aboutsummaryrefslogtreecommitdiffstats
path: root/virt/kvm/kvm_main.c
diff options
context:
space:
mode:
Diffstat (limited to 'virt/kvm/kvm_main.c')
-rw-r--r--virt/kvm/kvm_main.c961
1 files changed, 147 insertions, 814 deletions
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 7495ce347344..f92ba138007a 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -43,6 +43,7 @@
43#include <linux/swap.h> 43#include <linux/swap.h>
44#include <linux/bitops.h> 44#include <linux/bitops.h>
45#include <linux/spinlock.h> 45#include <linux/spinlock.h>
46#include <linux/compat.h>
46 47
47#include <asm/processor.h> 48#include <asm/processor.h>
48#include <asm/io.h> 49#include <asm/io.h>
@@ -53,12 +54,6 @@
53#include "coalesced_mmio.h" 54#include "coalesced_mmio.h"
54#endif 55#endif
55 56
56#ifdef KVM_CAP_DEVICE_ASSIGNMENT
57#include <linux/pci.h>
58#include <linux/interrupt.h>
59#include "irq.h"
60#endif
61
62#define CREATE_TRACE_POINTS 57#define CREATE_TRACE_POINTS
63#include <trace/events/kvm.h> 58#include <trace/events/kvm.h>
64 59
@@ -75,6 +70,8 @@ DEFINE_SPINLOCK(kvm_lock);
75LIST_HEAD(vm_list); 70LIST_HEAD(vm_list);
76 71
77static cpumask_var_t cpus_hardware_enabled; 72static cpumask_var_t cpus_hardware_enabled;
73static int kvm_usage_count = 0;
74static atomic_t hardware_enable_failed;
78 75
79struct kmem_cache *kvm_vcpu_cache; 76struct kmem_cache *kvm_vcpu_cache;
80EXPORT_SYMBOL_GPL(kvm_vcpu_cache); 77EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
@@ -85,615 +82,13 @@ struct dentry *kvm_debugfs_dir;
85 82
86static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, 83static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 unsigned long arg); 84 unsigned long arg);
85static int hardware_enable_all(void);
86static void hardware_disable_all(void);
88 87
89static bool kvm_rebooting; 88static bool kvm_rebooting;
90 89
91static bool largepages_enabled = true; 90static bool largepages_enabled = true;
92 91
93#ifdef KVM_CAP_DEVICE_ASSIGNMENT
94static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
95 int assigned_dev_id)
96{
97 struct list_head *ptr;
98 struct kvm_assigned_dev_kernel *match;
99
100 list_for_each(ptr, head) {
101 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
102 if (match->assigned_dev_id == assigned_dev_id)
103 return match;
104 }
105 return NULL;
106}
107
108static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
109 *assigned_dev, int irq)
110{
111 int i, index;
112 struct msix_entry *host_msix_entries;
113
114 host_msix_entries = assigned_dev->host_msix_entries;
115
116 index = -1;
117 for (i = 0; i < assigned_dev->entries_nr; i++)
118 if (irq == host_msix_entries[i].vector) {
119 index = i;
120 break;
121 }
122 if (index < 0) {
123 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
124 return 0;
125 }
126
127 return index;
128}
129
130static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
131{
132 struct kvm_assigned_dev_kernel *assigned_dev;
133 struct kvm *kvm;
134 int i;
135
136 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
137 interrupt_work);
138 kvm = assigned_dev->kvm;
139
140 mutex_lock(&kvm->irq_lock);
141 spin_lock_irq(&assigned_dev->assigned_dev_lock);
142 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
143 struct kvm_guest_msix_entry *guest_entries =
144 assigned_dev->guest_msix_entries;
145 for (i = 0; i < assigned_dev->entries_nr; i++) {
146 if (!(guest_entries[i].flags &
147 KVM_ASSIGNED_MSIX_PENDING))
148 continue;
149 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
150 kvm_set_irq(assigned_dev->kvm,
151 assigned_dev->irq_source_id,
152 guest_entries[i].vector, 1);
153 }
154 } else
155 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
156 assigned_dev->guest_irq, 1);
157
158 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
159 mutex_unlock(&assigned_dev->kvm->irq_lock);
160}
161
162static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
163{
164 unsigned long flags;
165 struct kvm_assigned_dev_kernel *assigned_dev =
166 (struct kvm_assigned_dev_kernel *) dev_id;
167
168 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
169 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
170 int index = find_index_from_host_irq(assigned_dev, irq);
171 if (index < 0)
172 goto out;
173 assigned_dev->guest_msix_entries[index].flags |=
174 KVM_ASSIGNED_MSIX_PENDING;
175 }
176
177 schedule_work(&assigned_dev->interrupt_work);
178
179 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
180 disable_irq_nosync(irq);
181 assigned_dev->host_irq_disabled = true;
182 }
183
184out:
185 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
186 return IRQ_HANDLED;
187}
188
189/* Ack the irq line for an assigned device */
190static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
191{
192 struct kvm_assigned_dev_kernel *dev;
193 unsigned long flags;
194
195 if (kian->gsi == -1)
196 return;
197
198 dev = container_of(kian, struct kvm_assigned_dev_kernel,
199 ack_notifier);
200
201 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
202
203 /* The guest irq may be shared so this ack may be
204 * from another device.
