KVM: remove the usage of the mmap_sem for the protection of the memory slots.
[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 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
123 {
124         struct page *page;
125         int r;
126
127         mutex_init(&vcpu->mutex);
128         vcpu->cpu = -1;
129         vcpu->kvm = kvm;
130         vcpu->vcpu_id = id;
131         init_waitqueue_head(&vcpu->wq);
132
133         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
134         if (!page) {
135                 r = -ENOMEM;
136                 goto fail;
137         }
138         vcpu->run = page_address(page);
139
140         r = kvm_arch_vcpu_init(vcpu);
141         if (r < 0)
142                 goto fail_free_run;
143         return 0;
144
145 fail_free_run:
146         free_page((unsigned long)vcpu->run);
147 fail:
148         return r;
149 }
150 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
151
152 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
153 {
154         kvm_arch_vcpu_uninit(vcpu);
155         free_page((unsigned long)vcpu->run);
156 }
157 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
158
159 static struct kvm *kvm_create_vm(void)
160 {
161         struct kvm *kvm = kvm_arch_create_vm();
162
163         if (IS_ERR(kvm))
164                 goto out;
165
166         kvm->mm = current->mm;
167         atomic_inc(&kvm->mm->mm_count);
168         spin_lock_init(&kvm->mmu_lock);
169         kvm_io_bus_init(&kvm->pio_bus);
170         mutex_init(&kvm->lock);
171         kvm_io_bus_init(&kvm->mmio_bus);
172         init_rwsem(&kvm->slots_lock);
173         spin_lock(&kvm_lock);
174         list_add(&kvm->vm_list, &vm_list);
175         spin_unlock(&kvm_lock);
176 out:
177         return kvm;
178 }
179
180 /*
181  * Free any memory in @free but not in @dont.
182  */
183 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
184                                   struct kvm_memory_slot *dont)
185 {
186         if (!dont || free->rmap != dont->rmap)
187                 vfree(free->rmap);
188
189         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
190                 vfree(free->dirty_bitmap);
191
192         free->npages = 0;
193         free->dirty_bitmap = NULL;
194         free->rmap = NULL;
195 }
196
197 void kvm_free_physmem(struct kvm *kvm)
198 {
199         int i;
200
201         for (i = 0; i < kvm->nmemslots; ++i)
202                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
203 }
204
205 static void kvm_destroy_vm(struct kvm *kvm)
206 {
207         struct mm_struct *mm = kvm->mm;
208
209         spin_lock(&kvm_lock);
210         list_del(&kvm->vm_list);
211         spin_unlock(&kvm_lock);
212         kvm_io_bus_destroy(&kvm->pio_bus);
213         kvm_io_bus_destroy(&kvm->mmio_bus);
214         kvm_arch_destroy_vm(kvm);
215         mmdrop(mm);
216 }
217
218 static int kvm_vm_release(struct inode *inode, struct file *filp)
219 {
220         struct kvm *kvm = filp->private_data;
221
222         kvm_destroy_vm(kvm);
223         return 0;
224 }
225
226 /*
227  * Allocate some memory and give it an address in the guest physical address
228  * space.
229  *
230  * Discontiguous memory is allowed, mostly for framebuffers.
231  *
232  * Must be called holding mmap_sem for write.
