?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Now it works.
[cbs-scheduler.git] / virt / kvm / kvm_main.c
blob9138d0d75539be80bbf5304d614bf39e8512be03
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18 #include "iodev.h"
19
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44
45 #include <asm/processor.h>
46 #include <asm/io.h>
47 #include <asm/uaccess.h>
48 #include <asm/pgtable.h>
49
50 #ifdef CONFIG_X86
51 #include <asm/msidef.h>
52 #endif
53
54 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
55 #include "coalesced_mmio.h"
56 #endif
57
58 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
59 #include <linux/pci.h>
60 #include <linux/interrupt.h>
61 #include "irq.h"
62 #endif
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 static int msi2intx = 1;
68 module_param(msi2intx, bool, 0);
69
70 DEFINE_SPINLOCK(kvm_lock);
71 LIST_HEAD(vm_list);
72
73 static cpumask_var_t cpus_hardware_enabled;
74
75 struct kmem_cache *kvm_vcpu_cache;
76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
77
78 static __read_mostly struct preempt_ops kvm_preempt_ops;
79
80 struct dentry *kvm_debugfs_dir;
81
82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
83 unsigned long arg);
84
85 static bool kvm_rebooting;
86
87 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
88
89 #ifdef CONFIG_X86
90 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev)
91 {
92 int vcpu_id;
93 struct kvm_vcpu *vcpu;
94 struct kvm_ioapic *ioapic = ioapic_irqchip(dev->kvm);
95 int dest_id = (dev->guest_msi.address_lo & MSI_ADDR_DEST_ID_MASK)
96 >> MSI_ADDR_DEST_ID_SHIFT;
97 int vector = (dev->guest_msi.data & MSI_DATA_VECTOR_MASK)
98 >> MSI_DATA_VECTOR_SHIFT;
99 int dest_mode = test_bit(MSI_ADDR_DEST_MODE_SHIFT,
100 (unsigned long *)&dev->guest_msi.address_lo);
101 int trig_mode = test_bit(MSI_DATA_TRIGGER_SHIFT,
102 (unsigned long *)&dev->guest_msi.data);
103 int delivery_mode = test_bit(MSI_DATA_DELIVERY_MODE_SHIFT,
104 (unsigned long *)&dev->guest_msi.data);
105 u32 deliver_bitmask;
106
107 BUG_ON(!ioapic);
108
109 deliver_bitmask = kvm_ioapic_get_delivery_bitmask(ioapic,
110 dest_id, dest_mode);
111 /* IOAPIC delivery mode value is the same as MSI here */
112 switch (delivery_mode) {
113 case IOAPIC_LOWEST_PRIORITY:
114 vcpu = kvm_get_lowest_prio_vcpu(ioapic->kvm, vector,
115 deliver_bitmask);
116 if (vcpu != NULL)
117 kvm_apic_set_irq(vcpu, vector, trig_mode);
118 else
119 printk(KERN_INFO "kvm: null lowest priority vcpu!\n");
120 break;
121 case IOAPIC_FIXED:
122 for (vcpu_id = 0; deliver_bitmask != 0; vcpu_id++) {
123 if (!(deliver_bitmask & (1 << vcpu_id)))
124 continue;
125 deliver_bitmask &= ~(1 << vcpu_id);
126 vcpu = ioapic->kvm->vcpus[vcpu_id];
127 if (vcpu)
128 kvm_apic_set_irq(vcpu, vector, trig_mode);
129 }
130 break;
131 default:
132 printk(KERN_INFO "kvm: unsupported MSI delivery mode\n");
133 }
134 }
135 #else
136 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev) {}
137 #endif
138
139 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
140 int assigned_dev_id)
141 {
142 struct list_head *ptr;
143 struct kvm_assigned_dev_kernel *match;
144
145 list_for_each(ptr, head) {
146 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
147 if (match->assigned_dev_id == assigned_dev_id)
148 return match;
149 }
150 return NULL;
151 }
152
153 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
154 {
155 struct kvm_assigned_dev_kernel *assigned_dev;
156
157 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
158 interrupt_work);
159
160 /* This is taken to safely inject irq inside the guest. When
161 * the interrupt injection (or the ioapic code) uses a
162 * finer-grained lock, update this
163 */
164 mutex_lock(&assigned_dev->kvm->lock);
165 if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_INTX)
166 kvm_set_irq(assigned_dev->kvm,
167 assigned_dev->irq_source_id,
168 assigned_dev->guest_irq, 1);
169 else if (assigned_dev->irq_requested_type &
170 KVM_ASSIGNED_DEV_GUEST_MSI) {
171 assigned_device_msi_dispatch(assigned_dev);
172 enable_irq(assigned_dev->host_irq);
173 assigned_dev->host_irq_disabled = false;
174 }
175 mutex_unlock(&assigned_dev->kvm->lock);
176 }
177
178 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
179 {
180 struct kvm_assigned_dev_kernel *assigned_dev =
181 (struct kvm_assigned_dev_kernel *) dev_id;
182
183 schedule_work(&assigned_dev->interrupt_work);
184
185 disable_irq_nosync(irq);
186 assigned_dev->host_irq_disabled = true;
187
188 return IRQ_HANDLED;
189 }
190
191 /* Ack the irq line for an assigned device */
192 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
193 {
194 struct kvm_assigned_dev_kernel *dev;
195
196 if (kian->gsi == -1)
197 return;
198
199 dev = container_of(kian, struct kvm_assigned_dev_kernel,
200 ack_notifier);
201
202 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
203
204 /* The guest irq may be shared so this ack may be
205 * from another device.
206 */
207 if (dev->host_irq_disabled) {
208 enable_irq(dev->host_irq);
209 dev->host_irq_disabled = false;
210 }
211 }
212
213 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
214 static void kvm_free_assigned_irq(struct kvm *kvm,
215 struct kvm_assigned_dev_kernel *assigned_dev)
216 {
217 if (!irqchip_in_kernel(kvm))
218 return;
219
220 kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
221
222 if (assigned_dev->irq_source_id != -1)
223 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
224 assigned_dev->irq_source_id = -1;
225
226 if (!assigned_dev->irq_requested_type)
227 return;
228
229 /*
230 * In kvm_free_device_irq, cancel_work_sync return true if:
231 * 1. work is scheduled, and then cancelled.
232 * 2. work callback is executed.
233 *
234 * The first one ensured that the irq is disabled and no more events
235 * would happen. But for the second one, the irq may be enabled (e.g.
236 * for MSI). So we disable irq here to prevent further events.
237 *
238 * Notice this maybe result in nested disable if the interrupt type is
239 * INTx, but it's OK for we are going to free it.
240 *
241 * If this function is a part of VM destroy, please ensure that till
242 * now, the kvm state is still legal for probably we also have to wait
243 * interrupt_work done.
244 */
245 disable_irq_nosync(assigned_dev->host_irq);
246 cancel_work_sync(&assigned_dev->interrupt_work);
247
248 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
249
250 if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
251 pci_disable_msi(assigned_dev->dev);
252
253 assigned_dev->irq_requested_type = 0;
254 }
255
256
257 static void kvm_free_assigned_device(struct kvm *kvm,
258 struct kvm_assigned_dev_kernel
259 *assigned_dev)
260 {
261 kvm_free_assigned_irq(kvm, assigned_dev);
262
263 pci_reset_function(assigned_dev->dev);
264
265 pci_release_regions(assigned_dev->dev);
266 pci_disable_device(assigned_dev->dev);
267 pci_dev_put(assigned_dev->dev);
268
269 list_del(&assigned_dev->list);
270 kfree(assigned_dev);
271 }
272
273 void kvm_free_all_assigned_devices(struct kvm *kvm)
274 {
275 struct list_head *ptr, *ptr2;
276 struct kvm_assigned_dev_kernel *assigned_dev;
277
278 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
279 assigned_dev = list_entry(ptr,
280 struct kvm_assigned_dev_kernel,
281 list);
282
283 kvm_free_assigned_device(kvm, assigned_dev);
284 }
285 }
286
287 static int assigned_device_update_intx(struct kvm *kvm,
288 struct kvm_assigned_dev_kernel *adev,
289 struct kvm_assigned_irq *airq)
290 {
291 adev->guest_irq = airq->guest_irq;
292 adev->ack_notifier.gsi = airq->guest_irq;
293
294 if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
295 return 0;
296
297 if (irqchip_in_kernel(kvm)) {
298 if (!msi2intx &&
299 (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)) {
300 free_irq(adev->host_irq, (void *)adev);
301 pci_disable_msi(adev->dev);
302 }
303
304 if (!capable(CAP_SYS_RAWIO))
305 return -EPERM;
306
307 if (airq->host_irq)
308 adev->host_irq = airq->host_irq;
309 else
310 adev->host_irq = adev->dev->irq;
311
312 /* Even though this is PCI, we don't want to use shared
313 * interrupts. Sharing host devices with guest-assigned devices
314 * on the same interrupt line is not a happy situation: there
315 * are going to be long delays in accepting, acking, etc.
