_SIZE - 1))
275                 goto out;
276         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
277                 goto out;
278         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
279                 goto out;
280         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
281                 goto out;
282
283         memslot = &kvm->memslots[mem->slot];
284         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
285         npages = mem->memory_size >> PAGE_SHIFT;
286
287         if (!npages)
288                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
289
290         new = old = *memslot;
291
292         new.base_gfn = base_gfn;
293         new.npages = npages;
294         new.flags = mem->flags;
295
296         /* Disallow changing a memory slot's size. */
297         r = -EINVAL;
298         if (npages && old.npages && npages != old.npages)
299                 goto out_free;
300
301         /* Check for overlaps */
302         r = -EEXIST;
303         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
304                 struct kvm_memory_slot *s = &kvm->memslots[i];
305
306                 if (s == memslot)
307                         continue;
308                 if (!((base_gfn + npages <= s->base_gfn) ||
309                       (base_gfn >= s->base_gfn + s->npages)))
310                         goto out_free;
311         }
312
313         /* Free page dirty bitmap if unneeded */
314         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
315                 new.dirty_bitmap = NULL;
316
317         r = -ENOMEM;
318
319         /* Allocate if a slot is being created */
320         if (npages && !new.rmap) {
321                 new.rmap = vmalloc(npages * sizeof(struct page *));
322
323                 if (!new.rmap)
324                         goto out_free;
325
326                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
327
328                 new.user_alloc = user_alloc;
329                 new.userspace_addr = mem->userspace_addr;
330         }
331         if (npages && !new.lpage_info) {
332                 int largepages = npages / KVM_PAGES_PER_HPAGE;
333                 if (npages % KVM_PAGES_PER_HPAGE)
334                         largepages++;
335                 if (base_gfn % KVM_PAGES_PER_HPAGE)
336                         largepages++;
337
338                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
339
340                 if (!new.lpage_info)
341                         goto out_free;
342
343                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
344
345                 if (base_gfn % KVM_PAGES_PER_HPAGE)
346                         new.lpage_info[0].write_count = 1;
347                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
348                         new.lpage_info[largepages-1].write_count = 1;
349         }
350
351         /* Allocate page dirty bitmap if needed */
352         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
353                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
354
355                 new.dirty_bitmap = vmalloc(dirty_bytes);
356                 if (!new.dirty_bitmap)
357                         goto out_free;
358                 memset(new.dirty_bitmap, 0, dirty_bytes);
359         }
360
361         if (mem->slot >= kvm->nmemslots)
362                 kvm->nmemslots = mem->slot + 1;
363
364         *memslot = new;
365
366         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
367         if (r) {
368                 *memslot = old;
369                 goto out_free;
370         }
371
372         kvm_free_physmem_slot(&old, &new);
373         return 0;
374
375 out_free:
376         kvm_free_physmem_slot(&new, &old);
377 out:
378         return r;
379
380 }
381 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
382
383 int kvm_set_memory_region(struct kvm *kvm,
384                           struct kvm_userspace_memory_region *mem,
385                           int user_alloc)
386 {
387         int r;
388
389         down_write(&kvm->slots_lock);
390         r = __kvm_set_memory_region(kvm, mem, user_alloc);
391         up_write(&kvm->slots_lock);
392         return r;
393 }
394 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
395
396 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
397                                    struct
398                                    kvm_userspace_memory_region *mem,
399                                    int user_alloc)
400 {
401         if (mem->slot >= KVM_MEMORY_SLOTS)
402                 return -EINVAL;
403         return kvm_set_memory_region(kvm, mem, user_alloc);
404 }
405
406 int kvm_get_dirty_log(struct kvm *kvm,
407                         struct kvm_dirty_log *log, int *is_dirty)
408 {
409         struct kvm_memory_slot *memslot;
410         int r, i;
411         int n;
412         unsigned long any = 0;
413
414         r = -EINVAL;
415         if (log->slot >= KVM_MEMORY_SLOTS)
416                 goto out;
417
418         memslot = &kvm->memslots[log->slot];
