KVM: Fix overlapping check for 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 #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 KVM_COALESCED_MMIO_PAGE_OFFSET
51 #include "coalesced_mmio.h"
52 #endif
53
54 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
55 #include <linux/pci.h>
56 #include <linux/interrupt.h>
57 #include "irq.h"
58 #endif
59
60 MODULE_AUTHOR("Qumranet");
61 MODULE_LICENSE("GPL");
62
63 static int msi2intx = 1;
64 module_param(msi2intx, bool, 0);
65
66 DEFINE_SPINLOCK(kvm_lock);
67 LIST_HEAD(vm_list);
68
69 static cpumask_var_t cpus_hardware_enabled;
70
71 struct kmem_cache *kvm_vcpu_cache;
72 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
73
74 static __read_mostly struct preempt_ops kvm_preempt_ops;
75
76 struct dentry *kvm_debugfs_dir;
77
78 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
79                            unsigned long arg);
80
81 static bool kvm_rebooting;
82
83 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
84 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
85                                                       int assigned_dev_id)
86 {
87         struct list_head *ptr;
88         struct kvm_assigned_dev_kernel *match;
89
90         list_for_each(ptr, head) {
91                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
92                 if (match->assigned_dev_id == assigned_dev_id)
93                         return match;
94         }
95         return NULL;
96 }
97
98 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
99 {
100         struct kvm_assigned_dev_kernel *assigned_dev;
101
102         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
103                                     interrupt_work);
104
105         /* This is taken to safely inject irq inside the guest. When
106          * the interrupt injection (or the ioapic code) uses a
107          * finer-grained lock, update this
108          */
109         mutex_lock(&assigned_dev->kvm->lock);
110         kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
111                     assigned_dev->guest_irq, 1);
112
113         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_MSI) {
114                 enable_irq(assigned_dev->host_irq);
115                 assigned_dev->host_irq_disabled = false;
116         }
117         mutex_unlock(&assigned_dev->kvm->lock);
118 }
119
120 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
121 {
122         struct kvm_assigned_dev_kernel *assigned_dev =
123                 (struct kvm_assigned_dev_kernel *) dev_id;
124
125         schedule_work(&assigned_dev->interrupt_work);
126
127         disable_irq_nosync(irq);
128         assigned_dev->host_irq_disabled = true;
129
130         return IRQ_HANDLED;
131 }
132
133 /* Ack the irq line for an assigned device */
134 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
135 {
136         struct kvm_assigned_dev_kernel *dev;
137
138         if (kian->gsi == -1)
139                 return;
140
141         dev = container_of(kian, struct kvm_assigned_dev_kernel,
142                            ack_notifier);
143
144         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
145
146         /* The guest irq may be shared so this ack may be
147          * from another device.
148          */
149         if (dev->host_irq_disabled) {
150                 enable_irq(dev->host_irq);
151                 dev->host_irq_disabled = false;
152         }
153 }
154
155 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
156 static void kvm_free_assigned_irq(struct kvm *kvm,
157                                   struct kvm_assigned_dev_kernel *assigned_dev)
158 {
159         if (!irqchip_in_kernel(kvm))
160                 return;
161
162         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
163
164         if (assigned_dev->irq_source_id != -1)
165                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
166         assigned_dev->irq_source_id = -1;
167
168         if (!assigned_dev->irq_requested_type)
169                 return;
170
171         /*
172          * In kvm_free_device_irq, cancel_work_sync return true if:
173          * 1. work is scheduled, and then cancelled.
174          * 2. work callback is executed.
175          *
176          * The first one ensured that the irq is disabled and no more events
177          * would happen. But for the second one, the irq may be enabled (e.g.
178          * for MSI). So we disable irq here to prevent further events.
179          *
180          * Notice this maybe result in nested disable if the interrupt type is
181          * INTx, but it's OK for we are going to free it.
182          *
183          * If this function is a part of VM destroy, please ensure that till
184          * now, the kvm state is still legal for probably we also have to wait
185          * interrupt_work done.
186          */
187         disable_irq_nosync(assigned_dev->host_irq);
188         cancel_work_sync(&assigned_dev->interrupt_work);
189
190         free_irq(assigned_dev->host_irq, (void *)assigned_dev);
191
192         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
193                 pci_disable_msi(assigned_dev->dev);
194
195         assigned_dev->irq_requested_type = 0;
196 }
197
198
199 static void kvm_free_assigned_device(struct kvm *kvm,
200                                      struct kvm_assigned_dev_kernel
201                                      *assigned_dev)
202 {
203         kvm_free_assigned_irq(kvm, assigned_dev);
204
205         pci_reset_function(assigned_dev->dev);
206
207         pci_release_regions(assigned_dev->dev);
208         pci_disable_device(assigned_dev->dev);
209         pci_dev_put(assigned_dev->dev);
210
211         list_del(&assigned_dev->list);
212         kfree(assigned_dev);
213 }
214
215 void kvm_free_all_assigned_devices(struct kvm *kvm)
216 {
217         struct list_head *ptr, *ptr2;
218         struct kvm_assigned_dev_kernel *assigned_dev;
219
220         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
221                 assigned_dev = list_entry(ptr,
222                                           struct kvm_assigned_dev_kernel,
223                                           list);
224
225                 kvm_free_assigned_device(kvm, assigned_dev);
226         }
227 }
228
229 static int assigned_device_update_intx(struct kvm *kvm,
230                         struct kvm_assigned_dev_kernel *adev,
231                         struct kvm_assigned_irq *airq)
232 {
233         adev->guest_irq = airq->guest_irq;
234         adev->ack_notifier.gsi = airq->guest_irq;
235
236         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
237                 return 0;
238
239         if (irqchip_in_kernel(kvm)) {
240                 if (!msi2intx &&
241                     (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)) {
242                         free_irq(adev->host_irq, (void *)adev);
243                         pci_disable_msi(adev->dev);
244                 }
245
246                 if (!capable(CAP_SYS_RAWIO))
247                         return -EPERM;
248
249                 if (airq->host_irq)
250                         adev->host_irq = airq->host_irq;
251                 else
252                         adev->host_irq = adev->dev->irq;
253
254                 /* Even though this is PCI, we don't want to use shared
255                  * interrupts. Sharing host devices with guest-assigned devices
256                  * on the same interrupt line is not a happy situation: there
257                  * are going to be long delays in accepting, acking, etc.
258                  */
259                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
260                                 0, "kvm_assigned_intx_device", (void *)adev))
261                         return -EIO;
262         }
263
264         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
265                                    KVM_ASSIGNED_DEV_HOST_INTX;
266         return 0;
267 }
268
269 #ifdef CONFIG_X86
270 static int assigned_device_update_msi(struct kvm *kvm,
271                         struct kvm_assigned_dev_kernel *adev,
272                         struct kvm_assigned_irq *airq)
273 {
274         int r;
275
276         adev->guest_irq = airq->guest_irq;
277         if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
278                 /* x86 don't care upper address of guest msi message addr */
279                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
280                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
281                 adev->ack_notifier.gsi = -1;
282         } else if (msi2intx) {
283                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
284                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
285                 adev->ack_notifier.gsi = airq->guest_irq;
286         } else {
287                 /*
288                  * Guest require to disable device MSI, we disable MSI and
289                  * re-enable INTx by default again. Notice it's only for
290                  * non-msi2intx.
