?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge commit 'refs/merge-requests/1' of git://gitorious.org/linux-on-wince-htc/linux_...
[htc-linux.git] / virt / kvm / kvm_main.c
blob7495ce3473448cd45ee1dba9e0ae441c83245ca4
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 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46
47 #include <asm/processor.h>
48 #include <asm/io.h>
49 #include <asm/uaccess.h>
50 #include <asm/pgtable.h>
51
52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
53 #include "coalesced_mmio.h"
54 #endif
55
56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
57 #include <linux/pci.h>
58 #include <linux/interrupt.h>
59 #include "irq.h"
60 #endif
61
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/kvm.h>
64
65 MODULE_AUTHOR("Qumranet");
66 MODULE_LICENSE("GPL");
67
68 /*
69 * Ordering of locks:
70 *
71 * kvm->slots_lock --> kvm->lock --> kvm->irq_lock
72 */
73
74 DEFINE_SPINLOCK(kvm_lock);
75 LIST_HEAD(vm_list);
76
77 static cpumask_var_t cpus_hardware_enabled;
78
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
83
84 struct dentry *kvm_debugfs_dir;
85
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 unsigned long arg);
88
89 static bool kvm_rebooting;
90
91 static bool largepages_enabled = true;
92
93 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
94 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
95 int assigned_dev_id)
96 {
97 struct list_head *ptr;
98 struct kvm_assigned_dev_kernel *match;
99
100 list_for_each(ptr, head) {
101 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
102 if (match->assigned_dev_id == assigned_dev_id)
103 return match;
104 }
105 return NULL;
106 }
107
108 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
109 *assigned_dev, int irq)
110 {
111 int i, index;
112 struct msix_entry *host_msix_entries;
113
114 host_msix_entries = assigned_dev->host_msix_entries;
115
116 index = -1;
117 for (i = 0; i < assigned_dev->entries_nr; i++)
118 if (irq == host_msix_entries[i].vector) {
119 index = i;
120 break;
121 }
122 if (index < 0) {
123 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
124 return 0;
125 }
126
127 return index;
128 }
129
130 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
131 {
132 struct kvm_assigned_dev_kernel *assigned_dev;
133 struct kvm *kvm;
134 int i;
135
136 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
137 interrupt_work);
138 kvm = assigned_dev->kvm;
139
140 mutex_lock(&kvm->irq_lock);
141 spin_lock_irq(&assigned_dev->assigned_dev_lock);
142 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
143 struct kvm_guest_msix_entry *guest_entries =
144 assigned_dev->guest_msix_entries;
145 for (i = 0; i < assigned_dev->entries_nr; i++) {
146 if (!(guest_entries[i].flags &
147 KVM_ASSIGNED_MSIX_PENDING))
148 continue;
149 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
150 kvm_set_irq(assigned_dev->kvm,
151 assigned_dev->irq_source_id,
152 guest_entries[i].vector, 1);
153 }
154 } else
155 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
156 assigned_dev->guest_irq, 1);
157
158 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
159 mutex_unlock(&assigned_dev->kvm->irq_lock);
160 }
161
162 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
163 {
164 unsigned long flags;
165 struct kvm_assigned_dev_kernel *assigned_dev =
166 (struct kvm_assigned_dev_kernel *) dev_id;
167
168 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
169 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
170 int index = find_index_from_host_irq(assigned_dev, irq);
171 if (index < 0)
172 goto out;
173 assigned_dev->guest_msix_entries[index].flags |=
174 KVM_ASSIGNED_MSIX_PENDING;
175 }
176
177 schedule_work(&assigned_dev->interrupt_work);
178
179 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
180 disable_irq_nosync(irq);
181 assigned_dev->host_irq_disabled = true;
182 }
183
184 out:
185 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
186 return IRQ_HANDLED;
187 }
188
189 /* Ack the irq line for an assigned device */
190 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
191 {
192 struct kvm_assigned_dev_kernel *dev;
193 unsigned long flags;
194
195 if (kian->gsi == -1)
196 return;
197
198 dev = container_of(kian, struct kvm_assigned_dev_kernel,
199 ack_notifier);
200
201 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
202
203 /* The guest irq may be shared so this ack may be
204 * from another device.
205 */
206 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
207 if (dev->host_irq_disabled) {
208 enable_irq(dev->host_irq);
209 dev->host_irq_disabled = false;
210 }
211 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
212 }
213
214 static void deassign_guest_irq(struct kvm *kvm,
215 struct kvm_assigned_dev_kernel *assigned_dev)
216 {
217 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
218 assigned_dev->ack_notifier.gsi = -1;
219
220 if (assigned_dev->irq_source_id != -1)
221 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
222 assigned_dev->irq_source_id = -1;
223 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
224 }
225
226 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
227 static void deassign_host_irq(struct kvm *kvm,
228 struct kvm_assigned_dev_kernel *assigned_dev)
229 {
230 /*
231 * In kvm_free_device_irq, cancel_work_sync return true if:
232 * 1. work is scheduled, and then cancelled.
233 * 2. work callback is executed.
234 *
235 * The first one ensured that the irq is disabled and no more events
236 * would happen. But for the second one, the irq may be enabled (e.g.
237 * for MSI). So we disable irq here to prevent further events.
238 *
239 * Notice this maybe result in nested disable if the interrupt type is
240 * INTx, but it's OK for we are going to free it.
241 *
242 * If this function is a part of VM destroy, please ensure that till
243 * now, the kvm state is still legal for probably we also have to wait
244 * interrupt_work done.
245 */
246 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
247 int i;
248 for (i = 0; i < assigned_dev->entries_nr; i++)
249 disable_irq_nosync(assigned_dev->
250 host_msix_entries[i].vector);
251
252 cancel_work_sync(&assigned_dev->interrupt_work);
253
254 for (i = 0; i < assigned_dev->entries_nr; i++)
255 free_irq(assigned_dev->host_msix_entries[i].vector,
256 (void *)assigned_dev);
257
258 assigned_dev->entries_nr = 0;
259 kfree(assigned_dev->host_msix_entries);
260 kfree(assigned_dev->guest_msix_entries);
261 pci_disable_msix(assigned_dev->dev);
262 } else {
263 /* Deal with MSI and INTx */
264 disable_irq_nosync(assigned_dev->host_irq);
265 cancel_work_sync(&assigned_dev->interrupt_work);
266
267 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
268
269 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
270 pci_disable_msi(assigned_dev->dev);
271 }
272
273 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
274 }
275
276 static int kvm_deassign_irq(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *assigned_dev,
278 unsigned long irq_requested_type)
279 {
280 unsigned long guest_irq_type, host_irq_type;
281
282 if (!irqchip_in_kernel(kvm))
283 return -EINVAL;
284 /* no irq assignment to deassign */
285 if (!assigned_dev->irq_requested_type)
286 return -ENXIO;
287
288 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
289 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
290
291 if (host_irq_type)
292 deassign_host_irq(kvm, assigned_dev);
293 if (guest_irq_type)
294 deassign_guest_irq(kvm, assigned_dev);
295
296 return 0;
297 }
298
299 static void kvm_free_assigned_irq(struct kvm *kvm,
300 struct kvm_assigned_dev_kernel *assigned_dev)
301 {
302 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
303 }
304
305 static void kvm_free_assigned_device(struct kvm *kvm,
306 struct kvm_assigned_dev_kernel
307 *assigned_dev)
308 {
309 kvm_free_assigned_irq(kvm, assigned_dev);
310
311 pci_reset_function(assigned_dev->dev);
312
313 pci_release_regions(assigned_dev->dev);
314 pci_disable_device(assigned_dev->dev);
315 pci_dev_put(assigned_dev->dev);
316
317 list_del(&assigned_dev->list);
318 kfree(assigned_dev);
319 }
320
321 void kvm_free_all_assigned_devices(struct kvm *kvm)
322 {
323 struct list_head *ptr, *ptr2;
324 struct kvm_assigned_dev_kernel *assigned_dev;
325
326 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
327 assigned_dev = list_entry(ptr,
328 struct kvm_assigned_dev_kernel,
329 list);
330
331 kvm_free_assigned_device(kvm, assigned_dev);
332 }
333 }
334
335 static int assigned_device_enable_host_intx(struct kvm *kvm,
336 struct kvm_assigned_dev_kernel *dev)
337 {
338 dev->host_irq = dev->dev->irq;
339 /* Even though this is PCI, we don't want to use shared
340 * interrupts. Sharing host devices with guest-assigned devices
341 * on the same interrupt line is not a happy situation: there
342 * are going to be long delays in accepting, acking, etc.
