xref: /linux/virt/kvm/kvm_main.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  *
9  * Authors:
10  *   Avi Kivity   <avi@qumranet.com>
11  *   Yaniv Kamay  <yaniv@qumranet.com>
12  *
13  * This work is licensed under the terms of the GNU GPL, version 2.  See
14  * the COPYING file in the top-level directory.
15  *
16  */
17 
18 #include "iodev.h"
19 
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 
45 #include <asm/processor.h>
46 #include <asm/io.h>
47 #include <asm/uaccess.h>
48 #include <asm/pgtable.h>
49 
50 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
51 #include "coalesced_mmio.h"
52 #endif
53 
54 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
55 #include <linux/pci.h>
56 #include <linux/interrupt.h>
57 #include "irq.h"
58 #endif
59 
60 MODULE_AUTHOR("Qumranet");
61 MODULE_LICENSE("GPL");
62 
63 static int msi2intx = 1;
64 module_param(msi2intx, bool, 0);
65 
66 DEFINE_SPINLOCK(kvm_lock);
67 LIST_HEAD(vm_list);
68 
69 static cpumask_var_t cpus_hardware_enabled;
70 
71 struct kmem_cache *kvm_vcpu_cache;
72 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
73 
74 static __read_mostly struct preempt_ops kvm_preempt_ops;
75 
76 struct dentry *kvm_debugfs_dir;
77 
78 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
79 			   unsigned long arg);
80 
81 static bool kvm_rebooting;
82 
83 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
84 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
85 						      int assigned_dev_id)
86 {
87 	struct list_head *ptr;
88 	struct kvm_assigned_dev_kernel *match;
89 
90 	list_for_each(ptr, head) {
91 		match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
92 		if (match->assigned_dev_id == assigned_dev_id)
93 			return match;
94 	}
95 	return NULL;
96 }
97 
98 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
99 {
100 	struct kvm_assigned_dev_kernel *assigned_dev;
101 
102 	assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
103 				    interrupt_work);
104 
105 	/* This is taken to safely inject irq inside the guest. When
106 	 * the interrupt injection (or the ioapic code) uses a
107 	 * finer-grained lock, update this
108 	 */
109 	mutex_lock(&assigned_dev->kvm->lock);
110 	kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
111 		    assigned_dev->guest_irq, 1);
112 
113 	if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_MSI) {
114 		enable_irq(assigned_dev->host_irq);
115 		assigned_dev->host_irq_disabled = false;
116 	}
117 	mutex_unlock(&assigned_dev->kvm->lock);
118 }
119 
120 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
121 {
122 	struct kvm_assigned_dev_kernel *assigned_dev =
123 		(struct kvm_assigned_dev_kernel *) dev_id;
124 
125 	schedule_work(&assigned_dev->interrupt_work);
126 
127 	disable_irq_nosync(irq);
128 	assigned_dev->host_irq_disabled = true;
129 
130 	return IRQ_HANDLED;
131 }
132 
133 /* Ack the irq line for an assigned device */
134 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
135 {
136 	struct kvm_assigned_dev_kernel *dev;
137 
138 	if (kian->gsi == -1)
139 		return;
140 
141 	dev = container_of(kian, struct kvm_assigned_dev_kernel,
142 			   ack_notifier);
143 
144 	kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
145 
146 	/* The guest irq may be shared so this ack may be
147 	 * from another device.
148 	 */
149 	if (dev->host_irq_disabled) {
150 		enable_irq(dev->host_irq);
151 		dev->host_irq_disabled = false;
152 	}
153 }
154 
155 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
156 static void kvm_free_assigned_irq(struct kvm *kvm,
157 				  struct kvm_assigned_dev_kernel *assigned_dev)
158 {
159 	if (!irqchip_in_kernel(kvm))
160 		return;
161 
162 	kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
163 
164 	if (assigned_dev->irq_source_id != -1)
165 		kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
166 	assigned_dev->irq_source_id = -1;
167 
168 	if (!assigned_dev->irq_requested_type)
169 		return;
170 
171 	/*
172 	 * In kvm_free_device_irq, cancel_work_sync return true if:
173 	 * 1. work is scheduled, and then cancelled.
174 	 * 2. work callback is executed.
175 	 *
176 	 * The first one ensured that the irq is disabled and no more events
177 	 * would happen. But for the second one, the irq may be enabled (e.g.
178 	 * for MSI). So we disable irq here to prevent further events.
179 	 *
180 	 * Notice this maybe result in nested disable if the interrupt type is
181 	 * INTx, but it's OK for we are going to free it.
182 	 *
183 	 * If this function is a part of VM destroy, please ensure that till
184 	 * now, the kvm state is still legal for probably we also have to wait
185 	 * interrupt_work done.
186 	 */
187 	disable_irq_nosync(assigned_dev->host_irq);
188 	cancel_work_sync(&assigned_dev->interrupt_work);
189 
190 	free_irq(assigned_dev->host_irq, (void *)assigned_dev);
191 
192 	if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
193 		pci_disable_msi(assigned_dev->dev);
194 
195 	assigned_dev->irq_requested_type = 0;
196 }
197 
198 
199 static void kvm_free_assigned_device(struct kvm *kvm,
200 				     struct kvm_assigned_dev_kernel
201 				     *assigned_dev)
202 {
203 	kvm_free_assigned_irq(kvm, assigned_dev);
204 
205 	pci_reset_function(assigned_dev->dev);
206 
207 	pci_release_regions(assigned_dev->dev);
208 	pci_disable_device(assigned_dev->dev);
209 	pci_dev_put(assigned_dev->dev);
210 
211 	list_del(&assigned_dev->list);
212 	kfree(assigned_dev);
213 }
214 
215 void kvm_free_all_assigned_devices(struct kvm *kvm)
216 {
217 	struct list_head *ptr, *ptr2;
218 	struct kvm_assigned_dev_kernel *assigned_dev;
219 
220 	list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
221 		assigned_dev = list_entry(ptr,
222 					  struct kvm_assigned_dev_kernel,
223 					  list);
224 
225 		kvm_free_assigned_device(kvm, assigned_dev);
226 	}
227 }
228 
229 static int assigned_device_update_intx(struct kvm *kvm,
230 			struct kvm_assigned_dev_kernel *adev,
231 			struct kvm_assigned_irq *airq)
232 {
233 	adev->guest_irq = airq->guest_irq;
234 	adev->ack_notifier.gsi = airq->guest_irq;
235 
236 	if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
237 		return 0;
238 
239 	if (irqchip_in_kernel(kvm)) {
240 		if (!msi2intx &&
241 		    (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)) {
242 			free_irq(adev->host_irq, (void *)adev);
243 			pci_disable_msi(adev->dev);
244 		}
245 
246 		if (!capable(CAP_SYS_RAWIO))
247 			return -EPERM;
248 
249 		if (airq->host_irq)
250 			adev->host_irq = airq->host_irq;
251 		else
252 			adev->host_irq = adev->dev->irq;
253 
254 		/* Even though this is PCI, we don't want to use shared
255 		 * interrupts. Sharing host devices with guest-assigned devices
256 		 * on the same interrupt line is not a happy situation: there
257 		 * are going to be long delays in accepting, acking, etc.
258 		 */
259 		if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
260 				0, "kvm_assigned_intx_device", (void *)adev))
261 			return -EIO;
262 	}
263 
264 	adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
265 				   KVM_ASSIGNED_DEV_HOST_INTX;
266 	return 0;
267 }
268 
269 #ifdef CONFIG_X86
270 static int assigned_device_update_msi(struct kvm *kvm,
271 			struct kvm_assigned_dev_kernel *adev,
272 			struct kvm_assigned_irq *airq)
273 {
274 	int r;
275 
276 	adev->guest_irq = airq->guest_irq;
277 	if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
278 		/* x86 don't care upper address of guest msi message addr */
279 		adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
280 		adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
281 		adev->ack_notifier.gsi = -1;
282 	} else if (msi2intx) {
283 		adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
284 		adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
285 		adev->ack_notifier.gsi = airq->guest_irq;
286 	} else {
287 		/*
288 		 * Guest require to disable device MSI, we disable MSI and
289 		 * re-enable INTx by default again. Notice it's only for
290 		 * non-msi2intx.