291                  */
292                 assigned_device_update_intx(kvm, adev, airq);
293                 return 0;
294         }
295
296         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
297                 return 0;
298
299         if (irqchip_in_kernel(kvm)) {
300                 if (!msi2intx) {
301                         if (adev->irq_requested_type &
302                                         KVM_ASSIGNED_DEV_HOST_INTX)
303                                 free_irq(adev->host_irq, (void *)adev);
304
305                         r = pci_enable_msi(adev->dev);
306                         if (r)
307                                 return r;
308                 }
309
310                 adev->host_irq = adev->dev->irq;
311                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
312                                 "kvm_assigned_msi_device", (void *)adev))
313                         return -EIO;
314         }
315
316         if (!msi2intx)
317                 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
318
319         adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
320         return 0;
321 }
322 #endif
323
324 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
325                                    struct kvm_assigned_irq
326                                    *assigned_irq)
327 {
328         int r = 0;
329         struct kvm_assigned_dev_kernel *match;
330         u32 current_flags = 0, changed_flags;
331
332         mutex_lock(&kvm->lock);
333
334         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
335                                       assigned_irq->assigned_dev_id);
336         if (!match) {
337                 mutex_unlock(&kvm->lock);
338                 return -EINVAL;
339         }
340
341         if (!match->irq_requested_type) {
342                 INIT_WORK(&match->interrupt_work,
343                                 kvm_assigned_dev_interrupt_work_handler);
344                 if (irqchip_in_kernel(kvm)) {
345                         /* Register ack nofitier */
346                         match->ack_notifier.gsi = -1;
347                         match->ack_notifier.irq_acked =
348                                         kvm_assigned_dev_ack_irq;
349                         kvm_register_irq_ack_notifier(kvm,
350                                         &match->ack_notifier);
351
352                         /* Request IRQ source ID */
353                         r = kvm_request_irq_source_id(kvm);
354                         if (r < 0)
355                                 goto out_release;
356                         else
357                                 match->irq_source_id = r;
358
359 #ifdef CONFIG_X86
360                         /* Determine host device irq type, we can know the
361                          * result from dev->msi_enabled */
362                         if (msi2intx)
363                                 pci_enable_msi(match->dev);
364 #endif
365                 }
366         }
367
368         if ((match->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) &&
369                  (match->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_MSI))
370                 current_flags |= KVM_DEV_IRQ_ASSIGN_ENABLE_MSI;
371
372         changed_flags = assigned_irq->flags ^ current_flags;
373
374         if ((changed_flags & KVM_DEV_IRQ_ASSIGN_MSI_ACTION) ||
375             (msi2intx && match->dev->msi_enabled)) {
376 #ifdef CONFIG_X86
377                 r = assigned_device_update_msi(kvm, match, assigned_irq);
378                 if (r) {
379                         printk(KERN_WARNING "kvm: failed to enable "
380                                         "MSI device!\n");
381                         goto out_release;
382                 }
383 #else
384                 r = -ENOTTY;
385 #endif
386         } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
387                 /* Host device IRQ 0 means don't support INTx */
388                 if (!msi2intx) {
389                         printk(KERN_WARNING
390                                "kvm: wait device to enable MSI!\n");
391                         r = 0;
392                 } else {
393                         printk(KERN_WARNING
394                                "kvm: failed to enable MSI device!\n");
395                         r = -ENOTTY;
396                         goto out_release;
397                 }
398         } else {
399                 /* Non-sharing INTx mode */
400                 r = assigned_device_update_intx(kvm, match, assigned_irq);
401                 if (r) {
402                         printk(KERN_WARNING "kvm: failed to enable "
403                                         "INTx device!\n");
404                         goto out_release;
405                 }
406         }
407
408         mutex_unlock(&kvm->lock);
409         return r;
410 out_release:
411         mutex_unlock(&kvm->lock);
412         kvm_free_assigned_device(kvm, match);
413         return r;
414 }
415
416 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
417                                       struct kvm_assigned_pci_dev *assigned_dev)
418 {
419         int r = 0;
420         struct kvm_assigned_dev_kernel *match;
421         struct pci_dev *dev;
422
423         down_read(&kvm->slots_lock);
424         mutex_lock(&kvm->lock);
425
426         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
427                                       assigned_dev->assigned_dev_id);
428         if (match) {
429                 /* device already assigned */
430                 r = -EINVAL;
431                 goto out;
432         }
433
434         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
435         if (match == NULL) {
436                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
437                        __func__);
438                 r = -ENOMEM;
439                 goto out;
440         }
441         dev = pci_get_bus_and_slot(assigned_dev->busnr,
442                                    assigned_dev->devfn);
443         if (!dev) {
444                 printk(KERN_INFO "%s: host device not found\n", __func__);
445                 r = -EINVAL;
446                 goto out_free;
447         }
448         if (pci_enable_device(dev)) {
449                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
450                 r = -EBUSY;
451                 goto out_put;
452         }
453         r = pci_request_regions(dev, "kvm_assigned_device");
454         if (r) {
455                 printk(KERN_INFO "%s: Could not get access to device regions\n",
456                        __func__);
457                 goto out_disable;
458         }
459
460         pci_reset_function(dev);
461
462         match->assigned_dev_id = assigned_dev->assigned_dev_id;
463         match->host_busnr = assigned_dev->busnr;
464         match->host_devfn = assigned_dev->devfn;
465         match->flags = assigned_dev->flags;
466         match->dev = dev;
467         match->irq_source_id = -1;
468         match->kvm = kvm;
469
470         list_add(&match->list, &kvm->arch.assigned_dev_head);
471
472         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
473                 if (!kvm->arch.iommu_domain) {
474                         r = kvm_iommu_map_guest(kvm);
475                         if (r)
476                                 goto out_list_del;
477                 }
478                 r = kvm_assign_device(kvm, match);
479                 if (r)
480                         goto out_list_del;
481         }
482
483 out:
484         mutex_unlock(&kvm->lock);
485         up_read(&kvm->slots_lock);
486         return r;
487 out_list_del:
488         list_del(&match->list);
489         pci_release_regions(dev);
490 out_disable:
491         pci_disable_device(dev);
492 out_put:
493         pci_dev_put(dev);
494 out_free:
495         kfree(match);
496         mutex_unlock(&kvm->lock);
497         up_read(&kvm->slots_lock);
498         return r;
499 }
500 #endif
501
502 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
503 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
504                 struct kvm_assigned_pci_dev *assigned_dev)
505 {
506         int r = 0;
507         struct kvm_assigned_dev_kernel *match;
508
509         mutex_lock(&kvm->lock);
510
511         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
512                                       assigned_dev->assigned_dev_id);
513         if (!match) {
514                 printk(KERN_INFO "%s: device hasn't been assigned before, "
515                   "so cannot be deassigned\n", __func__);
516                 r = -EINVAL;
517                 goto out;
518         }
519
520         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
521                 kvm_deassign_device(kvm, match);
522
523         kvm_free_assigned_device(kvm, match);
524
525 out:
526         mutex_unlock(&kvm->lock);
527         return r;
528 }
529 #endif
530
531 static inline int valid_vcpu(int n)
532 {
533         return likely(n >= 0 && n < KVM_MAX_VCPUS);
534 }
535
536 inline int kvm_is_mmio_pfn(pfn_t pfn)
537 {
538         if (pfn_valid(pfn)) {
539                 struct page *page = compound_head(pfn_to_page(pfn));
540                 return PageReserved(page);
541         }
542
543         return true;
544 }
545
546 /*
547  * Switches to specified vcpu, until a matching vcpu_put()
548  */
549 void vcpu_load(struct kvm_vcpu *vcpu)
550 {
551         int cpu;
552
553         mutex_lock(&vcpu->mutex);
554         cpu = get_cpu();
555         preempt_notifier_register(&vcpu->preempt_notifier);
556         kvm_arch_vcpu_load(vcpu, cpu);
557         put_cpu();
558 }
559
560 void vcpu_put(struct kvm_vcpu *vcpu)
561 {
562         preempt_disable();
563         kvm_arch_vcpu_put(vcpu);
564         preempt_notifier_unregister(&vcpu->preempt_notifier);
565         preempt_enable();
566         mutex_unlock(&vcpu->mutex);
567 }
568
569 static void ack_flush(void *_completed)
570 {
571 }
572
573 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
574 {
575         int i, cpu, me;
576         cpumask_var_t cpus;
577         bool called = true;
578         struct kvm_vcpu *vcpu;
579
580         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
581                 cpumask_clear(cpus);
582
583         me = get_cpu();
584         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
585                 vcpu = kvm->vcpus[i];
586                 if (!vcpu)
587                         continue;
588                 if (test_and_set_bit(req, &vcpu->requests))
589                         continue;
590                 cpu = vcpu->cpu;
591                 if (cpus != NULL && cpu != -1 && cpu != me)
592                         cpumask_set_cpu(cpu, cpus);
593         }
594         if (unlikely(cpus == NULL))
595                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
596         else if (!cpumask_empty(cpus))
597                 smp_call_function_many(cpus, ack_flush, NULL, 1);
598         else
599                 called = false;
600         put_cpu();
601         free_cpumask_var(cpus);
602         return called;
603 }
604
605 void kvm_flush_remote_tlbs(struct kvm *kvm)
606 {
607         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
608                 ++kvm->stat.remote_tlb_flush;
609 }
610
611 void kvm_reload_remote_mmus(struct kvm *kvm)
612 {
613         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
614 }
615
616 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
617 {
618         struct page *page;
619         int r;
620
621         mutex_init(&vcpu->mutex);
622         vcpu->cpu = -1;
623         vcpu->kvm = kvm;
624         vcpu->vcpu_id = id;
625         init_waitqueue_head(&vcpu->wq);
626
627         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
628         if (!page) {
629                 r = -ENOMEM;
630                 goto fail;
631         }
632         vcpu->run = page_address(page);
633
634         r = kvm_arch_vcpu_init(vcpu);
635         if (r < 0)
636                 goto fail_free_run;
637         return 0;
638
639 fail_free_run:
640         free_page((unsigned long)vcpu->run);
641 fail:
642         return r;
643 }
644 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
645
646 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
647 {
648         kvm_arch_vcpu_uninit(vcpu);
649         free_page((unsigned long)vcpu->run);
650 }
651 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
652
653 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
654 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
655 {
656         return container_of(mn, struct kvm, mmu_notifier);
657 }
658
659 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
660                                              struct mm_struct *mm,
661                                              unsigned long address)
662 {
663         struct kvm *kvm = mmu_notifier_to_kvm(mn);
664         int need_tlb_flush;
665
666         /*
667          * When ->invalidate_page runs, the linux pte has been zapped
668          * already but the page is still allocated until
669          * ->invalidate_page returns. So if we increase the sequence
670          * here the kvm page fault will notice if the spte can't be
671          * established because the page is going to be freed. If
672          * instead the kvm page fault establishes the spte before
673          * ->invalidate_page runs, kvm_unmap_hva will release it
674          * before returning.
