aboutsummaryrefslogtreecommitdiffstats
path: root/virt
diff options
context:
space:
mode:
Diffstat (limited to 'virt')
-rw-r--r--virt/kvm/Kconfig3
-rw-r--r--virt/kvm/assigned-dev.c821
-rw-r--r--virt/kvm/coalesced_mmio.c44
-rw-r--r--virt/kvm/coalesced_mmio.h15
-rw-r--r--virt/kvm/eventfd.c42
-rw-r--r--virt/kvm/ioapic.c119
-rw-r--r--virt/kvm/ioapic.h7
-rw-r--r--virt/kvm/iommu.c36
-rw-r--r--virt/kvm/irq_comm.c236
-rw-r--r--virt/kvm/kvm_main.c1366
10 files changed, 1582 insertions, 1107 deletions
diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
index daece36c0a57..7f1178f6b839 100644
--- a/virt/kvm/Kconfig
+++ b/virt/kvm/Kconfig
@@ -12,3 +12,6 @@ config HAVE_KVM_EVENTFD
12 12
13config KVM_APIC_ARCHITECTURE 13config KVM_APIC_ARCHITECTURE
14 bool 14 bool
15
16config KVM_MMIO
17 bool
diff --git a/virt/kvm/assigned-dev.c b/virt/kvm/assigned-dev.c
new file mode 100644
index 000000000000..02ff2b19dbe2
--- /dev/null
+++ b/virt/kvm/assigned-dev.c
@@ -0,0 +1,821 @@
1/*
2 * Kernel-based Virtual Machine - device assignment support
3 *
4 * Copyright (C) 2006-9 Red Hat, Inc
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2. See
7 * the COPYING file in the top-level directory.
8 *
9 */
10
11#include <linux/kvm_host.h>
12#include <linux/kvm.h>
13#include <linux/uaccess.h>
14#include <linux/vmalloc.h>
15#include <linux/errno.h>
16#include <linux/spinlock.h>
17#include <linux/pci.h>
18#include <linux/interrupt.h>
19#include <linux/slab.h>
20#include "irq.h"
21
22static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
23 int assigned_dev_id)
24{
25 struct list_head *ptr;
26 struct kvm_assigned_dev_kernel *match;
27
28 list_for_each(ptr, head) {
29 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
30 if (match->assigned_dev_id == assigned_dev_id)
31 return match;
32 }
33 return NULL;
34}
35
36static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
37 *assigned_dev, int irq)
38{
39 int i, index;
40 struct msix_entry *host_msix_entries;
41
42 host_msix_entries = assigned_dev->host_msix_entries;
43
44 index = -1;
45 for (i = 0; i < assigned_dev->entries_nr; i++)
46 if (irq == host_msix_entries[i].vector) {
47 index = i;
48 break;
49 }
50 if (index < 0) {
51 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
52 return 0;
53 }
54
55 return index;
56}
57
58static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
59{
60 struct kvm_assigned_dev_kernel *assigned_dev;
61 struct kvm *kvm;
62 int i;
63
64 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
65 interrupt_work);
66 kvm = assigned_dev->kvm;
67
68 spin_lock_irq(&assigned_dev->assigned_dev_lock);
69 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
70 struct kvm_guest_msix_entry *guest_entries =
71 assigned_dev->guest_msix_entries;
72 for (i = 0; i < assigned_dev->entries_nr; i++) {
73 if (!(guest_entries[i].flags &
74 KVM_ASSIGNED_MSIX_PENDING))
75 continue;
76 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
77 kvm_set_irq(assigned_dev->kvm,
78 assigned_dev->irq_source_id,
79 guest_entries[i].vector, 1);
80 }
81 } else
82 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
83 assigned_dev->guest_irq, 1);
84
85 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
86}
87
88static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
89{
90 unsigned long flags;
91 struct kvm_assigned_dev_kernel *assigned_dev =
92 (struct kvm_assigned_dev_kernel *) dev_id;
93
94 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
95 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
96 int index = find_index_from_host_irq(assigned_dev, irq);
97 if (index < 0)
98 goto out;
99 assigned_dev->guest_msix_entries[index].flags |=
100 KVM_ASSIGNED_MSIX_PENDING;
101 }
102
103 schedule_work(&assigned_dev->interrupt_work);
104
105 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
106 disable_irq_nosync(irq);
107 assigned_dev->host_irq_disabled = true;
108 }
109
110out:
111 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
112 return IRQ_HANDLED;
113}
114
115/* Ack the irq line for an assigned device */
116static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
117{
118 struct kvm_assigned_dev_kernel *dev;
119 unsigned long flags;
120
121 if (kian->gsi == -1)
122 return;
123
124 dev = container_of(kian, struct kvm_assigned_dev_kernel,
125 ack_notifier);
126
127 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
128
129 /* The guest irq may be shared so this ack may be
130 * from another device.
131 */
132 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
133 if (dev->host_irq_disabled) {
134 enable_irq(dev->host_irq);
135 dev->host_irq_disabled = false;
136 }
137 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
138}
139
140static void deassign_guest_irq(struct kvm *kvm,
141 struct kvm_assigned_dev_kernel *assigned_dev)
142{
143 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
144 assigned_dev->ack_notifier.gsi = -1;
145
146 if (assigned_dev->irq_source_id != -1)
147 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
148 assigned_dev->irq_source_id = -1;
149 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
150}
151
152/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
153static void deassign_host_irq(struct kvm *kvm,
154 struct kvm_assigned_dev_kernel *assigned_dev)
155{
156 /*
157 * In kvm_free_device_irq, cancel_work_sync return true if:
158 * 1. work is scheduled, and then cancelled.
159 * 2. work callback is executed.
160 *
161 * The first one ensured that the irq is disabled and no more events
162 * would happen. But for the second one, the irq may be enabled (e.g.
163 * for MSI). So we disable irq here to prevent further events.
164 *
165 * Notice this maybe result in nested disable if the interrupt type is
166 * INTx, but it's OK for we are going to free it.
167 *
168 * If this function is a part of VM destroy, please ensure that till
169 * now, the kvm state is still legal for probably we also have to wait
170 * interrupt_work done.
171 */
172 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
173 int i;
174 for (i = 0; i < assigned_dev->entries_nr; i++)
175 disable_irq_nosync(assigned_dev->
176 host_msix_entries[i].vector);
177
178 cancel_work_sync(&assigned_dev->interrupt_work);
179
180 for (i = 0; i < assigned_dev->entries_nr; i++)
181 free_irq(assigned_dev->host_msix_entries[i].vector,
182 (void *)assigned_dev);
183
184 assigned_dev->entries_nr = 0;
185 kfree(assigned_dev->host_msix_entries);
186 kfree(assigned_dev->guest_msix_entries);
187 pci_disable_msix(assigned_dev->dev);
188 } else {
189 /* Deal with MSI and INTx */
190 disable_irq_nosync(assigned_dev->host_irq);
191 cancel_work_sync(&assigned_dev->interrupt_work);
192
193 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
194
195 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
196 pci_disable_msi(assigned_dev->dev);
197 }
198
199 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
200}
201
202static int kvm_deassign_irq(struct kvm *kvm,
203 struct kvm_assigned_dev_kernel *assigned_dev,
204 unsigned long irq_requested_type)
205{
206 unsigned long guest_irq_type, host_irq_type;
207
208 if (!irqchip_in_kernel(kvm))
209 return -EINVAL;
210 /* no irq assignment to deassign */
211 if (!assigned_dev->irq_requested_type)
212 return -ENXIO;
213
214 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
215 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
216
217 if (host_irq_type)
218 deassign_host_irq(kvm, assigned_dev);
219 if (guest_irq_type)
220 deassign_guest_irq(kvm, assigned_dev);
221
222 return 0;
223}
224
225static void kvm_free_assigned_irq(struct kvm *kvm,
226 struct kvm_assigned_dev_kernel *assigned_dev)
227{
228 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
229}
230
231static void kvm_free_assigned_device(struct kvm *kvm,
232 struct kvm_assigned_dev_kernel
233 *assigned_dev)
234{
235 kvm_free_assigned_irq(kvm, assigned_dev);
236
237 pci_reset_function(assigned_dev->dev);
238
239 pci_release_regions(assigned_dev->dev);
240 pci_disable_device(assigned_dev->dev);
241 pci_dev_put(assigned_dev->dev);
242
243 list_del(&assigned_dev->list);
244 kfree(assigned_dev);
245}
246
247void kvm_free_all_assigned_devices(struct kvm *kvm)
248{
249 struct list_head *ptr, *ptr2;
250 struct kvm_assigned_dev_kernel *assigned_dev;
251
252 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
253 assigned_dev = list_entry(ptr,
254 struct kvm_assigned_dev_kernel,
255 list);
256
257 kvm_free_assigned_device(kvm, assigned_dev);
258 }
259}
260
261static int assigned_device_enable_host_intx(struct kvm *kvm,
262 struct kvm_assigned_dev_kernel *dev)
263{
264 dev->host_irq = dev->dev->irq;
265 /* Even though this is PCI, we don't want to use shared
266 * interrupts. Sharing host devices with guest-assigned devices
267 * on the same interrupt line is not a happy situation: there
268 * are going to be long delays in accepting, acking, etc.
269 */
270 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
271 0, "kvm_assigned_intx_device", (void *)dev))
272 return -EIO;
273 return 0;
274}
275
276#ifdef __KVM_HAVE_MSI
277static int assigned_device_enable_host_msi(struct kvm *kvm,
278 struct kvm_assigned_dev_kernel *dev)
279{
280 int r;
281
282 if (!dev->dev->msi_enabled) {
283 r = pci_enable_msi(dev->dev);
284 if (r)
285 return r;
286 }
287
288 dev->host_irq = dev->dev->irq;
289 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
290 "kvm_assigned_msi_device", (void *)dev)) {
291 pci_disable_msi(dev->dev);
292 return -EIO;
293 }
294
295 return 0;
296}
297#endif
298
299#ifdef __KVM_HAVE_MSIX
300static int assigned_device_enable_host_msix(struct kvm *kvm,
301 struct kvm_assigned_dev_kernel *dev)
302{
303 int i, r = -EINVAL;
304
305 /* host_msix_entries and guest_msix_entries should have been
306 * initialized */
307 if (dev->entries_nr == 0)
308 return r;
309
310 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
311 if (r)
312 return r;
313
314 for (i = 0; i < dev->entries_nr; i++) {
315 r = request_irq(dev->host_msix_entries[i].vector,
316 kvm_assigned_dev_intr, 0,
317 "kvm_assigned_msix_device",
318 (void *)dev);
319 /* FIXME: free requested_irq's on failure */
320 if (r)
321 return r;
322 }
323
324 return 0;
325}
326
327#endif
328
329static int assigned_device_enable_guest_intx(struct kvm *kvm,
330 struct kvm_assigned_dev_kernel *dev,
331 struct kvm_assigned_irq *irq)
332{
333 dev->guest_irq = irq->guest_irq;
334 dev->ack_notifier.gsi = irq->guest_irq;
335 return 0;
336}
337
338#ifdef __KVM_HAVE_MSI
339static int assigned_device_enable_guest_msi(struct kvm *kvm,
340 struct kvm_assigned_dev_kernel *dev,
341 struct kvm_assigned_irq *irq)
342{
343 dev->guest_irq = irq->guest_irq;
344 dev->ack_notifier.gsi = -1;
345 dev->host_irq_disabled = false;
346 return 0;
347}
348#endif
349
350#ifdef __KVM_HAVE_MSIX
351static int assigned_device_enable_guest_msix(struct kvm *kvm,
352 struct kvm_assigned_dev_kernel *dev,
353 struct kvm_assigned_irq *irq)
354{
355 dev->guest_irq = irq->guest_irq;
356 dev->ack_notifier.gsi = -1;
357 dev->host_irq_disabled = false;
358 return 0;
359}
360#endif
361
362static int assign_host_irq(struct kvm *kvm,
363 struct kvm_assigned_dev_kernel *dev,
364 __u32 host_irq_type)
365{
366 int r = -EEXIST;
367
368 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
369 return r;
370
371 switch (host_irq_type) {
372 case KVM_DEV_IRQ_HOST_INTX:
373 r = assigned_device_enable_host_intx(kvm, dev);
374 break;
375#ifdef __KVM_HAVE_MSI
376 case KVM_DEV_IRQ_HOST_MSI:
377 r = assigned_device_enable_host_msi(kvm, dev);
378 break;
379#endif
380#ifdef __KVM_HAVE_MSIX
381 case KVM_DEV_IRQ_HOST_MSIX:
382 r = assigned_device_enable_host_msix(kvm, dev);
383 break;
384#endif
385 default:
386 r = -EINVAL;
387 }
388
389 if (!r)
390 dev->irq_requested_type |= host_irq_type;
391
392 return r;
393}
394
395static int assign_guest_irq(struct kvm *kvm,
396 struct kvm_assigned_dev_kernel *dev,
397 struct kvm_assigned_irq *irq,
398 unsigned long guest_irq_type)
399{
400 int id;
401 int r = -EEXIST;
402
403 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
404 return r;
405
406 id = kvm_request_irq_source_id(kvm);
407 if (id < 0)
408 return id;
409
410 dev->irq_source_id = id;
411
412 switch (guest_irq_type) {
413 case KVM_DEV_IRQ_GUEST_INTX:
414 r = assigned_device_enable_guest_intx(kvm, dev, irq);
415 break;
416#ifdef __KVM_HAVE_MSI
417 case KVM_DEV_IRQ_GUEST_MSI:
418 r = assigned_device_enable_guest_msi(kvm, dev, irq);
419 break;
420#endif
421#ifdef __KVM_HAVE_MSIX
422 case KVM_DEV_IRQ_GUEST_MSIX:
423 r = assigned_device_enable_guest_msix(kvm, dev, irq);
424 break;
425#endif
426 default:
427 r = -EINVAL;
428 }
429
430 if (!r) {
431 dev->irq_requested_type |= guest_irq_type;
432 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
433 } else
434 kvm_free_irq_source_id(kvm, dev->irq_source_id);
435
436 return r;
437}
438
439/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
440static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
441 struct kvm_assigned_irq *assigned_irq)
442{
443 int r = -EINVAL;
444 struct kvm_assigned_dev_kernel *match;
445 unsigned long host_irq_type, guest_irq_type;
446
447 if (!capable(CAP_SYS_RAWIO))
448 return -EPERM;
449
450 if (!irqchip_in_kernel(kvm))
451 return r;
452
453 mutex_lock(&kvm->lock);
454 r = -ENODEV;
455 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
456 assigned_irq->assigned_dev_id);
457 if (!match)
458 goto out;
459
460 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
461 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
462
463 r = -EINVAL;
464 /* can only assign one type at a time */
465 if (hweight_long(host_irq_type) > 1)
466 goto out;
467 if (hweight_long(guest_irq_type) > 1)
468 goto out;
469 if (host_irq_type == 0 && guest_irq_type == 0)
470 goto out;
471
472 r = 0;
473 if (host_irq_type)
474 r = assign_host_irq(kvm, match, host_irq_type);
475 if (r)
476 goto out;
477
478 if (guest_irq_type)
479 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
480out:
481 mutex_unlock(&kvm->lock);
482 return r;
483}
484
485static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
486 struct kvm_assigned_irq
487 *assigned_irq)
488{
489 int r = -ENODEV;
490 struct kvm_assigned_dev_kernel *match;
491
492 mutex_lock(&kvm->lock);
493
494 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
495 assigned_irq->assigned_dev_id);
496 if (!match)
497 goto out;
498
499 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
500out:
501 mutex_unlock(&kvm->lock);
502 return r;
503}
504
505static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
506 struct kvm_assigned_pci_dev *assigned_dev)
507{
508 int r = 0, idx;
509 struct kvm_assigned_dev_kernel *match;
510 struct pci_dev *dev;
511