419         r = -ENOENT;
420         if (!memslot->dirty_bitmap)
421                 goto out;
422
423         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
424
425         for (i = 0; !any && i < n/sizeof(long); ++i)
426                 any = memslot->dirty_bitmap[i];
427
428         r = -EFAULT;
429         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
430                 goto out;
431
432         if (any)
433                 *is_dirty = 1;
434
435         r = 0;
436 out:
437         return r;
438 }
439
440 int is_error_page(struct page *page)
441 {
442         return page == bad_page;
443 }
444 EXPORT_SYMBOL_GPL(is_error_page);
445
446 static inline unsigned long bad_hva(void)
447 {
448         return PAGE_OFFSET;
449 }
450
451 int kvm_is_error_hva(unsigned long addr)
452 {
453         return addr == bad_hva();
454 }
455 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
456
457 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
458 {
459         int i;
460
461         for (i = 0; i < kvm->nmemslots; ++i) {
462                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
463
464                 if (gfn >= memslot->base_gfn
465                     && gfn < memslot->base_gfn + memslot->npages)
466                         return memslot;
467         }
468         return NULL;
469 }
470
471 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
472 {
473         gfn = unalias_gfn(kvm, gfn);
474         return __gfn_to_memslot(kvm, gfn);
475 }
476
477 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
478 {
479         int i;
480
481         gfn = unalias_gfn(kvm, gfn);
482         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
483                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
484
485                 if (gfn >= memslot->base_gfn
486                     && gfn < memslot->base_gfn + memslot->npages)
487                         return 1;
488         }
489         return 0;
490 }
491 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
492
493 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
494 {
495         struct kvm_memory_slot *slot;
496
497         gfn = unalias_gfn(kvm, gfn);
498         slot = __gfn_to_memslot(kvm, gfn);
499         if (!slot)
500                 return bad_hva();
501         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
502 }
503
504 /*
505  * Requires current->mm->mmap_sem to be held
506  */
507 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
508 {
509         struct page *page[1];
510         unsigned long addr;
511         int npages;
512
513         might_sleep();
514
515         addr = gfn_to_hva(kvm, gfn);
516         if (kvm_is_error_hva(addr)) {
517                 get_page(bad_page);
518                 return bad_page;
519         }
520
521         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
522                                 NULL);
523
524         if (npages != 1) {
525                 get_page(bad_page);
526                 return bad_page;
527         }
528
529         return page[0];
530 }
531
532 EXPORT_SYMBOL_GPL(gfn_to_page);
533
534 void kvm_release_page_clean(struct page *page)
535 {
536         put_page(page);
537 }
538 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
539
540 void kvm_release_page_dirty(struct page *page)
541 {
542         if (!PageReserved(page))
543                 SetPageDirty(page);
544         put_page(page);
545 }
546 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
547
548 static int next_segment(unsigned long len, int offset)
549 {
550         if (len > PAGE_SIZE - offset)
551                 return PAGE_SIZE - offset;
552         else
553                 return len;
554 }
555
556 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
557                         int len)
558 {
559         int r;
560         unsigned long addr;
561
562         addr = gfn_to_hva(kvm, gfn);
563         if (kvm_is_error_hva(addr))
564                 return -EFAULT;
565         r = copy_from_user(data, (void __user *)addr + offset, len);
566         if (r)
567                 return -EFAULT;
568         return 0;
569 }
570 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
571
572 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
573 {
574         gfn_t gfn = gpa >> PAGE_SHIFT;
575         int seg;
576         int offset = offset_in_page(gpa);
577         int ret;
578
579         while ((seg = next_segment(len, offset)) != 0) {
580                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
581                 if (ret < 0)
582                         return ret;
583                 offset = 0;
584                 len -= seg;
585                 data += seg;
586                 ++gfn;
587         }
588         return 0;
589 }
590 EXPORT_SYMBOL_GPL(kvm_read_guest);
591
592 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
593                           unsigned long len)
594 {
595         int r;
596         unsigned long addr;
597         gfn_t gfn = gpa >> PAGE_SHIFT;
598         int offset = offset_in_page(gpa);
599
600         addr = gfn_to_hva(kvm, gfn);