675          *
676          * The sequence increase only need to be seen at spin_unlock
677          * time, and not at spin_lock time.
678          *
679          * Increasing the sequence after the spin_unlock would be
680          * unsafe because the kvm page fault could then establish the
681          * pte after kvm_unmap_hva returned, without noticing the page
682          * is going to be freed.
683          */
684         spin_lock(&kvm->mmu_lock);
685         kvm->mmu_notifier_seq++;
686         need_tlb_flush = kvm_unmap_hva(kvm, address);
687         spin_unlock(&kvm->mmu_lock);
688
689         /* we've to flush the tlb before the pages can be freed */
690         if (need_tlb_flush)
691                 kvm_flush_remote_tlbs(kvm);
692
693 }
694
695 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
696                                                     struct mm_struct *mm,
697                                                     unsigned long start,
698                                                     unsigned long end)
699 {
700         struct kvm *kvm = mmu_notifier_to_kvm(mn);
701         int need_tlb_flush = 0;
702
703         spin_lock(&kvm->mmu_lock);
704         /*
705          * The count increase must become visible at unlock time as no
706          * spte can be established without taking the mmu_lock and
707          * count is also read inside the mmu_lock critical section.
708          */
709         kvm->mmu_notifier_count++;
710         for (; start < end; start += PAGE_SIZE)
711                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
712         spin_unlock(&kvm->mmu_lock);
713
714         /* we've to flush the tlb before the pages can be freed */
715         if (need_tlb_flush)
716                 kvm_flush_remote_tlbs(kvm);
717 }
718
719 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
720                                                   struct mm_struct *mm,
721                                                   unsigned long start,
722                                                   unsigned long end)
723 {
724         struct kvm *kvm = mmu_notifier_to_kvm(mn);
725
726         spin_lock(&kvm->mmu_lock);
727         /*
728          * This sequence increase will notify the kvm page fault that
729          * the page that is going to be mapped in the spte could have
730          * been freed.
731          */
732         kvm->mmu_notifier_seq++;
733         /*
734          * The above sequence increase must be visible before the
735          * below count decrease but both values are read by the kvm
736          * page fault under mmu_lock spinlock so we don't need to add
737          * a smb_wmb() here in between the two.
738          */
739         kvm->mmu_notifier_count--;
740         spin_unlock(&kvm->mmu_lock);
741
742         BUG_ON(kvm->mmu_notifier_count < 0);
743 }
744
745 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
746                                               struct mm_struct *mm,
747                                               unsigned long address)
748 {
749         struct kvm *kvm = mmu_notifier_to_kvm(mn);
750         int young;
751
752         spin_lock(&kvm->mmu_lock);
753         young = kvm_age_hva(kvm, address);
754         spin_unlock(&kvm->mmu_lock);
755
756         if (young)
757                 kvm_flush_remote_tlbs(kvm);
758
759         return young;
760 }
761
762 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
763                                      struct mm_struct *mm)
764 {
765         struct kvm *kvm = mmu_notifier_to_kvm(mn);
766         kvm_arch_flush_shadow(kvm);
767 }
768
769 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
770         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
771         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
772         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
773         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
774         .release                = kvm_mmu_notifier_release,
775 };
776 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
777
778 static struct kvm *kvm_create_vm(void)
779 {
780         struct kvm *kvm = kvm_arch_create_vm();
781 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
782         struct page *page;
783 #endif
784
785         if (IS_ERR(kvm))
786                 goto out;
787 #ifdef CONFIG_HAVE_KVM_IRQCHIP
788         INIT_LIST_HEAD(&kvm->irq_routing);
789         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
790 #endif
791
792 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
793         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
794         if (!page) {
795                 kfree(kvm);
796                 return ERR_PTR(-ENOMEM);
797         }
798         kvm->coalesced_mmio_ring =
799                         (struct kvm_coalesced_mmio_ring *)page_address(page);
800 #endif
801
802 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
803         {
804                 int err;
805                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
806                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
807                 if (err) {
808 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
809                         put_page(page);
810 #endif
811                         kfree(kvm);
812                         return ERR_PTR(err);
813                 }
814         }
815 #endif
816
817         kvm->mm = current->mm;
818         atomic_inc(&kvm->mm->mm_count);
819         spin_lock_init(&kvm->mmu_lock);
820         kvm_io_bus_init(&kvm->pio_bus);
821         mutex_init(&kvm->lock);
822         kvm_io_bus_init(&kvm->mmio_bus);
823         init_rwsem(&kvm->slots_lock);
824         atomic_set(&kvm->users_count, 1);
825         spin_lock(&kvm_lock);
826         list_add(&kvm->vm_list, &vm_list);
827         spin_unlock(&kvm_lock);
828 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
829         kvm_coalesced_mmio_init(kvm);
830 #endif
831 out:
832         return kvm;
833 }
834
835 /*
836  * Free any memory in @free but not in @dont.
837  */
838 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
839                                   struct kvm_memory_slot *dont)
840 {
841         if (!dont || free->rmap != dont->rmap)
842                 vfree(free->rmap);
843
844         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
845                 vfree(free->dirty_bitmap);
846
847         if (!dont || free->lpage_info != dont->lpage_info)
848                 vfree(free->lpage_info);
849
850         free->npages = 0;
851         free->dirty_bitmap = NULL;
852         free->rmap = NULL;
853         free->lpage_info = NULL;
854 }
855
856 void kvm_free_physmem(struct kvm *kvm)
857 {
858         int i;
859
860         for (i = 0; i < kvm->nmemslots; ++i)
861                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
862 }
863
864 static void kvm_destroy_vm(struct kvm *kvm)
865 {
866         struct mm_struct *mm = kvm->mm;
867
868         kvm_arch_sync_events(kvm);
869         spin_lock(&kvm_lock);
870         list_del(&kvm->vm_list);
871         spin_unlock(&kvm_lock);
872         kvm_free_irq_routing(kvm);
873         kvm_io_bus_destroy(&kvm->pio_bus);
874         kvm_io_bus_destroy(&kvm->mmio_bus);
875 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
876         if (kvm->coalesced_mmio_ring != NULL)
877                 free_page((unsigned long)kvm->coalesced_mmio_ring);
878 #endif
879 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
880         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
881 #endif
882         kvm_arch_destroy_vm(kvm);
883         mmdrop(mm);
884 }
885
886 void kvm_get_kvm(struct kvm *kvm)
887 {
888         atomic_inc(&kvm->users_count);
889 }
890 EXPORT_SYMBOL_GPL(kvm_get_kvm);
891
892 void kvm_put_kvm(struct kvm *kvm)
893 {
894         if (atomic_dec_and_test(&kvm->users_count))
895                 kvm_destroy_vm(kvm);
896 }
897 EXPORT_SYMBOL_GPL(kvm_put_kvm);
898
899
900 static int kvm_vm_release(struct inode *inode, struct file *filp)
901 {
902         struct kvm *kvm = filp->private_data;
903
904         kvm_put_kvm(kvm);
905         return 0;
906 }
907
908 /*
909  * Allocate some memory and give it an address in the guest physical address
910  * space.