512 mutex_lock(&kvm->lock);
513 idx = srcu_read_lock(&kvm->srcu);
514
515 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
516 assigned_dev->assigned_dev_id);
517 if (match) {
518 /* device already assigned */
519 r = -EEXIST;
520 goto out;
521 }
522
523 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
524 if (match == NULL) {
525 printk(KERN_INFO "%s: Couldn't allocate memory\n",
526 __func__);
527 r = -ENOMEM;
528 goto out;
529 }
530 dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
531 assigned_dev->busnr,
532 assigned_dev->devfn);
533 if (!dev) {
534 printk(KERN_INFO "%s: host device not found\n", __func__);
535 r = -EINVAL;
536 goto out_free;
537 }
538 if (pci_enable_device(dev)) {
539 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
540 r = -EBUSY;
541 goto out_put;
542 }
543 r = pci_request_regions(dev, "kvm_assigned_device");
544 if (r) {
545 printk(KERN_INFO "%s: Could not get access to device regions\n",
546 __func__);
547 goto out_disable;
548 }
549
550 pci_reset_function(dev);
551
552 match->assigned_dev_id = assigned_dev->assigned_dev_id;
553 match->host_segnr = assigned_dev->segnr;
554 match->host_busnr = assigned_dev->busnr;
555 match->host_devfn = assigned_dev->devfn;
556 match->flags = assigned_dev->flags;
557 match->dev = dev;
558 spin_lock_init(&match->assigned_dev_lock);
559 match->irq_source_id = -1;
560 match->kvm = kvm;
561 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
562 INIT_WORK(&match->interrupt_work,
563 kvm_assigned_dev_interrupt_work_handler);
564
565 list_add(&match->list, &kvm->arch.assigned_dev_head);
566
567 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
568 if (!kvm->arch.iommu_domain) {
569 r = kvm_iommu_map_guest(kvm);
570 if (r)
571 goto out_list_del;
572 }
573 r = kvm_assign_device(kvm, match);
574 if (r)
575 goto out_list_del;
576 }
577
578out:
579 srcu_read_unlock(&kvm->srcu, idx);
580 mutex_unlock(&kvm->lock);
581 return r;
582out_list_del:
583 list_del(&match->list);
584 pci_release_regions(dev);
585out_disable:
586 pci_disable_device(dev);
587out_put:
588 pci_dev_put(dev);
589out_free:
590 kfree(match);
591 srcu_read_unlock(&kvm->srcu, idx);
592 mutex_unlock(&kvm->lock);
593 return r;
594}
595
596static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
597 struct kvm_assigned_pci_dev *assigned_dev)
598{
599 int r = 0;
600 struct kvm_assigned_dev_kernel *match;
601
602 mutex_lock(&kvm->lock);
603
604 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
605 assigned_dev->assigned_dev_id);
606 if (!match) {
607 printk(KERN_INFO "%s: device hasn't been assigned before, "
608 "so cannot be deassigned\n", __func__);
609 r = -EINVAL;
610 goto out;
611 }
612
613 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
614 kvm_deassign_device(kvm, match);
615
616 kvm_free_assigned_device(kvm, match);
617
618out:
619 mutex_unlock(&kvm->lock);
620 return r;
621}
622
623
624#ifdef __KVM_HAVE_MSIX
625static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
626 struct kvm_assigned_msix_nr *entry_nr)
627{
628 int r = 0;
629 struct kvm_assigned_dev_kernel *adev;
630
631 mutex_lock(&kvm->lock);
632
633 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
634 entry_nr->assigned_dev_id);
635 if (!adev) {
636 r = -EINVAL;
637 goto msix_nr_out;
638 }
639
640 if (adev->entries_nr == 0) {
641 adev->entries_nr = entry_nr->entry_nr;
642 if (adev->entries_nr == 0 ||
643 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
644 r = -EINVAL;
645 goto msix_nr_out;
646 }
647
648 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
649 entry_nr->entry_nr,
650 GFP_KERNEL);
651 if (!adev->host_msix_entries) {
652 r = -ENOMEM;
653 goto msix_nr_out;
654 }
655 adev->guest_msix_entries = kzalloc(
656 sizeof(struct kvm_guest_msix_entry) *
657 entry_nr->entry_nr, GFP_KERNEL);
658 if (!adev->guest_msix_entries) {
659 kfree(adev->host_msix_entries);
660 r = -ENOMEM;
661 goto msix_nr_out;
662 }
663 } else /* Not allowed set MSI-X number twice */
664 r = -EINVAL;
665msix_nr_out:
666 mutex_unlock(&kvm->lock);
667 return r;
668}
669
670static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
671 struct kvm_assigned_msix_entry *entry)
672{
673 int r = 0, i;
674 struct kvm_assigned_dev_kernel *adev;
675
676 mutex_lock(&kvm->lock);
677
678 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
679 entry->assigned_dev_id);
680
681 if (!adev) {
682 r = -EINVAL;
683 goto msix_entry_out;
684 }
685
686 for (i = 0; i < adev->entries_nr; i++)
687 if (adev->guest_msix_entries[i].vector == 0 ||
688 adev->guest_msix_entries[i].entry == entry->entry) {
689 adev->guest_msix_entries[i].entry = entry->entry;
690 adev->guest_msix_entries[i].vector = entry->gsi;
691 adev->host_msix_entries[i].entry = entry->entry;
692 break;
693 }
694 if (i == adev->entries_nr) {
695 r = -ENOSPC;
696 goto msix_entry_out;
697 }
698
699msix_entry_out:
700 mutex_unlock(&kvm->lock);
701
702 return r;
703}
704#endif
705
706long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
707 unsigned long arg)
708{
709 void __user *argp = (void __user *)arg;
710 int r = -ENOTTY;
711
712 switch (ioctl) {
713 case KVM_ASSIGN_PCI_DEVICE: {
714 struct kvm_assigned_pci_dev assigned_dev;
715
716 r = -EFAULT;
717 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
718 goto out;
719 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
720 if (r)
721 goto out;
722 break;
723 }
724 case KVM_ASSIGN_IRQ: {
725 r = -EOPNOTSUPP;
726 break;
727 }
728#ifdef KVM_CAP_ASSIGN_DEV_IRQ
729 case KVM_ASSIGN_DEV_IRQ: {
730 struct kvm_assigned_irq assigned_irq;
731
732 r = -EFAULT;
733 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
734 goto out;
735 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
736 if (r)
737 goto out;
738 break;
739 }
740 case KVM_DEASSIGN_DEV_IRQ: {
741 struct kvm_assigned_irq assigned_irq;
742
743 r = -EFAULT;
744 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
745 goto out;
746 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
747 if (r)
748 goto out;
749 break;
750 }
751#endif
752#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
753 case KVM_DEASSIGN_PCI_DEVICE: {
754 struct kvm_assigned_pci_dev assigned_dev;
755
756 r = -EFAULT;
757 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
758 goto out;
759 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
760 if (r)
761 goto out;
762 break;
763 }
764#endif
765#ifdef KVM_CAP_IRQ_ROUTING
766 case KVM_SET_GSI_ROUTING: {
767 struct kvm_irq_routing routing;
768 struct kvm_irq_routing __user *urouting;
769 struct kvm_irq_routing_entry *entries;
770
771 r = -EFAULT;
772 if (copy_from_user(&routing, argp, sizeof(routing)))
773 goto out;
774 r = -EINVAL;
775 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
776 goto out;
777 if (routing.flags)
778 goto out;
779 r = -ENOMEM;
780 entries = vmalloc(routing.nr * sizeof(*entries));
781 if (!entries)
782 goto out;
783 r = -EFAULT;
784 urouting = argp;
785 if (copy_from_user(entries, urouting->entries,
786 routing.nr * sizeof(*entries)))
787 goto out_free_irq_routing;
788 r = kvm_set_irq_routing(kvm, entries, routing.nr,
789 routing.flags);
790 out_free_irq_routing:
791 vfree(entries);
792 break;
793 }
794#endif /* KVM_CAP_IRQ_ROUTING */
795#ifdef __KVM_HAVE_MSIX
796 case KVM_ASSIGN_SET_MSIX_NR: {
797 struct kvm_assigned_msix_nr entry_nr;
798 r = -EFAULT;
799 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
800 goto out;
801 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
802 if (r)
803 goto out;
804 break;
805 }
806 case KVM_ASSIGN_SET_MSIX_ENTRY: {
807 struct kvm_assigned_msix_entry entry;
808 r = -EFAULT;
809 if (copy_from_user(&entry, argp, sizeof entry))
810 goto out;
811 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
812 if (r)
813 goto out;
814 break;
815 }
816#endif
817 }
818out:
819 return r;
820}
821
diff --git a/virt/kvm/coalesced_mmio.c b/virt/kvm/coalesced_mmio.c
index 04d69cd7049b..36e258029649 100644
--- a/virt/kvm/coalesced_mmio.c
+++ b/virt/kvm/coalesced_mmio.c
@@ -10,6 +10,7 @@
10#include "iodev.h" 10#include "iodev.h"
11 11
12#include <linux/kvm_host.h> 12#include <linux/kvm_host.h>
13#include <linux/slab.h>
13#include <linux/kvm.h> 14#include <linux/kvm.h>
14 15
15#include "coalesced_mmio.h" 16#include "coalesced_mmio.h"
@@ -92,41 +93,64 @@ static const struct kvm_io_device_ops coalesced_mmio_ops = {
92int kvm_coalesced_mmio_init(struct kvm *kvm) 93int kvm_coalesced_mmio_init(struct kvm *kvm)
93{ 94{
94 struct kvm_coalesced_mmio_dev *dev; 95 struct kvm_coalesced_mmio_dev *dev;
96 struct page *page;
95 int ret; 97 int ret;
96 98
99 ret = -ENOMEM;
100 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
101 if (!page)
102 goto out_err;
103 kvm->coalesced_mmio_ring = page_address(page);
104
105 ret = -ENOMEM;
97 dev = kzalloc(sizeof(struct kvm_coalesced_mmio_dev), GFP_KERNEL); 106 dev = kzalloc(sizeof(struct kvm_coalesced_mmio_dev), GFP_KERNEL);
98 if (!dev) 107 if (!dev)
99 return -ENOMEM; 108 goto out_free_page;
100 spin_lock_init(&dev->lock); 109 spin_lock_init(&dev->lock);
101 kvm_iodevice_init(&dev->dev, &coalesced_mmio_ops); 110 kvm_iodevice_init(&dev->dev, &coalesced_mmio_ops);
102 dev->kvm = kvm; 111 dev->kvm = kvm;
103 kvm->coalesced_mmio_dev = dev; 112 kvm->coalesced_mmio_dev = dev;
104 113
105 ret = kvm_io_bus_register_dev(kvm, &kvm->mmio_bus, &dev->dev); 114 mutex_lock(&kvm->slots_lock);
115 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, &dev->dev);
116 mutex_unlock(&kvm->slots_lock);
106 if (ret < 0) 117 if (ret < 0)
107 kfree(dev); 118 goto out_free_dev;
119
120 return ret;
108 121
122out_free_dev:
123 kfree(dev);
124out_free_page:
125 __free_page(page);
126out_err:
109 return ret; 127 return ret;
110} 128}
111 129
130void kvm_coalesced_mmio_free(struct kvm *kvm)
131{
132 if (kvm->coalesced_mmio_ring)
133 free_page((unsigned long)kvm->coalesced_mmio_ring);
134}
135
112int kvm_vm_ioctl_register_coalesced_mmio(struct kvm *kvm, 136int kvm_vm_ioctl_register_coalesced_mmio(struct kvm *kvm,
113 struct kvm_coalesced_mmio_zone *zone) 137 struct kvm_coalesced_mmio_zone *zone)
114{ 138{
115 struct kvm_coalesced_mmio_dev *dev = kvm->coalesced_mmio_dev; 139 struct kvm_coalesced_mmio_dev *dev = kvm->coalesced_mmio_dev;
116 140
117 if (dev == NULL) 141 if (dev == NULL)
118 return -EINVAL; 142 return -EINVAL;
119 143
120 down_write(&kvm->slots_lock); 144 mutex_lock(&kvm->slots_lock);
121 if (dev->nb_zones >= KVM_COALESCED_MMIO_ZONE_MAX) { 145 if (dev->nb_zones >= KVM_COALESCED_MMIO_ZONE_MAX) {
122 up_write(&kvm->slots_lock); 146 mutex_unlock(&kvm->slots_lock);
123 return -ENOBUFS; 147 return -ENOBUFS;
124 } 148 }
125 149
126 dev->zone[dev->nb_zones] = *zone; 150 dev->zone[dev->nb_zones] = *zone;
127 dev->nb_zones++; 151 dev->nb_zones++;
128 152
129 up_write(&kvm->slots_lock); 153 mutex_unlock(&kvm->slots_lock);
130 return 0; 154 return 0;
131} 155}
132 156
@@ -140,10 +164,10 @@ int kvm_vm_ioctl_unregister_coalesced_mmio(struct kvm *kvm,
140 if (dev == NULL) 164 if (dev == NULL)
141 return -EINVAL; 165 return -EINVAL;
142 166
143 down_write(&kvm->slots_lock); 167 mutex_lock(&kvm->slots_lock);
144 168
145 i = dev->nb_zones; 169 i = dev->nb_zones;
146 while(i) { 170 while (i) {
147 z = &dev->zone[i - 1]; 171 z = &dev->zone[i - 1];
148 172
149 /* unregister all zones 173 /* unregister all zones
@@ -158,7 +182,7 @@ int kvm_vm_ioctl_unregister_coalesced_mmio(struct kvm *kvm,
158 i--; 182 i--;
159 } 183 }
160 184
161 up_write(&kvm->slots_lock); 185 mutex_unlock(&kvm->slots_lock);
162 186
163 return 0; 187 return 0;
164} 188}
diff --git a/virt/kvm/coalesced_mmio.h b/virt/kvm/coalesced_mmio.h
index 4b49f27fa31e..8a5959e3535f 100644
--- a/virt/kvm/coalesced_mmio.h
+++ b/virt/kvm/coalesced_mmio.h
@@ -1,3 +1,6 @@
1#ifndef __KVM_COALESCED_MMIO_H__
2#define __KVM_COALESCED_MMIO_H__
3
1/* 4/*
2 * KVM coalesced MMIO 5 * KVM coalesced MMIO
3 * 6 *
@@ -7,6 +10,8 @@
7 * 10 *
8 */ 11 */
9 12
13#ifdef CONFIG_KVM_MMIO
14
10#define KVM_COALESCED_MMIO_ZONE_MAX 100 15#define KVM_COALESCED_MMIO_ZONE_MAX 100
11 16
12struct kvm_coalesced_mmio_dev { 17struct kvm_coalesced_mmio_dev {
@@ -18,7 +23,17 @@ struct kvm_coalesced_mmio_dev {
18}; 23};
19 24
20int kvm_coalesced_mmio_init(struct kvm *kvm); 25int kvm_coalesced_mmio_init(struct kvm *kvm);
26void kvm_coalesced_mmio_free(struct kvm *kvm);
21int kvm_vm_ioctl_register_coalesced_mmio(struct kvm *kvm, 27int kvm_vm_ioctl_register_coalesced_mmio(struct kvm *kvm,
22 struct kvm_coalesced_mmio_zone *zone); 28 struct kvm_coalesced_mmio_zone *zone);
23int kvm_vm_ioctl_unregister_coalesced_mmio(struct kvm *kvm, 29int kvm_vm_ioctl_unregister_coalesced_mmio(struct kvm *kvm,
24 struct kvm_coalesced_mmio_zone *zone); 30 struct kvm_coalesced_mmio_zone *zone);
31
32#else
33
34static inline int kvm_coalesced_mmio_init(struct kvm *kvm) { return 0; }
35static inline void kvm_coalesced_mmio_free(struct kvm *kvm) { }
36
37#endif
38
39#endif
diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
index bb4ebd89b9ff..b81f0ebbaaad 100644
--- a/virt/kvm/eventfd.c
+++ b/virt/kvm/eventfd.c
@@ -30,6 +30,7 @@
30#include <linux/list.h> 30#include <linux/list.h>
31#include <linux/eventfd.h> 31#include <linux/eventfd.h>
32#include <linux/kernel.h> 32#include <linux/kernel.h>
33#include <linux/slab.h>
33 34
34#include "iodev.h" 35#include "iodev.h"
35 36
@@ -47,7 +48,6 @@ struct _irqfd {
47 int gsi; 48 int gsi;
48 struct list_head list; 49 struct list_head list;
49 poll_table pt; 50 poll_table pt;
50 wait_queue_head_t *wqh;
51 wait_queue_t wait; 51 wait_queue_t wait;
52 struct work_struct inject; 52 struct work_struct inject;
53 struct work_struct shutdown; 53 struct work_struct shutdown;
@@ -61,10 +61,8 @@ irqfd_inject(struct work_struct *work)
61 struct _irqfd *irqfd = container_of(work, struct _irqfd, inject); 61 struct _irqfd *irqfd = container_of(work, struct _irqfd, inject);
62 struct kvm *kvm = irqfd->kvm; 62 struct kvm *kvm = irqfd->kvm;
63 63
64 mutex_lock(&kvm->irq_lock);
65 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1); 64 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1);
66 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0); 65 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0);
67 mutex_unlock(&kvm->irq_lock);
68} 66}
69 67
70/* 68/*
@@ -74,12 +72,13 @@ static void
74irqfd_shutdown(struct work_struct *work) 72irqfd_shutdown(struct work_struct *work)
75{ 73{
76 struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown); 74 struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown);
75 u64 cnt;
77 76
78 /* 77 /*
79 * Synchronize with the wait-queue and unhook ourselves to prevent 78 * Synchronize with the wait-queue and unhook ourselves to prevent
80 * further events. 79 * further events.
