?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: MMU: handle compound pages in kvm_is_mmio_pfn
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / virt / kvm / kvm_main.c
blob65c00b34248fd29c100b5ded64b559a85039a842
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 (assigned_dev->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 unsigned long i;
968 struct kvm_memory_slot *memslot;
969 struct kvm_memory_slot old, new;
970
971 r = -EINVAL;
972 /* General sanity checks */
973 if (mem->memory_size & (PAGE_SIZE - 1))
974 goto out;
975 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
976 goto out;
977 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
978 goto out;
979 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
980 goto out;
981 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
982 goto out;
983
984 memslot = &kvm->memslots[mem->slot];
985 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
986 npages = mem->memory_size >> PAGE_SHIFT;
987
988 if (!npages)
989 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
990
991 new = old = *memslot;
992
993 new.base_gfn = base_gfn;
994 new.npages = npages;
995 new.flags = mem->flags;
996
997 /* Disallow changing a memory slot's size. */
998 r = -EINVAL;
999 if (npages && old.npages && npages != old.npages)
1000 goto out_free;
1001
1002 /* Check for overlaps */
1003 r = -EEXIST;
1004 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1005 struct kvm_memory_slot *s = &kvm->memslots[i];
1006
1007 if (s == memslot)
1008 continue;
1009 if (!((base_gfn + npages <= s->base_gfn) ||
1010 (base_gfn >= s->base_gfn + s->npages)))
1011 goto out_free;
1012 }
1013
1014 /* Free page dirty bitmap if unneeded */
1015 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1016 new.dirty_bitmap = NULL;
1017
1018 r = -ENOMEM;
1019
1020 /* Allocate if a slot is being created */
1021 #ifndef CONFIG_S390
1022 if (npages && !new.rmap) {
1023 new.rmap = vmalloc(npages * sizeof(struct page *));
1024
1025 if (!new.rmap)
1026 goto out_free;
1027
1028 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1029
1030 new.user_alloc = user_alloc;
1031 /*
1032 * hva_to_rmmap() serialzies with the mmu_lock and to be
1033 * safe it has to ignore memslots with !user_alloc &&
1034 * !userspace_addr.
1035 */
1036 if (user_alloc)
1037 new.userspace_addr = mem->userspace_addr;
1038 else
1039 new.userspace_addr = 0;
1040 }
1041 if (npages && !new.lpage_info) {
1042 int largepages = npages / KVM_PAGES_PER_HPAGE;
1043 if (npages % KVM_PAGES_PER_HPAGE)
1044 largepages++;
1045 if (base_gfn % KVM_PAGES_PER_HPAGE)
1046 largepages++;
1047
1048 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1049
1050 if (!new.lpage_info)
1051 goto out_free;
1052
1053 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1054
1055 if (base_gfn % KVM_PAGES_PER_HPAGE)
1056 new.lpage_info[0].write_count = 1;
1057 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1058 new.lpage_info[largepages-1].write_count = 1;
1059 }
1060
1061 /* Allocate page dirty bitmap if needed */
1062 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1063 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1064
1065 new.dirty_bitmap = vmalloc(dirty_bytes);
1066 if (!new.dirty_bitmap)
1067 goto out_free;
1068 memset(new.dirty_bitmap, 0, dirty_bytes);
1069 }
1070 #endif /* not defined CONFIG_S390 */
1071
1072 if (!npages)
1073 kvm_arch_flush_shadow(kvm);
1074
1075 spin_lock(&kvm->mmu_lock);
1076 if (mem->slot >= kvm->nmemslots)
1077 kvm->nmemslots = mem->slot + 1;
1078
1079 *memslot = new;
1080 spin_unlock(&kvm->mmu_lock);
1081
1082 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1083 if (r) {
1084 spin_lock(&kvm->mmu_lock);
1085 *memslot = old;
1086 spin_unlock(&kvm->mmu_lock);
1087 goto out_free;
1088 }
1089
1090 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1091 /* Slot deletion case: we have to update the current slot */
1092 if (!npages)
1093 *memslot = old;
1094 #ifdef CONFIG_DMAR
1095 /* map the pages in iommu page table */
