KVM: Fix assigned devices circular locking dependency
[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_var_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 static 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                 assigned_dev->host_irq_disabled = false;
174         }
175         mutex_unlock(&assigned_dev->kvm->lock);
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         schedule_work(&assigned_dev->interrupt_work);
184
185         disable_irq_nosync(irq);
186         assigned_dev->host_irq_disabled = true;
187
188         return IRQ_HANDLED;
189 }
190
191 /* Ack the irq line for an assigned device */
192 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
193 {
194         struct kvm_assigned_dev_kernel *dev;
195
196         if (kian->gsi == -1)
197                 return;
198
199         dev = container_of(kian, struct kvm_assigned_dev_kernel,
200                            ack_notifier);
201
202         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
203
204         /* The guest irq may be shared so this ack may be
205          * from another device.
206          */
207         if (dev->host_irq_disabled) {
208                 enable_irq(dev->host_irq);
209                 dev->host_irq_disabled = false;
210         }
211 }
212
213 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
214 static void kvm_free_assigned_irq(struct kvm *kvm,
215                                   struct kvm_assigned_dev_kernel *assigned_dev)
216 {
217         if (!irqchip_in_kernel(kvm))
218                 return;
219
220         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
221
222         if (assigned_dev->irq_source_id != -1)
223                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
224         assigned_dev->irq_source_id = -1;
225
226         if (!assigned_dev->irq_requested_type)
227                 return;
228
229         /*
230          * In kvm_free_device_irq, cancel_work_sync return true if:
231          * 1. work is scheduled, and then cancelled.
232          * 2. work callback is executed.
233          *
234          * The first one ensured that the irq is disabled and no more events
235          * would happen. But for the second one, the irq may be enabled (e.g.
236          * for MSI). So we disable irq here to prevent further events.
237          *
238          * Notice this maybe result in nested disable if the interrupt type is
239          * INTx, but it's OK for we are going to free it.
240          *
241          * If this function is a part of VM destroy, please ensure that till
242          * now, the kvm state is still legal for probably we also have to wait
243          * interrupt_work done.
244          */
245         disable_irq_nosync(assigned_dev->host_irq);
246         cancel_work_sync(&assigned_dev->interrupt_work);
247
248         free_irq(assigned_dev->host_irq, (void *)assigned_dev);
249
250         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
251                 pci_disable_msi(assigned_dev->dev);
252
253         assigned_dev->irq_requested_type = 0;
254 }
255
256
257 static void kvm_free_assigned_device(struct kvm *kvm,
258                                      struct kvm_assigned_dev_kernel
259                                      *assigned_dev)
260 {
261         kvm_free_assigned_irq(kvm, assigned_dev);
262
263         pci_reset_function(assigned_dev->dev);
264
265         pci_release_regions(assigned_dev->dev);
266         pci_disable_device(assigned_dev->dev);
267         pci_dev_put(assigned_dev->dev);
268
269         list_del(&assigned_dev->list);
270         kfree(assigned_dev);
271 }
272
273 void kvm_free_all_assigned_devices(struct kvm *kvm)
274 {
275         struct list_head *ptr, *ptr2;
276         struct kvm_assigned_dev_kernel *assigned_dev;
277
278         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
279                 assigned_dev = list_entry(ptr,
280                                           struct kvm_assigned_dev_kernel,
281                                           list);
282
283                 kvm_free_assigned_device(kvm, assigned_dev);
284         }
285 }
286
287 static int assigned_device_update_intx(struct kvm *kvm,
288                         struct kvm_assigned_dev_kernel *adev,
289                         struct kvm_assigned_irq *airq)
290 {
291         adev->guest_irq = airq->guest_irq;
292         adev->ack_notifier.gsi = airq->guest_irq;
293
294         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
295                 return 0;
296
297         if (irqchip_in_kernel(kvm)) {
298                 if (!msi2intx &&
299                     (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)) {
300                         free_irq(adev->host_irq, (void *)adev);
301                         pci_disable_msi(adev->dev);
302                 }
303
304                 if (!capable(CAP_SYS_RAWIO))
305                         return -EPERM;
306
307                 if (airq->host_irq)
308                         adev->host_irq = airq->host_irq;
309                 else
310                         adev->host_irq = adev->dev->irq;
311
312                 /* Even though this is PCI, we don't want to use shared
313                  * interrupts. Sharing host devices with guest-assigned devices
314                  * on the same interrupt line is not a happy situation: there
315                  * are going to be long delays in accepting, acking, etc.
316                  */
317                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
318                                 0, "kvm_assigned_intx_device", (void *)adev))
319                         return -EIO;
320         }
321
322         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
323                                    KVM_ASSIGNED_DEV_HOST_INTX;
324         return 0;
325 }
326
327 #ifdef CONFIG_X86
328 static int assigned_device_update_msi(struct kvm *kvm,
329                         struct kvm_assigned_dev_kernel *adev,
330                         struct kvm_assigned_irq *airq)
331 {
332         int r;
333
334         if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
335                 /* x86 don't care upper address of guest msi message addr */
336                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
337                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
338                 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
339                 adev->guest_msi.data = airq->guest_msi.data;
340                 adev->ack_notifier.gsi = -1;
341         } else if (msi2intx) {
342                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
343                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
344                 adev->guest_irq = airq->guest_irq;
345                 adev->ack_notifier.gsi = airq->guest_irq;
346         }
347
348         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
349                 return 0;
350
351         if (irqchip_in_kernel(kvm)) {
352                 if (!msi2intx) {
353                         if (adev->irq_requested_type &
354                                         KVM_ASSIGNED_DEV_HOST_INTX)
355                                 free_irq(adev->host_irq, (void *)adev);
356
357                         r = pci_enable_msi(adev->dev);
358                         if (r)
359                                 return r;
360                 }
361
362                 adev->host_irq = adev->dev->irq;
363                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
364                                 "kvm_assigned_msi_device", (void *)adev))
365                         return -EIO;
366         }
367
368         if (!msi2intx)
369                 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
370
371         adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
372         return 0;
373 }
374 #endif
375
376 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
377                                    struct kvm_assigned_irq
378                                    *assigned_irq)
379 {
380         int r = 0;
381         struct kvm_assigned_dev_kernel *match;
382
383         mutex_lock(&kvm->lock);
384
385         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
386                                       assigned_irq->assigned_dev_id);
387         if (!match) {
388                 mutex_unlock(&kvm->lock);
389                 return -EINVAL;
390         }
391
392         if (!match->irq_requested_type) {
393                 INIT_WORK(&match->interrupt_work,
394                                 kvm_assigned_dev_interrupt_work_handler);
395                 if (irqchip_in_kernel(kvm)) {
396                         /* Register ack nofitier */
397                         match->ack_notifier.gsi = -1;
398                         match->ack_notifier.irq_acked =
399                                         kvm_assigned_dev_ack_irq;
400                         kvm_register_irq_ack_notifier(kvm,
401                                         &match->ack_notifier);
402
403                         /* Request IRQ source ID */
404                         r = kvm_request_irq_source_id(kvm);
405                         if (r < 0)
406                                 goto out_release;
407                         else
408                                 match->irq_source_id = r;
409
410 #ifdef CONFIG_X86
411                         /* Determine host device irq type, we can know the
412                          * result from dev->msi_enabled */
413                         if (msi2intx)
414                                 pci_enable_msi(match->dev);
415 #endif
416                 }
417         }
418
419         if ((!msi2intx &&
420              (assigned_irq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI)) ||
421             (msi2intx && match->dev->msi_enabled)) {
422 #ifdef CONFIG_X86
