xref: /openbmc/linux/virt/kvm/kvm_main.c (revision fd589a8f)
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 	if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
742 		cpumask_clear(cpus);
743 
744 	spin_lock(&kvm->requests_lock);
745 	me = smp_processor_id();
746 	kvm_for_each_vcpu(i, vcpu, kvm) {
747 		if (test_and_set_bit(req, &vcpu->requests))
748 			continue;
749 		cpu = vcpu->cpu;
750 		if (cpus != NULL && cpu != -1 && cpu != me)
751 			cpumask_set_cpu(cpu, cpus);
752 	}
753 	if (unlikely(cpus == NULL))
754 		smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
755 	else if (!cpumask_empty(cpus))
756 		smp_call_function_many(cpus, ack_flush, NULL, 1);
757 	else
758 		called = false;
759 	spin_unlock(&kvm->requests_lock);
760 	free_cpumask_var(cpus);
761 	return called;
762 }
763 
764 void kvm_flush_remote_tlbs(struct kvm *kvm)
765 {
766 	if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
767 		++kvm->stat.remote_tlb_flush;
768 }
769 
770 void kvm_reload_remote_mmus(struct kvm *kvm)
771 {
772 	make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
773 }
774 
775 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
776 {
777 	struct page *page;
778 	int r;
779 
780 	mutex_init(&vcpu->mutex);
781 	vcpu->cpu = -1;
782 	vcpu->kvm = kvm;
783 	vcpu->vcpu_id = id;
784 	init_waitqueue_head(&vcpu->wq);
785 
786 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
787 	if (!page) {
788 		r = -ENOMEM;
789 		goto fail;
790 	}
791 	vcpu->run = page_address(page);
792 
793 	r = kvm_arch_vcpu_init(vcpu);
794 	if (r < 0)
795 		goto fail_free_run;
796 	return 0;
797 
798 fail_free_run:
799 	free_page((unsigned long)vcpu->run);
800 fail:
801 	return r;
802 }
803 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
804 
805 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
806 {
807 	kvm_arch_vcpu_uninit(vcpu);
808 	free_page((unsigned long)vcpu->run);
809 }
810 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
811 
812 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
813 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
814 {
815 	return container_of(mn, struct kvm, mmu_notifier);
816 }
817 
818 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
819 					     struct mm_struct *mm,
820 					     unsigned long address)
821 {
822 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
823 	int need_tlb_flush;
824 
825 	/*
826 	 * When ->invalidate_page runs, the linux pte has been zapped
827 	 * already but the page is still allocated until
828 	 * ->invalidate_page returns. So if we increase the sequence
829 	 * here the kvm page fault will notice if the spte can't be
830 	 * established because the page is going to be freed. If
831 	 * instead the kvm page fault establishes the spte before
832 	 * ->invalidate_page runs, kvm_unmap_hva will release it
833 	 * before returning.
834 	 *
835 	 * The sequence increase only need to be seen at spin_unlock
836 	 * time, and not at spin_lock time.
837 	 *
838 	 * Increasing the sequence after the spin_unlock would be
839 	 * unsafe because the kvm page fault could then establish the
840 	 * pte after kvm_unmap_hva returned, without noticing the page
841 	 * is going to be freed.
842 	 */
843 	spin_lock(&kvm->mmu_lock);
844 	kvm->mmu_notifier_seq++;
845 	need_tlb_flush = kvm_unmap_hva(kvm, address);
846 	spin_unlock(&kvm->mmu_lock);
847 
848 	/* we've to flush the tlb before the pages can be freed */
849 	if (need_tlb_flush)
850 		kvm_flush_remote_tlbs(kvm);
851 
852 }
853 
854 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
855 						    struct mm_struct *mm,
856 						    unsigned long start,
857 						    unsigned long end)
858 {
859 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
860 	int need_tlb_flush = 0;
861 
862 	spin_lock(&kvm->mmu_lock);
863 	/*
864 	 * The count increase must become visible at unlock time as no
865 	 * spte can be established without taking the mmu_lock and
866 	 * count is also read inside the mmu_lock critical section.
867 	 */
868 	kvm->mmu_notifier_count++;
869 	for (; start < end; start += PAGE_SIZE)
870 		need_tlb_flush |= kvm_unmap_hva(kvm, start);
871 	spin_unlock(&kvm->mmu_lock);
872 
873 	/* we've to flush the tlb before the pages can be freed */
874 	if (need_tlb_flush)
875 		kvm_flush_remote_tlbs(kvm);
876 }
877 
878 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
879 						  struct mm_struct *mm,
880 						  unsigned long start,
881 						  unsigned long end)
882 {
883 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
884 
885 	spin_lock(&kvm->mmu_lock);
886 	/*
887 	 * This sequence increase will notify the kvm page fault that
888 	 * the page that is going to be mapped in the spte could have
889 	 * been freed.
890 	 */
891 	kvm->mmu_notifier_seq++;
892 	/*
893 	 * The above sequence increase must be visible before the
894 	 * below count decrease but both values are read by the kvm
895 	 * page fault under mmu_lock spinlock so we don't need to add
896 	 * a smb_wmb() here in between the two.
897 	 */
898 	kvm->mmu_notifier_count--;
899 	spin_unlock(&kvm->mmu_lock);
900 
901 	BUG_ON(kvm->mmu_notifier_count < 0);
902 }
903 
904 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
905 					      struct mm_struct *mm,
906 					      unsigned long address)
907 {
908 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
909 	int young;
910 
911 	spin_lock(&kvm->mmu_lock);
912 	young = kvm_age_hva(kvm, address);
913 	spin_unlock(&kvm->mmu_lock);
914 
915 	if (young)
916 		kvm_flush_remote_tlbs(kvm);
917 
918 	return young;
919 }
920 
921 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
922 				     struct mm_struct *mm)
923 {
924 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
925 	kvm_arch_flush_shadow(kvm);
926 }
927 
928 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
929 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
930 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
931 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
932 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
933 	.release		= kvm_mmu_notifier_release,
934 };
935 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
936 
937 static struct kvm *kvm_create_vm(void)
938 {
939 	struct kvm *kvm = kvm_arch_create_vm();
940 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
941 	struct page *page;
942 #endif
943 
944 	if (IS_ERR(kvm))
945 		goto out;
946 #ifdef CONFIG_HAVE_KVM_IRQCHIP
947 	INIT_LIST_HEAD(&kvm->irq_routing);
948 	INIT_HLIST_HEAD(&kvm->mask_notifier_list);
949 #endif
950 
951 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
952 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
953 	if (!page) {
954 		kfree(kvm);
955 		return ERR_PTR(-ENOMEM);
956 	}
957 	kvm->coalesced_mmio_ring =
958 			(struct kvm_coalesced_mmio_ring *)page_address(page);
959 #endif
960 
961 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
962 	{
963 		int err;
964 		kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
965 		err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
966 		if (err) {
967 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
968 			put_page(page);
969 #endif
970 			kfree(kvm);
971 			return ERR_PTR(err);
972 		}
973 	}
974 #endif
975 
976 	kvm->mm = current->mm;
977 	atomic_inc(&kvm->mm->mm_count);
978 	spin_lock_init(&kvm->mmu_lock);
979 	spin_lock_init(&kvm->requests_lock);
980 	kvm_io_bus_init(&kvm->pio_bus);
981 	kvm_eventfd_init(kvm);
982 	mutex_init(&kvm->lock);
983 	mutex_init(&kvm->irq_lock);
984 	kvm_io_bus_init(&kvm->mmio_bus);
985 	init_rwsem(&kvm->slots_lock);
986 	atomic_set(&kvm->users_count, 1);
987 	spin_lock(&kvm_lock);
988 	list_add(&kvm->vm_list, &vm_list);
989 	spin_unlock(&kvm_lock);
990 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
991 	kvm_coalesced_mmio_init(kvm);
992 #endif
993 out:
994 	return kvm;
995 }
996 
997 /*
998  * Free any memory in @free but not in @dont.