601         if (kvm_is_error_hva(addr))
602                 return -EFAULT;
603         pagefault_disable();
604         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
605         pagefault_enable();
606         if (r)
607                 return -EFAULT;
608         return 0;
609 }
610 EXPORT_SYMBOL(kvm_read_guest_atomic);
611
612 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
613                          int offset, int len)
614 {
615         int r;
616         unsigned long addr;
617
618         addr = gfn_to_hva(kvm, gfn);
619         if (kvm_is_error_hva(addr))
620                 return -EFAULT;
621         r = copy_to_user((void __user *)addr + offset, data, len);
622         if (r)
623                 return -EFAULT;
624         mark_page_dirty(kvm, gfn);
625         return 0;
626 }
627 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
628
629 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
630                     unsigned long len)
631 {
632         gfn_t gfn = gpa >> PAGE_SHIFT;
633         int seg;
634         int offset = offset_in_page(gpa);
635         int ret;
636
637         while ((seg = next_segment(len, offset)) != 0) {
638                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
639                 if (ret < 0)
640                         return ret;
641                 offset = 0;
642                 len -= seg;
643                 data += seg;
644                 ++gfn;
645         }
646         return 0;
647 }
648
649 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
650 {
651         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
652 }
653 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
654
655 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
656 {
657         gfn_t gfn = gpa >> PAGE_SHIFT;
658         int seg;
659         int offset = offset_in_page(gpa);
660         int ret;
661
662         while ((seg = next_segment(len, offset)) != 0) {
663                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
664                 if (ret < 0)
665                         return ret;
666                 offset = 0;
667                 len -= seg;
668                 ++gfn;
669         }
670         return 0;
671 }
672 EXPORT_SYMBOL_GPL(kvm_clear_guest);
673
674 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
675 {
676         struct kvm_memory_slot *memslot;
677
678         gfn = unalias_gfn(kvm, gfn);
679         memslot = __gfn_to_memslot(kvm, gfn);
680         if (memslot && memslot->dirty_bitmap) {
681                 unsigned long rel_gfn = gfn - memslot->base_gfn;
682
683                 /* avoid RMW */
684                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
685                         set_bit(rel_gfn, memslot->dirty_bitmap);
686         }
687 }
688
689 /*
690  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
691  */
692 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
693 {
694         DECLARE_WAITQUEUE(wait, current);
695
696         add_wait_queue(&vcpu->wq, &wait);
697
698         /*
699          * We will block until either an interrupt or a signal wakes us up
700          */
701         while (!kvm_cpu_has_interrupt(vcpu)
702                && !signal_pending(current)
703                && !kvm_arch_vcpu_runnable(vcpu)) {
704                 set_current_state(TASK_INTERRUPTIBLE);
705                 vcpu_put(vcpu);
706                 schedule();
707                 vcpu_load(vcpu);
708         }
709
710         __set_current_state(TASK_RUNNING);
711         remove_wait_queue(&vcpu->wq, &wait);
712 }
713
714 void kvm_resched(struct kvm_vcpu *vcpu)
715 {
716         if (!need_resched())
717                 return;
718         cond_resched();
719 }
720 EXPORT_SYMBOL_GPL(kvm_resched);
721
722 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
723 {
724         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
725         struct page *page;
726
727         if (vmf->pgoff == 0)
728                 page = virt_to_page(vcpu->run);
729 #ifdef CONFIG_X86
730         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
731                 page = virt_to_page(vcpu->arch.pio_data);
732 #endif
733         else
734                 return VM_FAULT_SIGBUS;
735         get_page(page);
736         vmf->page = page;
737         return 0;
738 }
739
740 static struct vm_operations_struct kvm_vcpu_vm_ops = {
741         .fault = kvm_vcpu_fault,
742 };
743
744 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
745 {
746         vma->vm_ops = &kvm_vcpu_vm_ops;
747         return 0;
748 }
749
750 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
751 {
752         struct kvm_vcpu *vcpu = filp->private_data;
753
754         fput(vcpu->kvm->filp);
755         return 0;
756 }
757
758 static const struct file_operations kvm_vcpu_fops = {
759         .release        = kvm_vcpu_release,
760         .unlocked_ioctl = kvm_vcpu_ioctl,
761         .compat_ioctl   = kvm_vcpu_ioctl,
762         .mmap           = kvm_vcpu_mmap,
763 };
764
765 /*
766  * Allocates an inode for the vcpu.
