KVM: MSI to INTx translate
[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 #include <linux/swap.h>
44
45 #include <asm/processor.h>
46 #include <asm/io.h>
47 #include <asm/uaccess.h>
48 #include <asm/pgtable.h>
49
50 #ifdef CONFIG_X86
51 #include <asm/msidef.h>
52 #endif
53
54 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
55 #include "coalesced_mmio.h"
56 #endif
57
58 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
59 #include <linux/pci.h>
60 #include <linux/interrupt.h>
61 #include "irq.h"
62 #endif
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 static int msi2intx = 1;
68 module_param(msi2intx, bool, 0);
69
70 DEFINE_SPINLOCK(kvm_lock);
71 LIST_HEAD(vm_list);
72
73 static cpumask_t cpus_hardware_enabled;
74
75 struct kmem_cache *kvm_vcpu_cache;
76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
77
78 static __read_mostly struct preempt_ops kvm_preempt_ops;
79
80 struct dentry *kvm_debugfs_dir;
81
82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
83                            unsigned long arg);
84
85 bool kvm_rebooting;
86
87 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
88
89 #ifdef CONFIG_X86
90 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev)
91 {
92         int vcpu_id;
93         struct kvm_vcpu *vcpu;
94         struct kvm_ioapic *ioapic = ioapic_irqchip(dev->kvm);
95         int dest_id = (dev->guest_msi.address_lo & MSI_ADDR_DEST_ID_MASK)
96                         >> MSI_ADDR_DEST_ID_SHIFT;
97         int vector = (dev->guest_msi.data & MSI_DATA_VECTOR_MASK)
98                         >> MSI_DATA_VECTOR_SHIFT;
99         int dest_mode = test_bit(MSI_ADDR_DEST_MODE_SHIFT,
100                                 (unsigned long *)&dev->guest_msi.address_lo);
101         int trig_mode = test_bit(MSI_DATA_TRIGGER_SHIFT,
102                                 (unsigned long *)&dev->guest_msi.data);
103         int delivery_mode = test_bit(MSI_DATA_DELIVERY_MODE_SHIFT,
104                                 (unsigned long *)&dev->guest_msi.data);
105         u32 deliver_bitmask;
106
107         BUG_ON(!ioapic);
108
109         deliver_bitmask = kvm_ioapic_get_delivery_bitmask(ioapic,
110                                 dest_id, dest_mode);
111         /* IOAPIC delivery mode value is the same as MSI here */
112         switch (delivery_mode) {
113         case IOAPIC_LOWEST_PRIORITY:
114                 vcpu = kvm_get_lowest_prio_vcpu(ioapic->kvm, vector,
115                                 deliver_bitmask);
116                 if (vcpu != NULL)
117                         kvm_apic_set_irq(vcpu, vector, trig_mode);
118                 else
119                         printk(KERN_INFO "kvm: null lowest priority vcpu!\n");
120                 break;
121         case IOAPIC_FIXED:
122                 for (vcpu_id = 0; deliver_bitmask != 0; vcpu_id++) {
123                         if (!(deliver_bitmask & (1 << vcpu_id)))
124                                 continue;
125                         deliver_bitmask &= ~(1 << vcpu_id);
126                         vcpu = ioapic->kvm->vcpus[vcpu_id];
127                         if (vcpu)
128                                 kvm_apic_set_irq(vcpu, vector, trig_mode);
129                 }
130                 break;
131         default:
132                 printk(KERN_INFO "kvm: unsupported MSI delivery mode\n");
133         }
134 }
135 #else
136 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev) {}
137 #endif
138
139 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
140                                                       int assigned_dev_id)
141 {
142         struct list_head *ptr;
143         struct kvm_assigned_dev_kernel *match;
144
145         list_for_each(ptr, head) {
146                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
147                 if (match->assigned_dev_id == assigned_dev_id)
148                         return match;
149         }
150         return NULL;
151 }
152
153 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
154 {
155         struct kvm_assigned_dev_kernel *assigned_dev;
156
157         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
158                                     interrupt_work);
159
160         /* This is taken to safely inject irq inside the guest. When
161          * the interrupt injection (or the ioapic code) uses a
162          * finer-grained lock, update this
163          */
164         mutex_lock(&assigned_dev->kvm->lock);
165         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_INTX)
166                 kvm_set_irq(assigned_dev->kvm,
167                             assigned_dev->irq_source_id,
168                             assigned_dev->guest_irq, 1);
169         else if (assigned_dev->irq_requested_type &
170                                 KVM_ASSIGNED_DEV_GUEST_MSI) {
171                 assigned_device_msi_dispatch(assigned_dev);
172                 enable_irq(assigned_dev->host_irq);
173         }
174         mutex_unlock(&assigned_dev->kvm->lock);
175         kvm_put_kvm(assigned_dev->kvm);
176 }
177
178 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
179 {
180         struct kvm_assigned_dev_kernel *assigned_dev =
181                 (struct kvm_assigned_dev_kernel *) dev_id;
182
183         kvm_get_kvm(assigned_dev->kvm);
184         schedule_work(&assigned_dev->interrupt_work);
185         disable_irq_nosync(irq);
186         return IRQ_HANDLED;
187 }
188
189 /* Ack the irq line for an assigned device */
190 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
191 {
192         struct kvm_assigned_dev_kernel *dev;
193
194         if (kian->gsi == -1)
195                 return;
196
197         dev = container_of(kian, struct kvm_assigned_dev_kernel,
198                            ack_notifier);
199         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
200         enable_irq(dev->host_irq);
201 }
202
203 static void kvm_free_assigned_device(struct kvm *kvm,
204                                      struct kvm_assigned_dev_kernel
205                                      *assigned_dev)
206 {
207         if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested_type)
208                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
209         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
210                 pci_disable_msi(assigned_dev->dev);
211
212         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
213         kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
214
215         if (cancel_work_sync(&assigned_dev->interrupt_work))
216                 /* We had pending work. That means we will have to take
217                  * care of kvm_put_kvm.
218                  */
219                 kvm_put_kvm(kvm);
220
221         pci_reset_function(assigned_dev->dev);
222
223         pci_release_regions(assigned_dev->dev);
224         pci_disable_device(assigned_dev->dev);
225         pci_dev_put(assigned_dev->dev);
226
227         list_del(&assigned_dev->list);
228         kfree(assigned_dev);
229 }
230
231 void kvm_free_all_assigned_devices(struct kvm *kvm)
232 {
233         struct list_head *ptr, *ptr2;
234         struct kvm_assigned_dev_kernel *assigned_dev;
235
236         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
237                 assigned_dev = list_entry(ptr,
238                                           struct kvm_assigned_dev_kernel,
239                                           list);
240
241                 kvm_free_assigned_device(kvm, assigned_dev);
242         }
243 }
244
245 static int assigned_device_update_intx(struct kvm *kvm,
246                         struct kvm_assigned_dev_kernel *adev,
247                         struct kvm_assigned_irq *airq)
248 {
249         adev->guest_irq = airq->guest_irq;
250         adev->ack_notifier.gsi = airq->guest_irq;
251
252         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
253                 return 0;
254
255         if (irqchip_in_kernel(kvm)) {
256                 if (!msi2intx &&
257                     adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) {
258                         free_irq(adev->host_irq, (void *)kvm);
259                         pci_disable_msi(adev->dev);
260                 }
261
262                 if (!capable(CAP_SYS_RAWIO))
263                         return -EPERM;
264
265                 if (airq->host_irq)
266                         adev->host_irq = airq->host_irq;
267                 else
268                         adev->host_irq = adev->dev->irq;
269
270                 /* Even though this is PCI, we don't want to use shared
271                  * interrupts. Sharing host devices with guest-assigned devices
272                  * on the same interrupt line is not a happy situation: there
273                  * are going to be long delays in accepting, acking, etc.