423                 r = assigned_device_update_msi(kvm, match, assigned_irq);
424                 if (r) {
425                         printk(KERN_WARNING "kvm: failed to enable "
426                                         "MSI device!\n");
427                         goto out_release;
428                 }
429 #else
430                 r = -ENOTTY;
431 #endif
432         } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
433                 /* Host device IRQ 0 means don't support INTx */
434                 if (!msi2intx) {
435                         printk(KERN_WARNING
436                                "kvm: wait device to enable MSI!\n");
437                         r = 0;
438                 } else {
439                         printk(KERN_WARNING
440                                "kvm: failed to enable MSI device!\n");
441                         r = -ENOTTY;
442                         goto out_release;
443                 }
444         } else {
445                 /* Non-sharing INTx mode */
446                 r = assigned_device_update_intx(kvm, match, assigned_irq);
447                 if (r) {
448                         printk(KERN_WARNING "kvm: failed to enable "
449                                         "INTx device!\n");
450                         goto out_release;
451                 }
452         }
453
454         mutex_unlock(&kvm->lock);
455         return r;
456 out_release:
457         mutex_unlock(&kvm->lock);
458         kvm_free_assigned_device(kvm, match);
459         return r;
460 }
461
462 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
463                                       struct kvm_assigned_pci_dev *assigned_dev)
464 {
465         int r = 0;
466         struct kvm_assigned_dev_kernel *match;
467         struct pci_dev *dev;
468
469         down_read(&kvm->slots_lock);
470         mutex_lock(&kvm->lock);
471
472         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
473                                       assigned_dev->assigned_dev_id);
474         if (match) {
475                 /* device already assigned */
476                 r = -EINVAL;
477                 goto out;
478         }
479
480         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
481         if (match == NULL) {
482                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
483                        __func__);
484                 r = -ENOMEM;
485                 goto out;
486         }
487         dev = pci_get_bus_and_slot(assigned_dev->busnr,
488                                    assigned_dev->devfn);
489         if (!dev) {
490                 printk(KERN_INFO "%s: host device not found\n", __func__);
491                 r = -EINVAL;
492                 goto out_free;
493         }
494         if (pci_enable_device(dev)) {
495                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
496                 r = -EBUSY;
497                 goto out_put;
498         }
499         r = pci_request_regions(dev, "kvm_assigned_device");
500         if (r) {
501                 printk(KERN_INFO "%s: Could not get access to device regions\n",
502                        __func__);
503                 goto out_disable;
504         }
505
506         pci_reset_function(dev);
507
508         match->assigned_dev_id = assigned_dev->assigned_dev_id;
509         match->host_busnr = assigned_dev->busnr;
510         match->host_devfn = assigned_dev->devfn;
511         match->flags = assigned_dev->flags;
512         match->dev = dev;
513         match->irq_source_id = -1;
514         match->kvm = kvm;
515
516         list_add(&match->list, &kvm->arch.assigned_dev_head);
517
518         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
519                 if (!kvm->arch.iommu_domain) {
520                         r = kvm_iommu_map_guest(kvm);
521                         if (r)
522                                 goto out_list_del;
523                 }
524                 r = kvm_assign_device(kvm, match);
525                 if (r)
526                         goto out_list_del;
527         }
528
529 out:
530         mutex_unlock(&kvm->lock);
531         up_read(&kvm->slots_lock);
532         return r;
533 out_list_del:
534         list_del(&match->list);
535         pci_release_regions(dev);
536 out_disable:
537         pci_disable_device(dev);
538 out_put:
539         pci_dev_put(dev);
540 out_free:
541         kfree(match);
542         mutex_unlock(&kvm->lock);
543         up_read(&kvm->slots_lock);
544         return r;
545 }
546 #endif
547
548 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
549 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
550                 struct kvm_assigned_pci_dev *assigned_dev)
551 {
552         int r = 0;
553         struct kvm_assigned_dev_kernel *match;
554
555         mutex_lock(&kvm->lock);
556
557         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
558                                       assigned_dev->assigned_dev_id);
559         if (!match) {
560                 printk(KERN_INFO "%s: device hasn't been assigned before, "
561                   "so cannot be deassigned\n", __func__);
562                 r = -EINVAL;
563                 goto out;
564         }
565
566         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
567                 kvm_deassign_device(kvm, match);
568
569         kvm_free_assigned_device(kvm, match);
570
571 out:
572         mutex_unlock(&kvm->lock);
573         return r;
574 }
575 #endif
576
577 static inline int valid_vcpu(int n)
578 {
579         return likely(n >= 0 && n < KVM_MAX_VCPUS);
580 }
581
582 inline int kvm_is_mmio_pfn(pfn_t pfn)
583 {
584         if (pfn_valid(pfn))
585                 return PageReserved(pfn_to_page(pfn));
586
587         return true;
588 }
589
590 /*
591  * Switches to specified vcpu, until a matching vcpu_put()
592  */
593 void vcpu_load(struct kvm_vcpu *vcpu)
594 {
595         int cpu;
596
597         mutex_lock(&vcpu->mutex);
598         cpu = get_cpu();
599         preempt_notifier_register(&vcpu->preempt_notifier);
600         kvm_arch_vcpu_load(vcpu, cpu);
601         put_cpu();
602 }
603
604 void vcpu_put(struct kvm_vcpu *vcpu)
605 {
606         preempt_disable();
607         kvm_arch_vcpu_put(vcpu);
608         preempt_notifier_unregister(&vcpu->preempt_notifier);
609         preempt_enable();
610         mutex_unlock(&vcpu->mutex);
611 }
612
613 static void ack_flush(void *_completed)
614 {
615 }
616
617 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
618 {
619         int i, cpu, me;
620         cpumask_var_t cpus;
621         bool called = true;
622         struct kvm_vcpu *vcpu;
623
624         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
625                 cpumask_clear(cpus);
626
627         me = get_cpu();
628         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
629                 vcpu = kvm->vcpus[i];
630                 if (!vcpu)
631                         continue;
632                 if (test_and_set_bit(req, &vcpu->requests))
633                         continue;
634                 cpu = vcpu->cpu;
635                 if (cpus != NULL && cpu != -1 && cpu != me)
636                         cpumask_set_cpu(cpu, cpus);
637         }
638         if (unlikely(cpus == NULL))
639                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
640         else if (!cpumask_empty(cpus))
641                 smp_call_function_many(cpus, ack_flush, NULL, 1);
642         else
643                 called = false;
644         put_cpu();
645         free_cpumask_var(cpus);
646         return called;
647 }
648
649 void kvm_flush_remote_tlbs(struct kvm *kvm)
650 {
651         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
652                 ++kvm->stat.remote_tlb_flush;
653 }
654
655 void kvm_reload_remote_mmus(struct kvm *kvm)
656 {
657         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
658 }
659
660 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
661 {
662         struct page *page;
663         int r;
664
665         mutex_init(&vcpu->mutex);
666         vcpu->cpu = -1;
667         vcpu->kvm = kvm;
668         vcpu->vcpu_id = id;
669         init_waitqueue_head(&vcpu->wq);
670
671         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
672         if (!page) {
673                 r = -ENOMEM;
674                 goto fail;
675         }
676         vcpu->run = page_address(page);
677
678         r = kvm_arch_vcpu_init(vcpu);
679         if (r < 0)
680                 goto fail_free_run;
681         return 0;
682
683 fail_free_run:
684         free_page((unsigned long)vcpu->run);
685 fail:
686         return r;
687 }
688 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
689
690 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
691 {
692         kvm_arch_vcpu_uninit(vcpu);
693         free_page((unsigned long)vcpu->run);
694 }
695 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
696
697 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
698 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
699 {
700         return container_of(mn, struct kvm, mmu_notifier);
701 }
702
703 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
704                                              struct mm_struct *mm,
705                                              unsigned long address)
706 {
707         struct kvm *kvm = mmu_notifier_to_kvm(mn);
708         int need_tlb_flush;
709
710         /*
711          * When ->invalidate_page runs, the linux pte has been zapped
712          * already but the page is still allocated until
713          * ->invalidate_page returns. So if we increase the sequence
714          * here the kvm page fault will notice if the spte can't be
715          * established because the page is going to be freed. If
716          * instead the kvm page fault establishes the spte before
717          * ->invalidate_page runs, kvm_unmap_hva will release it
718          * before returning.