81 */ 80 */
82 remove_wait_queue(irqfd->wqh, &irqfd->wait); 81 eventfd_ctx_remove_wait_queue(irqfd->eventfd, &irqfd->wait, &cnt);
83 82
84 /* 83 /*
85 * We know no new events will be scheduled at this point, so block 84 * We know no new events will be scheduled at this point, so block
@@ -160,15 +159,13 @@ irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh,
160 poll_table *pt) 159 poll_table *pt)
161{ 160{
162 struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt); 161 struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt);
163
164 irqfd->wqh = wqh;
165 add_wait_queue(wqh, &irqfd->wait); 162 add_wait_queue(wqh, &irqfd->wait);
166} 163}
167 164
168static int 165static int
169kvm_irqfd_assign(struct kvm *kvm, int fd, int gsi) 166kvm_irqfd_assign(struct kvm *kvm, int fd, int gsi)
170{ 167{
171 struct _irqfd *irqfd; 168 struct _irqfd *irqfd, *tmp;
172 struct file *file = NULL; 169 struct file *file = NULL;
173 struct eventfd_ctx *eventfd = NULL; 170 struct eventfd_ctx *eventfd = NULL;
174 int ret; 171 int ret;
@@ -205,9 +202,20 @@ kvm_irqfd_assign(struct kvm *kvm, int fd, int gsi)
205 init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup); 202 init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup);
206 init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc); 203 init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc);
207 204
205 spin_lock_irq(&kvm->irqfds.lock);
206
207 ret = 0;
208 list_for_each_entry(tmp, &kvm->irqfds.items, list) {
209 if (irqfd->eventfd != tmp->eventfd)
210 continue;
211 /* This fd is used for another irq already. */
212 ret = -EBUSY;
213 spin_unlock_irq(&kvm->irqfds.lock);
214 goto fail;
215 }
216
208 events = file->f_op->poll(file, &irqfd->pt); 217 events = file->f_op->poll(file, &irqfd->pt);
209 218
210 spin_lock_irq(&kvm->irqfds.lock);
211 list_add_tail(&irqfd->list, &kvm->irqfds.items); 219 list_add_tail(&irqfd->list, &kvm->irqfds.items);
212 spin_unlock_irq(&kvm->irqfds.lock); 220 spin_unlock_irq(&kvm->irqfds.lock);
213 221
@@ -453,7 +461,7 @@ static int
453kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) 461kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
454{ 462{
455 int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; 463 int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO;
456 struct kvm_io_bus *bus = pio ? &kvm->pio_bus : &kvm->mmio_bus; 464 enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS;
457 struct _ioeventfd *p; 465 struct _ioeventfd *p;
458 struct eventfd_ctx *eventfd; 466 struct eventfd_ctx *eventfd;
459 int ret; 467 int ret;
@@ -498,7 +506,7 @@ kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
498 else 506 else
499 p->wildcard = true; 507 p->wildcard = true;
500 508
501 down_write(&kvm->slots_lock); 509 mutex_lock(&kvm->slots_lock);
502 510
503 /* Verify that there isnt a match already */ 511 /* Verify that there isnt a match already */
504 if (ioeventfd_check_collision(kvm, p)) { 512 if (ioeventfd_check_collision(kvm, p)) {
@@ -508,18 +516,18 @@ kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
508 516
509 kvm_iodevice_init(&p->dev, &ioeventfd_ops); 517 kvm_iodevice_init(&p->dev, &ioeventfd_ops);
510 518
511 ret = __kvm_io_bus_register_dev(bus, &p->dev); 519 ret = kvm_io_bus_register_dev(kvm, bus_idx, &p->dev);
512 if (ret < 0) 520 if (ret < 0)
513 goto unlock_fail; 521 goto unlock_fail;
514 522
515 list_add_tail(&p->list, &kvm->ioeventfds); 523 list_add_tail(&p->list, &kvm->ioeventfds);
516 524
517 up_write(&kvm->slots_lock); 525 mutex_unlock(&kvm->slots_lock);
518 526
519 return 0; 527 return 0;
520 528
521unlock_fail: 529unlock_fail:
522 up_write(&kvm->slots_lock); 530 mutex_unlock(&kvm->slots_lock);
523 531
524fail: 532fail:
525 kfree(p); 533 kfree(p);
@@ -532,7 +540,7 @@ static int
532kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) 540kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
533{ 541{
534 int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; 542 int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO;
535 struct kvm_io_bus *bus = pio ? &kvm->pio_bus : &kvm->mmio_bus; 543 enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS;
536 struct _ioeventfd *p, *tmp; 544 struct _ioeventfd *p, *tmp;
537 struct eventfd_ctx *eventfd; 545 struct eventfd_ctx *eventfd;
538 int ret = -ENOENT; 546 int ret = -ENOENT;
@@ -541,7 +549,7 @@ kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
541 if (IS_ERR(eventfd)) 549 if (IS_ERR(eventfd))
542 return PTR_ERR(eventfd); 550 return PTR_ERR(eventfd);
543 551
544 down_write(&kvm->slots_lock); 552 mutex_lock(&kvm->slots_lock);
545 553
546 list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) { 554 list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) {
547 bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH); 555 bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH);
@@ -555,13 +563,13 @@ kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
555 if (!p->wildcard && p->datamatch != args->datamatch) 563 if (!p->wildcard && p->datamatch != args->datamatch)
556 continue; 564 continue;
557 565
558 __kvm_io_bus_unregister_dev(bus, &p->dev); 566 kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev);
559 ioeventfd_release(p); 567 ioeventfd_release(p);
560 ret = 0; 568 ret = 0;
561 break; 569 break;
562 } 570 }
563 571
564 up_write(&kvm->slots_lock); 572 mutex_unlock(&kvm->slots_lock);
565 573
566 eventfd_ctx_put(eventfd); 574 eventfd_ctx_put(eventfd);
567 575
diff --git a/virt/kvm/ioapic.c b/virt/kvm/ioapic.c
index 9fe140bb38ec..7c79c1d76d0c 100644
--- a/virt/kvm/ioapic.c
+++ b/virt/kvm/ioapic.c
@@ -33,6 +33,7 @@
33#include <linux/smp.h> 33#include <linux/smp.h>
34#include <linux/hrtimer.h> 34#include <linux/hrtimer.h>
35#include <linux/io.h> 35#include <linux/io.h>
36#include <linux/slab.h>
36#include <asm/processor.h> 37#include <asm/processor.h>
37#include <asm/page.h> 38#include <asm/page.h>
38#include <asm/current.h> 39#include <asm/current.h>
@@ -100,6 +101,19 @@ static int ioapic_service(struct kvm_ioapic *ioapic, unsigned int idx)
100 return injected; 101 return injected;
101} 102}
102 103
104static void update_handled_vectors(struct kvm_ioapic *ioapic)
105{
106 DECLARE_BITMAP(handled_vectors, 256);
107 int i;
108
109 memset(handled_vectors, 0, sizeof(handled_vectors));
110 for (i = 0; i < IOAPIC_NUM_PINS; ++i)
111 __set_bit(ioapic->redirtbl[i].fields.vector, handled_vectors);
112 memcpy(ioapic->handled_vectors, handled_vectors,
113 sizeof(handled_vectors));
114 smp_wmb();
115}
116
103static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val) 117static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
104{ 118{
105 unsigned index; 119 unsigned index;
@@ -134,6 +148,7 @@ static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
134 e->bits |= (u32) val; 148 e->bits |= (u32) val;
135 e->fields.remote_irr = 0; 149 e->fields.remote_irr = 0;
136 } 150 }
151 update_handled_vectors(ioapic);
137 mask_after = e->fields.mask; 152 mask_after = e->fields.mask;
138 if (mask_before != mask_after) 153 if (mask_before != mask_after)
139 kvm_fire_mask_notifiers(ioapic->kvm, index, mask_after); 154 kvm_fire_mask_notifiers(ioapic->kvm, index, mask_after);
@@ -182,6 +197,7 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)
182 union kvm_ioapic_redirect_entry entry; 197 union kvm_ioapic_redirect_entry entry;
183 int ret = 1; 198 int ret = 1;
184 199
200 spin_lock(&ioapic->lock);
185 if (irq >= 0 && irq < IOAPIC_NUM_PINS) { 201 if (irq >= 0 && irq < IOAPIC_NUM_PINS) {
186 entry = ioapic->redirtbl[irq]; 202 entry = ioapic->redirtbl[irq];
187 level ^= entry.fields.polarity; 203 level ^= entry.fields.polarity;
@@ -198,34 +214,54 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)
198 } 214 }
199 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0); 215 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
200 } 216 }
217 spin_unlock(&ioapic->lock);
218
201 return ret; 219 return ret;
202} 220}
203 221
204static void __kvm_ioapic_update_eoi(struct kvm_ioapic *ioapic, int pin, 222static void __kvm_ioapic_update_eoi(struct kvm_ioapic *ioapic, int vector,
205 int trigger_mode) 223 int trigger_mode)
206{ 224{
207 union kvm_ioapic_redirect_entry *ent; 225 int i;
226
227 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
228 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
208 229
209 ent = &ioapic->redirtbl[pin]; 230 if (ent->fields.vector != vector)
231 continue;
210 232
211 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, pin); 233 /*
234 * We are dropping lock while calling ack notifiers because ack
235 * notifier callbacks for assigned devices call into IOAPIC
236 * recursively. Since remote_irr is cleared only after call
237 * to notifiers if the same vector will be delivered while lock
238 * is dropped it will be put into irr and will be delivered
239 * after ack notifier returns.
