KVM: Fix racy in kvm_free_assigned_irq
[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 *)kvm);
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         mutex_lock(&kvm->lock);
470
471         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
472                                       assigned_dev->assigned_dev_id);
473         if (match) {
474                 /* device already assigned */
475                 r = -EINVAL;
476                 goto out;
477         }
478
479         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
480         if (match == NULL) {
481                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
482                        __func__);
483                 r = -ENOMEM;
484                 goto out;
485         }
486         dev = pci_get_bus_and_slot(assigned_dev->busnr,
487                                    assigned_dev->devfn);
488         if (!dev) {
489                 printk(KERN_INFO "%s: host device not found\n", __func__);
490                 r = -EINVAL;
491                 goto out_free;
492         }
493         if (pci_enable_device(dev)) {
494                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
495                 r = -EBUSY;
496                 goto out_put;
497         }
498         r = pci_request_regions(dev, "kvm_assigned_device");
499         if (r) {
500                 printk(KERN_INFO "%s: Could not get access to device regions\n",
501                        __func__);
502                 goto out_disable;
503         }
504
505         pci_reset_function(dev);
506
507         match->assigned_dev_id = assigned_dev->assigned_dev_id;
508         match->host_busnr = assigned_dev->busnr;
509         match->host_devfn = assigned_dev->devfn;
510         match->flags = assigned_dev->flags;
511         match->dev = dev;
512         match->irq_source_id = -1;
513         match->kvm = kvm;
514
515         list_add(&match->list, &kvm->arch.assigned_dev_head);
516
517         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
518                 if (!kvm->arch.iommu_domain) {
519                         r = kvm_iommu_map_guest(kvm);
520                         if (r)
521                                 goto out_list_del;
522                 }
523                 r = kvm_assign_device(kvm, match);
524                 if (r)
525                         goto out_list_del;
526         }
527
528 out:
529         mutex_unlock(&kvm->lock);
530         return r;
531 out_list_del:
532         list_del(&match->list);
533         pci_release_regions(dev);
534 out_disable:
535         pci_disable_device(dev);
536 out_put:
537         pci_dev_put(dev);
538 out_free:
539         kfree(match);
540         mutex_unlock(&kvm->lock);
541         return r;
542 }
543 #endif
544
545 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
546 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
547                 struct kvm_assigned_pci_dev *assigned_dev)
548 {
549         int r = 0;
550         struct kvm_assigned_dev_kernel *match;
551
552         mutex_lock(&kvm->lock);
553
554         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
555                                       assigned_dev->assigned_dev_id);
556         if (!match) {
557                 printk(KERN_INFO "%s: device hasn't been assigned before, "
558                   "so cannot be deassigned\n", __func__);
559                 r = -EINVAL;
560                 goto out;
561         }
562
563         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
564                 kvm_deassign_device(kvm, match);
565
566         kvm_free_assigned_device(kvm, match);
567
568 out:
569         mutex_unlock(&kvm->lock);
570         return r;
571 }
572 #endif
573
574 static inline int valid_vcpu(int n)
575 {
576         return likely(n >= 0 && n < KVM_MAX_VCPUS);
577 }
578
579 inline int kvm_is_mmio_pfn(pfn_t pfn)
580 {
581         if (pfn_valid(pfn))
582                 return PageReserved(pfn_to_page(pfn));
583
584         return true;
585 }
586
587 /*
588  * Switches to specified vcpu, until a matching vcpu_put()
589  */
590 void vcpu_load(struct kvm_vcpu *vcpu)
591 {
592         int cpu;
593
594         mutex_lock(&vcpu->mutex);
595         cpu = get_cpu();
596         preempt_notifier_register(&vcpu->preempt_notifier);
597         kvm_arch_vcpu_load(vcpu, cpu);
598         put_cpu();
599 }
600
601 void vcpu_put(struct kvm_vcpu *vcpu)
602 {
603         preempt_disable();
604         kvm_arch_vcpu_put(vcpu);
605         preempt_notifier_unregister(&vcpu->preempt_notifier);
606         preempt_enable();
607         mutex_unlock(&vcpu->mutex);
608 }
609
610 static void ack_flush(void *_completed)
611 {
612 }
613
614 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
615 {
616         int i, cpu, me;
617         cpumask_var_t cpus;
618         bool called = true;
619         struct kvm_vcpu *vcpu;
620
621         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
622                 cpumask_clear(cpus);
623
624         me = get_cpu();
625         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
626                 vcpu = kvm->vcpus[i];
627                 if (!vcpu)
628                         continue;
629                 if (test_and_set_bit(req, &vcpu->requests))
630                         continue;
631                 cpu = vcpu->cpu;
632                 if (cpus != NULL && cpu != -1 && cpu != me)
633                         cpumask_set_cpu(cpu, cpus);
634         }
635         if (unlikely(cpus == NULL))
636                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
637         else if (!cpumask_empty(cpus))
638                 smp_call_function_many(cpus, ack_flush, NULL, 1);
639         else
640                 called = false;
641         put_cpu();
642         free_cpumask_var(cpus);
643         return called;
644 }
645
646 void kvm_flush_remote_tlbs(struct kvm *kvm)
647 {
648         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
649                 ++kvm->stat.remote_tlb_flush;
650 }
651
652 void kvm_reload_remote_mmus(struct kvm *kvm)
653 {
654         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
655 }
656
657 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
658 {
659         struct page *page;
660         int r;
661
662         mutex_init(&vcpu->mutex);
663         vcpu->cpu = -1;
664         vcpu->kvm = kvm;
665         vcpu->vcpu_id = id;
666         init_waitqueue_head(&vcpu->wq);
667
668         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
669         if (!page) {
670                 r = -ENOMEM;
671                 goto fail;
672         }
673         vcpu->run = page_address(page);
674
675         r = kvm_arch_vcpu_init(vcpu);
676         if (r < 0)
677                 goto fail_free_run;
678         return 0;
679
680 fail_free_run:
681         free_page((unsigned long)vcpu->run);
682 fail:
683         return r;
684 }
685 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
686
687 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
688 {
689         kvm_arch_vcpu_uninit(vcpu);
690         free_page((unsigned long)vcpu->run);
691 }
692 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
693
694 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
695 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
696 {
697         return container_of(mn, struct kvm, mmu_notifier);
698 }
699
700 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
701                                              struct mm_struct *mm,
702                                              unsigned long address)
703 {
704         struct kvm *kvm = mmu_notifier_to_kvm(mn);
705         int need_tlb_flush;
706
707         /*
708          * When ->invalidate_page runs, the linux pte has been zapped
709          * already but the page is still allocated until
710          * ->invalidate_page returns. So if we increase the sequence
711          * here the kvm page fault will notice if the spte can't be
712          * established because the page is going to be freed. If
713          * instead the kvm page fault establishes the spte before
714          * ->invalidate_page runs, kvm_unmap_hva will release it
715          * before returning.