719          *
720          * The sequence increase only need to be seen at spin_unlock
721          * time, and not at spin_lock time.
722          *
723          * Increasing the sequence after the spin_unlock would be
724          * unsafe because the kvm page fault could then establish the
725          * pte after kvm_unmap_hva returned, without noticing the page
726          * is going to be freed.
727          */
728         spin_lock(&kvm->mmu_lock);
729         kvm->mmu_notifier_seq++;
730         need_tlb_flush = kvm_unmap_hva(kvm, address);
731         spin_unlock(&kvm->mmu_lock);
732
733         /* we've to flush the tlb before the pages can be freed */
734         if (need_tlb_flush)
735                 kvm_flush_remote_tlbs(kvm);
736
737 }
738
739 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
740                                                     struct mm_struct *mm,
741                                                     unsigned long start,
742                                                     unsigned long end)
743 {
744         struct kvm *kvm = mmu_notifier_to_kvm(mn);
745         int need_tlb_flush = 0;
746
747         spin_lock(&kvm->mmu_lock);
748         /*
749          * The count increase must become visible at unlock time as no
750          * spte can be established without taking the mmu_lock and
751          * count is also read inside the mmu_lock critical section.
752          */
753         kvm->mmu_notifier_count++;
754         for (; start < end; start += PAGE_SIZE)
755                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
756         spin_unlock(&kvm->mmu_lock);
757
758         /* we've to flush the tlb before the pages can be freed */
759         if (need_tlb_flush)
760                 kvm_flush_remote_tlbs(kvm);
761 }
762
763 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
764                                                   struct mm_struct *mm,
765                                                   unsigned long start,
766                                                   unsigned long end)
767 {
768         struct kvm *kvm = mmu_notifier_to_kvm(mn);
769
770         spin_lock(&kvm->mmu_lock);
771         /*
772          * This sequence increase will notify the kvm page fault that
773          * the page that is going to be mapped in the spte could have
774          * been freed.
775          */
776         kvm->mmu_notifier_seq++;
777         /*
778          * The above sequence increase must be visible before the
779          * below count decrease but both values are read by the kvm
780          * page fault under mmu_lock spinlock so we don't need to add
781          * a smb_wmb() here in between the two.
782          */
783         kvm->mmu_notifier_count--;
784         spin_unlock(&kvm->mmu_lock);
785
786         BUG_ON(kvm->mmu_notifier_count < 0);
787 }
788
789 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
790                                               struct mm_struct *mm,
791                                               unsigned long address)
792 {
793         struct kvm *kvm = mmu_notifier_to_kvm(mn);
794         int young;
795
796         spin_lock(&kvm->mmu_lock);
797         young = kvm_age_hva(kvm, address);
798         spin_unlock(&kvm->mmu_lock);
799
800         if (young)
801                 kvm_flush_remote_tlbs(kvm);
802
803         return young;
804 }
805
806 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
807                                      struct mm_struct *mm)
808 {
809         struct kvm *kvm = mmu_notifier_to_kvm(mn);
810         kvm_arch_flush_shadow(kvm);
811 }
812
813 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
814         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
815         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
816         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
817         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
818         .release                = kvm_mmu_notifier_release,
819 };
820 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
821
822 static struct kvm *kvm_create_vm(void)
823 {
824         struct kvm *kvm = kvm_arch_create_vm();
825 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
826         struct page *page;
827 #endif
828
829         if (IS_ERR(kvm))
830                 goto out;
831
832 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
833         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
834         if (!page) {
835                 kfree(kvm);
836                 return ERR_PTR(-ENOMEM);
837         }
838         kvm->coalesced_mmio_ring =
839                         (struct kvm_coalesced_mmio_ring *)page_address(page);
840 #endif
841
842 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
843         {
844                 int err;
845                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
846                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
847                 if (err) {
848 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
849                         put_page(page);
850 #endif
851                         kfree(kvm);
852                         return ERR_PTR(err);
853                 }
854         }
855 #endif
856
857         kvm->mm = current->mm;
858         atomic_inc(&kvm->mm->mm_count);
859         spin_lock_init(&kvm->mmu_lock);
860         kvm_io_bus_init(&kvm->pio_bus);
861         mutex_init(&kvm->lock);
862         kvm_io_bus_init(&kvm->mmio_bus);
863         init_rwsem(&kvm->slots_lock);
864         atomic_set(&kvm->users_count, 1);
865         spin_lock(&kvm_lock);
866         list_add(&kvm->vm_list, &vm_list);
867         spin_unlock(&kvm_lock);
868 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
869         kvm_coalesced_mmio_init(kvm);
870 #endif
871 out:
872         return kvm;
873 }
874
875 /*
876  * Free any memory in @free but not in @dont.
877  */
878 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
879                                   struct kvm_memory_slot *dont)
880 {
881         if (!dont || free->rmap != dont->rmap)
882                 vfree(free->rmap);
883
884         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
885                 vfree(free->dirty_bitmap);
886
887         if (!dont || free->lpage_info != dont->lpage_info)
888                 vfree(free->lpage_info);
889
890         free->npages = 0;
891         free->dirty_bitmap = NULL;
892         free->rmap = NULL;
893         free->lpage_info = NULL;
894 }
895
896 void kvm_free_physmem(struct kvm *kvm)
897 {
898         int i;
899
900         for (i = 0; i < kvm->nmemslots; ++i)
901                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
902 }
903
904 static void kvm_destroy_vm(struct kvm *kvm)
905 {
906         struct mm_struct *mm = kvm->mm;
907
908         kvm_arch_sync_events(kvm);
909         spin_lock(&kvm_lock);
910         list_del(&kvm->vm_list);
911         spin_unlock(&kvm_lock);
912         kvm_io_bus_destroy(&kvm->pio_bus);
913         kvm_io_bus_destroy(&kvm->mmio_bus);
914 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
915         if (kvm->coalesced_mmio_ring != NULL)
916                 free_page((unsigned long)kvm->coalesced_mmio_ring);
917 #endif
918 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
919         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
920 #endif
921         kvm_arch_destroy_vm(kvm);
922         mmdrop(mm);
923 }
924
925 void kvm_get_kvm(struct kvm *kvm)
926 {
927         atomic_inc(&kvm->users_count);
928 }
929 EXPORT_SYMBOL_GPL(kvm_get_kvm);
930
931 void kvm_put_kvm(struct kvm *kvm)
932 {
933         if (atomic_dec_and_test(&kvm->users_count))
934                 kvm_destroy_vm(kvm);
935 }
936 EXPORT_SYMBOL_GPL(kvm_put_kvm);
937
938
939 static int kvm_vm_release(struct inode *inode, struct file *filp)
940 {
941         struct kvm *kvm = filp->private_data;
942
943         kvm_put_kvm(kvm);
944         return 0;
945 }
946
947 /*
948  * Allocate some memory and give it an address in the guest physical address
949  * space.
950  *
951  * Discontiguous memory is allowed, mostly for framebuffers.
952  *
953  * Must be called holding mmap_sem for write.