274                  */
275                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
276                                 0, "kvm_assigned_intx_device", (void *)adev))
277                         return -EIO;
278         }
279
280         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
281                                    KVM_ASSIGNED_DEV_HOST_INTX;
282         return 0;
283 }
284
285 #ifdef CONFIG_X86
286 static int assigned_device_update_msi(struct kvm *kvm,
287                         struct kvm_assigned_dev_kernel *adev,
288                         struct kvm_assigned_irq *airq)
289 {
290         int r;
291
292         if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
293                 /* x86 don't care upper address of guest msi message addr */
294                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
295                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
296                 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
297                 adev->guest_msi.data = airq->guest_msi.data;
298                 adev->ack_notifier.gsi = -1;
299         } else if (msi2intx) {
300                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
301                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
302                 adev->guest_irq = airq->guest_irq;
303                 adev->ack_notifier.gsi = airq->guest_irq;
304         }
305
306         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
307                 return 0;
308
309         if (irqchip_in_kernel(kvm)) {
310                 if (!msi2intx) {
311                         if (adev->irq_requested_type &
312                                         KVM_ASSIGNED_DEV_HOST_INTX)
313                                 free_irq(adev->host_irq, (void *)adev);
314
315                         r = pci_enable_msi(adev->dev);
316                         if (r)
317                                 return r;
318                 }
319
320                 adev->host_irq = adev->dev->irq;
321                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
322                                 "kvm_assigned_msi_device", (void *)adev))
323                         return -EIO;
324         }
325
326         if (!msi2intx)
327                 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
328
329         adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
330         return 0;
331 }
332 #endif
333
334 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
335                                    struct kvm_assigned_irq
336                                    *assigned_irq)
337 {
338         int r = 0;
339         struct kvm_assigned_dev_kernel *match;
340
341         mutex_lock(&kvm->lock);
342
343         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
344                                       assigned_irq->assigned_dev_id);
345         if (!match) {
346                 mutex_unlock(&kvm->lock);
347                 return -EINVAL;
348         }
349
350         if (!match->irq_requested_type) {
351                 INIT_WORK(&match->interrupt_work,
352                                 kvm_assigned_dev_interrupt_work_handler);
353                 if (irqchip_in_kernel(kvm)) {
354                         /* Register ack nofitier */
355                         match->ack_notifier.gsi = -1;
356                         match->ack_notifier.irq_acked =
357                                         kvm_assigned_dev_ack_irq;
358                         kvm_register_irq_ack_notifier(kvm,
359                                         &match->ack_notifier);
360
361                         /* Request IRQ source ID */
362                         r = kvm_request_irq_source_id(kvm);
363                         if (r < 0)
364                                 goto out_release;
365                         else
366                                 match->irq_source_id = r;
367
368 #ifdef CONFIG_X86
369                         /* Determine host device irq type, we can know the
370                          * result from dev->msi_enabled */
371                         if (msi2intx)
372                                 pci_enable_msi(match->dev);
373 #endif
374                 }
375         }
376
377         if ((!msi2intx &&
378              (assigned_irq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI)) ||
379             (msi2intx && match->dev->msi_enabled)) {
380 #ifdef CONFIG_X86
381                 r = assigned_device_update_msi(kvm, match, assigned_irq);
382                 if (r) {
383                         printk(KERN_WARNING "kvm: failed to enable "
384                                         "MSI device!\n");
385                         goto out_release;
386                 }
387 #else
388                 r = -ENOTTY;
389 #endif
390         } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
391                 /* Host device IRQ 0 means don't support INTx */
392                 if (!msi2intx) {
393                         printk(KERN_WARNING
394                                "kvm: wait device to enable MSI!\n");
395                         r = 0;
396                 } else {
397                         printk(KERN_WARNING
398                                "kvm: failed to enable MSI device!\n");
399                         r = -ENOTTY;
400                         goto out_release;
401                 }
402         } else {
403                 /* Non-sharing INTx mode */
404                 r = assigned_device_update_intx(kvm, match, assigned_irq);
405                 if (r) {
406                         printk(KERN_WARNING "kvm: failed to enable "
407                                         "INTx device!\n");
408                         goto out_release;
409                 }
410         }
411
412         mutex_unlock(&kvm->lock);
413         return r;
414 out_release:
415         mutex_unlock(&kvm->lock);
416         kvm_free_assigned_device(kvm, match);
417         return r;
418 }
419
420 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
421                                       struct kvm_assigned_pci_dev *assigned_dev)
422 {
423         int r = 0;
424         struct kvm_assigned_dev_kernel *match;
425         struct pci_dev *dev;
426
427         mutex_lock(&kvm->lock);
428
429         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
430                                       assigned_dev->assigned_dev_id);
431         if (match) {
432                 /* device already assigned */
433                 r = -EINVAL;
434                 goto out;
435         }
436
437         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
438         if (match == NULL) {
439                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
440                        __func__);
441                 r = -ENOMEM;
442                 goto out;
443         }
444         dev = pci_get_bus_and_slot(assigned_dev->busnr,
445                                    assigned_dev->devfn);
446         if (!dev) {
447                 printk(KERN_INFO "%s: host device not found\n", __func__);
448                 r = -EINVAL;
449                 goto out_free;
450         }
451         if (pci_enable_device(dev)) {
452                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
453                 r = -EBUSY;
454                 goto out_put;
455         }
456         r = pci_request_regions(dev, "kvm_assigned_device");
457         if (r) {
458                 printk(KERN_INFO "%s: Could not get access to device regions\n",
459                        __func__);
460                 goto out_disable;
461         }
462
463         pci_reset_function(dev);
464
465         match->assigned_dev_id = assigned_dev->assigned_dev_id;
466         match->host_busnr = assigned_dev->busnr;
467         match->host_devfn = assigned_dev->devfn;
468         match->dev = dev;
469
470         match->kvm = kvm;
471
472         list_add(&match->list, &kvm->arch.assigned_dev_head);
473
474         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
475                 r = kvm_iommu_map_guest(kvm, match);
476                 if (r)
477                         goto out_list_del;
478         }
479
480 out:
481         mutex_unlock(&kvm->lock);
482         return r;
483 out_list_del:
484         list_del(&match->list);
485         pci_release_regions(dev);
486 out_disable:
487         pci_disable_device(dev);
488 out_put:
489         pci_dev_put(dev);
490 out_free:
491         kfree(match);
492         mutex_unlock(&kvm->lock);
493         return r;
494 }
495 #endif
496
497 static inline int valid_vcpu(int n)
498 {
499         return likely(n >= 0 && n < KVM_MAX_VCPUS);
500 }
501
502 inline int kvm_is_mmio_pfn(pfn_t pfn)
503 {
504         if (pfn_valid(pfn))
505                 return PageReserved(pfn_to_page(pfn));
506
507         return true;
508 }
509
510 /*
511  * Switches to specified vcpu, until a matching vcpu_put()
512  */
513 void vcpu_load(struct kvm_vcpu *vcpu)
514 {
515         int cpu;
516
517         mutex_lock(&vcpu->mutex);
518         cpu = get_cpu();
519         preempt_notifier_register(&vcpu->preempt_notifier);
520         kvm_arch_vcpu_load(vcpu, cpu);
521         put_cpu();
522 }
523
524 void vcpu_put(struct kvm_vcpu *vcpu)
525 {
526         preempt_disable();
527         kvm_arch_vcpu_put(vcpu);
528         preempt_notifier_unregister(&vcpu->preempt_notifier);
529         preempt_enable();
530         mutex_unlock(&vcpu->mutex);
531 }
532
533 static void ack_flush(void *_completed)
534 {
535 }
536
537 void kvm_flush_remote_tlbs(struct kvm *kvm)
538 {
539         int i, cpu, me;
540         cpumask_t cpus;
541         struct kvm_vcpu *vcpu;
542
543         me = get_cpu();
544         cpus_clear(cpus);
545         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
546                 vcpu = kvm->vcpus[i];
547                 if (!vcpu)
548                         continue;
549                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
550                         continue;
551                 cpu = vcpu->cpu;
552                 if (cpu != -1 && cpu != me)
553                         cpu_set(cpu, cpus);
554         }
555         if (cpus_empty(cpus))
556                 goto out;
557         ++kvm->stat.remote_tlb_flush;
558         smp_call_function_mask(cpus, ack_flush, NULL, 1);
559 out:
560         put_cpu();
561 }
562
563 void kvm_reload_remote_mmus(struct kvm *kvm)
564 {
565         int i, cpu, me;
566         cpumask_t cpus;
567         struct kvm_vcpu *vcpu;
568
569         me = get_cpu();
570         cpus_clear(cpus);
571         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
572                 vcpu = kvm->vcpus[i];
573                 if (!vcpu)
574                         continue;
575                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
576                         continue;
577                 cpu = vcpu->cpu;
578                 if (cpu != -1 && cpu != me)
579                         cpu_set(cpu, cpus);
580         }
581         if (cpus_empty(cpus))
582                 goto out;
583         smp_call_function_mask(cpus, ack_flush, NULL, 1);
584 out:
585         put_cpu();
586 }
587
588
589 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
590 {
591         struct page *page;
592         int r;
593
594         mutex_init(&vcpu->mutex);
595         vcpu->cpu = -1;
596         vcpu->kvm = kvm;
597         vcpu->vcpu_id = id;
598         init_waitqueue_head(&vcpu->wq);
599
600         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
601         if (!page) {
602                 r = -ENOMEM;
603                 goto fail;
604         }
605         vcpu->run = page_address(page);
606
607         r = kvm_arch_vcpu_init(vcpu);
608         if (r < 0)
609                 goto fail_free_run;
610         return 0;
611
612 fail_free_run:
613         free_page((unsigned long)vcpu->run);
614 fail:
615         return r;
616 }
617 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
618
619 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
620 {
621         kvm_arch_vcpu_uninit(vcpu);
622         free_page((unsigned long)vcpu->run);
623 }
624 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
625
626 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
627 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
628 {
629         return container_of(mn, struct kvm, mmu_notifier);
630 }
631
632 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
633                                              struct mm_struct *mm,
634                                              unsigned long address)
635 {
636         struct kvm *kvm = mmu_notifier_to_kvm(mn);
637         int need_tlb_flush;
638
639         /*
640          * When ->invalidate_page runs, the linux pte has been zapped
641          * already but the page is still allocated until
642          * ->invalidate_page returns. So if we increase the sequence
643          * here the kvm page fault will notice if the spte can't be
644          * established because the page is going to be freed. If
645          * instead the kvm page fault establishes the spte before
646          * ->invalidate_page runs, kvm_unmap_hva will release it
647          * before returning.