999  */
1000 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1001 				  struct kvm_memory_slot *dont)
1002 {
1003 	int i;
1004 
1005 	if (!dont || free->rmap != dont->rmap)
1006 		vfree(free->rmap);
1007 
1008 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1009 		vfree(free->dirty_bitmap);
1010 
1011 
1012 	for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1013 		if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1014 			vfree(free->lpage_info[i]);
1015 			free->lpage_info[i] = NULL;
1016 		}
1017 	}
1018 
1019 	free->npages = 0;
1020 	free->dirty_bitmap = NULL;
1021 	free->rmap = NULL;
1022 }
1023 
1024 void kvm_free_physmem(struct kvm *kvm)
1025 {
1026 	int i;
1027 
1028 	for (i = 0; i < kvm->nmemslots; ++i)
1029 		kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1030 }
1031 
1032 static void kvm_destroy_vm(struct kvm *kvm)
1033 {
1034 	struct mm_struct *mm = kvm->mm;
1035 
1036 	kvm_arch_sync_events(kvm);
1037 	spin_lock(&kvm_lock);
1038 	list_del(&kvm->vm_list);
1039 	spin_unlock(&kvm_lock);
1040 	kvm_free_irq_routing(kvm);
1041 	kvm_io_bus_destroy(&kvm->pio_bus);
1042 	kvm_io_bus_destroy(&kvm->mmio_bus);
1043 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1044 	if (kvm->coalesced_mmio_ring != NULL)
1045 		free_page((unsigned long)kvm->coalesced_mmio_ring);
1046 #endif
1047 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1048 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1049 #else
1050 	kvm_arch_flush_shadow(kvm);
1051 #endif
1052 	kvm_arch_destroy_vm(kvm);
1053 	mmdrop(mm);
1054 }
1055 
1056 void kvm_get_kvm(struct kvm *kvm)
1057 {
1058 	atomic_inc(&kvm->users_count);
1059 }
1060 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1061 
1062 void kvm_put_kvm(struct kvm *kvm)
1063 {
1064 	if (atomic_dec_and_test(&kvm->users_count))
1065 		kvm_destroy_vm(kvm);
1066 }
1067 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1068 
1069 
1070 static int kvm_vm_release(struct inode *inode, struct file *filp)
1071 {
1072 	struct kvm *kvm = filp->private_data;
1073 
1074 	kvm_irqfd_release(kvm);
1075 
1076 	kvm_put_kvm(kvm);
1077 	return 0;
1078 }
1079 
1080 /*
1081  * Allocate some memory and give it an address in the guest physical address
1082  * space.
1083  *
1084  * Discontiguous memory is allowed, mostly for framebuffers.
1085  *
1086  * Must be called holding mmap_sem for write.
1087  */
1088 int __kvm_set_memory_region(struct kvm *kvm,
1089 			    struct kvm_userspace_memory_region *mem,
1090 			    int user_alloc)
1091 {
1092 	int r;
1093 	gfn_t base_gfn;
1094 	unsigned long npages;
1095 	unsigned long i;
1096 	struct kvm_memory_slot *memslot;
1097 	struct kvm_memory_slot old, new;
1098 
1099 	r = -EINVAL;
1100 	/* General sanity checks */
1101 	if (mem->memory_size & (PAGE_SIZE - 1))
1102 		goto out;
1103 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1104 		goto out;
1105 	if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1106 		goto out;
1107 	if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1108 		goto out;
1109 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1110 		goto out;
1111 
1112 	memslot = &kvm->memslots[mem->slot];
1113 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1114 	npages = mem->memory_size >> PAGE_SHIFT;
1115 
1116 	if (!npages)
1117 		mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1118 
1119 	new = old = *memslot;
1120 
1121 	new.base_gfn = base_gfn;
1122 	new.npages = npages;
1123 	new.flags = mem->flags;
1124 
1125 	/* Disallow changing a memory slot's size. */
1126 	r = -EINVAL;
1127 	if (npages && old.npages && npages != old.npages)
1128 		goto out_free;
1129 
1130 	/* Check for overlaps */
1131 	r = -EEXIST;
1132 	for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1133 		struct kvm_memory_slot *s = &kvm->memslots[i];
1134 
1135 		if (s == memslot || !s->npages)
1136 			continue;
1137 		if (!((base_gfn + npages <= s->base_gfn) ||
1138 		      (base_gfn >= s->base_gfn + s->npages)))
1139 			goto out_free;
1140 	}
1141 
1142 	/* Free page dirty bitmap if unneeded */
1143 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1144 		new.dirty_bitmap = NULL;
1145 
1146 	r = -ENOMEM;
1147 
1148 	/* Allocate if a slot is being created */
1149 #ifndef CONFIG_S390
1150 	if (npages && !new.rmap) {
1151 		new.rmap = vmalloc(npages * sizeof(struct page *));
1152 
1153 		if (!new.rmap)
1154 			goto out_free;
1155 
1156 		memset(new.rmap, 0, npages * sizeof(*new.rmap));
1157 
1158 		new.user_alloc = user_alloc;
1159 		/*
1160 		 * hva_to_rmmap() serialzies with the mmu_lock and to be
1161 		 * safe it has to ignore memslots with !user_alloc &&
1162 		 * !userspace_addr.
1163 		 */
1164 		if (user_alloc)
1165 			new.userspace_addr = mem->userspace_addr;
1166 		else
1167 			new.userspace_addr = 0;
1168 	}
1169 	if (!npages)
1170 		goto skip_lpage;
1171 
1172 	for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1173 		unsigned long ugfn;
1174 		unsigned long j;
1175 		int lpages;
1176 		int level = i + 2;
1177 
1178 		/* Avoid unused variable warning if no large pages */
1179 		(void)level;
1180 
1181 		if (new.lpage_info[i])
1182 			continue;
1183 
1184 		lpages = 1 + (base_gfn + npages - 1) /
1185 			     KVM_PAGES_PER_HPAGE(level);
1186 		lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1187 
1188 		new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1189 
1190 		if (!new.lpage_info[i])
1191 			goto out_free;
1192 
1193 		memset(new.lpage_info[i], 0,
1194 		       lpages * sizeof(*new.lpage_info[i]));
1195 
1196 		if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1197 			new.lpage_info[i][0].write_count = 1;
1198 		if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1199 			new.lpage_info[i][lpages - 1].write_count = 1;
1200 		ugfn = new.userspace_addr >> PAGE_SHIFT;
1201 		/*
1202 		 * If the gfn and userspace address are not aligned wrt each
1203 		 * other, or if explicitly asked to, disable large page
1204 		 * support for this slot
1205 		 */
1206 		if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1207 		    !largepages_enabled)
1208 			for (j = 0; j < lpages; ++j)
1209 				new.lpage_info[i][j].write_count = 1;
1210 	}
1211 
1212 skip_lpage:
1213 
1214 	/* Allocate page dirty bitmap if needed */
1215 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1216 		unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1217 
1218 		new.dirty_bitmap = vmalloc(dirty_bytes);
1219 		if (!new.dirty_bitmap)
1220 			goto out_free;
1221 		memset(new.dirty_bitmap, 0, dirty_bytes);
1222 		if (old.npages)
1223 			kvm_arch_flush_shadow(kvm);
1224 	}
1225 #else  /* not defined CONFIG_S390 */
1226 	new.user_alloc = user_alloc;
1227 	if (user_alloc)
1228 		new.userspace_addr = mem->userspace_addr;