343 */
344 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
345 0, "kvm_assigned_intx_device", (void *)dev))
346 return -EIO;
347 return 0;
348 }
349
350 #ifdef __KVM_HAVE_MSI
351 static int assigned_device_enable_host_msi(struct kvm *kvm,
352 struct kvm_assigned_dev_kernel *dev)
353 {
354 int r;
355
356 if (!dev->dev->msi_enabled) {
357 r = pci_enable_msi(dev->dev);
358 if (r)
359 return r;
360 }
361
362 dev->host_irq = dev->dev->irq;
363 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
364 "kvm_assigned_msi_device", (void *)dev)) {
365 pci_disable_msi(dev->dev);
366 return -EIO;
367 }
368
369 return 0;
370 }
371 #endif
372
373 #ifdef __KVM_HAVE_MSIX
374 static int assigned_device_enable_host_msix(struct kvm *kvm,
375 struct kvm_assigned_dev_kernel *dev)
376 {
377 int i, r = -EINVAL;
378
379 /* host_msix_entries and guest_msix_entries should have been
380 * initialized */
381 if (dev->entries_nr == 0)
382 return r;
383
384 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
385 if (r)
386 return r;
387
388 for (i = 0; i < dev->entries_nr; i++) {
389 r = request_irq(dev->host_msix_entries[i].vector,
390 kvm_assigned_dev_intr, 0,
391 "kvm_assigned_msix_device",
392 (void *)dev);
393 /* FIXME: free requested_irq's on failure */
394 if (r)
395 return r;
396 }
397
398 return 0;
399 }
400
401 #endif
402
403 static int assigned_device_enable_guest_intx(struct kvm *kvm,
404 struct kvm_assigned_dev_kernel *dev,
405 struct kvm_assigned_irq *irq)
406 {
407 dev->guest_irq = irq->guest_irq;
408 dev->ack_notifier.gsi = irq->guest_irq;
409 return 0;
410 }
411
412 #ifdef __KVM_HAVE_MSI
413 static int assigned_device_enable_guest_msi(struct kvm *kvm,
414 struct kvm_assigned_dev_kernel *dev,
415 struct kvm_assigned_irq *irq)
416 {
417 dev->guest_irq = irq->guest_irq;
418 dev->ack_notifier.gsi = -1;
419 dev->host_irq_disabled = false;
420 return 0;
421 }
422 #endif
423 #ifdef __KVM_HAVE_MSIX
424 static int assigned_device_enable_guest_msix(struct kvm *kvm,
425 struct kvm_assigned_dev_kernel *dev,
426 struct kvm_assigned_irq *irq)
427 {
428 dev->guest_irq = irq->guest_irq;
429 dev->ack_notifier.gsi = -1;
430 dev->host_irq_disabled = false;
431 return 0;
432 }
433 #endif
434
435 static int assign_host_irq(struct kvm *kvm,
436 struct kvm_assigned_dev_kernel *dev,
437 __u32 host_irq_type)
438 {
439 int r = -EEXIST;
440
441 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
442 return r;
443
444 switch (host_irq_type) {
445 case KVM_DEV_IRQ_HOST_INTX:
446 r = assigned_device_enable_host_intx(kvm, dev);
447 break;
448 #ifdef __KVM_HAVE_MSI
449 case KVM_DEV_IRQ_HOST_MSI:
450 r = assigned_device_enable_host_msi(kvm, dev);
451 break;
452 #endif
453 #ifdef __KVM_HAVE_MSIX
454 case KVM_DEV_IRQ_HOST_MSIX:
455 r = assigned_device_enable_host_msix(kvm, dev);
456 break;
457 #endif
458 default:
459 r = -EINVAL;
460 }
461
462 if (!r)
463 dev->irq_requested_type |= host_irq_type;
464
465 return r;
466 }
467
468 static int assign_guest_irq(struct kvm *kvm,
469 struct kvm_assigned_dev_kernel *dev,
470 struct kvm_assigned_irq *irq,
471 unsigned long guest_irq_type)
472 {
473 int id;
474 int r = -EEXIST;
475
476 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
477 return r;
478
479 id = kvm_request_irq_source_id(kvm);
480 if (id < 0)
481 return id;
482
483 dev->irq_source_id = id;
484
485 switch (guest_irq_type) {
486 case KVM_DEV_IRQ_GUEST_INTX:
487 r = assigned_device_enable_guest_intx(kvm, dev, irq);
488 break;
489 #ifdef __KVM_HAVE_MSI
490 case KVM_DEV_IRQ_GUEST_MSI:
491 r = assigned_device_enable_guest_msi(kvm, dev, irq);
492 break;
493 #endif
494 #ifdef __KVM_HAVE_MSIX
495 case KVM_DEV_IRQ_GUEST_MSIX:
496 r = assigned_device_enable_guest_msix(kvm, dev, irq);
497 break;
498 #endif
499 default:
500 r = -EINVAL;
501 }
502
503 if (!r) {
504 dev->irq_requested_type |= guest_irq_type;
505 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
506 } else
507 kvm_free_irq_source_id(kvm, dev->irq_source_id);
508
509 return r;
510 }
511
512 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
513 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
514 struct kvm_assigned_irq *assigned_irq)
515 {
516 int r = -EINVAL;
517 struct kvm_assigned_dev_kernel *match;
518 unsigned long host_irq_type, guest_irq_type;
519
520 if (!capable(CAP_SYS_RAWIO))
521 return -EPERM;
522
523 if (!irqchip_in_kernel(kvm))
524 return r;
525
526 mutex_lock(&kvm->lock);
527 r = -ENODEV;
528 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
529 assigned_irq->assigned_dev_id);
530 if (!match)
531 goto out;
532
533 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
534 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
535
536 r = -EINVAL;
537 /* can only assign one type at a time */
538 if (hweight_long(host_irq_type) > 1)
539 goto out;
540 if (hweight_long(guest_irq_type) > 1)
541 goto out;
542 if (host_irq_type == 0 && guest_irq_type == 0)
543 goto out;
544
545 r = 0;
546 if (host_irq_type)
547 r = assign_host_irq(kvm, match, host_irq_type);
548 if (r)
549 goto out;
550
551 if (guest_irq_type)
552 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
553 out:
554 mutex_unlock(&kvm->lock);
555 return r;
556 }
557
558 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
559 struct kvm_assigned_irq
560 *assigned_irq)
561 {
562 int r = -ENODEV;
563 struct kvm_assigned_dev_kernel *match;
564
565 mutex_lock(&kvm->lock);
566
567 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
568 assigned_irq->assigned_dev_id);
569 if (!match)
570 goto out;
571
572 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
573 out:
574 mutex_unlock(&kvm->lock);
575 return r;
576 }
577
578 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
579 struct kvm_assigned_pci_dev *assigned_dev)
580 {
581 int r = 0;
582 struct kvm_assigned_dev_kernel *match;
583 struct pci_dev *dev;
584
585 down_read(&kvm->slots_lock);
586 mutex_lock(&kvm->lock);
587
588 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
589 assigned_dev->assigned_dev_id);
590 if (match) {
591 /* device already assigned */
592 r = -EEXIST;
593 goto out;
594 }
595
596 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
597 if (match == NULL) {
598 printk(KERN_INFO "%s: Couldn't allocate memory\n",
599 __func__);
600 r = -ENOMEM;
601 goto out;
602 }
603 dev = pci_get_bus_and_slot(assigned_dev->busnr,
604 assigned_dev->devfn);
605 if (!dev) {
606 printk(KERN_INFO "%s: host device not found\n", __func__);
607 r = -EINVAL;
608 goto out_free;
609 }
610 if (pci_enable_device(dev)) {
611 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
612 r = -EBUSY;
613 goto out_put;
614 }
615 r = pci_request_regions(dev, "kvm_assigned_device");
616 if (r) {
617 printk(KERN_INFO "%s: Could not get access to device regions\n",
618 __func__);
619 goto out_disable;
620 }
621
622 pci_reset_function(dev);
623
624 match->assigned_dev_id = assigned_dev->assigned_dev_id;
625 match->host_busnr = assigned_dev->busnr;
626 match->host_devfn = assigned_dev->devfn;
627 match->flags = assigned_dev->flags;
628 match->dev = dev;
629 spin_lock_init(&match->assigned_dev_lock);
630 match->irq_source_id = -1;
631 match->kvm = kvm;
632 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
633 INIT_WORK(&match->interrupt_work,
634 kvm_assigned_dev_interrupt_work_handler);
635
636 list_add(&match->list, &kvm->arch.assigned_dev_head);
637
638 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
639 if (!kvm->arch.iommu_domain) {
640 r = kvm_iommu_map_guest(kvm);
641 if (r)
642 goto out_list_del;
643 }
644 r = kvm_assign_device(kvm, match);
645 if (r)
646 goto out_list_del;
647 }
648
649 out:
650 mutex_unlock(&kvm->lock);
651 up_read(&kvm->slots_lock);
652 return r;
653 out_list_del:
654 list_del(&match->list);
655 pci_release_regions(dev);
656 out_disable:
657 pci_disable_device(dev);
658 out_put:
659 pci_dev_put(dev);
660 out_free:
661 kfree(match);
662 mutex_unlock(&kvm->lock);
663 up_read(&kvm->slots_lock);
664 return r;
665 }
666 #endif
667
668 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
669 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
670 struct kvm_assigned_pci_dev *assigned_dev)
671 {
672 int r = 0;
673 struct kvm_assigned_dev_kernel *match;
674
675 mutex_lock(&kvm->lock);
676
677 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
678 assigned_dev->assigned_dev_id);
679 if (!match) {
680 printk(KERN_INFO "%s: device hasn't been assigned before, "
681 "so cannot be deassigned\n", __func__);
682 r = -EINVAL;
683 goto out;
684 }
685
686 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
687 kvm_deassign_device(kvm, match);
688
689 kvm_free_assigned_device(kvm, match);
690
691 out:
692 mutex_unlock(&kvm->lock);
693 return r;
694 }
695 #endif
696
697 inline int kvm_is_mmio_pfn(pfn_t pfn)
698 {
699 if (pfn_valid(pfn)) {
700 struct page *page = compound_head(pfn_to_page(pfn));
701 return PageReserved(page);
702 }
703
704 return true;
705 }
706
707 /*
708 * Switches to specified vcpu, until a matching vcpu_put()
709 */
710 void vcpu_load(struct kvm_vcpu *vcpu)
711 {
712 int cpu;
713
714 mutex_lock(&vcpu->mutex);
715 cpu = get_cpu();
716 preempt_notifier_register(&vcpu->preempt_notifier);
717 kvm_arch_vcpu_load(vcpu, cpu);
718 put_cpu();
719 }
720
721 void vcpu_put(struct kvm_vcpu *vcpu)
722 {
723 preempt_disable();
724 kvm_arch_vcpu_put(vcpu);
725 preempt_notifier_unregister(&vcpu->preempt_notifier);
726 preempt_enable();
727 mutex_unlock(&vcpu->mutex);
728 }
729
730 static void ack_flush(void *_completed)
731 {
732 }
733
734 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
735 {
736 int i, cpu, me;
737 cpumask_var_t cpus;
738 bool called = true;
739 struct kvm_vcpu *vcpu;
740
741 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
742
743 spin_lock(&kvm->requests_lock);
744 me = smp_processor_id();
745 kvm_for_each_vcpu(i, vcpu, kvm) {
746 if (test_and_set_bit(req, &vcpu->requests))
747 continue;
748 cpu = vcpu->cpu;
749 if (cpus != NULL && cpu != -1 && cpu != me)
750 cpumask_set_cpu(cpu, cpus);
751 }
752 if (unlikely(cpus == NULL))
753 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
754 else if (!cpumask_empty(cpus))
755 smp_call_function_many(cpus, ack_flush, NULL, 1);
756 else
757 called = false;
758 spin_unlock(&kvm->requests_lock);
759 free_cpumask_var(cpus);
760 return called;
761 }
762
763 void kvm_flush_remote_tlbs(struct kvm *kvm)
764 {
765 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
766 ++kvm->stat.remote_tlb_flush;
767 }
768
769 void kvm_reload_remote_mmus(struct kvm *kvm)
770 {
771 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
772 }
773
774 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
775 {
776 struct page *page;
777 int r;
778
779 mutex_init(&vcpu->mutex);
780 vcpu->cpu = -1;
781 vcpu->kvm = kvm;
782 vcpu->vcpu_id = id;
783 init_waitqueue_head(&vcpu->wq);
784
785 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
786 if (!page) {
787 r = -ENOMEM;
788 goto fail;
789 }
790 vcpu->run = page_address(page);
791
792 r = kvm_arch_vcpu_init(vcpu);
793 if (r < 0)
794 goto fail_free_run;
795 return 0;
796
797 fail_free_run:
798 free_page((unsigned long)vcpu->run);
799 fail:
800 return r;
801 }
802 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
803
804 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
805 {
806 kvm_arch_vcpu_uninit(vcpu);
807 free_page((unsigned long)vcpu->run);
808 }
809 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
810
811 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
812 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
813 {
814 return container_of(mn, struct kvm, mmu_notifier);
815 }
816
817 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
818 struct mm_struct *mm,
819 unsigned long address)
820 {
821 struct kvm *kvm = mmu_notifier_to_kvm(mn);
822 int need_tlb_flush;
823
824 /*
825 * When ->invalidate_page runs, the linux pte has been zapped
826 * already but the page is still allocated until
827 * ->invalidate_page returns. So if we increase the sequence
828 * here the kvm page fault will notice if the spte can't be
829 * established because the page is going to be freed. If
830 * instead the kvm page fault establishes the spte before
831 * ->invalidate_page runs, kvm_unmap_hva will release it
832 * before returning.