716          *
717          * The sequence increase only need to be seen at spin_unlock
718          * time, and not at spin_lock time.
719          *
720          * Increasing the sequence after the spin_unlock would be
721          * unsafe because the kvm page fault could then establish the
722          * pte after kvm_unmap_hva returned, without noticing the page
723          * is going to be freed.
724          */
725         spin_lock(&kvm->mmu_lock);
726         kvm->mmu_notifier_seq++;
727         need_tlb_flush = kvm_unmap_hva(kvm, address);
728         spin_unlock(&kvm->mmu_lock);
729
730         /* we've to flush the tlb before the pages can be freed */
731         if (need_tlb_flush)
732                 kvm_flush_remote_tlbs(kvm);
733
734 }
735
736 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
737                                                     struct mm_struct *mm,
738                                                     unsigned long start,
739                                                     unsigned long end)
740 {
741         struct kvm *kvm = mmu_notifier_to_kvm(mn);
742         int need_tlb_flush = 0;
743
744         spin_lock(&kvm->mmu_lock);
745         /*
746          * The count increase must become visible at unlock time as no
747          * spte can be established without taking the mmu_lock and
748          * count is also read inside the mmu_lock critical section.
749          */
750         kvm->mmu_notifier_count++;
751         for (; start < end; start += PAGE_SIZE)
752                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
753         spin_unlock(&kvm->mmu_lock);
754
755         /* we've to flush the tlb before the pages can be freed */
756         if (need_tlb_flush)
757                 kvm_flush_remote_tlbs(kvm);
758 }
759
760 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
761                                                   struct mm_struct *mm,
762                                                   unsigned long start,
763                                                   unsigned long end)
764 {
765         struct kvm *kvm = mmu_notifier_to_kvm(mn);
766
767         spin_lock(&kvm->mmu_lock);
768         /*
769          * This sequence increase will notify the kvm page fault that
770          * the page that is going to be mapped in the spte could have
771          * been freed.
772          */
773         kvm->mmu_notifier_seq++;
774         /*
775          * The above sequence increase must be visible before the
776          * below count decrease but both values are read by the kvm
777          * page fault under mmu_lock spinlock so we don't need to add
778          * a smb_wmb() here in between the two.
779          */
780         kvm->mmu_notifier_count--;
781         spin_unlock(&kvm->mmu_lock);
782
783         BUG_ON(kvm->mmu_notifier_count < 0);
784 }
785
786 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
787                                               struct mm_struct *mm,
788                                               unsigned long address)
789 {
790         struct kvm *kvm = mmu_notifier_to_kvm(mn);
791         int young;
792
793         spin_lock(&kvm->mmu_lock);
794         young = kvm_age_hva(kvm, address);
795         spin_unlock(&kvm->mmu_lock);
796
797         if (young)
798                 kvm_flush_remote_tlbs(kvm);
799
800         return young;
801 }
802
803 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
804                                      struct mm_struct *mm)
805 {
806         struct kvm *kvm = mmu_notifier_to_kvm(mn);
807         kvm_arch_flush_shadow(kvm);
808 }
809
810 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
811         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
812         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
813         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
814         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
815         .release                = kvm_mmu_notifier_release,
816 };
817 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
818
819 static struct kvm *kvm_create_vm(void)
820 {
821         struct kvm *kvm = kvm_arch_create_vm();
822 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
823         struct page *page;
824 #endif
825
826         if (IS_ERR(kvm))
827                 goto out;
828
829 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
830         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
831         if (!page) {
832                 kfree(kvm);
833                 return ERR_PTR(-ENOMEM);
834         }
835         kvm->coalesced_mmio_ring =
836                         (struct kvm_coalesced_mmio_ring *)page_address(page);
837 #endif
838
839 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
840         {
841                 int err;
842                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
843                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
844                 if (err) {
845 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
846                         put_page(page);
847 #endif
848                         kfree(kvm);
849                         return ERR_PTR(err);
850                 }
851         }
852 #endif
853
854         kvm->mm = current->mm;
855         atomic_inc(&kvm->mm->mm_count);
856         spin_lock_init(&kvm->mmu_lock);
857         kvm_io_bus_init(&kvm->pio_bus);
858         mutex_init(&kvm->lock);
859         kvm_io_bus_init(&kvm->mmio_bus);
860         init_rwsem(&kvm->slots_lock);
861         atomic_set(&kvm->users_count, 1);
862         spin_lock(&kvm_lock);
863         list_add(&kvm->vm_list, &vm_list);
864         spin_unlock(&kvm_lock);
865 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
866         kvm_coalesced_mmio_init(kvm);
867 #endif
868 out:
869         return kvm;
870 }
871
872 /*
873  * Free any memory in @free but not in @dont.
874  */
875 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
876                                   struct kvm_memory_slot *dont)
877 {
878         if (!dont || free->rmap != dont->rmap)
879                 vfree(free->rmap);
880
881         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
882                 vfree(free->dirty_bitmap);
883
884         if (!dont || free->lpage_info != dont->lpage_info)
885                 vfree(free->lpage_info);
886
887         free->npages = 0;
888         free->dirty_bitmap = NULL;
889         free->rmap = NULL;
890         free->lpage_info = NULL;
891 }
892
893 void kvm_free_physmem(struct kvm *kvm)
894 {
895         int i;
896
897         for (i = 0; i < kvm->nmemslots; ++i)
898                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
899 }
900
901 static void kvm_destroy_vm(struct kvm *kvm)
902 {
903         struct mm_struct *mm = kvm->mm;
904
905         kvm_arch_sync_events(kvm);
906         spin_lock(&kvm_lock);
907         list_del(&kvm->vm_list);
908         spin_unlock(&kvm_lock);
909         kvm_io_bus_destroy(&kvm->pio_bus);
910         kvm_io_bus_destroy(&kvm->mmio_bus);
911 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
912         if (kvm->coalesced_mmio_ring != NULL)
913                 free_page((unsigned long)kvm->coalesced_mmio_ring);
914 #endif
915 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
916         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
917 #endif
918         kvm_arch_destroy_vm(kvm);
919         mmdrop(mm);
920 }
921
922 void kvm_get_kvm(struct kvm *kvm)
923 {
924         atomic_inc(&kvm->users_count);
925 }
926 EXPORT_SYMBOL_GPL(kvm_get_kvm);
927
928 void kvm_put_kvm(struct kvm *kvm)
929 {
930         if (atomic_dec_and_test(&kvm->users_count))
931                 kvm_destroy_vm(kvm);
932 }
933 EXPORT_SYMBOL_GPL(kvm_put_kvm);
934
935
936 static int kvm_vm_release(struct inode *inode, struct file *filp)
937 {
938         struct kvm *kvm = filp->private_data;
939
940         kvm_put_kvm(kvm);
941         return 0;
942 }
943
944 /*
945  * Allocate some memory and give it an address in the guest physical address
946  * space.