911  *
912  * Discontiguous memory is allowed, mostly for framebuffers.
913  *
914  * Must be called holding mmap_sem for write.
915  */
916 int __kvm_set_memory_region(struct kvm *kvm,
917                             struct kvm_userspace_memory_region *mem,
918                             int user_alloc)
919 {
920         int r;
921         gfn_t base_gfn;
922         unsigned long npages;
923         int largepages;
924         unsigned long i;
925         struct kvm_memory_slot *memslot;
926         struct kvm_memory_slot old, new;
927
928         r = -EINVAL;
929         /* General sanity checks */
930         if (mem->memory_size & (PAGE_SIZE - 1))
931                 goto out;
932         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
933                 goto out;
934         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
935                 goto out;
936         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
937                 goto out;
938         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
939                 goto out;
940
941         memslot = &kvm->memslots[mem->slot];
942         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
943         npages = mem->memory_size >> PAGE_SHIFT;
944
945         if (!npages)
946                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
947
948         new = old = *memslot;
949
950         new.base_gfn = base_gfn;
951         new.npages = npages;
952         new.flags = mem->flags;
953
954         /* Disallow changing a memory slot's size. */
955         r = -EINVAL;
956         if (npages && old.npages && npages != old.npages)
957                 goto out_free;
958
959         /* Check for overlaps */
960         r = -EEXIST;
961         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
962                 struct kvm_memory_slot *s = &kvm->memslots[i];
963
964                 if (s == memslot || !s->npages)
965                         continue;
966                 if (!((base_gfn + npages <= s->base_gfn) ||
967                       (base_gfn >= s->base_gfn + s->npages)))
968                         goto out_free;
969         }
970
971         /* Free page dirty bitmap if unneeded */
972         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
973                 new.dirty_bitmap = NULL;
974
975         r = -ENOMEM;
976
977         /* Allocate if a slot is being created */
978 #ifndef CONFIG_S390
979         if (npages && !new.rmap) {
980                 new.rmap = vmalloc(npages * sizeof(struct page *));
981
982                 if (!new.rmap)
983                         goto out_free;
984
985                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
986
987                 new.user_alloc = user_alloc;
988                 /*
989                  * hva_to_rmmap() serialzies with the mmu_lock and to be
990                  * safe it has to ignore memslots with !user_alloc &&
991                  * !userspace_addr.
992                  */
993                 if (user_alloc)
994                         new.userspace_addr = mem->userspace_addr;
995                 else
996                         new.userspace_addr = 0;
997         }
998         if (npages && !new.lpage_info) {
999                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1000                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1001
1002                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1003
1004                 if (!new.lpage_info)
1005                         goto out_free;
1006
1007                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1008
1009                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1010                         new.lpage_info[0].write_count = 1;
1011                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1012                         new.lpage_info[largepages-1].write_count = 1;
1013         }
1014
1015         /* Allocate page dirty bitmap if needed */
1016         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1017                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1018
1019                 new.dirty_bitmap = vmalloc(dirty_bytes);
1020                 if (!new.dirty_bitmap)
1021                         goto out_free;
1022                 memset(new.dirty_bitmap, 0, dirty_bytes);
1023         }
1024 #endif /* not defined CONFIG_S390 */
1025
1026         if (!npages)
1027                 kvm_arch_flush_shadow(kvm);
1028
1029         spin_lock(&kvm->mmu_lock);
1030         if (mem->slot >= kvm->nmemslots)
1031                 kvm->nmemslots = mem->slot + 1;
1032
1033         *memslot = new;
1034         spin_unlock(&kvm->mmu_lock);
1035
1036         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1037         if (r) {
1038                 spin_lock(&kvm->mmu_lock);
1039                 *memslot = old;
1040                 spin_unlock(&kvm->mmu_lock);
1041                 goto out_free;
1042         }
1043
1044         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1045         /* Slot deletion case: we have to update the current slot */
1046         if (!npages)
1047                 *memslot = old;
1048 #ifdef CONFIG_DMAR
1049         /* map the pages in iommu page table */
1050         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1051         if (r)
1052                 goto out;
1053 #endif
1054         return 0;
1055
1056 out_free:
1057         kvm_free_physmem_slot(&new, &old);
1058 out:
1059         return r;
1060
1061 }
1062 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1063
1064 int kvm_set_memory_region(struct kvm *kvm,
1065                           struct kvm_userspace_memory_region *mem,
1066                           int user_alloc)
1067 {
1068         int r;
1069
1070         down_write(&kvm->slots_lock);
1071         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1072         up_write(&kvm->slots_lock);
1073         return r;
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1076
1077 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1078                                    struct
1079                                    kvm_userspace_memory_region *mem,
1080                                    int user_alloc)
1081 {
1082         if (mem->slot >= KVM_MEMORY_SLOTS)
1083                 return -EINVAL;
1084         return kvm_set_memory_region(kvm, mem, user_alloc);
1085 }
1086
1087 int kvm_get_dirty_log(struct kvm *kvm,
1088                         struct kvm_dirty_log *log, int *is_dirty)
1089 {
1090         struct kvm_memory_slot *memslot;
1091         int r, i;
1092         int n;
1093         unsigned long any = 0;
1094
1095         r = -EINVAL;
1096         if (log->slot >= KVM_MEMORY_SLOTS)
1097                 goto out;
1098
1099         memslot = &kvm->memslots[log->slot];
1100         r = -ENOENT;
1101         if (!memslot->dirty_bitmap)
1102                 goto out;
1103
1104         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1105
1106         for (i = 0; !any && i < n/sizeof(long); ++i)
1107                 any = memslot->dirty_bitmap[i];
1108
1109         r = -EFAULT;
1110         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1111                 goto out;
1112
1113         if (any)
1114                 *is_dirty = 1;
1115
1116         r = 0;
1117 out:
1118         return r;
1119 }
1120
1121 int is_error_page(struct page *page)
1122 {
1123         return page == bad_page;
1124 }
1125 EXPORT_SYMBOL_GPL(is_error_page);
1126
1127 int is_error_pfn(pfn_t pfn)
1128 {
1129         return pfn == bad_pfn;
1130 }
1131 EXPORT_SYMBOL_GPL(is_error_pfn);
1132
1133 static inline unsigned long bad_hva(void)
1134 {
1135         return PAGE_OFFSET;
1136 }
1137
1138 int kvm_is_error_hva(unsigned long addr)
1139 {
1140         return addr == bad_hva();
1141 }
1142 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1143
1144 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1145 {
1146         int i;
1147
1148         for (i = 0; i < kvm->nmemslots; ++i) {
1149                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1150
1151                 if (gfn >= memslot->base_gfn
1152                     && gfn < memslot->base_gfn + memslot->npages)
1153                         return memslot;
1154         }
1155         return NULL;
1156 }
1157 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1158
1159 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1160 {
1161         gfn = unalias_gfn(kvm, gfn);
1162         return gfn_to_memslot_unaliased(kvm, gfn);
1163 }
1164
1165 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1166 {
1167         int i;
1168
1169         gfn = unalias_gfn(kvm, gfn);
1170         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1171                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1172
1173                 if (gfn >= memslot->base_gfn
1174                     && gfn < memslot->base_gfn + memslot->npages)
1175                         return 1;
1176         }
1177         return 0;
1178 }
1179 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1180
1181 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1182 {
1183         struct kvm_memory_slot *slot;
1184
1185         gfn = unalias_gfn(kvm, gfn);
1186         slot = gfn_to_memslot_unaliased(kvm, gfn);
1187         if (!slot)
1188                 return bad_hva();
1189         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1190 }
1191 EXPORT_SYMBOL_GPL(gfn_to_hva);
1192
1193 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1194 {
1195         struct page *page[1];
1196         unsigned long addr;
1197         int npages;
1198         pfn_t pfn;
1199
1200         might_sleep();
1201
1202         addr = gfn_to_hva(kvm, gfn);
1203         if (kvm_is_error_hva(addr)) {
1204                 get_page(bad_page);
1205                 return page_to_pfn(bad_page);
1206         }
1207
1208         npages = get_user_pages_fast(addr, 1, 1, page);
1209
1210         if (unlikely(npages != 1)) {