833 *
834 * The sequence increase only need to be seen at spin_unlock
835 * time, and not at spin_lock time.
836 *
837 * Increasing the sequence after the spin_unlock would be
838 * unsafe because the kvm page fault could then establish the
839 * pte after kvm_unmap_hva returned, without noticing the page
840 * is going to be freed.
841 */
842 spin_lock(&kvm->mmu_lock);
843 kvm->mmu_notifier_seq++;
844 need_tlb_flush = kvm_unmap_hva(kvm, address);
845 spin_unlock(&kvm->mmu_lock);
846
847 /* we've to flush the tlb before the pages can be freed */
848 if (need_tlb_flush)
849 kvm_flush_remote_tlbs(kvm);
850
851 }
852
853 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
854 struct mm_struct *mm,
855 unsigned long address,
856 pte_t pte)
857 {
858 struct kvm *kvm = mmu_notifier_to_kvm(mn);
859
860 spin_lock(&kvm->mmu_lock);
861 kvm->mmu_notifier_seq++;
862 kvm_set_spte_hva(kvm, address, pte);
863 spin_unlock(&kvm->mmu_lock);
864 }
865
866 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
867 struct mm_struct *mm,
868 unsigned long start,
869 unsigned long end)
870 {
871 struct kvm *kvm = mmu_notifier_to_kvm(mn);
872 int need_tlb_flush = 0;
873
874 spin_lock(&kvm->mmu_lock);
875 /*
876 * The count increase must become visible at unlock time as no
877 * spte can be established without taking the mmu_lock and
878 * count is also read inside the mmu_lock critical section.
879 */
880 kvm->mmu_notifier_count++;
881 for (; start < end; start += PAGE_SIZE)
882 need_tlb_flush |= kvm_unmap_hva(kvm, start);
883 spin_unlock(&kvm->mmu_lock);
884
885 /* we've to flush the tlb before the pages can be freed */
886 if (need_tlb_flush)
887 kvm_flush_remote_tlbs(kvm);
888 }
889
890 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
891 struct mm_struct *mm,
892 unsigned long start,
893 unsigned long end)
894 {
895 struct kvm *kvm = mmu_notifier_to_kvm(mn);
896
897 spin_lock(&kvm->mmu_lock);
898 /*
899 * This sequence increase will notify the kvm page fault that
900 * the page that is going to be mapped in the spte could have
901 * been freed.
902 */
903 kvm->mmu_notifier_seq++;
904 /*
905 * The above sequence increase must be visible before the
906 * below count decrease but both values are read by the kvm
907 * page fault under mmu_lock spinlock so we don't need to add
908 * a smb_wmb() here in between the two.
909 */
910 kvm->mmu_notifier_count--;
911 spin_unlock(&kvm->mmu_lock);
912
913 BUG_ON(kvm->mmu_notifier_count < 0);
914 }
915
916 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
917 struct mm_struct *mm,
918 unsigned long address)
919 {
920 struct kvm *kvm = mmu_notifier_to_kvm(mn);
921 int young;
922
923 spin_lock(&kvm->mmu_lock);
924 young = kvm_age_hva(kvm, address);
925 spin_unlock(&kvm->mmu_lock);
926
927 if (young)
928 kvm_flush_remote_tlbs(kvm);
929
930 return young;
931 }
932
933 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
934 struct mm_struct *mm)
935 {
936 struct kvm *kvm = mmu_notifier_to_kvm(mn);
937 kvm_arch_flush_shadow(kvm);
938 }
939
940 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
941 .invalidate_page = kvm_mmu_notifier_invalidate_page,
942 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
943 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
944 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
945 .change_pte = kvm_mmu_notifier_change_pte,
946 .release = kvm_mmu_notifier_release,
947 };
948 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
949
950 static struct kvm *kvm_create_vm(void)
951 {
952 struct kvm *kvm = kvm_arch_create_vm();
953 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
954 struct page *page;
955 #endif
956
957 if (IS_ERR(kvm))
958 goto out;
959 #ifdef CONFIG_HAVE_KVM_IRQCHIP
960 INIT_LIST_HEAD(&kvm->irq_routing);
961 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
962 #endif
963
964 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
965 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
966 if (!page) {
967 kfree(kvm);
968 return ERR_PTR(-ENOMEM);
969 }
970 kvm->coalesced_mmio_ring =
971 (struct kvm_coalesced_mmio_ring *)page_address(page);
972 #endif
973
974 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
975 {
976 int err;
977 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
978 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
979 if (err) {
980 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
981 put_page(page);
982 #endif
983 kfree(kvm);
984 return ERR_PTR(err);
985 }
986 }
987 #endif
988
989 kvm->mm = current->mm;
990 atomic_inc(&kvm->mm->mm_count);
991 spin_lock_init(&kvm->mmu_lock);
992 spin_lock_init(&kvm->requests_lock);
993 kvm_io_bus_init(&kvm->pio_bus);
994 kvm_eventfd_init(kvm);
995 mutex_init(&kvm->lock);
996 mutex_init(&kvm->irq_lock);
997 kvm_io_bus_init(&kvm->mmio_bus);
998 init_rwsem(&kvm->slots_lock);
999 atomic_set(&kvm->users_count, 1);
1000 spin_lock(&kvm_lock);
1001 list_add(&kvm->vm_list, &vm_list);
1002 spin_unlock(&kvm_lock);
1003 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1004 kvm_coalesced_mmio_init(kvm);
1005 #endif
1006 out:
1007 return kvm;
1008 }
1009
1010 /*
1011 * Free any memory in @free but not in @dont.
1012 */
1013 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1014 struct kvm_memory_slot *dont)
1015 {
1016 int i;
1017
1018 if (!dont || free->rmap != dont->rmap)
1019 vfree(free->rmap);
1020
1021 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1022 vfree(free->dirty_bitmap);
1023
1024
1025 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1026 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1027 vfree(free->lpage_info[i]);
1028 free->lpage_info[i] = NULL;
1029 }
1030 }
1031
1032 free->npages = 0;
1033 free->dirty_bitmap = NULL;
1034 free->rmap = NULL;
1035 }
1036
1037 void kvm_free_physmem(struct kvm *kvm)
1038 {
1039 int i;
1040
1041 for (i = 0; i < kvm->nmemslots; ++i)
1042 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1043 }
1044
1045 static void kvm_destroy_vm(struct kvm *kvm)
1046 {
1047 struct mm_struct *mm = kvm->mm;
1048
1049 kvm_arch_sync_events(kvm);
1050 spin_lock(&kvm_lock);
1051 list_del(&kvm->vm_list);
1052 spin_unlock(&kvm_lock);
1053 kvm_free_irq_routing(kvm);
1054 kvm_io_bus_destroy(&kvm->pio_bus);
1055 kvm_io_bus_destroy(&kvm->mmio_bus);
1056 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1057 if (kvm->coalesced_mmio_ring != NULL)
1058 free_page((unsigned long)kvm->coalesced_mmio_ring);
1059 #endif
1060 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1061 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1062 #else
1063 kvm_arch_flush_shadow(kvm);
1064 #endif
1065 kvm_arch_destroy_vm(kvm);
1066 mmdrop(mm);
1067 }
1068
1069 void kvm_get_kvm(struct kvm *kvm)
1070 {
1071 atomic_inc(&kvm->users_count);
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1074
1075 void kvm_put_kvm(struct kvm *kvm)
1076 {
1077 if (atomic_dec_and_test(&kvm->users_count))
1078 kvm_destroy_vm(kvm);
1079 }
1080 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1081
1082
1083 static int kvm_vm_release(struct inode *inode, struct file *filp)
1084 {
1085 struct kvm *kvm = filp->private_data;
1086
1087 kvm_irqfd_release(kvm);
1088
1089 kvm_put_kvm(kvm);
1090 return 0;
1091 }
1092
1093 /*
1094 * Allocate some memory and give it an address in the guest physical address
1095 * space.