947  *
948  * Discontiguous memory is allowed, mostly for framebuffers.
949  *
950  * Must be called holding mmap_sem for write.
951  */
952 int __kvm_set_memory_region(struct kvm *kvm,
953                             struct kvm_userspace_memory_region *mem,
954                             int user_alloc)
955 {
956         int r;
957         gfn_t base_gfn;
958         unsigned long npages;
959         unsigned long i;
960         struct kvm_memory_slot *memslot;
961         struct kvm_memory_slot old, new;
962
963         r = -EINVAL;
964         /* General sanity checks */
965         if (mem->memory_size & (PAGE_SIZE - 1))
966                 goto out;
967         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
968                 goto out;
969         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
970                 goto out;
971         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
972                 goto out;
973         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
974                 goto out;
975
976         memslot = &kvm->memslots[mem->slot];
977         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
978         npages = mem->memory_size >> PAGE_SHIFT;
979
980         if (!npages)
981                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
982
983         new = old = *memslot;
984
985         new.base_gfn = base_gfn;
986         new.npages = npages;
987         new.flags = mem->flags;
988
989         /* Disallow changing a memory slot's size. */
990         r = -EINVAL;
991         if (npages && old.npages && npages != old.npages)
992                 goto out_free;
993
994         /* Check for overlaps */
995         r = -EEXIST;
996         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
997                 struct kvm_memory_slot *s = &kvm->memslots[i];
998
999                 if (s == memslot)
1000                         continue;
1001                 if (!((base_gfn + npages <= s->base_gfn) ||
1002                       (base_gfn >= s->base_gfn + s->npages)))
1003                         goto out_free;
1004         }
1005
1006         /* Free page dirty bitmap if unneeded */
1007         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1008                 new.dirty_bitmap = NULL;
1009
1010         r = -ENOMEM;
1011
1012         /* Allocate if a slot is being created */
1013 #ifndef CONFIG_S390
1014         if (npages && !new.rmap) {
1015                 new.rmap = vmalloc(npages * sizeof(struct page *));
1016
1017                 if (!new.rmap)
1018                         goto out_free;
1019
1020                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1021
1022                 new.user_alloc = user_alloc;
1023                 /*
1024                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1025                  * safe it has to ignore memslots with !user_alloc &&
1026                  * !userspace_addr.
1027                  */
1028                 if (user_alloc)
1029                         new.userspace_addr = mem->userspace_addr;
1030                 else
1031                         new.userspace_addr = 0;
1032         }
1033         if (npages && !new.lpage_info) {
1034                 int largepages = npages / KVM_PAGES_PER_HPAGE;
1035                 if (npages % KVM_PAGES_PER_HPAGE)
1036                         largepages++;
1037                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1038                         largepages++;
1039
1040                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1041
1042                 if (!new.lpage_info)
1043                         goto out_free;
1044
1045                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1046
1047                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1048                         new.lpage_info[0].write_count = 1;
1049                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1050                         new.lpage_info[largepages-1].write_count = 1;
1051         }
1052
1053         /* Allocate page dirty bitmap if needed */
1054         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1055                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1056
1057                 new.dirty_bitmap = vmalloc(dirty_bytes);
1058                 if (!new.dirty_bitmap)
1059                         goto out_free;
1060                 memset(new.dirty_bitmap, 0, dirty_bytes);
1061         }
1062 #endif /* not defined CONFIG_S390 */
1063
1064         if (!npages)
1065                 kvm_arch_flush_shadow(kvm);
1066
1067         spin_lock(&kvm->mmu_lock);
1068         if (mem->slot >= kvm->nmemslots)
1069                 kvm->nmemslots = mem->slot + 1;
1070
1071         *memslot = new;
1072         spin_unlock(&kvm->mmu_lock);
1073
1074         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1075         if (r) {
1076                 spin_lock(&kvm->mmu_lock);
1077                 *memslot = old;
1078                 spin_unlock(&kvm->mmu_lock);
1079                 goto out_free;
1080         }
1081
1082         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1083         /* Slot deletion case: we have to update the current slot */
1084         if (!npages)
1085                 *memslot = old;
1086 #ifdef CONFIG_DMAR
1087         /* map the pages in iommu page table */
1088         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1089         if (r)
1090                 goto out;
1091 #endif
1092         return 0;
1093
1094 out_free:
1095         kvm_free_physmem_slot(&new, &old);
1096 out:
1097         return r;
1098
1099 }
1100 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1101
1102 int kvm_set_memory_region(struct kvm *kvm,
1103                           struct kvm_userspace_memory_region *mem,
1104                           int user_alloc)
1105 {
1106         int r;
1107
1108         down_write(&kvm->slots_lock);
1109         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1110         up_write(&kvm->slots_lock);
1111         return r;
1112 }
1113 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1114
1115 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1116                                    struct
1117                                    kvm_userspace_memory_region *mem,
1118                                    int user_alloc)
1119 {
1120         if (mem->slot >= KVM_MEMORY_SLOTS)
1121                 return -EINVAL;
1122         return kvm_set_memory_region(kvm, mem, user_alloc);
1123 }
1124
1125 int kvm_get_dirty_log(struct kvm *kvm,
1126                         struct kvm_dirty_log *log, int *is_dirty)
1127 {
1128         struct kvm_memory_slot *memslot;
1129         int r, i;
1130         int n;
1131         unsigned long any = 0;
1132
1133         r = -EINVAL;
1134         if (log->slot >= KVM_MEMORY_SLOTS)
1135                 goto out;
1136
1137         memslot = &kvm->memslots[log->slot];
1138         r = -ENOENT;
1139         if (!memslot->dirty_bitmap)
1140                 goto out;
1141
1142         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1143
1144         for (i = 0; !any && i < n/sizeof(long); ++i)
1145                 any = memslot->dirty_bitmap[i];
1146
1147         r = -EFAULT;
1148         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1149                 goto out;
1150
1151         if (any)
1152                 *is_dirty = 1;
1153
1154         r = 0;
1155 out:
1156         return r;
1157 }
1158
1159 int is_error_page(struct page *page)
1160 {
1161         return page == bad_page;
1162 }
1163 EXPORT_SYMBOL_GPL(is_error_page);
1164
1165 int is_error_pfn(pfn_t pfn)
1166 {
1167         return pfn == bad_pfn;
1168 }
1169 EXPORT_SYMBOL_GPL(is_error_pfn);
1170
1171 static inline unsigned long bad_hva(void)
1172 {
1173         return PAGE_OFFSET;
1174 }
1175
1176 int kvm_is_error_hva(unsigned long addr)
1177 {