240 */
241 spin_unlock(&ioapic->lock);
242 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
243 spin_lock(&ioapic->lock);
244
245 if (trigger_mode != IOAPIC_LEVEL_TRIG)
246 continue;
212 247
213 if (trigger_mode == IOAPIC_LEVEL_TRIG) {
214 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG); 248 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
215 ent->fields.remote_irr = 0; 249 ent->fields.remote_irr = 0;
216 if (!ent->fields.mask && (ioapic->irr & (1 << pin))) 250 if (!ent->fields.mask && (ioapic->irr & (1 << i)))
217 ioapic_service(ioapic, pin); 251 ioapic_service(ioapic, i);
218 } 252 }
219} 253}
220 254
221void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode) 255void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode)
222{ 256{
223 struct kvm_ioapic *ioapic = kvm->arch.vioapic; 257 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
224 int i;
225 258
226 for (i = 0; i < IOAPIC_NUM_PINS; i++) 259 smp_rmb();
227 if (ioapic->redirtbl[i].fields.vector == vector) 260 if (!test_bit(vector, ioapic->handled_vectors))
228 __kvm_ioapic_update_eoi(ioapic, i, trigger_mode); 261 return;
262 spin_lock(&ioapic->lock);
263 __kvm_ioapic_update_eoi(ioapic, vector, trigger_mode);
264 spin_unlock(&ioapic->lock);
229} 265}
230 266
231static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev) 267static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
@@ -250,8 +286,8 @@ static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
250 ioapic_debug("addr %lx\n", (unsigned long)addr); 286 ioapic_debug("addr %lx\n", (unsigned long)addr);
251 ASSERT(!(addr & 0xf)); /* check alignment */ 287 ASSERT(!(addr & 0xf)); /* check alignment */
252 288
253 mutex_lock(&ioapic->kvm->irq_lock);
254 addr &= 0xff; 289 addr &= 0xff;
290 spin_lock(&ioapic->lock);
255 switch (addr) { 291 switch (addr) {
256 case IOAPIC_REG_SELECT: 292 case IOAPIC_REG_SELECT:
257 result = ioapic->ioregsel; 293 result = ioapic->ioregsel;
@@ -265,6 +301,8 @@ static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
265 result = 0; 301 result = 0;
266 break; 302 break;
267 } 303 }
304 spin_unlock(&ioapic->lock);
305
268 switch (len) { 306 switch (len) {
269 case 8: 307 case 8:
270 *(u64 *) val = result; 308 *(u64 *) val = result;
@@ -277,7 +315,6 @@ static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
277 default: 315 default:
278 printk(KERN_WARNING "ioapic: wrong length %d\n", len); 316 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
279 } 317 }
280 mutex_unlock(&ioapic->kvm->irq_lock);
281 return 0; 318 return 0;
282} 319}
283 320
@@ -293,15 +330,15 @@ static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
293 (void*)addr, len, val); 330 (void*)addr, len, val);
294 ASSERT(!(addr & 0xf)); /* check alignment */ 331 ASSERT(!(addr & 0xf)); /* check alignment */
295 332
296 mutex_lock(&ioapic->kvm->irq_lock);
297 if (len == 4 || len == 8) 333 if (len == 4 || len == 8)
298 data = *(u32 *) val; 334 data = *(u32 *) val;
299 else { 335 else {
300 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len); 336 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
301 goto unlock; 337 return 0;
302 } 338 }
303 339
304 addr &= 0xff; 340 addr &= 0xff;
341 spin_lock(&ioapic->lock);
305 switch (addr) { 342 switch (addr) {
306 case IOAPIC_REG_SELECT: 343 case IOAPIC_REG_SELECT:
307 ioapic->ioregsel = data; 344 ioapic->ioregsel = data;
@@ -312,15 +349,14 @@ static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
312 break; 349 break;
313#ifdef CONFIG_IA64 350#ifdef CONFIG_IA64
314 case IOAPIC_REG_EOI: 351 case IOAPIC_REG_EOI:
315 kvm_ioapic_update_eoi(ioapic->kvm, data, IOAPIC_LEVEL_TRIG); 352 __kvm_ioapic_update_eoi(ioapic, data, IOAPIC_LEVEL_TRIG);
316 break; 353 break;
317#endif 354#endif
318 355
319 default: 356 default:
320 break; 357 break;
321 } 358 }
322unlock: 359 spin_unlock(&ioapic->lock);
323 mutex_unlock(&ioapic->kvm->irq_lock);
324 return 0; 360 return 0;
325} 361}
326 362
@@ -334,6 +370,7 @@ void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
334 ioapic->ioregsel = 0; 370 ioapic->ioregsel = 0;
335 ioapic->irr = 0; 371 ioapic->irr = 0;
336 ioapic->id = 0; 372 ioapic->id = 0;
373 update_handled_vectors(ioapic);
337} 374}
338 375
339static const struct kvm_io_device_ops ioapic_mmio_ops = { 376static const struct kvm_io_device_ops ioapic_mmio_ops = {
@@ -349,14 +386,54 @@ int kvm_ioapic_init(struct kvm *kvm)
349 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL); 386 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
350 if (!ioapic) 387 if (!ioapic)
351 return -ENOMEM; 388 return -ENOMEM;
389 spin_lock_init(&ioapic->lock);
352 kvm->arch.vioapic = ioapic; 390 kvm->arch.vioapic = ioapic;
353 kvm_ioapic_reset(ioapic); 391 kvm_ioapic_reset(ioapic);
354 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops); 392 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
355 ioapic->kvm = kvm; 393 ioapic->kvm = kvm;
356 ret = kvm_io_bus_register_dev(kvm, &kvm->mmio_bus, &ioapic->dev); 394 mutex_lock(&kvm->slots_lock);
357 if (ret < 0) 395 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
396 mutex_unlock(&kvm->slots_lock);
397 if (ret < 0) {
398 kvm->arch.vioapic = NULL;
358 kfree(ioapic); 399 kfree(ioapic);
400 }
359 401
360 return ret; 402 return ret;
361} 403}
362 404
405void kvm_ioapic_destroy(struct kvm *kvm)
406{
407 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
408
409 if (ioapic) {
410 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
411 kvm->arch.vioapic = NULL;
412 kfree(ioapic);
413 }
414}
415
416int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
417{
418 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
419 if (!ioapic)
420 return -EINVAL;
421
422 spin_lock(&ioapic->lock);
423 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
424 spin_unlock(&ioapic->lock);
425 return 0;
426}
427
428int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
429{
430 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
431 if (!ioapic)
432 return -EINVAL;
433
434 spin_lock(&ioapic->lock);
435 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
436 update_handled_vectors(ioapic);
437 spin_unlock(&ioapic->lock);
438 return 0;
439}
diff --git a/virt/kvm/ioapic.h b/virt/kvm/ioapic.h
index 7080b713c160..0b190c34ccc3 100644
--- a/virt/kvm/ioapic.h
+++ b/virt/kvm/ioapic.h
@@ -41,9 +41,12 @@ struct kvm_ioapic {
41 u32 irr; 41 u32 irr;
42 u32 pad; 42 u32 pad;
43 union kvm_ioapic_redirect_entry redirtbl[IOAPIC_NUM_PINS]; 43 union kvm_ioapic_redirect_entry redirtbl[IOAPIC_NUM_PINS];
44 unsigned long irq_states[IOAPIC_NUM_PINS];
44 struct kvm_io_device dev; 45 struct kvm_io_device dev;
45 struct kvm *kvm; 46 struct kvm *kvm;
46 void (*ack_notifier)(void *opaque, int irq); 47 void (*ack_notifier)(void *opaque, int irq);
48 spinlock_t lock;
49 DECLARE_BITMAP(handled_vectors, 256);
47}; 50};
48 51
49#ifdef DEBUG 52#ifdef DEBUG
@@ -69,8 +72,12 @@ int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
69int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2); 72int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2);
70void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode); 73void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);
71int kvm_ioapic_init(struct kvm *kvm); 74int kvm_ioapic_init(struct kvm *kvm);
75void kvm_ioapic_destroy(struct kvm *kvm);
72int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level); 76int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level);
73void kvm_ioapic_reset(struct kvm_ioapic *ioapic); 77void kvm_ioapic_reset(struct kvm_ioapic *ioapic);
74int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src, 78int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
75 struct kvm_lapic_irq *irq); 79 struct kvm_lapic_irq *irq);
80int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state);
81int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state);
82
76#endif 83#endif
diff --git a/virt/kvm/iommu.c b/virt/kvm/iommu.c
index 15147583abd1..80fd3ad3b2de 100644
--- a/virt/kvm/iommu.c
+++ b/virt/kvm/iommu.c
@@ -32,10 +32,10 @@ static int kvm_iommu_unmap_memslots(struct kvm *kvm);
32static void kvm_iommu_put_pages(struct kvm *kvm, 32static void kvm_iommu_put_pages(struct kvm *kvm,
33 gfn_t base_gfn, unsigned long npages); 33 gfn_t base_gfn, unsigned long npages);
34 34
35int kvm_iommu_map_pages(struct kvm *kvm, 35int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
36 gfn_t base_gfn, unsigned long npages)
37{ 36{
38 gfn_t gfn = base_gfn; 37 gfn_t gfn = slot->base_gfn;
38 unsigned long npages = slot->npages;
39 pfn_t pfn; 39 pfn_t pfn;
40 int i, r = 0; 40 int i, r = 0;
41 struct iommu_domain *domain = kvm->arch.iommu_domain; 41 struct iommu_domain *domain = kvm->arch.iommu_domain;
@@ -54,7 +54,7 @@ int kvm_iommu_map_pages(struct kvm *kvm,
54 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) 54 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn)))
55 continue; 55 continue;
56 56
57 pfn = gfn_to_pfn(kvm, gfn); 57 pfn = gfn_to_pfn_memslot(kvm, slot, gfn);
58 r = iommu_map_range(domain, 58 r = iommu_map_range(domain,
59 gfn_to_gpa(gfn), 59 gfn_to_gpa(gfn),
60 pfn_to_hpa(pfn), 60 pfn_to_hpa(pfn),
@@ -69,17 +69,19 @@ int kvm_iommu_map_pages(struct kvm *kvm,
69 return 0; 69 return 0;
70 70
71unmap_pages: 71unmap_pages:
72 kvm_iommu_put_pages(kvm, base_gfn, i); 72 kvm_iommu_put_pages(kvm, slot->base_gfn, i);
73 return r; 73 return r;
74} 74}
75 75
76static int kvm_iommu_map_memslots(struct kvm *kvm) 76static int kvm_iommu_map_memslots(struct kvm *kvm)
77{ 77{
78 int i, r = 0; 78 int i, r = 0;
79 struct kvm_memslots *slots;
80
81 slots = rcu_dereference(kvm->memslots);
79 82
80 for (i = 0; i < kvm->nmemslots; i++) { 83 for (i = 0; i < slots->nmemslots; i++) {
81 r = kvm_iommu_map_pages(kvm, kvm->memslots[i].base_gfn, 84 r = kvm_iommu_map_pages(kvm, &slots->memslots[i]);
82 kvm->memslots[i].npages);
83 if (r) 85 if (r)
84 break; 86 break;
85 } 87 }
@@ -104,7 +106,8 @@ int kvm_assign_device(struct kvm *kvm,
104 106
105 r = iommu_attach_device(domain, &pdev->dev); 107 r = iommu_attach_device(domain, &pdev->dev);
106 if (r) { 108 if (r) {
107 printk(KERN_ERR "assign device %x:%x.%x failed", 109 printk(KERN_ERR "assign device %x:%x:%x.%x failed",
110 pci_domain_nr(pdev->bus),
108 pdev->bus->number, 111 pdev->bus->number,
109 PCI_SLOT(pdev->devfn), 112 PCI_SLOT(pdev->devfn),
110 PCI_FUNC(pdev->devfn)); 113 PCI_FUNC(pdev->devfn));
@@ -125,7 +128,8 @@ int kvm_assign_device(struct kvm *kvm,
125 goto out_unmap; 128 goto out_unmap;
126 } 129 }
127 130
128 printk(KERN_DEBUG "assign device: host bdf = %x:%x:%x\n", 131 printk(KERN_DEBUG "assign device %x:%x:%x.%x\n",
132 assigned_dev->host_segnr,
129 assigned_dev->host_busnr, 133 assigned_dev->host_busnr,
130 PCI_SLOT(assigned_dev->host_devfn), 134 PCI_SLOT(assigned_dev->host_devfn),
131 PCI_FUNC(assigned_dev->host_devfn)); 135 PCI_FUNC(assigned_dev->host_devfn));
@@ -152,7 +156,8 @@ int kvm_deassign_device(struct kvm *kvm,
152 156
153 iommu_detach_device(domain, &pdev->dev); 157 iommu_detach_device(domain, &pdev->dev);
154 158
155 printk(KERN_DEBUG "deassign device: host bdf = %x:%x:%x\n", 159 printk(KERN_DEBUG "deassign device %x:%x:%x.%x\n",
160 assigned_dev->host_segnr,
156 assigned_dev->host_busnr, 161 assigned_dev->host_busnr,
157 PCI_SLOT(assigned_dev->host_devfn), 162 PCI_SLOT(assigned_dev->host_devfn),
158 PCI_FUNC(assigned_dev->host_devfn)); 163 PCI_FUNC(assigned_dev->host_devfn));
@@ -210,10 +215,13 @@ static void kvm_iommu_put_pages(struct kvm *kvm,
210static int kvm_iommu_unmap_memslots(struct kvm *kvm) 215static int kvm_iommu_unmap_memslots(struct kvm *kvm)
211{ 216{
212 int i; 217 int i;
218 struct kvm_memslots *slots;
219
220 slots = rcu_dereference(kvm->memslots);
213 221
214 for (i = 0; i < kvm->nmemslots; i++) { 222 for (i = 0; i < slots->nmemslots; i++) {
215 kvm_iommu_put_pages(kvm, kvm->memslots[i].base_gfn, 223 kvm_iommu_put_pages(kvm, slots->memslots[i].base_gfn,
216 kvm->memslots[i].npages); 224 slots->memslots[i].npages);
217 } 225 }
218 226
219 return 0; 227 return 0;
diff --git a/virt/kvm/irq_comm.c b/virt/kvm/irq_comm.c
index 001663ff401a..a0e88809e45e 100644
--- a/virt/kvm/irq_comm.c
+++ b/virt/kvm/irq_comm.c
@@ -20,6 +20,7 @@
20 */ 20 */
21 21
22#include <linux/kvm_host.h> 22#include <linux/kvm_host.h>
23#include <linux/slab.h>
23#include <trace/events/kvm.h> 24#include <trace/events/kvm.h>
24 25
25#include <asm/msidef.h> 26#include <asm/msidef.h>
@@ -31,20 +32,39 @@
31 32
32#include "ioapic.h" 33#include "ioapic.h"
33 34
35static inline int kvm_irq_line_state(unsigned long *irq_state,
36 int irq_source_id, int level)
37{
38 /* Logical OR for level trig interrupt */
39 if (level)
40 set_bit(irq_source_id, irq_state);
41 else
42 clear_bit(irq_source_id, irq_state);
43
44 return !!(*irq_state);
45}
46
34static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e, 47static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e,
35 struct kvm *kvm, int level) 48 struct kvm *kvm, int irq_source_id, int level)
36{ 49{
37#ifdef CONFIG_X86 50#ifdef CONFIG_X86
38 return kvm_pic_set_irq(pic_irqchip(kvm), e->irqchip.pin, level); 51 struct kvm_pic *pic = pic_irqchip(kvm);
52 level = kvm_irq_line_state(&pic->irq_states[e->irqchip.pin],
53 irq_source_id, level);
54 return kvm_pic_set_irq(pic, e->irqchip.pin, level);
39#else 55#else
40 return -1; 56 return -1;
41#endif 57#endif
42} 58}
43 59
44static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, 60static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e,
45 struct kvm *kvm, int level) 61 struct kvm *kvm, int irq_source_id, int level)
46{ 62{
47 return kvm_ioapic_set_irq(kvm->arch.vioapic, e->irqchip.pin, level); 63 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
64 level = kvm_irq_line_state(&ioapic->irq_states[e->irqchip.pin],
65 irq_source_id, level);
66
67 return kvm_ioapic_set_irq(ioapic, e->irqchip.pin, level);
48} 68}
49 69
50inline static bool kvm_is_dm_lowest_prio(struct kvm_lapic_irq *irq) 70inline static bool kvm_is_dm_lowest_prio(struct kvm_lapic_irq *irq)
@@ -63,8 +83,6 @@ int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
63 int i, r = -1; 83 int i, r = -1;
64 struct kvm_vcpu *vcpu, *lowest = NULL; 84 struct kvm_vcpu *vcpu, *lowest = NULL;
65 85
66 WARN_ON(!mutex_is_locked(&kvm->irq_lock));
67
68 if (irq->dest_mode == 0 && irq->dest_id == 0xff && 86 if (irq->dest_mode == 0 && irq->dest_id == 0xff &&
69 kvm_is_dm_lowest_prio(irq)) 87 kvm_is_dm_lowest_prio(irq))
70 printk(KERN_INFO "kvm: apic: phys broadcast and lowest prio\n"); 88 printk(KERN_INFO "kvm: apic: phys broadcast and lowest prio\n");
@@ -96,10 +114,13 @@ int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
96} 114}
97 115
98static int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, 116static int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
99 struct kvm *kvm, int level) 117 struct kvm *kvm, int irq_source_id, int level)
100{ 118{
101 struct kvm_lapic_irq irq; 119 struct kvm_lapic_irq irq;
102 120
121 if (!level)
122 return -1;
123
103 trace_kvm_msi_set_irq(e->msi.address_lo, e->msi.data); 124 trace_kvm_msi_set_irq(e->msi.address_lo, e->msi.data);
104 125
105 irq.dest_id = (e->msi.address_lo & 126 irq.dest_id = (e->msi.address_lo &
@@ -116,78 +137,67 @@ static int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
116 return kvm_irq_delivery_to_apic(kvm, NULL, &irq); 137 return kvm_irq_delivery_to_apic(kvm, NULL, &irq);
117} 138}
118 139
119/* This should be called with the kvm->irq_lock mutex held 140/*
120 * Return value: 141 * Return value:
121 * < 0 Interrupt was ignored (masked or not delivered for other reasons) 142 * < 0 Interrupt was ignored (masked or not delivered for other reasons)
122 * = 0 Interrupt was coalesced (previous irq is still pending) 143 * = 0 Interrupt was coalesced (previous irq is still pending)
123 * > 0 Number of CPUs interrupt was delivered to 144 * > 0 Number of CPUs interrupt was delivered to
124 */ 145 */
125int kvm_set_irq(struct kvm *kvm, int irq_source_id, int irq, int level) 146int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level)
126{ 147{
127 struct kvm_kernel_irq_routing_entry *e; 148 struct kvm_kernel_irq_routing_entry *e, irq_set[KVM_NR_IRQCHIPS];
128 unsigned long *irq_state, sig_level; 149 int ret = -1, i = 0;
129 int ret = -1; 150 struct kvm_irq_routing_table *irq_rt;
151 struct hlist_node *n;
130 152
131 trace_kvm_set_irq(irq, level, irq_source_id); 153 trace_kvm_set_irq(irq, level, irq_source_id);
132 154
133 WARN_ON(!mutex_is_locked(&kvm->irq_lock));
134
135 if (irq < KVM_IOAPIC_NUM_PINS) {
136 irq_state = (unsigned long *)&kvm->arch.irq_states[irq];
137
138 /* Logical OR for level trig interrupt */
139 if (level)
140 set_bit(irq_source_id, irq_state);
141 else
142 clear_bit(irq_source_id, irq_state);
143 sig_level = !!(*irq_state);
144 } else if (!level)
145 return ret;
146 else /* Deal with MSI/MSI-X */
147 sig_level = 1;
148
149 /* Not possible to detect if the guest uses the PIC or the 155 /* Not possible to detect if the guest uses the PIC or the
150 * IOAPIC. So set the bit in both. The guest will ignore 156 * IOAPIC. So set the bit in both. The guest will ignore
151 * writes to the unused one. 157 * writes to the unused one.