1096 *
1097 * Discontiguous memory is allowed, mostly for framebuffers.
1098 *
1099 * Must be called holding mmap_sem for write.
1100 */
1101 int __kvm_set_memory_region(struct kvm *kvm,
1102 struct kvm_userspace_memory_region *mem,
1103 int user_alloc)
1104 {
1105 int r;
1106 gfn_t base_gfn;
1107 unsigned long npages;
1108 unsigned long i;
1109 struct kvm_memory_slot *memslot;
1110 struct kvm_memory_slot old, new;
1111
1112 r = -EINVAL;
1113 /* General sanity checks */
1114 if (mem->memory_size & (PAGE_SIZE - 1))
1115 goto out;
1116 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1117 goto out;
1118 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1119 goto out;
1120 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1121 goto out;
1122 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1123 goto out;
1124
1125 memslot = &kvm->memslots[mem->slot];
1126 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1127 npages = mem->memory_size >> PAGE_SHIFT;
1128
1129 if (!npages)
1130 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1131
1132 new = old = *memslot;
1133
1134 new.base_gfn = base_gfn;
1135 new.npages = npages;
1136 new.flags = mem->flags;
1137
1138 /* Disallow changing a memory slot's size. */
1139 r = -EINVAL;
1140 if (npages && old.npages && npages != old.npages)
1141 goto out_free;
1142
1143 /* Check for overlaps */
1144 r = -EEXIST;
1145 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1146 struct kvm_memory_slot *s = &kvm->memslots[i];
1147
1148 if (s == memslot || !s->npages)
1149 continue;
1150 if (!((base_gfn + npages <= s->base_gfn) ||
1151 (base_gfn >= s->base_gfn + s->npages)))
1152 goto out_free;
1153 }
1154
1155 /* Free page dirty bitmap if unneeded */
1156 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1157 new.dirty_bitmap = NULL;
1158
1159 r = -ENOMEM;
1160
1161 /* Allocate if a slot is being created */
1162 #ifndef CONFIG_S390
1163 if (npages && !new.rmap) {
1164 new.rmap = vmalloc(npages * sizeof(struct page *));
1165
1166 if (!new.rmap)
1167 goto out_free;
1168
1169 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1170
1171 new.user_alloc = user_alloc;
1172 /*
1173 * hva_to_rmmap() serialzies with the mmu_lock and to be
1174 * safe it has to ignore memslots with !user_alloc &&
1175 * !userspace_addr.
1176 */
1177 if (user_alloc)
1178 new.userspace_addr = mem->userspace_addr;
1179 else
1180 new.userspace_addr = 0;
1181 }
1182 if (!npages)
1183 goto skip_lpage;
1184
1185 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1186 unsigned long ugfn;
1187 unsigned long j;
1188 int lpages;
1189 int level = i + 2;
1190
1191 /* Avoid unused variable warning if no large pages */
1192 (void)level;
1193
1194 if (new.lpage_info[i])
1195 continue;
1196
1197 lpages = 1 + (base_gfn + npages - 1) /
1198 KVM_PAGES_PER_HPAGE(level);
1199 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1200
1201 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1202
1203 if (!new.lpage_info[i])
1204 goto out_free;
1205
1206 memset(new.lpage_info[i], 0,
1207 lpages * sizeof(*new.lpage_info[i]));
1208
1209 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1210 new.lpage_info[i][0].write_count = 1;
1211 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1212 new.lpage_info[i][lpages - 1].write_count = 1;
1213 ugfn = new.userspace_addr >> PAGE_SHIFT;
1214 /*
1215 * If the gfn and userspace address are not aligned wrt each
1216 * other, or if explicitly asked to, disable large page
1217 * support for this slot
1218 */
1219 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1220 !largepages_enabled)
1221 for (j = 0; j < lpages; ++j)
1222 new.lpage_info[i][j].write_count = 1;
1223 }
1224
1225 skip_lpage:
1226
1227 /* Allocate page dirty bitmap if needed */
1228 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1229 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1230
1231 new.dirty_bitmap = vmalloc(dirty_bytes);
1232 if (!new.dirty_bitmap)
1233 goto out_free;
1234 memset(new.dirty_bitmap, 0, dirty_bytes);
1235 if (old.npages)
1236 kvm_arch_flush_shadow(kvm);
1237 }
1238 #else /* not defined CONFIG_S390 */
1239 new.user_alloc = user_alloc;
1240 if (user_alloc)
1241 new.userspace_addr = mem->userspace_addr;
1242 #endif /* not defined CONFIG_S390 */
1243
1244 if (!npages)
1245 kvm_arch_flush_shadow(kvm);
1246
1247 spin_lock(&kvm->mmu_lock);
1248 if (mem->slot >= kvm->nmemslots)
1249 kvm->nmemslots = mem->slot + 1;
1250
1251 *memslot = new;
1252 spin_unlock(&kvm->mmu_lock);
1253
1254 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1255 if (r) {
1256 spin_lock(&kvm->mmu_lock);
1257 *memslot = old;
1258 spin_unlock(&kvm->mmu_lock);
1259 goto out_free;
1260 }
1261
1262 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1263 /* Slot deletion case: we have to update the current slot */
1264 spin_lock(&kvm->mmu_lock);
1265 if (!npages)
1266 *memslot = old;
1267 spin_unlock(&kvm->mmu_lock);
1268 #ifdef CONFIG_DMAR
1269 /* map the pages in iommu page table */
1270 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1271 if (r)
1272 goto out;
1273 #endif
1274 return 0;
1275
1276 out_free:
1277 kvm_free_physmem_slot(&new, &old);
1278 out:
1279 return r;
1280
1281 }
1282 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1283
1284 int kvm_set_memory_region(struct kvm *kvm,
1285 struct kvm_userspace_memory_region *mem,
1286 int user_alloc)
1287 {
1288 int r;
1289
1290 down_write(&kvm->slots_lock);
1291 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1292 up_write(&kvm->slots_lock);
1293 return r;
1294 }
1295 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1296
1297 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1298 struct
1299 kvm_userspace_memory_region *mem,
1300 int user_alloc)
1301 {
1302 if (mem->slot >= KVM_MEMORY_SLOTS)
1303 return -EINVAL;
1304 return kvm_set_memory_region(kvm, mem, user_alloc);
1305 }
1306
1307 int kvm_get_dirty_log(struct kvm *kvm,
1308 struct kvm_dirty_log *log, int *is_dirty)
1309 {
1310 struct kvm_memory_slot *memslot;
1311 int r, i;
1312 int n;
1313 unsigned long any = 0;
1314
1315 r = -EINVAL;
1316 if (log->slot >= KVM_MEMORY_SLOTS)
1317 goto out;
1318
1319 memslot = &kvm->memslots[log->slot];
1320 r = -ENOENT;
1321 if (!memslot->dirty_bitmap)
1322 goto out;
1323
1324 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1325
1326 for (i = 0; !any && i < n/sizeof(long); ++i)
1327 any = memslot->dirty_bitmap[i];
1328
1329 r = -EFAULT;
1330 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1331 goto out;
1332
1333 if (any)
1334 *is_dirty = 1;
1335
1336 r = 0;
1337 out:
1338 return r;
1339 }
1340
1341 void kvm_disable_largepages(void)
1342 {
1343 largepages_enabled = false;
1344 }
1345 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1346
1347 int is_error_page(struct page *page)
1348 {
1349 return page == bad_page;
1350 }
1351 EXPORT_SYMBOL_GPL(is_error_page);
1352
1353 int is_error_pfn(pfn_t pfn)
1354 {
1355 return pfn == bad_pfn;
1356 }
1357 EXPORT_SYMBOL_GPL(is_error_pfn);
1358
1359 static inline unsigned long bad_hva(void)
1360 {
1361 return PAGE_OFFSET;
1362 }
1363
1364 int kvm_is_error_hva(unsigned long addr)
1365 {
1366 return addr == bad_hva();
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1369
1370 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1371 {
1372 int i;
1373
1374 for (i = 0; i < kvm->nmemslots; ++i) {
1375 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1376
1377 if (gfn >= memslot->base_gfn
1378 && gfn < memslot->base_gfn + memslot->npages)
1379 return memslot;
1380 }
1381 return NULL;
1382 }
1383 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1384
1385 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1386 {
1387 gfn = unalias_gfn(kvm, gfn);
1388 return gfn_to_memslot_unaliased(kvm, gfn);
1389 }
1390
1391 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1392 {
1393 int i;
1394
1395 gfn = unalias_gfn(kvm, gfn);
1396 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1397 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1398
1399 if (gfn >= memslot->base_gfn
1400 && gfn < memslot->base_gfn + memslot->npages)
1401 return 1;
1402 }
1403 return 0;
1404 }
1405 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1406
1407 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1408 {
1409 struct kvm_memory_slot *slot;
1410
1411 gfn = unalias_gfn(kvm, gfn);
1412 slot = gfn_to_memslot_unaliased(kvm, gfn);
1413 if (!slot)
1414 return bad_hva();
1415 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1416 }
1417 EXPORT_SYMBOL_GPL(gfn_to_hva);