1178         return addr == bad_hva();
1179 }
1180 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1181
1182 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1183 {
1184         int i;
1185
1186         for (i = 0; i < kvm->nmemslots; ++i) {
1187                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1188
1189                 if (gfn >= memslot->base_gfn
1190                     && gfn < memslot->base_gfn + memslot->npages)
1191                         return memslot;
1192         }
1193         return NULL;
1194 }
1195 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1196
1197 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1198 {
1199         gfn = unalias_gfn(kvm, gfn);
1200         return gfn_to_memslot_unaliased(kvm, gfn);
1201 }
1202
1203 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1204 {
1205         int i;
1206
1207         gfn = unalias_gfn(kvm, gfn);
1208         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1209                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1210
1211                 if (gfn >= memslot->base_gfn
1212                     && gfn < memslot->base_gfn + memslot->npages)
1213                         return 1;
1214         }
1215         return 0;
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1218
1219 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1220 {
1221         struct kvm_memory_slot *slot;
1222
1223         gfn = unalias_gfn(kvm, gfn);
1224         slot = gfn_to_memslot_unaliased(kvm, gfn);
1225         if (!slot)
1226                 return bad_hva();
1227         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1228 }
1229 EXPORT_SYMBOL_GPL(gfn_to_hva);
1230
1231 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1232 {
1233         struct page *page[1];
1234         unsigned long addr;
1235         int npages;
1236         pfn_t pfn;
1237
1238         might_sleep();
1239
1240         addr = gfn_to_hva(kvm, gfn);
1241         if (kvm_is_error_hva(addr)) {
1242                 get_page(bad_page);
1243                 return page_to_pfn(bad_page);
1244         }
1245
1246         npages = get_user_pages_fast(addr, 1, 1, page);
1247
1248         if (unlikely(npages != 1)) {
1249                 struct vm_area_struct *vma;
1250
1251                 down_read(&current->mm->mmap_sem);
1252                 vma = find_vma(current->mm, addr);
1253
1254                 if (vma == NULL || addr < vma->vm_start ||
1255                     !(vma->vm_flags & VM_PFNMAP)) {
1256                         up_read(&current->mm->mmap_sem);
1257                         get_page(bad_page);
1258                         return page_to_pfn(bad_page);
1259                 }
1260
1261                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1262                 up_read(&current->mm->mmap_sem);
1263                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1264         } else
1265                 pfn = page_to_pfn(page[0]);
1266
1267         return pfn;
1268 }
1269
1270 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1271
1272 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1273 {
1274         pfn_t pfn;
1275
1276         pfn = gfn_to_pfn(kvm, gfn);
1277         if (!kvm_is_mmio_pfn(pfn))
1278                 return pfn_to_page(pfn);
1279
1280         WARN_ON(kvm_is_mmio_pfn(pfn));
1281
1282         get_page(bad_page);
1283         return bad_page;
1284 }
1285
1286 EXPORT_SYMBOL_GPL(gfn_to_page);
1287
1288 void kvm_release_page_clean(struct page *page)
1289 {
1290         kvm_release_pfn_clean(page_to_pfn(page));
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1293
1294 void kvm_release_pfn_clean(pfn_t pfn)
1295 {
1296         if (!kvm_is_mmio_pfn(pfn))
1297                 put_page(pfn_to_page(pfn));
1298 }
1299 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1300
1301 void kvm_release_page_dirty(struct page *page)
1302 {
1303         kvm_release_pfn_dirty(page_to_pfn(page));
1304 }
1305 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1306
1307 void kvm_release_pfn_dirty(pfn_t pfn)
1308 {
1309         kvm_set_pfn_dirty(pfn);
1310         kvm_release_pfn_clean(pfn);
1311 }
1312 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1313
1314 void kvm_set_page_dirty(struct page *page)
1315 {
1316         kvm_set_pfn_dirty(page_to_pfn(page));
1317 }
1318 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1319
1320 void kvm_set_pfn_dirty(pfn_t pfn)
1321 {
1322         if (!kvm_is_mmio_pfn(pfn)) {
1323                 struct page *page = pfn_to_page(pfn);
1324                 if (!PageReserved(page))
1325                         SetPageDirty(page);
1326         }
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1329
1330 void kvm_set_pfn_accessed(pfn_t pfn)
1331 {
1332         if (!kvm_is_mmio_pfn(pfn))
1333                 mark_page_accessed(pfn_to_page(pfn));
1334 }
1335 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1336
1337 void kvm_get_pfn(pfn_t pfn)
1338 {
1339         if (!kvm_is_mmio_pfn(pfn))
1340                 get_page(pfn_to_page(pfn));
1341 }
1342 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1343
1344 static int next_segment(unsigned long len, int offset)
1345 {
1346         if (len > PAGE_SIZE - offset)
1347                 return PAGE_SIZE - offset;
1348         else
1349                 return len;
1350 }
1351
1352 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1353                         int len)
1354 {
1355         int r;
1356         unsigned long addr;
1357
1358         addr = gfn_to_hva(kvm, gfn);
1359         if (kvm_is_error_hva(addr))
1360                 return -EFAULT;
1361         r = copy_from_user(data, (void __user *)addr + offset, len);
1362         if (r)
1363                 return -EFAULT;
1364         return 0;
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1367
1368 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1369 {
1370         gfn_t gfn = gpa >> PAGE_SHIFT;
1371         int seg;
1372         int offset = offset_in_page(gpa);
1373         int ret;
1374
1375         while ((seg = next_segment(len, offset)) != 0) {
1376                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1377                 if (ret < 0)
1378                         return ret;
1379                 offset = 0;
1380                 len -= seg;
1381                 data += seg;
1382                 ++gfn;
1383         }
1384         return 0;
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_read_guest);
1387
1388 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1389                           unsigned long len)
1390 {
1391         int r;
1392         unsigned long addr;
1393         gfn_t gfn = gpa >> PAGE_SHIFT;
1394         int offset = offset_in_page(gpa);
1395
1396         addr = gfn_to_hva(kvm, gfn);
1397         if (kvm_is_error_hva(addr))
1398                 return -EFAULT;
1399         pagefault_disable();
1400         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1401         pagefault_enable();
1402         if (r)
1403                 return -EFAULT;
1404         return 0;
1405 }
1406 EXPORT_SYMBOL(kvm_read_guest_atomic);
1407
1408 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1409                          int offset, int len)
1410 {
1411         int r;
1412         unsigned long addr;
1413
1414         addr = gfn_to_hva(kvm, gfn);
1415         if (kvm_is_error_hva(addr))
1416                 return -EFAULT;
1417         r = copy_to_user((void __user *)addr + offset, data, len);
1418         if (r)
1419                 return -EFAULT;
1420         mark_page_dirty(kvm, gfn);
1421         return 0;
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1424
1425 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1426                     unsigned long len)
1427 {
1428         gfn_t gfn = gpa >> PAGE_SHIFT;
1429         int seg;
1430         int offset = offset_in_page(gpa);
1431         int ret;
1432