291 		 */
292 		assigned_device_update_intx(kvm, adev, airq);
293 		return 0;
294 	}
295 
296 	if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
297 		return 0;
298 
299 	if (irqchip_in_kernel(kvm)) {
300 		if (!msi2intx) {
301 			if (adev->irq_requested_type &
302 					KVM_ASSIGNED_DEV_HOST_INTX)
303 				free_irq(adev->host_irq, (void *)adev);
304 
305 			r = pci_enable_msi(adev->dev);
306 			if (r)
307 				return r;
308 		}
309 
310 		adev->host_irq = adev->dev->irq;
311 		if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
312 				"kvm_assigned_msi_device", (void *)adev))
313 			return -EIO;
314 	}
315 
316 	if (!msi2intx)
317 		adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
318 
319 	adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
320 	return 0;
321 }
322 #endif
323 
324 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
325 				   struct kvm_assigned_irq
326 				   *assigned_irq)
327 {
328 	int r = 0;
329 	struct kvm_assigned_dev_kernel *match;
330 	u32 current_flags = 0, changed_flags;
331 
332 	mutex_lock(&kvm->lock);
333 
334 	match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
335 				      assigned_irq->assigned_dev_id);
336 	if (!match) {
337 		mutex_unlock(&kvm->lock);
338 		return -EINVAL;
339 	}
340 
341 	if (!match->irq_requested_type) {
342 		INIT_WORK(&match->interrupt_work,
343 				kvm_assigned_dev_interrupt_work_handler);
344 		if (irqchip_in_kernel(kvm)) {
345 			/* Register ack nofitier */
346 			match->ack_notifier.gsi = -1;
347 			match->ack_notifier.irq_acked =
348 					kvm_assigned_dev_ack_irq;
349 			kvm_register_irq_ack_notifier(kvm,
350 					&match->ack_notifier);
351 
352 			/* Request IRQ source ID */
353 			r = kvm_request_irq_source_id(kvm);
354 			if (r < 0)
355 				goto out_release;
356 			else
357 				match->irq_source_id = r;
358 
359 #ifdef CONFIG_X86
360 			/* Determine host device irq type, we can know the
361 			 * result from dev->msi_enabled */
362 			if (msi2intx)
363 				pci_enable_msi(match->dev);
364 #endif
365 		}
366 	}
367 
368 	if ((match->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) &&
369 		 (match->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_MSI))
370 		current_flags |= KVM_DEV_IRQ_ASSIGN_ENABLE_MSI;
371 
372 	changed_flags = assigned_irq->flags ^ current_flags;
373 
374 	if ((changed_flags & KVM_DEV_IRQ_ASSIGN_MSI_ACTION) ||
375 	    (msi2intx && match->dev->msi_enabled)) {
376 #ifdef CONFIG_X86
377 		r = assigned_device_update_msi(kvm, match, assigned_irq);
378 		if (r) {
379 			printk(KERN_WARNING "kvm: failed to enable "
380 					"MSI device!\n");
381 			goto out_release;
382 		}
383 #else
384 		r = -ENOTTY;
385 #endif
386 	} else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
387 		/* Host device IRQ 0 means don't support INTx */
388 		if (!msi2intx) {
389 			printk(KERN_WARNING
390 			       "kvm: wait device to enable MSI!\n");
391 			r = 0;
392 		} else {
393 			printk(KERN_WARNING
394 			       "kvm: failed to enable MSI device!\n");
395 			r = -ENOTTY;
396 			goto out_release;
397 		}
398 	} else {
399 		/* Non-sharing INTx mode */
400 		r = assigned_device_update_intx(kvm, match, assigned_irq);
401 		if (r) {
402 			printk(KERN_WARNING "kvm: failed to enable "
403 					"INTx device!\n");
404 			goto out_release;
405 		}
406 	}
407 
408 	mutex_unlock(&kvm->lock);
409 	return r;
410 out_release:
411 	mutex_unlock(&kvm->lock);
412 	kvm_free_assigned_device(kvm, match);
413 	return r;
414 }
415 
416 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
417 				      struct kvm_assigned_pci_dev *assigned_dev)
418 {
419 	int r = 0;
420 	struct kvm_assigned_dev_kernel *match;
421 	struct pci_dev *dev;
422 
423 	down_read(&kvm->slots_lock);
424 	mutex_lock(&kvm->lock);
425 
426 	match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
427 				      assigned_dev->assigned_dev_id);
428 	if (match) {
429 		/* device already assigned */
430 		r = -EINVAL;
431 		goto out;
432 	}
433 
434 	match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
435 	if (match == NULL) {
436 		printk(KERN_INFO "%s: Couldn't allocate memory\n",
437 		       __func__);
438 		r = -ENOMEM;
439 		goto out;
440 	}
441 	dev = pci_get_bus_and_slot(assigned_dev->busnr,
442 				   assigned_dev->devfn);
443 	if (!dev) {
444 		printk(KERN_INFO "%s: host device not found\n", __func__);
445 		r = -EINVAL;
446 		goto out_free;
447 	}
448 	if (pci_enable_device(dev)) {
449 		printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
450 		r = -EBUSY;
451 		goto out_put;
452 	}
453 	r = pci_request_regions(dev, "kvm_assigned_device");
454 	if (r) {
455 		printk(KERN_INFO "%s: Could not get access to device regions\n",
456 		       __func__);
457 		goto out_disable;
458 	}
459 
460 	pci_reset_function(dev);
461 
462 	match->assigned_dev_id = assigned_dev->assigned_dev_id;
463 	match->host_busnr = assigned_dev->busnr;
464 	match->host_devfn = assigned_dev->devfn;
465 	match->flags = assigned_dev->flags;
466 	match->dev = dev;
467 	match->irq_source_id = -1;
468 	match->kvm = kvm;
469 
470 	list_add(&match->list, &kvm->arch.assigned_dev_head);
471 
472 	if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
473 		if (!kvm->arch.iommu_domain) {
474 			r = kvm_iommu_map_guest(kvm);
475 			if (r)
476 				goto out_list_del;
477 		}
478 		r = kvm_assign_device(kvm, match);
479 		if (r)
480 			goto out_list_del;
481 	}
482 
483 out:
484 	mutex_unlock(&kvm->lock);
485 	up_read(&kvm->slots_lock);
486 	return r;
487 out_list_del:
488 	list_del(&match->list);
489 	pci_release_regions(dev);
490 out_disable:
491 	pci_disable_device(dev);
492 out_put:
493 	pci_dev_put(dev);
494 out_free:
495 	kfree(match);
496 	mutex_unlock(&kvm->lock);
497 	up_read(&kvm->slots_lock);
498 	return r;
499 }
500 #endif
501 
502 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
503 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
504 		struct kvm_assigned_pci_dev *assigned_dev)
505 {
506 	int r = 0;
507 	struct kvm_assigned_dev_kernel *match;
508 
509 	mutex_lock(&kvm->lock);
510 
511 	match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
512 				      assigned_dev->assigned_dev_id);
513 	if (!match) {
514 		printk(KERN_INFO "%s: device hasn't been assigned before, "
515 		  "so cannot be deassigned\n", __func__);
516 		r = -EINVAL;
517 		goto out;
518 	}
519 
520 	if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
521 		kvm_deassign_device(kvm, match);
522 
523 	kvm_free_assigned_device(kvm, match);
524 
525 out:
526 	mutex_unlock(&kvm->lock);
527 	return r;
528 }
529 #endif
530 
531 static inline int valid_vcpu(int n)
532 {
533 	return likely(n >= 0 && n < KVM_MAX_VCPUS);
534 }
535 
536 inline int kvm_is_mmio_pfn(pfn_t pfn)
537 {
538 	if (pfn_valid(pfn)) {
539 		struct page *page = compound_head(pfn_to_page(pfn));
540 		return PageReserved(page);
541 	}
542 
543 	return true;
544 }
545 
546 /*
547  * Switches to specified vcpu, until a matching vcpu_put()
548  */
549 void vcpu_load(struct kvm_vcpu *vcpu)
550 {
551 	int cpu;
552 
553 	mutex_lock(&vcpu->mutex);
554 	cpu = get_cpu();
555 	preempt_notifier_register(&vcpu->preempt_notifier);
556 	kvm_arch_vcpu_load(vcpu, cpu);
557 	put_cpu();
558 }
559 
560 void vcpu_put(struct kvm_vcpu *vcpu)
561 {
562 	preempt_disable();
563 	kvm_arch_vcpu_put(vcpu);
564 	preempt_notifier_unregister(&vcpu->preempt_notifier);
565 	preempt_enable();
566 	mutex_unlock(&vcpu->mutex);
567 }
568 
569 static void ack_flush(void *_completed)
570 {
571 }
572 
573 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
574 {
575 	int i, cpu, me;
576 	cpumask_var_t cpus;
577 	bool called = true;
578 	struct kvm_vcpu *vcpu;
579 
580 	if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
581 		cpumask_clear(cpus);
582 
583 	me = get_cpu();
584 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
585 		vcpu = kvm->vcpus[i];
586 		if (!vcpu)
587 			continue;
588 		if (test_and_set_bit(req, &vcpu->requests))
589 			continue;
590 		cpu = vcpu->cpu;
591 		if (cpus != NULL && cpu != -1 && cpu != me)
592 			cpumask_set_cpu(cpu, cpus);
593 	}
594 	if (unlikely(cpus == NULL))
595 		smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
596 	else if (!cpumask_empty(cpus))
597 		smp_call_function_many(cpus, ack_flush, NULL, 1);
598 	else
599 		called = false;
600 	put_cpu();
601 	free_cpumask_var(cpus);
602 	return called;
603 }
604 
605 void kvm_flush_remote_tlbs(struct kvm *kvm)
606 {
607 	if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
608 		++kvm->stat.remote_tlb_flush;
609 }
610 
611 void kvm_reload_remote_mmus(struct kvm *kvm)
612 {
613 	make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
614 }
615 
616 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
617 {
618 	struct page *page;
619 	int r;
620 
621 	mutex_init(&vcpu->mutex);
622 	vcpu->cpu = -1;
623 	vcpu->kvm = kvm;
624 	vcpu->vcpu_id = id;
625 	init_waitqueue_head(&vcpu->wq);
626 
627 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
628 	if (!page) {
629 		r = -ENOMEM;
630 		goto fail;
631 	}
632 	vcpu->run = page_address(page);
633 
634 	r = kvm_arch_vcpu_init(vcpu);
635 	if (r < 0)
636 		goto fail_free_run;
637 	return 0;
638 
639 fail_free_run:
640 	free_page((unsigned long)vcpu->run);
641 fail:
642 	return r;
643 }
644 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
645 
646 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
647 {
648 	kvm_arch_vcpu_uninit(vcpu);
649 	free_page((unsigned long)vcpu->run);
650 }
651 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
652 
653 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
654 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
655 {
656 	return container_of(mn, struct kvm, mmu_notifier);
657 }
658 
659 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
660 					     struct mm_struct *mm,
661 					     unsigned long address)
662 {
663 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
664 	int need_tlb_flush;
665 
666 	/*
667 	 * When ->invalidate_page runs, the linux pte has been zapped
668 	 * already but the page is still allocated until
669 	 * ->invalidate_page returns. So if we increase the sequence
670 	 * here the kvm page fault will notice if the spte can't be
671 	 * established because the page is going to be freed. If
672 	 * instead the kvm page fault establishes the spte before
673 	 * ->invalidate_page runs, kvm_unmap_hva will release it
674 	 * before returning.