1096 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1097 if (r)
1098 goto out;
1099 #endif
1100 return 0;
1101
1102 out_free:
1103 kvm_free_physmem_slot(&new, &old);
1104 out:
1105 return r;
1106
1107 }
1108 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1109
1110 int kvm_set_memory_region(struct kvm *kvm,
1111 struct kvm_userspace_memory_region *mem,
1112 int user_alloc)
1113 {
1114 int r;
1115
1116 down_write(&kvm->slots_lock);
1117 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1118 up_write(&kvm->slots_lock);
1119 return r;
1120 }
1121 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1122
1123 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1124 struct
1125 kvm_userspace_memory_region *mem,
1126 int user_alloc)
1127 {
1128 if (mem->slot >= KVM_MEMORY_SLOTS)
1129 return -EINVAL;
1130 return kvm_set_memory_region(kvm, mem, user_alloc);
1131 }
1132
1133 int kvm_get_dirty_log(struct kvm *kvm,
1134 struct kvm_dirty_log *log, int *is_dirty)
1135 {
1136 struct kvm_memory_slot *memslot;
1137 int r, i;
1138 int n;
1139 unsigned long any = 0;
1140
1141 r = -EINVAL;
1142 if (log->slot >= KVM_MEMORY_SLOTS)
1143 goto out;
1144
1145 memslot = &kvm->memslots[log->slot];
1146 r = -ENOENT;
1147 if (!memslot->dirty_bitmap)
1148 goto out;
1149
1150 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1151
1152 for (i = 0; !any && i < n/sizeof(long); ++i)
1153 any = memslot->dirty_bitmap[i];
1154
1155 r = -EFAULT;
1156 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1157 goto out;
1158
1159 if (any)
1160 *is_dirty = 1;
1161
1162 r = 0;
1163 out:
1164 return r;
1165 }
1166
1167 int is_error_page(struct page *page)
1168 {
1169 return page == bad_page;
1170 }
1171 EXPORT_SYMBOL_GPL(is_error_page);
1172
1173 int is_error_pfn(pfn_t pfn)
1174 {
1175 return pfn == bad_pfn;
1176 }
1177 EXPORT_SYMBOL_GPL(is_error_pfn);
1178
1179 static inline unsigned long bad_hva(void)
1180 {
1181 return PAGE_OFFSET;
1182 }
1183
1184 int kvm_is_error_hva(unsigned long addr)
1185 {
1186 return addr == bad_hva();
1187 }
1188 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1189
1190 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1191 {
1192 int i;
1193
1194 for (i = 0; i < kvm->nmemslots; ++i) {
1195 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1196
1197 if (gfn >= memslot->base_gfn
1198 && gfn < memslot->base_gfn + memslot->npages)
1199 return memslot;
1200 }
1201 return NULL;
1202 }
1203 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1204
1205 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1206 {
1207 gfn = unalias_gfn(kvm, gfn);
1208 return gfn_to_memslot_unaliased(kvm, gfn);
1209 }
1210
1211 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1212 {
1213 int i;
1214
1215 gfn = unalias_gfn(kvm, gfn);
1216 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1217 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1218
1219 if (gfn >= memslot->base_gfn
1220 && gfn < memslot->base_gfn + memslot->npages)
1221 return 1;
1222 }
1223 return 0;
1224 }
1225 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1226
1227 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1228 {
1229 struct kvm_memory_slot *slot;
1230
1231 gfn = unalias_gfn(kvm, gfn);
1232 slot = gfn_to_memslot_unaliased(kvm, gfn);
1233 if (!slot)
1234 return bad_hva();
1235 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1236 }
1237 EXPORT_SYMBOL_GPL(gfn_to_hva);
1238
1239 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1240 {
1241 struct page *page[1];
1242 unsigned long addr;
1243 int npages;
1244 pfn_t pfn;
1245
1246 might_sleep();
1247
1248 addr = gfn_to_hva(kvm, gfn);
1249 if (kvm_is_error_hva(addr)) {
1250 get_page(bad_page);
1251 return page_to_pfn(bad_page);
1252 }
1253
1254 npages = get_user_pages_fast(addr, 1, 1, page);
1255
1256 if (unlikely(npages != 1)) {
1257 struct vm_area_struct *vma;
1258
1259 down_read(&current->mm->mmap_sem);
1260 vma = find_vma(current->mm, addr);
1261
1262 if (vma == NULL || addr < vma->vm_start ||
1263 !(vma->vm_flags & VM_PFNMAP)) {