152 */ 158 */
153 list_for_each_entry(e, &kvm->irq_routing, link) 159 rcu_read_lock();
154 if (e->gsi == irq) { 160 irq_rt = rcu_dereference(kvm->irq_routing);
155 int r = e->set(e, kvm, sig_level); 161 if (irq < irq_rt->nr_rt_entries)
156 if (r < 0) 162 hlist_for_each_entry(e, n, &irq_rt->map[irq], link)
157 continue; 163 irq_set[i++] = *e;
164 rcu_read_unlock();
165
166 while(i--) {
167 int r;
168 r = irq_set[i].set(&irq_set[i], kvm, irq_source_id, level);
169 if (r < 0)
170 continue;
171
172 ret = r + ((ret < 0) ? 0 : ret);
173 }
158 174
159 ret = r + ((ret < 0) ? 0 : ret);
160 }
161 return ret; 175 return ret;
162} 176}
163 177
164void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin) 178void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin)
165{ 179{
166 struct kvm_kernel_irq_routing_entry *e;
167 struct kvm_irq_ack_notifier *kian; 180 struct kvm_irq_ack_notifier *kian;
168 struct hlist_node *n; 181 struct hlist_node *n;
169 unsigned gsi = pin; 182 int gsi;
170 183
171 trace_kvm_ack_irq(irqchip, pin); 184 trace_kvm_ack_irq(irqchip, pin);
172 185
173 list_for_each_entry(e, &kvm->irq_routing, link) 186 rcu_read_lock();
174 if (e->type == KVM_IRQ_ROUTING_IRQCHIP && 187 gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
175 e->irqchip.irqchip == irqchip && 188 if (gsi != -1)
176 e->irqchip.pin == pin) { 189 hlist_for_each_entry_rcu(kian, n, &kvm->irq_ack_notifier_list,
177 gsi = e->gsi; 190 link)
178 break; 191 if (kian->gsi == gsi)
179 } 192 kian->irq_acked(kian);
180 193 rcu_read_unlock();
181 hlist_for_each_entry(kian, n, &kvm->arch.irq_ack_notifier_list, link)
182 if (kian->gsi == gsi)
183 kian->irq_acked(kian);
184} 194}
185 195
186void kvm_register_irq_ack_notifier(struct kvm *kvm, 196void kvm_register_irq_ack_notifier(struct kvm *kvm,
187 struct kvm_irq_ack_notifier *kian) 197 struct kvm_irq_ack_notifier *kian)
188{ 198{
189 mutex_lock(&kvm->irq_lock); 199 mutex_lock(&kvm->irq_lock);
190 hlist_add_head(&kian->link, &kvm->arch.irq_ack_notifier_list); 200 hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list);
191 mutex_unlock(&kvm->irq_lock); 201 mutex_unlock(&kvm->irq_lock);
192} 202}
193 203
@@ -195,8 +205,9 @@ void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
195 struct kvm_irq_ack_notifier *kian) 205 struct kvm_irq_ack_notifier *kian)
196{ 206{
197 mutex_lock(&kvm->irq_lock); 207 mutex_lock(&kvm->irq_lock);
198 hlist_del_init(&kian->link); 208 hlist_del_init_rcu(&kian->link);
199 mutex_unlock(&kvm->irq_lock); 209 mutex_unlock(&kvm->irq_lock);
210 synchronize_rcu();
200} 211}
201 212
202int kvm_request_irq_source_id(struct kvm *kvm) 213int kvm_request_irq_source_id(struct kvm *kvm)
@@ -205,16 +216,17 @@ int kvm_request_irq_source_id(struct kvm *kvm)
205 int irq_source_id; 216 int irq_source_id;
206 217
207 mutex_lock(&kvm->irq_lock); 218 mutex_lock(&kvm->irq_lock);
208 irq_source_id = find_first_zero_bit(bitmap, 219 irq_source_id = find_first_zero_bit(bitmap, BITS_PER_LONG);
209 sizeof(kvm->arch.irq_sources_bitmap));
210 220
211 if (irq_source_id >= sizeof(kvm->arch.irq_sources_bitmap)) { 221 if (irq_source_id >= BITS_PER_LONG) {
212 printk(KERN_WARNING "kvm: exhaust allocatable IRQ sources!\n"); 222 printk(KERN_WARNING "kvm: exhaust allocatable IRQ sources!\n");
213 return -EFAULT; 223 irq_source_id = -EFAULT;
224 goto unlock;
214 } 225 }
215 226
216 ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID); 227 ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
217 set_bit(irq_source_id, bitmap); 228 set_bit(irq_source_id, bitmap);
229unlock:
218 mutex_unlock(&kvm->irq_lock); 230 mutex_unlock(&kvm->irq_lock);
219 231
220 return irq_source_id; 232 return irq_source_id;
@@ -228,13 +240,23 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)
228 240
229 mutex_lock(&kvm->irq_lock); 241 mutex_lock(&kvm->irq_lock);
230 if (irq_source_id < 0 || 242 if (irq_source_id < 0 ||
231 irq_source_id >= sizeof(kvm->arch.irq_sources_bitmap)) { 243 irq_source_id >= BITS_PER_LONG) {
232 printk(KERN_ERR "kvm: IRQ source ID out of range!\n"); 244 printk(KERN_ERR "kvm: IRQ source ID out of range!\n");
233 return; 245 goto unlock;
234 } 246 }
235 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
236 clear_bit(irq_source_id, &kvm->arch.irq_states[i]);
237 clear_bit(irq_source_id, &kvm->arch.irq_sources_bitmap); 247 clear_bit(irq_source_id, &kvm->arch.irq_sources_bitmap);
248 if (!irqchip_in_kernel(kvm))
249 goto unlock;
250
251 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++) {
252 clear_bit(irq_source_id, &kvm->arch.vioapic->irq_states[i]);
253 if (i >= 16)
254 continue;
255#ifdef CONFIG_X86
256 clear_bit(irq_source_id, &pic_irqchip(kvm)->irq_states[i]);
257#endif
258 }
259unlock:
238 mutex_unlock(&kvm->irq_lock); 260 mutex_unlock(&kvm->irq_lock);
239} 261}
240 262
@@ -243,7 +265,7 @@ void kvm_register_irq_mask_notifier(struct kvm *kvm, int irq,
243{ 265{
244 mutex_lock(&kvm->irq_lock); 266 mutex_lock(&kvm->irq_lock);
245 kimn->irq = irq; 267 kimn->irq = irq;
246 hlist_add_head(&kimn->link, &kvm->mask_notifier_list); 268 hlist_add_head_rcu(&kimn->link, &kvm->mask_notifier_list);
247 mutex_unlock(&kvm->irq_lock); 269 mutex_unlock(&kvm->irq_lock);
248} 270}
249 271
@@ -251,8 +273,9 @@ void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq,
251 struct kvm_irq_mask_notifier *kimn) 273 struct kvm_irq_mask_notifier *kimn)
252{ 274{
253 mutex_lock(&kvm->irq_lock); 275 mutex_lock(&kvm->irq_lock);
254 hlist_del(&kimn->link); 276 hlist_del_rcu(&kimn->link);
255 mutex_unlock(&kvm->irq_lock); 277 mutex_unlock(&kvm->irq_lock);
278 synchronize_rcu();
256} 279}
257 280
258void kvm_fire_mask_notifiers(struct kvm *kvm, int irq, bool mask) 281void kvm_fire_mask_notifiers(struct kvm *kvm, int irq, bool mask)
@@ -260,33 +283,38 @@ void kvm_fire_mask_notifiers(struct kvm *kvm, int irq, bool mask)
260 struct kvm_irq_mask_notifier *kimn; 283 struct kvm_irq_mask_notifier *kimn;
261 struct hlist_node *n; 284 struct hlist_node *n;
262 285
263 WARN_ON(!mutex_is_locked(&kvm->irq_lock)); 286 rcu_read_lock();
264 287 hlist_for_each_entry_rcu(kimn, n, &kvm->mask_notifier_list, link)
265 hlist_for_each_entry(kimn, n, &kvm->mask_notifier_list, link)
266 if (kimn->irq == irq) 288 if (kimn->irq == irq)
267 kimn->func(kimn, mask); 289 kimn->func(kimn, mask);
268} 290 rcu_read_unlock();
269
270static void __kvm_free_irq_routing(struct list_head *irq_routing)
271{
272 struct kvm_kernel_irq_routing_entry *e, *n;
273
274 list_for_each_entry_safe(e, n, irq_routing, link)
275 kfree(e);
276} 291}
277 292
278void kvm_free_irq_routing(struct kvm *kvm) 293void kvm_free_irq_routing(struct kvm *kvm)
279{ 294{
280 mutex_lock(&kvm->irq_lock); 295 /* Called only during vm destruction. Nobody can use the pointer
281 __kvm_free_irq_routing(&kvm->irq_routing); 296 at this stage */
282 mutex_unlock(&kvm->irq_lock); 297 kfree(kvm->irq_routing);
283} 298}
284 299
285static int setup_routing_entry(struct kvm_kernel_irq_routing_entry *e, 300static int setup_routing_entry(struct kvm_irq_routing_table *rt,
301 struct kvm_kernel_irq_routing_entry *e,
286 const struct kvm_irq_routing_entry *ue) 302 const struct kvm_irq_routing_entry *ue)
287{ 303{
288 int r = -EINVAL; 304 int r = -EINVAL;
289 int delta; 305 int delta;
306 unsigned max_pin;
307 struct kvm_kernel_irq_routing_entry *ei;
308 struct hlist_node *n;
309
310 /*
311 * Do not allow GSI to be mapped to the same irqchip more than once.
312 * Allow only one to one mapping between GSI and MSI.
313 */
314 hlist_for_each_entry(ei, n, &rt->map[ue->gsi], link)
315 if (ei->type == KVM_IRQ_ROUTING_MSI ||
316 ue->u.irqchip.irqchip == ei->irqchip.irqchip)
317 return r;
290 318
291 e->gsi = ue->gsi; 319 e->gsi = ue->gsi;
292 e->type = ue->type; 320 e->type = ue->type;
@@ -296,12 +324,15 @@ static int setup_routing_entry(struct kvm_kernel_irq_routing_entry *e,
296 switch (ue->u.irqchip.irqchip) { 324 switch (ue->u.irqchip.irqchip) {
297 case KVM_IRQCHIP_PIC_MASTER: 325 case KVM_IRQCHIP_PIC_MASTER:
298 e->set = kvm_set_pic_irq; 326 e->set = kvm_set_pic_irq;
327 max_pin = 16;
299 break; 328 break;
300 case KVM_IRQCHIP_PIC_SLAVE: 329 case KVM_IRQCHIP_PIC_SLAVE:
301 e->set = kvm_set_pic_irq; 330 e->set = kvm_set_pic_irq;
331 max_pin = 16;
302 delta = 8; 332 delta = 8;
303 break; 333 break;
304 case KVM_IRQCHIP_IOAPIC: 334 case KVM_IRQCHIP_IOAPIC:
335 max_pin = KVM_IOAPIC_NUM_PINS;
305 e->set = kvm_set_ioapic_irq; 336 e->set = kvm_set_ioapic_irq;
306 break; 337 break;
307 default: 338 default:
@@ -309,6 +340,9 @@ static int setup_routing_entry(struct kvm_kernel_irq_routing_entry *e,
309 } 340 }
310 e->irqchip.irqchip = ue->u.irqchip.irqchip; 341 e->irqchip.irqchip = ue->u.irqchip.irqchip;
311 e->irqchip.pin = ue->u.irqchip.pin + delta; 342 e->irqchip.pin = ue->u.irqchip.pin + delta;
343 if (e->irqchip.pin >= max_pin)
344 goto out;
345 rt->chip[ue->u.irqchip.irqchip][e->irqchip.pin] = ue->gsi;
312 break; 346 break;
313 case KVM_IRQ_ROUTING_MSI: 347 case KVM_IRQ_ROUTING_MSI:
314 e->set = kvm_set_msi; 348 e->set = kvm_set_msi;
@@ -319,6 +353,8 @@ static int setup_routing_entry(struct kvm_kernel_irq_routing_entry *e,
319 default: 353 default:
320 goto out; 354 goto out;
321 } 355 }
356
357 hlist_add_head(&e->link, &rt->map[e->gsi]);
322 r = 0; 358 r = 0;
323out: 359out:
324 return r; 360 return r;
@@ -330,43 +366,53 @@ int kvm_set_irq_routing(struct kvm *kvm,
330 unsigned nr, 366 unsigned nr,
331 unsigned flags) 367 unsigned flags)
332{ 368{
333 struct list_head irq_list = LIST_HEAD_INIT(irq_list); 369 struct kvm_irq_routing_table *new, *old;
334 struct list_head tmp = LIST_HEAD_INIT(tmp); 370 u32 i, j, nr_rt_entries = 0;
335 struct kvm_kernel_irq_routing_entry *e = NULL;
336 unsigned i;
337 int r; 371 int r;
338 372
339 for (i = 0; i < nr; ++i) { 373 for (i = 0; i < nr; ++i) {
374 if (ue[i].gsi >= KVM_MAX_IRQ_ROUTES)
375 return -EINVAL;
376 nr_rt_entries = max(nr_rt_entries, ue[i].gsi);
377 }
378
379 nr_rt_entries += 1;
380
381 new = kzalloc(sizeof(*new) + (nr_rt_entries * sizeof(struct hlist_head))
382 + (nr * sizeof(struct kvm_kernel_irq_routing_entry)),
383 GFP_KERNEL);
384
385 if (!new)
386 return -ENOMEM;
387
388 new->rt_entries = (void *)&new->map[nr_rt_entries];
389
390 new->nr_rt_entries = nr_rt_entries;
391 for (i = 0; i < 3; i++)
392 for (j = 0; j < KVM_IOAPIC_NUM_PINS; j++)
393 new->chip[i][j] = -1;
394
395 for (i = 0; i < nr; ++i) {
340 r = -EINVAL; 396 r = -EINVAL;
341 if (ue->gsi >= KVM_MAX_IRQ_ROUTES)
342 goto out;
343 if (ue->flags) 397 if (ue->flags)
344 goto out; 398 goto out;
345 r = -ENOMEM; 399 r = setup_routing_entry(new, &new->rt_entries[i], ue);
346 e = kzalloc(sizeof(*e), GFP_KERNEL);
347 if (!e)
348 goto out;
349 r = setup_routing_entry(e, ue);
350 if (r) 400 if (r)
351 goto out; 401 goto out;
352 ++ue; 402 ++ue;
353 list_add(&e->link, &irq_list);
354 e = NULL;
355 } 403 }
356 404
357 mutex_lock(&kvm->irq_lock); 405 mutex_lock(&kvm->irq_lock);
358 list_splice(&kvm->irq_routing, &tmp); 406 old = kvm->irq_routing;
359 INIT_LIST_HEAD(&kvm->irq_routing); 407 rcu_assign_pointer(kvm->irq_routing, new);
360 list_splice(&irq_list, &kvm->irq_routing);
361 INIT_LIST_HEAD(&irq_list);
362 list_splice(&tmp, &irq_list);
363 mutex_unlock(&kvm->irq_lock); 408 mutex_unlock(&kvm->irq_lock);
409 synchronize_rcu();
364 410
411 new = old;
365 r = 0; 412 r = 0;
366 413
367out: 414out:
368 kfree(e); 415 kfree(new);
369 __kvm_free_irq_routing(&irq_list);
370 return r; 416 return r;
371} 417}
372 418
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 7495ce347344..c82ae2492634 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -22,7 +22,6 @@
22#include <linux/module.h> 22#include <linux/module.h>
23#include <linux/errno.h> 23#include <linux/errno.h>
24#include <linux/percpu.h> 24#include <linux/percpu.h>
25#include <linux/gfp.h>
26#include <linux/mm.h> 25#include <linux/mm.h>
27#include <linux/miscdevice.h> 26#include <linux/miscdevice.h>
28#include <linux/vmalloc.h> 27#include <linux/vmalloc.h>
@@ -43,21 +42,18 @@
43#include <linux/swap.h> 42#include <linux/swap.h>
44#include <linux/bitops.h> 43#include <linux/bitops.h>
45#include <linux/spinlock.h> 44#include <linux/spinlock.h>
45#include <linux/compat.h>
46#include <linux/srcu.h>
47#include <linux/hugetlb.h>
48#include <linux/slab.h>
46 49
47#include <asm/processor.h> 50#include <asm/processor.h>
48#include <asm/io.h> 51#include <asm/io.h>
49#include <asm/uaccess.h> 52#include <asm/uaccess.h>
50#include <asm/pgtable.h> 53#include <asm/pgtable.h>
54#include <asm-generic/bitops/le.h>
51 55
52#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
53#include "coalesced_mmio.h" 56#include "coalesced_mmio.h"
54#endif
55
56#ifdef KVM_CAP_DEVICE_ASSIGNMENT
57#include <linux/pci.h>
58#include <linux/interrupt.h>
59#include "irq.h"
60#endif
61 57
62#define CREATE_TRACE_POINTS 58#define CREATE_TRACE_POINTS
63#include <trace/events/kvm.h> 59#include <trace/events/kvm.h>
@@ -68,13 +64,15 @@ MODULE_LICENSE("GPL");
68/* 64/*
69 * Ordering of locks: 65 * Ordering of locks:
70 * 66 *
71 * kvm->slots_lock --> kvm->lock --> kvm->irq_lock 67 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
72 */ 68 */
73 69
74DEFINE_SPINLOCK(kvm_lock); 70DEFINE_SPINLOCK(kvm_lock);
75LIST_HEAD(vm_list); 71LIST_HEAD(vm_list);
76 72
77static cpumask_var_t cpus_hardware_enabled; 73static cpumask_var_t cpus_hardware_enabled;
74static int kvm_usage_count = 0;
75static atomic_t hardware_enable_failed;
78 76
79struct kmem_cache *kvm_vcpu_cache; 77struct kmem_cache *kvm_vcpu_cache;
80EXPORT_SYMBOL_GPL(kvm_vcpu_cache); 78EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
@@ -85,615 +83,15 @@ struct dentry *kvm_debugfs_dir;
85 83