675 	 *
676 	 * The sequence increase only need to be seen at spin_unlock
677 	 * time, and not at spin_lock time.
678 	 *
679 	 * Increasing the sequence after the spin_unlock would be
680 	 * unsafe because the kvm page fault could then establish the
681 	 * pte after kvm_unmap_hva returned, without noticing the page
682 	 * is going to be freed.
683 	 */
684 	spin_lock(&kvm->mmu_lock);
685 	kvm->mmu_notifier_seq++;
686 	need_tlb_flush = kvm_unmap_hva(kvm, address);
687 	spin_unlock(&kvm->mmu_lock);
688 
689 	/* we've to flush the tlb before the pages can be freed */
690 	if (need_tlb_flush)
691 		kvm_flush_remote_tlbs(kvm);
692 
693 }
694 
695 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
696 						    struct mm_struct *mm,
697 						    unsigned long start,
698 						    unsigned long end)
699 {
700 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
701 	int need_tlb_flush = 0;
702 
703 	spin_lock(&kvm->mmu_lock);
704 	/*
705 	 * The count increase must become visible at unlock time as no
706 	 * spte can be established without taking the mmu_lock and
707 	 * count is also read inside the mmu_lock critical section.
708 	 */
709 	kvm->mmu_notifier_count++;
710 	for (; start < end; start += PAGE_SIZE)
711 		need_tlb_flush |= kvm_unmap_hva(kvm, start);
712 	spin_unlock(&kvm->mmu_lock);
713 
714 	/* we've to flush the tlb before the pages can be freed */
715 	if (need_tlb_flush)
716 		kvm_flush_remote_tlbs(kvm);
717 }
718 
719 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
720 						  struct mm_struct *mm,
721 						  unsigned long start,
722 						  unsigned long end)
723 {
724 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
725 
726 	spin_lock(&kvm->mmu_lock);
727 	/*
728 	 * This sequence increase will notify the kvm page fault that
729 	 * the page that is going to be mapped in the spte could have
730 	 * been freed.
731 	 */
732 	kvm->mmu_notifier_seq++;
733 	/*
734 	 * The above sequence increase must be visible before the
735 	 * below count decrease but both values are read by the kvm
736 	 * page fault under mmu_lock spinlock so we don't need to add
737 	 * a smb_wmb() here in between the two.
738 	 */
739 	kvm->mmu_notifier_count--;
740 	spin_unlock(&kvm->mmu_lock);
741 
742 	BUG_ON(kvm->mmu_notifier_count < 0);
743 }
744 
745 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
746 					      struct mm_struct *mm,
747 					      unsigned long address)
748 {
749 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
750 	int young;
751 
752 	spin_lock(&kvm->mmu_lock);
753 	young = kvm_age_hva(kvm, address);
754 	spin_unlock(&kvm->mmu_lock);
755 
756 	if (young)
757 		kvm_flush_remote_tlbs(kvm);
758 
759 	return young;
760 }
761 
762 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
763 				     struct mm_struct *mm)
764 {
765 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
766 	kvm_arch_flush_shadow(kvm);
767 }
768 
769 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
770 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
771 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
772 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
773 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
774 	.release		= kvm_mmu_notifier_release,
775 };
776 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
777 
778 static struct kvm *kvm_create_vm(void)
779 {
780 	struct kvm *kvm = kvm_arch_create_vm();
781 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
782 	struct page *page;
783 #endif
784 
785 	if (IS_ERR(kvm))
786 		goto out;
787 #ifdef CONFIG_HAVE_KVM_IRQCHIP
788 	INIT_LIST_HEAD(&kvm->irq_routing);
789 	INIT_HLIST_HEAD(&kvm->mask_notifier_list);
790 #endif
791 
792 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
793 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
794 	if (!page) {
795 		kfree(kvm);
796 		return ERR_PTR(-ENOMEM);
797 	}
798 	kvm->coalesced_mmio_ring =
799 			(struct kvm_coalesced_mmio_ring *)page_address(page);
800 #endif
801 
802 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
803 	{
804 		int err;
805 		kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
806 		err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
807 		if (err) {
808 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
809 			put_page(page);
810 #endif
811 			kfree(kvm);
812 			return ERR_PTR(err);
813 		}
814 	}
815 #endif
816 
817 	kvm->mm = current->mm;
818 	atomic_inc(&kvm->mm->mm_count);
819 	spin_lock_init(&kvm->mmu_lock);
820 	kvm_io_bus_init(&kvm->pio_bus);
821 	mutex_init(&kvm->lock);
822 	kvm_io_bus_init(&kvm->mmio_bus);
823 	init_rwsem(&kvm->slots_lock);
824 	atomic_set(&kvm->users_count, 1);
825 	spin_lock(&kvm_lock);
826 	list_add(&kvm->vm_list, &vm_list);
827 	spin_unlock(&kvm_lock);
828 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
829 	kvm_coalesced_mmio_init(kvm);
830 #endif
831 out:
832 	return kvm;
833 }
834 
835 /*
836  * Free any memory in @free but not in @dont.
837  */
838 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
839 				  struct kvm_memory_slot *dont)
840 {
841 	if (!dont || free->rmap != dont->rmap)
842 		vfree(free->rmap);
843 
844 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
845 		vfree(free->dirty_bitmap);
846 
847 	if (!dont || free->lpage_info != dont->lpage_info)
848 		vfree(free->lpage_info);
849 
850 	free->npages = 0;
851 	free->dirty_bitmap = NULL;
852 	free->rmap = NULL;
853 	free->lpage_info = NULL;
854 }
855 
856 void kvm_free_physmem(struct kvm *kvm)
857 {
858 	int i;
859 
860 	for (i = 0; i < kvm->nmemslots; ++i)
861 		kvm_free_physmem_slot(&kvm->memslots[i], NULL);
862 }
863 
864 static void kvm_destroy_vm(struct kvm *kvm)
865 {
866 	struct mm_struct *mm = kvm->mm;
867 
868 	kvm_arch_sync_events(kvm);
869 	spin_lock(&kvm_lock);
870 	list_del(&kvm->vm_list);
871 	spin_unlock(&kvm_lock);
872 	kvm_free_irq_routing(kvm);
873 	kvm_io_bus_destroy(&kvm->pio_bus);
874 	kvm_io_bus_destroy(&kvm->mmio_bus);
875 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
876 	if (kvm->coalesced_mmio_ring != NULL)
877 		free_page((unsigned long)kvm->coalesced_mmio_ring);
878 #endif
879 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
880 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
881 #endif
882 	kvm_arch_destroy_vm(kvm);
883 	mmdrop(mm);
884 }
885 
886 void kvm_get_kvm(struct kvm *kvm)
887 {
888 	atomic_inc(&kvm->users_count);
889 }
890 EXPORT_SYMBOL_GPL(kvm_get_kvm);
891 
892 void kvm_put_kvm(struct kvm *kvm)
893 {
894 	if (atomic_dec_and_test(&kvm->users_count))
895 		kvm_destroy_vm(kvm);
896 }
897 EXPORT_SYMBOL_GPL(kvm_put_kvm);
898 
899 
900 static int kvm_vm_release(struct inode *inode, struct file *filp)
901 {
902 	struct kvm *kvm = filp->private_data;
903 
904 	kvm_put_kvm(kvm);
905 	return 0;
906 }
907 
908 /*
909  * Allocate some memory and give it an address in the guest physical address
910  * space.