648          *
649          * The sequence increase only need to be seen at spin_unlock
650          * time, and not at spin_lock time.
651          *
652          * Increasing the sequence after the spin_unlock would be
653          * unsafe because the kvm page fault could then establish the
654          * pte after kvm_unmap_hva returned, without noticing the page
655          * is going to be freed.
656          */
657         spin_lock(&kvm->mmu_lock);
658         kvm->mmu_notifier_seq++;
659         need_tlb_flush = kvm_unmap_hva(kvm, address);
660         spin_unlock(&kvm->mmu_lock);
661
662         /* we've to flush the tlb before the pages can be freed */
663         if (need_tlb_flush)
664                 kvm_flush_remote_tlbs(kvm);
665
666 }
667
668 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
669                                                     struct mm_struct *mm,
670                                                     unsigned long start,
671                                                     unsigned long end)
672 {
673         struct kvm *kvm = mmu_notifier_to_kvm(mn);
674         int need_tlb_flush = 0;
675
676         spin_lock(&kvm->mmu_lock);
677         /*
678          * The count increase must become visible at unlock time as no
679          * spte can be established without taking the mmu_lock and
680          * count is also read inside the mmu_lock critical section.
681          */
682         kvm->mmu_notifier_count++;
683         for (; start < end; start += PAGE_SIZE)
684                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
685         spin_unlock(&kvm->mmu_lock);
686
687         /* we've to flush the tlb before the pages can be freed */
688         if (need_tlb_flush)
689                 kvm_flush_remote_tlbs(kvm);
690 }
691
692 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
693                                                   struct mm_struct *mm,
694                                                   unsigned long start,
695                                                   unsigned long end)
696 {
697         struct kvm *kvm = mmu_notifier_to_kvm(mn);
698
699         spin_lock(&kvm->mmu_lock);
700         /*
701          * This sequence increase will notify the kvm page fault that
702          * the page that is going to be mapped in the spte could have
703          * been freed.
704          */
705         kvm->mmu_notifier_seq++;
706         /*
707          * The above sequence increase must be visible before the
708          * below count decrease but both values are read by the kvm
709          * page fault under mmu_lock spinlock so we don't need to add
710          * a smb_wmb() here in between the two.
711          */
712         kvm->mmu_notifier_count--;
713         spin_unlock(&kvm->mmu_lock);
714
715         BUG_ON(kvm->mmu_notifier_count < 0);
716 }
717
718 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
719                                               struct mm_struct *mm,
720                                               unsigned long address)
721 {
722         struct kvm *kvm = mmu_notifier_to_kvm(mn);
723         int young;
724
725         spin_lock(&kvm->mmu_lock);
726         young = kvm_age_hva(kvm, address);
727         spin_unlock(&kvm->mmu_lock);
728
729         if (young)
730                 kvm_flush_remote_tlbs(kvm);
731
732         return young;
733 }
734
735 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
736         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
737         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
738         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
739         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
740 };
741 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
742
743 static struct kvm *kvm_create_vm(void)
744 {
745         struct kvm *kvm = kvm_arch_create_vm();
746 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
747         struct page *page;
748 #endif
749
750         if (IS_ERR(kvm))
751                 goto out;
752
753 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
754         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
755         if (!page) {
756                 kfree(kvm);
757                 return ERR_PTR(-ENOMEM);
758         }
759         kvm->coalesced_mmio_ring =
760                         (struct kvm_coalesced_mmio_ring *)page_address(page);
761 #endif
762
763 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
764         {
765                 int err;
766                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
767                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
768                 if (err) {
769 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
770                         put_page(page);
771 #endif
772                         kfree(kvm);
773                         return ERR_PTR(err);
774                 }
775         }
776 #endif
777
778         kvm->mm = current->mm;
779         atomic_inc(&kvm->mm->mm_count);
780         spin_lock_init(&kvm->mmu_lock);
781         kvm_io_bus_init(&kvm->pio_bus);
782         mutex_init(&kvm->lock);
783         kvm_io_bus_init(&kvm->mmio_bus);
784         init_rwsem(&kvm->slots_lock);
785         atomic_set(&kvm->users_count, 1);
786         spin_lock(&kvm_lock);
787         list_add(&kvm->vm_list, &vm_list);
788         spin_unlock(&kvm_lock);
789 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
790         kvm_coalesced_mmio_init(kvm);
791 #endif
792 out:
793         return kvm;
794 }
795
796 /*
797  * Free any memory in @free but not in @dont.
798  */
799 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
800                                   struct kvm_memory_slot *dont)
801 {
802         if (!dont || free->rmap != dont->rmap)
803                 vfree(free->rmap);
804
805         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
806                 vfree(free->dirty_bitmap);
807
808         if (!dont || free->lpage_info != dont->lpage_info)
809                 vfree(free->lpage_info);
810
811         free->npages = 0;
812         free->dirty_bitmap = NULL;
813         free->rmap = NULL;
814         free->lpage_info = NULL;
815 }
816
817 void kvm_free_physmem(struct kvm *kvm)
818 {
819         int i;
820
821         for (i = 0; i < kvm->nmemslots; ++i)
822                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
823 }
824
825 static void kvm_destroy_vm(struct kvm *kvm)
826 {
827         struct mm_struct *mm = kvm->mm;
828
829         spin_lock(&kvm_lock);
830         list_del(&kvm->vm_list);
831         spin_unlock(&kvm_lock);
832         kvm_io_bus_destroy(&kvm->pio_bus);
833         kvm_io_bus_destroy(&kvm->mmio_bus);
834 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
835         if (kvm->coalesced_mmio_ring != NULL)
836                 free_page((unsigned long)kvm->coalesced_mmio_ring);
837 #endif
838 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
839         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
840 #endif
841         kvm_arch_destroy_vm(kvm);
842         mmdrop(mm);
843 }
844
845 void kvm_get_kvm(struct kvm *kvm)
846 {
847         atomic_inc(&kvm->users_count);
848 }
849 EXPORT_SYMBOL_GPL(kvm_get_kvm);
850
851 void kvm_put_kvm(struct kvm *kvm)
852 {
853         if (atomic_dec_and_test(&kvm->users_count))
854                 kvm_destroy_vm(kvm);
855 }
856 EXPORT_SYMBOL_GPL(kvm_put_kvm);
857
858
859 static int kvm_vm_release(struct inode *inode, struct file *filp)
860 {
861         struct kvm *kvm = filp->private_data;
862
863         kvm_put_kvm(kvm);
864         return 0;
865 }
866
867 /*
868  * Allocate some memory and give it an address in the guest physical address
869  * space.