86static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, 84static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 unsigned long arg); 85 unsigned long arg);
86static int hardware_enable_all(void);
87static void hardware_disable_all(void);
88
89static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
88 90
89static bool kvm_rebooting; 91static bool kvm_rebooting;
90 92
91static bool largepages_enabled = true; 93static bool largepages_enabled = true;
92 94
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) 95inline int kvm_is_mmio_pfn(pfn_t pfn)
698{ 96{
699 if (pfn_valid(pfn)) { 97 if (pfn_valid(pfn)) {
@@ -740,7 +138,7 @@ static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
740 138
741 zalloc_cpumask_var(&cpus, GFP_ATOMIC); 139 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
742 140
743 spin_lock(&kvm->requests_lock); 141 raw_spin_lock(&kvm->requests_lock);
744 me = smp_processor_id(); 142 me = smp_processor_id();
745 kvm_for_each_vcpu(i, vcpu, kvm) { 143 kvm_for_each_vcpu(i, vcpu, kvm) {
746 if (test_and_set_bit(req, &vcpu->requests)) 144 if (test_and_set_bit(req, &vcpu->requests))
@@ -755,7 +153,7 @@ static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
755 smp_call_function_many(cpus, ack_flush, NULL, 1); 153 smp_call_function_many(cpus, ack_flush, NULL, 1);
756 else 154 else
757 called = false; 155 called = false;
758 spin_unlock(&kvm->requests_lock); 156 raw_spin_unlock(&kvm->requests_lock);
759 free_cpumask_var(cpus); 157 free_cpumask_var(cpus);
760 return called; 158 return called;
761} 159}
@@ -819,7 +217,7 @@ static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
819 unsigned long address) 217 unsigned long address)
820{ 218{
821 struct kvm *kvm = mmu_notifier_to_kvm(mn); 219 struct kvm *kvm = mmu_notifier_to_kvm(mn);
822 int need_tlb_flush; 220 int need_tlb_flush, idx;
823 221
824 /* 222 /*
825 * When ->invalidate_page runs, the linux pte has been zapped 223 * When ->invalidate_page runs, the linux pte has been zapped
@@ -839,10 +237,12 @@ static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
839 * pte after kvm_unmap_hva returned, without noticing the page 237 * pte after kvm_unmap_hva returned, without noticing the page
840 * is going to be freed. 238 * is going to be freed.
841 */ 239 */
240 idx = srcu_read_lock(&kvm->srcu);
842 spin_lock(&kvm->mmu_lock); 241 spin_lock(&kvm->mmu_lock);
843 kvm->mmu_notifier_seq++; 242 kvm->mmu_notifier_seq++;
844 need_tlb_flush = kvm_unmap_hva(kvm, address); 243 need_tlb_flush = kvm_unmap_hva(kvm, address);
845 spin_unlock(&kvm->mmu_lock); 244 spin_unlock(&kvm->mmu_lock);
245 srcu_read_unlock(&kvm->srcu, idx);
846 246
847 /* we've to flush the tlb before the pages can be freed */ 247 /* we've to flush the tlb before the pages can be freed */
848 if (need_tlb_flush) 248 if (need_tlb_flush)
@@ -856,11 +256,14 @@ static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
856 pte_t pte) 256 pte_t pte)
857{ 257{
858 struct kvm *kvm = mmu_notifier_to_kvm(mn); 258 struct kvm *kvm = mmu_notifier_to_kvm(mn);
259 int idx;
859 260
261 idx = srcu_read_lock(&kvm->srcu);
860 spin_lock(&kvm->mmu_lock); 262 spin_lock(&kvm->mmu_lock);
861 kvm->mmu_notifier_seq++; 263 kvm->mmu_notifier_seq++;
862 kvm_set_spte_hva(kvm, address, pte); 264 kvm_set_spte_hva(kvm, address, pte);
863 spin_unlock(&kvm->mmu_lock); 265 spin_unlock(&kvm->mmu_lock);
266 srcu_read_unlock(&kvm->srcu, idx);
864} 267}
865 268
866static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, 269static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
@@ -869,8 +272,9 @@ static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
869 unsigned long end) 272 unsigned long end)
870{ 273{
871 struct kvm *kvm = mmu_notifier_to_kvm(mn); 274 struct kvm *kvm = mmu_notifier_to_kvm(mn);
872 int need_tlb_flush = 0; 275 int need_tlb_flush = 0, idx;
873 276
277 idx = srcu_read_lock(&kvm->srcu);
874 spin_lock(&kvm->mmu_lock); 278 spin_lock(&kvm->mmu_lock);
875 /* 279 /*
876 * The count increase must become visible at unlock time as no 280 * The count increase must become visible at unlock time as no
@@ -881,6 +285,7 @@ static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
881 for (; start < end; start += PAGE_SIZE) 285 for (; start < end; start += PAGE_SIZE)
882 need_tlb_flush |= kvm_unmap_hva(kvm, start); 286 need_tlb_flush |= kvm_unmap_hva(kvm, start);
883 spin_unlock(&kvm->mmu_lock); 287 spin_unlock(&kvm->mmu_lock);
288 srcu_read_unlock(&kvm->srcu, idx);
884 289
885 /* we've to flush the tlb before the pages can be freed */ 290 /* we've to flush the tlb before the pages can be freed */
886 if (need_tlb_flush) 291 if (need_tlb_flush)
@@ -918,11 +323,13 @@ static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
918 unsigned long address) 323 unsigned long address)
919{ 324{
920 struct kvm *kvm = mmu_notifier_to_kvm(mn); 325 struct kvm *kvm = mmu_notifier_to_kvm(mn);
921 int young; 326 int young, idx;
922 327
328 idx = srcu_read_lock(&kvm->srcu);
923 spin_lock(&kvm->mmu_lock); 329 spin_lock(&kvm->mmu_lock);
924 young = kvm_age_hva(kvm, address); 330 young = kvm_age_hva(kvm, address);
925 spin_unlock(&kvm->mmu_lock); 331 spin_unlock(&kvm->mmu_lock);
332 srcu_read_unlock(&kvm->srcu, idx);
926 333
927 if (young) 334 if (young)
928 kvm_flush_remote_tlbs(kvm); 335 kvm_flush_remote_tlbs(kvm);
@@ -934,7 +341,11 @@ static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
934 struct mm_struct *mm) 341 struct mm_struct *mm)
935{ 342{
936 struct kvm *kvm = mmu_notifier_to_kvm(mn); 343 struct kvm *kvm = mmu_notifier_to_kvm(mn);
344 int idx;
345
346 idx = srcu_read_lock(&kvm->srcu);
937 kvm_arch_flush_shadow(kvm); 347 kvm_arch_flush_shadow(kvm);
348 srcu_read_unlock(&kvm->srcu, idx);
938} 349}
939 350
940static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { 351static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
@@ -945,57 +356,68 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
945 .change_pte = kvm_mmu_notifier_change_pte, 356 .change_pte = kvm_mmu_notifier_change_pte,
946 .release = kvm_mmu_notifier_release, 357 .release = kvm_mmu_notifier_release,
947}; 358};
359
360static int kvm_init_mmu_notifier(struct kvm *kvm)
361{
362 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
363 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
364}
365
366#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
367
368static int kvm_init_mmu_notifier(struct kvm *kvm)
369{
370 return 0;
371}
372
948#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ 373#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
949 374
950static struct kvm *kvm_create_vm(void) 375static struct kvm *kvm_create_vm(void)
951{ 376{
377 int r = 0, i;
952 struct kvm *kvm = kvm_arch_create_vm(); 378 struct kvm *kvm = kvm_arch_create_vm();
953#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
954 struct page *page;
955#endif
956 379
957 if (IS_ERR(kvm)) 380 if (IS_ERR(kvm))
958 goto out; 381 goto out;
382
383 r = hardware_enable_all();
384 if (r)
385 goto out_err_nodisable;
386
959#ifdef CONFIG_HAVE_KVM_IRQCHIP 387#ifdef CONFIG_HAVE_KVM_IRQCHIP
960 INIT_LIST_HEAD(&kvm->irq_routing);
961 INIT_HLIST_HEAD(&kvm->mask_notifier_list); 388 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
389 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
962#endif 390#endif
963 391
964#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 392 r = -ENOMEM;
965 page = alloc_page(GFP_KERNEL | __GFP_ZERO); 393 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
966 if (!page) { 394 if (!kvm->memslots)
967 kfree(kvm); 395 goto out_err;
968 return ERR_PTR(-ENOMEM); 396 if (init_srcu_struct(&kvm->srcu))
397 goto out_err;
398 for (i = 0; i < KVM_NR_BUSES; i++) {
399 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
400 GFP_KERNEL);
401 if (!kvm->buses[i]) {
402 cleanup_srcu_struct(&kvm->srcu);
403 goto out_err;
404 }
969 } 405 }
970 kvm->coalesced_mmio_ring =
971 (struct kvm_coalesced_mmio_ring *)page_address(page);
972#endif
973 406
974#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 407 r = kvm_init_mmu_notifier(kvm);
975 { 408 if (r) {
976 int err; 409 cleanup_srcu_struct(&kvm->srcu);
977 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; 410 goto out_err;
978 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
979 if (err) {
980#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
981 put_page(page);
982#endif
983 kfree(kvm);
984 return ERR_PTR(err);
985 }
986 } 411 }
987#endif
988 412
989 kvm->mm = current->mm; 413 kvm->mm = current->mm;
990 atomic_inc(&kvm->mm->mm_count); 414 atomic_inc(&kvm->mm->mm_count);
991 spin_lock_init(&kvm->mmu_lock); 415 spin_lock_init(&kvm->mmu_lock);
992 spin_lock_init(&kvm->requests_lock); 416 raw_spin_lock_init(&kvm->requests_lock);
993 kvm_io_bus_init(&kvm->pio_bus);
994 kvm_eventfd_init(kvm); 417 kvm_eventfd_init(kvm);
995 mutex_init(&kvm->lock); 418 mutex_init(&kvm->lock);
996 mutex_init(&kvm->irq_lock); 419 mutex_init(&kvm->irq_lock);
997 kvm_io_bus_init(&kvm->mmio_bus); 420 mutex_init(&kvm->slots_lock);
998 init_rwsem(&kvm->slots_lock);
999 atomic_set(&kvm->users_count, 1); 421 atomic_set(&kvm->users_count, 1);
1000 spin_lock(&kvm_lock); 422 spin_lock(&kvm_lock);
1001 list_add(&kvm->vm_list, &vm_list); 423 list_add(&kvm->vm_list, &vm_list);
@@ -1005,6 +427,15 @@ static struct kvm *kvm_create_vm(void)
1005#endif 427#endif
1006out: 428out:
1007 return kvm; 429 return kvm;
430
431out_err:
432 hardware_disable_all();
433out_err_nodisable:
434 for (i = 0; i < KVM_NR_BUSES; i++)
435 kfree(kvm->buses[i]);
436 kfree(kvm->memslots);
437 kfree(kvm);
438 return ERR_PTR(r);
1008} 439}
1009 440
1010/* 441/*
@@ -1037,13 +468,17 @@ static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1037void kvm_free_physmem(struct kvm *kvm) 468void kvm_free_physmem(struct kvm *kvm)
1038{ 469{
1039 int i; 470 int i;
471 struct kvm_memslots *slots = kvm->memslots;
472
473 for (i = 0; i < slots->nmemslots; ++i)
474 kvm_free_physmem_slot(&slots->memslots[i], NULL);
1040 475
1041 for (i = 0; i < kvm->nmemslots; ++i) 476 kfree(kvm->memslots);
1042 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1043} 477}
1044 478
1045static void kvm_destroy_vm(struct kvm *kvm) 479static void kvm_destroy_vm(struct kvm *kvm)
1046{ 480{
481 int i;
1047 struct mm_struct *mm = kvm->mm; 482 struct mm_struct *mm = kvm->mm;
1048 483
1049 kvm_arch_sync_events(kvm); 484 kvm_arch_sync_events(kvm);
@@ -1051,18 +486,16 @@ static void kvm_destroy_vm(struct kvm *kvm)
1051 list_del(&kvm->vm_list); 486 list_del(&kvm->vm_list);
1052 spin_unlock(&kvm_lock); 487 spin_unlock(&kvm_lock);
1053 kvm_free_irq_routing(kvm); 488 kvm_free_irq_routing(kvm);
1054 kvm_io_bus_destroy(&kvm->pio_bus); 489 for (i = 0; i < KVM_NR_BUSES; i++)
1055 kvm_io_bus_destroy(&kvm->mmio_bus); 490 kvm_io_bus_destroy(kvm->buses[i]);
1056#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 491 kvm_coalesced_mmio_free(kvm);
1057 if (kvm->coalesced_mmio_ring != NULL)
1058 free_page((unsigned long)kvm->coalesced_mmio_ring);
1059#endif
1060#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 492#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1061 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); 493 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1062#else 494#else
1063 kvm_arch_flush_shadow(kvm); 495 kvm_arch_flush_shadow(kvm);
1064#endif 496#endif
1065 kvm_arch_destroy_vm(kvm); 497 kvm_arch_destroy_vm(kvm);
498 hardware_disable_all();
1066 mmdrop(mm); 499 mmdrop(mm);
1067} 500}
1068 501
@@ -1102,12 +535,13 @@ int __kvm_set_memory_region(struct kvm *kvm,
1102 struct kvm_userspace_memory_region *mem, 535 struct kvm_userspace_memory_region *mem,
1103 int user_alloc) 536 int user_alloc)
1104{ 537{
1105 int r; 538 int r, flush_shadow = 0;
1106 gfn_t base_gfn; 539 gfn_t base_gfn;
1107 unsigned long npages; 540 unsigned long npages;
1108 unsigned long i; 541 unsigned long i;
1109 struct kvm_memory_slot *memslot; 542 struct kvm_memory_slot *memslot;
1110 struct kvm_memory_slot old, new; 543 struct kvm_memory_slot old, new;
544 struct kvm_memslots *slots, *old_memslots;
1111 545
1112 r = -EINVAL; 546 r = -EINVAL;
1113 /* General sanity checks */ 547 /* General sanity checks */
@@ -1122,7 +556,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
1122 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) 556 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1123 goto out; 557 goto out;
1124 558
1125 memslot = &kvm->memslots[mem->slot]; 559 memslot = &kvm->memslots->memslots[mem->slot];
1126 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; 560 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1127 npages = mem->memory_size >> PAGE_SHIFT; 561 npages = mem->memory_size >> PAGE_SHIFT;
1128 562
@@ -1143,7 +577,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
1143 /* Check for overlaps */ 577 /* Check for overlaps */
1144 r = -EEXIST; 578 r = -EEXIST;
1145 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { 579 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1146 struct kvm_memory_slot *s = &kvm->memslots[i]; 580 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
1147 581
1148 if (s == memslot || !s->npages) 582 if (s == memslot || !s->npages)
1149 continue; 583 continue;
@@ -1169,15 +603,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
1169 memset(new.rmap, 0, npages * sizeof(*new.rmap)); 603 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1170 604
1171 new.user_alloc = user_alloc; 605 new.user_alloc = user_alloc;
1172 /* 606 new.userspace_addr = mem->userspace_addr;
1173 * hva_to_rmmap() serialzies with the mmu_lock and to be
1174 * safe it has to ignore memslots with !user_alloc &&
1175 * !userspace_addr.