911  *
912  * Discontiguous memory is allowed, mostly for framebuffers.
913  *
914  * Must be called holding mmap_sem for write.
915  */
916 int __kvm_set_memory_region(struct kvm *kvm,
917 			    struct kvm_userspace_memory_region *mem,
918 			    int user_alloc)
919 {
920 	int r;
921 	gfn_t base_gfn;
922 	unsigned long npages;
923 	unsigned long i;
924 	struct kvm_memory_slot *memslot;
925 	struct kvm_memory_slot old, new;
926 
927 	r = -EINVAL;
928 	/* General sanity checks */
929 	if (mem->memory_size & (PAGE_SIZE - 1))
930 		goto out;
931 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
932 		goto out;
933 	if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
934 		goto out;
935 	if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
936 		goto out;
937 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
938 		goto out;
939 
940 	memslot = &kvm->memslots[mem->slot];
941 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
942 	npages = mem->memory_size >> PAGE_SHIFT;
943 
944 	if (!npages)
945 		mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
946 
947 	new = old = *memslot;
948 
949 	new.base_gfn = base_gfn;
950 	new.npages = npages;
951 	new.flags = mem->flags;
952 
953 	/* Disallow changing a memory slot's size. */
954 	r = -EINVAL;
955 	if (npages && old.npages && npages != old.npages)
956 		goto out_free;
957 
958 	/* Check for overlaps */
959 	r = -EEXIST;
960 	for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
961 		struct kvm_memory_slot *s = &kvm->memslots[i];
962 
963 		if (s == memslot)
964 			continue;
965 		if (!((base_gfn + npages <= s->base_gfn) ||
966 		      (base_gfn >= s->base_gfn + s->npages)))
967 			goto out_free;
968 	}
969 
970 	/* Free page dirty bitmap if unneeded */
971 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
972 		new.dirty_bitmap = NULL;
973 
974 	r = -ENOMEM;
975 
976 	/* Allocate if a slot is being created */
977 #ifndef CONFIG_S390
978 	if (npages && !new.rmap) {
979 		new.rmap = vmalloc(npages * sizeof(struct page *));
980 
981 		if (!new.rmap)
982 			goto out_free;
983 
984 		memset(new.rmap, 0, npages * sizeof(*new.rmap));
985 
986 		new.user_alloc = user_alloc;
987 		/*
988 		 * hva_to_rmmap() serialzies with the mmu_lock and to be
989 		 * safe it has to ignore memslots with !user_alloc &&
990 		 * !userspace_addr.
991 		 */
992 		if (user_alloc)
993 			new.userspace_addr = mem->userspace_addr;
994 		else
995 			new.userspace_addr = 0;
996 	}
997 	if (npages && !new.lpage_info) {
998 		int largepages = npages / KVM_PAGES_PER_HPAGE;
999 		if (npages % KVM_PAGES_PER_HPAGE)
1000 			largepages++;
1001 		if (base_gfn % KVM_PAGES_PER_HPAGE)
1002 			largepages++;
1003 
1004 		new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1005 
1006 		if (!new.lpage_info)
1007 			goto out_free;
1008 
1009 		memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1010 
1011 		if (base_gfn % KVM_PAGES_PER_HPAGE)
1012 			new.lpage_info[0].write_count = 1;
1013 		if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1014 			new.lpage_info[largepages-1].write_count = 1;
1015 	}
1016 
1017 	/* Allocate page dirty bitmap if needed */
1018 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1019 		unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1020 
1021 		new.dirty_bitmap = vmalloc(dirty_bytes);
1022 		if (!new.dirty_bitmap)
1023 			goto out_free;
1024 		memset(new.dirty_bitmap, 0, dirty_bytes);
1025 	}
1026 #endif /* not defined CONFIG_S390 */
1027 
1028 	if (!npages)
1029 		kvm_arch_flush_shadow(kvm);
1030 
1031 	spin_lock(&kvm->mmu_lock);
1032 	if (mem->slot >= kvm->nmemslots)
1033 		kvm->nmemslots = mem->slot + 1;
1034 
1035 	*memslot = new;
1036 	spin_unlock(&kvm->mmu_lock);
1037 
1038 	r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1039 	if (r) {
1040 		spin_lock(&kvm->mmu_lock);
1041 		*memslot = old;
1042 		spin_unlock(&kvm->mmu_lock);
1043 		goto out_free;
1044 	}
1045 
1046 	kvm_free_physmem_slot(&old, npages ? &new : NULL);
1047 	/* Slot deletion case: we have to update the current slot */
1048 	if (!npages)
1049 		*memslot = old;
1050 #ifdef CONFIG_DMAR
1051 	/* map the pages in iommu page table */
1052 	r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1053 	if (r)
1054 		goto out;
1055 #endif
1056 	return 0;
1057 
1058 out_free:
1059 	kvm_free_physmem_slot(&new, &old);
1060 out:
1061 	return r;
1062 
1063 }
1064 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1065 
1066 int kvm_set_memory_region(struct kvm *kvm,
1067 			  struct kvm_userspace_memory_region *mem,
1068 			  int user_alloc)
1069 {
1070 	int r;
1071 
1072 	down_write(&kvm->slots_lock);
1073 	r = __kvm_set_memory_region(kvm, mem, user_alloc);
1074 	up_write(&kvm->slots_lock);
1075 	return r;
1076 }
1077 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1078 
1079 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1080 				   struct
1081 				   kvm_userspace_memory_region *mem,
1082 				   int user_alloc)
1083 {
1084 	if (mem->slot >= KVM_MEMORY_SLOTS)
1085 		return -EINVAL;
1086 	return kvm_set_memory_region(kvm, mem, user_alloc);
1087 }
1088 
1089 int kvm_get_dirty_log(struct kvm *kvm,
1090 			struct kvm_dirty_log *log, int *is_dirty)
1091 {
1092 	struct kvm_memory_slot *memslot;
1093 	int r, i;
1094 	int n;
1095 	unsigned long any = 0;
1096 
1097 	r = -EINVAL;
1098 	if (log->slot >= KVM_MEMORY_SLOTS)
1099 		goto out;
1100 
1101 	memslot = &kvm->memslots[log->slot];
1102 	r = -ENOENT;
1103 	if (!memslot->dirty_bitmap)
1104 		goto out;
1105 
1106 	n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1107 
1108 	for (i = 0; !any && i < n/sizeof(long); ++i)
1109 		any = memslot->dirty_bitmap[i];
1110 
1111 	r = -EFAULT;
1112 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1113 		goto out;
1114 
1115 	if (any)
1116 		*is_dirty = 1;
1117 
1118 	r = 0;
1119 out:
1120 	return r;
1121 }
1122 
1123 int is_error_page(struct page *page)
1124 {
1125 	return page == bad_page;
1126 }
1127 EXPORT_SYMBOL_GPL(is_error_page);
1128 
1129 int is_error_pfn(pfn_t pfn)
1130 {
1131 	return pfn == bad_pfn;
1132 }
1133 EXPORT_SYMBOL_GPL(is_error_pfn);
1134 
1135 static inline unsigned long bad_hva(void)
1136 {
1137 	return PAGE_OFFSET;
1138 }
1139 
1140 int kvm_is_error_hva(unsigned long addr)
1141 {
1142 	return addr == bad_hva();
1143 }
1144 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1145 
1146 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1147 {
1148 	int i;
1149 
1150 	for (i = 0; i < kvm->nmemslots; ++i) {
1151 		struct kvm_memory_slot *memslot = &kvm->memslots[i];
1152 
1153 		if (gfn >= memslot->base_gfn
1154 		    && gfn < memslot->base_gfn + memslot->npages)
1155 			return memslot;
1156 	}
1157 	return NULL;
1158 }
1159 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1160 
1161 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1162 {
1163 	gfn = unalias_gfn(kvm, gfn);
1164 	return gfn_to_memslot_unaliased(kvm, gfn);
1165 }
1166 
1167 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1168 {
1169 	int i;
1170 
1171 	gfn = unalias_gfn(kvm, gfn);
1172 	for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1173 		struct kvm_memory_slot *memslot = &kvm->memslots[i];
1174 
1175 		if (gfn >= memslot->base_gfn
1176 		    && gfn < memslot->base_gfn + memslot->npages)
1177 			return 1;
1178 	}
1179 	return 0;
1180 }
1181 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1182 
1183 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1184 {
1185 	struct kvm_memory_slot *slot;
1186 
1187 	gfn = unalias_gfn(kvm, gfn);
1188 	slot = gfn_to_memslot_unaliased(kvm, gfn);
1189 	if (!slot)
1190 		return bad_hva();
1191 	return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1192 }
1193 EXPORT_SYMBOL_GPL(gfn_to_hva);
1194 
1195 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1196 {
1197 	struct page *page[1];
1198 	unsigned long addr;
1199 	int npages;
1200 	pfn_t pfn;
1201 
1202 	might_sleep();
1203 
1204 	addr = gfn_to_hva(kvm, gfn);
1205 	if (kvm_is_error_hva(addr)) {
1206 		get_page(bad_page);
1207 		return page_to_pfn(bad_page);
1208 	}
1209 
1210 	npages = get_user_pages_fast(addr, 1, 1, page);
1211 
1212 	if (unlikely(npages != 1)) {
1213 		struct vm_area_struct *vma;
1214 
1215 		down_read(&current->mm->mmap_sem);