1433         while ((seg = next_segment(len, offset)) != 0) {
1434                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1435                 if (ret < 0)
1436                         return ret;
1437                 offset = 0;
1438                 len -= seg;
1439                 data += seg;
1440                 ++gfn;
1441         }
1442         return 0;
1443 }
1444
1445 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1446 {
1447         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1450
1451 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1452 {
1453         gfn_t gfn = gpa >> PAGE_SHIFT;
1454         int seg;
1455         int offset = offset_in_page(gpa);
1456         int ret;
1457
1458         while ((seg = next_segment(len, offset)) != 0) {
1459                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1460                 if (ret < 0)
1461                         return ret;
1462                 offset = 0;
1463                 len -= seg;
1464                 ++gfn;
1465         }
1466         return 0;
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1469
1470 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1471 {
1472         struct kvm_memory_slot *memslot;
1473
1474         gfn = unalias_gfn(kvm, gfn);
1475         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1476         if (memslot && memslot->dirty_bitmap) {
1477                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1478
1479                 /* avoid RMW */
1480                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1481                         set_bit(rel_gfn, memslot->dirty_bitmap);
1482         }
1483 }
1484
1485 /*
1486  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1487  */
1488 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1489 {
1490         DEFINE_WAIT(wait);
1491
1492         for (;;) {
1493                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1494
1495                 if (kvm_cpu_has_interrupt(vcpu) ||
1496                     kvm_cpu_has_pending_timer(vcpu) ||
1497                     kvm_arch_vcpu_runnable(vcpu)) {
1498                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1499                         break;
1500                 }
1501                 if (signal_pending(current))
1502                         break;
1503
1504                 vcpu_put(vcpu);
1505                 schedule();
1506                 vcpu_load(vcpu);
1507         }
1508
1509         finish_wait(&vcpu->wq, &wait);
1510 }
1511
1512 void kvm_resched(struct kvm_vcpu *vcpu)
1513 {
1514         if (!need_resched())
1515                 return;
1516         cond_resched();
1517 }
1518 EXPORT_SYMBOL_GPL(kvm_resched);
1519
1520 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1521 {
1522         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1523         struct page *page;
1524
1525         if (vmf->pgoff == 0)
1526                 page = virt_to_page(vcpu->run);
1527 #ifdef CONFIG_X86
1528         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1529                 page = virt_to_page(vcpu->arch.pio_data);
1530 #endif
1531 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1532         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1533                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1534 #endif
1535         else
1536                 return VM_FAULT_SIGBUS;
1537         get_page(page);
1538         vmf->page = page;
1539         return 0;
1540 }
1541
1542 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1543         .fault = kvm_vcpu_fault,
1544 };
1545
1546 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1547 {
1548         vma->vm_ops = &kvm_vcpu_vm_ops;
1549         return 0;
1550 }
1551
1552 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1553 {
1554         struct kvm_vcpu *vcpu = filp->private_data;
1555
1556         kvm_put_kvm(vcpu->kvm);
1557         return 0;
1558 }
1559
1560 static struct file_operations kvm_vcpu_fops = {
1561         .release        = kvm_vcpu_release,
1562         .unlocked_ioctl = kvm_vcpu_ioctl,
1563         .compat_ioctl   = kvm_vcpu_ioctl,
1564         .mmap           = kvm_vcpu_mmap,
1565 };
1566
1567 /*
1568  * Allocates an inode for the vcpu.
1569  */
1570 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1571 {
1572         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1573         if (fd < 0)
1574                 kvm_put_kvm(vcpu->kvm);
1575         return fd;
1576 }
1577
1578 /*
1579  * Creates some virtual cpus.  Good luck creating more than one.
1580  */
1581 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1582 {
1583         int r;
1584         struct kvm_vcpu *vcpu;
1585
1586         if (!valid_vcpu(n))
1587                 return -EINVAL;
1588
1589         vcpu = kvm_arch_vcpu_create(kvm, n);
1590         if (IS_ERR(vcpu))
1591                 return PTR_ERR(vcpu);
1592
1593         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1594
1595         r = kvm_arch_vcpu_setup(vcpu);
1596         if (r)
1597                 return r;
1598
1599         mutex_lock(&kvm->lock);
1600         if (kvm->vcpus[n]) {
1601                 r = -EEXIST;
1602                 goto vcpu_destroy;
1603         }
1604         kvm->vcpus[n] = vcpu;
1605         mutex_unlock(&kvm->lock);
1606
1607         /* Now it's all set up, let userspace reach it */
1608         kvm_get_kvm(kvm);
1609         r = create_vcpu_fd(vcpu);
1610         if (r < 0)
1611                 goto unlink;
1612         return r;
1613
1614 unlink:
1615         mutex_lock(&kvm->lock);
1616         kvm->vcpus[n] = NULL;
1617 vcpu_destroy:
1618         mutex_unlock(&kvm->lock);
1619         kvm_arch_vcpu_destroy(vcpu);
1620         return r;
1621 }
1622
1623 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1624 {
1625         if (sigset) {
1626                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1627                 vcpu->sigset_active = 1;
1628                 vcpu->sigset = *sigset;
1629         } else
1630                 vcpu->sigset_active = 0;
1631         return 0;
1632 }
1633
1634 static long kvm_vcpu_ioctl(struct file *filp,
1635                            unsigned int ioctl, unsigned long arg)
1636 {
1637         struct kvm_vcpu *vcpu = filp->private_data;
1638         void __user *argp = (void __user *)arg;
1639         int r;
1640         struct kvm_fpu *fpu = NULL;
1641         struct kvm_sregs *kvm_sregs = NULL;
1642
1643         if (vcpu->kvm->mm != current->mm)
1644                 return -EIO;
1645         switch (ioctl) {
1646         case KVM_RUN:
1647                 r = -EINVAL;
1648                 if (arg)
1649                         goto out;
1650                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1651                 break;
1652         case KVM_GET_REGS: {
1653                 struct kvm_regs *kvm_regs;
1654
1655                 r = -ENOMEM;
1656                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1657                 if (!kvm_regs)
1658                         goto out;
1659                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1660                 if (r)
1661                         goto out_free1;
1662                 r = -EFAULT;
1663                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1664                         goto out_free1;
1665                 r = 0;
1666 out_free1:
1667                 kfree(kvm_regs);
1668                 break;
1669         }
1670         case KVM_SET_REGS: {
1671                 struct kvm_regs *kvm_regs;
1672
1673                 r = -ENOMEM;
1674                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1675                 if (!kvm_regs)
1676                         goto out;
1677                 r = -EFAULT;
1678                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1679                         goto out_free2;