1229 #endif /* not defined CONFIG_S390 */
1230 
1231 	if (!npages)
1232 		kvm_arch_flush_shadow(kvm);
1233 
1234 	spin_lock(&kvm->mmu_lock);
1235 	if (mem->slot >= kvm->nmemslots)
1236 		kvm->nmemslots = mem->slot + 1;
1237 
1238 	*memslot = new;
1239 	spin_unlock(&kvm->mmu_lock);
1240 
1241 	r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1242 	if (r) {
1243 		spin_lock(&kvm->mmu_lock);
1244 		*memslot = old;
1245 		spin_unlock(&kvm->mmu_lock);
1246 		goto out_free;
1247 	}
1248 
1249 	kvm_free_physmem_slot(&old, npages ? &new : NULL);
1250 	/* Slot deletion case: we have to update the current slot */
1251 	spin_lock(&kvm->mmu_lock);
1252 	if (!npages)
1253 		*memslot = old;
1254 	spin_unlock(&kvm->mmu_lock);
1255 #ifdef CONFIG_DMAR
1256 	/* map the pages in iommu page table */
1257 	r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1258 	if (r)
1259 		goto out;
1260 #endif
1261 	return 0;
1262 
1263 out_free:
1264 	kvm_free_physmem_slot(&new, &old);
1265 out:
1266 	return r;
1267 
1268 }
1269 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1270 
1271 int kvm_set_memory_region(struct kvm *kvm,
1272 			  struct kvm_userspace_memory_region *mem,
1273 			  int user_alloc)
1274 {
1275 	int r;
1276 
1277 	down_write(&kvm->slots_lock);
1278 	r = __kvm_set_memory_region(kvm, mem, user_alloc);
1279 	up_write(&kvm->slots_lock);
1280 	return r;
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1283 
1284 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1285 				   struct
1286 				   kvm_userspace_memory_region *mem,
1287 				   int user_alloc)
1288 {
1289 	if (mem->slot >= KVM_MEMORY_SLOTS)
1290 		return -EINVAL;
1291 	return kvm_set_memory_region(kvm, mem, user_alloc);
1292 }
1293 
1294 int kvm_get_dirty_log(struct kvm *kvm,
1295 			struct kvm_dirty_log *log, int *is_dirty)
1296 {
1297 	struct kvm_memory_slot *memslot;
1298 	int r, i;
1299 	int n;
1300 	unsigned long any = 0;
1301 
1302 	r = -EINVAL;
1303 	if (log->slot >= KVM_MEMORY_SLOTS)
1304 		goto out;
1305 
1306 	memslot = &kvm->memslots[log->slot];
1307 	r = -ENOENT;
1308 	if (!memslot->dirty_bitmap)
1309 		goto out;
1310 
1311 	n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1312 
1313 	for (i = 0; !any && i < n/sizeof(long); ++i)
1314 		any = memslot->dirty_bitmap[i];
1315 
1316 	r = -EFAULT;
1317 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1318 		goto out;
1319 
1320 	if (any)
1321 		*is_dirty = 1;
1322 
1323 	r = 0;
1324 out:
1325 	return r;
1326 }
1327 
1328 void kvm_disable_largepages(void)
1329 {
1330 	largepages_enabled = false;
1331 }
1332 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1333 
1334 int is_error_page(struct page *page)
1335 {
1336 	return page == bad_page;
1337 }
1338 EXPORT_SYMBOL_GPL(is_error_page);
1339 
1340 int is_error_pfn(pfn_t pfn)
1341 {
1342 	return pfn == bad_pfn;
1343 }
1344 EXPORT_SYMBOL_GPL(is_error_pfn);
1345 
1346 static inline unsigned long bad_hva(void)
1347 {
1348 	return PAGE_OFFSET;
1349 }
1350 
1351 int kvm_is_error_hva(unsigned long addr)
1352 {
1353 	return addr == bad_hva();
1354 }
1355 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1356 
1357 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1358 {
1359 	int i;
1360 
1361 	for (i = 0; i < kvm->nmemslots; ++i) {
1362 		struct kvm_memory_slot *memslot = &kvm->memslots[i];
1363 
1364 		if (gfn >= memslot->base_gfn
1365 		    && gfn < memslot->base_gfn + memslot->npages)
1366 			return memslot;
1367 	}
1368 	return NULL;
1369 }
1370 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1371 
1372 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1373 {
1374 	gfn = unalias_gfn(kvm, gfn);
1375 	return gfn_to_memslot_unaliased(kvm, gfn);
1376 }
1377 
1378 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1379 {
1380 	int i;
1381 
1382 	gfn = unalias_gfn(kvm, gfn);
1383 	for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1384 		struct kvm_memory_slot *memslot = &kvm->memslots[i];
1385 
1386 		if (gfn >= memslot->base_gfn
1387 		    && gfn < memslot->base_gfn + memslot->npages)
1388 			return 1;
1389 	}
1390 	return 0;
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1393 
1394 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1395 {
1396 	struct kvm_memory_slot *slot;
1397 
1398 	gfn = unalias_gfn(kvm, gfn);
1399 	slot = gfn_to_memslot_unaliased(kvm, gfn);
1400 	if (!slot)
1401 		return bad_hva();
1402 	return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1403 }
1404 EXPORT_SYMBOL_GPL(gfn_to_hva);
1405 
1406 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1407 {
1408 	struct page *page[1];
1409 	unsigned long addr;
1410 	int npages;
1411 	pfn_t pfn;
1412 
1413 	might_sleep();
1414 
1415 	addr = gfn_to_hva(kvm, gfn);
1416 	if (kvm_is_error_hva(addr)) {
1417 		get_page(bad_page);
1418 		return page_to_pfn(bad_page);
1419 	}
1420 
1421 	npages = get_user_pages_fast(addr, 1, 1, page);
1422 
1423 	if (unlikely(npages != 1)) {
1424 		struct vm_area_struct *vma;
1425 
1426 		down_read(&current->mm->mmap_sem);
1427 		vma = find_vma(current->mm, addr);
1428 
1429 		if (vma == NULL || addr < vma->vm_start ||
1430 		    !(vma->vm_flags & VM_PFNMAP)) {
1431 			up_read(&current->mm->mmap_sem);
1432 			get_page(bad_page);
1433 			return page_to_pfn(bad_page);
1434 		}
1435 
1436 		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1437 		up_read(&current->mm->mmap_sem);
1438 		BUG_ON(!kvm_is_mmio_pfn(pfn));
1439 	} else
1440 		pfn = page_to_pfn(page[0]);
1441 
1442 	return pfn;
1443 }
1444 
1445 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1446 
1447 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1448 {
1449 	pfn_t pfn;
1450 
1451 	pfn = gfn_to_pfn(kvm, gfn);
1452 	if (!kvm_is_mmio_pfn(pfn))
1453 		return pfn_to_page(pfn);
1454 
1455 	WARN_ON(kvm_is_mmio_pfn(pfn));
1456 
1457 	get_page(bad_page);
1458 	return bad_page;
1459 }
1460 
1461 EXPORT_SYMBOL_GPL(gfn_to_page);
1462 
1463 void kvm_release_page_clean(struct page *page)
1464 {
1465 	kvm_release_pfn_clean(page_to_pfn(page));
1466 }
1467 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1468 
1469 void kvm_release_pfn_clean(pfn_t pfn)
1470 {
1471 	if (!kvm_is_mmio_pfn(pfn))
1472 		put_page(pfn_to_page(pfn));
1473 }
1474 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1475 
1476 void kvm_release_page_dirty(struct page *page)
1477 {
1478 	kvm_release_pfn_dirty(page_to_pfn(page));
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1481 
1482 void kvm_release_pfn_dirty(pfn_t pfn)
1483 {
1484 	kvm_set_pfn_dirty(pfn);
1485 	kvm_release_pfn_clean(pfn);
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1488 
1489 void kvm_set_page_dirty(struct page *page)
1490 {
1491 	kvm_set_pfn_dirty(page_to_pfn(page));
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1494 