1418
1419 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1420 {
1421 struct page *page[1];
1422 unsigned long addr;
1423 int npages;
1424 pfn_t pfn;
1425
1426 might_sleep();
1427
1428 addr = gfn_to_hva(kvm, gfn);
1429 if (kvm_is_error_hva(addr)) {
1430 get_page(bad_page);
1431 return page_to_pfn(bad_page);
1432 }
1433
1434 npages = get_user_pages_fast(addr, 1, 1, page);
1435
1436 if (unlikely(npages != 1)) {
1437 struct vm_area_struct *vma;
1438
1439 down_read(&current->mm->mmap_sem);
1440 vma = find_vma(current->mm, addr);
1441
1442 if (vma == NULL || addr < vma->vm_start ||
1443 !(vma->vm_flags & VM_PFNMAP)) {
1444 up_read(&current->mm->mmap_sem);
1445 get_page(bad_page);
1446 return page_to_pfn(bad_page);
1447 }
1448
1449 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1450 up_read(&current->mm->mmap_sem);
1451 BUG_ON(!kvm_is_mmio_pfn(pfn));
1452 } else
1453 pfn = page_to_pfn(page[0]);
1454
1455 return pfn;
1456 }
1457
1458 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1459
1460 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1461 {
1462 pfn_t pfn;
1463
1464 pfn = gfn_to_pfn(kvm, gfn);
1465 if (!kvm_is_mmio_pfn(pfn))
1466 return pfn_to_page(pfn);
1467
1468 WARN_ON(kvm_is_mmio_pfn(pfn));
1469
1470 get_page(bad_page);
1471 return bad_page;
1472 }
1473
1474 EXPORT_SYMBOL_GPL(gfn_to_page);
1475
1476 void kvm_release_page_clean(struct page *page)
1477 {
1478 kvm_release_pfn_clean(page_to_pfn(page));
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1481
1482 void kvm_release_pfn_clean(pfn_t pfn)
1483 {
1484 if (!kvm_is_mmio_pfn(pfn))
1485 put_page(pfn_to_page(pfn));
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1488
1489 void kvm_release_page_dirty(struct page *page)
1490 {
1491 kvm_release_pfn_dirty(page_to_pfn(page));
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1494
1495 void kvm_release_pfn_dirty(pfn_t pfn)
1496 {
1497 kvm_set_pfn_dirty(pfn);
1498 kvm_release_pfn_clean(pfn);
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1501
1502 void kvm_set_page_dirty(struct page *page)
1503 {
1504 kvm_set_pfn_dirty(page_to_pfn(page));
1505 }
1506 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1507
1508 void kvm_set_pfn_dirty(pfn_t pfn)
1509 {
1510 if (!kvm_is_mmio_pfn(pfn)) {
1511 struct page *page = pfn_to_page(pfn);
1512 if (!PageReserved(page))
1513 SetPageDirty(page);
1514 }
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1517
1518 void kvm_set_pfn_accessed(pfn_t pfn)
1519 {
1520 if (!kvm_is_mmio_pfn(pfn))
1521 mark_page_accessed(pfn_to_page(pfn));
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1524
1525 void kvm_get_pfn(pfn_t pfn)
1526 {
1527 if (!kvm_is_mmio_pfn(pfn))
1528 get_page(pfn_to_page(pfn));
1529 }
1530 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1531
1532 static int next_segment(unsigned long len, int offset)
1533 {
1534 if (len > PAGE_SIZE - offset)
1535 return PAGE_SIZE - offset;
1536 else
1537 return len;
1538 }
1539
1540 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1541 int len)
1542 {
1543 int r;
1544 unsigned long addr;
1545
1546 addr = gfn_to_hva(kvm, gfn);
1547 if (kvm_is_error_hva(addr))
1548 return -EFAULT;
1549 r = copy_from_user(data, (void __user *)addr + offset, len);
1550 if (r)
1551 return -EFAULT;
1552 return 0;
1553 }
1554 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1555
1556 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1557 {
1558 gfn_t gfn = gpa >> PAGE_SHIFT;
1559 int seg;
1560 int offset = offset_in_page(gpa);
1561 int ret;
1562
1563 while ((seg = next_segment(len, offset)) != 0) {
1564 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1565 if (ret < 0)
1566 return ret;
1567 offset = 0;
1568 len -= seg;
1569 data += seg;
1570 ++gfn;
1571 }
1572 return 0;
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_read_guest);
1575
1576 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1577 unsigned long len)
1578 {
1579 int r;
1580 unsigned long addr;
1581 gfn_t gfn = gpa >> PAGE_SHIFT;
1582 int offset = offset_in_page(gpa);
1583
1584 addr = gfn_to_hva(kvm, gfn);
1585 if (kvm_is_error_hva(addr))
1586 return -EFAULT;
1587 pagefault_disable();
1588 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1589 pagefault_enable();
1590 if (r)
1591 return -EFAULT;
1592 return 0;
1593 }
1594 EXPORT_SYMBOL(kvm_read_guest_atomic);
1595
1596 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1597 int offset, int len)
1598 {
1599 int r;
1600 unsigned long addr;
1601
1602 addr = gfn_to_hva(kvm, gfn);
1603 if (kvm_is_error_hva(addr))
1604 return -EFAULT;
1605 r = copy_to_user((void __user *)addr + offset, data, len);
1606 if (r)
1607 return -EFAULT;
1608 mark_page_dirty(kvm, gfn);
1609 return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1612
1613 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1614 unsigned long len)
1615 {
1616 gfn_t gfn = gpa >> PAGE_SHIFT;
1617 int seg;
1618 int offset = offset_in_page(gpa);
1619 int ret;
1620
1621 while ((seg = next_segment(len, offset)) != 0) {
1622 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1623 if (ret < 0)
1624 return ret;
1625 offset = 0;
1626 len -= seg;
1627 data += seg;
1628 ++gfn;
1629 }
1630 return 0;
1631 }
1632
1633 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1634 {
1635 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1638
1639 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1640 {
1641 gfn_t gfn = gpa >> PAGE_SHIFT;
1642 int seg;
1643 int offset = offset_in_page(gpa);
1644 int ret;
1645
1646 while ((seg = next_segment(len, offset)) != 0) {
1647 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1648 if (ret < 0)
1649 return ret;
1650 offset = 0;
1651 len -= seg;
1652 ++gfn;
1653 }
1654 return 0;
1655 }
1656 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1657
1658 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1659 {
1660 struct kvm_memory_slot *memslot;
1661
1662 gfn = unalias_gfn(kvm, gfn);
1663 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1664 if (memslot && memslot->dirty_bitmap) {
1665 unsigned long rel_gfn = gfn - memslot->base_gfn;
1666
1667 /* avoid RMW */
1668 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1669 set_bit(rel_gfn, memslot->dirty_bitmap);
1670 }
1671 }
1672
1673 /*
1674 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1675 */
1676 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1677 {
1678 DEFINE_WAIT(wait);
1679
1680 for (;;) {
1681 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1682
1683 if (kvm_arch_vcpu_runnable(vcpu)) {
1684 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1685 break;
1686 }
1687 if (kvm_cpu_has_pending_timer(vcpu))
1688 break;
1689 if (signal_pending(current))
1690 break;
1691
1692 vcpu_put(vcpu);
1693 schedule();
1694 vcpu_load(vcpu);
1695 }
1696
1697 finish_wait(&vcpu->wq, &wait);
1698 }
1699
1700 void kvm_resched(struct kvm_vcpu *vcpu)
1701 {
1702 if (!need_resched())
1703 return;
1704 cond_resched();
1705 }
1706 EXPORT_SYMBOL_GPL(kvm_resched);
1707
1708 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1709 {
1710 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1711 struct page *page;
1712
1713 if (vmf->pgoff == 0)
1714 page = virt_to_page(vcpu->run);
1715 #ifdef CONFIG_X86
1716 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1717 page = virt_to_page(vcpu->arch.pio_data);
1718 #endif
1719 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1720 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1721 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1722 #endif
1723 else
1724 return VM_FAULT_SIGBUS;
1725 get_page(page);
1726 vmf->page = page;
1727 return 0;
1728 }
1729
1730 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1731 .fault = kvm_vcpu_fault,
1732 };
1733
1734 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1735 {
1736 vma->vm_ops = &kvm_vcpu_vm_ops;
1737 return 0;
1738 }
1739
1740 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1741 {
1742 struct kvm_vcpu *vcpu = filp->private_data;
1743
1744 kvm_put_kvm(vcpu->kvm);
1745 return 0;
1746 }
1747
1748 static struct file_operations kvm_vcpu_fops = {
1749 .release = kvm_vcpu_release,
1750 .unlocked_ioctl = kvm_vcpu_ioctl,
1751 .compat_ioctl = kvm_vcpu_ioctl,
1752 .mmap = kvm_vcpu_mmap,
1753 };
1754
1755 /*
1756 * Allocates an inode for the vcpu.
1757 */
1758 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1759 {
1760 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1761 }
1762
1763 /*
1764 * Creates some virtual cpus. Good luck creating more than one.