316 */
317 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
318 0, "kvm_assigned_intx_device", (void *)adev))
319 return -EIO;
320 }
321
322 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
323 KVM_ASSIGNED_DEV_HOST_INTX;
324 return 0;
325 }
326
327 #ifdef CONFIG_X86
328 static int assigned_device_update_msi(struct kvm *kvm,
329 struct kvm_assigned_dev_kernel *adev,
330 struct kvm_assigned_irq *airq)
331 {
332 int r;
333
334 if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
335 /* x86 don't care upper address of guest msi message addr */
336 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
337 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
338 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
339 adev->guest_msi.data = airq->guest_msi.data;
340 adev->ack_notifier.gsi = -1;
341 } else if (msi2intx) {
342 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
343 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
344 adev->guest_irq = airq->guest_irq;
345 adev->ack_notifier.gsi = airq->guest_irq;
346 }
347
348 if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
349 return 0;
350
351 if (irqchip_in_kernel(kvm)) {
352 if (!msi2intx) {
353 if (adev->irq_requested_type &
354 KVM_ASSIGNED_DEV_HOST_INTX)
355 free_irq(adev->host_irq, (void *)adev);
356
357 r = pci_enable_msi(adev->dev);
358 if (r)
359 return r;
360 }
361
362 adev->host_irq = adev->dev->irq;
363 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
364 "kvm_assigned_msi_device", (void *)adev))
365 return -EIO;
366 }
367
368 if (!msi2intx)
369 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
370
371 adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
372 return 0;
373 }
374 #endif
375
376 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
377 struct kvm_assigned_irq
378 *assigned_irq)
379 {
380 int r = 0;
381 struct kvm_assigned_dev_kernel *match;
382
383 mutex_lock(&kvm->lock);
384
385 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
386 assigned_irq->assigned_dev_id);
387 if (!match) {
388 mutex_unlock(&kvm->lock);
389 return -EINVAL;
390 }
391
392 if (!match->irq_requested_type) {
393 INIT_WORK(&match->interrupt_work,
394 kvm_assigned_dev_interrupt_work_handler);
395 if (irqchip_in_kernel(kvm)) {
396 /* Register ack nofitier */
397 match->ack_notifier.gsi = -1;
398 match->ack_notifier.irq_acked =
399 kvm_assigned_dev_ack_irq;
400 kvm_register_irq_ack_notifier(kvm,
401 &match->ack_notifier);
402
403 /* Request IRQ source ID */
404 r = kvm_request_irq_source_id(kvm);
405 if (r < 0)
406 goto out_release;
407 else
408 match->irq_source_id = r;
409
410 #ifdef CONFIG_X86
411 /* Determine host device irq type, we can know the
412 * result from dev->msi_enabled */
413 if (msi2intx)
414 pci_enable_msi(match->dev);
415 #endif
416 }
417 }
418
419 if ((!msi2intx &&
420 (assigned_irq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI)) ||
421 (msi2intx && match->dev->msi_enabled)) {
422 #ifdef CONFIG_X86
423 r = assigned_device_update_msi(kvm, match, assigned_irq);
424 if (r) {
425 printk(KERN_WARNING "kvm: failed to enable "
426 "MSI device!\n");
427 goto out_release;
428 }
429 #else
430 r = -ENOTTY;
431 #endif
432 } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
433 /* Host device IRQ 0 means don't support INTx */
434 if (!msi2intx) {
435 printk(KERN_WARNING
436 "kvm: wait device to enable MSI!\n");
437 r = 0;
438 } else {
439 printk(KERN_WARNING
440 "kvm: failed to enable MSI device!\n");
441 r = -ENOTTY;
442 goto out_release;
443 }
444 } else {
445 /* Non-sharing INTx mode */
446 r = assigned_device_update_intx(kvm, match, assigned_irq);
447 if (r) {
448 printk(KERN_WARNING "kvm: failed to enable "
449 "INTx device!\n");
450 goto out_release;
451 }
452 }
453
454 mutex_unlock(&kvm->lock);
455 return r;
456 out_release:
457 mutex_unlock(&kvm->lock);
458 kvm_free_assigned_device(kvm, match);
459 return r;
460 }
461
462 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
463 struct kvm_assigned_pci_dev *assigned_dev)
464 {
465 int r = 0;
466 struct kvm_assigned_dev_kernel *match;
467 struct pci_dev *dev;
468
469 down_read(&kvm->slots_lock);
470 mutex_lock(&kvm->lock);
471
472 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
473 assigned_dev->assigned_dev_id);
474 if (match) {
475 /* device already assigned */
476 r = -EINVAL;
477 goto out;
478 }
479
480 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
481 if (match == NULL) {
482 printk(KERN_INFO "%s: Couldn't allocate memory\n",
483 __func__);
484 r = -ENOMEM;
485 goto out;
486 }
487 dev = pci_get_bus_and_slot(assigned_dev->busnr,
488 assigned_dev->devfn);
489 if (!dev) {
490 printk(KERN_INFO "%s: host device not found\n", __func__);
491 r = -EINVAL;
492 goto out_free;
493 }
494 if (pci_enable_device(dev)) {
495 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
496 r = -EBUSY;
497 goto out_put;
498 }
499 r = pci_request_regions(dev, "kvm_assigned_device");
500 if (r) {
501 printk(KERN_INFO "%s: Could not get access to device regions\n",
502 __func__);
503 goto out_disable;
504 }
505
506 pci_reset_function(dev);
507
508 match->assigned_dev_id = assigned_dev->assigned_dev_id;
509 match->host_busnr = assigned_dev->busnr;
510 match->host_devfn = assigned_dev->devfn;
511 match->flags = assigned_dev->flags;
512 match->dev = dev;
513 match->irq_source_id = -1;
514 match->kvm = kvm;
515
516 list_add(&match->list, &kvm->arch.assigned_dev_head);
517
518 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
519 if (!kvm->arch.iommu_domain) {
520 r = kvm_iommu_map_guest(kvm);
521 if (r)
522 goto out_list_del;
523 }
524 r = kvm_assign_device(kvm, match);
525 if (r)
526 goto out_list_del;
527 }
528
529 out:
530 mutex_unlock(&kvm->lock);
531 up_read(&kvm->slots_lock);
532 return r;
533 out_list_del:
534 list_del(&match->list);
535 pci_release_regions(dev);
536 out_disable:
537 pci_disable_device(dev);
538 out_put:
539 pci_dev_put(dev);
540 out_free:
541 kfree(match);
542 mutex_unlock(&kvm->lock);
543 up_read(&kvm->slots_lock);
544 return r;
545 }
546 #endif
547
548 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
549 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
550 struct kvm_assigned_pci_dev *assigned_dev)
551 {
552 int r = 0;
553 struct kvm_assigned_dev_kernel *match;
554
555 mutex_lock(&kvm->lock);
556
557 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
558 assigned_dev->assigned_dev_id);
559 if (!match) {
560 printk(KERN_INFO "%s: device hasn't been assigned before, "
561 "so cannot be deassigned\n", __func__);
562 r = -EINVAL;
563 goto out;
564 }
565
566 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
567 kvm_deassign_device(kvm, match);
568
569 kvm_free_assigned_device(kvm, match);
570
571 out:
572 mutex_unlock(&kvm->lock);
573 return r;
574 }
575 #endif
576
577 static inline int valid_vcpu(int n)
578 {
579 return likely(n >= 0 && n < KVM_MAX_VCPUS);
580 }
581
582 inline int kvm_is_mmio_pfn(pfn_t pfn)
583 {
584 if (pfn_valid(pfn)) {
585 struct page *page = compound_head(pfn_to_page(pfn));
586 return PageReserved(page);
587 }
588
589 return true;
590 }
591
592 /*
593 * Switches to specified vcpu, until a matching vcpu_put()
594 */
595 void vcpu_load(struct kvm_vcpu *vcpu)
596 {
597 int cpu;
598
599 mutex_lock(&vcpu->mutex);
600 cpu = get_cpu();
601 preempt_notifier_register(&vcpu->preempt_notifier);
602 kvm_arch_vcpu_load(vcpu, cpu);
603 put_cpu();
604 }
605
606 void vcpu_put(struct kvm_vcpu *vcpu)
607 {
608 preempt_disable();
609 kvm_arch_vcpu_put(vcpu);
610 preempt_notifier_unregister(&vcpu->preempt_notifier);