1495 void kvm_set_pfn_dirty(pfn_t pfn)
1496 {
1497 	if (!kvm_is_mmio_pfn(pfn)) {
1498 		struct page *page = pfn_to_page(pfn);
1499 		if (!PageReserved(page))
1500 			SetPageDirty(page);
1501 	}
1502 }
1503 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1504 
1505 void kvm_set_pfn_accessed(pfn_t pfn)
1506 {
1507 	if (!kvm_is_mmio_pfn(pfn))
1508 		mark_page_accessed(pfn_to_page(pfn));
1509 }
1510 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1511 
1512 void kvm_get_pfn(pfn_t pfn)
1513 {
1514 	if (!kvm_is_mmio_pfn(pfn))
1515 		get_page(pfn_to_page(pfn));
1516 }
1517 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1518 
1519 static int next_segment(unsigned long len, int offset)
1520 {
1521 	if (len > PAGE_SIZE - offset)
1522 		return PAGE_SIZE - offset;
1523 	else
1524 		return len;
1525 }
1526 
1527 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1528 			int len)
1529 {
1530 	int r;
1531 	unsigned long addr;
1532 
1533 	addr = gfn_to_hva(kvm, gfn);
1534 	if (kvm_is_error_hva(addr))
1535 		return -EFAULT;
1536 	r = copy_from_user(data, (void __user *)addr + offset, len);
1537 	if (r)
1538 		return -EFAULT;
1539 	return 0;
1540 }
1541 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1542 
1543 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1544 {
1545 	gfn_t gfn = gpa >> PAGE_SHIFT;
1546 	int seg;
1547 	int offset = offset_in_page(gpa);
1548 	int ret;
1549 
1550 	while ((seg = next_segment(len, offset)) != 0) {
1551 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1552 		if (ret < 0)
1553 			return ret;
1554 		offset = 0;
1555 		len -= seg;
1556 		data += seg;
1557 		++gfn;
1558 	}
1559 	return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_read_guest);
1562 
1563 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1564 			  unsigned long len)
1565 {
1566 	int r;
1567 	unsigned long addr;
1568 	gfn_t gfn = gpa >> PAGE_SHIFT;
1569 	int offset = offset_in_page(gpa);
1570 
1571 	addr = gfn_to_hva(kvm, gfn);
1572 	if (kvm_is_error_hva(addr))
1573 		return -EFAULT;
1574 	pagefault_disable();
1575 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1576 	pagefault_enable();
1577 	if (r)
1578 		return -EFAULT;
1579 	return 0;
1580 }
1581 EXPORT_SYMBOL(kvm_read_guest_atomic);
1582 
1583 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1584 			 int offset, int len)
1585 {
1586 	int r;
1587 	unsigned long addr;
1588 
1589 	addr = gfn_to_hva(kvm, gfn);
1590 	if (kvm_is_error_hva(addr))
1591 		return -EFAULT;
1592 	r = copy_to_user((void __user *)addr + offset, data, len);
1593 	if (r)
1594 		return -EFAULT;
1595 	mark_page_dirty(kvm, gfn);
1596 	return 0;
1597 }
1598 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1599 
1600 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1601 		    unsigned long len)
1602 {
1603 	gfn_t gfn = gpa >> PAGE_SHIFT;
1604 	int seg;
1605 	int offset = offset_in_page(gpa);
1606 	int ret;
1607 
1608 	while ((seg = next_segment(len, offset)) != 0) {
1609 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1610 		if (ret < 0)
1611 			return ret;
1612 		offset = 0;
1613 		len -= seg;
1614 		data += seg;
1615 		++gfn;
1616 	}
1617 	return 0;
1618 }
1619 
1620 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1621 {
1622 	return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1623 }
1624 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1625 
1626 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1627 {
1628 	gfn_t gfn = gpa >> PAGE_SHIFT;
1629 	int seg;
1630 	int offset = offset_in_page(gpa);
1631 	int ret;
1632 
1633         while ((seg = next_segment(len, offset)) != 0) {
1634 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1635 		if (ret < 0)
1636 			return ret;
1637 		offset = 0;
1638 		len -= seg;
1639 		++gfn;
1640 	}
1641 	return 0;
1642 }
1643 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1644 
1645 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1646 {
1647 	struct kvm_memory_slot *memslot;
1648 
1649 	gfn = unalias_gfn(kvm, gfn);
1650 	memslot = gfn_to_memslot_unaliased(kvm, gfn);
1651 	if (memslot && memslot->dirty_bitmap) {
1652 		unsigned long rel_gfn = gfn - memslot->base_gfn;
1653 
1654 		/* avoid RMW */
1655 		if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1656 			set_bit(rel_gfn, memslot->dirty_bitmap);
1657 	}
1658 }
1659 
1660 /*
1661  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1662  */
1663 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1664 {
1665 	DEFINE_WAIT(wait);
1666 
1667 	for (;;) {
1668 		prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1669 
1670 		if (kvm_arch_vcpu_runnable(vcpu)) {
1671 			set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1672 			break;
1673 		}
1674 		if (kvm_cpu_has_pending_timer(vcpu))
1675 			break;
1676 		if (signal_pending(current))
1677 			break;
1678 
1679 		vcpu_put(vcpu);
1680 		schedule();
1681 		vcpu_load(vcpu);
1682 	}
1683 
1684 	finish_wait(&vcpu->wq, &wait);
1685 }
1686 
1687 void kvm_resched(struct kvm_vcpu *vcpu)
1688 {
1689 	if (!need_resched())
1690 		return;
1691 	cond_resched();
1692 }
1693 EXPORT_SYMBOL_GPL(kvm_resched);
1694 
1695 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1696 {
1697 	struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1698 	struct page *page;
1699 
1700 	if (vmf->pgoff == 0)
1701 		page = virt_to_page(vcpu->run);
1702 #ifdef CONFIG_X86
1703 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1704 		page = virt_to_page(vcpu->arch.pio_data);
1705 #endif
1706 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1707 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1708 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1709 #endif
1710 	else
1711 		return VM_FAULT_SIGBUS;
1712 	get_page(page);
1713 	vmf->page = page;
1714 	return 0;
1715 }
1716 
1717 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1718 	.fault = kvm_vcpu_fault,
1719 };
1720 
1721 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1722 {
1723 	vma->vm_ops = &kvm_vcpu_vm_ops;
1724 	return 0;
1725 }
1726 
1727 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1728 {
1729 	struct kvm_vcpu *vcpu = filp->private_data;
1730 
1731 	kvm_put_kvm(vcpu->kvm);
1732 	return 0;
1733 }
1734 
1735 static struct file_operations kvm_vcpu_fops = {
1736 	.release        = kvm_vcpu_release,
1737 	.unlocked_ioctl = kvm_vcpu_ioctl,
1738 	.compat_ioctl   = kvm_vcpu_ioctl,
1739 	.mmap           = kvm_vcpu_mmap,
1740 };
1741 
1742 /*
1743  * Allocates an inode for the vcpu.
1744  */
1745 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1746 {
1747 	return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1748 }
1749 
1750 /*
1751  * Creates some virtual cpus.  Good luck creating more than one.