870  *
871  * Discontiguous memory is allowed, mostly for framebuffers.
872  *
873  * Must be called holding mmap_sem for write.
874  */
875 int __kvm_set_memory_region(struct kvm *kvm,
876                             struct kvm_userspace_memory_region *mem,
877                             int user_alloc)
878 {
879         int r;
880         gfn_t base_gfn;
881         unsigned long npages;
882         unsigned long i;
883         struct kvm_memory_slot *memslot;
884         struct kvm_memory_slot old, new;
885
886         r = -EINVAL;
887         /* General sanity checks */
888         if (mem->memory_size & (PAGE_SIZE - 1))
889                 goto out;
890         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
891                 goto out;
892         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
893                 goto out;
894         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
895                 goto out;
896         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
897                 goto out;
898
899         memslot = &kvm->memslots[mem->slot];
900         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
901         npages = mem->memory_size >> PAGE_SHIFT;
902
903         if (!npages)
904                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
905
906         new = old = *memslot;
907
908         new.base_gfn = base_gfn;
909         new.npages = npages;
910         new.flags = mem->flags;
911
912         /* Disallow changing a memory slot's size. */
913         r = -EINVAL;
914         if (npages && old.npages && npages != old.npages)
915                 goto out_free;
916
917         /* Check for overlaps */
918         r = -EEXIST;
919         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
920                 struct kvm_memory_slot *s = &kvm->memslots[i];
921
922                 if (s == memslot)
923                         continue;
924                 if (!((base_gfn + npages <= s->base_gfn) ||
925                       (base_gfn >= s->base_gfn + s->npages)))
926                         goto out_free;
927         }
928
929         /* Free page dirty bitmap if unneeded */
930         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
931                 new.dirty_bitmap = NULL;
932
933         r = -ENOMEM;
934
935         /* Allocate if a slot is being created */
936 #ifndef CONFIG_S390
937         if (npages && !new.rmap) {
938                 new.rmap = vmalloc(npages * sizeof(struct page *));
939
940                 if (!new.rmap)
941                         goto out_free;
942
943                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
944
945                 new.user_alloc = user_alloc;
946                 /*
947                  * hva_to_rmmap() serialzies with the mmu_lock and to be
948                  * safe it has to ignore memslots with !user_alloc &&
949                  * !userspace_addr.
950                  */
951                 if (user_alloc)
952                         new.userspace_addr = mem->userspace_addr;
953                 else
954                         new.userspace_addr = 0;
955         }
956         if (npages && !new.lpage_info) {
957                 int largepages = npages / KVM_PAGES_PER_HPAGE;
958                 if (npages % KVM_PAGES_PER_HPAGE)
959                         largepages++;
960                 if (base_gfn % KVM_PAGES_PER_HPAGE)
961                         largepages++;
962
963                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
964
965                 if (!new.lpage_info)
966                         goto out_free;
967
968                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
969
970                 if (base_gfn % KVM_PAGES_PER_HPAGE)
971                         new.lpage_info[0].write_count = 1;
972                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
973                         new.lpage_info[largepages-1].write_count = 1;
974         }
975
976         /* Allocate page dirty bitmap if needed */
977         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
978                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
979
980                 new.dirty_bitmap = vmalloc(dirty_bytes);
981                 if (!new.dirty_bitmap)
982                         goto out_free;
983                 memset(new.dirty_bitmap, 0, dirty_bytes);
984         }
985 #endif /* not defined CONFIG_S390 */
986
987         if (!npages)
988                 kvm_arch_flush_shadow(kvm);
989
990         spin_lock(&kvm->mmu_lock);
991         if (mem->slot >= kvm->nmemslots)
992                 kvm->nmemslots = mem->slot + 1;
993
994         *memslot = new;
995         spin_unlock(&kvm->mmu_lock);
996
997         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
998         if (r) {
999                 spin_lock(&kvm->mmu_lock);
1000                 *memslot = old;
1001                 spin_unlock(&kvm->mmu_lock);
1002                 goto out_free;
1003         }
1004
1005         kvm_free_physmem_slot(&old, &new);
1006 #ifdef CONFIG_DMAR
1007         /* map the pages in iommu page table */
1008         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1009         if (r)
1010                 goto out;
1011 #endif
1012         return 0;
1013
1014 out_free:
1015         kvm_free_physmem_slot(&new, &old);
1016 out:
1017         return r;
1018
1019 }
1020 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1021
1022 int kvm_set_memory_region(struct kvm *kvm,
1023                           struct kvm_userspace_memory_region *mem,
1024                           int user_alloc)
1025 {
1026         int r;
1027
1028         down_write(&kvm->slots_lock);
1029         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1030         up_write(&kvm->slots_lock);
1031         return r;
1032 }
1033 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1034
1035 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1036                                    struct
1037                                    kvm_userspace_memory_region *mem,
1038                                    int user_alloc)
1039 {
1040         if (mem->slot >= KVM_MEMORY_SLOTS)
1041                 return -EINVAL;
1042         return kvm_set_memory_region(kvm, mem, user_alloc);
1043 }
1044
1045 int kvm_get_dirty_log(struct kvm *kvm,
1046                         struct kvm_dirty_log *log, int *is_dirty)
1047 {
1048         struct kvm_memory_slot *memslot;
1049         int r, i;
1050         int n;
1051         unsigned long any = 0;
1052
1053         r = -EINVAL;
1054         if (log->slot >= KVM_MEMORY_SLOTS)
1055                 goto out;
1056
1057         memslot = &kvm->memslots[log->slot];
1058         r = -ENOENT;
1059         if (!memslot->dirty_bitmap)
1060                 goto out;
1061
1062         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1063
1064         for (i = 0; !any && i < n/sizeof(long); ++i)
1065                 any = memslot->dirty_bitmap[i];
1066
1067         r = -EFAULT;
1068         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1069                 goto out;
1070
1071         if (any)
1072                 *is_dirty = 1;
1073
1074         r = 0;
1075 out:
1076         return r;
1077 }
1078
1079 int is_error_page(struct page *page)
1080 {
1081         return page == bad_page;
1082 }
1083 EXPORT_SYMBOL_GPL(is_error_page);
1084
1085 int is_error_pfn(pfn_t pfn)
1086 {
1087         return pfn == bad_pfn;
1088 }
1089 EXPORT_SYMBOL_GPL(is_error_pfn);
1090
1091 static inline unsigned long bad_hva(void)
1092 {
1093         return PAGE_OFFSET;
1094 }
1095
1096 int kvm_is_error_hva(unsigned long addr)
1097 {
1098         return addr == bad_hva();
1099 }
1100 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1101
1102 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1103 {
1104         int i;
1105
1106         for (i = 0; i < kvm->nmemslots; ++i) {
1107                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1108
1109                 if (gfn >= memslot->base_gfn
1110                     && gfn < memslot->base_gfn + memslot->npages)
1111                         return memslot;
1112         }
1113         return NULL;
1114 }
1115 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1116
1117 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1118 {
1119         gfn = unalias_gfn(kvm, gfn);
1120         return gfn_to_memslot_unaliased(kvm, gfn);
1121 }
1122
1123 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1124 {
1125         int i;
1126
1127         gfn = unalias_gfn(kvm, gfn);
1128         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1129                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1130
1131                 if (gfn >= memslot->base_gfn
1132                     && gfn < memslot->base_gfn + memslot->npages)
1133                         return 1;
1134         }
1135         return 0;
1136 }
1137 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1138
1139 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1140 {
1141         struct kvm_memory_slot *slot;
1142
1143         gfn = unalias_gfn(kvm, gfn);
1144         slot = gfn_to_memslot_unaliased(kvm, gfn);
1145         if (!slot)
1146                 return bad_hva();
1147         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1148 }
1149 EXPORT_SYMBOL_GPL(gfn_to_hva);
1150
1151 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1152 {
1153         struct page *page[1];
1154         unsigned long addr;
1155         int npages;
1156         pfn_t pfn;
1157
1158         might_sleep();
1159
1160         addr = gfn_to_hva(kvm, gfn);
1161         if (kvm_is_error_hva(addr)) {
1162                 get_page(bad_page);
1163                 return page_to_pfn(bad_page);
1164         }
1165
1166         npages = get_user_pages_fast(addr, 1, 1, page);
1167
1168         if (unlikely(npages != 1)) {
1169                 struct vm_area_struct *vma;
1170
1171                 down_read(&current->mm->mmap_sem);