611 preempt_enable();
612 mutex_unlock(&vcpu->mutex);
613 }
614
615 static void ack_flush(void *_completed)
616 {
617 }
618
619 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
620 {
621 int i, cpu, me;
622 cpumask_var_t cpus;
623 bool called = true;
624 struct kvm_vcpu *vcpu;
625
626 if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
627 cpumask_clear(cpus);
628
629 me = get_cpu();
630 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
631 vcpu = kvm->vcpus[i];
632 if (!vcpu)
633 continue;
634 if (test_and_set_bit(req, &vcpu->requests))
635 continue;
636 cpu = vcpu->cpu;
637 if (cpus != NULL && cpu != -1 && cpu != me)
638 cpumask_set_cpu(cpu, cpus);
639 }
640 if (unlikely(cpus == NULL))
641 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
642 else if (!cpumask_empty(cpus))
643 smp_call_function_many(cpus, ack_flush, NULL, 1);
644 else
645 called = false;
646 put_cpu();
647 free_cpumask_var(cpus);
648 return called;
649 }
650
651 void kvm_flush_remote_tlbs(struct kvm *kvm)
652 {
653 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
654 ++kvm->stat.remote_tlb_flush;
655 }
656
657 void kvm_reload_remote_mmus(struct kvm *kvm)
658 {
659 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
660 }
661
662 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
663 {
664 struct page *page;
665 int r;
666
667 mutex_init(&vcpu->mutex);
668 vcpu->cpu = -1;
669 vcpu->kvm = kvm;
670 vcpu->vcpu_id = id;
671 init_waitqueue_head(&vcpu->wq);
672
673 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
674 if (!page) {
675 r = -ENOMEM;
676 goto fail;
677 }
678 vcpu->run = page_address(page);
679
680 r = kvm_arch_vcpu_init(vcpu);
681 if (r < 0)
682 goto fail_free_run;
683 return 0;
684
685 fail_free_run:
686 free_page((unsigned long)vcpu->run);
687 fail:
688 return r;
689 }
690 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
691
692 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
693 {
694 kvm_arch_vcpu_uninit(vcpu);
695 free_page((unsigned long)vcpu->run);
696 }
697 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
698
699 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
700 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
701 {
702 return container_of(mn, struct kvm, mmu_notifier);
703 }
704
705 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
706 struct mm_struct *mm,
707 unsigned long address)
708 {
709 struct kvm *kvm = mmu_notifier_to_kvm(mn);
710 int need_tlb_flush;
711
712 /*
713 * When ->invalidate_page runs, the linux pte has been zapped
714 * already but the page is still allocated until
715 * ->invalidate_page returns. So if we increase the sequence
716 * here the kvm page fault will notice if the spte can't be
717 * established because the page is going to be freed. If
718 * instead the kvm page fault establishes the spte before
719 * ->invalidate_page runs, kvm_unmap_hva will release it
720 * before returning.
721 *
722 * The sequence increase only need to be seen at spin_unlock
723 * time, and not at spin_lock time.
724 *
725 * Increasing the sequence after the spin_unlock would be
726 * unsafe because the kvm page fault could then establish the
727 * pte after kvm_unmap_hva returned, without noticing the page
728 * is going to be freed.
729 */
730 spin_lock(&kvm->mmu_lock);
731 kvm->mmu_notifier_seq++;
732 need_tlb_flush = kvm_unmap_hva(kvm, address);
733 spin_unlock(&kvm->mmu_lock);
734
735 /* we've to flush the tlb before the pages can be freed */
736 if (need_tlb_flush)
737 kvm_flush_remote_tlbs(kvm);
738
739 }
740
741 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
742 struct mm_struct *mm,
743 unsigned long start,
744 unsigned long end)
745 {
746 struct kvm *kvm = mmu_notifier_to_kvm(mn);
747 int need_tlb_flush = 0;
748
749 spin_lock(&kvm->mmu_lock);
750 /*
751 * The count increase must become visible at unlock time as no
752 * spte can be established without taking the mmu_lock and
753 * count is also read inside the mmu_lock critical section.
754 */
755 kvm->mmu_notifier_count++;
756 for (; start < end; start += PAGE_SIZE)
757 need_tlb_flush |= kvm_unmap_hva(kvm, start);
758 spin_unlock(&kvm->mmu_lock);
759
760 /* we've to flush the tlb before the pages can be freed */
761 if (need_tlb_flush)
762 kvm_flush_remote_tlbs(kvm);
763 }
764
765 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
766 struct mm_struct *mm,
767 unsigned long start,
768 unsigned long end)
769 {
770 struct kvm *kvm = mmu_notifier_to_kvm(mn);
771
772 spin_lock(&kvm->mmu_lock);
773 /*
774 * This sequence increase will notify the kvm page fault that
775 * the page that is going to be mapped in the spte could have
776 * been freed.
777 */
778 kvm->mmu_notifier_seq++;
779 /*
780 * The above sequence increase must be visible before the
781 * below count decrease but both values are read by the kvm
782 * page fault under mmu_lock spinlock so we don't need to add
783 * a smb_wmb() here in between the two.
784 */
785 kvm->mmu_notifier_count--;
786 spin_unlock(&kvm->mmu_lock);
787
788 BUG_ON(kvm->mmu_notifier_count < 0);
789 }
790
791 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
792 struct mm_struct *mm,
793 unsigned long address)
794 {
795 struct kvm *kvm = mmu_notifier_to_kvm(mn);
796 int young;
797
798 spin_lock(&kvm->mmu_lock);
799 young = kvm_age_hva(kvm, address);
800 spin_unlock(&kvm->mmu_lock);
801
802 if (young)
803 kvm_flush_remote_tlbs(kvm);
804
805 return young;
806 }
807
808 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
809 struct mm_struct *mm)
810 {
811 struct kvm *kvm = mmu_notifier_to_kvm(mn);
812 kvm_arch_flush_shadow(kvm);
813 }
814
815 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
816 .invalidate_page = kvm_mmu_notifier_invalidate_page,
817 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
818 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
819 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
820 .release = kvm_mmu_notifier_release,
821 };
822 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
823
824 static struct kvm *kvm_create_vm(void)
825 {
826 struct kvm *kvm = kvm_arch_create_vm();
827 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
828 struct page *page;
829 #endif
830
831 if (IS_ERR(kvm))
832 goto out;
833 #ifdef CONFIG_HAVE_KVM_IRQCHIP
834 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
835 #endif
836
837 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
838 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
839 if (!page) {
840 kfree(kvm);
841 return ERR_PTR(-ENOMEM);
842 }
843 kvm->coalesced_mmio_ring =
844 (struct kvm_coalesced_mmio_ring *)page_address(page);
845 #endif
846
847 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
848 {
849 int err;
850 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
851 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
852 if (err) {
853 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
854 put_page(page);
855 #endif
856 kfree(kvm);
857 return ERR_PTR(err);
858 }
859 }
860 #endif
861
862 kvm->mm = current->mm;
863 atomic_inc(&kvm->mm->mm_count);
864 spin_lock_init(&kvm->mmu_lock);
865 kvm_io_bus_init(&kvm->pio_bus);
866 mutex_init(&kvm->lock);
867 kvm_io_bus_init(&kvm->mmio_bus);
868 init_rwsem(&kvm->slots_lock);
869 atomic_set(&kvm->users_count, 1);
870 spin_lock(&kvm_lock);
871 list_add(&kvm->vm_list, &vm_list);
872 spin_unlock(&kvm_lock);
873 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
874 kvm_coalesced_mmio_init(kvm);
875 #endif
876 out:
877 return kvm;
878 }
879
880 /*
881 * Free any memory in @free but not in @dont.