1211                 struct vm_area_struct *vma;
1212
1213                 down_read(&current->mm->mmap_sem);
1214                 vma = find_vma(current->mm, addr);
1215
1216                 if (vma == NULL || addr < vma->vm_start ||
1217                     !(vma->vm_flags & VM_PFNMAP)) {
1218                         up_read(&current->mm->mmap_sem);
1219                         get_page(bad_page);
1220                         return page_to_pfn(bad_page);
1221                 }
1222
1223                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1224                 up_read(&current->mm->mmap_sem);
1225                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1226         } else
1227                 pfn = page_to_pfn(page[0]);
1228
1229         return pfn;
1230 }
1231
1232 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1233
1234 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1235 {
1236         pfn_t pfn;
1237
1238         pfn = gfn_to_pfn(kvm, gfn);
1239         if (!kvm_is_mmio_pfn(pfn))
1240                 return pfn_to_page(pfn);
1241
1242         WARN_ON(kvm_is_mmio_pfn(pfn));
1243
1244         get_page(bad_page);
1245         return bad_page;
1246 }
1247
1248 EXPORT_SYMBOL_GPL(gfn_to_page);
1249
1250 void kvm_release_page_clean(struct page *page)
1251 {
1252         kvm_release_pfn_clean(page_to_pfn(page));
1253 }
1254 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1255
1256 void kvm_release_pfn_clean(pfn_t pfn)
1257 {
1258         if (!kvm_is_mmio_pfn(pfn))
1259                 put_page(pfn_to_page(pfn));
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1262
1263 void kvm_release_page_dirty(struct page *page)
1264 {
1265         kvm_release_pfn_dirty(page_to_pfn(page));
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1268
1269 void kvm_release_pfn_dirty(pfn_t pfn)
1270 {
1271         kvm_set_pfn_dirty(pfn);
1272         kvm_release_pfn_clean(pfn);
1273 }
1274 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1275
1276 void kvm_set_page_dirty(struct page *page)
1277 {
1278         kvm_set_pfn_dirty(page_to_pfn(page));
1279 }
1280 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1281
1282 void kvm_set_pfn_dirty(pfn_t pfn)
1283 {
1284         if (!kvm_is_mmio_pfn(pfn)) {
1285                 struct page *page = pfn_to_page(pfn);
1286                 if (!PageReserved(page))
1287                         SetPageDirty(page);
1288         }
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1291
1292 void kvm_set_pfn_accessed(pfn_t pfn)
1293 {
1294         if (!kvm_is_mmio_pfn(pfn))
1295                 mark_page_accessed(pfn_to_page(pfn));
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1298
1299 void kvm_get_pfn(pfn_t pfn)
1300 {
1301         if (!kvm_is_mmio_pfn(pfn))
1302                 get_page(pfn_to_page(pfn));
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1305
1306 static int next_segment(unsigned long len, int offset)
1307 {
1308         if (len > PAGE_SIZE - offset)
1309                 return PAGE_SIZE - offset;
1310         else
1311                 return len;
1312 }
1313
1314 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1315                         int len)
1316 {
1317         int r;
1318         unsigned long addr;
1319
1320         addr = gfn_to_hva(kvm, gfn);
1321         if (kvm_is_error_hva(addr))
1322                 return -EFAULT;
1323         r = copy_from_user(data, (void __user *)addr + offset, len);
1324         if (r)
1325                 return -EFAULT;
1326         return 0;
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1329
1330 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1331 {
1332         gfn_t gfn = gpa >> PAGE_SHIFT;
1333         int seg;
1334         int offset = offset_in_page(gpa);
1335         int ret;
1336
1337         while ((seg = next_segment(len, offset)) != 0) {
1338                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1339                 if (ret < 0)
1340                         return ret;
1341                 offset = 0;
1342                 len -= seg;
1343                 data += seg;
1344                 ++gfn;
1345         }
1346         return 0;
1347 }
1348 EXPORT_SYMBOL_GPL(kvm_read_guest);
1349
1350 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1351                           unsigned long len)
1352 {
1353         int r;
1354         unsigned long addr;
1355         gfn_t gfn = gpa >> PAGE_SHIFT;
1356         int offset = offset_in_page(gpa);
1357
1358         addr = gfn_to_hva(kvm, gfn);
1359         if (kvm_is_error_hva(addr))
1360                 return -EFAULT;
1361         pagefault_disable();
1362         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1363         pagefault_enable();
1364         if (r)
1365                 return -EFAULT;
1366         return 0;
1367 }
1368 EXPORT_SYMBOL(kvm_read_guest_atomic);
1369
1370 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1371                          int offset, int len)
1372 {
1373         int r;
1374         unsigned long addr;
1375
1376         addr = gfn_to_hva(kvm, gfn);
1377         if (kvm_is_error_hva(addr))
1378                 return -EFAULT;
1379         r = copy_to_user((void __user *)addr + offset, data, len);
1380         if (r)
1381                 return -EFAULT;
1382         mark_page_dirty(kvm, gfn);
1383         return 0;
1384 }
1385 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1386
1387 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1388                     unsigned long len)
1389 {
1390         gfn_t gfn = gpa >> PAGE_SHIFT;
1391         int seg;
1392         int offset = offset_in_page(gpa);
1393         int ret;
1394
1395         while ((seg = next_segment(len, offset)) != 0) {
1396                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1397                 if (ret < 0)
1398                         return ret;
1399                 offset = 0;
1400                 len -= seg;
1401                 data += seg;
1402                 ++gfn;
1403         }
1404         return 0;
1405 }
1406
1407 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1408 {
1409         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1410 }
1411 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1412
1413 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1414 {
1415         gfn_t gfn = gpa >> PAGE_SHIFT;
1416         int seg;
1417         int offset = offset_in_page(gpa);
1418         int ret;
1419
1420         while ((seg = next_segment(len, offset)) != 0) {
1421                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1422                 if (ret < 0)
1423                         return ret;
1424                 offset = 0;
1425                 len -= seg;
1426                 ++gfn;
1427         }
1428         return 0;
1429 }
1430 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1431
1432 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1433 {
1434         struct kvm_memory_slot *memslot;
1435
1436         gfn = unalias_gfn(kvm, gfn);
1437         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1438         if (memslot && memslot->dirty_bitmap) {
1439                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1440
1441                 /* avoid RMW */
1442                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1443                         set_bit(rel_gfn, memslot->dirty_bitmap);
1444         }
1445 }
1446
1447 /*
1448  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1449  */
1450 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1451 {
1452         DEFINE_WAIT(wait);
1453
1454         for (;;) {
1455                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1456
1457                 if (kvm_cpu_has_interrupt(vcpu) ||
1458                     kvm_cpu_has_pending_timer(vcpu) ||
1459                     kvm_arch_vcpu_runnable(vcpu)) {
1460                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1461                         break;
1462                 }
1463                 if (signal_pending(current))
1464                         break;
1465
1466                 vcpu_put(vcpu);
1467                 schedule();
1468                 vcpu_load(vcpu);
1469         }
1470
1471         finish_wait(&vcpu->wq, &wait);
1472 }
1473
1474 void kvm_resched(struct kvm_vcpu *vcpu)
1475 {
1476         if (!need_resched())
1477                 return;
1478         cond_resched();
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_resched);
1481
1482 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1483 {
1484         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1485         struct page *page;
1486
1487         if (vmf->pgoff == 0)
1488                 page = virt_to_page(vcpu->run);
1489 #ifdef CONFIG_X86
1490         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1491                 page = virt_to_page(vcpu->arch.pio_data);
1492 #endif
1493 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1494         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1495                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1496 #endif
1497         else
1498                 return VM_FAULT_SIGBUS;
1499         get_page(page);
1500         vmf->page = page;
1501         return 0;
1502 }
1503
1504 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1505         .fault = kvm_vcpu_fault,
1506 };
1507
1508 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1509 {
1510         vma->vm_ops = &kvm_vcpu_vm_ops;
1511         return 0;
1512 }
1513
1514 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1515 {
1516         struct kvm_vcpu *vcpu = filp->private_data;
1517
1518         kvm_put_kvm(vcpu->kvm);
1519         return 0;
1520 }
1521
1522 static struct file_operations kvm_vcpu_fops = {
1523         .release        = kvm_vcpu_release,
1524         .unlocked_ioctl = kvm_vcpu_ioctl,
1525         .compat_ioctl   = kvm_vcpu_ioctl,
1526         .mmap           = kvm_vcpu_mmap,
1527 };
1528
1529 /*
1530  * Allocates an inode for the vcpu.