1216 		vma = find_vma(current->mm, addr);
1217 
1218 		if (vma == NULL || addr < vma->vm_start ||
1219 		    !(vma->vm_flags & VM_PFNMAP)) {
1220 			up_read(&current->mm->mmap_sem);
1221 			get_page(bad_page);
1222 			return page_to_pfn(bad_page);
1223 		}
1224 
1225 		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1226 		up_read(&current->mm->mmap_sem);
1227 		BUG_ON(!kvm_is_mmio_pfn(pfn));
1228 	} else
1229 		pfn = page_to_pfn(page[0]);
1230 
1231 	return pfn;
1232 }
1233 
1234 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1235 
1236 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1237 {
1238 	pfn_t pfn;
1239 
1240 	pfn = gfn_to_pfn(kvm, gfn);
1241 	if (!kvm_is_mmio_pfn(pfn))
1242 		return pfn_to_page(pfn);
1243 
1244 	WARN_ON(kvm_is_mmio_pfn(pfn));
1245 
1246 	get_page(bad_page);
1247 	return bad_page;
1248 }
1249 
1250 EXPORT_SYMBOL_GPL(gfn_to_page);
1251 
1252 void kvm_release_page_clean(struct page *page)
1253 {
1254 	kvm_release_pfn_clean(page_to_pfn(page));
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1257 
1258 void kvm_release_pfn_clean(pfn_t pfn)
1259 {
1260 	if (!kvm_is_mmio_pfn(pfn))
1261 		put_page(pfn_to_page(pfn));
1262 }
1263 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1264 
1265 void kvm_release_page_dirty(struct page *page)
1266 {
1267 	kvm_release_pfn_dirty(page_to_pfn(page));
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1270 
1271 void kvm_release_pfn_dirty(pfn_t pfn)
1272 {
1273 	kvm_set_pfn_dirty(pfn);
1274 	kvm_release_pfn_clean(pfn);
1275 }
1276 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1277 
1278 void kvm_set_page_dirty(struct page *page)
1279 {
1280 	kvm_set_pfn_dirty(page_to_pfn(page));
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1283 
1284 void kvm_set_pfn_dirty(pfn_t pfn)
1285 {
1286 	if (!kvm_is_mmio_pfn(pfn)) {
1287 		struct page *page = pfn_to_page(pfn);
1288 		if (!PageReserved(page))
1289 			SetPageDirty(page);
1290 	}
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1293 
1294 void kvm_set_pfn_accessed(pfn_t pfn)
1295 {
1296 	if (!kvm_is_mmio_pfn(pfn))
1297 		mark_page_accessed(pfn_to_page(pfn));
1298 }
1299 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1300 
1301 void kvm_get_pfn(pfn_t pfn)
1302 {
1303 	if (!kvm_is_mmio_pfn(pfn))
1304 		get_page(pfn_to_page(pfn));
1305 }
1306 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1307 
1308 static int next_segment(unsigned long len, int offset)
1309 {
1310 	if (len > PAGE_SIZE - offset)
1311 		return PAGE_SIZE - offset;
1312 	else
1313 		return len;
1314 }
1315 
1316 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1317 			int len)
1318 {
1319 	int r;
1320 	unsigned long addr;
1321 
1322 	addr = gfn_to_hva(kvm, gfn);
1323 	if (kvm_is_error_hva(addr))
1324 		return -EFAULT;
1325 	r = copy_from_user(data, (void __user *)addr + offset, len);
1326 	if (r)
1327 		return -EFAULT;
1328 	return 0;
1329 }
1330 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1331 
1332 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1333 {
1334 	gfn_t gfn = gpa >> PAGE_SHIFT;
1335 	int seg;
1336 	int offset = offset_in_page(gpa);
1337 	int ret;
1338 
1339 	while ((seg = next_segment(len, offset)) != 0) {
1340 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1341 		if (ret < 0)
1342 			return ret;
1343 		offset = 0;
1344 		len -= seg;
1345 		data += seg;
1346 		++gfn;
1347 	}
1348 	return 0;
1349 }
1350 EXPORT_SYMBOL_GPL(kvm_read_guest);
1351 
1352 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1353 			  unsigned long len)
1354 {
1355 	int r;
1356 	unsigned long addr;
1357 	gfn_t gfn = gpa >> PAGE_SHIFT;
1358 	int offset = offset_in_page(gpa);
1359 
1360 	addr = gfn_to_hva(kvm, gfn);
1361 	if (kvm_is_error_hva(addr))
1362 		return -EFAULT;
1363 	pagefault_disable();
1364 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1365 	pagefault_enable();
1366 	if (r)
1367 		return -EFAULT;
1368 	return 0;
1369 }
1370 EXPORT_SYMBOL(kvm_read_guest_atomic);
1371 
1372 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1373 			 int offset, int len)
1374 {
1375 	int r;
1376 	unsigned long addr;
1377 
1378 	addr = gfn_to_hva(kvm, gfn);
1379 	if (kvm_is_error_hva(addr))
1380 		return -EFAULT;
1381 	r = copy_to_user((void __user *)addr + offset, data, len);
1382 	if (r)
1383 		return -EFAULT;
1384 	mark_page_dirty(kvm, gfn);
1385 	return 0;
1386 }
1387 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1388 
1389 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1390 		    unsigned long len)
1391 {
1392 	gfn_t gfn = gpa >> PAGE_SHIFT;
1393 	int seg;
1394 	int offset = offset_in_page(gpa);
1395 	int ret;
1396 
1397 	while ((seg = next_segment(len, offset)) != 0) {
1398 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1399 		if (ret < 0)
1400 			return ret;
1401 		offset = 0;
1402 		len -= seg;
1403 		data += seg;
1404 		++gfn;
1405 	}
1406 	return 0;
1407 }
1408 
1409 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1410 {
1411 	return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1412 }
1413 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1414 
1415 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1416 {
1417 	gfn_t gfn = gpa >> PAGE_SHIFT;
1418 	int seg;
1419 	int offset = offset_in_page(gpa);
1420 	int ret;
1421 
1422         while ((seg = next_segment(len, offset)) != 0) {
1423 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1424 		if (ret < 0)
1425 			return ret;
1426 		offset = 0;
1427 		len -= seg;
1428 		++gfn;
1429 	}
1430 	return 0;
1431 }
1432 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1433 
1434 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1435 {
1436 	struct kvm_memory_slot *memslot;
1437 
1438 	gfn = unalias_gfn(kvm, gfn);
1439 	memslot = gfn_to_memslot_unaliased(kvm, gfn);
1440 	if (memslot && memslot->dirty_bitmap) {
1441 		unsigned long rel_gfn = gfn - memslot->base_gfn;
1442 
1443 		/* avoid RMW */
1444 		if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1445 			set_bit(rel_gfn, memslot->dirty_bitmap);
1446 	}
1447 }
1448 
1449 /*
1450  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1451  */
1452 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1453 {
1454 	DEFINE_WAIT(wait);
1455 
1456 	for (;;) {
1457 		prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1458 
1459 		if (kvm_cpu_has_interrupt(vcpu) ||
1460 		    kvm_cpu_has_pending_timer(vcpu) ||
1461 		    kvm_arch_vcpu_runnable(vcpu)) {
1462 			set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1463 			break;
1464 		}
1465 		if (signal_pending(current))
1466 			break;
1467 
1468 		vcpu_put(vcpu);
1469 		schedule();
1470 		vcpu_load(vcpu);
1471 	}
1472 
1473 	finish_wait(&vcpu->wq, &wait);
1474 }
1475 
1476 void kvm_resched(struct kvm_vcpu *vcpu)
1477 {
1478 	if (!need_resched())
1479 		return;
1480 	cond_resched();
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_resched);
1483 
1484 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1485 {
1486 	struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1487 	struct page *page;
1488 
1489 	if (vmf->pgoff == 0)
1490 		page = virt_to_page(vcpu->run);
1491 #ifdef CONFIG_X86
1492 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1493 		page = virt_to_page(vcpu->arch.pio_data);
1494 #endif
1495 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1496 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1497 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1498 #endif
1499 	else
1500 		return VM_FAULT_SIGBUS;
1501 	get_page(page);
1502 	vmf->page = page;
1503 	return 0;
1504 }
1505 
1506 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1507 	.fault = kvm_vcpu_fault,
1508 };
1509 
1510 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1511 {
1512 	vma->vm_ops = &kvm_vcpu_vm_ops;
1513 	return 0;
1514 }
1515 
1516 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1517 {
1518 	struct kvm_vcpu *vcpu = filp->private_data;
1519 
1520 	kvm_put_kvm(vcpu->kvm);
1521 	return 0;
1522 }
1523 
1524 static struct file_operations kvm_vcpu_fops = {
1525 	.release        = kvm_vcpu_release,
1526 	.unlocked_ioctl = kvm_vcpu_ioctl,
1527 	.compat_ioctl   = kvm_vcpu_ioctl,
1528 	.mmap           = kvm_vcpu_mmap,
1529 };
1530 
1531 /*
1532  * Allocates an inode for the vcpu.