767  */
768 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
769 {
770         int fd, r;
771         struct inode *inode;
772         struct file *file;
773
774         r = anon_inode_getfd(&fd, &inode, &file,
775                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
776         if (r)
777                 return r;
778         atomic_inc(&vcpu->kvm->filp->f_count);
779         return fd;
780 }
781
782 /*
783  * Creates some virtual cpus.  Good luck creating more than one.
784  */
785 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
786 {
787         int r;
788         struct kvm_vcpu *vcpu;
789
790         if (!valid_vcpu(n))
791                 return -EINVAL;
792
793         vcpu = kvm_arch_vcpu_create(kvm, n);
794         if (IS_ERR(vcpu))
795                 return PTR_ERR(vcpu);
796
797         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
798
799         r = kvm_arch_vcpu_setup(vcpu);
800         if (r)
801                 goto vcpu_destroy;
802
803         mutex_lock(&kvm->lock);
804         if (kvm->vcpus[n]) {
805                 r = -EEXIST;
806                 mutex_unlock(&kvm->lock);
807                 goto vcpu_destroy;
808         }
809         kvm->vcpus[n] = vcpu;
810         mutex_unlock(&kvm->lock);
811
812         /* Now it's all set up, let userspace reach it */
813         r = create_vcpu_fd(vcpu);
814         if (r < 0)
815                 goto unlink;
816         return r;
817
818 unlink:
819         mutex_lock(&kvm->lock);
820         kvm->vcpus[n] = NULL;
821         mutex_unlock(&kvm->lock);
822 vcpu_destroy:
823         kvm_arch_vcpu_destroy(vcpu);
824         return r;
825 }
826
827 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
828 {
829         if (sigset) {
830                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
831                 vcpu->sigset_active = 1;
832                 vcpu->sigset = *sigset;
833         } else
834                 vcpu->sigset_active = 0;
835         return 0;
836 }
837
838 static long kvm_vcpu_ioctl(struct file *filp,
839                            unsigned int ioctl, unsigned long arg)
840 {
841         struct kvm_vcpu *vcpu = filp->private_data;
842         void __user *argp = (void __user *)arg;
843         int r;
844
845         if (vcpu->kvm->mm != current->mm)
846                 return -EIO;
847         switch (ioctl) {
848         case KVM_RUN:
849                 r = -EINVAL;
850                 if (arg)
851                         goto out;
852                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
853                 break;
854         case KVM_GET_REGS: {
855                 struct kvm_regs kvm_regs;
856
857                 memset(&kvm_regs, 0, sizeof kvm_regs);
858                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
859                 if (r)
860                         goto out;
861                 r = -EFAULT;
862                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
863                         goto out;
864                 r = 0;
865                 break;
866         }
867         case KVM_SET_REGS: {
868                 struct kvm_regs kvm_regs;
869
870                 r = -EFAULT;
871                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
872                         goto out;
873                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
874                 if (r)
875                         goto out;
876                 r = 0;
877                 break;
878         }
879         case KVM_GET_SREGS: {
880                 struct kvm_sregs kvm_sregs;
881
882                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
883                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
884                 if (r)
885                         goto out;
886                 r = -EFAULT;
887                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
888                         goto out;
889                 r = 0;
890                 break;
891         }
892         case KVM_SET_SREGS: {
893                 struct kvm_sregs kvm_sregs;
894
895                 r = -EFAULT;
896                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
897                         goto out;
898                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
899                 if (r)
900                         goto out;
901                 r = 0;
902                 break;
903         }
904         case KVM_TRANSLATE: {
905                 struct kvm_translation tr;
906
907                 r = -EFAULT;
908                 if (copy_from_user(&tr, argp, sizeof tr))
909                         goto out;
910                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