954  */
955 int __kvm_set_memory_region(struct kvm *kvm,
956                             struct kvm_userspace_memory_region *mem,
957                             int user_alloc)
958 {
959         int r;
960         gfn_t base_gfn;
961         unsigned long npages;
962         unsigned long i;
963         struct kvm_memory_slot *memslot;
964         struct kvm_memory_slot old, new;
965
966         r = -EINVAL;
967         /* General sanity checks */
968         if (mem->memory_size & (PAGE_SIZE - 1))
969                 goto out;
970         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
971                 goto out;
972         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
973                 goto out;
974         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
975                 goto out;
976         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
977                 goto out;
978
979         memslot = &kvm->memslots[mem->slot];
980         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
981         npages = mem->memory_size >> PAGE_SHIFT;
982
983         if (!npages)
984                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
985
986         new = old = *memslot;
987
988         new.base_gfn = base_gfn;
989         new.npages = npages;
990         new.flags = mem->flags;
991
992         /* Disallow changing a memory slot's size. */
993         r = -EINVAL;
994         if (npages && old.npages && npages != old.npages)
995                 goto out_free;
996
997         /* Check for overlaps */
998         r = -EEXIST;
999         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1000                 struct kvm_memory_slot *s = &kvm->memslots[i];
1001
1002                 if (s == memslot)
1003                         continue;
1004                 if (!((base_gfn + npages <= s->base_gfn) ||
1005                       (base_gfn >= s->base_gfn + s->npages)))
1006                         goto out_free;
1007         }
1008
1009         /* Free page dirty bitmap if unneeded */
1010         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1011                 new.dirty_bitmap = NULL;
1012
1013         r = -ENOMEM;
1014
1015         /* Allocate if a slot is being created */
1016 #ifndef CONFIG_S390
1017         if (npages && !new.rmap) {
1018                 new.rmap = vmalloc(npages * sizeof(struct page *));
1019
1020                 if (!new.rmap)
1021                         goto out_free;
1022
1023                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1024
1025                 new.user_alloc = user_alloc;
1026                 /*
1027                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1028                  * safe it has to ignore memslots with !user_alloc &&
1029                  * !userspace_addr.
1030                  */
1031                 if (user_alloc)
1032                         new.userspace_addr = mem->userspace_addr;
1033                 else
1034                         new.userspace_addr = 0;
1035         }
1036         if (npages && !new.lpage_info) {
1037                 int largepages = npages / KVM_PAGES_PER_HPAGE;
1038                 if (npages % KVM_PAGES_PER_HPAGE)
1039                         largepages++;
1040                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1041                         largepages++;
1042
1043                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1044
1045                 if (!new.lpage_info)
1046                         goto out_free;
1047
1048                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1049
1050                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1051                         new.lpage_info[0].write_count = 1;
1052                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1053                         new.lpage_info[largepages-1].write_count = 1;
1054         }
1055
1056         /* Allocate page dirty bitmap if needed */
1057         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1058                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1059
1060                 new.dirty_bitmap = vmalloc(dirty_bytes);
1061                 if (!new.dirty_bitmap)
1062                         goto out_free;
1063                 memset(new.dirty_bitmap, 0, dirty_bytes);
1064         }
1065 #endif /* not defined CONFIG_S390 */
1066
1067         if (!npages)
1068                 kvm_arch_flush_shadow(kvm);
1069
1070         spin_lock(&kvm->mmu_lock);
1071         if (mem->slot >= kvm->nmemslots)
1072                 kvm->nmemslots = mem->slot + 1;
1073
1074         *memslot = new;
1075         spin_unlock(&kvm->mmu_lock);
1076
1077         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1078         if (r) {
1079                 spin_lock(&kvm->mmu_lock);
1080                 *memslot = old;
1081                 spin_unlock(&kvm->mmu_lock);
1082                 goto out_free;
1083         }
1084
1085         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1086         /* Slot deletion case: we have to update the current slot */
1087         if (!npages)
1088                 *memslot = old;
1089 #ifdef CONFIG_DMAR
1090         /* map the pages in iommu page table */
1091         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1092         if (r)
1093                 goto out;
1094 #endif
1095         return 0;
1096
1097 out_free:
1098         kvm_free_physmem_slot(&new, &old);
1099 out:
1100         return r;
1101
1102 }
1103 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1104
1105 int kvm_set_memory_region(struct kvm *kvm,
1106                           struct kvm_userspace_memory_region *mem,
1107                           int user_alloc)
1108 {
1109         int r;
1110
1111         down_write(&kvm->slots_lock);
1112         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1113         up_write(&kvm->slots_lock);
1114         return r;
1115 }
1116 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1117
1118 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1119                                    struct
1120                                    kvm_userspace_memory_region *mem,
1121                                    int user_alloc)
1122 {
1123         if (mem->slot >= KVM_MEMORY_SLOTS)
1124                 return -EINVAL;
1125         return kvm_set_memory_region(kvm, mem, user_alloc);
1126 }
1127
1128 int kvm_get_dirty_log(struct kvm *kvm,
1129                         struct kvm_dirty_log *log, int *is_dirty)
1130 {
1131         struct kvm_memory_slot *memslot;
1132         int r, i;
1133         int n;
1134         unsigned long any = 0;
1135
1136         r = -EINVAL;
1137         if (log->slot >= KVM_MEMORY_SLOTS)
1138                 goto out;
1139
1140         memslot = &kvm->memslots[log->slot];
1141         r = -ENOENT;
1142         if (!memslot->dirty_bitmap)
1143                 goto out;
1144
1145         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1146
1147         for (i = 0; !any && i < n/sizeof(long); ++i)
1148                 any = memslot->dirty_bitmap[i];
1149
1150         r = -EFAULT;
1151         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1152                 goto out;
1153
1154         if (any)
1155                 *is_dirty = 1;
1156
1157         r = 0;
1158 out:
1159         return r;
1160 }
1161
1162 int is_error_page(struct page *page)
1163 {
1164         return page == bad_page;
1165 }
1166 EXPORT_SYMBOL_GPL(is_error_page);
1167
1168 int is_error_pfn(pfn_t pfn)
1169 {
1170         return pfn == bad_pfn;
1171 }
1172 EXPORT_SYMBOL_GPL(is_error_pfn);
1173
1174 static inline unsigned long bad_hva(void)
1175 {
1176         return PAGE_OFFSET;
1177 }
1178
1179 int kvm_is_error_hva(unsigned long addr)
1180 {
1181         return addr == bad_hva();
1182 }
1183 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1184
1185 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1186 {
1187         int i;
1188
1189         for (i = 0; i < kvm->nmemslots; ++i) {
1190                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1191
1192                 if (gfn >= memslot->base_gfn
1193                     && gfn < memslot->base_gfn + memslot->npages)
1194                         return memslot;
1195         }
1196         return NULL;
1197 }
1198 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1199
1200 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1201 {
1202         gfn = unalias_gfn(kvm, gfn);
1203         return gfn_to_memslot_unaliased(kvm, gfn);
1204 }
1205
1206 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1207 {
1208         int i;
1209
1210         gfn = unalias_gfn(kvm, gfn);
1211         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1212                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1213
1214                 if (gfn >= memslot->base_gfn
1215                     && gfn < memslot->base_gfn + memslot->npages)
1216                         return 1;
1217         }
1218         return 0;
1219 }
1220 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1221
1222 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1223 {
1224         struct kvm_memory_slot *slot;
1225
1226         gfn = unalias_gfn(kvm, gfn);
1227         slot = gfn_to_memslot_unaliased(kvm, gfn);
1228         if (!slot)
1229                 return bad_hva();
1230         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1231 }
1232 EXPORT_SYMBOL_GPL(gfn_to_hva);
1233
1234 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1235 {
1236         struct page *page[1];
1237         unsigned long addr;
1238         int npages;
1239         pfn_t pfn;
1240
1241         might_sleep();
1242
1243         addr = gfn_to_hva(kvm, gfn);
1244         if (kvm_is_error_hva(addr)) {
1245                 get_page(bad_page);
1246                 return page_to_pfn(bad_page);
1247         }
1248
1249         npages = get_user_pages_fast(addr, 1, 1, page);
1250
1251         if (unlikely(npages != 1)) {
1252                 struct vm_area_struct *vma;
1253