1533  */
1534 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1535 {
1536 	int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1537 	if (fd < 0)
1538 		kvm_put_kvm(vcpu->kvm);
1539 	return fd;
1540 }
1541 
1542 /*
1543  * Creates some virtual cpus.  Good luck creating more than one.
1544  */
1545 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1546 {
1547 	int r;
1548 	struct kvm_vcpu *vcpu;
1549 
1550 	if (!valid_vcpu(n))
1551 		return -EINVAL;
1552 
1553 	vcpu = kvm_arch_vcpu_create(kvm, n);
1554 	if (IS_ERR(vcpu))
1555 		return PTR_ERR(vcpu);
1556 
1557 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1558 
1559 	r = kvm_arch_vcpu_setup(vcpu);
1560 	if (r)
1561 		return r;
1562 
1563 	mutex_lock(&kvm->lock);
1564 	if (kvm->vcpus[n]) {
1565 		r = -EEXIST;
1566 		goto vcpu_destroy;
1567 	}
1568 	kvm->vcpus[n] = vcpu;
1569 	mutex_unlock(&kvm->lock);
1570 
1571 	/* Now it's all set up, let userspace reach it */
1572 	kvm_get_kvm(kvm);
1573 	r = create_vcpu_fd(vcpu);
1574 	if (r < 0)
1575 		goto unlink;
1576 	return r;
1577 
1578 unlink:
1579 	mutex_lock(&kvm->lock);
1580 	kvm->vcpus[n] = NULL;
1581 vcpu_destroy:
1582 	mutex_unlock(&kvm->lock);
1583 	kvm_arch_vcpu_destroy(vcpu);
1584 	return r;
1585 }
1586 
1587 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1588 {
1589 	if (sigset) {
1590 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1591 		vcpu->sigset_active = 1;
1592 		vcpu->sigset = *sigset;
1593 	} else
1594 		vcpu->sigset_active = 0;
1595 	return 0;
1596 }
1597 
1598 static long kvm_vcpu_ioctl(struct file *filp,
1599 			   unsigned int ioctl, unsigned long arg)
1600 {
1601 	struct kvm_vcpu *vcpu = filp->private_data;
1602 	void __user *argp = (void __user *)arg;
1603 	int r;
1604 	struct kvm_fpu *fpu = NULL;
1605 	struct kvm_sregs *kvm_sregs = NULL;
1606 
1607 	if (vcpu->kvm->mm != current->mm)
1608 		return -EIO;
1609 	switch (ioctl) {
1610 	case KVM_RUN:
1611 		r = -EINVAL;
1612 		if (arg)
1613 			goto out;
1614 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1615 		break;
1616 	case KVM_GET_REGS: {
1617 		struct kvm_regs *kvm_regs;
1618 
1619 		r = -ENOMEM;
1620 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1621 		if (!kvm_regs)
1622 			goto out;
1623 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1624 		if (r)
1625 			goto out_free1;
1626 		r = -EFAULT;
1627 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1628 			goto out_free1;
1629 		r = 0;
1630 out_free1:
1631 		kfree(kvm_regs);
1632 		break;
1633 	}
1634 	case KVM_SET_REGS: {
1635 		struct kvm_regs *kvm_regs;
1636 
1637 		r = -ENOMEM;
1638 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1639 		if (!kvm_regs)
1640 			goto out;
1641 		r = -EFAULT;
1642 		if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1643 			goto out_free2;
1644 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1645 		if (r)
1646 			goto out_free2;
1647 		r = 0;
1648 out_free2:
1649 		kfree(kvm_regs);
1650 		break;
1651 	}
1652 	case KVM_GET_SREGS: {
1653 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1654 		r = -ENOMEM;
1655 		if (!kvm_sregs)
1656 			goto out;
1657 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1658 		if (r)
1659 			goto out;
1660 		r = -EFAULT;
1661 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1662 			goto out;
1663 		r = 0;
1664 		break;
1665 	}
1666 	case KVM_SET_SREGS: {
1667 		kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1668 		r = -ENOMEM;
1669 		if (!kvm_sregs)
1670 			goto out;
1671 		r = -EFAULT;
1672 		if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1673 			goto out;
1674 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1675 		if (r)
1676 			goto out;
1677 		r = 0;
1678 		break;
1679 	}
1680 	case KVM_GET_MP_STATE: {
1681 		struct kvm_mp_state mp_state;
1682 
1683 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1684 		if (r)
1685 			goto out;
1686 		r = -EFAULT;
1687 		if (copy_to_user(argp, &mp_state, sizeof mp_state))
1688 			goto out;
1689 		r = 0;
1690 		break;
1691 	}
1692 	case KVM_SET_MP_STATE: {
1693 		struct kvm_mp_state mp_state;
1694 
1695 		r = -EFAULT;
1696 		if (copy_from_user(&mp_state, argp, sizeof mp_state))
1697 			goto out;
1698 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1699 		if (r)
1700 			goto out;
1701 		r = 0;
1702 		break;
1703 	}
1704 	case KVM_TRANSLATE: {
1705 		struct kvm_translation tr;
1706 
1707 		r = -EFAULT;
1708 		if (copy_from_user(&tr, argp, sizeof tr))
1709 			goto out;
1710 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1711 		if (r)
1712 			goto out;
1713 		r = -EFAULT;
1714 		if (copy_to_user(argp, &tr, sizeof tr))
1715 			goto out;
1716 		r = 0;
1717 		break;
1718 	}
1719 	case KVM_SET_GUEST_DEBUG: {
1720 		struct kvm_guest_debug dbg;
1721 
1722 		r = -EFAULT;
1723 		if (copy_from_user(&dbg, argp, sizeof dbg))
1724 			goto out;
1725 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1726 		if (r)
1727 			goto out;
1728 		r = 0;
1729 		break;
1730 	}
1731 	case KVM_SET_SIGNAL_MASK: {
1732 		struct kvm_signal_mask __user *sigmask_arg = argp;
1733 		struct kvm_signal_mask kvm_sigmask;
1734 		sigset_t sigset, *p;
1735 
1736 		p = NULL;
1737 		if (argp) {
1738 			r = -EFAULT;
1739 			if (copy_from_user(&kvm_sigmask, argp,
1740 					   sizeof kvm_sigmask))
1741 				goto out;
1742 			r = -EINVAL;
1743 			if (kvm_sigmask.len != sizeof sigset)
1744 				goto out;
1745 			r = -EFAULT;
1746 			if (copy_from_user(&sigset, sigmask_arg->sigset,
1747 					   sizeof sigset))
1748 				goto out;
1749 			p = &sigset;
1750 		}
1751 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1752 		break;
1753 	}
1754 	case KVM_GET_FPU: {
1755 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1756 		r = -ENOMEM;
1757 		if (!fpu)
1758 			goto out;
1759 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1760 		if (r)
1761 			goto out;
1762 		r = -EFAULT;
1763 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1764 			goto out;
1765 		r = 0;
1766 		break;
1767 	}
1768 	case KVM_SET_FPU: {
1769 		fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1770 		r = -ENOMEM;
1771 		if (!fpu)
1772 			goto out;
1773 		r = -EFAULT;
1774 		if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1775 			goto out;
1776 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1777 		if (r)
1778 			goto out;
1779 		r = 0;
1780 		break;
1781 	}
1782 	default:
1783 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1784 	}
1785 out:
1786 	kfree(fpu);
1787 	kfree(kvm_sregs);
1788 	return r;
1789 }
1790 
1791 static long kvm_vm_ioctl(struct file *filp,
1792 			   unsigned int ioctl, unsigned long arg)
1793 {
1794 	struct kvm *kvm = filp->private_data;
1795 	void __user *argp = (void __user *)arg;
1796 	int r;
1797 
1798 	if (kvm->mm != current->mm)
1799 		return -EIO;
1800 	switch (ioctl) {
1801 	case KVM_CREATE_VCPU:
1802 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1803 		if (r < 0)
1804 			goto out;
1805 		break;
1806 	case KVM_SET_USER_MEMORY_REGION: {
1807 		struct kvm_userspace_memory_region kvm_userspace_mem;
1808 
1809 		r = -EFAULT;
1810 		if (copy_from_user(&kvm_userspace_mem, argp,
1811 						sizeof kvm_userspace_mem))
1812 			goto out;
1813 
1814 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1815 		if (r)
1816 			goto out;
1817 		break;
1818 	}
1819 	case KVM_GET_DIRTY_LOG: {
1820 		struct kvm_dirty_log log;
1821 
1822 		r = -EFAULT;
1823 		if (copy_from_user(&log, argp, sizeof log))
1824 			goto out;
1825 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1826 		if (r)
1827 			goto out;
1828 		break;
1829 	}
1830 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1831 	case KVM_REGISTER_COALESCED_MMIO: {