233  */
234 int __kvm_set_memory_region(struct kvm *kvm,
235                             struct kvm_userspace_memory_region *mem,
236                             int user_alloc)
237 {
238         int r;
239         gfn_t base_gfn;
240         unsigned long npages;
241         unsigned long i;
242         struct kvm_memory_slot *memslot;
243         struct kvm_memory_slot old, new;
244
245         r = -EINVAL;
246         /* General sanity checks */
247         if (mem->memory_size & (PAGE_SIZE - 1))
248                 goto out;
249         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
250                 goto out;
251         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
252                 goto out;
253         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
254                 goto out;
255
256         memslot = &kvm->memslots[mem->slot];
257         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
258         npages = mem->memory_size >> PAGE_SHIFT;
259
260         if (!npages)
261                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
262
263         new = old = *memslot;
264
265         new.base_gfn = base_gfn;
266         new.npages = npages;
267         new.flags = mem->flags;
268
269         /* Disallow changing a memory slot's size. */
270         r = -EINVAL;
271         if (npages && old.npages && npages != old.npages)
272                 goto out_free;
273
274         /* Check for overlaps */
275         r = -EEXIST;
276         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
277                 struct kvm_memory_slot *s = &kvm->memslots[i];
278
279                 if (s == memslot)
280                         continue;
281                 if (!((base_gfn + npages <= s->base_gfn) ||
282                       (base_gfn >= s->base_gfn + s->npages)))
283                         goto out_free;
284         }
285
286         /* Free page dirty bitmap if unneeded */
287         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
288                 new.dirty_bitmap = NULL;
289
290         r = -ENOMEM;
291
292         /* Allocate if a slot is being created */
293         if (npages && !new.rmap) {
294                 new.rmap = vmalloc(npages * sizeof(struct page *));
295
296                 if (!new.rmap)
297                         goto out_free;
298
299                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
300
301                 new.user_alloc = user_alloc;
302                 new.userspace_addr = mem->userspace_addr;
303         }
304
305         /* Allocate page dirty bitmap if needed */
306         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
307                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
308
309                 new.dirty_bitmap = vmalloc(dirty_bytes);
310                 if (!new.dirty_bitmap)
311                         goto out_free;
312                 memset(new.dirty_bitmap, 0, dirty_bytes);
313         }
314
315         if (mem->slot >= kvm->nmemslots)
316                 kvm->nmemslots = mem->slot + 1;
317
318         *memslot = new;
319
320         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
321         if (r) {
322                 *memslot = old;
323                 goto out_free;
324         }
325
326         kvm_free_physmem_slot(&old, &new);
327         return 0;
328
329 out_free:
330         kvm_free_physmem_slot(&new, &old);
331 out:
332         return r;
333
334 }
335 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
336
337 int kvm_set_memory_region(struct kvm *kvm,
338                           struct kvm_userspace_memory_region *mem,
339                           int user_alloc)
340 {
341         int r;
342
343         down_write(&kvm->slots_lock);
344         r = __kvm_set_memory_region(kvm, mem, user_alloc);
345         up_write(&kvm->slots_lock);
346         return r;
347 }
348 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
349
350 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
351                                    struct
352                                    kvm_userspace_memory_region *mem,
353                                    int user_alloc)
354 {
355         if (mem->slot >= KVM_MEMORY_SLOTS)
356                 return -EINVAL;
357         return kvm_set_memory_region(kvm, mem, user_alloc);
358 }
359
360 int kvm_get_dirty_log(struct kvm *kvm,
361                         struct kvm_dirty_log *log, int *is_dirty)
362 {
363         struct kvm_memory_slot *memslot;
364         int r, i;
365         int n;
366         unsigned long any = 0;
367
368         r = -EINVAL;
369         if (log->slot >= KVM_MEMORY_SLOTS)
370                 goto out;
371
372         memslot = &kvm->memslots[log->slot];
373         r = -ENOENT;
374         if (!memslot->dirty_bitmap)
375                 goto out;
376
377         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
378
379         for (i = 0; !any && i < n/sizeof(long); ++i)
380                 any = memslot->dirty_bitmap[i];
381
382         r = -EFAULT;
383         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
384                 goto out;
385
386         if (any)
387                 *is_dirty = 1;
388
389         r = 0;
390 out:
391         return r;
392 }
393
394 int is_error_page(struct page *page)
395 {
396         return page == bad_page;
397 }
398 EXPORT_SYMBOL_GPL(is_error_page);
399
400 static inline unsigned long bad_hva(void)
401 {
402         return PAGE_OFFSET;
403 }
404
405 int kvm_is_error_hva(unsigned long addr)
406 {
407         return addr == bad_hva();
408 }
409 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
410
411 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
412 {
413         int i;
414
415         for (i = 0; i < kvm->nmemslots; ++i) {