882 */
883 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
884 struct kvm_memory_slot *dont)
885 {
886 if (!dont || free->rmap != dont->rmap)
887 vfree(free->rmap);
888
889 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
890 vfree(free->dirty_bitmap);
891
892 if (!dont || free->lpage_info != dont->lpage_info)
893 vfree(free->lpage_info);
894
895 free->npages = 0;
896 free->dirty_bitmap = NULL;
897 free->rmap = NULL;
898 free->lpage_info = NULL;
899 }
900
901 void kvm_free_physmem(struct kvm *kvm)
902 {
903 int i;
904
905 for (i = 0; i < kvm->nmemslots; ++i)
906 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
907 }
908
909 static void kvm_destroy_vm(struct kvm *kvm)
910 {
911 struct mm_struct *mm = kvm->mm;
912
913 kvm_arch_sync_events(kvm);
914 spin_lock(&kvm_lock);
915 list_del(&kvm->vm_list);
916 spin_unlock(&kvm_lock);
917 kvm_io_bus_destroy(&kvm->pio_bus);
918 kvm_io_bus_destroy(&kvm->mmio_bus);
919 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
920 if (kvm->coalesced_mmio_ring != NULL)
921 free_page((unsigned long)kvm->coalesced_mmio_ring);
922 #endif
923 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
924 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
925 #endif
926 kvm_arch_destroy_vm(kvm);
927 mmdrop(mm);
928 }
929
930 void kvm_get_kvm(struct kvm *kvm)
931 {
932 atomic_inc(&kvm->users_count);
933 }
934 EXPORT_SYMBOL_GPL(kvm_get_kvm);
935
936 void kvm_put_kvm(struct kvm *kvm)
937 {
938 if (atomic_dec_and_test(&kvm->users_count))
939 kvm_destroy_vm(kvm);
940 }
941 EXPORT_SYMBOL_GPL(kvm_put_kvm);
942
943
944 static int kvm_vm_release(struct inode *inode, struct file *filp)
945 {
946 struct kvm *kvm = filp->private_data;
947
948 kvm_put_kvm(kvm);
949 return 0;
950 }
951
952 /*
953 * Allocate some memory and give it an address in the guest physical address
954 * space.
955 *
956 * Discontiguous memory is allowed, mostly for framebuffers.
957 *
958 * Must be called holding mmap_sem for write.
959 */
960 int __kvm_set_memory_region(struct kvm *kvm,
961 struct kvm_userspace_memory_region *mem,
962 int user_alloc)
963 {
964 int r;
965 gfn_t base_gfn;
966 unsigned long npages;
967 int largepages;
968 unsigned long i;
969 struct kvm_memory_slot *memslot;
970 struct kvm_memory_slot old, new;
971
972 r = -EINVAL;
973 /* General sanity checks */
974 if (mem->memory_size & (PAGE_SIZE - 1))
975 goto out;
976 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
977 goto out;
978 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
979 goto out;
980 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
981 goto out;
982 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
983 goto out;
984
985 memslot = &kvm->memslots[mem->slot];
986 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
987 npages = mem->memory_size >> PAGE_SHIFT;
988
989 if (!npages)
990 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
991
992 new = old = *memslot;
993
994 new.base_gfn = base_gfn;
995 new.npages = npages;
996 new.flags = mem->flags;
997
998 /* Disallow changing a memory slot's size. */
999 r = -EINVAL;
1000 if (npages && old.npages && npages != old.npages)
1001 goto out_free;
1002
1003 /* Check for overlaps */
1004 r = -EEXIST;
1005 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1006 struct kvm_memory_slot *s = &kvm->memslots[i];
1007
1008 if (s == memslot || !s->npages)
1009 continue;
1010 if (!((base_gfn + npages <= s->base_gfn) ||
1011 (base_gfn >= s->base_gfn + s->npages)))
1012 goto out_free;
1013 }
1014
1015 /* Free page dirty bitmap if unneeded */
1016 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1017 new.dirty_bitmap = NULL;
1018
1019 r = -ENOMEM;
1020
1021 /* Allocate if a slot is being created */
1022 #ifndef CONFIG_S390
1023 if (npages && !new.rmap) {
1024 new.rmap = vmalloc(npages * sizeof(struct page *));
1025
1026 if (!new.rmap)
1027 goto out_free;
1028
1029 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1030
1031 new.user_alloc = user_alloc;
1032 /*
1033 * hva_to_rmmap() serialzies with the mmu_lock and to be
1034 * safe it has to ignore memslots with !user_alloc &&
1035 * !userspace_addr.
1036 */
1037 if (user_alloc)
1038 new.userspace_addr = mem->userspace_addr;
1039 else
1040 new.userspace_addr = 0;
1041 }
1042 if (npages && !new.lpage_info) {
1043 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1044 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1045
1046 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1047
1048 if (!new.lpage_info)
1049 goto out_free;
1050
1051 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1052
1053 if (base_gfn % KVM_PAGES_PER_HPAGE)
1054 new.lpage_info[0].write_count = 1;
1055 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1056 new.lpage_info[largepages-1].write_count = 1;
1057 }
1058
1059 /* Allocate page dirty bitmap if needed */
1060 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1061 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1062
1063 new.dirty_bitmap = vmalloc(dirty_bytes);
1064 if (!new.dirty_bitmap)
1065 goto out_free;
1066 memset(new.dirty_bitmap, 0, dirty_bytes);
1067 }
1068 #endif /* not defined CONFIG_S390 */
1069
1070 if (!npages)
1071 kvm_arch_flush_shadow(kvm);
1072
1073 spin_lock(&kvm->mmu_lock);
1074 if (mem->slot >= kvm->nmemslots)
1075 kvm->nmemslots = mem->slot + 1;
1076
1077 *memslot = new;
1078 spin_unlock(&kvm->mmu_lock);
1079
1080 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1081 if (r) {
1082 spin_lock(&kvm->mmu_lock);
1083 *memslot = old;
1084 spin_unlock(&kvm->mmu_lock);
1085 goto out_free;
1086 }
1087
1088 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1089 /* Slot deletion case: we have to update the current slot */
1090 if (!npages)
1091 *memslot = old;
1092 #ifdef CONFIG_DMAR
1093 /* map the pages in iommu page table */
1094 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1095 if (r)
1096 goto out;
1097 #endif
1098 return 0;
1099
1100 out_free:
1101 kvm_free_physmem_slot(&new, &old);
1102 out:
1103 return r;
1104
1105 }
1106 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1107
1108 int kvm_set_memory_region(struct kvm *kvm,
1109 struct kvm_userspace_memory_region *mem,
1110 int user_alloc)
1111 {
1112 int r;
1113
1114 down_write(&kvm->slots_lock);
1115 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1116 up_write(&kvm->slots_lock);
1117 return r;
1118 }
1119 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1120
1121 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1122 struct
1123 kvm_userspace_memory_region *mem,
1124 int user_alloc)
1125 {
1126 if (mem->slot >= KVM_MEMORY_SLOTS)
1127 return -EINVAL;
1128 return kvm_set_memory_region(kvm, mem, user_alloc);
1129 }
1130
1131 int kvm_get_dirty_log(struct kvm *kvm,
1132 struct kvm_dirty_log *log, int *is_dirty)
1133 {
1134 struct kvm_memory_slot *memslot;
1135 int r, i;
1136 int n;
1137 unsigned long any = 0;
1138
1139 r = -EINVAL;
1140 if (log->slot >= KVM_MEMORY_SLOTS)
1141 goto out;
1142
1143 memslot = &kvm->memslots[log->slot];
1144 r = -ENOENT;
1145 if (!memslot->dirty_bitmap)
1146 goto out;
1147
1148 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1149
1150 for (i = 0; !any && i < n/sizeof(long); ++i)
1151 any = memslot->dirty_bitmap[i];
1152
1153 r = -EFAULT;
1154 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1155 goto out;
1156
1157 if (any)
1158 *is_dirty = 1;
1159
1160 r = 0;
1161 out:
1162 return r;
1163 }
1164
1165 int is_error_page(struct page *page)
1166 {
1167 return page == bad_page;
1168 }
1169 EXPORT_SYMBOL_GPL(is_error_page);
1170
1171 int is_error_pfn(pfn_t pfn)
1172 {
1173 return pfn == bad_pfn;
1174 }
1175 EXPORT_SYMBOL_GPL(is_error_pfn);
1176
1177 static inline unsigned long bad_hva(void)
1178 {
1179 return PAGE_OFFSET;
1180 }
1181
1182 int kvm_is_error_hva(unsigned long addr)
1183 {
1184 return addr == bad_hva();
1185 }
1186 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1187
1188 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1189 {
1190 int i;
1191
1192 for (i = 0; i < kvm->nmemslots; ++i) {
1193 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1194
1195 if (gfn >= memslot->base_gfn
1196 && gfn < memslot->base_gfn + memslot->npages)
1197 return memslot;
1198 }
1199 return NULL;
1200 }
1201 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1202
1203 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1204 {
1205 gfn = unalias_gfn(kvm, gfn);
1206 return gfn_to_memslot_unaliased(kvm, gfn);
1207 }
1208
1209 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1210 {
1211 int i;
1212
1213 gfn = unalias_gfn(kvm, gfn);
1214 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1215 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1216
1217 if (gfn >= memslot->base_gfn
1218 && gfn < memslot->base_gfn + memslot->npages)
1219 return 1;
1220 }
1221 return 0;
1222 }