205 */
206 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
207 if (dev->host_irq_disabled) {
208 enable_irq(dev->host_irq);
209 dev->host_irq_disabled = false;
210 }
211 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
212}
213
214static void deassign_guest_irq(struct kvm *kvm,
215 struct kvm_assigned_dev_kernel *assigned_dev)
216{
217 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
218 assigned_dev->ack_notifier.gsi = -1;
219
220 if (assigned_dev->irq_source_id != -1)
221 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
222 assigned_dev->irq_source_id = -1;
223 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
224}
225
226/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
227static void deassign_host_irq(struct kvm *kvm,
228 struct kvm_assigned_dev_kernel *assigned_dev)
229{
230 /*
231 * In kvm_free_device_irq, cancel_work_sync return true if:
232 * 1. work is scheduled, and then cancelled.
233 * 2. work callback is executed.
234 *
235 * The first one ensured that the irq is disabled and no more events
236 * would happen. But for the second one, the irq may be enabled (e.g.
237 * for MSI). So we disable irq here to prevent further events.
238 *
239 * Notice this maybe result in nested disable if the interrupt type is
240 * INTx, but it's OK for we are going to free it.
241 *
242 * If this function is a part of VM destroy, please ensure that till
243 * now, the kvm state is still legal for probably we also have to wait
244 * interrupt_work done.
245 */
246 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
247 int i;
248 for (i = 0; i < assigned_dev->entries_nr; i++)
249 disable_irq_nosync(assigned_dev->
250 host_msix_entries[i].vector);
251
252 cancel_work_sync(&assigned_dev->interrupt_work);
253
254 for (i = 0; i < assigned_dev->entries_nr; i++)
255 free_irq(assigned_dev->host_msix_entries[i].vector,
256 (void *)assigned_dev);
257
258 assigned_dev->entries_nr = 0;
259 kfree(assigned_dev->host_msix_entries);
260 kfree(assigned_dev->guest_msix_entries);
261 pci_disable_msix(assigned_dev->dev);
262 } else {
263 /* Deal with MSI and INTx */
264 disable_irq_nosync(assigned_dev->host_irq);
265 cancel_work_sync(&assigned_dev->interrupt_work);
266
267 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
268
269 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
270 pci_disable_msi(assigned_dev->dev);
271 }
272
273 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
274}
275
276static int kvm_deassign_irq(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *assigned_dev,
278 unsigned long irq_requested_type)
279{
280 unsigned long guest_irq_type, host_irq_type;
281
282 if (!irqchip_in_kernel(kvm))
283 return -EINVAL;
284 /* no irq assignment to deassign */
285 if (!assigned_dev->irq_requested_type)
286 return -ENXIO;
287
288 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
289 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
290
291 if (host_irq_type)
292 deassign_host_irq(kvm, assigned_dev);
293 if (guest_irq_type)
294 deassign_guest_irq(kvm, assigned_dev);
295
296 return 0;
297}
298
299static void kvm_free_assigned_irq(struct kvm *kvm,
300 struct kvm_assigned_dev_kernel *assigned_dev)
301{
302 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
303}
304
305static void kvm_free_assigned_device(struct kvm *kvm,
306 struct kvm_assigned_dev_kernel
307 *assigned_dev)
308{
309 kvm_free_assigned_irq(kvm, assigned_dev);
310
311 pci_reset_function(assigned_dev->dev);
312
313 pci_release_regions(assigned_dev->dev);
314 pci_disable_device(assigned_dev->dev);
315 pci_dev_put(assigned_dev->dev);
316
317 list_del(&assigned_dev->list);
318 kfree(assigned_dev);
319}
320
321void kvm_free_all_assigned_devices(struct kvm *kvm)
322{
323 struct list_head *ptr, *ptr2;
324 struct kvm_assigned_dev_kernel *assigned_dev;
325
326 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
327 assigned_dev = list_entry(ptr,
328 struct kvm_assigned_dev_kernel,
329 list);
330
331 kvm_free_assigned_device(kvm, assigned_dev);
332 }
333}
334
335static int assigned_device_enable_host_intx(struct kvm *kvm,
336 struct kvm_assigned_dev_kernel *dev)
337{
338 dev->host_irq = dev->dev->irq;
339 /* Even though this is PCI, we don't want to use shared
340 * interrupts. Sharing host devices with guest-assigned devices
341 * on the same interrupt line is not a happy situation: there
342 * are going to be long delays in accepting, acking, etc.