1172                 vma = find_vma(current->mm, addr);
1173
1174                 if (vma == NULL || addr < vma->vm_start ||
1175                     !(vma->vm_flags & VM_PFNMAP)) {
1176                         up_read(&current->mm->mmap_sem);
1177                         get_page(bad_page);
1178                         return page_to_pfn(bad_page);
1179                 }
1180
1181                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1182                 up_read(&current->mm->mmap_sem);
1183                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1184         } else
1185                 pfn = page_to_pfn(page[0]);
1186
1187         return pfn;
1188 }
1189
1190 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1191
1192 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1193 {
1194         pfn_t pfn;
1195
1196         pfn = gfn_to_pfn(kvm, gfn);
1197         if (!kvm_is_mmio_pfn(pfn))
1198                 return pfn_to_page(pfn);
1199
1200         WARN_ON(kvm_is_mmio_pfn(pfn));
1201
1202         get_page(bad_page);
1203         return bad_page;
1204 }
1205
1206 EXPORT_SYMBOL_GPL(gfn_to_page);
1207
1208 void kvm_release_page_clean(struct page *page)
1209 {
1210         kvm_release_pfn_clean(page_to_pfn(page));
1211 }
1212 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1213
1214 void kvm_release_pfn_clean(pfn_t pfn)
1215 {
1216         if (!kvm_is_mmio_pfn(pfn))
1217                 put_page(pfn_to_page(pfn));
1218 }
1219 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1220
1221 void kvm_release_page_dirty(struct page *page)
1222 {
1223         kvm_release_pfn_dirty(page_to_pfn(page));
1224 }
1225 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1226
1227 void kvm_release_pfn_dirty(pfn_t pfn)
1228 {
1229         kvm_set_pfn_dirty(pfn);
1230         kvm_release_pfn_clean(pfn);
1231 }
1232 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1233
1234 void kvm_set_page_dirty(struct page *page)
1235 {
1236         kvm_set_pfn_dirty(page_to_pfn(page));
1237 }
1238 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1239
1240 void kvm_set_pfn_dirty(pfn_t pfn)
1241 {
1242         if (!kvm_is_mmio_pfn(pfn)) {
1243                 struct page *page = pfn_to_page(pfn);
1244                 if (!PageReserved(page))
1245                         SetPageDirty(page);
1246         }
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1249
1250 void kvm_set_pfn_accessed(pfn_t pfn)
1251 {
1252         if (!kvm_is_mmio_pfn(pfn))
1253                 mark_page_accessed(pfn_to_page(pfn));
1254 }
1255 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1256
1257 void kvm_get_pfn(pfn_t pfn)
1258 {
1259         if (!kvm_is_mmio_pfn(pfn))
1260                 get_page(pfn_to_page(pfn));
1261 }
1262 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1263
1264 static int next_segment(unsigned long len, int offset)
1265 {
1266         if (len > PAGE_SIZE - offset)
1267                 return PAGE_SIZE - offset;
1268         else
1269                 return len;
1270 }
1271
1272 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1273                         int len)
1274 {
1275         int r;
1276         unsigned long addr;
1277
1278         addr = gfn_to_hva(kvm, gfn);
1279         if (kvm_is_error_hva(addr))
1280                 return -EFAULT;
1281         r = copy_from_user(data, (void __user *)addr + offset, len);
1282         if (r)
1283                 return -EFAULT;
1284         return 0;
1285 }
1286 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1287
1288 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1289 {
1290         gfn_t gfn = gpa >> PAGE_SHIFT;
1291         int seg;
1292         int offset = offset_in_page(gpa);
1293         int ret;
1294
1295         while ((seg = next_segment(len, offset)) != 0) {
1296                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1297                 if (ret < 0)
1298                         return ret;
1299                 offset = 0;
1300                 len -= seg;
1301                 data += seg;
1302                 ++gfn;
1303         }
1304         return 0;
1305 }
1306 EXPORT_SYMBOL_GPL(kvm_read_guest);
1307
1308 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1309                           unsigned long len)
1310 {
1311         int r;
1312         unsigned long addr;
1313         gfn_t gfn = gpa >> PAGE_SHIFT;
1314         int offset = offset_in_page(gpa);
1315
1316         addr = gfn_to_hva(kvm, gfn);
1317         if (kvm_is_error_hva(addr))
1318                 return -EFAULT;
1319         pagefault_disable();
1320         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1321         pagefault_enable();
1322         if (r)
1323                 return -EFAULT;
1324         return 0;
1325 }
1326 EXPORT_SYMBOL(kvm_read_guest_atomic);
1327
1328 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1329                          int offset, int len)
1330 {
1331         int r;
1332         unsigned long addr;
1333
1334         addr = gfn_to_hva(kvm, gfn);
1335         if (kvm_is_error_hva(addr))
1336                 return -EFAULT;
1337         r = copy_to_user((void __user *)addr + offset, data, len);
1338         if (r)
1339                 return -EFAULT;
1340         mark_page_dirty(kvm, gfn);
1341         return 0;
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1344
1345 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1346                     unsigned long len)
1347 {
1348         gfn_t gfn = gpa >> PAGE_SHIFT;
1349         int seg;
1350         int offset = offset_in_page(gpa);
1351         int ret;
1352
1353         while ((seg = next_segment(len, offset)) != 0) {
1354                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1355                 if (ret < 0)
1356                         return ret;
1357                 offset = 0;
1358                 len -= seg;
1359                 data += seg;
1360                 ++gfn;
1361         }
1362         return 0;
1363 }
1364
1365 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1366 {
1367         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1368 }
1369 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1370
1371 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1372 {
1373         gfn_t gfn = gpa >> PAGE_SHIFT;
1374         int seg;
1375         int offset = offset_in_page(gpa);
1376         int ret;
1377
1378         while ((seg = next_segment(len, offset)) != 0) {
1379                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1380                 if (ret < 0)
1381                         return ret;
1382                 offset = 0;
1383                 len -= seg;
1384                 ++gfn;
1385         }
1386         return 0;
1387 }
1388 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1389
1390 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1391 {
1392         struct kvm_memory_slot *memslot;
1393
1394         gfn = unalias_gfn(kvm, gfn);
1395         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1396         if (memslot && memslot->dirty_bitmap) {
1397                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1398
1399                 /* avoid RMW */
1400                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1401                         set_bit(rel_gfn, memslot->dirty_bitmap);
1402         }
1403 }
1404
1405 /*
1406  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1407  */
1408 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1409 {
1410         DEFINE_WAIT(wait);
1411
1412         for (;;) {
1413                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1414
1415                 if (kvm_cpu_has_interrupt(vcpu) ||
1416                     kvm_cpu_has_pending_timer(vcpu) ||
1417                     kvm_arch_vcpu_runnable(vcpu)) {
1418                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1419                         break;
1420                 }
1421                 if (signal_pending(current))
1422                         break;
1423
1424                 vcpu_put(vcpu);
1425                 schedule();
1426                 vcpu_load(vcpu);
1427         }
1428
1429         finish_wait(&vcpu->wq, &wait);
1430 }
1431
1432 void kvm_resched(struct kvm_vcpu *vcpu)
1433 {
1434         if (!need_resched())
1435                 return;
1436         cond_resched();
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_resched);
1439
1440 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1441 {
1442         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1443         struct page *page;
1444
1445         if (vmf->pgoff == 0)
1446                 page = virt_to_page(vcpu->run);
1447 #ifdef CONFIG_X86
1448         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1449                 page = virt_to_page(vcpu->arch.pio_data);
1450 #endif
1451 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1452         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1453                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1454 #endif
1455         else
1456                 return VM_FAULT_SIGBUS;
1457         get_page(page);
1458         vmf->page = page;
1459         return 0;
1460 }
1461
1462 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1463         .fault = kvm_vcpu_fault,
1464 };
1465
1466 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1467 {
1468         vma->vm_ops = &kvm_vcpu_vm_ops;
1469         return 0;
1470 }
1471
1472 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1473 {
1474         struct kvm_vcpu *vcpu = filp->private_data;
1475
1476         kvm_put_kvm(vcpu->kvm);
1477         return 0;
1478 }
1479
1480 static const struct file_operations kvm_vcpu_fops = {
1481         .release        = kvm_vcpu_release,
1482         .unlocked_ioctl = kvm_vcpu_ioctl,
1483         .compat_ioctl   = kvm_vcpu_ioctl,
1484         .mmap           = kvm_vcpu_mmap,
1485 };
1486
1487 /*
1488  * Allocates an inode for the vcpu.