1223 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1224
1225 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1226 {
1227 struct kvm_memory_slot *slot;
1228
1229 gfn = unalias_gfn(kvm, gfn);
1230 slot = gfn_to_memslot_unaliased(kvm, gfn);
1231 if (!slot)
1232 return bad_hva();
1233 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1234 }
1235 EXPORT_SYMBOL_GPL(gfn_to_hva);
1236
1237 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1238 {
1239 struct page *page[1];
1240 unsigned long addr;
1241 int npages;
1242 pfn_t pfn;
1243
1244 might_sleep();
1245
1246 addr = gfn_to_hva(kvm, gfn);
1247 if (kvm_is_error_hva(addr)) {
1248 get_page(bad_page);
1249 return page_to_pfn(bad_page);
1250 }
1251
1252 npages = get_user_pages_fast(addr, 1, 1, page);
1253
1254 if (unlikely(npages != 1)) {
1255 struct vm_area_struct *vma;
1256
1257 down_read(&current->mm->mmap_sem);
1258 vma = find_vma(current->mm, addr);
1259
1260 if (vma == NULL || addr < vma->vm_start ||
1261 !(vma->vm_flags & VM_PFNMAP)) {
1262 up_read(&current->mm->mmap_sem);
1263 get_page(bad_page);
1264 return page_to_pfn(bad_page);
1265 }
1266
1267 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1268 up_read(&current->mm->mmap_sem);
1269 BUG_ON(!kvm_is_mmio_pfn(pfn));
1270 } else
1271 pfn = page_to_pfn(page[0]);
1272
1273 return pfn;
1274 }
1275
1276 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1277
1278 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1279 {
1280 pfn_t pfn;
1281
1282 pfn = gfn_to_pfn(kvm, gfn);
1283 if (!kvm_is_mmio_pfn(pfn))
1284 return pfn_to_page(pfn);
1285
1286 WARN_ON(kvm_is_mmio_pfn(pfn));
1287
1288 get_page(bad_page);
1289 return bad_page;
1290 }
1291
1292 EXPORT_SYMBOL_GPL(gfn_to_page);
1293
1294 void kvm_release_page_clean(struct page *page)
1295 {
1296 kvm_release_pfn_clean(page_to_pfn(page));
1297 }
1298 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1299
1300 void kvm_release_pfn_clean(pfn_t pfn)
1301 {
1302 if (!kvm_is_mmio_pfn(pfn))
1303 put_page(pfn_to_page(pfn));
1304 }
1305 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1306
1307 void kvm_release_page_dirty(struct page *page)
1308 {
1309 kvm_release_pfn_dirty(page_to_pfn(page));
1310 }
1311 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1312
1313 void kvm_release_pfn_dirty(pfn_t pfn)
1314 {
1315 kvm_set_pfn_dirty(pfn);
1316 kvm_release_pfn_clean(pfn);
1317 }
1318 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1319
1320 void kvm_set_page_dirty(struct page *page)
1321 {
1322 kvm_set_pfn_dirty(page_to_pfn(page));
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1325
1326 void kvm_set_pfn_dirty(pfn_t pfn)
1327 {
1328 if (!kvm_is_mmio_pfn(pfn)) {
1329 struct page *page = pfn_to_page(pfn);
1330 if (!PageReserved(page))
1331 SetPageDirty(page);
1332 }
1333 }
1334 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1335
1336 void kvm_set_pfn_accessed(pfn_t pfn)
1337 {
1338 if (!kvm_is_mmio_pfn(pfn))
1339 mark_page_accessed(pfn_to_page(pfn));
1340 }
1341 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1342
1343 void kvm_get_pfn(pfn_t pfn)
1344 {
1345 if (!kvm_is_mmio_pfn(pfn))
1346 get_page(pfn_to_page(pfn));
1347 }
1348 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1349
1350 static int next_segment(unsigned long len, int offset)
1351 {
1352 if (len > PAGE_SIZE - offset)
1353 return PAGE_SIZE - offset;
1354 else
1355 return len;
1356 }
1357
1358 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1359 int len)
1360 {
1361 int r;
1362 unsigned long addr;
1363
1364 addr = gfn_to_hva(kvm, gfn);
1365 if (kvm_is_error_hva(addr))
1366 return -EFAULT;
1367 r = copy_from_user(data, (void __user *)addr + offset, len);
1368 if (r)
1369 return -EFAULT;
1370 return 0;
1371 }
1372 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1373
1374 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1375 {
1376 gfn_t gfn = gpa >> PAGE_SHIFT;
1377 int seg;
1378 int offset = offset_in_page(gpa);
1379 int ret;
1380
1381 while ((seg = next_segment(len, offset)) != 0) {
1382 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1383 if (ret < 0)
1384 return ret;
1385 offset = 0;
1386 len -= seg;
1387 data += seg;
1388 ++gfn;
1389 }
1390 return 0;
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_read_guest);
1393
1394 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1395 unsigned long len)
1396 {
1397 int r;
1398 unsigned long addr;
1399 gfn_t gfn = gpa >> PAGE_SHIFT;
1400 int offset = offset_in_page(gpa);
1401
1402 addr = gfn_to_hva(kvm, gfn);
1403 if (kvm_is_error_hva(addr))
1404 return -EFAULT;
1405 pagefault_disable();
1406 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1407 pagefault_enable();
1408 if (r)
1409 return -EFAULT;
1410 return 0;
1411 }
1412 EXPORT_SYMBOL(kvm_read_guest_atomic);
1413
1414 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1415 int offset, int len)
1416 {
1417 int r;
1418 unsigned long addr;
1419
1420 addr = gfn_to_hva(kvm, gfn);
1421 if (kvm_is_error_hva(addr))
1422 return -EFAULT;
1423 r = copy_to_user((void __user *)addr + offset, data, len);
1424 if (r)
1425 return -EFAULT;
1426 mark_page_dirty(kvm, gfn);
1427 return 0;
1428 }
1429 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1430
1431 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1432 unsigned long len)
1433 {
1434 gfn_t gfn = gpa >> PAGE_SHIFT;
1435 int seg;
1436 int offset = offset_in_page(gpa);
1437 int ret;
1438
1439 while ((seg = next_segment(len, offset)) != 0) {
1440 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1441 if (ret < 0)
1442 return ret;
1443 offset = 0;
1444 len -= seg;
1445 data += seg;
1446 ++gfn;
1447 }
1448 return 0;
1449 }
1450
1451 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1452 {
1453 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1454 }
1455 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1456
1457 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1458 {
1459 gfn_t gfn = gpa >> PAGE_SHIFT;
1460 int seg;
1461 int offset = offset_in_page(gpa);
1462 int ret;
1463
1464 while ((seg = next_segment(len, offset)) != 0) {
1465 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1466 if (ret < 0)
1467 return ret;
1468 offset = 0;
1469 len -= seg;
1470 ++gfn;
1471 }
1472 return 0;
1473 }
1474 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1475
1476 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1477 {
1478 struct kvm_memory_slot *memslot;
1479
1480 gfn = unalias_gfn(kvm, gfn);
1481 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1482 if (memslot && memslot->dirty_bitmap) {
1483 unsigned long rel_gfn = gfn - memslot->base_gfn;
1484
1485 /* avoid RMW */
1486 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1487 set_bit(rel_gfn, memslot->dirty_bitmap);
1488 }
1489 }
1490
1491 /*
1492 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1493 */
1494 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1495 {
1496 DEFINE_WAIT(wait);
1497
1498 for (;;) {
1499 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1500
1501 if (kvm_cpu_has_interrupt(vcpu) ||
1502 kvm_cpu_has_pending_timer(vcpu) ||
1503 kvm_arch_vcpu_runnable(vcpu)) {
1504 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1505 break;
1506 }
1507 if (signal_pending(current))
1508 break;
1509
1510 vcpu_put(vcpu);
1511 schedule();
1512 vcpu_load(vcpu);
1513 }
1514
1515 finish_wait(&vcpu->wq, &wait);
1516 }
1517
1518 void kvm_resched(struct kvm_vcpu *vcpu)
1519 {
1520 if (!need_resched())
1521 return;
1522 cond_resched();
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_resched);
1525
1526 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1527 {
1528 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1529 struct page *page;
1530
1531 if (vmf->pgoff == 0)
1532 page = virt_to_page(vcpu->run);
1533 #ifdef CONFIG_X86
1534 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1535 page = virt_to_page(vcpu->arch.pio_data);
1536 #endif
1537 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1538 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1539 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1540 #endif
1541 else
1542 return VM_FAULT_SIGBUS;
1543 get_page(page);
1544 vmf->page = page;
1545 return 0;
1546 }
1547
1548 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1549 .fault = kvm_vcpu_fault,
1550 };
1551
1552 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1553 {
1554 vma->vm_ops = &kvm_vcpu_vm_ops;
1555 return 0;
1556 }
1557
1558 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1559 {
1560 struct kvm_vcpu *vcpu = filp->private_data;
1561
1562 kvm_put_kvm(vcpu->kvm);
1563 return 0;
1564 }
1565
1566 static struct file_operations kvm_vcpu_fops = {
1567 .release = kvm_vcpu_release,
1568 .unlocked_ioctl = kvm_vcpu_ioctl,
1569 .compat_ioctl = kvm_vcpu_ioctl,
1570 .mmap = kvm_vcpu_mmap,
1571 };
1572
1573 /*
1574 * Allocates an inode for the vcpu.