1832 		struct kvm_coalesced_mmio_zone zone;
1833 		r = -EFAULT;
1834 		if (copy_from_user(&zone, argp, sizeof zone))
1835 			goto out;
1836 		r = -ENXIO;
1837 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1838 		if (r)
1839 			goto out;
1840 		r = 0;
1841 		break;
1842 	}
1843 	case KVM_UNREGISTER_COALESCED_MMIO: {
1844 		struct kvm_coalesced_mmio_zone zone;
1845 		r = -EFAULT;
1846 		if (copy_from_user(&zone, argp, sizeof zone))
1847 			goto out;
1848 		r = -ENXIO;
1849 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1850 		if (r)
1851 			goto out;
1852 		r = 0;
1853 		break;
1854 	}
1855 #endif
1856 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1857 	case KVM_ASSIGN_PCI_DEVICE: {
1858 		struct kvm_assigned_pci_dev assigned_dev;
1859 
1860 		r = -EFAULT;
1861 		if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1862 			goto out;
1863 		r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1864 		if (r)
1865 			goto out;
1866 		break;
1867 	}
1868 	case KVM_ASSIGN_IRQ: {
1869 		struct kvm_assigned_irq assigned_irq;
1870 
1871 		r = -EFAULT;
1872 		if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1873 			goto out;
1874 		r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1875 		if (r)
1876 			goto out;
1877 		break;
1878 	}
1879 #endif
1880 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1881 	case KVM_DEASSIGN_PCI_DEVICE: {
1882 		struct kvm_assigned_pci_dev assigned_dev;
1883 
1884 		r = -EFAULT;
1885 		if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1886 			goto out;
1887 		r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1888 		if (r)
1889 			goto out;
1890 		break;
1891 	}
1892 #endif
1893 #ifdef KVM_CAP_IRQ_ROUTING
1894 	case KVM_SET_GSI_ROUTING: {
1895 		struct kvm_irq_routing routing;
1896 		struct kvm_irq_routing __user *urouting;
1897 		struct kvm_irq_routing_entry *entries;
1898 
1899 		r = -EFAULT;
1900 		if (copy_from_user(&routing, argp, sizeof(routing)))
1901 			goto out;
1902 		r = -EINVAL;
1903 		if (routing.nr >= KVM_MAX_IRQ_ROUTES)
1904 			goto out;
1905 		if (routing.flags)
1906 			goto out;
1907 		r = -ENOMEM;
1908 		entries = vmalloc(routing.nr * sizeof(*entries));
1909 		if (!entries)
1910 			goto out;
1911 		r = -EFAULT;
1912 		urouting = argp;
1913 		if (copy_from_user(entries, urouting->entries,
1914 				   routing.nr * sizeof(*entries)))
1915 			goto out_free_irq_routing;
1916 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
1917 					routing.flags);
1918 	out_free_irq_routing:
1919 		vfree(entries);
1920 		break;
1921 	}
1922 #endif
1923 	default:
1924 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1925 	}
1926 out:
1927 	return r;
1928 }
1929 
1930 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1931 {
1932 	struct page *page[1];
1933 	unsigned long addr;
1934 	int npages;
1935 	gfn_t gfn = vmf->pgoff;
1936 	struct kvm *kvm = vma->vm_file->private_data;
1937 
1938 	addr = gfn_to_hva(kvm, gfn);
1939 	if (kvm_is_error_hva(addr))
1940 		return VM_FAULT_SIGBUS;
1941 
1942 	npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1943 				NULL);
1944 	if (unlikely(npages != 1))
1945 		return VM_FAULT_SIGBUS;
1946 
1947 	vmf->page = page[0];
1948 	return 0;
1949 }
1950 
1951 static struct vm_operations_struct kvm_vm_vm_ops = {
1952 	.fault = kvm_vm_fault,
1953 };
1954 
1955 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1956 {
1957 	vma->vm_ops = &kvm_vm_vm_ops;
1958 	return 0;
1959 }
1960 
1961 static struct file_operations kvm_vm_fops = {
1962 	.release        = kvm_vm_release,
1963 	.unlocked_ioctl = kvm_vm_ioctl,
1964 	.compat_ioctl   = kvm_vm_ioctl,
1965 	.mmap           = kvm_vm_mmap,
1966 };
1967 
1968 static int kvm_dev_ioctl_create_vm(void)
1969 {
1970 	int fd;
1971 	struct kvm *kvm;
1972 
1973 	kvm = kvm_create_vm();
1974 	if (IS_ERR(kvm))
1975 		return PTR_ERR(kvm);
1976 	fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1977 	if (fd < 0)
1978 		kvm_put_kvm(kvm);
1979 
1980 	return fd;
1981 }
1982 
1983 static long kvm_dev_ioctl_check_extension_generic(long arg)
1984 {
1985 	switch (arg) {
1986 	case KVM_CAP_USER_MEMORY:
1987 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1988 		return 1;
1989 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1990 	case KVM_CAP_IRQ_ROUTING:
1991 		return KVM_MAX_IRQ_ROUTES;
1992 #endif
1993 	default:
1994 		break;
1995 	}
1996 	return kvm_dev_ioctl_check_extension(arg);
1997 }
1998 
1999 static long kvm_dev_ioctl(struct file *filp,
2000 			  unsigned int ioctl, unsigned long arg)
2001 {
2002 	long r = -EINVAL;
2003 
2004 	switch (ioctl) {
2005 	case KVM_GET_API_VERSION:
2006 		r = -EINVAL;
2007 		if (arg)
2008 			goto out;
2009 		r = KVM_API_VERSION;
2010 		break;
2011 	case KVM_CREATE_VM:
2012 		r = -EINVAL;
2013 		if (arg)
2014 			goto out;
2015 		r = kvm_dev_ioctl_create_vm();
2016 		break;
2017 	case KVM_CHECK_EXTENSION:
2018 		r = kvm_dev_ioctl_check_extension_generic(arg);
2019 		break;
2020 	case KVM_GET_VCPU_MMAP_SIZE:
2021 		r = -EINVAL;
2022 		if (arg)
2023 			goto out;
2024 		r = PAGE_SIZE;     /* struct kvm_run */
2025 #ifdef CONFIG_X86
2026 		r += PAGE_SIZE;    /* pio data page */
2027 #endif
2028 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2029 		r += PAGE_SIZE;    /* coalesced mmio ring page */
2030 #endif
2031 		break;
2032 	case KVM_TRACE_ENABLE:
2033 	case KVM_TRACE_PAUSE:
2034 	case KVM_TRACE_DISABLE:
2035 		r = kvm_trace_ioctl(ioctl, arg);
2036 		break;
2037 	default:
2038 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
2039 	}
2040 out:
2041 	return r;
2042 }
2043 
2044 static struct file_operations kvm_chardev_ops = {
2045 	.unlocked_ioctl = kvm_dev_ioctl,
2046 	.compat_ioctl   = kvm_dev_ioctl,
2047 };
2048 
2049 static struct miscdevice kvm_dev = {
2050 	KVM_MINOR,
2051 	"kvm",
2052 	&kvm_chardev_ops,
2053 };
2054 
2055 static void hardware_enable(void *junk)
2056 {
2057 	int cpu = raw_smp_processor_id();
2058 
2059 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2060 		return;
2061 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
2062 	kvm_arch_hardware_enable(NULL);
2063 }
2064 
2065 static void hardware_disable(void *junk)
2066 {
2067 	int cpu = raw_smp_processor_id();
2068 
2069 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2070 		return;
2071 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2072 	kvm_arch_hardware_disable(NULL);
2073 }
2074 
2075 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2076 			   void *v)
2077 {
2078 	int cpu = (long)v;
2079 
2080 	val &= ~CPU_TASKS_FROZEN;
2081 	switch (val) {
2082 	case CPU_DYING:
2083 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2084 		       cpu);
2085 		hardware_disable(NULL);
2086 		break;
2087 	case CPU_UP_CANCELED:
2088 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2089 		       cpu);
2090 		smp_call_function_single(cpu, hardware_disable, NULL, 1);
2091 		break;
2092 	case CPU_ONLINE:
2093 		printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2094 		       cpu);
2095 		smp_call_function_single(cpu, hardware_enable, NULL, 1);
2096 		break;
2097 	}
2098 	return NOTIFY_OK;
2099 }
2100 
2101 
2102 asmlinkage void kvm_handle_fault_on_reboot(void)
2103 {
2104 	if (kvm_rebooting)
2105 		/* spin while reset goes on */
2106 		while (true)
2107 			;
2108 	/* Fault while not rebooting.  We want the trace. */
2109 	BUG();
2110 }
2111 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2112 
2113 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2114 		      void *v)
2115 {
2116 	if (val == SYS_RESTART) {
2117 		/*
2118 		 * Some (well, at least mine) BIOSes hang on reboot if
2119 		 * in vmx root mode.