1176 */
1177 if (user_alloc)
1178 new.userspace_addr = mem->userspace_addr;
1179 else
1180 new.userspace_addr = 0;
1181 } 607 }
1182 if (!npages) 608 if (!npages)
1183 goto skip_lpage; 609 goto skip_lpage;
@@ -1226,14 +652,15 @@ skip_lpage:
1226 652
1227 /* Allocate page dirty bitmap if needed */ 653 /* Allocate page dirty bitmap if needed */
1228 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { 654 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1229 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8; 655 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
1230 656
1231 new.dirty_bitmap = vmalloc(dirty_bytes); 657 new.dirty_bitmap = vmalloc(dirty_bytes);
1232 if (!new.dirty_bitmap) 658 if (!new.dirty_bitmap)
1233 goto out_free; 659 goto out_free;
1234 memset(new.dirty_bitmap, 0, dirty_bytes); 660 memset(new.dirty_bitmap, 0, dirty_bytes);
661 /* destroy any largepage mappings for dirty tracking */
1235 if (old.npages) 662 if (old.npages)
1236 kvm_arch_flush_shadow(kvm); 663 flush_shadow = 1;
1237 } 664 }
1238#else /* not defined CONFIG_S390 */ 665#else /* not defined CONFIG_S390 */
1239 new.user_alloc = user_alloc; 666 new.user_alloc = user_alloc;
@@ -1241,36 +668,72 @@ skip_lpage:
1241 new.userspace_addr = mem->userspace_addr; 668 new.userspace_addr = mem->userspace_addr;
1242#endif /* not defined CONFIG_S390 */ 669#endif /* not defined CONFIG_S390 */
1243 670
1244 if (!npages) 671 if (!npages) {
672 r = -ENOMEM;
673 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
674 if (!slots)
675 goto out_free;
676 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
677 if (mem->slot >= slots->nmemslots)
678 slots->nmemslots = mem->slot + 1;
679 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
680
681 old_memslots = kvm->memslots;
682 rcu_assign_pointer(kvm->memslots, slots);
683 synchronize_srcu_expedited(&kvm->srcu);
684 /* From this point no new shadow pages pointing to a deleted
685 * memslot will be created.
686 *
687 * validation of sp->gfn happens in:
688 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
689 * - kvm_is_visible_gfn (mmu_check_roots)
690 */
1245 kvm_arch_flush_shadow(kvm); 691 kvm_arch_flush_shadow(kvm);
692 kfree(old_memslots);
693 }
1246 694
1247 spin_lock(&kvm->mmu_lock); 695 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
1248 if (mem->slot >= kvm->nmemslots) 696 if (r)
1249 kvm->nmemslots = mem->slot + 1;
1250
1251 *memslot = new;
1252 spin_unlock(&kvm->mmu_lock);
1253
1254 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1255 if (r) {
1256 spin_lock(&kvm->mmu_lock);
1257 *memslot = old;
1258 spin_unlock(&kvm->mmu_lock);
1259 goto out_free; 697 goto out_free;
1260 }
1261 698
1262 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1263 /* Slot deletion case: we have to update the current slot */
1264 spin_lock(&kvm->mmu_lock);
1265 if (!npages)
1266 *memslot = old;
1267 spin_unlock(&kvm->mmu_lock);
1268#ifdef CONFIG_DMAR 699#ifdef CONFIG_DMAR
1269 /* map the pages in iommu page table */ 700 /* map the pages in iommu page table */
1270 r = kvm_iommu_map_pages(kvm, base_gfn, npages); 701 if (npages) {
1271 if (r) 702 r = kvm_iommu_map_pages(kvm, &new);
1272 goto out; 703 if (r)
704 goto out_free;
705 }
1273#endif 706#endif
707
708 r = -ENOMEM;
709 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
710 if (!slots)
711 goto out_free;
712 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
713 if (mem->slot >= slots->nmemslots)
714 slots->nmemslots = mem->slot + 1;
715
716 /* actual memory is freed via old in kvm_free_physmem_slot below */
717 if (!npages) {
718 new.rmap = NULL;
719 new.dirty_bitmap = NULL;
720 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
721 new.lpage_info[i] = NULL;
722 }
723
724 slots->memslots[mem->slot] = new;
725 old_memslots = kvm->memslots;
726 rcu_assign_pointer(kvm->memslots, slots);
727 synchronize_srcu_expedited(&kvm->srcu);
728
729 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
730
731 kvm_free_physmem_slot(&old, &new);
732 kfree(old_memslots);
733
734 if (flush_shadow)
735 kvm_arch_flush_shadow(kvm);
736
1274 return 0; 737 return 0;
1275 738
1276out_free: 739out_free:
@@ -1287,9 +750,9 @@ int kvm_set_memory_region(struct kvm *kvm,
1287{ 750{
1288 int r; 751 int r;
1289 752
1290 down_write(&kvm->slots_lock); 753 mutex_lock(&kvm->slots_lock);
1291 r = __kvm_set_memory_region(kvm, mem, user_alloc); 754 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1292 up_write(&kvm->slots_lock); 755 mutex_unlock(&kvm->slots_lock);
1293 return r; 756 return r;
1294} 757}
1295EXPORT_SYMBOL_GPL(kvm_set_memory_region); 758EXPORT_SYMBOL_GPL(kvm_set_memory_region);
@@ -1309,19 +772,19 @@ int kvm_get_dirty_log(struct kvm *kvm,
1309{ 772{
1310 struct kvm_memory_slot *memslot; 773 struct kvm_memory_slot *memslot;
1311 int r, i; 774 int r, i;
1312 int n; 775 unsigned long n;
1313 unsigned long any = 0; 776 unsigned long any = 0;
1314 777
1315 r = -EINVAL; 778 r = -EINVAL;
1316 if (log->slot >= KVM_MEMORY_SLOTS) 779 if (log->slot >= KVM_MEMORY_SLOTS)
1317 goto out; 780 goto out;
1318 781
1319 memslot = &kvm->memslots[log->slot]; 782 memslot = &kvm->memslots->memslots[log->slot];
1320 r = -ENOENT; 783 r = -ENOENT;
1321 if (!memslot->dirty_bitmap) 784 if (!memslot->dirty_bitmap)
1322 goto out; 785 goto out;
1323 786
1324 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8; 787 n = kvm_dirty_bitmap_bytes(memslot);
1325 788
1326 for (i = 0; !any && i < n/sizeof(long); ++i) 789 for (i = 0; !any && i < n/sizeof(long); ++i)
1327 any = memslot->dirty_bitmap[i]; 790 any = memslot->dirty_bitmap[i];
@@ -1370,9 +833,10 @@ EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1370struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn) 833struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1371{ 834{
1372 int i; 835 int i;
836 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
1373 837
1374 for (i = 0; i < kvm->nmemslots; ++i) { 838 for (i = 0; i < slots->nmemslots; ++i) {
1375 struct kvm_memory_slot *memslot = &kvm->memslots[i]; 839 struct kvm_memory_slot *memslot = &slots->memslots[i];
1376 840
1377 if (gfn >= memslot->base_gfn 841 if (gfn >= memslot->base_gfn
1378 && gfn < memslot->base_gfn + memslot->npages) 842 && gfn < memslot->base_gfn + memslot->npages)
@@ -1391,10 +855,14 @@ struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1391int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) 855int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1392{ 856{
1393 int i; 857 int i;
858 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
1394 859
1395 gfn = unalias_gfn(kvm, gfn); 860 gfn = unalias_gfn_instantiation(kvm, gfn);
1396 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { 861 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1397 struct kvm_memory_slot *memslot = &kvm->memslots[i]; 862 struct kvm_memory_slot *memslot = &slots->memslots[i];
863
864 if (memslot->flags & KVM_MEMSLOT_INVALID)
865 continue;
1398 866
1399 if (gfn >= memslot->base_gfn 867 if (gfn >= memslot->base_gfn
1400 && gfn < memslot->base_gfn + memslot->npages) 868 && gfn < memslot->base_gfn + memslot->npages)
@@ -1404,33 +872,68 @@ int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1404} 872}
1405EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); 873EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1406 874
875unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
876{
877 struct vm_area_struct *vma;
878 unsigned long addr, size;
879
880 size = PAGE_SIZE;
881
882 addr = gfn_to_hva(kvm, gfn);
883 if (kvm_is_error_hva(addr))
884 return PAGE_SIZE;
885
886 down_read(&current->mm->mmap_sem);
887 vma = find_vma(current->mm, addr);
888 if (!vma)
889 goto out;
890
891 size = vma_kernel_pagesize(vma);
892
893out:
894 up_read(&current->mm->mmap_sem);
895
896 return size;
897}
898
899int memslot_id(struct kvm *kvm, gfn_t gfn)
900{
901 int i;
902 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
903 struct kvm_memory_slot *memslot = NULL;
904
905 gfn = unalias_gfn(kvm, gfn);
906 for (i = 0; i < slots->nmemslots; ++i) {
907 memslot = &slots->memslots[i];
908
909 if (gfn >= memslot->base_gfn
910 && gfn < memslot->base_gfn + memslot->npages)
911 break;
912 }
913
914 return memslot - slots->memslots;
915}
916
1407unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) 917unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1408{ 918{
1409 struct kvm_memory_slot *slot; 919 struct kvm_memory_slot *slot;
1410 920
1411 gfn = unalias_gfn(kvm, gfn); 921 gfn = unalias_gfn_instantiation(kvm, gfn);
1412 slot = gfn_to_memslot_unaliased(kvm, gfn); 922 slot = gfn_to_memslot_unaliased(kvm, gfn);
1413 if (!slot) 923 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1414 return bad_hva(); 924 return bad_hva();
1415 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE); 925 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1416} 926}
1417EXPORT_SYMBOL_GPL(gfn_to_hva); 927EXPORT_SYMBOL_GPL(gfn_to_hva);
1418 928
1419pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) 929static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
1420{ 930{
1421 struct page *page[1]; 931 struct page *page[1];
1422 unsigned long addr;
1423 int npages; 932 int npages;
1424 pfn_t pfn; 933 pfn_t pfn;
1425 934
1426 might_sleep(); 935 might_sleep();
1427 936
1428 addr = gfn_to_hva(kvm, gfn);
1429 if (kvm_is_error_hva(addr)) {
1430 get_page(bad_page);
1431 return page_to_pfn(bad_page);
1432 }
1433
1434 npages = get_user_pages_fast(addr, 1, 1, page); 937 npages = get_user_pages_fast(addr, 1, 1, page);
1435 938
1436 if (unlikely(npages != 1)) { 939 if (unlikely(npages != 1)) {
@@ -1455,8 +958,32 @@ pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1455 return pfn; 958 return pfn;
1456} 959}
1457 960
961pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
962{
963 unsigned long addr;
964
965 addr = gfn_to_hva(kvm, gfn);
966 if (kvm_is_error_hva(addr)) {
967 get_page(bad_page);
968 return page_to_pfn(bad_page);
969 }
970
971 return hva_to_pfn(kvm, addr);
972}
1458EXPORT_SYMBOL_GPL(gfn_to_pfn); 973EXPORT_SYMBOL_GPL(gfn_to_pfn);
1459 974
975static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
976{
977 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
978}
979
980pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
981 struct kvm_memory_slot *slot, gfn_t gfn)
982{
983 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
984 return hva_to_pfn(kvm, addr);
985}
986
1460struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) 987struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1461{ 988{
1462 pfn_t pfn; 989 pfn_t pfn;
@@ -1663,10 +1190,13 @@ void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1663 memslot = gfn_to_memslot_unaliased(kvm, gfn); 1190 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1664 if (memslot && memslot->dirty_bitmap) { 1191 if (memslot && memslot->dirty_bitmap) {
1665 unsigned long rel_gfn = gfn - memslot->base_gfn; 1192 unsigned long rel_gfn = gfn - memslot->base_gfn;
1193 unsigned long *p = memslot->dirty_bitmap +
1194 rel_gfn / BITS_PER_LONG;
1195 int offset = rel_gfn % BITS_PER_LONG;
1666 1196
1667 /* avoid RMW */ 1197 /* avoid RMW */
1668 if (!test_bit(rel_gfn, memslot->dirty_bitmap)) 1198 if (!generic_test_le_bit(offset, p))
1669 set_bit(rel_gfn, memslot->dirty_bitmap); 1199 generic___set_le_bit(offset, p);
1670 } 1200 }
1671} 1201}
1672 1202
@@ -1689,9 +1219,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1689 if (signal_pending(current)) 1219 if (signal_pending(current))
1690 break; 1220 break;
1691 1221
1692 vcpu_put(vcpu);
1693 schedule(); 1222 schedule();
1694 vcpu_load(vcpu);
1695 } 1223 }
1696 1224
1697 finish_wait(&vcpu->wq, &wait); 1225 finish_wait(&vcpu->wq, &wait);
@@ -1705,6 +1233,21 @@ void kvm_resched(struct kvm_vcpu *vcpu)
1705} 1233}
1706EXPORT_SYMBOL_GPL(kvm_resched); 1234EXPORT_SYMBOL_GPL(kvm_resched);
1707 1235
1236void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1237{
1238 ktime_t expires;
1239 DEFINE_WAIT(wait);
1240
1241 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1242
1243 /* Sleep for 100 us, and hope lock-holder got scheduled */
1244 expires = ktime_add_ns(ktime_get(), 100000UL);
1245 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1246
1247 finish_wait(&vcpu->wq, &wait);
1248}
1249EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1250
1708static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1251static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1709{ 1252{
1710 struct kvm_vcpu *vcpu = vma->vm_file->private_data; 1253 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