1765 */
1766 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1767 {
1768 int r;
1769 struct kvm_vcpu *vcpu, *v;
1770
1771 vcpu = kvm_arch_vcpu_create(kvm, id);
1772 if (IS_ERR(vcpu))
1773 return PTR_ERR(vcpu);
1774
1775 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1776
1777 r = kvm_arch_vcpu_setup(vcpu);
1778 if (r)
1779 return r;
1780
1781 mutex_lock(&kvm->lock);
1782 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1783 r = -EINVAL;
1784 goto vcpu_destroy;
1785 }
1786
1787 kvm_for_each_vcpu(r, v, kvm)
1788 if (v->vcpu_id == id) {
1789 r = -EEXIST;
1790 goto vcpu_destroy;
1791 }
1792
1793 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1794
1795 /* Now it's all set up, let userspace reach it */
1796 kvm_get_kvm(kvm);
1797 r = create_vcpu_fd(vcpu);
1798 if (r < 0) {
1799 kvm_put_kvm(kvm);
1800 goto vcpu_destroy;
1801 }
1802
1803 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1804 smp_wmb();
1805 atomic_inc(&kvm->online_vcpus);
1806
1807 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1808 if (kvm->bsp_vcpu_id == id)
1809 kvm->bsp_vcpu = vcpu;
1810 #endif
1811 mutex_unlock(&kvm->lock);
1812 return r;
1813
1814 vcpu_destroy:
1815 mutex_unlock(&kvm->lock);
1816 kvm_arch_vcpu_destroy(vcpu);
1817 return r;
1818 }
1819
1820 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1821 {
1822 if (sigset) {
1823 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1824 vcpu->sigset_active = 1;
1825 vcpu->sigset = *sigset;
1826 } else
1827 vcpu->sigset_active = 0;
1828 return 0;
1829 }
1830
1831 #ifdef __KVM_HAVE_MSIX
1832 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1833 struct kvm_assigned_msix_nr *entry_nr)
1834 {
1835 int r = 0;
1836 struct kvm_assigned_dev_kernel *adev;
1837
1838 mutex_lock(&kvm->lock);
1839
1840 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1841 entry_nr->assigned_dev_id);
1842 if (!adev) {
1843 r = -EINVAL;
1844 goto msix_nr_out;
1845 }
1846
1847 if (adev->entries_nr == 0) {
1848 adev->entries_nr = entry_nr->entry_nr;
1849 if (adev->entries_nr == 0 ||
1850 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1851 r = -EINVAL;
1852 goto msix_nr_out;
1853 }
1854
1855 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1856 entry_nr->entry_nr,
1857 GFP_KERNEL);
1858 if (!adev->host_msix_entries) {
1859 r = -ENOMEM;
1860 goto msix_nr_out;
1861 }
1862 adev->guest_msix_entries = kzalloc(
1863 sizeof(struct kvm_guest_msix_entry) *
1864 entry_nr->entry_nr, GFP_KERNEL);
1865 if (!adev->guest_msix_entries) {
1866 kfree(adev->host_msix_entries);
1867 r = -ENOMEM;
1868 goto msix_nr_out;
1869 }
1870 } else /* Not allowed set MSI-X number twice */
1871 r = -EINVAL;
1872 msix_nr_out:
1873 mutex_unlock(&kvm->lock);
1874 return r;
1875 }
1876
1877 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1878 struct kvm_assigned_msix_entry *entry)
1879 {
1880 int r = 0, i;
1881 struct kvm_assigned_dev_kernel *adev;
1882
1883 mutex_lock(&kvm->lock);
1884
1885 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1886 entry->assigned_dev_id);
1887
1888 if (!adev) {
1889 r = -EINVAL;
1890 goto msix_entry_out;
1891 }
1892
1893 for (i = 0; i < adev->entries_nr; i++)
1894 if (adev->guest_msix_entries[i].vector == 0 ||
1895 adev->guest_msix_entries[i].entry == entry->entry) {
1896 adev->guest_msix_entries[i].entry = entry->entry;
1897 adev->guest_msix_entries[i].vector = entry->gsi;
1898 adev->host_msix_entries[i].entry = entry->entry;
1899 break;
1900 }
1901 if (i == adev->entries_nr) {
1902 r = -ENOSPC;
1903 goto msix_entry_out;
1904 }
1905
1906 msix_entry_out:
1907 mutex_unlock(&kvm->lock);
1908
1909 return r;
1910 }
1911 #endif
1912
1913 static long kvm_vcpu_ioctl(struct file *filp,
1914 unsigned int ioctl, unsigned long arg)
1915 {
1916 struct kvm_vcpu *vcpu = filp->private_data;
1917 void __user *argp = (void __user *)arg;
1918 int r;
1919 struct kvm_fpu *fpu = NULL;
1920 struct kvm_sregs *kvm_sregs = NULL;
1921
1922 if (vcpu->kvm->mm != current->mm)
1923 return -EIO;
1924 switch (ioctl) {
1925 case KVM_RUN:
1926 r = -EINVAL;
1927 if (arg)
1928 goto out;
1929 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1930 break;
1931 case KVM_GET_REGS: {
1932 struct kvm_regs *kvm_regs;
1933
1934 r = -ENOMEM;
1935 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1936 if (!kvm_regs)
1937 goto out;
1938 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1939 if (r)
1940 goto out_free1;
1941 r = -EFAULT;
1942 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1943 goto out_free1;
1944 r = 0;
1945 out_free1:
1946 kfree(kvm_regs);
1947 break;
1948 }
1949 case KVM_SET_REGS: {
1950 struct kvm_regs *kvm_regs;
1951
1952 r = -ENOMEM;
1953 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1954 if (!kvm_regs)
1955 goto out;
1956 r = -EFAULT;
1957 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1958 goto out_free2;
1959 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1960 if (r)
1961 goto out_free2;
1962 r = 0;
1963 out_free2:
1964 kfree(kvm_regs);
1965 break;
1966 }
1967 case KVM_GET_SREGS: {
1968 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1969 r = -ENOMEM;
1970 if (!kvm_sregs)
1971 goto out;
1972 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1973 if (r)
1974 goto out;
1975 r = -EFAULT;
1976 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1977 goto out;
1978 r = 0;
1979 break;
1980 }
1981 case KVM_SET_SREGS: {
1982 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1983 r = -ENOMEM;
1984 if (!kvm_sregs)
1985 goto out;
1986 r = -EFAULT;
1987 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1988 goto out;
1989 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1990 if (r)
1991 goto out;
1992 r = 0;
1993 break;
1994 }
1995 case KVM_GET_MP_STATE: {
1996 struct kvm_mp_state mp_state;
1997
1998 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1999 if (r)
2000 goto out;
2001 r = -EFAULT;
2002 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2003 goto out;
2004 r = 0;
2005 break;
2006 }
2007 case KVM_SET_MP_STATE: {
2008 struct kvm_mp_state mp_state;
2009
2010 r = -EFAULT;
2011 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2012 goto out;
2013 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2014 if (r)
2015 goto out;
2016 r = 0;
2017 break;
2018 }
2019 case KVM_TRANSLATE: {
2020 struct kvm_translation tr;
2021
2022 r = -EFAULT;
2023 if (copy_from_user(&tr, argp, sizeof tr))
2024 goto out;
2025 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2026 if (r)
2027 goto out;
2028 r = -EFAULT;
2029 if (copy_to_user(argp, &tr, sizeof tr))
2030 goto out;
2031 r = 0;
2032 break;
2033 }
2034 case KVM_SET_GUEST_DEBUG: {
2035 struct kvm_guest_debug dbg;
2036
2037 r = -EFAULT;
2038 if (copy_from_user(&dbg, argp, sizeof dbg))
2039 goto out;
2040 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2041 if (r)
2042 goto out;
2043 r = 0;
2044 break;
2045 }
2046 case KVM_SET_SIGNAL_MASK: {
2047 struct kvm_signal_mask __user *sigmask_arg = argp;
2048 struct kvm_signal_mask kvm_sigmask;
2049 sigset_t sigset, *p;
2050
2051 p = NULL;
2052 if (argp) {
2053 r = -EFAULT;
2054 if (copy_from_user(&kvm_sigmask, argp,
2055 sizeof kvm_sigmask))
2056 goto out;
2057 r = -EINVAL;
2058 if (kvm_sigmask.len != sizeof sigset)
2059 goto out;
2060 r = -EFAULT;
2061 if (copy_from_user(&sigset, sigmask_arg->sigset,
2062 sizeof sigset))
2063 goto out;
2064 p = &sigset;
2065 }
2066 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2067 break;
2068 }
2069 case KVM_GET_FPU: {
2070 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2071 r = -ENOMEM;
2072 if (!fpu)
2073 goto out;
2074 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2075 if (r)
2076 goto out;
2077 r = -EFAULT;
2078 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2079 goto out;
2080 r = 0;
2081 break;
2082 }
2083 case KVM_SET_FPU: {
2084 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2085 r = -ENOMEM;
2086 if (!fpu)
2087 goto out;
2088 r = -EFAULT;
2089 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2090 goto out;
2091 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2092 if (r)
2093 goto out;
2094 r = 0;
2095 break;
2096 }
2097 default:
2098 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2099 }
2100 out:
2101 kfree(fpu);
2102 kfree(kvm_sregs);
2103 return r;
2104 }
2105
2106 static long kvm_vm_ioctl(struct file *filp,
2107 unsigned int ioctl, unsigned long arg)
2108 {
2109 struct kvm *kvm = filp->private_data;
2110 void __user *argp = (void __user *)arg;
2111 int r;
2112
2113 if (kvm->mm != current->mm)
2114 return -EIO;
2115 switch (ioctl) {
2116 case KVM_CREATE_VCPU:
2117 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2118 if (r < 0)
2119 goto out;
2120 break;
2121 case KVM_SET_USER_MEMORY_REGION: {
2122 struct kvm_userspace_memory_region kvm_userspace_mem;
2123
2124 r = -EFAULT;
2125 if (copy_from_user(&kvm_userspace_mem, argp,
2126 sizeof kvm_userspace_mem))
2127 goto out;
2128
2129 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2130 if (r)
2131 goto out;
2132 break;
2133 }
2134 case KVM_GET_DIRTY_LOG: {
2135 struct kvm_dirty_log log;
2136
2137 r = -EFAULT;
2138 if (copy_from_user(&log, argp, sizeof log))
2139 goto out;
2140 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2141 if (r)