1680                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1681                 if (r)
1682                         goto out_free2;
1683                 r = 0;
1684 out_free2:
1685                 kfree(kvm_regs);
1686                 break;
1687         }
1688         case KVM_GET_SREGS: {
1689                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1690                 r = -ENOMEM;
1691                 if (!kvm_sregs)
1692                         goto out;
1693                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1694                 if (r)
1695                         goto out;
1696                 r = -EFAULT;
1697                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1698                         goto out;
1699                 r = 0;
1700                 break;
1701         }
1702         case KVM_SET_SREGS: {
1703                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1704                 r = -ENOMEM;
1705                 if (!kvm_sregs)
1706                         goto out;
1707                 r = -EFAULT;
1708                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1709                         goto out;
1710                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1711                 if (r)
1712                         goto out;
1713                 r = 0;
1714                 break;
1715         }
1716         case KVM_GET_MP_STATE: {
1717                 struct kvm_mp_state mp_state;
1718
1719                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1720                 if (r)
1721                         goto out;
1722                 r = -EFAULT;
1723                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1724                         goto out;
1725                 r = 0;
1726                 break;
1727         }
1728         case KVM_SET_MP_STATE: {
1729                 struct kvm_mp_state mp_state;
1730
1731                 r = -EFAULT;
1732                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1733                         goto out;
1734                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1735                 if (r)
1736                         goto out;
1737                 r = 0;
1738                 break;
1739         }
1740         case KVM_TRANSLATE: {
1741                 struct kvm_translation tr;
1742
1743                 r = -EFAULT;
1744                 if (copy_from_user(&tr, argp, sizeof tr))
1745                         goto out;
1746                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1747                 if (r)
1748                         goto out;
1749                 r = -EFAULT;
1750                 if (copy_to_user(argp, &tr, sizeof tr))
1751                         goto out;
1752                 r = 0;
1753                 break;
1754         }
1755         case KVM_DEBUG_GUEST: {
1756                 struct kvm_debug_guest dbg;
1757
1758                 r = -EFAULT;
1759                 if (copy_from_user(&dbg, argp, sizeof dbg))
1760                         goto out;
1761                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1762                 if (r)
1763                         goto out;
1764                 r = 0;
1765                 break;
1766         }
1767         case KVM_SET_SIGNAL_MASK: {
1768                 struct kvm_signal_mask __user *sigmask_arg = argp;
1769                 struct kvm_signal_mask kvm_sigmask;
1770                 sigset_t sigset, *p;
1771
1772                 p = NULL;
1773                 if (argp) {
1774                         r = -EFAULT;
1775                         if (copy_from_user(&kvm_sigmask, argp,
1776                                            sizeof kvm_sigmask))
1777                                 goto out;
1778                         r = -EINVAL;
1779                         if (kvm_sigmask.len != sizeof sigset)
1780                                 goto out;
1781                         r = -EFAULT;
1782                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1783                                            sizeof sigset))
1784                                 goto out;
1785                         p = &sigset;
1786                 }
1787                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1788                 break;
1789         }
1790         case KVM_GET_FPU: {
1791                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1792                 r = -ENOMEM;
1793                 if (!fpu)
1794                         goto out;
1795                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1796                 if (r)
1797                         goto out;
1798                 r = -EFAULT;
1799                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1800                         goto out;
1801                 r = 0;
1802                 break;
1803         }
1804         case KVM_SET_FPU: {
1805                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1806                 r = -ENOMEM;
1807                 if (!fpu)
1808                         goto out;
1809                 r = -EFAULT;
1810                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1811                         goto out;
1812                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1813                 if (r)
1814                         goto out;
1815                 r = 0;
1816                 break;
1817         }
1818         default:
1819                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1820         }
1821 out:
1822         kfree(fpu);
1823         kfree(kvm_sregs);
1824         return r;
1825 }
1826
1827 static long kvm_vm_ioctl(struct file *filp,
1828                            unsigned int ioctl, unsigned long arg)
1829 {
1830         struct kvm *kvm = filp->private_data;
1831         void __user *argp = (void __user *)arg;
1832         int r;
1833
1834         if (kvm->mm != current->mm)
1835                 return -EIO;
1836         switch (ioctl) {
1837         case KVM_CREATE_VCPU:
1838                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1839                 if (r < 0)
1840                         goto out;
1841                 break;
1842         case KVM_SET_USER_MEMORY_REGION: {
1843                 struct kvm_userspace_memory_region kvm_userspace_mem;
1844
1845                 r = -EFAULT;
1846                 if (copy_from_user(&kvm_userspace_mem, argp,
1847                                                 sizeof kvm_userspace_mem))
1848                         goto out;
1849
1850                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1851                 if (r)
1852                         goto out;
1853                 break;
1854         }
1855         case KVM_GET_DIRTY_LOG: {
1856                 struct kvm_dirty_log log;
1857
1858                 r = -EFAULT;
1859                 if (copy_from_user(&log, argp, sizeof log))
1860                         goto out;
1861                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1862                 if (r)
1863                         goto out;
1864                 break;
1865         }
1866 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1867         case KVM_REGISTER_COALESCED_MMIO: {
1868                 struct kvm_coalesced_mmio_zone zone;
1869                 r = -EFAULT;
1870                 if (copy_from_user(&zone, argp, sizeof zone))
1871                         goto out;
1872                 r = -ENXIO;
1873                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1874                 if (r)
1875                         goto out;
1876                 r = 0;
1877                 break;
1878         }
1879         case KVM_UNREGISTER_COALESCED_MMIO: {
1880                 struct kvm_coalesced_mmio_zone zone;
1881                 r = -EFAULT;
1882                 if (copy_from_user(&zone, argp, sizeof zone))