1752  */
1753 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1754 {
1755 	int r;
1756 	struct kvm_vcpu *vcpu, *v;
1757 
1758 	vcpu = kvm_arch_vcpu_create(kvm, id);
1759 	if (IS_ERR(vcpu))
1760 		return PTR_ERR(vcpu);
1761 
1762 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1763 
1764 	r = kvm_arch_vcpu_setup(vcpu);
1765 	if (r)
1766 		return r;
1767 
1768 	mutex_lock(&kvm->lock);
1769 	if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1770 		r = -EINVAL;
1771 		goto vcpu_destroy;
1772 	}
1773 
1774 	kvm_for_each_vcpu(r, v, kvm)
1775 		if (v->vcpu_id == id) {
1776 			r = -EEXIST;
1777 			goto vcpu_destroy;
1778 		}
1779 
1780 	BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1781 
1782 	/* Now it's all set up, let userspace reach it */
1783 	kvm_get_kvm(kvm);
1784 	r = create_vcpu_fd(vcpu);
1785 	if (r < 0) {
1786 		kvm_put_kvm(kvm);
1787 		goto vcpu_destroy;
1788 	}
1789 
1790 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1791 	smp_wmb();
1792 	atomic_inc(&kvm->online_vcpus);
1793 
1794 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1795 	if (kvm->bsp_vcpu_id == id)
1796 		kvm->bsp_vcpu = vcpu;
1797 #endif
1798 	mutex_unlock(&kvm->lock);
1799 	return r;
1800 
1801 vcpu_destroy:
1802 	mutex_unlock(&kvm->lock);
1803 	kvm_arch_vcpu_destroy(vcpu);
1804 	return r;
1805 }
1806 
1807 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1808 {
1809 	if (sigset) {
1810 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1811 		vcpu->sigset_active = 1;
1812 		vcpu->sigset = *sigset;
1813 	} else
1814 		vcpu->sigset_active = 0;
1815 	return 0;
1816 }
1817 
1818 #ifdef __KVM_HAVE_MSIX
1819 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1820 				    struct kvm_assigned_msix_nr *entry_nr)
1821 {
1822 	int r = 0;
1823 	struct kvm_assigned_dev_kernel *adev;
1824 
1825 	mutex_lock(&kvm->lock);
1826 
1827 	adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1828 				      entry_nr->assigned_dev_id);
1829 	if (!adev) {
1830 		r = -EINVAL;
1831 		goto msix_nr_out;
1832 	}
1833 
1834 	if (adev->entries_nr == 0) {
1835 		adev->entries_nr = entry_nr->entry_nr;
1836 		if (adev->entries_nr == 0 ||
1837 		    adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1838 			r = -EINVAL;
1839 			goto msix_nr_out;
1840 		}
1841 
1842 		adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1843 						entry_nr->entry_nr,
1844 						GFP_KERNEL);
1845 		if (!adev->host_msix_entries) {
1846 			r = -ENOMEM;
1847 			goto msix_nr_out;
1848 		}
1849 		adev->guest_msix_entries = kzalloc(
1850 				sizeof(struct kvm_guest_msix_entry) *
1851 				entry_nr->entry_nr, GFP_KERNEL);
1852 		if (!adev->guest_msix_entries) {
1853 			kfree(adev->host_msix_entries);
1854 			r = -ENOMEM;
1855 			goto msix_nr_out;
1856 		}
1857 	} else /* Not allowed set MSI-X number twice */
1858 		r = -EINVAL;
1859 msix_nr_out:
1860 	mutex_unlock(&kvm->lock);
1861 	return r;
1862 }
1863 
1864 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1865 				       struct kvm_assigned_msix_entry *entry)
1866 {
1867 	int r = 0, i;
1868 	struct kvm_assigned_dev_kernel *adev;
1869 
1870 	mutex_lock(&kvm->lock);
1871 
1872 	adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1873 				      entry->assigned_dev_id);
1874 
1875 	if (!adev) {
1876 		r = -EINVAL;
1877 		goto msix_entry_out;
1878 	}
1879 
1880 	for (i = 0; i < adev->entries_nr; i++)
1881 		if (adev->guest_msix_entries[i].vector == 0 ||
1882 		    adev->guest_msix_entries[i].entry == entry->entry) {
1883 			adev->guest_msix_entries[i].entry = entry->entry;
1884 			adev->guest_msix_entries[i].vector = entry->gsi;
1885 			adev->host_msix_entries[i].entry = entry->entry;
1886 			break;
1887 		}
1888 	if (i == adev->entries_nr) {
1889 		r = -ENOSPC;
1890 		goto msix_entry_out;
1891 	}
1892 
1893 msix_entry_out:
1894 	mutex_unlock(&kvm->lock);
1895 
1896 	return r;
1897 }
1898 #endif
1899 
1900 static long kvm_vcpu_ioctl(struct file *filp,
1901 			   unsigned int ioctl, unsigned long arg)
1902 {
1903 	struct kvm_vcpu *vcpu = filp->private_data;
1904 	void __user *argp = (void __user *)arg;
1905 	int r;
1906 	struct kvm_fpu *fpu = NULL;
1907 	struct kvm_sregs *kvm_sregs = NULL;
1908 
1909 	if (vcpu->kvm->mm != current->mm)
1910 		return -EIO;
1911 	switch (ioctl) {
1912 	case KVM_RUN:
1913 		r = -EINVAL;
1914 		if (arg)
1915 			goto out;
1916 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1917 		break;
1918 	case KVM_GET_REGS: {
1919 		struct kvm_regs *kvm_regs;
1920 
1921 		r = -ENOMEM;
1922 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1923 		if (!kvm_regs)
1924 			goto out;
1925 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1926 		if (r)
1927 			goto out_free1;
1928 		r = -EFAULT;
1929 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1930 			goto out_free1;
1931 		r = 0;
1932 out_free1:
1933 		kfree(kvm_regs);
1934 		break;
1935 	}
1936 	case KVM_SET_REGS: {
1937 		struct kvm_regs *kvm_regs;
1938 
1939 		r = -ENOMEM;
1940 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1941 		if (!kvm_regs)
1942 			goto out;
1943 		r = -EFAULT;
1944 		if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1945 			goto out_free2;
1946 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1947 		if (r)
1948 			goto out_free2;
1949 		r = 0;
1950 out_free2:
1951 		kfree(kvm_regs);
1952 		break;
1953 	}
1954 	case KVM_GET_SREGS: {
1955 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1956 		r = -ENOMEM;
1957 		if (!kvm_sregs)
1958 			goto out;
1959 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1960 		if (r)
1961 			goto out;
1962 		r = -EFAULT;
1963 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1964 			goto out;
1965 		r = 0;
1966 		break;
1967 	}
1968 	case KVM_SET_SREGS: {
1969 		kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1970 		r = -ENOMEM;
1971 		if (!kvm_sregs)
1972 			goto out;
1973 		r = -EFAULT;
1974 		if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1975 			goto out;
1976 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1977 		if (r)
1978 			goto out;
1979 		r = 0;
1980 		break;
1981 	}
1982 	case KVM_GET_MP_STATE: {
1983 		struct kvm_mp_state mp_state;
1984 
1985 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1986 		if (r)
1987 			goto out;
1988 		r = -EFAULT;
1989 		if (copy_to_user(argp, &mp_state, sizeof mp_state))
1990 			goto out;
1991 		r = 0;
1992 		break;
1993 	}
1994 	case KVM_SET_MP_STATE: {
1995 		struct kvm_mp_state mp_state;
1996 
1997 		r = -EFAULT;
1998 		if (copy_from_user(&mp_state, argp, sizeof mp_state))
1999 			goto out;
2000 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2001 		if (r)
2002 			goto out;
2003 		r = 0;
2004 		break;
2005 	}
2006 	case KVM_TRANSLATE: {
2007 		struct kvm_translation tr;
2008 
2009 		r = -EFAULT;
2010 		if (copy_from_user(&tr, argp, sizeof tr))
2011 			goto out;
2012 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2013 		if (r)
2014 			goto out;
2015 		r = -EFAULT;
2016 		if (copy_to_user(argp, &tr, sizeof tr))
2017 			goto out;
2018 		r = 0;
2019 		break;
2020 	}