343 */
344 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
345 0, "kvm_assigned_intx_device", (void *)dev))
346 return -EIO;
347 return 0;
348}
349
350#ifdef __KVM_HAVE_MSI
351static int assigned_device_enable_host_msi(struct kvm *kvm,
352 struct kvm_assigned_dev_kernel *dev)
353{
354 int r;
355
356 if (!dev->dev->msi_enabled) {
357 r = pci_enable_msi(dev->dev);
358 if (r)
359 return r;
360 }
361
362 dev->host_irq = dev->dev->irq;
363 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
364 "kvm_assigned_msi_device", (void *)dev)) {
365 pci_disable_msi(dev->dev);
366 return -EIO;
367 }
368
369 return 0;
370}
371#endif
372
373#ifdef __KVM_HAVE_MSIX
374static int assigned_device_enable_host_msix(struct kvm *kvm,
375 struct kvm_assigned_dev_kernel *dev)
376{
377 int i, r = -EINVAL;
378
379 /* host_msix_entries and guest_msix_entries should have been
380 * initialized */
381 if (dev->entries_nr == 0)
382 return r;
383
384 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
385 if (r)
386 return r;
387
388 for (i = 0; i < dev->entries_nr; i++) {
389 r = request_irq(dev->host_msix_entries[i].vector,
390 kvm_assigned_dev_intr, 0,
391 "kvm_assigned_msix_device",
392 (void *)dev);
393 /* FIXME: free requested_irq's on failure */
394 if (r)
395 return r;
396 }
397
398 return 0;
399}
400
401#endif
402
403static int assigned_device_enable_guest_intx(struct kvm *kvm,
404 struct kvm_assigned_dev_kernel *dev,
405 struct kvm_assigned_irq *irq)
406{
407 dev->guest_irq = irq->guest_irq;
408 dev->ack_notifier.gsi = irq->guest_irq;
409 return 0;
410}
411
412#ifdef __KVM_HAVE_MSI
413static int assigned_device_enable_guest_msi(struct kvm *kvm,
414 struct kvm_assigned_dev_kernel *dev,
415 struct kvm_assigned_irq *irq)
416{
417 dev->guest_irq = irq->guest_irq;
418 dev->ack_notifier.gsi = -1;
419 dev->host_irq_disabled = false;
420 return 0;
421}
422#endif
423#ifdef __KVM_HAVE_MSIX
424static int assigned_device_enable_guest_msix(struct kvm *kvm,
425 struct kvm_assigned_dev_kernel *dev,
426 struct kvm_assigned_irq *irq)
427{
428 dev->guest_irq = irq->guest_irq;
429 dev->ack_notifier.gsi = -1;
430 dev->host_irq_disabled = false;
431 return 0;
432}
433#endif
434
435static int assign_host_irq(struct kvm *kvm,
436 struct kvm_assigned_dev_kernel *dev,
437 __u32 host_irq_type)
438{
439 int r = -EEXIST;
440
441 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
442 return r;
443
444 switch (host_irq_type) {
445 case KVM_DEV_IRQ_HOST_INTX:
446 r = assigned_device_enable_host_intx(kvm, dev);
447 break;
448#ifdef __KVM_HAVE_MSI
449 case KVM_DEV_IRQ_HOST_MSI:
450 r = assigned_device_enable_host_msi(kvm, dev);
451 break;
452#endif
453#ifdef __KVM_HAVE_MSIX
454 case KVM_DEV_IRQ_HOST_MSIX:
455 r = assigned_device_enable_host_msix(kvm, dev);
456 break;
457#endif
458 default:
459 r = -EINVAL;
460 }
461
462 if (!r)
463 dev->irq_requested_type |= host_irq_type;
464
465 return r;
466}
467
468static int assign_guest_irq(struct kvm *kvm,
469 struct kvm_assigned_dev_kernel *dev,
470 struct kvm_assigned_irq *irq,
471 unsigned long guest_irq_type)
472{
473 int id;
474 int r = -EEXIST;
475
476 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
477 return r;
478
479 id = kvm_request_irq_source_id(kvm);
480 if (id < 0)
481 return id;
482
483 dev->irq_source_id = id;
484
485 switch (guest_irq_type) {
486 case KVM_DEV_IRQ_GUEST_INTX:
487 r = assigned_device_enable_guest_intx(kvm, dev, irq);
488 break;
489#ifdef __KVM_HAVE_MSI
490 case KVM_DEV_IRQ_GUEST_MSI:
491 r = assigned_device_enable_guest_msi(kvm, dev, irq);
492 break;
493#endif
494#ifdef __KVM_HAVE_MSIX
495 case KVM_DEV_IRQ_GUEST_MSIX:
496 r = assigned_device_enable_guest_msix(kvm, dev, irq);
497 break;
498#endif
499 default:
500 r = -EINVAL;
501 }
502
503 if (!r) {
504 dev->irq_requested_type |= guest_irq_type;
505 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
506 } else
507 kvm_free_irq_source_id(kvm, dev->irq_source_id);
508
509 return r;
510}
511
512/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
513static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
514 struct kvm_assigned_irq *assigned_irq)
515{