@@ -1757,7 +1300,7 @@ static struct file_operations kvm_vcpu_fops = {
1757 */ 1300 */
1758static int create_vcpu_fd(struct kvm_vcpu *vcpu) 1301static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1759{ 1302{
1760 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0); 1303 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1761} 1304}
1762 1305
1763/* 1306/*
@@ -1828,88 +1371,6 @@ static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1828 return 0; 1371 return 0;
1829} 1372}
1830 1373
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, 1374static long kvm_vcpu_ioctl(struct file *filp,
1914 unsigned int ioctl, unsigned long arg) 1375 unsigned int ioctl, unsigned long arg)
1915{ 1376{
@@ -2168,112 +1629,6 @@ static long kvm_vm_ioctl(struct file *filp,
2168 break; 1629 break;
2169 } 1630 }
2170#endif 1631#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: { 1632 case KVM_IRQFD: {
2278 struct kvm_irqfd data; 1633 struct kvm_irqfd data;
2279 1634
@@ -2305,10 +1660,58 @@ static long kvm_vm_ioctl(struct file *filp,
2305#endif 1660#endif
2306 default: 1661 default:
2307 r = kvm_arch_vm_ioctl(filp, ioctl, arg); 1662 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1663 if (r == -ENOTTY)
1664 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1665 }
1666out:
1667 return r;
1668}
1669
1670#ifdef CONFIG_COMPAT
1671struct compat_kvm_dirty_log {
1672 __u32 slot;
1673 __u32 padding1;
1674 union {
1675 compat_uptr_t dirty_bitmap; /* one bit per page */
1676 __u64 padding2;
1677 };
1678};
1679
1680static long kvm_vm_compat_ioctl(struct file *filp,
1681 unsigned int ioctl, unsigned long arg)
1682{
1683 struct kvm *kvm = filp->private_data;
1684 int r;
1685
1686 if (kvm->mm != current->mm)
1687 return -EIO;
1688 switch (ioctl) {
1689 case KVM_GET_DIRTY_LOG: {
1690 struct compat_kvm_dirty_log compat_log;
1691 struct kvm_dirty_log log;
1692
1693 r = -EFAULT;
1694 if (copy_from_user(&compat_log, (void __user *)arg,
1695 sizeof(compat_log)))
1696 goto out;
1697 log.slot = compat_log.slot;
1698 log.padding1 = compat_log.padding1;
1699 log.padding2 = compat_log.padding2;
1700 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1701
1702 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1703 if (r)
1704 goto out;
1705 break;
1706 }
1707 default:
1708 r = kvm_vm_ioctl(filp, ioctl, arg);
2308 } 1709 }
1710
2309out: 1711out:
2310 return r; 1712 return r;
2311} 1713}
1714#endif
2312 1715
2313static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1716static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2314{ 1717{
@@ -2344,7 +1747,9 @@ static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2344static struct file_operations kvm_vm_fops = { 1747static struct file_operations kvm_vm_fops = {
2345 .release = kvm_vm_release, 1748 .release = kvm_vm_release,
2346 .unlocked_ioctl = kvm_vm_ioctl, 1749 .unlocked_ioctl = kvm_vm_ioctl,
2347 .compat_ioctl = kvm_vm_ioctl, 1750#ifdef CONFIG_COMPAT
1751 .compat_ioctl = kvm_vm_compat_ioctl,
1752#endif
2348 .mmap = kvm_vm_mmap, 1753 .mmap = kvm_vm_mmap,
2349}; 1754};
2350 1755
@@ -2356,7 +1761,7 @@ static int kvm_dev_ioctl_create_vm(void)
2356 kvm = kvm_create_vm(); 1761 kvm = kvm_create_vm();
2357 if (IS_ERR(kvm)) 1762 if (IS_ERR(kvm))
2358 return PTR_ERR(kvm); 1763 return PTR_ERR(kvm);
2359 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0); 1764 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2360 if (fd < 0) 1765 if (fd < 0)
2361 kvm_put_kvm(kvm); 1766 kvm_put_kvm(kvm);
2362 1767
@@ -2372,6 +1777,7 @@ static long kvm_dev_ioctl_check_extension_generic(long arg)
2372#ifdef CONFIG_KVM_APIC_ARCHITECTURE 1777#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2373 case KVM_CAP_SET_BOOT_CPU_ID: 1778 case KVM_CAP_SET_BOOT_CPU_ID:
2374#endif 1779#endif
1780 case KVM_CAP_INTERNAL_ERROR_DATA:
2375 return 1; 1781 return 1;
2376#ifdef CONFIG_HAVE_KVM_IRQCHIP 1782#ifdef CONFIG_HAVE_KVM_IRQCHIP
2377 case KVM_CAP_IRQ_ROUTING: 1783 case KVM_CAP_IRQ_ROUTING:
@@ -2442,11 +1848,21 @@ static struct miscdevice kvm_dev = {
2442static void hardware_enable(void *junk) 1848static void hardware_enable(void *junk)
2443{ 1849{
2444 int cpu = raw_smp_processor_id(); 1850 int cpu = raw_smp_processor_id();
1851 int r;
2445 1852
2446 if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) 1853 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2447 return; 1854 return;
1855
2448 cpumask_set_cpu(cpu, cpus_hardware_enabled); 1856 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2449 kvm_arch_hardware_enable(NULL); 1857
1858 r = kvm_arch_hardware_enable(NULL);
1859
1860 if (r) {
1861 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1862 atomic_inc(&hardware_enable_failed);
1863 printk(KERN_INFO "kvm: enabling virtualization on "
1864 "CPU%d failed\n", cpu);
1865 }
2450} 1866}
2451 1867
2452static void hardware_disable(void *junk) 1868static void hardware_disable(void *junk)
@@ -2459,11 +1875,52 @@ static void hardware_disable(void *junk)
2459 kvm_arch_hardware_disable(NULL); 1875 kvm_arch_hardware_disable(NULL);
2460} 1876}
2461 1877
1878static void hardware_disable_all_nolock(void)
1879{
1880 BUG_ON(!kvm_usage_count);
1881
1882 kvm_usage_count--;
1883 if (!kvm_usage_count)
1884 on_each_cpu(hardware_disable, NULL, 1);
1885}
1886
1887static void hardware_disable_all(void)
1888{
1889 spin_lock(&kvm_lock);
1890 hardware_disable_all_nolock();
1891 spin_unlock(&kvm_lock);
1892}
1893
1894static int hardware_enable_all(void)
1895{
1896 int r = 0;
1897
1898 spin_lock(&kvm_lock);
1899
1900 kvm_usage_count++;
1901 if (kvm_usage_count == 1) {
1902 atomic_set(&hardware_enable_failed, 0);
1903 on_each_cpu(hardware_enable, NULL, 1);
1904
1905 if (atomic_read(&hardware_enable_failed)) {
1906 hardware_disable_all_nolock();
1907 r = -EBUSY;
1908 }
1909 }
1910
1911 spin_unlock(&kvm_lock);
1912
1913 return r;
1914}
1915
2462static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, 1916static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2463 void *v) 1917 void *v)
2464{ 1918{
2465 int cpu = (long)v; 1919 int cpu = (long)v;
2466 1920
1921 if (!kvm_usage_count)
1922 return NOTIFY_OK;
1923
2467 val &= ~CPU_TASKS_FROZEN; 1924 val &= ~CPU_TASKS_FROZEN;
2468 switch (val) { 1925 switch (val) {
2469 case CPU_DYING: 1926 case CPU_DYING:
@@ -2517,12 +1974,7 @@ static struct notifier_block kvm_reboot_notifier = {
2517 .priority = 0, 1974 .priority = 0,
2518}; 1975};
2519 1976
2520void kvm_io_bus_init(struct kvm_io_bus *bus) 1977static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2521{
2522 memset(bus, 0, sizeof(*bus));
2523}
2524
2525void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2526{ 1978{
2527 int i; 1979 int i;
2528 1980
@@ -2531,13 +1983,15 @@ void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2531 1983
2532 kvm_iodevice_destructor(pos); 1984 kvm_iodevice_destructor(pos);
2533 } 1985 }
1986 kfree(bus);
2534} 1987}
2535 1988
2536/* kvm_io_bus_write - called under kvm->slots_lock */ 1989/* kvm_io_bus_write - called under kvm->slots_lock */
2537int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr, 1990int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2538 int len, const void *val) 1991 int len, const void *val)
2539{ 1992{
2540 int i; 1993 int i;
1994 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
2541 for (i = 0; i < bus->dev_count; i++) 1995 for (i = 0; i < bus->dev_count; i++)
2542 if (!kvm_iodevice_write(bus->devs[i], addr, len, val)) 1996 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2543 return 0; 1997 return 0;
@@ -2545,59 +1999,71 @@ int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2545} 1999}
2546 2000
2547/* kvm_io_bus_read - called under kvm->slots_lock */ 2001/* kvm_io_bus_read - called under kvm->slots_lock */
2548int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val) 2002int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2003 int len, void *val)
2549{ 2004{
2550 int i; 2005 int i;
2006 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
2007
2551 for (i = 0; i < bus->dev_count; i++) 2008 for (i = 0; i < bus->dev_count; i++)
2552 if (!kvm_iodevice_read(bus->devs[i], addr, len, val)) 2009 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2553 return 0; 2010 return 0;
2554 return -EOPNOTSUPP; 2011 return -EOPNOTSUPP;
2555} 2012}
2556 2013
2557int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus, 2014/* Caller must hold slots_lock. */
2558 struct kvm_io_device *dev) 2015int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2016 struct kvm_io_device *dev)
2559{ 2017{
2560 int ret; 2018 struct kvm_io_bus *new_bus, *bus;
2561
2562 down_write(&kvm->slots_lock);
2563 ret = __kvm_io_bus_register_dev(bus, dev);
2564 up_write(&kvm->slots_lock);
2565
2566 return ret;
2567}
2568 2019
2569/* An unlocked version. Caller must have write lock on slots_lock. */ 2020 bus = kvm->buses[bus_idx];
2570int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2571 struct kvm_io_device *dev)
2572{
2573 if (bus->dev_count > NR_IOBUS_DEVS-1) 2021 if (bus->dev_count > NR_IOBUS_DEVS-1)
2574 return -ENOSPC; 2022 return -ENOSPC;
2575 2023
2576 bus->devs[bus->dev_count++] = dev; 2024 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2025 if (!new_bus)
2026 return -ENOMEM;
2027 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2028 new_bus->devs[new_bus->dev_count++] = dev;
2029 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2030 synchronize_srcu_expedited(&kvm->srcu);
2031 kfree(bus);
2577 2032
2578 return 0; 2033 return 0;
2579} 2034}
2580 2035
2581void kvm_io_bus_unregister_dev(struct kvm *kvm, 2036/* Caller must hold slots_lock. */
2582 struct kvm_io_bus *bus, 2037int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2583 struct kvm_io_device *dev) 2038 struct kvm_io_device *dev)
2584{ 2039{
2585 down_write(&kvm->slots_lock); 2040 int i, r;
2586 __kvm_io_bus_unregister_dev(bus, dev); 2041 struct kvm_io_bus *new_bus, *bus;
2587 up_write(&kvm->slots_lock);
2588}
2589 2042
2590/* An unlocked version. Caller must have write lock on slots_lock. */ 2043 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2591void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus, 2044 if (!new_bus)
2592 struct kvm_io_device *dev) 2045 return -ENOMEM;
2593{
2594 int i;
2595 2046
2596 for (i = 0; i < bus->dev_count; i++) 2047 bus = kvm->buses[bus_idx];
2597 if (bus->devs[i] == dev) { 2048 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2598 bus->devs[i] = bus->devs[--bus->dev_count]; 2049
2050 r = -ENOENT;
2051 for (i = 0; i < new_bus->dev_count; i++)
2052 if (new_bus->devs[i] == dev) {
2053 r = 0;
2054 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2599 break; 2055 break;
2600 } 2056 }
2057
2058 if (r) {
2059 kfree(new_bus);
2060 return r;
2061 }
2062
2063 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2064 synchronize_srcu_expedited(&kvm->srcu);
2065 kfree(bus);
2066 return r;
2601} 2067}
2602 2068
2603static struct notifier_block kvm_cpu_notifier = { 2069static struct notifier_block kvm_cpu_notifier = {
@@ -2666,13 +2132,15 @@ static void kvm_exit_debug(void)
2666 2132
2667static int kvm_suspend(struct sys_device *dev, pm_message_t state) 2133static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2668{ 2134{
2669 hardware_disable(NULL); 2135 if (kvm_usage_count)
2136 hardware_disable(NULL);
2670 return 0; 2137 return 0;
2671} 2138}
2672 2139
2673static int kvm_resume(struct sys_device *dev) 2140static int kvm_resume(struct sys_device *dev)
2674{ 2141{
2675 hardware_enable(NULL); 2142 if (kvm_usage_count)
2143 hardware_enable(NULL);
2676 return 0; 2144 return 0;
2677} 2145}
2678 2146
@@ -2747,7 +2215,6 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2747 goto out_free_1; 2215 goto out_free_1;
2748 } 2216 }
2749 2217
2750 on_each_cpu(hardware_enable, NULL, 1);
2751 r = register_cpu_notifier(&kvm_cpu_notifier); 2218 r = register_cpu_notifier(&kvm_cpu_notifier);
2752 if (r) 2219 if (r)
2753 goto out_free_2; 2220 goto out_free_2;
@@ -2797,7 +2264,6 @@ out_free_3:
2797 unregister_reboot_notifier(&kvm_reboot_notifier); 2264 unregister_reboot_notifier(&kvm_reboot_notifier);
2798 unregister_cpu_notifier(&kvm_cpu_notifier); 2265 unregister_cpu_notifier(&kvm_cpu_notifier);
2799out_free_2: 2266out_free_2:
2800 on_each_cpu(hardware_disable, NULL, 1);
2801out_free_1: 2267out_free_1:
2802 kvm_arch_hardware_unsetup(); 2268 kvm_arch_hardware_unsetup();
2803out_free_0a: 2269out_free_0a:
generated by cgit v1.2.2 (git 2.25.0) at 2025-04-04 22:30:18 -0400