2021 	case KVM_SET_GUEST_DEBUG: {
2022 		struct kvm_guest_debug dbg;
2023 
2024 		r = -EFAULT;
2025 		if (copy_from_user(&dbg, argp, sizeof dbg))
2026 			goto out;
2027 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2028 		if (r)
2029 			goto out;
2030 		r = 0;
2031 		break;
2032 	}
2033 	case KVM_SET_SIGNAL_MASK: {
2034 		struct kvm_signal_mask __user *sigmask_arg = argp;
2035 		struct kvm_signal_mask kvm_sigmask;
2036 		sigset_t sigset, *p;
2037 
2038 		p = NULL;
2039 		if (argp) {
2040 			r = -EFAULT;
2041 			if (copy_from_user(&kvm_sigmask, argp,
2042 					   sizeof kvm_sigmask))
2043 				goto out;
2044 			r = -EINVAL;
2045 			if (kvm_sigmask.len != sizeof sigset)
2046 				goto out;
2047 			r = -EFAULT;
2048 			if (copy_from_user(&sigset, sigmask_arg->sigset,
2049 					   sizeof sigset))
2050 				goto out;
2051 			p = &sigset;
2052 		}
2053 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2054 		break;
2055 	}
2056 	case KVM_GET_FPU: {
2057 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2058 		r = -ENOMEM;
2059 		if (!fpu)
2060 			goto out;
2061 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2062 		if (r)
2063 			goto out;
2064 		r = -EFAULT;
2065 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2066 			goto out;
2067 		r = 0;
2068 		break;
2069 	}
2070 	case KVM_SET_FPU: {
2071 		fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2072 		r = -ENOMEM;
2073 		if (!fpu)
2074 			goto out;
2075 		r = -EFAULT;
2076 		if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2077 			goto out;
2078 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2079 		if (r)
2080 			goto out;
2081 		r = 0;
2082 		break;
2083 	}
2084 	default:
2085 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2086 	}
2087 out:
2088 	kfree(fpu);
2089 	kfree(kvm_sregs);
2090 	return r;
2091 }
2092 
2093 static long kvm_vm_ioctl(struct file *filp,
2094 			   unsigned int ioctl, unsigned long arg)
2095 {
2096 	struct kvm *kvm = filp->private_data;
2097 	void __user *argp = (void __user *)arg;
2098 	int r;
2099 
2100 	if (kvm->mm != current->mm)
2101 		return -EIO;
2102 	switch (ioctl) {
2103 	case KVM_CREATE_VCPU:
2104 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2105 		if (r < 0)
2106 			goto out;
2107 		break;
2108 	case KVM_SET_USER_MEMORY_REGION: {
2109 		struct kvm_userspace_memory_region kvm_userspace_mem;
2110 
2111 		r = -EFAULT;
2112 		if (copy_from_user(&kvm_userspace_mem, argp,
2113 						sizeof kvm_userspace_mem))
2114 			goto out;
2115 
2116 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2117 		if (r)
2118 			goto out;
2119 		break;
2120 	}
2121 	case KVM_GET_DIRTY_LOG: {
2122 		struct kvm_dirty_log log;
2123 
2124 		r = -EFAULT;
2125 		if (copy_from_user(&log, argp, sizeof log))
2126 			goto out;
2127 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2128 		if (r)
2129 			goto out;
2130 		break;
2131 	}
2132 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2133 	case KVM_REGISTER_COALESCED_MMIO: {
2134 		struct kvm_coalesced_mmio_zone zone;
2135 		r = -EFAULT;
2136 		if (copy_from_user(&zone, argp, sizeof zone))
2137 			goto out;
2138 		r = -ENXIO;
2139 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2140 		if (r)
2141 			goto out;
2142 		r = 0;
2143 		break;
2144 	}
2145 	case KVM_UNREGISTER_COALESCED_MMIO: {
2146 		struct kvm_coalesced_mmio_zone zone;
2147 		r = -EFAULT;
2148 		if (copy_from_user(&zone, argp, sizeof zone))
2149 			goto out;
2150 		r = -ENXIO;
2151 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2152 		if (r)
2153 			goto out;
2154 		r = 0;
2155 		break;
2156 	}
2157 #endif
2158 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2159 	case KVM_ASSIGN_PCI_DEVICE: {
2160 		struct kvm_assigned_pci_dev assigned_dev;
2161 
2162 		r = -EFAULT;
2163 		if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2164 			goto out;
2165 		r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2166 		if (r)
2167 			goto out;
2168 		break;
2169 	}
2170 	case KVM_ASSIGN_IRQ: {
2171 		r = -EOPNOTSUPP;
2172 		break;
2173 	}
2174 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2175 	case KVM_ASSIGN_DEV_IRQ: {
2176 		struct kvm_assigned_irq assigned_irq;
2177 
2178 		r = -EFAULT;
2179 		if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2180 			goto out;
2181 		r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2182 		if (r)
2183 			goto out;
2184 		break;
2185 	}
2186 	case KVM_DEASSIGN_DEV_IRQ: {
2187 		struct kvm_assigned_irq assigned_irq;
2188 
2189 		r = -EFAULT;
2190 		if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2191 			goto out;
2192 		r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2193 		if (r)
2194 			goto out;
2195 		break;
2196 	}
2197 #endif
2198 #endif
2199 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2200 	case KVM_DEASSIGN_PCI_DEVICE: {
2201 		struct kvm_assigned_pci_dev assigned_dev;
2202 
2203 		r = -EFAULT;
2204 		if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2205 			goto out;
2206 		r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2207 		if (r)
2208 			goto out;
2209 		break;
2210 	}
2211 #endif
2212 #ifdef KVM_CAP_IRQ_ROUTING
2213 	case KVM_SET_GSI_ROUTING: {
2214 		struct kvm_irq_routing routing;
2215 		struct kvm_irq_routing __user *urouting;
2216 		struct kvm_irq_routing_entry *entries;
2217 
2218 		r = -EFAULT;
2219 		if (copy_from_user(&routing, argp, sizeof(routing)))
2220 			goto out;
2221 		r = -EINVAL;
2222 		if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2223 			goto out;
2224 		if (routing.flags)
2225 			goto out;
2226 		r = -ENOMEM;
2227 		entries = vmalloc(routing.nr * sizeof(*entries));
2228 		if (!entries)
2229 			goto out;
2230 		r = -EFAULT;
2231 		urouting = argp;
2232 		if (copy_from_user(entries, urouting->entries,
2233 				   routing.nr * sizeof(*entries)))
2234 			goto out_free_irq_routing;
2235 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
2236 					routing.flags);
2237 	out_free_irq_routing:
2238 		vfree(entries);
2239 		break;
2240 	}
2241 #endif /* KVM_CAP_IRQ_ROUTING */
2242 #ifdef __KVM_HAVE_MSIX
2243 	case KVM_ASSIGN_SET_MSIX_NR: {
2244 		struct kvm_assigned_msix_nr entry_nr;
2245 		r = -EFAULT;
2246 		if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2247 			goto out;
2248 		r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2249 		if (r)
2250 			goto out;
2251 		break;
2252 	}
2253 	case KVM_ASSIGN_SET_MSIX_ENTRY: {
2254 		struct kvm_assigned_msix_entry entry;
2255 		r = -EFAULT;
2256 		if (copy_from_user(&entry, argp, sizeof entry))
2257 			goto out;
2258 		r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2259 		if (r)
2260 			goto out;
2261 		break;
2262 	}
2263 #endif
2264 	case KVM_IRQFD: {
2265 		struct kvm_irqfd data;
2266 
2267 		r = -EFAULT;
2268 		if (copy_from_user(&data, argp, sizeof data))
2269 			goto out;
2270 		r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2271 		break;
2272 	}
2273 	case KVM_IOEVENTFD: {
2274 		struct kvm_ioeventfd data;
2275 
2276 		r = -EFAULT;
2277 		if (copy_from_user(&data, argp, sizeof data))
2278 			goto out;
