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