1531  */
1532 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1533 {
1534         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1535         if (fd < 0)
1536                 kvm_put_kvm(vcpu->kvm);
1537         return fd;
1538 }
1539
1540 /*
1541  * Creates some virtual cpus.  Good luck creating more than one.
1542  */
1543 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1544 {
1545         int r;
1546         struct kvm_vcpu *vcpu;
1547
1548         if (!valid_vcpu(n))
1549                 return -EINVAL;
1550
1551         vcpu = kvm_arch_vcpu_create(kvm, n);
1552         if (IS_ERR(vcpu))
1553                 return PTR_ERR(vcpu);
1554
1555         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1556
1557         r = kvm_arch_vcpu_setup(vcpu);
1558         if (r)
1559                 return r;
1560
1561         mutex_lock(&kvm->lock);
1562         if (kvm->vcpus[n]) {
1563                 r = -EEXIST;
1564                 goto vcpu_destroy;
1565         }
1566         kvm->vcpus[n] = vcpu;
1567         mutex_unlock(&kvm->lock);
1568
1569         /* Now it's all set up, let userspace reach it */
1570         kvm_get_kvm(kvm);
1571         r = create_vcpu_fd(vcpu);
1572         if (r < 0)
1573                 goto unlink;
1574         return r;
1575
1576 unlink:
1577         mutex_lock(&kvm->lock);
1578         kvm->vcpus[n] = NULL;
1579 vcpu_destroy:
1580         mutex_unlock(&kvm->lock);
1581         kvm_arch_vcpu_destroy(vcpu);
1582         return r;
1583 }
1584
1585 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1586 {
1587         if (sigset) {
1588                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1589                 vcpu->sigset_active = 1;
1590                 vcpu->sigset = *sigset;
1591         } else
1592                 vcpu->sigset_active = 0;
1593         return 0;
1594 }
1595
1596 static long kvm_vcpu_ioctl(struct file *filp,
1597                            unsigned int ioctl, unsigned long arg)
1598 {
1599         struct kvm_vcpu *vcpu = filp->private_data;
1600         void __user *argp = (void __user *)arg;
1601         int r;
1602         struct kvm_fpu *fpu = NULL;
1603         struct kvm_sregs *kvm_sregs = NULL;
1604
1605         if (vcpu->kvm->mm != current->mm)
1606                 return -EIO;
1607         switch (ioctl) {
1608         case KVM_RUN:
1609                 r = -EINVAL;
1610                 if (arg)
1611                         goto out;
1612                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1613                 break;
1614         case KVM_GET_REGS: {
1615                 struct kvm_regs *kvm_regs;
1616
1617                 r = -ENOMEM;
1618                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1619                 if (!kvm_regs)
1620                         goto out;
1621                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1622                 if (r)
1623                         goto out_free1;
1624                 r = -EFAULT;
1625                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1626                         goto out_free1;
1627                 r = 0;
1628 out_free1:
1629                 kfree(kvm_regs);
1630                 break;
1631         }
1632         case KVM_SET_REGS: {
1633                 struct kvm_regs *kvm_regs;
1634
1635                 r = -ENOMEM;
1636                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1637                 if (!kvm_regs)
1638                         goto out;
1639                 r = -EFAULT;
1640                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1641                         goto out_free2;
1642                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1643                 if (r)
1644                         goto out_free2;
1645                 r = 0;
1646 out_free2:
1647                 kfree(kvm_regs);
1648                 break;
1649         }
1650         case KVM_GET_SREGS: {
1651                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1652                 r = -ENOMEM;
1653                 if (!kvm_sregs)
1654                         goto out;
1655                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1656                 if (r)
1657                         goto out;
1658                 r = -EFAULT;
1659                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1660                         goto out;
1661                 r = 0;
1662                 break;
1663         }
1664         case KVM_SET_SREGS: {
1665                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1666                 r = -ENOMEM;
1667                 if (!kvm_sregs)
1668                         goto out;
1669                 r = -EFAULT;
1670                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1671                         goto out;
1672                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1673                 if (r)
1674                         goto out;
1675                 r = 0;
1676                 break;
1677         }
1678         case KVM_GET_MP_STATE: {
1679                 struct kvm_mp_state mp_state;
1680
1681                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1682                 if (r)
1683                         goto out;
1684                 r = -EFAULT;
1685                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1686                         goto out;
1687                 r = 0;
1688                 break;
1689         }
1690         case KVM_SET_MP_STATE: {
1691                 struct kvm_mp_state mp_state;
1692
1693                 r = -EFAULT;
1694                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1695                         goto out;
1696                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1697                 if (r)
1698                         goto out;
1699                 r = 0;
1700                 break;
1701         }
1702         case KVM_TRANSLATE: {
1703                 struct kvm_translation tr;
1704
1705                 r = -EFAULT;
1706                 if (copy_from_user(&tr, argp, sizeof tr))
1707                         goto out;
1708                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1709                 if (r)
1710                         goto out;
1711                 r = -EFAULT;
1712                 if (copy_to_user(argp, &tr, sizeof tr))
1713                         goto out;
1714                 r = 0;
1715                 break;
1716         }
1717         case KVM_SET_GUEST_DEBUG: {
1718                 struct kvm_guest_debug dbg;
1719
1720                 r = -EFAULT;
1721                 if (copy_from_user(&dbg, argp, sizeof dbg))
1722                         goto out;
1723                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1724                 if (r)
1725                         goto out;
1726                 r = 0;
1727                 break;
1728         }
1729         case KVM_SET_SIGNAL_MASK: {
1730                 struct kvm_signal_mask __user *sigmask_arg = argp;
1731                 struct kvm_signal_mask kvm_sigmask;
1732                 sigset_t sigset, *p;
1733
1734                 p = NULL;
1735                 if (argp) {
1736                         r = -EFAULT;
1737                         if (copy_from_user(&kvm_sigmask, argp,
1738                                            sizeof kvm_sigmask))
1739                                 goto out;
1740                         r = -EINVAL;
1741                         if (kvm_sigmask.len != sizeof sigset)
1742                                 goto out;
1743                         r = -EFAULT;
1744                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1745                                            sizeof sigset))
1746                                 goto out;
1747                         p = &sigset;
1748                 }
1749                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1750                 break;
1751         }
1752         case KVM_GET_FPU: {
1753                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1754                 r = -ENOMEM;
1755                 if (!fpu)
1756                         goto out;
1757                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1758                 if (r)
1759                         goto out;
1760                 r = -EFAULT;
1761                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1762                         goto out;
1763                 r = 0;
1764                 break;
1765         }
1766         case KVM_SET_FPU: {
1767                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1768                 r = -ENOMEM;
1769                 if (!fpu)
1770                         goto out;
1771                 r = -EFAULT;
1772                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1773                         goto out;
1774                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1775                 if (r)
1776                         goto out;
1777                 r = 0;
1778                 break;
1779         }
1780         default:
1781                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1782         }
1783 out:
1784         kfree(fpu);
1785         kfree(kvm_sregs);
1786         return r;
1787 }
1788
1789 static long kvm_vm_ioctl(struct file *filp,
1790                            unsigned int ioctl, unsigned long arg)
1791 {
1792         struct kvm *kvm = filp->private_data;
1793         void __user *argp = (void __user *)arg;
1794         int r;
1795
1796         if (kvm->mm != current->mm)
1797                 return -EIO;
1798         switch (ioctl) {
1799         case KVM_CREATE_VCPU:
1800                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1801                 if (r < 0)
1802                         goto out;
1803                 break;
1804         case KVM_SET_USER_MEMORY_REGION: {
1805                 struct kvm_userspace_memory_region kvm_userspace_mem;
1806
1807                 r = -EFAULT;
1808                 if (copy_from_user(&kvm_userspace_mem, argp,
1809                                                 sizeof kvm_userspace_mem))
1810                         goto out;
1811
1812                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1813                 if (r)
1814                         goto out;
1815                 break;
1816         }
1817         case KVM_GET_DIRTY_LOG: {
1818                 struct kvm_dirty_log log;