416                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
417
418                 if (gfn >= memslot->base_gfn
419                     && gfn < memslot->base_gfn + memslot->npages)
420                         return memslot;
421         }
422         return NULL;
423 }
424
425 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
426 {
427         gfn = unalias_gfn(kvm, gfn);
428         return __gfn_to_memslot(kvm, gfn);
429 }
430
431 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
432 {
433         int i;
434
435         gfn = unalias_gfn(kvm, gfn);
436         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
437                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
438
439                 if (gfn >= memslot->base_gfn
440                     && gfn < memslot->base_gfn + memslot->npages)
441                         return 1;
442         }
443         return 0;
444 }
445 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
446
447 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
448 {
449         struct kvm_memory_slot *slot;
450
451         gfn = unalias_gfn(kvm, gfn);
452         slot = __gfn_to_memslot(kvm, gfn);
453         if (!slot)
454                 return bad_hva();
455         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
456 }
457
458 /*
459  * Requires current->mm->mmap_sem to be held
460  */
461 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
462 {
463         struct page *page[1];
464         unsigned long addr;
465         int npages;
466
467         might_sleep();
468
469         addr = gfn_to_hva(kvm, gfn);
470         if (kvm_is_error_hva(addr)) {
471                 get_page(bad_page);
472                 return bad_page;
473         }
474
475         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
476                                 NULL);
477
478         if (npages != 1) {
479                 get_page(bad_page);
480                 return bad_page;
481         }
482
483         return page[0];
484 }
485
486 EXPORT_SYMBOL_GPL(gfn_to_page);
487
488 void kvm_release_page_clean(struct page *page)
489 {
490         put_page(page);
491 }
492 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
493
494 void kvm_release_page_dirty(struct page *page)
495 {
496         if (!PageReserved(page))
497                 SetPageDirty(page);
498         put_page(page);
499 }
500 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
501
502 static int next_segment(unsigned long len, int offset)
503 {
504         if (len > PAGE_SIZE - offset)
505                 return PAGE_SIZE - offset;
506         else
507                 return len;
508 }
509
510 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
511                         int len)
512 {
513         int r;
514         unsigned long addr;
515
516         addr = gfn_to_hva(kvm, gfn);
517         if (kvm_is_error_hva(addr))
518                 return -EFAULT;
519         r = copy_from_user(data, (void __user *)addr + offset, len);
520         if (r)
521                 return -EFAULT;
522         return 0;
523 }
524 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
525
526 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
527 {
528         gfn_t gfn = gpa >> PAGE_SHIFT;
529         int seg;
530         int offset = offset_in_page(gpa);
531         int ret;
532
533         while ((seg = next_segment(len, offset)) != 0) {
534                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
535                 if (ret < 0)
536                         return ret;
537                 offset = 0;
538                 len -= seg;
539                 data += seg;
540                 ++gfn;
541         }
542         return 0;
543 }
544 EXPORT_SYMBOL_GPL(kvm_read_guest);
545
546 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
547                           unsigned long len)
548 {
549         int r;
550         unsigned long addr;
551         gfn_t gfn = gpa >> PAGE_SHIFT;
552         int offset = offset_in_page(gpa);
553
554         addr = gfn_to_hva(kvm, gfn);
555         if (kvm_is_error_hva(addr))
556                 return -EFAULT;
557         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
558         if (r)
559                 return -EFAULT;
560         return 0;
561 }
562 EXPORT_SYMBOL(kvm_read_guest_atomic);
563
564 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
565                          int offset, int len)
566 {
567         int r;
568         unsigned long addr;
569
570         addr = gfn_to_hva(kvm, gfn);
571         if (kvm_is_error_hva(addr))
572                 return -EFAULT;
573         r = copy_to_user((void __user *)addr + offset, data, len);
574         if (r)
575                 return -EFAULT;
576         mark_page_dirty(kvm, gfn);
577         return 0;
578 }
579 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
580
581 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
582                     unsigned long len)
583 {
584         gfn_t gfn = gpa >> PAGE_SHIFT;
585         int seg;
586         int offset = offset_in_page(gpa);
587         int ret;
588
589         while ((seg = next_segment(len, offset)) != 0) {
590                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
591                 if (ret < 0)
592                         return ret;
593                 offset = 0;
594                 len -= seg;
595                 data += seg;
596                 ++gfn;
597         }
598         return 0;
599 }
600
601 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
602 {
603         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
604 }
605 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
606
607 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
608 {