1489  */
1490 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1491 {
1492         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1493         if (fd < 0)
1494                 kvm_put_kvm(vcpu->kvm);
1495         return fd;
1496 }
1497
1498 /*
1499  * Creates some virtual cpus.  Good luck creating more than one.
1500  */
1501 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1502 {
1503         int r;
1504         struct kvm_vcpu *vcpu;
1505
1506         if (!valid_vcpu(n))
1507                 return -EINVAL;
1508
1509         vcpu = kvm_arch_vcpu_create(kvm, n);
1510         if (IS_ERR(vcpu))
1511                 return PTR_ERR(vcpu);
1512
1513         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1514
1515         r = kvm_arch_vcpu_setup(vcpu);
1516         if (r)
1517                 return r;
1518
1519         mutex_lock(&kvm->lock);
1520         if (kvm->vcpus[n]) {
1521                 r = -EEXIST;
1522                 goto vcpu_destroy;
1523         }
1524         kvm->vcpus[n] = vcpu;
1525         mutex_unlock(&kvm->lock);
1526
1527         /* Now it's all set up, let userspace reach it */
1528         kvm_get_kvm(kvm);
1529         r = create_vcpu_fd(vcpu);
1530         if (r < 0)
1531                 goto unlink;
1532         return r;
1533
1534 unlink:
1535         mutex_lock(&kvm->lock);
1536         kvm->vcpus[n] = NULL;
1537 vcpu_destroy:
1538         mutex_unlock(&kvm->lock);
1539         kvm_arch_vcpu_destroy(vcpu);
1540         return r;
1541 }
1542
1543 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1544 {
1545         if (sigset) {
1546                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1547                 vcpu->sigset_active = 1;
1548                 vcpu->sigset = *sigset;
1549         } else
1550                 vcpu->sigset_active = 0;
1551         return 0;
1552 }
1553
1554 static long kvm_vcpu_ioctl(struct file *filp,
1555                            unsigned int ioctl, unsigned long arg)
1556 {
1557         struct kvm_vcpu *vcpu = filp->private_data;
1558         void __user *argp = (void __user *)arg;
1559         int r;
1560         struct kvm_fpu *fpu = NULL;
1561         struct kvm_sregs *kvm_sregs = NULL;
1562
1563         if (vcpu->kvm->mm != current->mm)
1564                 return -EIO;
1565         switch (ioctl) {
1566         case KVM_RUN:
1567                 r = -EINVAL;
1568                 if (arg)
1569                         goto out;
1570                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1571                 break;
1572         case KVM_GET_REGS: {
1573                 struct kvm_regs *kvm_regs;
1574
1575                 r = -ENOMEM;
1576                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1577                 if (!kvm_regs)
1578                         goto out;
1579                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1580                 if (r)
1581                         goto out_free1;
1582                 r = -EFAULT;
1583                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1584                         goto out_free1;
1585                 r = 0;
1586 out_free1:
1587                 kfree(kvm_regs);
1588                 break;
1589         }
1590         case KVM_SET_REGS: {
1591                 struct kvm_regs *kvm_regs;
1592
1593                 r = -ENOMEM;
1594                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1595                 if (!kvm_regs)
1596                         goto out;
1597                 r = -EFAULT;
1598                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1599                         goto out_free2;
1600                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1601                 if (r)
1602                         goto out_free2;
1603                 r = 0;
1604 out_free2:
1605                 kfree(kvm_regs);
1606                 break;
1607         }
1608         case KVM_GET_SREGS: {
1609                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1610                 r = -ENOMEM;
1611                 if (!kvm_sregs)
1612                         goto out;
1613                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1614                 if (r)
1615                         goto out;
1616                 r = -EFAULT;
1617                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1618                         goto out;
1619                 r = 0;
1620                 break;
1621         }
1622         case KVM_SET_SREGS: {
1623                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1624                 r = -ENOMEM;
1625                 if (!kvm_sregs)
1626                         goto out;
1627                 r = -EFAULT;
1628                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1629                         goto out;
1630                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1631                 if (r)
1632                         goto out;
1633                 r = 0;
1634                 break;
1635         }
1636         case KVM_GET_MP_STATE: {
1637                 struct kvm_mp_state mp_state;
1638
1639                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1640                 if (r)
1641                         goto out;
1642                 r = -EFAULT;
1643                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1644                         goto out;
1645                 r = 0;
1646                 break;
1647         }
1648         case KVM_SET_MP_STATE: {
1649                 struct kvm_mp_state mp_state;
1650
1651                 r = -EFAULT;
1652                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1653                         goto out;
1654                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1655                 if (r)
1656                         goto out;
1657                 r = 0;
1658                 break;
1659         }
1660         case KVM_TRANSLATE: {
1661                 struct kvm_translation tr;
1662
1663                 r = -EFAULT;
1664                 if (copy_from_user(&tr, argp, sizeof tr))
1665                         goto out;
1666                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1667                 if (r)
1668                         goto out;
1669                 r = -EFAULT;
1670                 if (copy_to_user(argp, &tr, sizeof tr))
1671                         goto out;
1672                 r = 0;
1673                 break;
1674         }
1675         case KVM_DEBUG_GUEST: {
1676                 struct kvm_debug_guest dbg;
1677
1678                 r = -EFAULT;
1679                 if (copy_from_user(&dbg, argp, sizeof dbg))
1680                         goto out;
1681                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1682                 if (r)
1683                         goto out;
1684                 r = 0;
1685                 break;
1686         }
1687         case KVM_SET_SIGNAL_MASK: {
1688                 struct kvm_signal_mask __user *sigmask_arg = argp;
1689                 struct kvm_signal_mask kvm_sigmask;
1690                 sigset_t sigset, *p;
1691
1692                 p = NULL;
1693                 if (argp) {
1694                         r = -EFAULT;
1695                         if (copy_from_user(&kvm_sigmask, argp,
1696                                            sizeof kvm_sigmask))
1697                                 goto out;
1698                         r = -EINVAL;
1699                         if (kvm_sigmask.len != sizeof sigset)
1700                                 goto out;
1701                         r = -EFAULT;
1702                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1703                                            sizeof sigset))
1704                                 goto out;
1705                         p = &sigset;
1706                 }
1707                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1708                 break;
1709         }
1710         case KVM_GET_FPU: {
1711                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1712                 r = -ENOMEM;
1713                 if (!fpu)
1714                         goto out;
1715                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1716                 if (r)
1717                         goto out;
1718                 r = -EFAULT;
1719                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1720                         goto out;
1721                 r = 0;
1722                 break;
1723         }
1724         case KVM_SET_FPU: {
1725                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1726                 r = -ENOMEM;
1727                 if (!fpu)
1728                         goto out;
1729                 r = -EFAULT;
1730                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1731                         goto out;
1732                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1733                 if (r)
1734                         goto out;
1735                 r = 0;
1736                 break;
1737         }
1738         default:
1739                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1740         }
1741 out:
1742         kfree(fpu);