516 int r = -EINVAL;
517 struct kvm_assigned_dev_kernel *match;
518 unsigned long host_irq_type, guest_irq_type;
519
520 if (!capable(CAP_SYS_RAWIO))
521 return -EPERM;
522
523 if (!irqchip_in_kernel(kvm))
524 return r;
525
526 mutex_lock(&kvm->lock);
527 r = -ENODEV;
528 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
529 assigned_irq->assigned_dev_id);
530 if (!match)
531 goto out;
532
533 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
534 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
535
536 r = -EINVAL;
537 /* can only assign one type at a time */
538 if (hweight_long(host_irq_type) > 1)
539 goto out;
540 if (hweight_long(guest_irq_type) > 1)
541 goto out;
542 if (host_irq_type == 0 && guest_irq_type == 0)
543 goto out;
544
545 r = 0;
546 if (host_irq_type)
547 r = assign_host_irq(kvm, match, host_irq_type);
548 if (r)
549 goto out;
550
551 if (guest_irq_type)
552 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
553out:
554 mutex_unlock(&kvm->lock);
555 return r;
556}
557
558static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
559 struct kvm_assigned_irq
560 *assigned_irq)
561{
562 int r = -ENODEV;
563 struct kvm_assigned_dev_kernel *match;
564
565 mutex_lock(&kvm->lock);
566
567 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
568 assigned_irq->assigned_dev_id);
569 if (!match)
570 goto out;
571
572 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
573out:
574 mutex_unlock(&kvm->lock);
575 return r;
576}
577
578static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
579 struct kvm_assigned_pci_dev *assigned_dev)
580{
581 int r = 0;
582 struct kvm_assigned_dev_kernel *match;
583 struct pci_dev *dev;
584
585 down_read(&kvm->slots_lock);
586 mutex_lock(&kvm->lock);
587
588 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
589 assigned_dev->assigned_dev_id);
590 if (match) {
591 /* device already assigned */
592 r = -EEXIST;
593 goto out;
594 }
595
596 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
597 if (match == NULL) {
598 printk(KERN_INFO "%s: Couldn't allocate memory\n",
599 __func__);
600 r = -ENOMEM;
601 goto out;
602 }
603 dev = pci_get_bus_and_slot(assigned_dev->busnr,
604 assigned_dev->devfn);
605 if (!dev) {
606 printk(KERN_INFO "%s: host device not found\n", __func__);
607 r = -EINVAL;
608 goto out_free;
609 }
610 if (pci_enable_device(dev)) {
611 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
612 r = -EBUSY;
613 goto out_put;
614 }
615 r = pci_request_regions(dev, "kvm_assigned_device");
616 if (r) {
617 printk(KERN_INFO "%s: Could not get access to device regions\n",
618 __func__);
619 goto out_disable;
620 }
621
622 pci_reset_function(dev);
623
624 match->assigned_dev_id = assigned_dev->assigned_dev_id;
625 match->host_busnr = assigned_dev->busnr;
626 match->host_devfn = assigned_dev->devfn;
627 match->flags = assigned_dev->flags;
628 match->dev = dev;
629 spin_lock_init(&match->assigned_dev_lock);
630 match->irq_source_id = -1;
631 match->kvm = kvm;
632 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
633 INIT_WORK(&match->interrupt_work,
634 kvm_assigned_dev_interrupt_work_handler);
635
636 list_add(&match->list, &kvm->arch.assigned_dev_head);
637
638 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
639 if (!kvm->arch.iommu_domain) {
640 r = kvm_iommu_map_guest(kvm);
641 if (r)
642 goto out_list_del;
643 }
644 r = kvm_assign_device(kvm, match);
645 if (r)
646 goto out_list_del;
647 }
648
649out:
650 mutex_unlock(&kvm->lock);
651 up_read(&kvm->slots_lock);
652 return r;
653out_list_del:
654 list_del(&match->list);
655 pci_release_regions(dev);
656out_disable:
657 pci_disable_device(dev);
658out_put:
659 pci_dev_put(dev);
660out_free:
661 kfree(match);
662 mutex_unlock(&kvm->lock);
663 up_read(&kvm->slots_lock);
664 return r;
665}
666#endif
667
668#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
669static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
670 struct kvm_assigned_pci_dev *assigned_dev)
671{
672 int r = 0;
673 struct kvm_assigned_dev_kernel *match;
674
675 mutex_lock(&kvm->lock);
676
677 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