1254                 down_read(&current->mm->mmap_sem);
1255                 vma = find_vma(current->mm, addr);
1256
1257                 if (vma == NULL || addr < vma->vm_start ||
1258                     !(vma->vm_flags & VM_PFNMAP)) {
1259                         up_read(&current->mm->mmap_sem);
1260                         get_page(bad_page);
1261                         return page_to_pfn(bad_page);
1262                 }
1263
1264                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1265                 up_read(&current->mm->mmap_sem);
1266                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1267         } else
1268                 pfn = page_to_pfn(page[0]);
1269
1270         return pfn;
1271 }
1272
1273 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1274
1275 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1276 {
1277         pfn_t pfn;
1278
1279         pfn = gfn_to_pfn(kvm, gfn);
1280         if (!kvm_is_mmio_pfn(pfn))
1281                 return pfn_to_page(pfn);
1282
1283         WARN_ON(kvm_is_mmio_pfn(pfn));
1284
1285         get_page(bad_page);
1286         return bad_page;
1287 }
1288
1289 EXPORT_SYMBOL_GPL(gfn_to_page);
1290
1291 void kvm_release_page_clean(struct page *page)
1292 {
1293         kvm_release_pfn_clean(page_to_pfn(page));
1294 }
1295 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1296
1297 void kvm_release_pfn_clean(pfn_t pfn)
1298 {
1299         if (!kvm_is_mmio_pfn(pfn))
1300                 put_page(pfn_to_page(pfn));
1301 }
1302 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1303
1304 void kvm_release_page_dirty(struct page *page)
1305 {
1306         kvm_release_pfn_dirty(page_to_pfn(page));
1307 }
1308 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1309
1310 void kvm_release_pfn_dirty(pfn_t pfn)
1311 {
1312         kvm_set_pfn_dirty(pfn);
1313         kvm_release_pfn_clean(pfn);
1314 }
1315 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1316
1317 void kvm_set_page_dirty(struct page *page)
1318 {
1319         kvm_set_pfn_dirty(page_to_pfn(page));
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1322
1323 void kvm_set_pfn_dirty(pfn_t pfn)
1324 {
1325         if (!kvm_is_mmio_pfn(pfn)) {
1326                 struct page *page = pfn_to_page(pfn);
1327                 if (!PageReserved(page))
1328                         SetPageDirty(page);
1329         }
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1332
1333 void kvm_set_pfn_accessed(pfn_t pfn)
1334 {
1335         if (!kvm_is_mmio_pfn(pfn))
1336                 mark_page_accessed(pfn_to_page(pfn));
1337 }
1338 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1339
1340 void kvm_get_pfn(pfn_t pfn)
1341 {
1342         if (!kvm_is_mmio_pfn(pfn))
1343                 get_page(pfn_to_page(pfn));
1344 }
1345 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1346
1347 static int next_segment(unsigned long len, int offset)
1348 {
1349         if (len > PAGE_SIZE - offset)
1350                 return PAGE_SIZE - offset;
1351         else
1352                 return len;
1353 }
1354
1355 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1356                         int len)
1357 {
1358         int r;
1359         unsigned long addr;
1360
1361         addr = gfn_to_hva(kvm, gfn);
1362         if (kvm_is_error_hva(addr))
1363                 return -EFAULT;
1364         r = copy_from_user(data, (void __user *)addr + offset, len);
1365         if (r)
1366                 return -EFAULT;
1367         return 0;
1368 }
1369 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1370
1371 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, 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_read_guest_page(kvm, gfn, data, offset, seg);
1380                 if (ret < 0)
1381                         return ret;
1382                 offset = 0;
1383                 len -= seg;
1384                 data += seg;
1385                 ++gfn;
1386         }
1387         return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_read_guest);
1390
1391 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1392                           unsigned long len)
1393 {
1394         int r;
1395         unsigned long addr;
1396         gfn_t gfn = gpa >> PAGE_SHIFT;
1397         int offset = offset_in_page(gpa);
1398
1399         addr = gfn_to_hva(kvm, gfn);
1400         if (kvm_is_error_hva(addr))
1401                 return -EFAULT;
1402         pagefault_disable();
1403         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1404         pagefault_enable();
1405         if (r)
1406                 return -EFAULT;
1407         return 0;
1408 }
1409 EXPORT_SYMBOL(kvm_read_guest_atomic);
1410
1411 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1412                          int offset, int len)
1413 {
1414         int r;
1415         unsigned long addr;
1416
1417         addr = gfn_to_hva(kvm, gfn);
1418         if (kvm_is_error_hva(addr))
1419                 return -EFAULT;
1420         r = copy_to_user((void __user *)addr + offset, data, len);
1421         if (r)
1422                 return -EFAULT;
1423         mark_page_dirty(kvm, gfn);
1424         return 0;
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1427
1428 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1429                     unsigned long len)
1430 {
1431         gfn_t gfn = gpa >> PAGE_SHIFT;
1432         int seg;
1433         int offset = offset_in_page(gpa);
1434         int ret;
1435
1436         while ((seg = next_segment(len, offset)) != 0) {
1437                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1438                 if (ret < 0)
1439                         return ret;
1440                 offset = 0;
1441                 len -= seg;
1442                 data += seg;
1443                 ++gfn;
1444         }
1445         return 0;
1446 }
1447
1448 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1449 {
1450         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1453
1454 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1455 {
1456         gfn_t gfn = gpa >> PAGE_SHIFT;
1457         int seg;
1458         int offset = offset_in_page(gpa);
1459         int ret;
1460
1461         while ((seg = next_segment(len, offset)) != 0) {
1462                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1463                 if (ret < 0)
1464                         return ret;
1465                 offset = 0;
1466                 len -= seg;
1467                 ++gfn;
1468         }
1469         return 0;
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1472
1473 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1474 {
1475         struct kvm_memory_slot *memslot;
1476
1477         gfn = unalias_gfn(kvm, gfn);
1478         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1479         if (memslot && memslot->dirty_bitmap) {
1480                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1481
1482                 /* avoid RMW */
1483                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1484                         set_bit(rel_gfn, memslot->dirty_bitmap);
1485         }
1486 }
1487
1488 /*
1489  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1490  */
1491 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1492 {
1493         DEFINE_WAIT(wait);
1494
1495         for (;;) {
1496                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1497
1498                 if (kvm_cpu_has_interrupt(vcpu) ||
1499                     kvm_cpu_has_pending_timer(vcpu) ||
1500                     kvm_arch_vcpu_runnable(vcpu)) {
1501                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1502                         break;
1503                 }
1504                 if (signal_pending(current))
1505                         break;
1506
1507                 vcpu_put(vcpu);
1508                 schedule();
1509                 vcpu_load(vcpu);
1510         }
1511
1512         finish_wait(&vcpu->wq, &wait);
1513 }
1514
1515 void kvm_resched(struct kvm_vcpu *vcpu)
1516 {
1517         if (!need_resched())
1518                 return;
1519         cond_resched();
1520 }
1521 EXPORT_SYMBOL_GPL(kvm_resched);
1522
1523 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1524 {
1525         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1526         struct page *page;
1527
1528         if (vmf->pgoff == 0)
1529                 page = virt_to_page(vcpu->run);
1530 #ifdef CONFIG_X86
1531         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1532                 page = virt_to_page(vcpu->arch.pio_data);
1533 #endif
1534 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1535         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1536                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1537 #endif
1538         else
1539                 return VM_FAULT_SIGBUS;
1540         get_page(page);
1541         vmf->page = page;
1542         return 0;
1543 }
1544
1545 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1546         .fault = kvm_vcpu_fault,
1547 };
1548
1549 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1550 {
1551         vma->vm_ops = &kvm_vcpu_vm_ops;
1552         return 0;
1553 }
1554
1555 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1556 {
1557         struct kvm_vcpu *vcpu = filp->private_data;
1558
1559         kvm_put_kvm(vcpu->kvm);
1560         return 0;
1561 }
1562
1563 static struct file_operations kvm_vcpu_fops = {
1564         .release        = kvm_vcpu_release,
1565         .unlocked_ioctl = kvm_vcpu_ioctl,
1566         .compat_ioctl   = kvm_vcpu_ioctl,
1567         .mmap           = kvm_vcpu_mmap,
1568 };
1569
1570 /*
1571  * Allocates an inode for the vcpu.
1572  */
1573 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1574 {
1575         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1576         if (fd < 0)
1577                 kvm_put_kvm(vcpu->kvm);
1578         return fd;
1579 }
1580
1581 /*
1582  * Creates some virtual cpus.  Good luck creating more than one.