911                 if (r)
912                         goto out;
913                 r = -EFAULT;
914                 if (copy_to_user(argp, &tr, sizeof tr))
915                         goto out;
916                 r = 0;
917                 break;
918         }
919         case KVM_DEBUG_GUEST: {
920                 struct kvm_debug_guest dbg;
921
922                 r = -EFAULT;
923                 if (copy_from_user(&dbg, argp, sizeof dbg))
924                         goto out;
925                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
926                 if (r)
927                         goto out;
928                 r = 0;
929                 break;
930         }
931         case KVM_SET_SIGNAL_MASK: {
932                 struct kvm_signal_mask __user *sigmask_arg = argp;
933                 struct kvm_signal_mask kvm_sigmask;
934                 sigset_t sigset, *p;
935
936                 p = NULL;
937                 if (argp) {
938                         r = -EFAULT;
939                         if (copy_from_user(&kvm_sigmask, argp,
940                                            sizeof kvm_sigmask))
941                                 goto out;
942                         r = -EINVAL;
943                         if (kvm_sigmask.len != sizeof sigset)
944                                 goto out;
945                         r = -EFAULT;
946                         if (copy_from_user(&sigset, sigmask_arg->sigset,
947                                            sizeof sigset))
948                                 goto out;
949                         p = &sigset;
950                 }
951                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
952                 break;
953         }
954         case KVM_GET_FPU: {
955                 struct kvm_fpu fpu;
956
957                 memset(&fpu, 0, sizeof fpu);
958                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
959                 if (r)
960                         goto out;
961                 r = -EFAULT;
962                 if (copy_to_user(argp, &fpu, sizeof fpu))
963                         goto out;
964                 r = 0;
965                 break;
966         }
967         case KVM_SET_FPU: {
968                 struct kvm_fpu fpu;
969
970                 r = -EFAULT;
971                 if (copy_from_user(&fpu, argp, sizeof fpu))
972                         goto out;
973                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
974                 if (r)
975                         goto out;
976                 r = 0;
977                 break;
978         }
979         default:
980                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
981         }
982 out:
983         return r;
984 }
985
986 static long kvm_vm_ioctl(struct file *filp,
987                            unsigned int ioctl, unsigned long arg)
988 {
989         struct kvm *kvm = filp->private_data;
990         void __user *argp = (void __user *)arg;
991         int r;
992
993         if (kvm->mm != current->mm)
994                 return -EIO;
995         switch (ioctl) {
996         case KVM_CREATE_VCPU:
997                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
998                 if (r < 0)
999                         goto out;
1000                 break;
1001         case KVM_SET_USER_MEMORY_REGION: {
1002                 struct kvm_userspace_memory_region kvm_userspace_mem;
1003
1004                 r = -EFAULT;
1005                 if (copy_from_user(&kvm_userspace_mem, argp,
1006                                                 sizeof kvm_userspace_mem))
1007                         goto out;
1008
1009                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1010                 if (r)
1011                         goto out;
1012                 break;
1013         }
1014         case KVM_GET_DIRTY_LOG: {
1015                 struct kvm_dirty_log log;
1016
1017                 r = -EFAULT;
1018                 if (copy_from_user(&log, argp, sizeof log))
1019                         goto out;
1020                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1021                 if (r)
1022                         goto out;
1023                 break;
1024         }
1025         default:
1026                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1027         }
1028 out:
1029         return r;
1030 }
1031
1032 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1033 {
1034         struct kvm *kvm = vma->vm_file->private_data;
1035         struct page *page;
1036
1037         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1038                 return VM_FAULT_SIGBUS;