2279 		r = kvm_ioeventfd(kvm, &data);
2280 		break;
2281 	}
2282 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2283 	case KVM_SET_BOOT_CPU_ID:
2284 		r = 0;
2285 		mutex_lock(&kvm->lock);
2286 		if (atomic_read(&kvm->online_vcpus) != 0)
2287 			r = -EBUSY;
2288 		else
2289 			kvm->bsp_vcpu_id = arg;
2290 		mutex_unlock(&kvm->lock);
2291 		break;
2292 #endif
2293 	default:
2294 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2295 	}
2296 out:
2297 	return r;
2298 }
2299 
2300 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2301 {
2302 	struct page *page[1];
2303 	unsigned long addr;
2304 	int npages;
2305 	gfn_t gfn = vmf->pgoff;
2306 	struct kvm *kvm = vma->vm_file->private_data;
2307 
2308 	addr = gfn_to_hva(kvm, gfn);
2309 	if (kvm_is_error_hva(addr))
2310 		return VM_FAULT_SIGBUS;
2311 
2312 	npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2313 				NULL);
2314 	if (unlikely(npages != 1))
2315 		return VM_FAULT_SIGBUS;
2316 
2317 	vmf->page = page[0];
2318 	return 0;
2319 }
2320 
2321 static struct vm_operations_struct kvm_vm_vm_ops = {
2322 	.fault = kvm_vm_fault,
2323 };
2324 
2325 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2326 {
2327 	vma->vm_ops = &kvm_vm_vm_ops;
2328 	return 0;
2329 }
2330 
2331 static struct file_operations kvm_vm_fops = {
2332 	.release        = kvm_vm_release,
2333 	.unlocked_ioctl = kvm_vm_ioctl,
2334 	.compat_ioctl   = kvm_vm_ioctl,
2335 	.mmap           = kvm_vm_mmap,
2336 };
2337 
2338 static int kvm_dev_ioctl_create_vm(void)
2339 {
2340 	int fd;
2341 	struct kvm *kvm;
2342 
2343 	kvm = kvm_create_vm();
2344 	if (IS_ERR(kvm))
2345 		return PTR_ERR(kvm);
2346 	fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2347 	if (fd < 0)
2348 		kvm_put_kvm(kvm);
2349 
2350 	return fd;
2351 }
2352 
2353 static long kvm_dev_ioctl_check_extension_generic(long arg)
2354 {
2355 	switch (arg) {
2356 	case KVM_CAP_USER_MEMORY:
2357 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2358 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2359 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2360 	case KVM_CAP_SET_BOOT_CPU_ID:
2361 #endif
2362 		return 1;
2363 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2364 	case KVM_CAP_IRQ_ROUTING:
2365 		return KVM_MAX_IRQ_ROUTES;
2366 #endif
2367 	default:
2368 		break;
2369 	}
2370 	return kvm_dev_ioctl_check_extension(arg);
2371 }
2372 
2373 static long kvm_dev_ioctl(struct file *filp,
2374 			  unsigned int ioctl, unsigned long arg)
2375 {
2376 	long r = -EINVAL;
2377 
2378 	switch (ioctl) {
2379 	case KVM_GET_API_VERSION:
2380 		r = -EINVAL;
2381 		if (arg)
2382 			goto out;
2383 		r = KVM_API_VERSION;
2384 		break;
2385 	case KVM_CREATE_VM:
2386 		r = -EINVAL;
2387 		if (arg)
2388 			goto out;
2389 		r = kvm_dev_ioctl_create_vm();
2390 		break;
2391 	case KVM_CHECK_EXTENSION:
2392 		r = kvm_dev_ioctl_check_extension_generic(arg);
2393 		break;
2394 	case KVM_GET_VCPU_MMAP_SIZE:
2395 		r = -EINVAL;
2396 		if (arg)
2397 			goto out;
2398 		r = PAGE_SIZE;     /* struct kvm_run */
2399 #ifdef CONFIG_X86
2400 		r += PAGE_SIZE;    /* pio data page */
2401 #endif
2402 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2403 		r += PAGE_SIZE;    /* coalesced mmio ring page */
2404 #endif
2405 		break;
2406 	case KVM_TRACE_ENABLE:
2407 	case KVM_TRACE_PAUSE:
2408 	case KVM_TRACE_DISABLE:
2409 		r = -EOPNOTSUPP;
2410 		break;
2411 	default:
2412 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
2413 	}
2414 out:
2415 	return r;
2416 }
2417 
2418 static struct file_operations kvm_chardev_ops = {
2419 	.unlocked_ioctl = kvm_dev_ioctl,
2420 	.compat_ioctl   = kvm_dev_ioctl,
2421 };
2422 
2423 static struct miscdevice kvm_dev = {
2424 	KVM_MINOR,
2425 	"kvm",
2426 	&kvm_chardev_ops,
2427 };
2428 
2429 static void hardware_enable(void *junk)
2430 {
2431 	int cpu = raw_smp_processor_id();
2432 
2433 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2434 		return;
2435 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
2436 	kvm_arch_hardware_enable(NULL);
2437 }
2438 
2439 static void hardware_disable(void *junk)
2440 {
2441 	int cpu = raw_smp_processor_id();
2442 
2443 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2444 		return;
2445 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2446 	kvm_arch_hardware_disable(NULL);
2447 }
2448 
2449 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2450 			   void *v)
2451 {
2452 	int cpu = (long)v;
2453 
2454 	val &= ~CPU_TASKS_FROZEN;
2455 	switch (val) {
2456 	case CPU_DYING:
2457 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2458 		       cpu);
2459 		hardware_disable(NULL);
2460 		break;
2461 	case CPU_UP_CANCELED:
2462 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2463 		       cpu);
2464 		smp_call_function_single(cpu, hardware_disable, NULL, 1);
2465 		break;
2466 	case CPU_ONLINE:
2467 		printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2468 		       cpu);
2469 		smp_call_function_single(cpu, hardware_enable, NULL, 1);
2470 		break;
2471 	}
2472 	return NOTIFY_OK;
2473 }
2474 
2475 
2476 asmlinkage void kvm_handle_fault_on_reboot(void)
2477 {
2478 	if (kvm_rebooting)
2479 		/* spin while reset goes on */
2480 		while (true)
2481 			;
2482 	/* Fault while not rebooting.  We want the trace. */
2483 	BUG();
2484 }
2485 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2486 
2487 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2488 		      void *v)
2489 {
2490 	/*
2491 	 * Some (well, at least mine) BIOSes hang on reboot if
2492 	 * in vmx root mode.
2493 	 *
2494 	 * And Intel TXT required VMX off for all cpu when system shutdown.
2495 	 */
2496 	printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2497 	kvm_rebooting = true;
2498 	on_each_cpu(hardware_disable, NULL, 1);
2499 	return NOTIFY_OK;
2500 }
2501 
2502 static struct notifier_block kvm_reboot_notifier = {
2503 	.notifier_call = kvm_reboot,
2504 	.priority = 0,
2505 };
2506 
2507 void kvm_io_bus_init(struct kvm_io_bus *bus)
2508 {
2509 	memset(bus, 0, sizeof(*bus));
2510 }
2511 
2512 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2513 {
2514 	int i;
2515 
2516 	for (i = 0; i < bus->dev_count; i++) {
2517 		struct kvm_io_device *pos = bus->devs[i];
2518 
2519 		kvm_iodevice_destructor(pos);
2520 	}
2521 }
2522 
2523 /* kvm_io_bus_write - called under kvm->slots_lock */
2524 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2525 		     int len, const void *val)
2526 {
2527 	int i;
2528 	for (i = 0; i < bus->dev_count; i++)
2529 		if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2530 			return 0;
2531 	return -EOPNOTSUPP;
2532 }
2533 
2534 /* kvm_io_bus_read - called under kvm->slots_lock */
2535 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
2536 {
2537 	int i;
2538 	for (i = 0; i < bus->dev_count; i++)
2539 		if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2540 			return 0;
2541 	return -EOPNOTSUPP;
2542 }
2543 
2544 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
2545 			     struct kvm_io_device *dev)
2546 {
2547 	int ret;
2548 