1583  */
1584 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1585 {
1586         int r;
1587         struct kvm_vcpu *vcpu;
1588
1589         if (!valid_vcpu(n))
1590                 return -EINVAL;
1591
1592         vcpu = kvm_arch_vcpu_create(kvm, n);
1593         if (IS_ERR(vcpu))
1594                 return PTR_ERR(vcpu);
1595
1596         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1597
1598         r = kvm_arch_vcpu_setup(vcpu);
1599         if (r)
1600                 return r;
1601
1602         mutex_lock(&kvm->lock);
1603         if (kvm->vcpus[n]) {
1604                 r = -EEXIST;
1605                 goto vcpu_destroy;
1606         }
1607         kvm->vcpus[n] = vcpu;
1608         mutex_unlock(&kvm->lock);
1609
1610         /* Now it's all set up, let userspace reach it */
1611         kvm_get_kvm(kvm);
1612         r = create_vcpu_fd(vcpu);
1613         if (r < 0)
1614                 goto unlink;
1615         return r;
1616
1617 unlink:
1618         mutex_lock(&kvm->lock);
1619         kvm->vcpus[n] = NULL;
1620 vcpu_destroy:
1621         mutex_unlock(&kvm->lock);
1622         kvm_arch_vcpu_destroy(vcpu);
1623         return r;
1624 }
1625
1626 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1627 {
1628         if (sigset) {
1629                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1630                 vcpu->sigset_active = 1;
1631                 vcpu->sigset = *sigset;
1632         } else
1633                 vcpu->sigset_active = 0;
1634         return 0;
1635 }
1636
1637 static long kvm_vcpu_ioctl(struct file *filp,
1638                            unsigned int ioctl, unsigned long arg)
1639 {
1640         struct kvm_vcpu *vcpu = filp->private_data;
1641         void __user *argp = (void __user *)arg;
1642         int r;
1643         struct kvm_fpu *fpu = NULL;
1644         struct kvm_sregs *kvm_sregs = NULL;
1645
1646         if (vcpu->kvm->mm != current->mm)
1647                 return -EIO;
1648         switch (ioctl) {
1649         case KVM_RUN:
1650                 r = -EINVAL;
1651                 if (arg)
1652                         goto out;
1653                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1654                 break;
1655         case KVM_GET_REGS: {
1656                 struct kvm_regs *kvm_regs;
1657
1658                 r = -ENOMEM;
1659                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1660                 if (!kvm_regs)
1661                         goto out;
1662                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1663                 if (r)
1664                         goto out_free1;
1665                 r = -EFAULT;
1666                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1667                         goto out_free1;
1668                 r = 0;
1669 out_free1:
1670                 kfree(kvm_regs);
1671                 break;
1672         }
1673         case KVM_SET_REGS: {
1674                 struct kvm_regs *kvm_regs;
1675
1676                 r = -ENOMEM;
1677                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1678                 if (!kvm_regs)
1679                         goto out;
1680                 r = -EFAULT;
1681                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1682                         goto out_free2;
1683                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1684                 if (r)
1685                         goto out_free2;
1686                 r = 0;
1687 out_free2:
1688                 kfree(kvm_regs);
1689                 break;
1690         }
1691         case KVM_GET_SREGS: {
1692                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1693                 r = -ENOMEM;
1694                 if (!kvm_sregs)
1695                         goto out;
1696                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1697                 if (r)
1698                         goto out;
1699                 r = -EFAULT;
1700                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1701                         goto out;
1702                 r = 0;
1703                 break;
1704         }
1705         case KVM_SET_SREGS: {
1706                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1707                 r = -ENOMEM;
1708                 if (!kvm_sregs)
1709                         goto out;
1710                 r = -EFAULT;
1711                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1712                         goto out;
1713                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1714                 if (r)
1715                         goto out;
1716                 r = 0;
1717                 break;
1718         }
1719         case KVM_GET_MP_STATE: {
1720                 struct kvm_mp_state mp_state;
1721
1722                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1723                 if (r)
1724                         goto out;
1725                 r = -EFAULT;
1726                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1727                         goto out;
1728                 r = 0;
1729                 break;
1730         }
1731         case KVM_SET_MP_STATE: {
1732                 struct kvm_mp_state mp_state;
1733
1734                 r = -EFAULT;
1735                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1736                         goto out;
1737                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1738                 if (r)
1739                         goto out;
1740                 r = 0;
1741                 break;
1742         }
1743         case KVM_TRANSLATE: {
1744                 struct kvm_translation tr;
1745
1746                 r = -EFAULT;
1747                 if (copy_from_user(&tr, argp, sizeof tr))
1748                         goto out;
1749                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1750                 if (r)
1751                         goto out;
1752                 r = -EFAULT;
1753                 if (copy_to_user(argp, &tr, sizeof tr))
1754                         goto out;
1755                 r = 0;
1756                 break;
1757         }
1758         case KVM_DEBUG_GUEST: {
1759                 struct kvm_debug_guest dbg;
1760
1761                 r = -EFAULT;
1762                 if (copy_from_user(&dbg, argp, sizeof dbg))
1763                         goto out;
1764                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1765                 if (r)
1766                         goto out;
1767                 r = 0;
1768                 break;
1769         }
1770         case KVM_SET_SIGNAL_MASK: {
1771                 struct kvm_signal_mask __user *sigmask_arg = argp;
1772                 struct kvm_signal_mask kvm_sigmask;
1773                 sigset_t sigset, *p;
1774
1775                 p = NULL;
1776                 if (argp) {
1777                         r = -EFAULT;
1778                         if (copy_from_user(&kvm_sigmask, argp,
1779                                            sizeof kvm_sigmask))
1780                                 goto out;
1781                         r = -EINVAL;
1782                         if (kvm_sigmask.len != sizeof sigset)
1783                                 goto out;
1784                         r = -EFAULT;
1785                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1786                                            sizeof sigset))
1787                                 goto out;
1788                         p = &sigset;
1789                 }
1790                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1791                 break;
1792         }
1793         case KVM_GET_FPU: {
1794                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1795                 r = -ENOMEM;
1796                 if (!fpu)
1797                         goto out;
1798                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1799                 if (r)
1800                         goto out;
1801                 r = -EFAULT;
1802                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1803                         goto out;
1804                 r = 0;
1805                 break;
1806         }
1807         case KVM_SET_FPU: {
1808                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1809                 r = -ENOMEM;
1810                 if (!fpu)
1811                         goto out;
1812                 r = -EFAULT;
1813                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1814                         goto out;
1815                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1816                 if (r)
1817                         goto out;
1818                 r = 0;
1819                 break;
1820         }
1821         default:
1822                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1823         }
1824 out:
1825         kfree(fpu);
1826         kfree(kvm_sregs);
1827         return r;
1828 }
1829
1830 static long kvm_vm_ioctl(struct file *filp,
1831                            unsigned int ioctl, unsigned long arg)
1832 {
1833         struct kvm *kvm = filp->private_data;
1834         void __user *argp = (void __user *)arg;
1835         int r;
1836
1837         if (kvm->mm != current->mm)
1838                 return -EIO;
1839         switch (ioctl) {
1840         case KVM_CREATE_VCPU:
1841                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1842                 if (r < 0)
1843                         goto out;