1039         page = gfn_to_page(kvm, vmf->pgoff);
1040         if (is_error_page(page)) {
1041                 kvm_release_page_clean(page);
1042                 return VM_FAULT_SIGBUS;
1043         }
1044         vmf->page = page;
1045         return 0;
1046 }
1047
1048 static struct vm_operations_struct kvm_vm_vm_ops = {
1049         .fault = kvm_vm_fault,
1050 };
1051
1052 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1053 {
1054         vma->vm_ops = &kvm_vm_vm_ops;
1055         return 0;
1056 }
1057
1058 static const struct file_operations kvm_vm_fops = {
1059         .release        = kvm_vm_release,
1060         .unlocked_ioctl = kvm_vm_ioctl,
1061         .compat_ioctl   = kvm_vm_ioctl,
1062         .mmap           = kvm_vm_mmap,
1063 };
1064
1065 static int kvm_dev_ioctl_create_vm(void)
1066 {
1067         int fd, r;
1068         struct inode *inode;
1069         struct file *file;
1070         struct kvm *kvm;
1071
1072         kvm = kvm_create_vm();
1073         if (IS_ERR(kvm))
1074                 return PTR_ERR(kvm);
1075         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1076         if (r) {
1077                 kvm_destroy_vm(kvm);
1078                 return r;
1079         }
1080
1081         kvm->filp = file;
1082
1083         return fd;
1084 }
1085
1086 static long kvm_dev_ioctl(struct file *filp,
1087                           unsigned int ioctl, unsigned long arg)
1088 {
1089         void __user *argp = (void __user *)arg;
1090         long r = -EINVAL;
1091
1092         switch (ioctl) {
1093         case KVM_GET_API_VERSION:
1094                 r = -EINVAL;
1095                 if (arg)
1096                         goto out;
1097                 r = KVM_API_VERSION;
1098                 break;
1099         case KVM_CREATE_VM:
1100                 r = -EINVAL;
1101                 if (arg)
1102                         goto out;
1103                 r = kvm_dev_ioctl_create_vm();
1104                 break;
1105         case KVM_CHECK_EXTENSION:
1106                 r = kvm_dev_ioctl_check_extension((long)argp);
1107                 break;
1108         case KVM_GET_VCPU_MMAP_SIZE:
1109                 r = -EINVAL;
1110                 if (arg)
1111                         goto out;
1112                 r = PAGE_SIZE;     /* struct kvm_run */
1113 #ifdef CONFIG_X86
1114                 r += PAGE_SIZE;    /* pio data page */
1115 #endif
1116                 break;
1117         default:
1118                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1119         }
1120 out:
1121         return r;
1122 }
1123
1124 static struct file_operations kvm_chardev_ops = {
1125         .unlocked_ioctl = kvm_dev_ioctl,
1126         .compat_ioctl   = kvm_dev_ioctl,
1127 };
1128
1129 static struct miscdevice kvm_dev = {
1130         KVM_MINOR,
1131         "kvm",
1132         &kvm_chardev_ops,
1133 };
1134
1135 static void hardware_enable(void *junk)
1136 {
1137         int cpu = raw_smp_processor_id();
1138
1139         if (cpu_isset(cpu, cpus_hardware_enabled))
1140                 return;
1141         cpu_set(cpu, cpus_hardware_enabled);
1142         kvm_arch_hardware_enable(NULL);
1143 }
1144
1145 static void hardware_disable(void *junk)
1146 {
1147         int cpu = raw_smp_processor_id();
1148
1149         if (!cpu_isset(cpu, cpus_hardware_enabled))
1150                 return;
1151         cpu_clear(cpu, cpus_hardware_enabled);
1152         decache_vcpus_on_cpu(cpu);
1153         kvm_arch_hardware_disable(NULL);
1154 }
1155
1156 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1157                            void *v)
1158 {
1159         int cpu = (long)v;
1160
1161         val &= ~CPU_TASKS_FROZEN;
1162         switch (val) {
1163         case CPU_DYING:
1164                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1165                        cpu);
1166                 hardware_disable(NULL);
1167                 break;
1168         case CPU_UP_CANCELED:
1169                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1170                        cpu);
1171                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1172                 break;
1173         case CPU_ONLINE:
1174                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1175                        cpu);
1176                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1177                 break;
1178         }
1179         return NOTIFY_OK;
1180 }
1181
1182 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1183                       void *v)
1184 {
1185         if (val == SYS_RESTART) {
1186                 /*
1187                  * Some (well, at least mine) BIOSes hang on reboot if
1188                  * in vmx root mode.