1743         kfree(kvm_sregs);
1744         return r;
1745 }
1746
1747 static long kvm_vm_ioctl(struct file *filp,
1748                            unsigned int ioctl, unsigned long arg)
1749 {
1750         struct kvm *kvm = filp->private_data;
1751         void __user *argp = (void __user *)arg;
1752         int r;
1753
1754         if (kvm->mm != current->mm)
1755                 return -EIO;
1756         switch (ioctl) {
1757         case KVM_CREATE_VCPU:
1758                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1759                 if (r < 0)
1760                         goto out;
1761                 break;
1762         case KVM_SET_USER_MEMORY_REGION: {
1763                 struct kvm_userspace_memory_region kvm_userspace_mem;
1764
1765                 r = -EFAULT;
1766                 if (copy_from_user(&kvm_userspace_mem, argp,
1767                                                 sizeof kvm_userspace_mem))
1768                         goto out;
1769
1770                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1771                 if (r)
1772                         goto out;
1773                 break;
1774         }
1775         case KVM_GET_DIRTY_LOG: {
1776                 struct kvm_dirty_log log;
1777
1778                 r = -EFAULT;
1779                 if (copy_from_user(&log, argp, sizeof log))
1780                         goto out;
1781                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1782                 if (r)
1783                         goto out;
1784                 break;
1785         }
1786 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1787         case KVM_REGISTER_COALESCED_MMIO: {
1788                 struct kvm_coalesced_mmio_zone zone;
1789                 r = -EFAULT;
1790                 if (copy_from_user(&zone, argp, sizeof zone))
1791                         goto out;
1792                 r = -ENXIO;
1793                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1794                 if (r)
1795                         goto out;
1796                 r = 0;
1797                 break;
1798         }
1799         case KVM_UNREGISTER_COALESCED_MMIO: {
1800                 struct kvm_coalesced_mmio_zone zone;
1801                 r = -EFAULT;
1802                 if (copy_from_user(&zone, argp, sizeof zone))
1803                         goto out;
1804                 r = -ENXIO;
1805                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1806                 if (r)
1807                         goto out;
1808                 r = 0;
1809                 break;
1810         }
1811 #endif
1812 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1813         case KVM_ASSIGN_PCI_DEVICE: {
1814                 struct kvm_assigned_pci_dev assigned_dev;
1815
1816                 r = -EFAULT;
1817                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1818                         goto out;
1819                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1820                 if (r)
1821                         goto out;
1822                 break;
1823         }
1824         case KVM_ASSIGN_IRQ: {
1825                 struct kvm_assigned_irq assigned_irq;
1826
1827                 r = -EFAULT;
1828                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1829                         goto out;
1830                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1831                 if (r)
1832                         goto out;
1833                 break;
1834         }
1835 #endif
1836         default:
1837                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1838         }
1839 out:
1840         return r;
1841 }
1842
1843 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1844 {
1845         struct page *page[1];
1846         unsigned long addr;
1847         int npages;
1848         gfn_t gfn = vmf->pgoff;
1849         struct kvm *kvm = vma->vm_file->private_data;
1850
1851         addr = gfn_to_hva(kvm, gfn);
1852         if (kvm_is_error_hva(addr))
1853                 return VM_FAULT_SIGBUS;
1854
1855         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1856                                 NULL);
1857         if (unlikely(npages != 1))
1858                 return VM_FAULT_SIGBUS;
1859
1860         vmf->page = page[0];
1861         return 0;
1862 }
1863
1864 static struct vm_operations_struct kvm_vm_vm_ops = {
1865         .fault = kvm_vm_fault,
1866 };
1867
1868 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1869 {
1870         vma->vm_ops = &kvm_vm_vm_ops;
1871         return 0;
1872 }
1873
1874 static const struct file_operations kvm_vm_fops = {
1875         .release        = kvm_vm_release,
1876         .unlocked_ioctl = kvm_vm_ioctl,
1877         .compat_ioctl   = kvm_vm_ioctl,
1878         .mmap           = kvm_vm_mmap,
1879 };
1880
1881 static int kvm_dev_ioctl_create_vm(void)
1882 {
1883         int fd;
1884         struct kvm *kvm;
1885
1886         kvm = kvm_create_vm();
1887         if (IS_ERR(kvm))
1888                 return PTR_ERR(kvm);
1889         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1890         if (fd < 0)
1891                 kvm_put_kvm(kvm);
1892
1893         return fd;
1894 }
1895
1896 static long kvm_dev_ioctl(struct file *filp,
1897                           unsigned int ioctl, unsigned long arg)
1898 {
1899         long r = -EINVAL;
1900
1901         switch (ioctl) {
1902         case KVM_GET_API_VERSION:
1903                 r = -EINVAL;
1904                 if (arg)
1905                         goto out;
1906                 r = KVM_API_VERSION;
1907                 break;
1908         case KVM_CREATE_VM:
1909                 r = -EINVAL;
1910                 if (arg)
1911                         goto out;
1912                 r = kvm_dev_ioctl_create_vm();
1913                 break;
1914         case KVM_CHECK_EXTENSION:
1915                 r = kvm_dev_ioctl_check_extension(arg);
1916                 break;
1917         case KVM_GET_VCPU_MMAP_SIZE:
1918                 r = -EINVAL;
1919                 if (arg)
1920                         goto out;
1921                 r = PAGE_SIZE;     /* struct kvm_run */
1922 #ifdef CONFIG_X86
1923                 r += PAGE_SIZE;    /* pio data page */
1924 #endif
1925 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1926                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1927 #endif
1928                 break;
1929         case KVM_TRACE_ENABLE:
1930         case KVM_TRACE_PAUSE:
1931         case KVM_TRACE_DISABLE:
1932                 r = kvm_trace_ioctl(ioctl, arg);
1933                 break;
1934         default:
1935                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1936         }
1937 out:
1938         return r;
1939 }
1940
1941 static struct file_operations kvm_chardev_ops = {
1942         .unlocked_ioctl = kvm_dev_ioctl,
1943         .compat_ioctl   = kvm_dev_ioctl,
1944 };
1945
1946 static struct miscdevice kvm_dev = {
1947         KVM_MINOR,
1948         "kvm",
1949         &kvm_chardev_ops,
1950 };
1951
1952 static void hardware_enable(void *junk)
1953 {
1954         int cpu = raw_smp_processor_id();
1955
1956         if (cpu_isset(cpu, cpus_hardware_enabled))
1957                 return;
1958         cpu_set(cpu, cpus_hardware_enabled);
1959         kvm_arch_hardware_enable(NULL);
1960 }
1961
1962 static void hardware_disable(void *junk)
1963 {
1964         int cpu = raw_smp_processor_id();
1965
1966         if (!cpu_isset(cpu, cpus_hardware_enabled))
1967                 return;
1968         cpu_clear(cpu, cpus_hardware_enabled);
1969         kvm_arch_hardware_disable(NULL);
1970 }
1971
1972 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1973                            void *v)
1974 {
1975         int cpu = (long)v;
1976
1977         val &= ~CPU_TASKS_FROZEN;
1978         switch (val) {
1979         case CPU_DYING:
1980                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1981                        cpu);
1982                 hardware_disable(NULL);
1983                 break;
1984         case CPU_UP_CANCELED:
1985                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1986                        cpu);
1987                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1988                 break;
1989         case CPU_ONLINE:
1990                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1991                        cpu);
1992                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1993                 break;
1994         }
1995         return NOTIFY_OK;
1996 }
1997
1998
1999 asmlinkage void kvm_handle_fault_on_reboot(void)
2000 {
2001         if (kvm_rebooting)
2002                 /* spin while reset goes on */
2003                 while (true)
2004                         ;
2005         /* Fault while not rebooting.  We want the trace. */
2006         BUG();
2007 }
2008 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2009
2010 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2011                       void *v)
2012 {
2013         if (val == SYS_RESTART) {
2014                 /*
2015                  * Some (well, at least mine) BIOSes hang on reboot if
2016                  * in vmx root mode.