1819
1820                 r = -EFAULT;
1821                 if (copy_from_user(&log, argp, sizeof log))
1822                         goto out;
1823                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1824                 if (r)
1825                         goto out;
1826                 break;
1827         }
1828 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1829         case KVM_REGISTER_COALESCED_MMIO: {
1830                 struct kvm_coalesced_mmio_zone zone;
1831                 r = -EFAULT;
1832                 if (copy_from_user(&zone, argp, sizeof zone))
1833                         goto out;
1834                 r = -ENXIO;
1835                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1836                 if (r)
1837                         goto out;
1838                 r = 0;
1839                 break;
1840         }
1841         case KVM_UNREGISTER_COALESCED_MMIO: {
1842                 struct kvm_coalesced_mmio_zone zone;
1843                 r = -EFAULT;
1844                 if (copy_from_user(&zone, argp, sizeof zone))
1845                         goto out;
1846                 r = -ENXIO;
1847                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1848                 if (r)
1849                         goto out;
1850                 r = 0;
1851                 break;
1852         }
1853 #endif
1854 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1855         case KVM_ASSIGN_PCI_DEVICE: {
1856                 struct kvm_assigned_pci_dev assigned_dev;
1857
1858                 r = -EFAULT;
1859                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1860                         goto out;
1861                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1862                 if (r)
1863                         goto out;
1864                 break;
1865         }
1866         case KVM_ASSIGN_IRQ: {
1867                 struct kvm_assigned_irq assigned_irq;
1868
1869                 r = -EFAULT;
1870                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1871                         goto out;
1872                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1873                 if (r)
1874                         goto out;
1875                 break;
1876         }
1877 #endif
1878 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1879         case KVM_DEASSIGN_PCI_DEVICE: {
1880                 struct kvm_assigned_pci_dev assigned_dev;
1881
1882                 r = -EFAULT;
1883                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1884                         goto out;
1885                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1886                 if (r)
1887                         goto out;
1888                 break;
1889         }
1890 #endif
1891 #ifdef KVM_CAP_IRQ_ROUTING
1892         case KVM_SET_GSI_ROUTING: {
1893                 struct kvm_irq_routing routing;
1894                 struct kvm_irq_routing __user *urouting;
1895                 struct kvm_irq_routing_entry *entries;
1896
1897                 r = -EFAULT;
1898                 if (copy_from_user(&routing, argp, sizeof(routing)))
1899                         goto out;
1900                 r = -EINVAL;
1901                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
1902                         goto out;
1903                 if (routing.flags)
1904                         goto out;
1905                 r = -ENOMEM;
1906                 entries = vmalloc(routing.nr * sizeof(*entries));
1907                 if (!entries)
1908                         goto out;
1909                 r = -EFAULT;
1910                 urouting = argp;
1911                 if (copy_from_user(entries, urouting->entries,
1912                                    routing.nr * sizeof(*entries)))
1913                         goto out_free_irq_routing;
1914                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
1915                                         routing.flags);
1916         out_free_irq_routing:
1917                 vfree(entries);
1918                 break;
1919         }
1920 #endif
1921         default:
1922                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1923         }
1924 out:
1925         return r;
1926 }
1927
1928 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1929 {
1930         struct page *page[1];
1931         unsigned long addr;
1932         int npages;
1933         gfn_t gfn = vmf->pgoff;
1934         struct kvm *kvm = vma->vm_file->private_data;
1935
1936         addr = gfn_to_hva(kvm, gfn);
1937         if (kvm_is_error_hva(addr))
1938                 return VM_FAULT_SIGBUS;
1939
1940         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1941                                 NULL);
1942         if (unlikely(npages != 1))
1943                 return VM_FAULT_SIGBUS;
1944
1945         vmf->page = page[0];
1946         return 0;
1947 }
1948
1949 static struct vm_operations_struct kvm_vm_vm_ops = {
1950         .fault = kvm_vm_fault,
1951 };
1952
1953 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1954 {
1955         vma->vm_ops = &kvm_vm_vm_ops;
1956         return 0;
1957 }
1958
1959 static struct file_operations kvm_vm_fops = {
1960         .release        = kvm_vm_release,
1961         .unlocked_ioctl = kvm_vm_ioctl,
1962         .compat_ioctl   = kvm_vm_ioctl,
1963         .mmap           = kvm_vm_mmap,
1964 };
1965
1966 static int kvm_dev_ioctl_create_vm(void)
1967 {
1968         int fd;
1969         struct kvm *kvm;
1970
1971         kvm = kvm_create_vm();
1972         if (IS_ERR(kvm))
1973                 return PTR_ERR(kvm);
1974         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1975         if (fd < 0)
1976                 kvm_put_kvm(kvm);
1977
1978         return fd;
1979 }
1980
1981 static long kvm_dev_ioctl_check_extension_generic(long arg)
1982 {
1983         switch (arg) {
1984         case KVM_CAP_USER_MEMORY:
1985         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1986         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1987                 return 1;
1988 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1989         case KVM_CAP_IRQ_ROUTING:
1990                 return KVM_MAX_IRQ_ROUTES;
1991 #endif
1992         default:
1993                 break;
1994         }
1995         return kvm_dev_ioctl_check_extension(arg);
1996 }
1997
1998 static long kvm_dev_ioctl(struct file *filp,
1999                           unsigned int ioctl, unsigned long arg)
2000 {
2001         long r = -EINVAL;
2002
2003         switch (ioctl) {
2004         case KVM_GET_API_VERSION:
2005                 r = -EINVAL;
2006                 if (arg)
2007                         goto out;
2008                 r = KVM_API_VERSION;
2009                 break;
2010         case KVM_CREATE_VM:
2011                 r = -EINVAL;
2012                 if (arg)
2013                         goto out;
2014                 r = kvm_dev_ioctl_create_vm();
2015                 break;
2016         case KVM_CHECK_EXTENSION:
2017                 r = kvm_dev_ioctl_check_extension_generic(arg);
2018                 break;
2019         case KVM_GET_VCPU_MMAP_SIZE:
2020                 r = -EINVAL;
2021                 if (arg)
2022                         goto out;
2023                 r = PAGE_SIZE;     /* struct kvm_run */
2024 #ifdef CONFIG_X86
2025                 r += PAGE_SIZE;    /* pio data page */
2026 #endif
2027 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2028                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2029 #endif
2030                 break;
2031         case KVM_TRACE_ENABLE:
2032         case KVM_TRACE_PAUSE:
2033         case KVM_TRACE_DISABLE:
2034                 r = kvm_trace_ioctl(ioctl, arg);
2035                 break;
2036         default:
2037                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2038         }
2039 out:
2040         return r;
2041 }
2042
2043 static struct file_operations kvm_chardev_ops = {
2044         .unlocked_ioctl = kvm_dev_ioctl,
2045         .compat_ioctl   = kvm_dev_ioctl,
2046 };
2047
2048 static struct miscdevice kvm_dev = {
2049         KVM_MINOR,
2050         "kvm",
2051         &kvm_chardev_ops,
2052 };
2053
2054 static void hardware_enable(void *junk)
2055 {
2056         int cpu = raw_smp_processor_id();
2057
2058         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2059                 return;
2060         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2061         kvm_arch_hardware_enable(NULL);
2062 }
2063
2064 static void hardware_disable(void *junk)
2065 {
2066         int cpu = raw_smp_processor_id();
2067
2068         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2069                 return;
2070         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2071         kvm_arch_hardware_disable(NULL);
2072 }
2073
2074 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2075                            void *v)
2076 {
2077         int cpu = (long)v;
2078
2079         val &= ~CPU_TASKS_FROZEN;
2080         switch (val) {
2081         case CPU_DYING:
2082                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2083                        cpu);
2084                 hardware_disable(NULL);
2085                 break;
2086         case CPU_UP_CANCELED:
2087                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2088                        cpu);
2089                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2090                 break;
2091         case CPU_ONLINE:
2092                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2093                        cpu);
2094                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2095                 break;
2096         }
2097         return NOTIFY_OK;
2098 }
2099
2100
2101 asmlinkage void kvm_handle_fault_on_reboot(void)
2102 {
2103         if (kvm_rebooting)
2104                 /* spin while reset goes on */
2105                 while (true)
2106                         ;
2107         /* Fault while not rebooting.  We want the trace. */
2108         BUG();
2109 }
2110 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2111
2112 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2113                       void *v)
2114 {
2115         if (val == SYS_RESTART) {
2116                 /*
2117                  * Some (well, at least mine) BIOSes hang on reboot if
2118                  * in vmx root mode.