1844                 break;
1845         case KVM_SET_USER_MEMORY_REGION: {
1846                 struct kvm_userspace_memory_region kvm_userspace_mem;
1847
1848                 r = -EFAULT;
1849                 if (copy_from_user(&kvm_userspace_mem, argp,
1850                                                 sizeof kvm_userspace_mem))
1851                         goto out;
1852
1853                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1854                 if (r)
1855                         goto out;
1856                 break;
1857         }
1858         case KVM_GET_DIRTY_LOG: {
1859                 struct kvm_dirty_log log;
1860
1861                 r = -EFAULT;
1862                 if (copy_from_user(&log, argp, sizeof log))
1863                         goto out;
1864                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1865                 if (r)
1866                         goto out;
1867                 break;
1868         }
1869 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1870         case KVM_REGISTER_COALESCED_MMIO: {
1871                 struct kvm_coalesced_mmio_zone zone;
1872                 r = -EFAULT;
1873                 if (copy_from_user(&zone, argp, sizeof zone))
1874                         goto out;
1875                 r = -ENXIO;
1876                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1877                 if (r)
1878                         goto out;
1879                 r = 0;
1880                 break;
1881         }
1882         case KVM_UNREGISTER_COALESCED_MMIO: {
1883                 struct kvm_coalesced_mmio_zone zone;
1884                 r = -EFAULT;
1885                 if (copy_from_user(&zone, argp, sizeof zone))
1886                         goto out;
1887                 r = -ENXIO;
1888                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1889                 if (r)
1890                         goto out;
1891                 r = 0;
1892                 break;
1893         }
1894 #endif
1895 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1896         case KVM_ASSIGN_PCI_DEVICE: {
1897                 struct kvm_assigned_pci_dev assigned_dev;
1898
1899                 r = -EFAULT;
1900                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1901                         goto out;
1902                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1903                 if (r)
1904                         goto out;
1905                 break;
1906         }
1907         case KVM_ASSIGN_IRQ: {
1908                 struct kvm_assigned_irq assigned_irq;
1909
1910                 r = -EFAULT;
1911                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1912                         goto out;
1913                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1914                 if (r)
1915                         goto out;
1916                 break;
1917         }
1918 #endif
1919 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1920         case KVM_DEASSIGN_PCI_DEVICE: {
1921                 struct kvm_assigned_pci_dev assigned_dev;
1922
1923                 r = -EFAULT;
1924                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1925                         goto out;
1926                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1927                 if (r)
1928                         goto out;
1929                 break;
1930         }
1931 #endif
1932         default:
1933                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1934         }
1935 out:
1936         return r;
1937 }
1938
1939 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1940 {
1941         struct page *page[1];
1942         unsigned long addr;
1943         int npages;
1944         gfn_t gfn = vmf->pgoff;
1945         struct kvm *kvm = vma->vm_file->private_data;
1946
1947         addr = gfn_to_hva(kvm, gfn);
1948         if (kvm_is_error_hva(addr))
1949                 return VM_FAULT_SIGBUS;
1950
1951         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1952                                 NULL);
1953         if (unlikely(npages != 1))
1954                 return VM_FAULT_SIGBUS;
1955
1956         vmf->page = page[0];
1957         return 0;
1958 }
1959
1960 static struct vm_operations_struct kvm_vm_vm_ops = {
1961         .fault = kvm_vm_fault,
1962 };
1963
1964 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1965 {
1966         vma->vm_ops = &kvm_vm_vm_ops;
1967         return 0;
1968 }
1969
1970 static struct file_operations kvm_vm_fops = {
1971         .release        = kvm_vm_release,
1972         .unlocked_ioctl = kvm_vm_ioctl,
1973         .compat_ioctl   = kvm_vm_ioctl,
1974         .mmap           = kvm_vm_mmap,
1975 };
1976
1977 static int kvm_dev_ioctl_create_vm(void)
1978 {
1979         int fd;
1980         struct kvm *kvm;
1981
1982         kvm = kvm_create_vm();
1983         if (IS_ERR(kvm))
1984                 return PTR_ERR(kvm);
1985         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1986         if (fd < 0)
1987                 kvm_put_kvm(kvm);
1988
1989         return fd;
1990 }
1991
1992 static long kvm_dev_ioctl_check_extension_generic(long arg)
1993 {
1994         switch (arg) {
1995         case KVM_CAP_USER_MEMORY:
1996         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1997                 return 1;
1998         default:
1999                 break;
2000         }
2001         return kvm_dev_ioctl_check_extension(arg);
2002 }
2003
2004 static long kvm_dev_ioctl(struct file *filp,
2005                           unsigned int ioctl, unsigned long arg)
2006 {
2007         long r = -EINVAL;
2008
2009         switch (ioctl) {
2010         case KVM_GET_API_VERSION:
2011                 r = -EINVAL;
2012                 if (arg)
2013                         goto out;
2014                 r = KVM_API_VERSION;
2015                 break;
2016         case KVM_CREATE_VM:
2017                 r = -EINVAL;
2018                 if (arg)
2019                         goto out;
2020                 r = kvm_dev_ioctl_create_vm();
2021                 break;
2022         case KVM_CHECK_EXTENSION:
2023                 r = kvm_dev_ioctl_check_extension_generic(arg);
2024                 break;
2025         case KVM_GET_VCPU_MMAP_SIZE:
2026                 r = -EINVAL;
2027                 if (arg)
2028                         goto out;
2029                 r = PAGE_SIZE;     /* struct kvm_run */
2030 #ifdef CONFIG_X86
2031                 r += PAGE_SIZE;    /* pio data page */
2032 #endif
2033 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2034                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2035 #endif
2036                 break;
2037         case KVM_TRACE_ENABLE:
2038         case KVM_TRACE_PAUSE:
2039         case KVM_TRACE_DISABLE:
2040                 r = kvm_trace_ioctl(ioctl, arg);
2041                 break;
2042         default:
2043                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2044         }
2045 out:
2046         return r;
2047 }
2048
2049 static struct file_operations kvm_chardev_ops = {
2050         .unlocked_ioctl = kvm_dev_ioctl,
2051         .compat_ioctl   = kvm_dev_ioctl,
2052 };
2053
2054 static struct miscdevice kvm_dev = {
2055         KVM_MINOR,
2056         "kvm",
2057         &kvm_chardev_ops,
2058 };
2059
2060 static void hardware_enable(void *junk)
2061 {
2062         int cpu = raw_smp_processor_id();
2063
2064         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2065                 return;
2066         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2067         kvm_arch_hardware_enable(NULL);
2068 }
2069
2070 static void hardware_disable(void *junk)
2071 {
2072         int cpu = raw_smp_processor_id();
2073
2074         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2075                 return;
2076         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2077         kvm_arch_hardware_disable(NULL);
2078 }
2079
2080 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2081                            void *v)
2082 {
2083         int cpu = (long)v;
2084
2085         val &= ~CPU_TASKS_FROZEN;
2086         switch (val) {
2087         case CPU_DYING:
2088                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2089                        cpu);
2090                 hardware_disable(NULL);
2091                 break;
2092         case CPU_UP_CANCELED:
2093                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2094                        cpu);
2095                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2096                 break;
2097         case CPU_ONLINE:
2098                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2099                        cpu);
2100                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2101                 break;
2102         }
2103         return NOTIFY_OK;
2104 }
2105
2106
2107 asmlinkage void kvm_handle_fault_on_reboot(void)
2108 {
2109         if (kvm_rebooting)
2110                 /* spin while reset goes on */
2111                 while (true)
2112                         ;
2113         /* Fault while not rebooting.  We want the trace. */
2114         BUG();
2115 }
2116 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2117
2118 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2119                       void *v)
2120 {
2121         if (val == SYS_RESTART) {
2122                 /*
2123                  * Some (well, at least mine) BIOSes hang on reboot if
2124                  * in vmx root mode.