609         gfn_t gfn = gpa >> PAGE_SHIFT;
610         int seg;
611         int offset = offset_in_page(gpa);
612         int ret;
613
614         while ((seg = next_segment(len, offset)) != 0) {
615                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
616                 if (ret < 0)
617                         return ret;
618                 offset = 0;
619                 len -= seg;
620                 ++gfn;
621         }
622         return 0;
623 }
624 EXPORT_SYMBOL_GPL(kvm_clear_guest);
625
626 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
627 {
628         struct kvm_memory_slot *memslot;
629
630         gfn = unalias_gfn(kvm, gfn);
631         memslot = __gfn_to_memslot(kvm, gfn);
632         if (memslot && memslot->dirty_bitmap) {
633                 unsigned long rel_gfn = gfn - memslot->base_gfn;
634
635                 /* avoid RMW */
636                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
637                         set_bit(rel_gfn, memslot->dirty_bitmap);
638         }
639 }
640
641 /*
642  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
643  */
644 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
645 {
646         DECLARE_WAITQUEUE(wait, current);
647
648         add_wait_queue(&vcpu->wq, &wait);
649
650         /*
651          * We will block until either an interrupt or a signal wakes us up
652          */
653         while (!kvm_cpu_has_interrupt(vcpu)
654                && !signal_pending(current)
655                && !kvm_arch_vcpu_runnable(vcpu)) {
656                 set_current_state(TASK_INTERRUPTIBLE);
657                 vcpu_put(vcpu);
658                 schedule();
659                 vcpu_load(vcpu);
660         }
661
662         __set_current_state(TASK_RUNNING);
663         remove_wait_queue(&vcpu->wq, &wait);
664 }
665
666 void kvm_resched(struct kvm_vcpu *vcpu)
667 {
668         if (!need_resched())
669                 return;
670         cond_resched();
671 }
672 EXPORT_SYMBOL_GPL(kvm_resched);
673
674 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
675 {
676         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
677         struct page *page;
678
679         if (vmf->pgoff == 0)
680                 page = virt_to_page(vcpu->run);
681         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
682                 page = virt_to_page(vcpu->arch.pio_data);
683         else
684                 return VM_FAULT_SIGBUS;
685         get_page(page);
686         vmf->page = page;
687         return 0;
688 }
689
690 static struct vm_operations_struct kvm_vcpu_vm_ops = {
691         .fault = kvm_vcpu_fault,
692 };
693
694 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
695 {
696         vma->vm_ops = &kvm_vcpu_vm_ops;
697         return 0;
698 }
699
700 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
701 {
702         struct kvm_vcpu *vcpu = filp->private_data;
703
704         fput(vcpu->kvm->filp);
705         return 0;
706 }
707
708 static struct file_operations kvm_vcpu_fops = {
709         .release        = kvm_vcpu_release,
710         .unlocked_ioctl = kvm_vcpu_ioctl,
711         .compat_ioctl   = kvm_vcpu_ioctl,
712         .mmap           = kvm_vcpu_mmap,
713 };
714
715 /*
716  * Allocates an inode for the vcpu.
717  */
718 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
719 {
720         int fd, r;
721         struct inode *inode;
722         struct file *file;
723
724         r = anon_inode_getfd(&fd, &inode, &file,
725                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
726         if (r)
727                 return r;
728         atomic_inc(&vcpu->kvm->filp->f_count);
729         return fd;
730 }
731
732 /*
733  * Creates some virtual cpus.  Good luck creating more than one.
734  */
735 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
736 {
737         int r;
738         struct kvm_vcpu *vcpu;
739
740         if (!valid_vcpu(n))
741                 return -EINVAL;
742
743         vcpu = kvm_arch_vcpu_create(kvm, n);
744         if (IS_ERR(vcpu))
745                 return PTR_ERR(vcpu);
746
747         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
748
749         r = kvm_arch_vcpu_setup(vcpu);
750         if (r)
751                 goto vcpu_destroy;
752
753         mutex_lock(&kvm->lock);
754         if (kvm->vcpus[n]) {
755                 r = -EEXIST;
756                 mutex_unlock(&kvm->lock);
757                 goto vcpu_destroy;
758         }
759         kvm->vcpus[n] = vcpu;
760         mutex_unlock(&kvm->lock);
761
762         /* Now it's all set up, let userspace reach it */
763         r = create_vcpu_fd(vcpu);
764         if (r < 0)
765                 goto unlink;
766         return r;
767
768 unlink:
769         mutex_lock(&kvm->lock);
770         kvm->vcpus[n] = NULL;
771         mutex_unlock(&kvm->lock);
772 vcpu_destroy:
773         kvm_arch_vcpu_destroy(vcpu);
774         return r;
775 }
776
777 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
778 {
779         if (sigset) {
780                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
781                 vcpu->sigset_active = 1;
782                 vcpu->sigset = *sigset;
783         } else
784                 vcpu->sigset_active = 0;
785         return 0;
786 }
787
788 static long kvm_vcpu_ioctl(struct file *filp,
789                            unsigned int ioctl, unsigned long arg)
790 {
791         struct kvm_vcpu *vcpu = filp->private_data;
792         void __user *argp = (void __user *)arg;
793         int r;
794
795         if (vcpu->kvm->mm != current->mm)
796                 return -EIO;
797         switch (ioctl) {
798         case KVM_RUN:
799                 r = -EINVAL;
800                 if (arg)
801                         goto out;