2119                  */
2120                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2121                 kvm_rebooting = true;
2122                 on_each_cpu(hardware_disable, NULL, 1);
2123         }
2124         return NOTIFY_OK;
2125 }
2126
2127 static struct notifier_block kvm_reboot_notifier = {
2128         .notifier_call = kvm_reboot,
2129         .priority = 0,
2130 };
2131
2132 void kvm_io_bus_init(struct kvm_io_bus *bus)
2133 {
2134         memset(bus, 0, sizeof(*bus));
2135 }
2136
2137 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2138 {
2139         int i;
2140
2141         for (i = 0; i < bus->dev_count; i++) {
2142                 struct kvm_io_device *pos = bus->devs[i];
2143
2144                 kvm_iodevice_destructor(pos);
2145         }
2146 }
2147
2148 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2149                                           gpa_t addr, int len, int is_write)
2150 {
2151         int i;
2152
2153         for (i = 0; i < bus->dev_count; i++) {
2154                 struct kvm_io_device *pos = bus->devs[i];
2155
2156                 if (pos->in_range(pos, addr, len, is_write))
2157                         return pos;
2158         }
2159
2160         return NULL;
2161 }
2162
2163 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2164 {
2165         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2166
2167         bus->devs[bus->dev_count++] = dev;
2168 }
2169
2170 static struct notifier_block kvm_cpu_notifier = {
2171         .notifier_call = kvm_cpu_hotplug,
2172         .priority = 20, /* must be > scheduler priority */
2173 };
2174
2175 static int vm_stat_get(void *_offset, u64 *val)
2176 {
2177         unsigned offset = (long)_offset;
2178         struct kvm *kvm;
2179
2180         *val = 0;
2181         spin_lock(&kvm_lock);
2182         list_for_each_entry(kvm, &vm_list, vm_list)
2183                 *val += *(u32 *)((void *)kvm + offset);
2184         spin_unlock(&kvm_lock);
2185         return 0;
2186 }
2187
2188 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2189
2190 static int vcpu_stat_get(void *_offset, u64 *val)
2191 {
2192         unsigned offset = (long)_offset;
2193         struct kvm *kvm;
2194         struct kvm_vcpu *vcpu;
2195         int i;
2196
2197         *val = 0;
2198         spin_lock(&kvm_lock);
2199         list_for_each_entry(kvm, &vm_list, vm_list)
2200                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2201                         vcpu = kvm->vcpus[i];
2202                         if (vcpu)
2203                                 *val += *(u32 *)((void *)vcpu + offset);
2204                 }
2205         spin_unlock(&kvm_lock);
2206         return 0;
2207 }
2208
2209 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2210
2211 static struct file_operations *stat_fops[] = {
2212         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2213         [KVM_STAT_VM]   = &vm_stat_fops,
2214 };
2215
2216 static void kvm_init_debug(void)
2217 {
2218         struct kvm_stats_debugfs_item *p;
2219
2220         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2221         for (p = debugfs_entries; p->name; ++p)
2222                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2223                                                 (void *)(long)p->offset,
2224                                                 stat_fops[p->kind]);
2225 }
2226
2227 static void kvm_exit_debug(void)
2228 {
2229         struct kvm_stats_debugfs_item *p;
2230
2231         for (p = debugfs_entries; p->name; ++p)
2232                 debugfs_remove(p->dentry);
2233         debugfs_remove(kvm_debugfs_dir);
2234 }
2235
2236 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2237 {
2238         hardware_disable(NULL);
2239         return 0;
2240 }
2241
2242 static int kvm_resume(struct sys_device *dev)
2243 {
2244         hardware_enable(NULL);
2245         return 0;
2246 }
2247
2248 static struct sysdev_class kvm_sysdev_class = {
2249         .name = "kvm",
2250         .suspend = kvm_suspend,
2251         .resume = kvm_resume,
2252 };
2253
2254 static struct sys_device kvm_sysdev = {
2255         .id = 0,
2256         .cls = &kvm_sysdev_class,
2257 };
2258
2259 struct page *bad_page;
2260 pfn_t bad_pfn;
2261
2262 static inline
2263 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2264 {
2265         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2266 }
2267
2268 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2269 {
2270         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2271
2272         kvm_arch_vcpu_load(vcpu, cpu);
2273 }
2274
2275 static void kvm_sched_out(struct preempt_notifier *pn,
2276                           struct task_struct *next)
2277 {
2278         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2279
2280         kvm_arch_vcpu_put(vcpu);
2281 }
2282
2283 int kvm_init(void *opaque, unsigned int vcpu_size,
2284                   struct module *module)
2285 {
2286         int r;
2287         int cpu;
2288
2289         kvm_init_debug();
2290
2291         r = kvm_arch_init(opaque);
2292         if (r)
2293                 goto out_fail;
2294
2295         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2296
2297         if (bad_page == NULL) {
2298                 r = -ENOMEM;
2299                 goto out;
2300         }
2301
2302         bad_pfn = page_to_pfn(bad_page);
2303
2304         if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2305                 r = -ENOMEM;
2306                 goto out_free_0;
2307         }
2308
2309         r = kvm_arch_hardware_setup();
2310         if (r < 0)
2311                 goto out_free_0a;
2312
2313         for_each_online_cpu(cpu) {
2314                 smp_call_function_single(cpu,
2315                                 kvm_arch_check_processor_compat,
2316                                 &r, 1);
2317                 if (r < 0)
2318                         goto out_free_1;
2319         }
2320
2321         on_each_cpu(hardware_enable, NULL, 1);
2322         r = register_cpu_notifier(&kvm_cpu_notifier);
2323         if (r)
2324                 goto out_free_2;
2325         register_reboot_notifier(&kvm_reboot_notifier);
2326
2327         r = sysdev_class_register(&kvm_sysdev_class);
2328         if (r)
2329                 goto out_free_3;
2330
2331         r = sysdev_register(&kvm_sysdev);
2332         if (r)
2333                 goto out_free_4;
2334
2335         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2336         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2337                                            __alignof__(struct kvm_vcpu),
2338                                            0, NULL);
2339         if (!kvm_vcpu_cache) {
2340                 r = -ENOMEM;
2341                 goto out_free_5;
2342         }
2343
2344         kvm_chardev_ops.owner = module;
2345         kvm_vm_fops.owner = module;
2346         kvm_vcpu_fops.owner = module;
2347
2348         r = misc_register(&kvm_dev);
2349         if (r) {
2350                 printk(KERN_ERR "kvm: misc device register failed\n");
2351                 goto out_free;
2352         }
2353
2354         kvm_preempt_ops.sched_in = kvm_sched_in;
2355         kvm_preempt_ops.sched_out = kvm_sched_out;
2356 #ifndef CONFIG_X86
2357         msi2intx = 0;
2358 #endif
2359
2360         return 0;
2361
2362 out_free:
2363         kmem_cache_destroy(kvm_vcpu_cache);
2364 out_free_5:
2365         sysdev_unregister(&kvm_sysdev);
2366 out_free_4:
2367         sysdev_class_unregister(&kvm_sysdev_class);
2368 out_free_3:
2369         unregister_reboot_notifier(&kvm_reboot_notifier);
2370         unregister_cpu_notifier(&kvm_cpu_notifier);
2371 out_free_2:
2372         on_each_cpu(hardware_disable, NULL, 1);
2373 out_free_1:
2374         kvm_arch_hardware_unsetup();
2375 out_free_0a:
2376         free_cpumask_var(cpus_hardware_enabled);
2377 out_free_0:
2378         __free_page(bad_page);
2379 out:
2380         kvm_arch_exit();
2381         kvm_exit_debug();
2382 out_fail:
2383         return r;
2384 }
2385 EXPORT_SYMBOL_GPL(kvm_init);
2386
2387 void kvm_exit(void)
2388 {
2389         kvm_trace_cleanup();
2390         misc_deregister(&kvm_dev);
2391         kmem_cache_destroy(kvm_vcpu_cache);
2392         sysdev_unregister(&kvm_sysdev);
2393         sysdev_class_unregister(&kvm_sysdev_class);
2394         unregister_reboot_notifier(&kvm_reboot_notifier);
2395         unregister_cpu_notifier(&kvm_cpu_notifier);
2396         on_each_cpu(hardware_disable, NULL, 1);
2397         kvm_arch_hardware_unsetup();
2398         kvm_arch_exit();
2399         kvm_exit_debug();
2400         free_cpumask_var(cpus_hardware_enabled);
2401         __free_page(bad_page);
2402 }
2403 EXPORT_SYMBOL_GPL(kvm_exit);