1264 up_read(&current->mm->mmap_sem);
1265 get_page(bad_page);
1266 return page_to_pfn(bad_page);
1267 }
1268
1269 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1270 up_read(&current->mm->mmap_sem);
1271 BUG_ON(!kvm_is_mmio_pfn(pfn));
1272 } else
1273 pfn = page_to_pfn(page[0]);
1274
1275 return pfn;
1276 }
1277
1278 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1279
1280 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1281 {
1282 pfn_t pfn;
1283
1284 pfn = gfn_to_pfn(kvm, gfn);
1285 if (!kvm_is_mmio_pfn(pfn))
1286 return pfn_to_page(pfn);
1287
1288 WARN_ON(kvm_is_mmio_pfn(pfn));
1289
1290 get_page(bad_page);
1291 return bad_page;
1292 }
1293
1294 EXPORT_SYMBOL_GPL(gfn_to_page);
1295
1296 void kvm_release_page_clean(struct page *page)
1297 {
1298 kvm_release_pfn_clean(page_to_pfn(page));
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1301
1302 void kvm_release_pfn_clean(pfn_t pfn)
1303 {
1304 if (!kvm_is_mmio_pfn(pfn))
1305 put_page(pfn_to_page(pfn));
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1308
1309 void kvm_release_page_dirty(struct page *page)
1310 {
1311 kvm_release_pfn_dirty(page_to_pfn(page));
1312 }
1313 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1314
1315 void kvm_release_pfn_dirty(pfn_t pfn)
1316 {
1317 kvm_set_pfn_dirty(pfn);
1318 kvm_release_pfn_clean(pfn);
1319 }
1320 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1321
1322 void kvm_set_page_dirty(struct page *page)
1323 {
1324 kvm_set_pfn_dirty(page_to_pfn(page));
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1327
1328 void kvm_set_pfn_dirty(pfn_t pfn)
1329 {
1330 if (!kvm_is_mmio_pfn(pfn)) {
1331 struct page *page = pfn_to_page(pfn);
1332 if (!PageReserved(page))
1333 SetPageDirty(page);
1334 }
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1337
1338 void kvm_set_pfn_accessed(pfn_t pfn)
1339 {
1340 if (!kvm_is_mmio_pfn(pfn))
1341 mark_page_accessed(pfn_to_page(pfn));
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1344
1345 void kvm_get_pfn(pfn_t pfn)
1346 {
1347 if (!kvm_is_mmio_pfn(pfn))
1348 get_page(pfn_to_page(pfn));
1349 }
1350 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1351
1352 static int next_segment(unsigned long len, int offset)
1353 {
1354 if (len > PAGE_SIZE - offset)
1355 return PAGE_SIZE - offset;
1356 else
1357 return len;
1358 }
1359
1360 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1361 int len)
1362 {
1363 int r;
1364 unsigned long addr;
1365
1366 addr = gfn_to_hva(kvm, gfn);
1367 if (kvm_is_error_hva(addr))
1368 return -EFAULT;
1369 r = copy_from_user(data, (void __user *)addr + offset, len);
1370 if (r)
1371 return -EFAULT;
1372 return 0;
1373 }
1374 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1375
1376 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1377 {
1378 gfn_t gfn = gpa >> PAGE_SHIFT;
1379 int seg;
1380 int offset = offset_in_page(gpa);
1381 int ret;
1382
1383 while ((seg = next_segment(len, offset)) != 0) {
1384 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1385 if (ret < 0)
1386 return ret;
1387 offset = 0;
1388 len -= seg;
1389 data += seg;
1390 ++gfn;
1391 }
1392 return 0;
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_read_guest);
1395
1396 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1397 unsigned long len)
1398 {
1399 int r;
1400 unsigned long addr;
1401 gfn_t gfn = gpa >> PAGE_SHIFT;
1402 int offset = offset_in_page(gpa);
1403
1404 addr = gfn_to_hva(kvm, gfn);
1405 if (kvm_is_error_hva(addr))
1406 return -EFAULT;
1407 pagefault_disable();
1408 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1409 pagefault_enable();
1410 if (r)
1411 return -EFAULT;
1412 return 0;
1413 }
1414 EXPORT_SYMBOL(kvm_read_guest_atomic);
1415
1416 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1417 int offset, int len)
1418 {
1419 int r;
1420 unsigned long addr;
1421
1422 addr = gfn_to_hva(kvm, gfn);
1423 if (kvm_is_error_hva(addr))
1424 return -EFAULT;
1425 r = copy_to_user((void __user *)addr + offset, data, len);
1426 if (r)