802                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
803                 break;
804         case KVM_GET_REGS: {
805                 struct kvm_regs kvm_regs;
806
807                 memset(&kvm_regs, 0, sizeof kvm_regs);
808                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
809                 if (r)
810                         goto out;
811                 r = -EFAULT;
812                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
813                         goto out;
814                 r = 0;
815                 break;
816         }
817         case KVM_SET_REGS: {
818                 struct kvm_regs kvm_regs;
819
820                 r = -EFAULT;
821                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
822                         goto out;
823                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
824                 if (r)
825                         goto out;
826                 r = 0;
827                 break;
828         }
829         case KVM_GET_SREGS: {
830                 struct kvm_sregs kvm_sregs;
831
832                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
833                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
834                 if (r)
835                         goto out;
836                 r = -EFAULT;
837                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
838                         goto out;
839                 r = 0;
840                 break;
841         }
842         case KVM_SET_SREGS: {
843                 struct kvm_sregs kvm_sregs;
844
845                 r = -EFAULT;
846                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
847                         goto out;
848                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
849                 if (r)
850                         goto out;
851                 r = 0;
852                 break;
853         }
854         case KVM_TRANSLATE: {
855                 struct kvm_translation tr;
856
857                 r = -EFAULT;
858                 if (copy_from_user(&tr, argp, sizeof tr))
859                         goto out;
860                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
861                 if (r)
862                         goto out;
863                 r = -EFAULT;
864                 if (copy_to_user(argp, &tr, sizeof tr))
865                         goto out;
866                 r = 0;
867                 break;
868         }
869         case KVM_DEBUG_GUEST: {
870                 struct kvm_debug_guest dbg;
871
872                 r = -EFAULT;
873                 if (copy_from_user(&dbg, argp, sizeof dbg))
874                         goto out;
875                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
876                 if (r)
877                         goto out;
878                 r = 0;
879                 break;
880         }
881         case KVM_SET_SIGNAL_MASK: {
882                 struct kvm_signal_mask __user *sigmask_arg = argp;
883                 struct kvm_signal_mask kvm_sigmask;
884                 sigset_t sigset, *p;
885
886                 p = NULL;
887                 if (argp) {
888                         r = -EFAULT;
889                         if (copy_from_user(&kvm_sigmask, argp,
890                                            sizeof kvm_sigmask))
891                                 goto out;
892                         r = -EINVAL;
893                         if (kvm_sigmask.len != sizeof sigset)
894                                 goto out;
895                         r = -EFAULT;
896                         if (copy_from_user(&sigset, sigmask_arg->sigset,
897                                            sizeof sigset))
898                                 goto out;
899                         p = &sigset;
900                 }
901                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
902                 break;
903         }
904         case KVM_GET_FPU: {
905                 struct kvm_fpu fpu;
906
907                 memset(&fpu, 0, sizeof fpu);
908                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
909                 if (r)
910                         goto out;
911                 r = -EFAULT;
912                 if (copy_to_user(argp, &fpu, sizeof fpu))
913                         goto out;
914                 r = 0;
915                 break;
916         }
917         case KVM_SET_FPU: {
918                 struct kvm_fpu fpu;
919
920                 r = -EFAULT;
921                 if (copy_from_user(&fpu, argp, sizeof fpu))
922                         goto out;
923                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
924                 if (r)
925                         goto out;
926                 r = 0;
927                 break;
928         }
929         default:
930                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
931         }
932 out:
933         return r;
934 }
935
936 static long kvm_vm_ioctl(struct file *filp,
937                            unsigned int ioctl, unsigned long arg)
938 {
939         struct kvm *kvm = filp->private_data;
940         void __user *argp = (void __user *)arg;
941         int r;
942
943         if (kvm->mm != current->mm)
944                 return -EIO;
945         switch (ioctl) {
946         case KVM_CREATE_VCPU:
947                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
948                 if (r < 0)
949                         goto out;
950                 break;
951         case KVM_SET_USER_MEMORY_REGION: {
952                 struct kvm_userspace_memory_region kvm_userspace_mem;
953
954                 r = -EFAULT;
955                 if (copy_from_user(&kvm_userspace_mem, argp,