2125                  */
2126                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2127                 kvm_rebooting = true;
2128                 on_each_cpu(hardware_disable, NULL, 1);
2129         }
2130         return NOTIFY_OK;
2131 }
2132
2133 static struct notifier_block kvm_reboot_notifier = {
2134         .notifier_call = kvm_reboot,
2135         .priority = 0,
2136 };
2137
2138 void kvm_io_bus_init(struct kvm_io_bus *bus)
2139 {
2140         memset(bus, 0, sizeof(*bus));
2141 }
2142
2143 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2144 {
2145         int i;
2146
2147         for (i = 0; i < bus->dev_count; i++) {
2148                 struct kvm_io_device *pos = bus->devs[i];
2149
2150                 kvm_iodevice_destructor(pos);
2151         }
2152 }
2153
2154 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2155                                           gpa_t addr, int len, int is_write)
2156 {
2157         int i;
2158
2159         for (i = 0; i < bus->dev_count; i++) {
2160                 struct kvm_io_device *pos = bus->devs[i];
2161
2162                 if (pos->in_range(pos, addr, len, is_write))
2163                         return pos;
2164         }
2165
2166         return NULL;
2167 }
2168
2169 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2170 {
2171         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2172
2173         bus->devs[bus->dev_count++] = dev;
2174 }
2175
2176 static struct notifier_block kvm_cpu_notifier = {
2177         .notifier_call = kvm_cpu_hotplug,
2178         .priority = 20, /* must be > scheduler priority */
2179 };
2180
2181 static int vm_stat_get(void *_offset, u64 *val)
2182 {
2183         unsigned offset = (long)_offset;
2184         struct kvm *kvm;
2185
2186         *val = 0;
2187         spin_lock(&kvm_lock);
2188         list_for_each_entry(kvm, &vm_list, vm_list)
2189                 *val += *(u32 *)((void *)kvm + offset);
2190         spin_unlock(&kvm_lock);
2191         return 0;
2192 }
2193
2194 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2195
2196 static int vcpu_stat_get(void *_offset, u64 *val)
2197 {
2198         unsigned offset = (long)_offset;
2199         struct kvm *kvm;
2200         struct kvm_vcpu *vcpu;
2201         int i;
2202
2203         *val = 0;
2204         spin_lock(&kvm_lock);
2205         list_for_each_entry(kvm, &vm_list, vm_list)
2206                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2207                         vcpu = kvm->vcpus[i];
2208                         if (vcpu)
2209                                 *val += *(u32 *)((void *)vcpu + offset);
2210                 }
2211         spin_unlock(&kvm_lock);
2212         return 0;
2213 }
2214
2215 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2216
2217 static struct file_operations *stat_fops[] = {
2218         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2219         [KVM_STAT_VM]   = &vm_stat_fops,
2220 };
2221
2222 static void kvm_init_debug(void)
2223 {
2224         struct kvm_stats_debugfs_item *p;
2225
2226         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2227         for (p = debugfs_entries; p->name; ++p)
2228                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2229                                                 (void *)(long)p->offset,
2230                                                 stat_fops[p->kind]);
2231 }
2232
2233 static void kvm_exit_debug(void)
2234 {
2235         struct kvm_stats_debugfs_item *p;
2236
2237         for (p = debugfs_entries; p->name; ++p)
2238                 debugfs_remove(p->dentry);
2239         debugfs_remove(kvm_debugfs_dir);
2240 }
2241
2242 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2243 {
2244         hardware_disable(NULL);
2245         return 0;
2246 }
2247
2248 static int kvm_resume(struct sys_device *dev)
2249 {
2250         hardware_enable(NULL);
2251         return 0;
2252 }
2253
2254 static struct sysdev_class kvm_sysdev_class = {
2255         .name = "kvm",
2256         .suspend = kvm_suspend,
2257         .resume = kvm_resume,
2258 };
2259
2260 static struct sys_device kvm_sysdev = {
2261         .id = 0,
2262         .cls = &kvm_sysdev_class,
2263 };
2264
2265 struct page *bad_page;
2266 pfn_t bad_pfn;
2267
2268 static inline
2269 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2270 {
2271         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2272 }
2273
2274 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2275 {
2276         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2277
2278         kvm_arch_vcpu_load(vcpu, cpu);
2279 }
2280
2281 static void kvm_sched_out(struct preempt_notifier *pn,
2282                           struct task_struct *next)
2283 {
2284         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2285
2286         kvm_arch_vcpu_put(vcpu);
2287 }
2288
2289 int kvm_init(void *opaque, unsigned int vcpu_size,
2290                   struct module *module)
2291 {
2292         int r;
2293         int cpu;
2294
2295         kvm_init_debug();
2296
2297         r = kvm_arch_init(opaque);
2298         if (r)
2299                 goto out_fail;
2300
2301         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2302
2303         if (bad_page == NULL) {
2304                 r = -ENOMEM;
2305                 goto out;
2306         }
2307
2308         bad_pfn = page_to_pfn(bad_page);
2309
2310         if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2311                 r = -ENOMEM;
2312                 goto out_free_0;
2313         }
2314
2315         r = kvm_arch_hardware_setup();
2316         if (r < 0)
2317                 goto out_free_0a;
2318
2319         for_each_online_cpu(cpu) {
2320                 smp_call_function_single(cpu,
2321                                 kvm_arch_check_processor_compat,
2322                                 &r, 1);
2323                 if (r < 0)
2324                         goto out_free_1;
2325         }
2326
2327         on_each_cpu(hardware_enable, NULL, 1);
2328         r = register_cpu_notifier(&kvm_cpu_notifier);
2329         if (r)
2330                 goto out_free_2;
2331         register_reboot_notifier(&kvm_reboot_notifier);
2332
2333         r = sysdev_class_register(&kvm_sysdev_class);
2334         if (r)
2335                 goto out_free_3;
2336
2337         r = sysdev_register(&kvm_sysdev);
2338         if (r)
2339                 goto out_free_4;
2340
2341         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2342         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2343                                            __alignof__(struct kvm_vcpu),
2344                                            0, NULL);
2345         if (!kvm_vcpu_cache) {
2346                 r = -ENOMEM;
2347                 goto out_free_5;
2348         }
2349
2350         kvm_chardev_ops.owner = module;
2351         kvm_vm_fops.owner = module;
2352         kvm_vcpu_fops.owner = module;
2353
2354         r = misc_register(&kvm_dev);
2355         if (r) {
2356                 printk(KERN_ERR "kvm: misc device register failed\n");
2357                 goto out_free;
2358         }
2359
2360         kvm_preempt_ops.sched_in = kvm_sched_in;
2361         kvm_preempt_ops.sched_out = kvm_sched_out;
2362 #ifndef CONFIG_X86
2363         msi2intx = 0;
2364 #endif
2365
2366         return 0;
2367
2368 out_free:
2369         kmem_cache_destroy(kvm_vcpu_cache);
2370 out_free_5:
2371         sysdev_unregister(&kvm_sysdev);
2372 out_free_4:
2373         sysdev_class_unregister(&kvm_sysdev_class);
2374 out_free_3:
2375         unregister_reboot_notifier(&kvm_reboot_notifier);
2376         unregister_cpu_notifier(&kvm_cpu_notifier);
2377 out_free_2:
2378         on_each_cpu(hardware_disable, NULL, 1);
2379 out_free_1:
2380         kvm_arch_hardware_unsetup();
2381 out_free_0a:
2382         free_cpumask_var(cpus_hardware_enabled);
2383 out_free_0:
2384         __free_page(bad_page);
2385 out:
2386         kvm_arch_exit();
2387         kvm_exit_debug();
2388 out_fail:
2389         return r;
2390 }
2391 EXPORT_SYMBOL_GPL(kvm_init);
2392
2393 void kvm_exit(void)
2394 {
2395         kvm_trace_cleanup();
2396         misc_deregister(&kvm_dev);
2397         kmem_cache_destroy(kvm_vcpu_cache);
2398         sysdev_unregister(&kvm_sysdev);
2399         sysdev_class_unregister(&kvm_sysdev_class);
2400         unregister_reboot_notifier(&kvm_reboot_notifier);
2401         unregister_cpu_notifier(&kvm_cpu_notifier);
2402         on_each_cpu(hardware_disable, NULL, 1);
2403         kvm_arch_hardware_unsetup();
2404         kvm_arch_exit();
2405         kvm_exit_debug();
2406         free_cpumask_var(cpus_hardware_enabled);
2407         __free_page(bad_page);
2408 }
2409 EXPORT_SYMBOL_GPL(kvm_exit);