2142 goto out;
2143 break;
2144 }
2145 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2146 case KVM_REGISTER_COALESCED_MMIO: {
2147 struct kvm_coalesced_mmio_zone zone;
2148 r = -EFAULT;
2149 if (copy_from_user(&zone, argp, sizeof zone))
2150 goto out;
2151 r = -ENXIO;
2152 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2153 if (r)
2154 goto out;
2155 r = 0;
2156 break;
2157 }
2158 case KVM_UNREGISTER_COALESCED_MMIO: {
2159 struct kvm_coalesced_mmio_zone zone;
2160 r = -EFAULT;
2161 if (copy_from_user(&zone, argp, sizeof zone))
2162 goto out;
2163 r = -ENXIO;
2164 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2165 if (r)
2166 goto out;
2167 r = 0;
2168 break;
2169 }
2170 #endif
2171 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2172 case KVM_ASSIGN_PCI_DEVICE: {
2173 struct kvm_assigned_pci_dev assigned_dev;
2174
2175 r = -EFAULT;
2176 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2177 goto out;
2178 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2179 if (r)
2180 goto out;
2181 break;
2182 }
2183 case KVM_ASSIGN_IRQ: {
2184 r = -EOPNOTSUPP;
2185 break;
2186 }
2187 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2188 case KVM_ASSIGN_DEV_IRQ: {
2189 struct kvm_assigned_irq assigned_irq;
2190
2191 r = -EFAULT;
2192 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2193 goto out;
2194 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2195 if (r)
2196 goto out;
2197 break;
2198 }
2199 case KVM_DEASSIGN_DEV_IRQ: {
2200 struct kvm_assigned_irq assigned_irq;
2201
2202 r = -EFAULT;
2203 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2204 goto out;
2205 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2206 if (r)
2207 goto out;
2208 break;
2209 }
2210 #endif
2211 #endif
2212 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2213 case KVM_DEASSIGN_PCI_DEVICE: {
2214 struct kvm_assigned_pci_dev assigned_dev;
2215
2216 r = -EFAULT;
2217 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2218 goto out;
2219 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2220 if (r)
2221 goto out;
2222 break;
2223 }
2224 #endif
2225 #ifdef KVM_CAP_IRQ_ROUTING
2226 case KVM_SET_GSI_ROUTING: {
2227 struct kvm_irq_routing routing;
2228 struct kvm_irq_routing __user *urouting;
2229 struct kvm_irq_routing_entry *entries;
2230
2231 r = -EFAULT;
2232 if (copy_from_user(&routing, argp, sizeof(routing)))
2233 goto out;
2234 r = -EINVAL;
2235 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2236 goto out;
2237 if (routing.flags)
2238 goto out;
2239 r = -ENOMEM;
2240 entries = vmalloc(routing.nr * sizeof(*entries));
2241 if (!entries)
2242 goto out;
2243 r = -EFAULT;
2244 urouting = argp;
2245 if (copy_from_user(entries, urouting->entries,
2246 routing.nr * sizeof(*entries)))
2247 goto out_free_irq_routing;
2248 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2249 routing.flags);
2250 out_free_irq_routing:
2251 vfree(entries);
2252 break;
2253 }
2254 #endif /* KVM_CAP_IRQ_ROUTING */
2255 #ifdef __KVM_HAVE_MSIX
2256 case KVM_ASSIGN_SET_MSIX_NR: {
2257 struct kvm_assigned_msix_nr entry_nr;
2258 r = -EFAULT;
2259 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2260 goto out;
2261 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2262 if (r)
2263 goto out;
2264 break;
2265 }
2266 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2267 struct kvm_assigned_msix_entry entry;
2268 r = -EFAULT;
2269 if (copy_from_user(&entry, argp, sizeof entry))
2270 goto out;
2271 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2272 if (r)
2273 goto out;
2274 break;
2275 }
2276 #endif
2277 case KVM_IRQFD: {
2278 struct kvm_irqfd data;
2279
2280 r = -EFAULT;
2281 if (copy_from_user(&data, argp, sizeof data))
2282 goto out;
2283 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2284 break;
2285 }
2286 case KVM_IOEVENTFD: {
2287 struct kvm_ioeventfd data;
2288
2289 r = -EFAULT;
2290 if (copy_from_user(&data, argp, sizeof data))
2291 goto out;
2292 r = kvm_ioeventfd(kvm, &data);
2293 break;
2294 }
2295 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2296 case KVM_SET_BOOT_CPU_ID:
2297 r = 0;
2298 mutex_lock(&kvm->lock);
2299 if (atomic_read(&kvm->online_vcpus) != 0)
2300 r = -EBUSY;
2301 else
2302 kvm->bsp_vcpu_id = arg;
2303 mutex_unlock(&kvm->lock);
2304 break;
2305 #endif
2306 default:
2307 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2308 }
2309 out:
2310 return r;
2311 }
2312
2313 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2314 {
2315 struct page *page[1];
2316 unsigned long addr;
2317 int npages;
2318 gfn_t gfn = vmf->pgoff;
2319 struct kvm *kvm = vma->vm_file->private_data;
2320
2321 addr = gfn_to_hva(kvm, gfn);
2322 if (kvm_is_error_hva(addr))
2323 return VM_FAULT_SIGBUS;
2324
2325 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2326 NULL);
2327 if (unlikely(npages != 1))
2328 return VM_FAULT_SIGBUS;
2329
2330 vmf->page = page[0];
2331 return 0;
2332 }
2333
2334 static const struct vm_operations_struct kvm_vm_vm_ops = {
2335 .fault = kvm_vm_fault,
2336 };
2337
2338 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2339 {
2340 vma->vm_ops = &kvm_vm_vm_ops;
2341 return 0;
2342 }
2343
2344 static struct file_operations kvm_vm_fops = {
2345 .release = kvm_vm_release,
2346 .unlocked_ioctl = kvm_vm_ioctl,
2347 .compat_ioctl = kvm_vm_ioctl,
2348 .mmap = kvm_vm_mmap,
2349 };
2350
2351 static int kvm_dev_ioctl_create_vm(void)
2352 {
2353 int fd;
2354 struct kvm *kvm;
2355
2356 kvm = kvm_create_vm();
2357 if (IS_ERR(kvm))
2358 return PTR_ERR(kvm);
2359 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2360 if (fd < 0)
2361 kvm_put_kvm(kvm);
2362
2363 return fd;
2364 }
2365
2366 static long kvm_dev_ioctl_check_extension_generic(long arg)
2367 {
2368 switch (arg) {
2369 case KVM_CAP_USER_MEMORY:
2370 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2371 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2372 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2373 case KVM_CAP_SET_BOOT_CPU_ID:
2374 #endif
2375 return 1;
2376 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2377 case KVM_CAP_IRQ_ROUTING:
2378 return KVM_MAX_IRQ_ROUTES;
2379 #endif
2380 default:
2381 break;
2382 }
2383 return kvm_dev_ioctl_check_extension(arg);
2384 }
2385
2386 static long kvm_dev_ioctl(struct file *filp,
2387 unsigned int ioctl, unsigned long arg)
2388 {
2389 long r = -EINVAL;
2390
2391 switch (ioctl) {
2392 case KVM_GET_API_VERSION:
2393 r = -EINVAL;
2394 if (arg)
2395 goto out;
2396 r = KVM_API_VERSION;
2397 break;
2398 case KVM_CREATE_VM:
2399 r = -EINVAL;
2400 if (arg)
2401 goto out;
2402 r = kvm_dev_ioctl_create_vm();
2403 break;
2404 case KVM_CHECK_EXTENSION:
2405 r = kvm_dev_ioctl_check_extension_generic(arg);
2406 break;
2407 case KVM_GET_VCPU_MMAP_SIZE:
2408 r = -EINVAL;
2409 if (arg)
2410 goto out;
2411 r = PAGE_SIZE; /* struct kvm_run */
2412 #ifdef CONFIG_X86
2413 r += PAGE_SIZE; /* pio data page */
2414 #endif
2415 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2416 r += PAGE_SIZE; /* coalesced mmio ring page */
2417 #endif
2418 break;
2419 case KVM_TRACE_ENABLE:
2420 case KVM_TRACE_PAUSE:
2421 case KVM_TRACE_DISABLE:
2422 r = -EOPNOTSUPP;
2423 break;
2424 default:
2425 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2426 }
2427 out:
2428 return r;
2429 }
2430
2431 static struct file_operations kvm_chardev_ops = {
2432 .unlocked_ioctl = kvm_dev_ioctl,
2433 .compat_ioctl = kvm_dev_ioctl,
2434 };
2435
2436 static struct miscdevice kvm_dev = {
2437 KVM_MINOR,
2438 "kvm",
2439 &kvm_chardev_ops,
2440 };
2441
2442 static void hardware_enable(void *junk)
2443 {
2444 int cpu = raw_smp_processor_id();
2445
2446 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2447 return;
2448 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2449 kvm_arch_hardware_enable(NULL);
2450 }
2451
2452 static void hardware_disable(void *junk)
2453 {
2454 int cpu = raw_smp_processor_id();
2455
2456 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2457 return;
2458 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2459 kvm_arch_hardware_disable(NULL);
2460 }
2461
2462 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2463 void *v)
2464 {
2465 int cpu = (long)v;
2466
2467 val &= ~CPU_TASKS_FROZEN;
2468 switch (val) {
2469 case CPU_DYING:
2470 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2471 cpu);
2472 hardware_disable(NULL);
2473 break;
2474 case CPU_UP_CANCELED:
2475 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2476 cpu);
2477 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2478 break;
2479 case CPU_ONLINE:
2480 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2481 cpu);
2482 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2483 break;
2484 }
2485 return NOTIFY_OK;
2486 }
2487
2488
2489 asmlinkage void kvm_handle_fault_on_reboot(void)
2490 {
2491 if (kvm_rebooting)
2492 /* spin while reset goes on */
2493 while (true)
2494 ;
2495 /* Fault while not rebooting. We want the trace. */
2496 BUG();
2497 }
2498 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2499
2500 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2501 void *v)
2502 {
2503 /*
2504 * Some (well, at least mine) BIOSes hang on reboot if
2505 * in vmx root mode.