1427 return -EFAULT;
1428 mark_page_dirty(kvm, gfn);
1429 return 0;
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1432
1433 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1434 unsigned long len)
1435 {
1436 gfn_t gfn = gpa >> PAGE_SHIFT;
1437 int seg;
1438 int offset = offset_in_page(gpa);
1439 int ret;
1440
1441 while ((seg = next_segment(len, offset)) != 0) {
1442 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1443 if (ret < 0)
1444 return ret;
1445 offset = 0;
1446 len -= seg;
1447 data += seg;
1448 ++gfn;
1449 }
1450 return 0;
1451 }
1452
1453 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1454 {
1455 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1456 }
1457 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1458
1459 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1460 {
1461 gfn_t gfn = gpa >> PAGE_SHIFT;
1462 int seg;
1463 int offset = offset_in_page(gpa);
1464 int ret;
1465
1466 while ((seg = next_segment(len, offset)) != 0) {
1467 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1468 if (ret < 0)
1469 return ret;
1470 offset = 0;
1471 len -= seg;
1472 ++gfn;
1473 }
1474 return 0;
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1477
1478 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1479 {
1480 struct kvm_memory_slot *memslot;
1481
1482 gfn = unalias_gfn(kvm, gfn);
1483 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1484 if (memslot && memslot->dirty_bitmap) {
1485 unsigned long rel_gfn = gfn - memslot->base_gfn;
1486
1487 /* avoid RMW */
1488 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1489 set_bit(rel_gfn, memslot->dirty_bitmap);
1490 }
1491 }
1492
1493 /*
1494 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1495 */
1496 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1497 {
1498 DEFINE_WAIT(wait);
1499
1500 for (;;) {
1501 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1502
1503 if (kvm_cpu_has_interrupt(vcpu) ||
1504 kvm_cpu_has_pending_timer(vcpu) ||
1505 kvm_arch_vcpu_runnable(vcpu)) {
1506 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1507 break;
1508 }
1509 if (signal_pending(current))
1510 break;
1511
1512 vcpu_put(vcpu);
1513 schedule();
1514 vcpu_load(vcpu);
1515 }
1516
1517 finish_wait(&vcpu->wq, &wait);
1518 }
1519
1520 void kvm_resched(struct kvm_vcpu *vcpu)
1521 {
1522 if (!need_resched())
1523 return;
1524 cond_resched();
1525 }
1526 EXPORT_SYMBOL_GPL(kvm_resched);
1527
1528 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1529 {
1530 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1531 struct page *page;
1532
1533 if (vmf->pgoff == 0)
1534 page = virt_to_page(vcpu->run);
1535 #ifdef CONFIG_X86
1536 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1537 page = virt_to_page(vcpu->arch.pio_data);
1538 #endif
1539 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1540 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1541 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1542 #endif
1543 else
1544 return VM_FAULT_SIGBUS;
1545 get_page(page);
1546 vmf->page = page;
1547 return 0;
1548 }
1549
1550 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1551 .fault = kvm_vcpu_fault,
1552 };
1553
1554 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1555 {
1556 vma->vm_ops = &kvm_vcpu_vm_ops;
1557 return 0;
1558 }
1559
1560 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1561 {
1562 struct kvm_vcpu *vcpu = filp->private_data;
1563
1564 kvm_put_kvm(vcpu->kvm);
1565 return 0;
1566 }
1567
1568 static struct file_operations kvm_vcpu_fops = {
1569 .release = kvm_vcpu_release,
1570 .unlocked_ioctl = kvm_vcpu_ioctl,
1571 .compat_ioctl = kvm_vcpu_ioctl,
1572 .mmap = kvm_vcpu_mmap,
1573 };
1574
1575 /*
1576 * Allocates an inode for the vcpu.
1577 */
1578 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1579 {
1580 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1581 if (fd < 0)
1582 kvm_put_kvm(vcpu->kvm);
1583 return fd;
1584 }
1585
1586 /*
1587 * Creates some virtual cpus. Good luck creating more than one.