678 assigned_dev->assigned_dev_id);
679 if (!match) {
680 printk(KERN_INFO "%s: device hasn't been assigned before, "
681 "so cannot be deassigned\n", __func__);
682 r = -EINVAL;
683 goto out;
684 }
685
686 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
687 kvm_deassign_device(kvm, match);
688
689 kvm_free_assigned_device(kvm, match);
690
691out:
692 mutex_unlock(&kvm->lock);
693 return r;
694}
695#endif
696
697inline int kvm_is_mmio_pfn(pfn_t pfn) 92inline int kvm_is_mmio_pfn(pfn_t pfn)
698{ 93{
699 if (pfn_valid(pfn)) { 94 if (pfn_valid(pfn)) {
@@ -949,6 +344,7 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
949 344
950static struct kvm *kvm_create_vm(void) 345static struct kvm *kvm_create_vm(void)
951{ 346{
347 int r = 0;
952 struct kvm *kvm = kvm_arch_create_vm(); 348 struct kvm *kvm = kvm_arch_create_vm();
953#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 349#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
954 struct page *page; 350 struct page *page;
@@ -956,16 +352,21 @@ static struct kvm *kvm_create_vm(void)
956 352
957 if (IS_ERR(kvm)) 353 if (IS_ERR(kvm))
958 goto out; 354 goto out;
355
356 r = hardware_enable_all();
357 if (r)
358 goto out_err_nodisable;
359
959#ifdef CONFIG_HAVE_KVM_IRQCHIP 360#ifdef CONFIG_HAVE_KVM_IRQCHIP
960 INIT_LIST_HEAD(&kvm->irq_routing);
961 INIT_HLIST_HEAD(&kvm->mask_notifier_list); 361 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
362 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
962#endif 363#endif
963 364
964#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 365#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
965 page = alloc_page(GFP_KERNEL | __GFP_ZERO); 366 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
966 if (!page) { 367 if (!page) {
967 kfree(kvm); 368 r = -ENOMEM;
968 return ERR_PTR(-ENOMEM); 369 goto out_err;
969 } 370 }
970 kvm->coalesced_mmio_ring = 371 kvm->coalesced_mmio_ring =
971 (struct kvm_coalesced_mmio_ring *)page_address(page); 372 (struct kvm_coalesced_mmio_ring *)page_address(page);
@@ -973,15 +374,13 @@ static struct kvm *kvm_create_vm(void)
973 374
974#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 375#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
975 { 376 {
976 int err;
977 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; 377 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
978 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm); 378 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
979 if (err) { 379 if (r) {
980#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 380#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
981 put_page(page); 381 put_page(page);
982#endif 382#endif
983 kfree(kvm); 383 goto out_err;
984 return ERR_PTR(err);
985 } 384 }
986 } 385 }
987#endif 386#endif
@@ -1005,6 +404,12 @@ static struct kvm *kvm_create_vm(void)
1005#endif 404#endif
1006out: 405out:
1007 return kvm; 406 return kvm;
407
408out_err:
409 hardware_disable_all();
410out_err_nodisable:
411 kfree(kvm);
412 return ERR_PTR(r);
1008} 413}
1009 414
1010/* 415/*
@@ -1063,6 +468,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
1063 kvm_arch_flush_shadow(kvm); 468 kvm_arch_flush_shadow(kvm);
1064#endif 469#endif
1065 kvm_arch_destroy_vm(kvm); 470 kvm_arch_destroy_vm(kvm);
471 hardware_disable_all();
1066 mmdrop(mm); 472 mmdrop(mm);
1067} 473}
1068 474
@@ -1689,9 +1095,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1689 if (signal_pending(current)) 1095 if (signal_pending(current))
1690 break; 1096 break;
1691 1097
1692 vcpu_put(vcpu);
1693 schedule(); 1098 schedule();
1694 vcpu_load(vcpu);
1695 } 1099 }
1696 1100
1697 finish_wait(&vcpu->wq, &wait); 1101 finish_wait(&vcpu->wq, &wait);
@@ -1705,6 +1109,21 @@ void kvm_resched(struct kvm_vcpu *vcpu)
1705} 1109}
1706EXPORT_SYMBOL_GPL(kvm_resched); 1110EXPORT_SYMBOL_GPL(kvm_resched);
1707 1111
1112void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1113{
1114 ktime_t expires;
1115 DEFINE_WAIT(wait);
1116
1117 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1118
1119 /* Sleep for 100 us, and hope lock-holder got scheduled */