2506 *
2507 * And Intel TXT required VMX off for all cpu when system shutdown.
2508 */
2509 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2510 kvm_rebooting = true;
2511 on_each_cpu(hardware_disable, NULL, 1);
2512 return NOTIFY_OK;
2513 }
2514
2515 static struct notifier_block kvm_reboot_notifier = {
2516 .notifier_call = kvm_reboot,
2517 .priority = 0,
2518 };
2519
2520 void kvm_io_bus_init(struct kvm_io_bus *bus)
2521 {
2522 memset(bus, 0, sizeof(*bus));
2523 }
2524
2525 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2526 {
2527 int i;
2528
2529 for (i = 0; i < bus->dev_count; i++) {
2530 struct kvm_io_device *pos = bus->devs[i];
2531
2532 kvm_iodevice_destructor(pos);
2533 }
2534 }
2535
2536 /* kvm_io_bus_write - called under kvm->slots_lock */
2537 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2538 int len, const void *val)
2539 {
2540 int i;
2541 for (i = 0; i < bus->dev_count; i++)
2542 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2543 return 0;
2544 return -EOPNOTSUPP;
2545 }
2546
2547 /* kvm_io_bus_read - called under kvm->slots_lock */
2548 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
2549 {
2550 int i;
2551 for (i = 0; i < bus->dev_count; i++)
2552 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2553 return 0;
2554 return -EOPNOTSUPP;
2555 }
2556
2557 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
2558 struct kvm_io_device *dev)
2559 {
2560 int ret;
2561
2562 down_write(&kvm->slots_lock);
2563 ret = __kvm_io_bus_register_dev(bus, dev);
2564 up_write(&kvm->slots_lock);
2565
2566 return ret;
2567 }
2568
2569 /* An unlocked version. Caller must have write lock on slots_lock. */
2570 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2571 struct kvm_io_device *dev)
2572 {
2573 if (bus->dev_count > NR_IOBUS_DEVS-1)
2574 return -ENOSPC;
2575
2576 bus->devs[bus->dev_count++] = dev;
2577
2578 return 0;
2579 }
2580
2581 void kvm_io_bus_unregister_dev(struct kvm *kvm,
2582 struct kvm_io_bus *bus,
2583 struct kvm_io_device *dev)
2584 {
2585 down_write(&kvm->slots_lock);
2586 __kvm_io_bus_unregister_dev(bus, dev);
2587 up_write(&kvm->slots_lock);
2588 }
2589
2590 /* An unlocked version. Caller must have write lock on slots_lock. */
2591 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2592 struct kvm_io_device *dev)
2593 {
2594 int i;
2595
2596 for (i = 0; i < bus->dev_count; i++)
2597 if (bus->devs[i] == dev) {
2598 bus->devs[i] = bus->devs[--bus->dev_count];
2599 break;
2600 }
2601 }
2602
2603 static struct notifier_block kvm_cpu_notifier = {
2604 .notifier_call = kvm_cpu_hotplug,
2605 .priority = 20, /* must be > scheduler priority */
2606 };
2607
2608 static int vm_stat_get(void *_offset, u64 *val)
2609 {
2610 unsigned offset = (long)_offset;
2611 struct kvm *kvm;
2612
2613 *val = 0;
2614 spin_lock(&kvm_lock);
2615 list_for_each_entry(kvm, &vm_list, vm_list)
2616 *val += *(u32 *)((void *)kvm + offset);
2617 spin_unlock(&kvm_lock);
2618 return 0;
2619 }
2620
2621 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2622
2623 static int vcpu_stat_get(void *_offset, u64 *val)
2624 {
2625 unsigned offset = (long)_offset;
2626 struct kvm *kvm;
2627 struct kvm_vcpu *vcpu;
2628 int i;
2629
2630 *val = 0;
2631 spin_lock(&kvm_lock);
2632 list_for_each_entry(kvm, &vm_list, vm_list)
2633 kvm_for_each_vcpu(i, vcpu, kvm)
2634 *val += *(u32 *)((void *)vcpu + offset);
2635
2636 spin_unlock(&kvm_lock);
2637 return 0;
2638 }
2639
2640 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2641
2642 static const struct file_operations *stat_fops[] = {
2643 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2644 [KVM_STAT_VM] = &vm_stat_fops,
2645 };
2646
2647 static void kvm_init_debug(void)
2648 {
2649 struct kvm_stats_debugfs_item *p;
2650
2651 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2652 for (p = debugfs_entries; p->name; ++p)
2653 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2654 (void *)(long)p->offset,
2655 stat_fops[p->kind]);
2656 }
2657
2658 static void kvm_exit_debug(void)
2659 {
2660 struct kvm_stats_debugfs_item *p;
2661
2662 for (p = debugfs_entries; p->name; ++p)
2663 debugfs_remove(p->dentry);
2664 debugfs_remove(kvm_debugfs_dir);
2665 }
2666
2667 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2668 {
2669 hardware_disable(NULL);
2670 return 0;
2671 }
2672
2673 static int kvm_resume(struct sys_device *dev)
2674 {
2675 hardware_enable(NULL);
2676 return 0;
2677 }
2678
2679 static struct sysdev_class kvm_sysdev_class = {
2680 .name = "kvm",
2681 .suspend = kvm_suspend,
2682 .resume = kvm_resume,
2683 };
2684
2685 static struct sys_device kvm_sysdev = {
2686 .id = 0,
2687 .cls = &kvm_sysdev_class,
2688 };
2689
2690 struct page *bad_page;
2691 pfn_t bad_pfn;
2692
2693 static inline
2694 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2695 {
2696 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2697 }
2698
2699 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2700 {
2701 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2702
2703 kvm_arch_vcpu_load(vcpu, cpu);
2704 }
2705
2706 static void kvm_sched_out(struct preempt_notifier *pn,
2707 struct task_struct *next)
2708 {
2709 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2710
2711 kvm_arch_vcpu_put(vcpu);
2712 }
2713
2714 int kvm_init(void *opaque, unsigned int vcpu_size,
2715 struct module *module)
2716 {
2717 int r;
2718 int cpu;
2719
2720 r = kvm_arch_init(opaque);
2721 if (r)
2722 goto out_fail;
2723
2724 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2725
2726 if (bad_page == NULL) {
2727 r = -ENOMEM;
2728 goto out;
2729 }
2730
2731 bad_pfn = page_to_pfn(bad_page);
2732
2733 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2734 r = -ENOMEM;
2735 goto out_free_0;
2736 }
2737
2738 r = kvm_arch_hardware_setup();
2739 if (r < 0)
2740 goto out_free_0a;
2741
2742 for_each_online_cpu(cpu) {
2743 smp_call_function_single(cpu,
2744 kvm_arch_check_processor_compat,
2745 &r, 1);
2746 if (r < 0)
2747 goto out_free_1;
2748 }
2749
2750 on_each_cpu(hardware_enable, NULL, 1);
2751 r = register_cpu_notifier(&kvm_cpu_notifier);
2752 if (r)
2753 goto out_free_2;
2754 register_reboot_notifier(&kvm_reboot_notifier);
2755
2756 r = sysdev_class_register(&kvm_sysdev_class);
2757 if (r)
2758 goto out_free_3;
2759
2760 r = sysdev_register(&kvm_sysdev);
2761 if (r)
2762 goto out_free_4;
2763
2764 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2765 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2766 __alignof__(struct kvm_vcpu),
2767 0, NULL);
2768 if (!kvm_vcpu_cache) {
2769 r = -ENOMEM;
2770 goto out_free_5;
2771 }
2772
2773 kvm_chardev_ops.owner = module;
2774 kvm_vm_fops.owner = module;
2775 kvm_vcpu_fops.owner = module;
2776
2777 r = misc_register(&kvm_dev);
2778 if (r) {
2779 printk(KERN_ERR "kvm: misc device register failed\n");
2780 goto out_free;
2781 }
2782
2783 kvm_preempt_ops.sched_in = kvm_sched_in;
2784 kvm_preempt_ops.sched_out = kvm_sched_out;
2785
2786 kvm_init_debug();
2787
2788 return 0;
2789
2790 out_free:
2791 kmem_cache_destroy(kvm_vcpu_cache);
2792 out_free_5:
2793 sysdev_unregister(&kvm_sysdev);
2794 out_free_4:
2795 sysdev_class_unregister(&kvm_sysdev_class);
2796 out_free_3:
2797 unregister_reboot_notifier(&kvm_reboot_notifier);
2798 unregister_cpu_notifier(&kvm_cpu_notifier);
2799 out_free_2:
2800 on_each_cpu(hardware_disable, NULL, 1);
2801 out_free_1:
2802 kvm_arch_hardware_unsetup();
2803 out_free_0a:
2804 free_cpumask_var(cpus_hardware_enabled);
2805 out_free_0:
2806 __free_page(bad_page);
2807 out:
2808 kvm_arch_exit();
2809 out_fail:
2810 return r;
2811 }
2812 EXPORT_SYMBOL_GPL(kvm_init);
2813
2814 void kvm_exit(void)
2815 {
2816 tracepoint_synchronize_unregister();
2817 kvm_exit_debug();
2818 misc_deregister(&kvm_dev);
2819 kmem_cache_destroy(kvm_vcpu_cache);
2820 sysdev_unregister(&kvm_sysdev);
2821 sysdev_class_unregister(&kvm_sysdev_class);
2822 unregister_reboot_notifier(&kvm_reboot_notifier);
2823 unregister_cpu_notifier(&kvm_cpu_notifier);
2824 on_each_cpu(hardware_disable, NULL, 1);
2825 kvm_arch_hardware_unsetup();
2826 kvm_arch_exit();
2827 free_cpumask_var(cpus_hardware_enabled);
2828 __free_page(bad_page);
2829 }
2830 EXPORT_SYMBOL_GPL(kvm_exit);
kernels