1575 */
1576 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1577 {
1578 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1579 if (fd < 0)
1580 kvm_put_kvm(vcpu->kvm);
1581 return fd;
1582 }
1583
1584 /*
1585 * Creates some virtual cpus. Good luck creating more than one.
1586 */
1587 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1588 {
1589 int r;
1590 struct kvm_vcpu *vcpu;
1591
1592 if (!valid_vcpu(n))
1593 return -EINVAL;
1594
1595 vcpu = kvm_arch_vcpu_create(kvm, n);
1596 if (IS_ERR(vcpu))
1597 return PTR_ERR(vcpu);
1598
1599 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1600
1601 r = kvm_arch_vcpu_setup(vcpu);
1602 if (r)
1603 return r;
1604
1605 mutex_lock(&kvm->lock);
1606 if (kvm->vcpus[n]) {
1607 r = -EEXIST;
1608 goto vcpu_destroy;
1609 }
1610 kvm->vcpus[n] = vcpu;
1611 mutex_unlock(&kvm->lock);
1612
1613 /* Now it's all set up, let userspace reach it */
1614 kvm_get_kvm(kvm);
1615 r = create_vcpu_fd(vcpu);
1616 if (r < 0)
1617 goto unlink;
1618 return r;
1619
1620 unlink:
1621 mutex_lock(&kvm->lock);
1622 kvm->vcpus[n] = NULL;
1623 vcpu_destroy:
1624 mutex_unlock(&kvm->lock);
1625 kvm_arch_vcpu_destroy(vcpu);
1626 return r;
1627 }
1628
1629 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1630 {
1631 if (sigset) {
1632 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1633 vcpu->sigset_active = 1;
1634 vcpu->sigset = *sigset;
1635 } else
1636 vcpu->sigset_active = 0;
1637 return 0;
1638 }
1639
1640 static long kvm_vcpu_ioctl(struct file *filp,
1641 unsigned int ioctl, unsigned long arg)
1642 {
1643 struct kvm_vcpu *vcpu = filp->private_data;
1644 void __user *argp = (void __user *)arg;
1645 int r;
1646 struct kvm_fpu *fpu = NULL;
1647 struct kvm_sregs *kvm_sregs = NULL;
1648
1649 if (vcpu->kvm->mm != current->mm)
1650 return -EIO;
1651 switch (ioctl) {
1652 case KVM_RUN:
1653 r = -EINVAL;
1654 if (arg)
1655 goto out;
1656 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1657 break;
1658 case KVM_GET_REGS: {
1659 struct kvm_regs *kvm_regs;
1660
1661 r = -ENOMEM;
1662 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1663 if (!kvm_regs)
1664 goto out;
1665 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1666 if (r)
1667 goto out_free1;
1668 r = -EFAULT;
1669 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1670 goto out_free1;
1671 r = 0;
1672 out_free1:
1673 kfree(kvm_regs);
1674 break;
1675 }
1676 case KVM_SET_REGS: {
1677 struct kvm_regs *kvm_regs;
1678
1679 r = -ENOMEM;
1680 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1681 if (!kvm_regs)
1682 goto out;
1683 r = -EFAULT;
1684 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1685 goto out_free2;
1686 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1687 if (r)
1688 goto out_free2;
1689 r = 0;
1690 out_free2:
1691 kfree(kvm_regs);
1692 break;
1693 }
1694 case KVM_GET_SREGS: {
1695 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1696 r = -ENOMEM;
1697 if (!kvm_sregs)
1698 goto out;
1699 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1700 if (r)
1701 goto out;
1702 r = -EFAULT;
1703 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1704 goto out;
1705 r = 0;
1706 break;
1707 }
1708 case KVM_SET_SREGS: {
1709 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1710 r = -ENOMEM;
1711 if (!kvm_sregs)
1712 goto out;
1713 r = -EFAULT;
1714 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1715 goto out;
1716 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1717 if (r)
1718 goto out;
1719 r = 0;
1720 break;
1721 }
1722 case KVM_GET_MP_STATE: {
1723 struct kvm_mp_state mp_state;
1724
1725 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1726 if (r)
1727 goto out;
1728 r = -EFAULT;
1729 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1730 goto out;
1731 r = 0;
1732 break;
1733 }
1734 case KVM_SET_MP_STATE: {
1735 struct kvm_mp_state mp_state;
1736
1737 r = -EFAULT;
1738 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1739 goto out;
1740 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1741 if (r)
1742 goto out;
1743 r = 0;
1744 break;
1745 }
1746 case KVM_TRANSLATE: {
1747 struct kvm_translation tr;
1748
1749 r = -EFAULT;
1750 if (copy_from_user(&tr, argp, sizeof tr))
1751 goto out;
1752 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1753 if (r)
1754 goto out;
1755 r = -EFAULT;
1756 if (copy_to_user(argp, &tr, sizeof tr))
1757 goto out;
1758 r = 0;
1759 break;
1760 }
1761 case KVM_DEBUG_GUEST: {
1762 struct kvm_debug_guest dbg;
1763
1764 r = -EFAULT;
1765 if (copy_from_user(&dbg, argp, sizeof dbg))
1766 goto out;
1767 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1768 if (r)
1769 goto out;
1770 r = 0;
1771 break;
1772 }
1773 case KVM_SET_SIGNAL_MASK: {
1774 struct kvm_signal_mask __user *sigmask_arg = argp;
1775 struct kvm_signal_mask kvm_sigmask;
1776 sigset_t sigset, *p;
1777
1778 p = NULL;
1779 if (argp) {
1780 r = -EFAULT;
1781 if (copy_from_user(&kvm_sigmask, argp,
1782 sizeof kvm_sigmask))
1783 goto out;
1784 r = -EINVAL;
1785 if (kvm_sigmask.len != sizeof sigset)
1786 goto out;
1787 r = -EFAULT;
1788 if (copy_from_user(&sigset, sigmask_arg->sigset,
1789 sizeof sigset))
1790 goto out;
1791 p = &sigset;
1792 }
1793 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1794 break;
1795 }
1796 case KVM_GET_FPU: {
1797 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1798 r = -ENOMEM;
1799 if (!fpu)
1800 goto out;
1801 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1802 if (r)
1803 goto out;
1804 r = -EFAULT;
1805 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1806 goto out;
1807 r = 0;
1808 break;
1809 }
1810 case KVM_SET_FPU: {
1811 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1812 r = -ENOMEM;
1813 if (!fpu)
1814 goto out;
1815 r = -EFAULT;
1816 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1817 goto out;
1818 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1819 if (r)
1820 goto out;
1821 r = 0;
1822 break;
1823 }
1824 default:
1825 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1826 }
1827 out:
1828 kfree(fpu);
1829 kfree(kvm_sregs);
1830 return r;
1831 }
1832
1833 static long kvm_vm_ioctl(struct file *filp,
1834 unsigned int ioctl, unsigned long arg)
1835 {
1836 struct kvm *kvm = filp->private_data;
1837 void __user *argp = (void __user *)arg;
1838 int r;
1839
1840 if (kvm->mm != current->mm)
1841 return -EIO;
1842 switch (ioctl) {
1843 case KVM_CREATE_VCPU:
1844 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1845 if (r < 0)
1846 goto out;
1847 break;
1848 case KVM_SET_USER_MEMORY_REGION: {
1849 struct kvm_userspace_memory_region kvm_userspace_mem;
1850
1851 r = -EFAULT;
1852 if (copy_from_user(&kvm_userspace_mem, argp,
1853 sizeof kvm_userspace_mem))
1854 goto out;
1855