1120 expires = ktime_add_ns(ktime_get(), 100000UL);
1121 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1122
1123 finish_wait(&vcpu->wq, &wait);
1124}
1125EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1126
1708static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1127static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1709{ 1128{
1710 struct kvm_vcpu *vcpu = vma->vm_file->private_data; 1129 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
@@ -1828,88 +1247,6 @@ static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1828 return 0; 1247 return 0;
1829} 1248}
1830 1249
1831#ifdef __KVM_HAVE_MSIX
1832static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1833 struct kvm_assigned_msix_nr *entry_nr)
1834{
1835 int r = 0;
1836 struct kvm_assigned_dev_kernel *adev;
1837
1838 mutex_lock(&kvm->lock);
1839
1840 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1841 entry_nr->assigned_dev_id);
1842 if (!adev) {
1843 r = -EINVAL;
1844 goto msix_nr_out;
1845 }
1846
1847 if (adev->entries_nr == 0) {
1848 adev->entries_nr = entry_nr->entry_nr;
1849 if (adev->entries_nr == 0 ||
1850 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1851 r = -EINVAL;
1852 goto msix_nr_out;
1853 }
1854
1855 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1856 entry_nr->entry_nr,
1857 GFP_KERNEL);
1858 if (!adev->host_msix_entries) {
1859 r = -ENOMEM;
1860 goto msix_nr_out;
1861 }
1862 adev->guest_msix_entries = kzalloc(
1863 sizeof(struct kvm_guest_msix_entry) *
1864 entry_nr->entry_nr, GFP_KERNEL);
1865 if (!adev->guest_msix_entries) {
1866 kfree(adev->host_msix_entries);
1867 r = -ENOMEM;
1868 goto msix_nr_out;
1869 }
1870 } else /* Not allowed set MSI-X number twice */
1871 r = -EINVAL;
1872msix_nr_out:
1873 mutex_unlock(&kvm->lock);
1874 return r;
1875}
1876
1877static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1878 struct kvm_assigned_msix_entry *entry)
1879{
1880 int r = 0, i;
1881 struct kvm_assigned_dev_kernel *adev;
1882
1883 mutex_lock(&kvm->lock);
1884
1885 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1886 entry->assigned_dev_id);
1887
1888 if (!adev) {
1889 r = -EINVAL;
1890 goto msix_entry_out;
1891 }
1892
1893 for (i = 0; i < adev->entries_nr; i++)
1894 if (adev->guest_msix_entries[i].vector == 0 ||
1895 adev->guest_msix_entries[i].entry == entry->entry) {
1896 adev->guest_msix_entries[i].entry = entry->entry;
1897 adev->guest_msix_entries[i].vector = entry->gsi;
1898 adev->host_msix_entries[i].entry = entry->entry;
1899 break;
1900 }
1901 if (i == adev->entries_nr) {
1902 r = -ENOSPC;
1903 goto msix_entry_out;
1904 }
1905
1906msix_entry_out:
1907 mutex_unlock(&kvm->lock);
1908
1909 return r;
1910}
1911#endif
1912
1913static long kvm_vcpu_ioctl(struct file *filp, 1250static long kvm_vcpu_ioctl(struct file *filp,
1914 unsigned int ioctl, unsigned long arg) 1251 unsigned int ioctl, unsigned long arg)
1915{ 1252{
@@ -2168,112 +1505,6 @@ static long kvm_vm_ioctl(struct file *filp,
2168 break; 1505 break;
2169 } 1506 }
2170#endif 1507#endif
2171#ifdef KVM_CAP_DEVICE_ASSIGNMENT
2172 case KVM_ASSIGN_PCI_DEVICE: {
2173 struct kvm_assigned_pci_dev assigned_dev;
2174
2175 r = -EFAULT;
2176 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2177 goto out;
2178 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2179 if (r)
2180 goto out;
2181 break;
2182 }
2183 case KVM_ASSIGN_IRQ: {
2184 r = -EOPNOTSUPP;
2185 break;
2186 }
2187#ifdef KVM_CAP_ASSIGN_DEV_IRQ
2188 case KVM_ASSIGN_DEV_IRQ: {
2189 struct kvm_assigned_irq assigned_irq;
2190
2191 r = -EFAULT;
2192 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2193 goto out;
2194 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2195 if (r)
2196 goto out;
2197 break;
2198 }
2199 case KVM_DEASSIGN_DEV_IRQ: {
2200 struct kvm_assigned_irq assigned_irq;
2201
2202 r = -EFAULT;
2203 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2204 goto out;
2205 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2206 if (r)
2207 goto out;
2208 break;
2209 }
2210#endif
2211#endif
2212#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2213 case KVM_DEASSIGN_PCI_DEVICE: {