956                                                 sizeof kvm_userspace_mem))
957                         goto out;
958
959                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
960                 if (r)
961                         goto out;
962                 break;
963         }
964         case KVM_GET_DIRTY_LOG: {
965                 struct kvm_dirty_log log;
966
967                 r = -EFAULT;
968                 if (copy_from_user(&log, argp, sizeof log))
969                         goto out;
970                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
971                 if (r)
972                         goto out;
973                 break;
974         }
975         default:
976                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
977         }
978 out:
979         return r;
980 }
981
982 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
983 {
984         struct kvm *kvm = vma->vm_file->private_data;
985         struct page *page;
986
987         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
988                 return VM_FAULT_SIGBUS;
989         page = gfn_to_page(kvm, vmf->pgoff);
990         if (is_error_page(page)) {
991                 kvm_release_page_clean(page);
992                 return VM_FAULT_SIGBUS;
993         }
994         vmf->page = page;
995         return 0;
996 }
997
998 static struct vm_operations_struct kvm_vm_vm_ops = {
999         .fault = kvm_vm_fault,
1000 };
1001
1002 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1003 {
1004         vma->vm_ops = &kvm_vm_vm_ops;
1005         return 0;
1006 }
1007
1008 static struct file_operations kvm_vm_fops = {
1009         .release        = kvm_vm_release,
1010         .unlocked_ioctl = kvm_vm_ioctl,
1011         .compat_ioctl   = kvm_vm_ioctl,
1012         .mmap           = kvm_vm_mmap,
1013 };
1014
1015 static int kvm_dev_ioctl_create_vm(void)
1016 {
1017         int fd, r;
1018         struct inode *inode;
1019         struct file *file;
1020         struct kvm *kvm;
1021
1022         kvm = kvm_create_vm();
1023         if (IS_ERR(kvm))
1024                 return PTR_ERR(kvm);
1025         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1026         if (r) {
1027                 kvm_destroy_vm(kvm);
1028                 return r;
1029         }
1030
1031         kvm->filp = file;
1032
1033         return fd;
1034 }
1035
1036 static long kvm_dev_ioctl(struct file *filp,
1037                           unsigned int ioctl, unsigned long arg)
1038 {
1039         void __user *argp = (void __user *)arg;
1040         long r = -EINVAL;
1041
1042         switch (ioctl) {
1043         case KVM_GET_API_VERSION:
1044                 r = -EINVAL;
1045                 if (arg)
1046                         goto out;
1047                 r = KVM_API_VERSION;
1048                 break;
1049         case KVM_CREATE_VM:
1050                 r = -EINVAL;
1051                 if (arg)
1052                         goto out;
1053                 r = kvm_dev_ioctl_create_vm();
1054                 break;
1055         case KVM_CHECK_EXTENSION:
1056                 r = kvm_dev_ioctl_check_extension((long)argp);
1057                 break;
1058         case KVM_GET_VCPU_MMAP_SIZE:
1059                 r = -EINVAL;
1060                 if (arg)
1061                         goto out;
1062                 r = 2 * PAGE_SIZE;
1063                 break;
1064         default:
1065                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1066         }
1067 out:
1068         return r;
1069 }
1070
1071 static struct file_operations kvm_chardev_ops = {
1072         .unlocked_ioctl = kvm_dev_ioctl,
1073         .compat_ioctl   = kvm_dev_ioctl,
1074 };
1075
1076 static struct miscdevice kvm_dev = {
1077         KVM_MINOR,
1078         "kvm",
1079         &kvm_chardev_ops,
1080 };
1081
1082 static void hardware_enable(void *junk)
1083 {
1084         int cpu = raw_smp_processor_id();
1085
1086         if (cpu_isset(cpu, cpus_hardware_enabled))
1087                 return;
1088         cpu_set(cpu, cpus_hardware_enabled);
1089         kvm_arch_hardware_enable(NULL);
1090 }
1091
1092 static void hardware_disable(void *junk)
1093 {
1094         int cpu = raw_smp_processor_id();
1095
1096         if (!cpu_isset(cpu, cpus_hardware_enabled))
1097                 return;
1098         cpu_clear(cpu, cpus_hardware_enabled);
1099         decache_vcpus_on_cpu(cpu);
1100         kvm_arch_hardware_disable(NULL);
1101 }
1102
1103 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1104                            void *v)
1105 {
1106         int cpu = (long)v;
1107
1108         val &= ~CPU_TASKS_FROZEN;
1109         switch (val) {
1110         case CPU_DYING:
1111                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1112                        cpu);
1113                 hardware_disable(NULL);
1114                 break;
1115         case CPU_UP_CANCELED:
1116                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1117                        cpu);
1118                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1119                 break;
1120         case CPU_ONLINE:
1121                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1122                        cpu);
1123                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1124                 break;
1125         }
1126         return NOTIFY_OK;
1127 }
1128
1129 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1130                       void *v)
1131 {
1132         if (val == SYS_RESTART) {
1133                 /*
1134                  * Some (well, at least mine) BIOSes hang on reboot if
1135                  * in vmx root mode.