1856 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1857 if (r)
1858 goto out;
1859 break;
1860 }
1861 case KVM_GET_DIRTY_LOG: {
1862 struct kvm_dirty_log log;
1863
1864 r = -EFAULT;
1865 if (copy_from_user(&log, argp, sizeof log))
1866 goto out;
1867 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1868 if (r)
1869 goto out;
1870 break;
1871 }
1872 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1873 case KVM_REGISTER_COALESCED_MMIO: {
1874 struct kvm_coalesced_mmio_zone zone;
1875 r = -EFAULT;
1876 if (copy_from_user(&zone, argp, sizeof zone))
1877 goto out;
1878 r = -ENXIO;
1879 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1880 if (r)
1881 goto out;
1882 r = 0;
1883 break;
1884 }
1885 case KVM_UNREGISTER_COALESCED_MMIO: {
1886 struct kvm_coalesced_mmio_zone zone;
1887 r = -EFAULT;
1888 if (copy_from_user(&zone, argp, sizeof zone))
1889 goto out;
1890 r = -ENXIO;
1891 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1892 if (r)
1893 goto out;
1894 r = 0;
1895 break;
1896 }
1897 #endif
1898 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1899 case KVM_ASSIGN_PCI_DEVICE: {
1900 struct kvm_assigned_pci_dev assigned_dev;
1901
1902 r = -EFAULT;
1903 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1904 goto out;
1905 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1906 if (r)
1907 goto out;
1908 break;
1909 }
1910 case KVM_ASSIGN_IRQ: {
1911 struct kvm_assigned_irq assigned_irq;
1912
1913 r = -EFAULT;
1914 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1915 goto out;
1916 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1917 if (r)
1918 goto out;
1919 break;
1920 }
1921 #endif
1922 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1923 case KVM_DEASSIGN_PCI_DEVICE: {
1924 struct kvm_assigned_pci_dev assigned_dev;
1925
1926 r = -EFAULT;
1927 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1928 goto out;
1929 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1930 if (r)
1931 goto out;
1932 break;
1933 }
1934 #endif
1935 default:
1936 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1937 }
1938 out:
1939 return r;
1940 }
1941
1942 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1943 {
1944 struct page *page[1];
1945 unsigned long addr;
1946 int npages;
1947 gfn_t gfn = vmf->pgoff;
1948 struct kvm *kvm = vma->vm_file->private_data;
1949
1950 addr = gfn_to_hva(kvm, gfn);
1951 if (kvm_is_error_hva(addr))
1952 return VM_FAULT_SIGBUS;
1953
1954 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1955 NULL);
1956 if (unlikely(npages != 1))
1957 return VM_FAULT_SIGBUS;
1958
1959 vmf->page = page[0];
1960 return 0;
1961 }
1962
1963 static struct vm_operations_struct kvm_vm_vm_ops = {
1964 .fault = kvm_vm_fault,
1965 };
1966
1967 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1968 {
1969 vma->vm_ops = &kvm_vm_vm_ops;
1970 return 0;
1971 }
1972
1973 static struct file_operations kvm_vm_fops = {
1974 .release = kvm_vm_release,
1975 .unlocked_ioctl = kvm_vm_ioctl,
1976 .compat_ioctl = kvm_vm_ioctl,
1977 .mmap = kvm_vm_mmap,
1978 };
1979
1980 static int kvm_dev_ioctl_create_vm(void)
1981 {
1982 int fd;
1983 struct kvm *kvm;
1984
1985 kvm = kvm_create_vm();
1986 if (IS_ERR(kvm))
1987 return PTR_ERR(kvm);
1988 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1989 if (fd < 0)
1990 kvm_put_kvm(kvm);
1991
1992 return fd;
1993 }
1994
1995 static long kvm_dev_ioctl_check_extension_generic(long arg)
1996 {
1997 switch (arg) {
1998 case KVM_CAP_USER_MEMORY:
1999 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2000 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2001 return 1;
2002 default:
2003 break;
2004 }
2005 return kvm_dev_ioctl_check_extension(arg);
2006 }
2007
2008 static long kvm_dev_ioctl(struct file *filp,
2009 unsigned int ioctl, unsigned long arg)
2010 {
2011 long r = -EINVAL;
2012
2013 switch (ioctl) {
2014 case KVM_GET_API_VERSION:
2015 r = -EINVAL;
2016 if (arg)
2017 goto out;
2018 r = KVM_API_VERSION;
2019 break;
2020 case KVM_CREATE_VM:
2021 r = -EINVAL;
2022 if (arg)
2023 goto out;
2024 r = kvm_dev_ioctl_create_vm();
2025 break;
2026 case KVM_CHECK_EXTENSION:
2027 r = kvm_dev_ioctl_check_extension_generic(arg);
2028 break;
2029 case KVM_GET_VCPU_MMAP_SIZE:
2030 r = -EINVAL;
2031 if (arg)
2032 goto out;
2033 r = PAGE_SIZE; /* struct kvm_run */
2034 #ifdef CONFIG_X86
2035 r += PAGE_SIZE; /* pio data page */
2036 #endif
2037 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2038 r += PAGE_SIZE; /* coalesced mmio ring page */
2039 #endif
2040 break;
2041 case KVM_TRACE_ENABLE:
2042 case KVM_TRACE_PAUSE:
2043 case KVM_TRACE_DISABLE:
2044 r = kvm_trace_ioctl(ioctl, arg);
2045 break;
2046 default:
2047 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2048 }
2049 out:
2050 return r;
2051 }
2052
2053 static struct file_operations kvm_chardev_ops = {
2054 .unlocked_ioctl = kvm_dev_ioctl,
2055 .compat_ioctl = kvm_dev_ioctl,
2056 };
2057
2058 static struct miscdevice kvm_dev = {
2059 KVM_MINOR,
2060 "kvm",
2061 &kvm_chardev_ops,
2062 };
2063
2064 static void hardware_enable(void *junk)
2065 {
2066 int cpu = raw_smp_processor_id();
2067
2068 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2069 return;
2070 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2071 kvm_arch_hardware_enable(NULL);
2072 }
2073
2074 static void hardware_disable(void *junk)
2075 {
2076 int cpu = raw_smp_processor_id();
2077
2078 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2079 return;
2080 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2081 kvm_arch_hardware_disable(NULL);
2082 }
2083
2084 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2085 void *v)
2086 {
2087 int cpu = (long)v;
2088
2089 val &= ~CPU_TASKS_FROZEN;
2090 switch (val) {
2091 case CPU_DYING:
2092 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2093 cpu);
2094 hardware_disable(NULL);
2095 break;
2096 case CPU_UP_CANCELED:
2097 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2098 cpu);
2099 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2100 break;
2101 case CPU_ONLINE:
2102 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2103 cpu);
2104 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2105 break;
2106 }
2107 return NOTIFY_OK;
2108 }
2109
2110
2111 asmlinkage void kvm_handle_fault_on_reboot(void)
2112 {
2113 if (kvm_rebooting)
2114 /* spin while reset goes on */
2115 while (true)
2116 ;
2117 /* Fault while not rebooting. We want the trace. */
2118 BUG();
2119 }
2120 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2121
2122 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2123 void *v)
2124 {
2125 if (val == SYS_RESTART) {
2126 /*
2127 * Some (well, at least mine) BIOSes hang on reboot if
2128 * in vmx root mode.