2120 		 */
2121 		printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2122 		kvm_rebooting = true;
2123 		on_each_cpu(hardware_disable, NULL, 1);
2124 	}
2125 	return NOTIFY_OK;
2126 }
2127 
2128 static struct notifier_block kvm_reboot_notifier = {
2129 	.notifier_call = kvm_reboot,
2130 	.priority = 0,
2131 };
2132 
2133 void kvm_io_bus_init(struct kvm_io_bus *bus)
2134 {
2135 	memset(bus, 0, sizeof(*bus));
2136 }
2137 
2138 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2139 {
2140 	int i;
2141 
2142 	for (i = 0; i < bus->dev_count; i++) {
2143 		struct kvm_io_device *pos = bus->devs[i];
2144 
2145 		kvm_iodevice_destructor(pos);
2146 	}
2147 }
2148 
2149 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2150 					  gpa_t addr, int len, int is_write)
2151 {
2152 	int i;
2153 
2154 	for (i = 0; i < bus->dev_count; i++) {
2155 		struct kvm_io_device *pos = bus->devs[i];
2156 
2157 		if (pos->in_range(pos, addr, len, is_write))
2158 			return pos;
2159 	}
2160 
2161 	return NULL;
2162 }
2163 
2164 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2165 {
2166 	BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2167 
2168 	bus->devs[bus->dev_count++] = dev;
2169 }
2170 
2171 static struct notifier_block kvm_cpu_notifier = {
2172 	.notifier_call = kvm_cpu_hotplug,
2173 	.priority = 20, /* must be > scheduler priority */
2174 };
2175 
2176 static int vm_stat_get(void *_offset, u64 *val)
2177 {
2178 	unsigned offset = (long)_offset;
2179 	struct kvm *kvm;
2180 
2181 	*val = 0;
2182 	spin_lock(&kvm_lock);
2183 	list_for_each_entry(kvm, &vm_list, vm_list)
2184 		*val += *(u32 *)((void *)kvm + offset);
2185 	spin_unlock(&kvm_lock);
2186 	return 0;
2187 }
2188 
2189 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2190 
2191 static int vcpu_stat_get(void *_offset, u64 *val)
2192 {
2193 	unsigned offset = (long)_offset;
2194 	struct kvm *kvm;
2195 	struct kvm_vcpu *vcpu;
2196 	int i;
2197 
2198 	*val = 0;
2199 	spin_lock(&kvm_lock);
2200 	list_for_each_entry(kvm, &vm_list, vm_list)
2201 		for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2202 			vcpu = kvm->vcpus[i];
2203 			if (vcpu)
2204 				*val += *(u32 *)((void *)vcpu + offset);
2205 		}
2206 	spin_unlock(&kvm_lock);
2207 	return 0;
2208 }
2209 
2210 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2211 
2212 static struct file_operations *stat_fops[] = {
2213 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
2214 	[KVM_STAT_VM]   = &vm_stat_fops,
2215 };
2216 
2217 static void kvm_init_debug(void)
2218 {
2219 	struct kvm_stats_debugfs_item *p;
2220 
2221 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2222 	for (p = debugfs_entries; p->name; ++p)
2223 		p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2224 						(void *)(long)p->offset,
2225 						stat_fops[p->kind]);
2226 }
2227 
2228 static void kvm_exit_debug(void)
2229 {
2230 	struct kvm_stats_debugfs_item *p;
2231 
2232 	for (p = debugfs_entries; p->name; ++p)
2233 		debugfs_remove(p->dentry);
2234 	debugfs_remove(kvm_debugfs_dir);
2235 }
2236 
2237 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2238 {
2239 	hardware_disable(NULL);
2240 	return 0;
2241 }
2242 
2243 static int kvm_resume(struct sys_device *dev)
2244 {
2245 	hardware_enable(NULL);
2246 	return 0;
2247 }
2248 
2249 static struct sysdev_class kvm_sysdev_class = {
2250 	.name = "kvm",
2251 	.suspend = kvm_suspend,
2252 	.resume = kvm_resume,
2253 };
2254 
2255 static struct sys_device kvm_sysdev = {
2256 	.id = 0,
2257 	.cls = &kvm_sysdev_class,
2258 };
2259 
2260 struct page *bad_page;
2261 pfn_t bad_pfn;
2262 
2263 static inline
2264 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2265 {
2266 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
2267 }
2268 
2269 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2270 {
2271 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2272 
2273 	kvm_arch_vcpu_load(vcpu, cpu);
2274 }
2275 
2276 static void kvm_sched_out(struct preempt_notifier *pn,
2277 			  struct task_struct *next)
2278 {
2279 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2280 
2281 	kvm_arch_vcpu_put(vcpu);
2282 }
2283 
2284 int kvm_init(void *opaque, unsigned int vcpu_size,
2285 		  struct module *module)
2286 {
2287 	int r;
2288 	int cpu;
2289 
2290 	kvm_init_debug();
2291 
2292 	r = kvm_arch_init(opaque);
2293 	if (r)
2294 		goto out_fail;
2295 
2296 	bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2297 
2298 	if (bad_page == NULL) {
2299 		r = -ENOMEM;
2300 		goto out;
2301 	}
2302 
2303 	bad_pfn = page_to_pfn(bad_page);
2304 
2305 	if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2306 		r = -ENOMEM;
2307 		goto out_free_0;
2308 	}
2309 
2310 	r = kvm_arch_hardware_setup();
2311 	if (r < 0)
2312 		goto out_free_0a;
2313 
2314 	for_each_online_cpu(cpu) {
2315 		smp_call_function_single(cpu,
2316 				kvm_arch_check_processor_compat,
2317 				&r, 1);
2318 		if (r < 0)
2319 			goto out_free_1;
2320 	}
2321 
2322 	on_each_cpu(hardware_enable, NULL, 1);
2323 	r = register_cpu_notifier(&kvm_cpu_notifier);
2324 	if (r)
2325 		goto out_free_2;
2326 	register_reboot_notifier(&kvm_reboot_notifier);
2327 
2328 	r = sysdev_class_register(&kvm_sysdev_class);
2329 	if (r)
2330 		goto out_free_3;
2331 
2332 	r = sysdev_register(&kvm_sysdev);
2333 	if (r)
2334 		goto out_free_4;
2335 
2336 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
2337 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2338 					   __alignof__(struct kvm_vcpu),
2339 					   0, NULL);
2340 	if (!kvm_vcpu_cache) {
2341 		r = -ENOMEM;
2342 		goto out_free_5;
2343 	}
2344 
2345 	kvm_chardev_ops.owner = module;
2346 	kvm_vm_fops.owner = module;
2347 	kvm_vcpu_fops.owner = module;
2348 
2349 	r = misc_register(&kvm_dev);
2350 	if (r) {
2351 		printk(KERN_ERR "kvm: misc device register failed\n");
2352 		goto out_free;
2353 	}
2354 
2355 	kvm_preempt_ops.sched_in = kvm_sched_in;
2356 	kvm_preempt_ops.sched_out = kvm_sched_out;
2357 #ifndef CONFIG_X86
2358 	msi2intx = 0;
2359 #endif
2360 
2361 	return 0;
2362 
2363 out_free:
2364 	kmem_cache_destroy(kvm_vcpu_cache);
2365 out_free_5:
2366 	sysdev_unregister(&kvm_sysdev);
2367 out_free_4:
2368 	sysdev_class_unregister(&kvm_sysdev_class);
2369 out_free_3:
2370 	unregister_reboot_notifier(&kvm_reboot_notifier);
2371 	unregister_cpu_notifier(&kvm_cpu_notifier);
2372 out_free_2:
2373 	on_each_cpu(hardware_disable, NULL, 1);
2374 out_free_1:
2375 	kvm_arch_hardware_unsetup();
2376 out_free_0a:
2377 	free_cpumask_var(cpus_hardware_enabled);
2378 out_free_0:
2379 	__free_page(bad_page);
2380 out:
2381 	kvm_arch_exit();
2382 	kvm_exit_debug();
2383 out_fail:
2384 	return r;
2385 }
2386 EXPORT_SYMBOL_GPL(kvm_init);
2387 
2388 void kvm_exit(void)
2389 {
2390 	kvm_trace_cleanup();
2391 	misc_deregister(&kvm_dev);
2392 	kmem_cache_destroy(kvm_vcpu_cache);
2393 	sysdev_unregister(&kvm_sysdev);
2394 	sysdev_class_unregister(&kvm_sysdev_class);
2395 	unregister_reboot_notifier(&kvm_reboot_notifier);
2396 	unregister_cpu_notifier(&kvm_cpu_notifier);
2397 	on_each_cpu(hardware_disable, NULL, 1);
2398 	kvm_arch_hardware_unsetup();
2399 	kvm_arch_exit();
2400 	kvm_exit_debug();
2401 	free_cpumask_var(cpus_hardware_enabled);
2402 	__free_page(bad_page);
2403 }
2404 EXPORT_SYMBOL_GPL(kvm_exit);
2405