2017                  */
2018                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2019                 kvm_rebooting = true;
2020                 on_each_cpu(hardware_disable, NULL, 1);
2021         }
2022         return NOTIFY_OK;
2023 }
2024
2025 static struct notifier_block kvm_reboot_notifier = {
2026         .notifier_call = kvm_reboot,
2027         .priority = 0,
2028 };
2029
2030 void kvm_io_bus_init(struct kvm_io_bus *bus)
2031 {
2032         memset(bus, 0, sizeof(*bus));
2033 }
2034
2035 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2036 {
2037         int i;
2038
2039         for (i = 0; i < bus->dev_count; i++) {
2040                 struct kvm_io_device *pos = bus->devs[i];
2041
2042                 kvm_iodevice_destructor(pos);
2043         }
2044 }
2045
2046 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2047                                           gpa_t addr, int len, int is_write)
2048 {
2049         int i;
2050
2051         for (i = 0; i < bus->dev_count; i++) {
2052                 struct kvm_io_device *pos = bus->devs[i];
2053
2054                 if (pos->in_range(pos, addr, len, is_write))
2055                         return pos;
2056         }
2057
2058         return NULL;
2059 }
2060
2061 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2062 {
2063         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2064
2065         bus->devs[bus->dev_count++] = dev;
2066 }
2067
2068 static struct notifier_block kvm_cpu_notifier = {
2069         .notifier_call = kvm_cpu_hotplug,
2070         .priority = 20, /* must be > scheduler priority */
2071 };
2072
2073 static int vm_stat_get(void *_offset, u64 *val)
2074 {
2075         unsigned offset = (long)_offset;
2076         struct kvm *kvm;
2077
2078         *val = 0;
2079         spin_lock(&kvm_lock);
2080         list_for_each_entry(kvm, &vm_list, vm_list)
2081                 *val += *(u32 *)((void *)kvm + offset);
2082         spin_unlock(&kvm_lock);
2083         return 0;
2084 }
2085
2086 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2087
2088 static int vcpu_stat_get(void *_offset, u64 *val)
2089 {
2090         unsigned offset = (long)_offset;
2091         struct kvm *kvm;
2092         struct kvm_vcpu *vcpu;
2093         int i;
2094
2095         *val = 0;
2096         spin_lock(&kvm_lock);
2097         list_for_each_entry(kvm, &vm_list, vm_list)
2098                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2099                         vcpu = kvm->vcpus[i];
2100                         if (vcpu)
2101                                 *val += *(u32 *)((void *)vcpu + offset);
2102                 }
2103         spin_unlock(&kvm_lock);
2104         return 0;
2105 }
2106
2107 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2108
2109 static struct file_operations *stat_fops[] = {
2110         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2111         [KVM_STAT_VM]   = &vm_stat_fops,
2112 };
2113
2114 static void kvm_init_debug(void)
2115 {
2116         struct kvm_stats_debugfs_item *p;
2117
2118         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2119         for (p = debugfs_entries; p->name; ++p)
2120                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2121                                                 (void *)(long)p->offset,
2122                                                 stat_fops[p->kind]);
2123 }
2124
2125 static void kvm_exit_debug(void)
2126 {
2127         struct kvm_stats_debugfs_item *p;
2128
2129         for (p = debugfs_entries; p->name; ++p)
2130                 debugfs_remove(p->dentry);
2131         debugfs_remove(kvm_debugfs_dir);
2132 }
2133
2134 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2135 {
2136         hardware_disable(NULL);
2137         return 0;
2138 }
2139
2140 static int kvm_resume(struct sys_device *dev)
2141 {
2142         hardware_enable(NULL);
2143         return 0;
2144 }
2145
2146 static struct sysdev_class kvm_sysdev_class = {
2147         .name = "kvm",
2148         .suspend = kvm_suspend,
2149         .resume = kvm_resume,
2150 };
2151
2152 static struct sys_device kvm_sysdev = {
2153         .id = 0,
2154         .cls = &kvm_sysdev_class,
2155 };
2156
2157 struct page *bad_page;
2158 pfn_t bad_pfn;
2159
2160 static inline
2161 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2162 {
2163         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2164 }
2165
2166 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2167 {
2168         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2169
2170         kvm_arch_vcpu_load(vcpu, cpu);
2171 }
2172
2173 static void kvm_sched_out(struct preempt_notifier *pn,
2174                           struct task_struct *next)
2175 {
2176         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2177
2178         kvm_arch_vcpu_put(vcpu);
2179 }
2180
2181 int kvm_init(void *opaque, unsigned int vcpu_size,
2182                   struct module *module)
2183 {
2184         int r;
2185         int cpu;
2186
2187         kvm_init_debug();
2188
2189         r = kvm_arch_init(opaque);
2190         if (r)
2191                 goto out_fail;
2192
2193         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2194
2195         if (bad_page == NULL) {
2196                 r = -ENOMEM;
2197                 goto out;
2198         }
2199
2200         bad_pfn = page_to_pfn(bad_page);
2201
2202         r = kvm_arch_hardware_setup();
2203         if (r < 0)
2204                 goto out_free_0;
2205
2206         for_each_online_cpu(cpu) {
2207                 smp_call_function_single(cpu,
2208                                 kvm_arch_check_processor_compat,
2209                                 &r, 1);
2210                 if (r < 0)
2211                         goto out_free_1;
2212         }
2213
2214         on_each_cpu(hardware_enable, NULL, 1);
2215         r = register_cpu_notifier(&kvm_cpu_notifier);
2216         if (r)
2217                 goto out_free_2;
2218         register_reboot_notifier(&kvm_reboot_notifier);
2219
2220         r = sysdev_class_register(&kvm_sysdev_class);
2221         if (r)
2222                 goto out_free_3;
2223
2224         r = sysdev_register(&kvm_sysdev);
2225         if (r)
2226                 goto out_free_4;
2227
2228         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2229         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2230                                            __alignof__(struct kvm_vcpu),
2231                                            0, NULL);
2232         if (!kvm_vcpu_cache) {
2233                 r = -ENOMEM;
2234                 goto out_free_5;
2235         }
2236
2237         kvm_chardev_ops.owner = module;
2238
2239         r = misc_register(&kvm_dev);
2240         if (r) {
2241                 printk(KERN_ERR "kvm: misc device register failed\n");
2242                 goto out_free;
2243         }
2244
2245         kvm_preempt_ops.sched_in = kvm_sched_in;
2246         kvm_preempt_ops.sched_out = kvm_sched_out;
2247 #ifndef CONFIG_X86
2248         msi2intx = 0;
2249 #endif
2250
2251         return 0;
2252
2253 out_free:
2254         kmem_cache_destroy(kvm_vcpu_cache);
2255 out_free_5:
2256         sysdev_unregister(&kvm_sysdev);
2257 out_free_4:
2258         sysdev_class_unregister(&kvm_sysdev_class);
2259 out_free_3:
2260         unregister_reboot_notifier(&kvm_reboot_notifier);
2261         unregister_cpu_notifier(&kvm_cpu_notifier);
2262 out_free_2:
2263         on_each_cpu(hardware_disable, NULL, 1);
2264 out_free_1:
2265         kvm_arch_hardware_unsetup();
2266 out_free_0:
2267         __free_page(bad_page);
2268 out:
2269         kvm_arch_exit();
2270         kvm_exit_debug();
2271 out_fail:
2272         return r;
2273 }
2274 EXPORT_SYMBOL_GPL(kvm_init);
2275
2276 void kvm_exit(void)
2277 {
2278         kvm_trace_cleanup();
2279         misc_deregister(&kvm_dev);
2280         kmem_cache_destroy(kvm_vcpu_cache);
2281         sysdev_unregister(&kvm_sysdev);
2282         sysdev_class_unregister(&kvm_sysdev_class);
2283         unregister_reboot_notifier(&kvm_reboot_notifier);
2284         unregister_cpu_notifier(&kvm_cpu_notifier);
2285         on_each_cpu(hardware_disable, NULL, 1);
2286         kvm_arch_hardware_unsetup();
2287         kvm_arch_exit();
2288         kvm_exit_debug();
2289         __free_page(bad_page);
2290 }
2291 EXPORT_SYMBOL_GPL(kvm_exit);