2129 */
2130 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2131 kvm_rebooting = true;
2132 on_each_cpu(hardware_disable, NULL, 1);
2133 }
2134 return NOTIFY_OK;
2135 }
2136
2137 static struct notifier_block kvm_reboot_notifier = {
2138 .notifier_call = kvm_reboot,
2139 .priority = 0,
2140 };
2141
2142 void kvm_io_bus_init(struct kvm_io_bus *bus)
2143 {
2144 memset(bus, 0, sizeof(*bus));
2145 }
2146
2147 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2148 {
2149 int i;
2150
2151 for (i = 0; i < bus->dev_count; i++) {
2152 struct kvm_io_device *pos = bus->devs[i];
2153
2154 kvm_iodevice_destructor(pos);
2155 }
2156 }
2157
2158 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2159 gpa_t addr, int len, int is_write)
2160 {
2161 int i;
2162
2163 for (i = 0; i < bus->dev_count; i++) {
2164 struct kvm_io_device *pos = bus->devs[i];
2165
2166 if (pos->in_range(pos, addr, len, is_write))
2167 return pos;
2168 }
2169
2170 return NULL;
2171 }
2172
2173 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2174 {
2175 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2176
2177 bus->devs[bus->dev_count++] = dev;
2178 }
2179
2180 static struct notifier_block kvm_cpu_notifier = {
2181 .notifier_call = kvm_cpu_hotplug,
2182 .priority = 20, /* must be > scheduler priority */
2183 };
2184
2185 static int vm_stat_get(void *_offset, u64 *val)
2186 {
2187 unsigned offset = (long)_offset;
2188 struct kvm *kvm;
2189
2190 *val = 0;
2191 spin_lock(&kvm_lock);
2192 list_for_each_entry(kvm, &vm_list, vm_list)
2193 *val += *(u32 *)((void *)kvm + offset);
2194 spin_unlock(&kvm_lock);
2195 return 0;
2196 }
2197
2198 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2199
2200 static int vcpu_stat_get(void *_offset, u64 *val)
2201 {
2202 unsigned offset = (long)_offset;
2203 struct kvm *kvm;
2204 struct kvm_vcpu *vcpu;
2205 int i;
2206
2207 *val = 0;
2208 spin_lock(&kvm_lock);
2209 list_for_each_entry(kvm, &vm_list, vm_list)
2210 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2211 vcpu = kvm->vcpus[i];
2212 if (vcpu)
2213 *val += *(u32 *)((void *)vcpu + offset);
2214 }
2215 spin_unlock(&kvm_lock);
2216 return 0;
2217 }
2218
2219 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2220
2221 static struct file_operations *stat_fops[] = {
2222 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2223 [KVM_STAT_VM] = &vm_stat_fops,
2224 };
2225
2226 static void kvm_init_debug(void)
2227 {
2228 struct kvm_stats_debugfs_item *p;
2229
2230 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2231 for (p = debugfs_entries; p->name; ++p)
2232 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2233 (void *)(long)p->offset,
2234 stat_fops[p->kind]);
2235 }
2236
2237 static void kvm_exit_debug(void)
2238 {
2239 struct kvm_stats_debugfs_item *p;
2240
2241 for (p = debugfs_entries; p->name; ++p)
2242 debugfs_remove(p->dentry);
2243 debugfs_remove(kvm_debugfs_dir);
2244 }
2245
2246 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2247 {
2248 hardware_disable(NULL);
2249 return 0;
2250 }
2251
2252 static int kvm_resume(struct sys_device *dev)
2253 {
2254 hardware_enable(NULL);
2255 return 0;
2256 }
2257
2258 static struct sysdev_class kvm_sysdev_class = {
2259 .name = "kvm",
2260 .suspend = kvm_suspend,
2261 .resume = kvm_resume,
2262 };
2263
2264 static struct sys_device kvm_sysdev = {
2265 .id = 0,
2266 .cls = &kvm_sysdev_class,
2267 };
2268
2269 struct page *bad_page;
2270 pfn_t bad_pfn;
2271
2272 static inline
2273 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2274 {
2275 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2276 }
2277
2278 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2279 {
2280 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2281
2282 kvm_arch_vcpu_load(vcpu, cpu);
2283 }
2284
2285 static void kvm_sched_out(struct preempt_notifier *pn,
2286 struct task_struct *next)
2287 {
2288 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2289
2290 kvm_arch_vcpu_put(vcpu);
2291 }
2292
2293 int kvm_init(void *opaque, unsigned int vcpu_size,
2294 struct module *module)
2295 {
2296 int r;
2297 int cpu;
2298
2299 kvm_init_debug();
2300
2301 r = kvm_arch_init(opaque);
2302 if (r)
2303 goto out_fail;
2304
2305 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2306
2307 if (bad_page == NULL) {
2308 r = -ENOMEM;
2309 goto out;
2310 }
2311
2312 bad_pfn = page_to_pfn(bad_page);
2313
2314 if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2315 r = -ENOMEM;
2316 goto out_free_0;
2317 }
2318 cpumask_clear(cpus_hardware_enabled);
2319
2320 r = kvm_arch_hardware_setup();
2321 if (r < 0)
2322 goto out_free_0a;
2323
2324 for_each_online_cpu(cpu) {
2325 smp_call_function_single(cpu,
2326 kvm_arch_check_processor_compat,
2327 &r, 1);
2328 if (r < 0)
2329 goto out_free_1;
2330 }
2331
2332 on_each_cpu(hardware_enable, NULL, 1);
2333 r = register_cpu_notifier(&kvm_cpu_notifier);
2334 if (r)
2335 goto out_free_2;
2336 register_reboot_notifier(&kvm_reboot_notifier);
2337
2338 r = sysdev_class_register(&kvm_sysdev_class);
2339 if (r)
2340 goto out_free_3;
2341
2342 r = sysdev_register(&kvm_sysdev);
2343 if (r)
2344 goto out_free_4;
2345
2346 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2347 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2348 __alignof__(struct kvm_vcpu),
2349 0, NULL);
2350 if (!kvm_vcpu_cache) {
2351 r = -ENOMEM;
2352 goto out_free_5;
2353 }
2354
2355 kvm_chardev_ops.owner = module;
2356 kvm_vm_fops.owner = module;
2357 kvm_vcpu_fops.owner = module;
2358
2359 r = misc_register(&kvm_dev);
2360 if (r) {
2361 printk(KERN_ERR "kvm: misc device register failed\n");
2362 goto out_free;
2363 }
2364
2365 kvm_preempt_ops.sched_in = kvm_sched_in;
2366 kvm_preempt_ops.sched_out = kvm_sched_out;
2367 #ifndef CONFIG_X86
2368 msi2intx = 0;
2369 #endif
2370
2371 return 0;
2372
2373 out_free:
2374 kmem_cache_destroy(kvm_vcpu_cache);
2375 out_free_5:
2376 sysdev_unregister(&kvm_sysdev);
2377 out_free_4:
2378 sysdev_class_unregister(&kvm_sysdev_class);
2379 out_free_3:
2380 unregister_reboot_notifier(&kvm_reboot_notifier);
2381 unregister_cpu_notifier(&kvm_cpu_notifier);
2382 out_free_2:
2383 on_each_cpu(hardware_disable, NULL, 1);
2384 out_free_1:
2385 kvm_arch_hardware_unsetup();
2386 out_free_0a:
2387 free_cpumask_var(cpus_hardware_enabled);
2388 out_free_0:
2389 __free_page(bad_page);
2390 out:
2391 kvm_arch_exit();
2392 kvm_exit_debug();
2393 out_fail:
2394 return r;
2395 }
2396 EXPORT_SYMBOL_GPL(kvm_init);
2397
2398 void kvm_exit(void)
2399 {
2400 kvm_trace_cleanup();
2401 misc_deregister(&kvm_dev);
2402 kmem_cache_destroy(kvm_vcpu_cache);
2403 sysdev_unregister(&kvm_sysdev);
2404 sysdev_class_unregister(&kvm_sysdev_class);
2405 unregister_reboot_notifier(&kvm_reboot_notifier);
2406 unregister_cpu_notifier(&kvm_cpu_notifier);
2407 on_each_cpu(hardware_disable, NULL, 1);
2408 kvm_arch_hardware_unsetup();
2409 kvm_arch_exit();
2410 kvm_exit_debug();
2411 free_cpumask_var(cpus_hardware_enabled);
2412 __free_page(bad_page);
2413 }
2414 EXPORT_SYMBOL_GPL(kvm_exit);
A cbs scheduler for rt-kernel