1883                         goto out;
1884                 r = -ENXIO;
1885                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1886                 if (r)
1887                         goto out;
1888                 r = 0;
1889                 break;
1890         }
1891 #endif
1892 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1893         case KVM_ASSIGN_PCI_DEVICE: {
1894                 struct kvm_assigned_pci_dev assigned_dev;
1895
1896                 r = -EFAULT;
1897                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1898                         goto out;
1899                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1900                 if (r)
1901                         goto out;
1902                 break;
1903         }
1904         case KVM_ASSIGN_IRQ: {
1905                 struct kvm_assigned_irq assigned_irq;
1906
1907                 r = -EFAULT;
1908                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1909                         goto out;
1910                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1911                 if (r)
1912                         goto out;
1913                 break;
1914         }
1915 #endif
1916 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1917         case KVM_DEASSIGN_PCI_DEVICE: {
1918                 struct kvm_assigned_pci_dev assigned_dev;
1919
1920                 r = -EFAULT;
1921                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1922                         goto out;
1923                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1924                 if (r)
1925                         goto out;
1926                 break;
1927         }
1928 #endif
1929         default:
1930                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1931         }
1932 out:
1933         return r;
1934 }
1935
1936 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1937 {
1938         struct page *page[1];
1939         unsigned long addr;
1940         int npages;
1941         gfn_t gfn = vmf->pgoff;
1942         struct kvm *kvm = vma->vm_file->private_data;
1943
1944         addr = gfn_to_hva(kvm, gfn);
1945         if (kvm_is_error_hva(addr))
1946                 return VM_FAULT_SIGBUS;
1947
1948         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1949                                 NULL);
1950         if (unlikely(npages != 1))
1951                 return VM_FAULT_SIGBUS;
1952
1953         vmf->page = page[0];
1954         return 0;
1955 }
1956
1957 static struct vm_operations_struct kvm_vm_vm_ops = {
1958         .fault = kvm_vm_fault,
1959 };
1960
1961 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1962 {
1963         vma->vm_ops = &kvm_vm_vm_ops;
1964         return 0;
1965 }
1966
1967 static struct file_operations kvm_vm_fops = {
1968         .release        = kvm_vm_release,
1969         .unlocked_ioctl = kvm_vm_ioctl,
1970         .compat_ioctl   = kvm_vm_ioctl,
1971         .mmap           = kvm_vm_mmap,
1972 };
1973
1974 static int kvm_dev_ioctl_create_vm(void)
1975 {
1976         int fd;
1977         struct kvm *kvm;
1978
1979         kvm = kvm_create_vm();
1980         if (IS_ERR(kvm))
1981                 return PTR_ERR(kvm);
1982         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1983         if (fd < 0)
1984                 kvm_put_kvm(kvm);
1985
1986         return fd;
1987 }
1988
1989 static long kvm_dev_ioctl_check_extension_generic(long arg)
1990 {
1991         switch (arg) {
1992         case KVM_CAP_USER_MEMORY:
1993         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1994                 return 1;
1995         default:
1996                 break;
1997         }
1998         return kvm_dev_ioctl_check_extension(arg);
1999 }
2000
2001 static long kvm_dev_ioctl(struct file *filp,
2002                           unsigned int ioctl, unsigned long arg)
2003 {
2004         long r = -EINVAL;
2005
2006         switch (ioctl) {
2007         case KVM_GET_API_VERSION:
2008                 r = -EINVAL;
2009                 if (arg)
2010                         goto out;
2011                 r = KVM_API_VERSION;
2012                 break;
2013         case KVM_CREATE_VM:
2014                 r = -EINVAL;
2015                 if (arg)
2016                         goto out;
2017                 r = kvm_dev_ioctl_create_vm();
2018                 break;
2019         case KVM_CHECK_EXTENSION:
2020                 r = kvm_dev_ioctl_check_extension_generic(arg);
2021                 break;
2022         case KVM_GET_VCPU_MMAP_SIZE:
2023                 r = -EINVAL;
2024                 if (arg)
2025                         goto out;
2026                 r = PAGE_SIZE;     /* struct kvm_run */
2027 #ifdef CONFIG_X86
2028                 r += PAGE_SIZE;    /* pio data page */
2029 #endif
2030 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2031                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2032 #endif
2033                 break;
2034         case KVM_TRACE_ENABLE:
2035         case KVM_TRACE_PAUSE:
2036         case KVM_TRACE_DISABLE:
2037                 r = kvm_trace_ioctl(ioctl, arg);
2038                 break;
2039         default:
2040                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2041         }
2042 out:
2043         return r;
2044 }
2045
2046 static struct file_operations kvm_chardev_ops = {
2047         .unlocked_ioctl = kvm_dev_ioctl,
2048         .compat_ioctl   = kvm_dev_ioctl,
2049 };
2050
2051 static struct miscdevice kvm_dev = {
2052         KVM_MINOR,
2053         "kvm",
2054         &kvm_chardev_ops,
2055 };
2056
2057 static void hardware_enable(void *junk)
2058 {
2059         int cpu = raw_smp_processor_id();
2060
2061         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2062                 return;
2063         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2064         kvm_arch_hardware_enable(NULL);
2065 }
2066
2067 static void hardware_disable(void *junk)
2068 {
2069         int cpu = raw_smp_processor_id();
2070
2071         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2072                 return;
2073         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2074         kvm_arch_hardware_disable(NULL);
2075 }
2076
2077 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2078                            void *v)
2079 {
2080         int cpu = (long)v;
2081
2082         val &= ~CPU_TASKS_FROZEN;
2083         switch (val) {
2084         case CPU_DYING:
2085                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2086                        cpu);
2087                 hardware_disable(NULL);
2088                 break;
2089         case CPU_UP_CANCELED:
2090                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2091                        cpu);
2092                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2093                 break;
2094         case CPU_ONLINE:
2095                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2096                        cpu);
2097                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2098                 break;
2099         }
2100         return NOTIFY_OK;
2101 }
2102
2103
2104 asmlinkage void kvm_handle_fault_on_reboot(void)
2105 {
2106         if (kvm_rebooting)
2107                 /* spin while reset goes on */
2108                 while (true)
2109                         ;
2110         /* Fault while not rebooting.  We want the trace. */
2111         BUG();
2112 }
2113 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2114
2115 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2116                       void *v)
2117 {
2118         if (val == SYS_RESTART) {
2119                 /*
2120                  * Some (well, at least mine) BIOSes hang on reboot if
2121                  * in vmx root mode.