1588 */
1589 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1590 {
1591 int r;
1592 struct kvm_vcpu *vcpu;
1593
1594 if (!valid_vcpu(n))
1595 return -EINVAL;
1596
1597 vcpu = kvm_arch_vcpu_create(kvm, n);
1598 if (IS_ERR(vcpu))
1599 return PTR_ERR(vcpu);
1600
1601 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1602
1603 r = kvm_arch_vcpu_setup(vcpu);
1604 if (r)
1605 return r;
1606
1607 mutex_lock(&kvm->lock);
1608 if (kvm->vcpus[n]) {
1609 r = -EEXIST;
1610 goto vcpu_destroy;
1611 }
1612 kvm->vcpus[n] = vcpu;
1613 mutex_unlock(&kvm->lock);
1614
1615 /* Now it's all set up, let userspace reach it */
1616 kvm_get_kvm(kvm);
1617 r = create_vcpu_fd(vcpu);
1618 if (r < 0)
1619 goto unlink;
1620 return r;
1621
1622 unlink:
1623 mutex_lock(&kvm->lock);
1624 kvm->vcpus[n] = NULL;
1625 vcpu_destroy:
1626 mutex_unlock(&kvm->lock);
1627 kvm_arch_vcpu_destroy(vcpu);
1628 return r;
1629 }
1630
1631 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1632 {
1633 if (sigset) {
1634 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1635 vcpu->sigset_active = 1;
1636 vcpu->sigset = *sigset;
1637 } else
1638 vcpu->sigset_active = 0;
1639 return 0;
1640 }
1641
1642 static long kvm_vcpu_ioctl(struct file *filp,
1643 unsigned int ioctl, unsigned long arg)
1644 {
1645 struct kvm_vcpu *vcpu = filp->private_data;
1646 void __user *argp = (void __user *)arg;
1647 int r;
1648 struct kvm_fpu *fpu = NULL;
1649 struct kvm_sregs *kvm_sregs = NULL;
1650
1651 if (vcpu->kvm->mm != current->mm)
1652 return -EIO;
1653 switch (ioctl) {
1654 case KVM_RUN:
1655 r = -EINVAL;
1656 if (arg)
1657 goto out;
1658 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1659 break;
1660 case KVM_GET_REGS: {
1661 struct kvm_regs *kvm_regs;
1662
1663 r = -ENOMEM;
1664 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1665 if (!kvm_regs)
1666 goto out;
1667 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1668 if (r)
1669 goto out_free1;
1670 r = -EFAULT;
1671 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1672 goto out_free1;
1673 r = 0;
1674 out_free1:
1675 kfree(kvm_regs);
1676 break;
1677 }
1678 case KVM_SET_REGS: {
1679 struct kvm_regs *kvm_regs;
1680
1681 r = -ENOMEM;
1682 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1683 if (!kvm_regs)
1684 goto out;
1685 r = -EFAULT;
1686 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1687 goto out_free2;
1688 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1689 if (r)
1690 goto out_free2;
1691 r = 0;
1692 out_free2:
1693 kfree(kvm_regs);
1694 break;
1695 }
1696 case KVM_GET_SREGS: {
1697 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1698 r = -ENOMEM;
1699 if (!kvm_sregs)
1700 goto out;
1701 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1702 if (r)
1703 goto out;
1704 r = -EFAULT;
1705 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1706 goto out;
1707 r = 0;
1708 break;
1709 }
1710 case KVM_SET_SREGS: {
1711 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1712 r = -ENOMEM;
1713 if (!kvm_sregs)
1714 goto out;
1715 r = -EFAULT;
1716 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1717 goto out;
1718 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1719 if (r)
1720 goto out;
1721 r = 0;
1722 break;
1723 }
1724 case KVM_GET_MP_STATE: {
1725 struct kvm_mp_state mp_state;
1726
1727 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1728 if (r)
1729 goto out;
1730 r = -EFAULT;
1731 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1732 goto out;
1733 r = 0;
1734 break;
1735 }
1736 case KVM_SET_MP_STATE: {
1737 struct kvm_mp_state mp_state;
1738
1739 r = -EFAULT;
1740 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1741 goto out;
1742 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1743 if (r)
1744 goto out;
1745 r = 0;
1746 break;
1747 }
1748 case KVM_TRANSLATE: {
1749 struct kvm_translation tr;
1750
1751 r = -EFAULT;
1752 if (copy_from_user(&tr, argp, sizeof tr))
1753 goto out;
1754 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1755 if (r)
1756 goto out;
1757 r = -EFAULT;
1758 if (copy_to_user(argp, &tr, sizeof tr))
1759 goto out;
1760 r = 0;
1761 break;
1762 }
1763 case KVM_DEBUG_GUEST: {
1764 struct kvm_debug_guest dbg;
1765
1766 r = -EFAULT;
1767 if (copy_from_user(&dbg, argp, sizeof dbg))
1768 goto out;
1769 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1770 if (r)
1771 goto out;