2214 struct kvm_assigned_pci_dev assigned_dev;
2215
2216 r = -EFAULT;
2217 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2218 goto out;
2219 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2220 if (r)
2221 goto out;
2222 break;
2223 }
2224#endif
2225#ifdef KVM_CAP_IRQ_ROUTING
2226 case KVM_SET_GSI_ROUTING: {
2227 struct kvm_irq_routing routing;
2228 struct kvm_irq_routing __user *urouting;
2229 struct kvm_irq_routing_entry *entries;
2230
2231 r = -EFAULT;
2232 if (copy_from_user(&routing, argp, sizeof(routing)))
2233 goto out;
2234 r = -EINVAL;
2235 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2236 goto out;
2237 if (routing.flags)
2238 goto out;
2239 r = -ENOMEM;
2240 entries = vmalloc(routing.nr * sizeof(*entries));
2241 if (!entries)
2242 goto out;
2243 r = -EFAULT;
2244 urouting = argp;
2245 if (copy_from_user(entries, urouting->entries,
2246 routing.nr * sizeof(*entries)))
2247 goto out_free_irq_routing;
2248 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2249 routing.flags);
2250 out_free_irq_routing:
2251 vfree(entries);
2252 break;
2253 }
2254#endif /* KVM_CAP_IRQ_ROUTING */
2255#ifdef __KVM_HAVE_MSIX
2256 case KVM_ASSIGN_SET_MSIX_NR: {
2257 struct kvm_assigned_msix_nr entry_nr;
2258 r = -EFAULT;
2259 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2260 goto out;
2261 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2262 if (r)
2263 goto out;
2264 break;
2265 }
2266 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2267 struct kvm_assigned_msix_entry entry;
2268 r = -EFAULT;
2269 if (copy_from_user(&entry, argp, sizeof entry))
2270 goto out;
2271 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2272 if (r)
2273 goto out;
2274 break;
2275 }
2276#endif
2277 case KVM_IRQFD: { 1508 case KVM_IRQFD: {
2278 struct kvm_irqfd data; 1509 struct kvm_irqfd data;
2279 1510
@@ -2305,11 +1536,59 @@ static long kvm_vm_ioctl(struct file *filp,
2305#endif 1536#endif
2306 default: 1537 default:
2307 r = kvm_arch_vm_ioctl(filp, ioctl, arg); 1538 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1539 if (r == -ENOTTY)
1540 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2308 } 1541 }
2309out: 1542out:
2310 return r; 1543 return r;
2311} 1544}
2312 1545
1546#ifdef CONFIG_COMPAT
1547struct compat_kvm_dirty_log {
1548 __u32 slot;
1549 __u32 padding1;
1550 union {
1551 compat_uptr_t dirty_bitmap; /* one bit per page */
1552 __u64 padding2;
1553 };
1554};
1555
1556static long kvm_vm_compat_ioctl(struct file *filp,
1557 unsigned int ioctl, unsigned long arg)
1558{
1559 struct kvm *kvm = filp->private_data;
1560 int r;
1561
1562 if (kvm->mm != current->mm)
1563 return -EIO;
1564 switch (ioctl) {
1565 case KVM_GET_DIRTY_LOG: {
1566 struct compat_kvm_dirty_log compat_log;
1567 struct kvm_dirty_log log;
1568
1569 r = -EFAULT;
1570 if (copy_from_user(&compat_log, (void __user *)arg,
1571 sizeof(compat_log)))
1572 goto out;
1573 log.slot = compat_log.slot;
1574 log.padding1 = compat_log.padding1;
1575 log.padding2 = compat_log.padding2;
1576 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1577
1578 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1579 if (r)
1580 goto out;
1581 break;
1582 }
1583 default:
1584 r = kvm_vm_ioctl(filp, ioctl, arg);
1585 }
1586
1587out:
1588 return r;
1589}
1590#endif
1591
2313static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1592static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2314{ 1593{
2315 struct page *page[1]; 1594 struct page *page[1];
@@ -2344,7 +1623,9 @@ static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2344static struct file_operations kvm_vm_fops = { 1623static struct file_operations kvm_vm_fops = {
2345 .release = kvm_vm_release, 1624 .release = kvm_vm_release,
2346 .unlocked_ioctl = kvm_vm_ioctl, 1625 .unlocked_ioctl = kvm_vm_ioctl,
2347 .compat_ioctl = kvm_vm_ioctl, 1626#ifdef CONFIG_COMPAT
1627 .compat_ioctl = kvm_vm_compat_ioctl,
1628#endif
2348 .mmap = kvm_vm_mmap, 1629 .mmap = kvm_vm_mmap,
2349}; 1630};
2350 1631
@@ -2372,6 +1653,7 @@ static long kvm_dev_ioctl_check_extension_generic(long arg)