2122                  */
2123                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2124                 kvm_rebooting = true;
2125                 on_each_cpu(hardware_disable, NULL, 1);
2126         }
2127         return NOTIFY_OK;
2128 }
2129
2130 static struct notifier_block kvm_reboot_notifier = {
2131         .notifier_call = kvm_reboot,
2132         .priority = 0,
2133 };
2134
2135 void kvm_io_bus_init(struct kvm_io_bus *bus)
2136 {
2137         memset(bus, 0, sizeof(*bus));
2138 }
2139
2140 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2141 {
2142         int i;
2143
2144         for (i = 0; i < bus->dev_count; i++) {
2145                 struct kvm_io_device *pos = bus->devs[i];
2146
2147                 kvm_iodevice_destructor(pos);
2148         }
2149 }
2150
2151 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2152                                           gpa_t addr, int len, int is_write)
2153 {
2154         int i;
2155
2156         for (i = 0; i < bus->dev_count; i++) {
2157                 struct kvm_io_device *pos = bus->devs[i];
2158
2159                 if (pos->in_range(pos, addr, len, is_write))
2160                         return pos;
2161         }
2162
2163         return NULL;
2164 }
2165
2166 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2167 {
2168         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2169
2170         bus->devs[bus->dev_count++] = dev;
2171 }
2172
2173 static struct notifier_block kvm_cpu_notifier = {
2174         .notifier_call = kvm_cpu_hotplug,
2175         .priority = 20, /* must be > scheduler priority */
2176 };
2177
2178 static int vm_stat_get(void *_offset, u64 *val)
2179 {
2180         unsigned offset = (long)_offset;
2181         struct kvm *kvm;
2182
2183         *val = 0;
2184         spin_lock(&kvm_lock);
2185         list_for_each_entry(kvm, &vm_list, vm_list)
2186                 *val += *(u32 *)((void *)kvm + offset);
2187         spin_unlock(&kvm_lock);
2188         return 0;
2189 }
2190
2191 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2192
2193 static int vcpu_stat_get(void *_offset, u64 *val)
2194 {
2195         unsigned offset = (long)_offset;
2196         struct kvm *kvm;
2197         struct kvm_vcpu *vcpu;
2198         int i;
2199
2200         *val = 0;
2201         spin_lock(&kvm_lock);
2202         list_for_each_entry(kvm, &vm_list, vm_list)
2203                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2204                         vcpu = kvm->vcpus[i];
2205                         if (vcpu)
2206                                 *val += *(u32 *)((void *)vcpu + offset);
2207                 }
2208         spin_unlock(&kvm_lock);
2209         return 0;
2210 }
2211
2212 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2213
2214 static struct file_operations *stat_fops[] = {
2215         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2216         [KVM_STAT_VM]   = &vm_stat_fops,
2217 };
2218
2219 static void kvm_init_debug(void)
2220 {
2221         struct kvm_stats_debugfs_item *p;
2222
2223         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2224         for (p = debugfs_entries; p->name; ++p)
2225                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2226                                                 (void *)(long)p->offset,
2227                                                 stat_fops[p->kind]);
2228 }
2229
2230 static void kvm_exit_debug(void)
2231 {
2232         struct kvm_stats_debugfs_item *p;
2233
2234         for (p = debugfs_entries; p->name; ++p)
2235                 debugfs_remove(p->dentry);
2236         debugfs_remove(kvm_debugfs_dir);
2237 }
2238
2239 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2240 {
2241         hardware_disable(NULL);
2242         return 0;
2243 }
2244
2245 static int kvm_resume(struct sys_device *dev)
2246 {
2247         hardware_enable(NULL);
2248         return 0;
2249 }
2250
2251 static struct sysdev_class kvm_sysdev_class = {
2252         .name = "kvm",
2253         .suspend = kvm_suspend,
2254         .resume = kvm_resume,
2255 };
2256
2257 static struct sys_device kvm_sysdev = {
2258         .id = 0,
2259         .cls = &kvm_sysdev_class,
2260 };
2261
2262 struct page *bad_page;
2263 pfn_t bad_pfn;
2264
2265 static inline
2266 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2267 {
2268         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2269 }
2270
2271 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2272 {
2273         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2274
2275         kvm_arch_vcpu_load(vcpu, cpu);
2276 }
2277
2278 static void kvm_sched_out(struct preempt_notifier *pn,
2279                           struct task_struct *next)
2280 {
2281         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2282
2283         kvm_arch_vcpu_put(vcpu);
2284 }
2285
2286 int kvm_init(void *opaque, unsigned int vcpu_size,
2287                   struct module *module)
2288 {
2289         int r;
2290         int cpu;
2291
2292         kvm_init_debug();
2293
2294         r = kvm_arch_init(opaque);
2295         if (r)
2296                 goto out_fail;
2297
2298         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2299
2300         if (bad_page == NULL) {
2301                 r = -ENOMEM;
2302                 goto out;
2303         }
2304
2305         bad_pfn = page_to_pfn(bad_page);
2306
2307         if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2308                 r = -ENOMEM;
2309                 goto out_free_0;
2310         }
2311
2312         r = kvm_arch_hardware_setup();
2313         if (r < 0)
2314                 goto out_free_0a;
2315
2316         for_each_online_cpu(cpu) {
2317                 smp_call_function_single(cpu,
2318                                 kvm_arch_check_processor_compat,
2319                                 &r, 1);
2320                 if (r < 0)
2321                         goto out_free_1;
2322         }
2323
2324         on_each_cpu(hardware_enable, NULL, 1);
2325         r = register_cpu_notifier(&kvm_cpu_notifier);
2326         if (r)
2327                 goto out_free_2;
2328         register_reboot_notifier(&kvm_reboot_notifier);
2329
2330         r = sysdev_class_register(&kvm_sysdev_class);
2331         if (r)
2332                 goto out_free_3;
2333
2334         r = sysdev_register(&kvm_sysdev);
2335         if (r)
2336                 goto out_free_4;
2337
2338         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2339         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2340                                            __alignof__(struct kvm_vcpu),
2341                                            0, NULL);
2342         if (!kvm_vcpu_cache) {
2343                 r = -ENOMEM;
2344                 goto out_free_5;
2345         }
2346
2347         kvm_chardev_ops.owner = module;
2348         kvm_vm_fops.owner = module;
2349         kvm_vcpu_fops.owner = module;
2350
2351         r = misc_register(&kvm_dev);
2352         if (r) {
2353                 printk(KERN_ERR "kvm: misc device register failed\n");
2354                 goto out_free;
2355         }
2356
2357         kvm_preempt_ops.sched_in = kvm_sched_in;
2358         kvm_preempt_ops.sched_out = kvm_sched_out;
2359 #ifndef CONFIG_X86
2360         msi2intx = 0;
2361 #endif
2362
2363         return 0;
2364
2365 out_free:
2366         kmem_cache_destroy(kvm_vcpu_cache);
2367 out_free_5:
2368         sysdev_unregister(&kvm_sysdev);
2369 out_free_4:
2370         sysdev_class_unregister(&kvm_sysdev_class);
2371 out_free_3:
2372         unregister_reboot_notifier(&kvm_reboot_notifier);
2373         unregister_cpu_notifier(&kvm_cpu_notifier);
2374 out_free_2:
2375         on_each_cpu(hardware_disable, NULL, 1);
2376 out_free_1:
2377         kvm_arch_hardware_unsetup();
2378 out_free_0a:
2379         free_cpumask_var(cpus_hardware_enabled);
2380 out_free_0:
2381         __free_page(bad_page);
2382 out:
2383         kvm_arch_exit();
2384         kvm_exit_debug();
2385 out_fail:
2386         return r;
2387 }
2388 EXPORT_SYMBOL_GPL(kvm_init);
2389
2390 void kvm_exit(void)
2391 {
2392         kvm_trace_cleanup();
2393         misc_deregister(&kvm_dev);
2394         kmem_cache_destroy(kvm_vcpu_cache);
2395         sysdev_unregister(&kvm_sysdev);
2396         sysdev_class_unregister(&kvm_sysdev_class);
2397         unregister_reboot_notifier(&kvm_reboot_notifier);
2398         unregister_cpu_notifier(&kvm_cpu_notifier);
2399         on_each_cpu(hardware_disable, NULL, 1);
2400         kvm_arch_hardware_unsetup();
2401         kvm_arch_exit();
2402         kvm_exit_debug();
2403         free_cpumask_var(cpus_hardware_enabled);
2404         __free_page(bad_page);
2405 }
2406 EXPORT_SYMBOL_GPL(kvm_exit);