1772 r = 0;
1773 break;
1774 }
1775 case KVM_SET_SIGNAL_MASK: {
1776 struct kvm_signal_mask __user *sigmask_arg = argp;
1777 struct kvm_signal_mask kvm_sigmask;
1778 sigset_t sigset, *p;
1779
1780 p = NULL;
1781 if (argp) {
1782 r = -EFAULT;
1783 if (copy_from_user(&kvm_sigmask, argp,
1784 sizeof kvm_sigmask))
1785 goto out;
1786 r = -EINVAL;
1787 if (kvm_sigmask.len != sizeof sigset)
1788 goto out;
1789 r = -EFAULT;
1790 if (copy_from_user(&sigset, sigmask_arg->sigset,
1791 sizeof sigset))
1792 goto out;
1793 p = &sigset;
1794 }
1795 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1796 break;
1797 }
1798 case KVM_GET_FPU: {
1799 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1800 r = -ENOMEM;
1801 if (!fpu)
1802 goto out;
1803 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1804 if (r)
1805 goto out;
1806 r = -EFAULT;
1807 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1808 goto out;
1809 r = 0;
1810 break;
1811 }
1812 case KVM_SET_FPU: {
1813 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1814 r = -ENOMEM;
1815 if (!fpu)
1816 goto out;
1817 r = -EFAULT;
1818 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1819 goto out;
1820 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1821 if (r)
1822 goto out;
1823 r = 0;
1824 break;
1825 }
1826 default:
1827 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1828 }
1829 out:
1830 kfree(fpu);
1831 kfree(kvm_sregs);
1832 return r;
1833 }
1834
1835 static long kvm_vm_ioctl(struct file *filp,
1836 unsigned int ioctl, unsigned long arg)
1837 {
1838 struct kvm *kvm = filp->private_data;
1839 void __user *argp = (void __user *)arg;
1840 int r;
1841
1842 if (kvm->mm != current->mm)
1843 return -EIO;
1844 switch (ioctl) {
1845 case KVM_CREATE_VCPU:
1846 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1847 if (r < 0)
1848 goto out;
1849 break;
1850 case KVM_SET_USER_MEMORY_REGION: {
1851 struct kvm_userspace_memory_region kvm_userspace_mem;
1852
1853 r = -EFAULT;
1854 if (copy_from_user(&kvm_userspace_mem, argp,
1855 sizeof kvm_userspace_mem))
1856 goto out;
1857
1858 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1859 if (r)
1860 goto out;
1861 break;
1862 }
1863 case KVM_GET_DIRTY_LOG: {
1864 struct kvm_dirty_log log;
1865
1866 r = -EFAULT;
1867 if (copy_from_user(&log, argp, sizeof log))
1868 goto out;
1869 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1870 if (r)
1871 goto out;
1872 break;
1873 }
1874 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1875 case KVM_REGISTER_COALESCED_MMIO: {
1876 struct kvm_coalesced_mmio_zone zone;
1877 r = -EFAULT;
1878 if (copy_from_user(&zone, argp, sizeof zone))
1879 goto out;
1880 r = -ENXIO;
1881 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1882 if (r)
1883 goto out;
1884 r = 0;
1885 break;
1886 }
1887 case KVM_UNREGISTER_COALESCED_MMIO: {
1888 struct kvm_coalesced_mmio_zone zone;
1889 r = -EFAULT;
1890 if (copy_from_user(&zone, argp, sizeof zone))
1891 goto out;
1892 r = -ENXIO;
1893 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1894 if (r)
1895 goto out;
1896 r = 0;
1897 break;
1898 }
1899 #endif
1900 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1901 case KVM_ASSIGN_PCI_DEVICE: {
1902 struct kvm_assigned_pci_dev assigned_dev;
1903
1904 r = -EFAULT;
1905 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1906 goto out;
1907 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1908 if (r)
1909 goto out;
1910 break;
1911 }
1912 case KVM_ASSIGN_IRQ: {
1913 struct kvm_assigned_irq assigned_irq;
1914
1915 r = -EFAULT;
1916 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1917 goto out;
1918 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1919 if (r)
1920 goto out;
1921 break;
1922 }
1923 #endif
1924 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1925 case KVM_DEASSIGN_PCI_DEVICE: {
1926 struct kvm_assigned_pci_dev assigned_dev;
1927
1928 r = -EFAULT;
1929 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1930 goto out;
1931 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1932 if (r)
1933 goto out;
1934 break;
1935 }
1936 #endif
1937 default:
1938 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1939 }
1940 out:
1941 return r;
1942 }
1943
1944 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1945 {
1946 struct page *page[1];
1947 unsigned long addr;
1948 int npages;
1949 gfn_t gfn = vmf->pgoff;