2549 	down_write(&kvm->slots_lock);
2550 	ret = __kvm_io_bus_register_dev(bus, dev);
2551 	up_write(&kvm->slots_lock);
2552 
2553 	return ret;
2554 }
2555 
2556 /* An unlocked version. Caller must have write lock on slots_lock. */
2557 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2558 			      struct kvm_io_device *dev)
2559 {
2560 	if (bus->dev_count > NR_IOBUS_DEVS-1)
2561 		return -ENOSPC;
2562 
2563 	bus->devs[bus->dev_count++] = dev;
2564 
2565 	return 0;
2566 }
2567 
2568 void kvm_io_bus_unregister_dev(struct kvm *kvm,
2569 			       struct kvm_io_bus *bus,
2570 			       struct kvm_io_device *dev)
2571 {
2572 	down_write(&kvm->slots_lock);
2573 	__kvm_io_bus_unregister_dev(bus, dev);
2574 	up_write(&kvm->slots_lock);
2575 }
2576 
2577 /* An unlocked version. Caller must have write lock on slots_lock. */
2578 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2579 				 struct kvm_io_device *dev)
2580 {
2581 	int i;
2582 
2583 	for (i = 0; i < bus->dev_count; i++)
2584 		if (bus->devs[i] == dev) {
2585 			bus->devs[i] = bus->devs[--bus->dev_count];
2586 			break;
2587 		}
2588 }
2589 
2590 static struct notifier_block kvm_cpu_notifier = {
2591 	.notifier_call = kvm_cpu_hotplug,
2592 	.priority = 20, /* must be > scheduler priority */
2593 };
2594 
2595 static int vm_stat_get(void *_offset, u64 *val)
2596 {
2597 	unsigned offset = (long)_offset;
2598 	struct kvm *kvm;
2599 
2600 	*val = 0;
2601 	spin_lock(&kvm_lock);
2602 	list_for_each_entry(kvm, &vm_list, vm_list)
2603 		*val += *(u32 *)((void *)kvm + offset);
2604 	spin_unlock(&kvm_lock);
2605 	return 0;
2606 }
2607 
2608 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2609 
2610 static int vcpu_stat_get(void *_offset, u64 *val)
2611 {
2612 	unsigned offset = (long)_offset;
2613 	struct kvm *kvm;
2614 	struct kvm_vcpu *vcpu;
2615 	int i;
2616 
2617 	*val = 0;
2618 	spin_lock(&kvm_lock);
2619 	list_for_each_entry(kvm, &vm_list, vm_list)
2620 		kvm_for_each_vcpu(i, vcpu, kvm)
2621 			*val += *(u32 *)((void *)vcpu + offset);
2622 
2623 	spin_unlock(&kvm_lock);
2624 	return 0;
2625 }
2626 
2627 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2628 
2629 static struct file_operations *stat_fops[] = {
2630 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
2631 	[KVM_STAT_VM]   = &vm_stat_fops,
2632 };
2633 
2634 static void kvm_init_debug(void)
2635 {
2636 	struct kvm_stats_debugfs_item *p;
2637 
2638 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2639 	for (p = debugfs_entries; p->name; ++p)
2640 		p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2641 						(void *)(long)p->offset,
2642 						stat_fops[p->kind]);
2643 }
2644 
2645 static void kvm_exit_debug(void)
2646 {
2647 	struct kvm_stats_debugfs_item *p;
2648 
2649 	for (p = debugfs_entries; p->name; ++p)
2650 		debugfs_remove(p->dentry);
2651 	debugfs_remove(kvm_debugfs_dir);
2652 }
2653 
2654 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2655 {
2656 	hardware_disable(NULL);
2657 	return 0;
2658 }
2659 
2660 static int kvm_resume(struct sys_device *dev)
2661 {
2662 	hardware_enable(NULL);
2663 	return 0;
2664 }
2665 
2666 static struct sysdev_class kvm_sysdev_class = {
2667 	.name = "kvm",
2668 	.suspend = kvm_suspend,
2669 	.resume = kvm_resume,
2670 };
2671 
2672 static struct sys_device kvm_sysdev = {
2673 	.id = 0,
2674 	.cls = &kvm_sysdev_class,
2675 };
2676 
2677 struct page *bad_page;
2678 pfn_t bad_pfn;
2679 
2680 static inline
2681 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2682 {
2683 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
2684 }
2685 
2686 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2687 {
2688 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2689 
2690 	kvm_arch_vcpu_load(vcpu, cpu);
2691 }
2692 
2693 static void kvm_sched_out(struct preempt_notifier *pn,
2694 			  struct task_struct *next)
2695 {
2696 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2697 
2698 	kvm_arch_vcpu_put(vcpu);
2699 }
2700 
2701 int kvm_init(void *opaque, unsigned int vcpu_size,
2702 		  struct module *module)
2703 {
2704 	int r;
2705 	int cpu;
2706 
2707 	kvm_init_debug();
2708 
2709 	r = kvm_arch_init(opaque);
2710 	if (r)
2711 		goto out_fail;
2712 
2713 	bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2714 
2715 	if (bad_page == NULL) {
2716 		r = -ENOMEM;
2717 		goto out;
2718 	}
2719 
2720 	bad_pfn = page_to_pfn(bad_page);
2721 
2722 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2723 		r = -ENOMEM;
2724 		goto out_free_0;
2725 	}
2726 
2727 	r = kvm_arch_hardware_setup();
2728 	if (r < 0)
2729 		goto out_free_0a;
2730 
2731 	for_each_online_cpu(cpu) {
2732 		smp_call_function_single(cpu,
2733 				kvm_arch_check_processor_compat,
2734 				&r, 1);
2735 		if (r < 0)
2736 			goto out_free_1;
2737 	}
2738 
2739 	on_each_cpu(hardware_enable, NULL, 1);
2740 	r = register_cpu_notifier(&kvm_cpu_notifier);
2741 	if (r)
2742 		goto out_free_2;
2743 	register_reboot_notifier(&kvm_reboot_notifier);
2744 
2745 	r = sysdev_class_register(&kvm_sysdev_class);
2746 	if (r)
2747 		goto out_free_3;
2748 
2749 	r = sysdev_register(&kvm_sysdev);
2750 	if (r)
2751 		goto out_free_4;
2752 
2753 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
2754 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2755 					   __alignof__(struct kvm_vcpu),
2756 					   0, NULL);
2757 	if (!kvm_vcpu_cache) {
2758 		r = -ENOMEM;
2759 		goto out_free_5;
2760 	}
2761 
2762 	kvm_chardev_ops.owner = module;
2763 	kvm_vm_fops.owner = module;
2764 	kvm_vcpu_fops.owner = module;
2765 
2766 	r = misc_register(&kvm_dev);
2767 	if (r) {
2768 		printk(KERN_ERR "kvm: misc device register failed\n");
2769 		goto out_free;
2770 	}
2771 
2772 	kvm_preempt_ops.sched_in = kvm_sched_in;
2773 	kvm_preempt_ops.sched_out = kvm_sched_out;
2774 
2775 	return 0;
2776 
2777 out_free:
2778 	kmem_cache_destroy(kvm_vcpu_cache);
2779 out_free_5:
2780 	sysdev_unregister(&kvm_sysdev);
2781 out_free_4:
2782 	sysdev_class_unregister(&kvm_sysdev_class);
2783 out_free_3:
2784 	unregister_reboot_notifier(&kvm_reboot_notifier);
2785 	unregister_cpu_notifier(&kvm_cpu_notifier);
2786 out_free_2:
2787 	on_each_cpu(hardware_disable, NULL, 1);
2788 out_free_1:
2789 	kvm_arch_hardware_unsetup();
2790 out_free_0a:
2791 	free_cpumask_var(cpus_hardware_enabled);
2792 out_free_0:
2793 	__free_page(bad_page);
2794 out:
2795 	kvm_arch_exit();
2796 out_fail:
2797 	kvm_exit_debug();
2798 	return r;
2799 }
2800 EXPORT_SYMBOL_GPL(kvm_init);
2801 
2802 void kvm_exit(void)
2803 {
2804 	tracepoint_synchronize_unregister();
2805 	misc_deregister(&kvm_dev);
2806 	kmem_cache_destroy(kvm_vcpu_cache);
2807 	sysdev_unregister(&kvm_sysdev);
2808 	sysdev_class_unregister(&kvm_sysdev_class);
2809 	unregister_reboot_notifier(&kvm_reboot_notifier);
2810 	unregister_cpu_notifier(&kvm_cpu_notifier);
2811 	on_each_cpu(hardware_disable, NULL, 1);
2812 	kvm_arch_hardware_unsetup();
2813 	kvm_arch_exit();
2814 	kvm_exit_debug();
2815 	free_cpumask_var(cpus_hardware_enabled);
2816 	__free_page(bad_page);
2817 }
2818 EXPORT_SYMBOL_GPL(kvm_exit);
2819