1136                  */
1137                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1138                 on_each_cpu(hardware_disable, NULL, 0, 1);
1139         }
1140         return NOTIFY_OK;
1141 }
1142
1143 static struct notifier_block kvm_reboot_notifier = {
1144         .notifier_call = kvm_reboot,
1145         .priority = 0,
1146 };
1147
1148 void kvm_io_bus_init(struct kvm_io_bus *bus)
1149 {
1150         memset(bus, 0, sizeof(*bus));
1151 }
1152
1153 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1154 {
1155         int i;
1156
1157         for (i = 0; i < bus->dev_count; i++) {
1158                 struct kvm_io_device *pos = bus->devs[i];
1159
1160                 kvm_iodevice_destructor(pos);
1161         }
1162 }
1163
1164 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1165 {
1166         int i;
1167
1168         for (i = 0; i < bus->dev_count; i++) {
1169                 struct kvm_io_device *pos = bus->devs[i];
1170
1171                 if (pos->in_range(pos, addr))
1172                         return pos;
1173         }
1174
1175         return NULL;
1176 }
1177
1178 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1179 {
1180         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1181
1182         bus->devs[bus->dev_count++] = dev;
1183 }
1184
1185 static struct notifier_block kvm_cpu_notifier = {
1186         .notifier_call = kvm_cpu_hotplug,
1187         .priority = 20, /* must be > scheduler priority */
1188 };
1189
1190 static int vm_stat_get(void *_offset, u64 *val)
1191 {
1192         unsigned offset = (long)_offset;
1193         struct kvm *kvm;
1194
1195         *val = 0;
1196         spin_lock(&kvm_lock);
1197         list_for_each_entry(kvm, &vm_list, vm_list)
1198                 *val += *(u32 *)((void *)kvm + offset);
1199         spin_unlock(&kvm_lock);
1200         return 0;
1201 }
1202
1203 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1204
1205 static int vcpu_stat_get(void *_offset, u64 *val)
1206 {
1207         unsigned offset = (long)_offset;
1208         struct kvm *kvm;
1209         struct kvm_vcpu *vcpu;
1210         int i;
1211
1212         *val = 0;
1213         spin_lock(&kvm_lock);
1214         list_for_each_entry(kvm, &vm_list, vm_list)
1215                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1216                         vcpu = kvm->vcpus[i];
1217                         if (vcpu)
1218                                 *val += *(u32 *)((void *)vcpu + offset);
1219                 }
1220         spin_unlock(&kvm_lock);
1221         return 0;
1222 }
1223
1224 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1225
1226 static struct file_operations *stat_fops[] = {
1227         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1228         [KVM_STAT_VM]   = &vm_stat_fops,
1229 };
1230
1231 static void kvm_init_debug(void)
1232 {
1233         struct kvm_stats_debugfs_item *p;
1234
1235         debugfs_dir = debugfs_create_dir("kvm", NULL);
1236         for (p = debugfs_entries; p->name; ++p)
1237                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1238                                                 (void *)(long)p->offset,
1239                                                 stat_fops[p->kind]);
1240 }
1241
1242 static void kvm_exit_debug(void)
1243 {
1244         struct kvm_stats_debugfs_item *p;
1245
1246         for (p = debugfs_entries; p->name; ++p)
1247                 debugfs_remove(p->dentry);
1248         debugfs_remove(debugfs_dir);
1249 }
1250
1251 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1252 {
1253         hardware_disable(NULL);
1254         return 0;
1255 }
1256
1257 static int kvm_resume(struct sys_device *dev)
1258 {
1259         hardware_enable(NULL);
1260         return 0;
1261 }
1262
1263 static struct sysdev_class kvm_sysdev_class = {
1264         .name = "kvm",
1265         .suspend = kvm_suspend,
1266         .resume = kvm_resume,
1267 };
1268
1269 static struct sys_device kvm_sysdev = {
1270         .id = 0,
1271         .cls = &kvm_sysdev_class,
1272 };
1273
1274 struct page *bad_page;
1275
1276 static inline
1277 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1278 {
1279         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1280 }
1281
1282 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1283 {
1284         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1285
1286         kvm_arch_vcpu_load(vcpu, cpu);
1287 }
1288
1289 static void kvm_sched_out(struct preempt_notifier *pn,
1290                           struct task_struct *next)
1291 {
1292         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1293
1294         kvm_arch_vcpu_put(vcpu);
1295 }
1296
1297 int kvm_init(void *opaque, unsigned int vcpu_size,
1298                   struct module *module)
1299 {
1300         int r;
1301         int cpu;
1302
1303         kvm_init_debug();
1304
1305         r = kvm_arch_init(opaque);
1306         if (r)
1307                 goto out_fail;
1308
1309         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1310
1311         if (bad_page == NULL) {
1312                 r = -ENOMEM;
1313                 goto out;
1314         }
1315
1316         r = kvm_arch_hardware_setup();
1317         if (r < 0)
1318                 goto out_free_0;
1319
1320         for_each_online_cpu(cpu) {
1321                 smp_call_function_single(cpu,
1322                                 kvm_arch_check_processor_compat,
1323                                 &r, 0, 1);
1324                 if (r < 0)
1325                         goto out_free_1;
1326         }
1327
1328         on_each_cpu(hardware_enable, NULL, 0, 1);
1329         r = register_cpu_notifier(&kvm_cpu_notifier);
1330         if (r)
1331                 goto out_free_2;
1332         register_reboot_notifier(&kvm_reboot_notifier);
1333
1334         r = sysdev_class_register(&kvm_sysdev_class);
1335         if (r)
1336                 goto out_free_3;
1337
1338         r = sysdev_register(&kvm_sysdev);
1339         if (r)
1340                 goto out_free_4;
1341
1342         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1343         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1344                                            __alignof__(struct kvm_vcpu),
1345                                            0, NULL);
1346         if (!kvm_vcpu_cache) {
1347                 r = -ENOMEM;
1348                 goto out_free_5;
1349         }
1350
1351         kvm_chardev_ops.owner = module;
1352
1353         r = misc_register(&kvm_dev);
1354         if (r) {
1355                 printk(KERN_ERR "kvm: misc device register failed\n");
1356                 goto out_free;
1357         }
1358
1359         kvm_preempt_ops.sched_in = kvm_sched_in;
1360         kvm_preempt_ops.sched_out = kvm_sched_out;
1361
1362         return 0;
1363
1364 out_free:
1365         kmem_cache_destroy(kvm_vcpu_cache);
1366 out_free_5:
1367         sysdev_unregister(&kvm_sysdev);
1368 out_free_4:
1369         sysdev_class_unregister(&kvm_sysdev_class);
1370 out_free_3:
1371         unregister_reboot_notifier(&kvm_reboot_notifier);
1372         unregister_cpu_notifier(&kvm_cpu_notifier);
1373 out_free_2:
1374         on_each_cpu(hardware_disable, NULL, 0, 1);
1375 out_free_1:
1376         kvm_arch_hardware_unsetup();
1377 out_free_0:
1378         __free_page(bad_page);
1379 out:
1380         kvm_arch_exit();
1381         kvm_exit_debug();
1382 out_fail:
1383         return r;
1384 }
1385 EXPORT_SYMBOL_GPL(kvm_init);
1386
1387 void kvm_exit(void)
1388 {
1389         misc_deregister(&kvm_dev);
1390         kmem_cache_destroy(kvm_vcpu_cache);
1391         sysdev_unregister(&kvm_sysdev);
1392         sysdev_class_unregister(&kvm_sysdev_class);
1393         unregister_reboot_notifier(&kvm_reboot_notifier);
1394         unregister_cpu_notifier(&kvm_cpu_notifier);
1395         on_each_cpu(hardware_disable, NULL, 0, 1);
1396         kvm_arch_hardware_unsetup();
1397         kvm_arch_exit();
1398         kvm_exit_debug();
1399         __free_page(bad_page);
1400 }
1401 EXPORT_SYMBOL_GPL(kvm_exit);