1950 struct kvm *kvm = vma->vm_file->private_data;
1951
1952 addr = gfn_to_hva(kvm, gfn);
1953 if (kvm_is_error_hva(addr))
1954 return VM_FAULT_SIGBUS;
1955
1956 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1957 NULL);
1958 if (unlikely(npages != 1))
1959 return VM_FAULT_SIGBUS;
1960
1961 vmf->page = page[0];
1962 return 0;
1963 }
1964
1965 static struct vm_operations_struct kvm_vm_vm_ops = {
1966 .fault = kvm_vm_fault,
1967 };
1968
1969 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1970 {
1971 vma->vm_ops = &kvm_vm_vm_ops;
1972 return 0;
1973 }
1974
1975 static struct file_operations kvm_vm_fops = {
1976 .release = kvm_vm_release,
1977 .unlocked_ioctl = kvm_vm_ioctl,
1978 .compat_ioctl = kvm_vm_ioctl,
1979 .mmap = kvm_vm_mmap,
1980 };
1981
1982 static int kvm_dev_ioctl_create_vm(void)
1983 {
1984 int fd;
1985 struct kvm *kvm;
1986
1987 kvm = kvm_create_vm();
1988 if (IS_ERR(kvm))
1989 return PTR_ERR(kvm);
1990 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1991 if (fd < 0)
1992 kvm_put_kvm(kvm);
1993
1994 return fd;
1995 }
1996
1997 static long kvm_dev_ioctl_check_extension_generic(long arg)
1998 {
1999 switch (arg) {
2000 case KVM_CAP_USER_MEMORY:
2001 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2002 return 1;
2003 default:
2004 break;
2005 }
2006 return kvm_dev_ioctl_check_extension(arg);
2007 }
2008
2009 static long kvm_dev_ioctl(struct file *filp,
2010 unsigned int ioctl, unsigned long arg)
2011 {
2012 long r = -EINVAL;
2013
2014 switch (ioctl) {
2015 case KVM_GET_API_VERSION:
2016 r = -EINVAL;
2017 if (arg)
2018 goto out;
2019 r = KVM_API_VERSION;
2020 break;
2021 case KVM_CREATE_VM:
2022 r = -EINVAL;
2023 if (arg)
2024 goto out;
2025 r = kvm_dev_ioctl_create_vm();
2026 break;
2027 case KVM_CHECK_EXTENSION:
2028 r = kvm_dev_ioctl_check_extension_generic(arg);
2029 break;
2030 case KVM_GET_VCPU_MMAP_SIZE:
2031 r = -EINVAL;
2032 if (arg)
2033 goto out;
2034 r = PAGE_SIZE; /* struct kvm_run */
2035 #ifdef CONFIG_X86
2036 r += PAGE_SIZE; /* pio data page */
2037 #endif
2038 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2039 r += PAGE_SIZE; /* coalesced mmio ring page */
2040 #endif
2041 break;
2042 case KVM_TRACE_ENABLE:
2043 case KVM_TRACE_PAUSE:
2044 case KVM_TRACE_DISABLE:
2045 r = kvm_trace_ioctl(ioctl, arg);
2046 break;
2047 default:
2048 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2049 }
2050 out:
2051 return r;
2052 }
2053
2054 static struct file_operations kvm_chardev_ops = {
2055 .unlocked_ioctl = kvm_dev_ioctl,
2056 .compat_ioctl = kvm_dev_ioctl,
2057 };
2058
2059 static struct miscdevice kvm_dev = {
2060 KVM_MINOR,
2061 "kvm",
2062 &kvm_chardev_ops,
2063 };
2064
2065 static void hardware_enable(void *junk)
2066 {
2067 int cpu = raw_smp_processor_id();
2068
2069 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2070 return;
2071 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2072 kvm_arch_hardware_enable(NULL);
2073 }
2074
2075 static void hardware_disable(void *junk)
2076 {
2077 int cpu = raw_smp_processor_id();
2078
2079 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2080 return;
2081 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2082 kvm_arch_hardware_disable(NULL);
2083 }
2084
2085 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2086 void *v)
2087 {
2088 int cpu = (long)v;
2089
2090 val &= ~CPU_TASKS_FROZEN;
2091 switch (val) {
2092 case CPU_DYING:
2093 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2094 cpu);
2095 hardware_disable(NULL);
2096 break;
2097 case CPU_UP_CANCELED:
2098 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2099 cpu);
2100 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2101 break;
2102 case CPU_ONLINE:
2103 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2104 cpu);
2105 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2106 break;
2107 }
2108 return NOTIFY_OK;
2109 }
2110
2111
2112 asmlinkage void kvm_handle_fault_on_reboot(void)
2113 {
2114 if (kvm_rebooting)
2115 /* spin while reset goes on */
2116 while (true)
2117 ;
2118 /* Fault while not rebooting. We want the trace. */
2119 BUG();
2120 }
2121 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2122
2123 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2124 void *v)
2125 {
2126 if (val == SYS_RESTART) {
2127 /*
2128 * Some (well, at least mine) BIOSes hang on reboot if
2129 * in vmx root mode.