2372#ifdef CONFIG_KVM_APIC_ARCHITECTURE 1653#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2373 case KVM_CAP_SET_BOOT_CPU_ID: 1654 case KVM_CAP_SET_BOOT_CPU_ID:
2374#endif 1655#endif
1656 case KVM_CAP_INTERNAL_ERROR_DATA:
2375 return 1; 1657 return 1;
2376#ifdef CONFIG_HAVE_KVM_IRQCHIP 1658#ifdef CONFIG_HAVE_KVM_IRQCHIP
2377 case KVM_CAP_IRQ_ROUTING: 1659 case KVM_CAP_IRQ_ROUTING:
@@ -2442,11 +1724,21 @@ static struct miscdevice kvm_dev = {
2442static void hardware_enable(void *junk) 1724static void hardware_enable(void *junk)
2443{ 1725{
2444 int cpu = raw_smp_processor_id(); 1726 int cpu = raw_smp_processor_id();
1727 int r;
2445 1728
2446 if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) 1729 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2447 return; 1730 return;
1731
2448 cpumask_set_cpu(cpu, cpus_hardware_enabled); 1732 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2449 kvm_arch_hardware_enable(NULL); 1733
1734 r = kvm_arch_hardware_enable(NULL);
1735
1736 if (r) {
1737 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1738 atomic_inc(&hardware_enable_failed);
1739 printk(KERN_INFO "kvm: enabling virtualization on "
1740 "CPU%d failed\n", cpu);
1741 }
2450} 1742}
2451 1743
2452static void hardware_disable(void *junk) 1744static void hardware_disable(void *junk)
@@ -2459,11 +1751,52 @@ static void hardware_disable(void *junk)
2459 kvm_arch_hardware_disable(NULL); 1751 kvm_arch_hardware_disable(NULL);
2460} 1752}
2461 1753
1754static void hardware_disable_all_nolock(void)
1755{
1756 BUG_ON(!kvm_usage_count);
1757
1758 kvm_usage_count--;
1759 if (!kvm_usage_count)
1760 on_each_cpu(hardware_disable, NULL, 1);
1761}
1762
1763static void hardware_disable_all(void)
1764{
1765 spin_lock(&kvm_lock);
1766 hardware_disable_all_nolock();
1767 spin_unlock(&kvm_lock);
1768}
1769
1770static int hardware_enable_all(void)
1771{
1772 int r = 0;
1773
1774 spin_lock(&kvm_lock);
1775
1776 kvm_usage_count++;
1777 if (kvm_usage_count == 1) {
1778 atomic_set(&hardware_enable_failed, 0);
1779 on_each_cpu(hardware_enable, NULL, 1);
1780
1781 if (atomic_read(&hardware_enable_failed)) {
1782 hardware_disable_all_nolock();
1783 r = -EBUSY;
1784 }
1785 }
1786
1787 spin_unlock(&kvm_lock);
1788
1789 return r;
1790}
1791
2462static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, 1792static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2463 void *v) 1793 void *v)
2464{ 1794{
2465 int cpu = (long)v; 1795 int cpu = (long)v;
2466 1796
1797 if (!kvm_usage_count)
1798 return NOTIFY_OK;
1799
2467 val &= ~CPU_TASKS_FROZEN; 1800 val &= ~CPU_TASKS_FROZEN;
2468 switch (val) { 1801 switch (val) {
2469 case CPU_DYING: 1802 case CPU_DYING:
@@ -2666,13 +1999,15 @@ static void kvm_exit_debug(void)
2666 1999
2667static int kvm_suspend(struct sys_device *dev, pm_message_t state) 2000static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2668{ 2001{
2669 hardware_disable(NULL); 2002 if (kvm_usage_count)
2003 hardware_disable(NULL);
2670 return 0; 2004 return 0;
2671} 2005}
2672 2006
2673static int kvm_resume(struct sys_device *dev) 2007static int kvm_resume(struct sys_device *dev)
2674{ 2008{
2675 hardware_enable(NULL); 2009 if (kvm_usage_count)
2010 hardware_enable(NULL);
2676 return 0; 2011 return 0;
2677} 2012}
2678 2013
@@ -2747,7 +2082,6 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2747 goto out_free_1; 2082 goto out_free_1;
2748 } 2083 }
2749 2084
2750 on_each_cpu(hardware_enable, NULL, 1);
2751 r = register_cpu_notifier(&kvm_cpu_notifier); 2085 r = register_cpu_notifier(&kvm_cpu_notifier);
2752 if (r) 2086 if (r)
2753 goto out_free_2; 2087 goto out_free_2;
@@ -2797,7 +2131,6 @@ out_free_3:
2797 unregister_reboot_notifier(&kvm_reboot_notifier); 2131 unregister_reboot_notifier(&kvm_reboot_notifier);
2798 unregister_cpu_notifier(&kvm_cpu_notifier); 2132 unregister_cpu_notifier(&kvm_cpu_notifier);
2799out_free_2: 2133out_free_2:
2800 on_each_cpu(hardware_disable, NULL, 1);
2801out_free_1: 2134out_free_1:
2802 kvm_arch_hardware_unsetup(); 2135 kvm_arch_hardware_unsetup();
2803out_free_0a: 2136out_free_0a:
generated by cgit v1.2.2 (git 2.25.0) at 2025-05-23 17:00:43 -0400