1189                  */
1190                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1191                 on_each_cpu(hardware_disable, NULL, 0, 1);
1192         }
1193         return NOTIFY_OK;
1194 }
1195
1196 static struct notifier_block kvm_reboot_notifier = {
1197         .notifier_call = kvm_reboot,
1198         .priority = 0,
1199 };
1200
1201 void kvm_io_bus_init(struct kvm_io_bus *bus)
1202 {
1203         memset(bus, 0, sizeof(*bus));
1204 }
1205
1206 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1207 {
1208         int i;
1209
1210         for (i = 0; i < bus->dev_count; i++) {
1211                 struct kvm_io_device *pos = bus->devs[i];
1212
1213                 kvm_iodevice_destructor(pos);
1214         }
1215 }
1216
1217 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1218 {
1219         int i;
1220
1221         for (i = 0; i < bus->dev_count; i++) {
1222                 struct kvm_io_device *pos = bus->devs[i];
1223
1224                 if (pos->in_range(pos, addr))
1225                         return pos;
1226         }
1227
1228         return NULL;
1229 }
1230
1231 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1232 {
1233         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1234
1235         bus->devs[bus->dev_count++] = dev;
1236 }
1237
1238 static struct notifier_block kvm_cpu_notifier = {
1239         .notifier_call = kvm_cpu_hotplug,
1240         .priority = 20, /* must be > scheduler priority */
1241 };
1242
1243 static int vm_stat_get(void *_offset, u64 *val)
1244 {
1245         unsigned offset = (long)_offset;
1246         struct kvm *kvm;
1247
1248         *val = 0;
1249         spin_lock(&kvm_lock);
1250         list_for_each_entry(kvm, &vm_list, vm_list)
1251                 *val += *(u32 *)((void *)kvm + offset);
1252         spin_unlock(&kvm_lock);
1253         return 0;
1254 }
1255
1256 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1257
1258 static int vcpu_stat_get(void *_offset, u64 *val)
1259 {
1260         unsigned offset = (long)_offset;
1261         struct kvm *kvm;
1262         struct kvm_vcpu *vcpu;
1263         int i;
1264
1265         *val = 0;
1266         spin_lock(&kvm_lock);
1267         list_for_each_entry(kvm, &vm_list, vm_list)
1268                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1269                         vcpu = kvm->vcpus[i];
1270                         if (vcpu)
1271                                 *val += *(u32 *)((void *)vcpu + offset);
1272                 }
1273         spin_unlock(&kvm_lock);
1274         return 0;
1275 }
1276
1277 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1278
1279 static struct file_operations *stat_fops[] = {
1280         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1281         [KVM_STAT_VM]   = &vm_stat_fops,
1282 };
1283
1284 static void kvm_init_debug(void)
1285 {
1286         struct kvm_stats_debugfs_item *p;
1287
1288         debugfs_dir = debugfs_create_dir("kvm", NULL);
1289         for (p = debugfs_entries; p->name; ++p)
1290                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1291                                                 (void *)(long)p->offset,
1292                                                 stat_fops[p->kind]);
1293 }
1294
1295 static void kvm_exit_debug(void)
1296 {
1297         struct kvm_stats_debugfs_item *p;
1298
1299         for (p = debugfs_entries; p->name; ++p)
1300                 debugfs_remove(p->dentry);
1301         debugfs_remove(debugfs_dir);
1302 }
1303
1304 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1305 {
1306         hardware_disable(NULL);
1307         return 0;
1308 }
1309
1310 static int kvm_resume(struct sys_device *dev)
1311 {
1312         hardware_enable(NULL);
1313         return 0;
1314 }
1315
1316 static struct sysdev_class kvm_sysdev_class = {
1317         .name = "kvm",
1318         .suspend = kvm_suspend,
1319         .resume = kvm_resume,
1320 };
1321
1322 static struct sys_device kvm_sysdev = {
1323         .id = 0,
1324         .cls = &kvm_sysdev_class,
1325 };
1326
1327 struct page *bad_page;
1328
1329 static inline
1330 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1331 {
1332         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1333 }
1334
1335 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1336 {
1337         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1338
1339         kvm_arch_vcpu_load(vcpu, cpu);
1340 }
1341
1342 static void kvm_sched_out(struct preempt_notifier *pn,
1343                           struct task_struct *next)
1344 {
1345         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1346
1347         kvm_arch_vcpu_put(vcpu);
1348 }
1349
1350 int kvm_init(void *opaque, unsigned int vcpu_size,
1351                   struct module *module)
1352 {
1353         int r;
1354         int cpu;
1355
1356         kvm_init_debug();
1357
1358         r = kvm_arch_init(opaque);
1359         if (r)
1360                 goto out_fail;
1361
1362         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1363
1364         if (bad_page == NULL) {
1365                 r = -ENOMEM;
1366                 goto out;
1367         }
1368
1369         r = kvm_arch_hardware_setup();
1370         if (r < 0)
1371                 goto out_free_0;
1372
1373         for_each_online_cpu(cpu) {
1374                 smp_call_function_single(cpu,
1375                                 kvm_arch_check_processor_compat,
1376                                 &r, 0, 1);
1377                 if (r < 0)
1378                         goto out_free_1;
1379         }
1380
1381         on_each_cpu(hardware_enable, NULL, 0, 1);
1382         r = register_cpu_notifier(&kvm_cpu_notifier);
1383         if (r)
1384                 goto out_free_2;
1385         register_reboot_notifier(&kvm_reboot_notifier);
1386
1387         r = sysdev_class_register(&kvm_sysdev_class);
1388         if (r)
1389                 goto out_free_3;
1390
1391         r = sysdev_register(&kvm_sysdev);
1392         if (r)
1393                 goto out_free_4;
1394
1395         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1396         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1397                                            __alignof__(struct kvm_vcpu),
1398                                            0, NULL);
1399         if (!kvm_vcpu_cache) {
1400                 r = -ENOMEM;
1401                 goto out_free_5;
1402         }
1403
1404         kvm_chardev_ops.owner = module;
1405
1406         r = misc_register(&kvm_dev);
1407         if (r) {
1408                 printk(KERN_ERR "kvm: misc device register failed\n");
1409                 goto out_free;
1410         }
1411
1412         kvm_preempt_ops.sched_in = kvm_sched_in;
1413         kvm_preempt_ops.sched_out = kvm_sched_out;
1414
1415         return 0;
1416
1417 out_free:
1418         kmem_cache_destroy(kvm_vcpu_cache);
1419 out_free_5:
1420         sysdev_unregister(&kvm_sysdev);
1421 out_free_4:
1422         sysdev_class_unregister(&kvm_sysdev_class);
1423 out_free_3:
1424         unregister_reboot_notifier(&kvm_reboot_notifier);
1425         unregister_cpu_notifier(&kvm_cpu_notifier);
1426 out_free_2:
1427         on_each_cpu(hardware_disable, NULL, 0, 1);
1428 out_free_1:
1429         kvm_arch_hardware_unsetup();
1430 out_free_0:
1431         __free_page(bad_page);
1432 out:
1433         kvm_arch_exit();
1434         kvm_exit_debug();
1435 out_fail:
1436         return r;
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_init);
1439
1440 void kvm_exit(void)
1441 {
1442         misc_deregister(&kvm_dev);
1443         kmem_cache_destroy(kvm_vcpu_cache);
1444         sysdev_unregister(&kvm_sysdev);
1445         sysdev_class_unregister(&kvm_sysdev_class);
1446         unregister_reboot_notifier(&kvm_reboot_notifier);
1447         unregister_cpu_notifier(&kvm_cpu_notifier);
1448         on_each_cpu(hardware_disable, NULL, 0, 1);
1449         kvm_arch_hardware_unsetup();
1450         kvm_arch_exit();
1451         kvm_exit_debug();
1452         __free_page(bad_page);
1453 }
1454 EXPORT_SYMBOL_GPL(kvm_exit);