2130 */
2131 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2132 kvm_rebooting = true;
2133 on_each_cpu(hardware_disable, NULL, 1);
2134 }
2135 return NOTIFY_OK;
2136 }
2137
2138 static struct notifier_block kvm_reboot_notifier = {
2139 .notifier_call = kvm_reboot,
2140 .priority = 0,
2141 };
2142
2143 void kvm_io_bus_init(struct kvm_io_bus *bus)
2144 {
2145 memset(bus, 0, sizeof(*bus));
2146 }
2147
2148 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2149 {
2150 int i;
2151
2152 for (i = 0; i < bus->dev_count; i++) {
2153 struct kvm_io_device *pos = bus->devs[i];
2154
2155 kvm_iodevice_destructor(pos);
2156 }
2157 }
2158
2159 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2160 gpa_t addr, int len, int is_write)
2161 {
2162 int i;
2163
2164 for (i = 0; i < bus->dev_count; i++) {
2165 struct kvm_io_device *pos = bus->devs[i];
2166
2167 if (pos->in_range(pos, addr, len, is_write))
2168 return pos;
2169 }
2170
2171 return NULL;
2172 }
2173
2174 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2175 {
2176 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2177
2178 bus->devs[bus->dev_count++] = dev;
2179 }
2180
2181 static struct notifier_block kvm_cpu_notifier = {
2182 .notifier_call = kvm_cpu_hotplug,
2183 .priority = 20, /* must be > scheduler priority */
2184 };
2185
2186 static int vm_stat_get(void *_offset, u64 *val)
2187 {
2188 unsigned offset = (long)_offset;
2189 struct kvm *kvm;
2190
2191 *val = 0;
2192 spin_lock(&kvm_lock);
2193 list_for_each_entry(kvm, &vm_list, vm_list)
2194 *val += *(u32 *)((void *)kvm + offset);
2195 spin_unlock(&kvm_lock);
2196 return 0;
2197 }
2198
2199 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2200
2201 static int vcpu_stat_get(void *_offset, u64 *val)
2202 {
2203 unsigned offset = (long)_offset;
2204 struct kvm *kvm;
2205 struct kvm_vcpu *vcpu;
2206 int i;
2207
2208 *val = 0;
2209 spin_lock(&kvm_lock);
2210 list_for_each_entry(kvm, &vm_list, vm_list)
2211 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2212 vcpu = kvm->vcpus[i];
2213 if (vcpu)
2214 *val += *(u32 *)((void *)vcpu + offset);
2215 }
2216 spin_unlock(&kvm_lock);
2217 return 0;
2218 }
2219
2220 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2221
2222 static struct file_operations *stat_fops[] = {
2223 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2224 [KVM_STAT_VM] = &vm_stat_fops,
2225 };
2226
2227 static void kvm_init_debug(void)
2228 {
2229 struct kvm_stats_debugfs_item *p;
2230
2231 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2232 for (p = debugfs_entries; p->name; ++p)
2233 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2234 (void *)(long)p->offset,
2235 stat_fops[p->kind]);
2236 }
2237
2238 static void kvm_exit_debug(void)
2239 {
2240 struct kvm_stats_debugfs_item *p;
2241
2242 for (p = debugfs_entries; p->name; ++p)
2243 debugfs_remove(p->dentry);
2244 debugfs_remove(kvm_debugfs_dir);
2245 }
2246
2247 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2248 {
2249 hardware_disable(NULL);
2250 return 0;
2251 }
2252
2253 static int kvm_resume(struct sys_device *dev)
2254 {
2255 hardware_enable(NULL);
2256 return 0;
2257 }
2258
2259 static struct sysdev_class kvm_sysdev_class = {
2260 .name = "kvm",
2261 .suspend = kvm_suspend,
2262 .resume = kvm_resume,
2263 };
2264
2265 static struct sys_device kvm_sysdev = {
2266 .id = 0,
2267 .cls = &kvm_sysdev_class,
2268 };
2269
2270 struct page *bad_page;
2271 pfn_t bad_pfn;
2272
2273 static inline
2274 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2275 {
2276 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2277 }
2278
2279 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2280 {
2281 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2282
2283 kvm_arch_vcpu_load(vcpu, cpu);
2284 }
2285
2286 static void kvm_sched_out(struct preempt_notifier *pn,
2287 struct task_struct *next)
2288 {
2289 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2290
2291 kvm_arch_vcpu_put(vcpu);
2292 }
2293
2294 int kvm_init(void *opaque, unsigned int vcpu_size,
2295 struct module *module)
2296 {
2297 int r;
2298 int cpu;
2299
2300 kvm_init_debug();
2301
2302 r = kvm_arch_init(opaque);
2303 if (r)
2304 goto out_fail;
2305
2306 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2307
2308 if (bad_page == NULL) {
2309 r = -ENOMEM;
2310 goto out;
2311 }
2312
2313 bad_pfn = page_to_pfn(bad_page);
2314
2315 if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2316 r = -ENOMEM;
2317 goto out_free_0;
2318 }
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);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree