aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAvi Kivity <avi@redhat.com>2009-08-26 07:57:50 -0400
committerAvi Kivity <avi@redhat.com>2009-12-03 02:32:09 -0500
commitbfd99ff5d483b11c32bca49fbff7a5ac59038b0a (patch)
treee48d9dbf89f6f3940133f7b80559740152af3f7b
parent367e1319b229110a27c53221c2fa32a6aa86d4a9 (diff)
KVM: Move assigned device code to own file
Signed-off-by: Avi Kivity <avi@redhat.com>
-rw-r--r--arch/ia64/kvm/Makefile2
-rw-r--r--arch/x86/kvm/Makefile3
-rw-r--r--include/linux/kvm_host.h17
-rw-r--r--virt/kvm/assigned-dev.c818
-rw-r--r--virt/kvm/kvm_main.c798
5 files changed, 840 insertions, 798 deletions
diff --git a/arch/ia64/kvm/Makefile b/arch/ia64/kvm/Makefile
index 0bb99b732908..1089b3e918ac 100644
--- a/arch/ia64/kvm/Makefile
+++ b/arch/ia64/kvm/Makefile
@@ -49,7 +49,7 @@ EXTRA_CFLAGS += -Ivirt/kvm -Iarch/ia64/kvm/
49EXTRA_AFLAGS += -Ivirt/kvm -Iarch/ia64/kvm/ 49EXTRA_AFLAGS += -Ivirt/kvm -Iarch/ia64/kvm/
50 50
51common-objs = $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \ 51common-objs = $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \
52 coalesced_mmio.o irq_comm.o) 52 coalesced_mmio.o irq_comm.o assigned-dev.o)
53 53
54ifeq ($(CONFIG_IOMMU_API),y) 54ifeq ($(CONFIG_IOMMU_API),y)
55common-objs += $(addprefix ../../../virt/kvm/, iommu.o) 55common-objs += $(addprefix ../../../virt/kvm/, iommu.o)
diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
index 0e7fe78d0f74..31a7035c4bd9 100644
--- a/arch/x86/kvm/Makefile
+++ b/arch/x86/kvm/Makefile
@@ -6,7 +6,8 @@ CFLAGS_svm.o := -I.
6CFLAGS_vmx.o := -I. 6CFLAGS_vmx.o := -I.
7 7
8kvm-y += $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \ 8kvm-y += $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \
9 coalesced_mmio.o irq_comm.o eventfd.o) 9 coalesced_mmio.o irq_comm.o eventfd.o \
10 assigned-dev.o)
10kvm-$(CONFIG_IOMMU_API) += $(addprefix ../../../virt/kvm/, iommu.o) 11kvm-$(CONFIG_IOMMU_API) += $(addprefix ../../../virt/kvm/, iommu.o)
11 12
12kvm-y += x86.o mmu.o emulate.o i8259.o irq.o lapic.o \ 13kvm-y += x86.o mmu.o emulate.o i8259.o irq.o lapic.o \
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 4aa5e1d9a797..c0a1cc35f080 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -577,4 +577,21 @@ static inline bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
577 return vcpu->kvm->bsp_vcpu_id == vcpu->vcpu_id; 577 return vcpu->kvm->bsp_vcpu_id == vcpu->vcpu_id;
578} 578}
579#endif 579#endif
580
581#ifdef __KVM_HAVE_DEVICE_ASSIGNMENT
582
583long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
584 unsigned long arg);
585
586#else
587
588static inline long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
589 unsigned long arg)
590{
591 return -ENOTTY;
592}
593
580#endif 594#endif
595
596#endif
597
diff --git a/virt/kvm/assigned-dev.c b/virt/kvm/assigned-dev.c
new file mode 100644
index 000000000000..fd9c097b760a
--- /dev/null
+++ b/virt/kvm/assigned-dev.c
@@ -0,0 +1,818 @@
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 "irq.h"
20
21static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
22 int assigned_dev_id)
23{
24 struct list_head *ptr;
25 struct kvm_assigned_dev_kernel *match;
26
27 list_for_each(ptr, head) {
28 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
29 if (match->assigned_dev_id == assigned_dev_id)
30 return match;
31 }
32 return NULL;
33}
34
35static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
36 *assigned_dev, int irq)
37{
38 int i, index;
39 struct msix_entry *host_msix_entries;
40
41 host_msix_entries = assigned_dev->host_msix_entries;
42
43 index = -1;
44 for (i = 0; i < assigned_dev->entries_nr; i++)
45 if (irq == host_msix_entries[i].vector) {
46 index = i;
47 break;
48 }
49 if (index < 0) {
50 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
51 return 0;
52 }
53
54 return index;
55}
56
57static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
58{
59 struct kvm_assigned_dev_kernel *assigned_dev;
60 struct kvm *kvm;
61 int i;
62
63 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
64 interrupt_work);
65 kvm = assigned_dev->kvm;
66
67 spin_lock_irq(&assigned_dev->assigned_dev_lock);
68 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
69 struct kvm_guest_msix_entry *guest_entries =
70 assigned_dev->guest_msix_entries;
71 for (i = 0; i < assigned_dev->entries_nr; i++) {
72 if (!(guest_entries[i].flags &
73 KVM_ASSIGNED_MSIX_PENDING))
74 continue;
75 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
76 kvm_set_irq(assigned_dev->kvm,
77 assigned_dev->irq_source_id,
78 guest_entries[i].vector, 1);
79 }
80 } else
81 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
82 assigned_dev->guest_irq, 1);
83
84 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
85}
86
87static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
88{
89 unsigned long flags;
90 struct kvm_assigned_dev_kernel *assigned_dev =
91 (struct kvm_assigned_dev_kernel *) dev_id;
92
93 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
94 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
95 int index = find_index_from_host_irq(assigned_dev, irq);
96 if (index < 0)
97 goto out;
98 assigned_dev->guest_msix_entries[index].flags |=
99 KVM_ASSIGNED_MSIX_PENDING;
100 }
101
102 schedule_work(&assigned_dev->interrupt_work);
103
104 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
105 disable_irq_nosync(irq);
106 assigned_dev->host_irq_disabled = true;
107 }
108
109out:
110 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
111 return IRQ_HANDLED;
112}
113
114/* Ack the irq line for an assigned device */
115static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
116{
117 struct kvm_assigned_dev_kernel *dev;
118 unsigned long flags;
119
120 if (kian->gsi == -1)
121 return;
122
123 dev = container_of(kian, struct kvm_assigned_dev_kernel,
124 ack_notifier);
125
126 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
127
128 /* The guest irq may be shared so this ack may be
129 * from another device.
130 */
131 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
132 if (dev->host_irq_disabled) {
133 enable_irq(dev->host_irq);
134 dev->host_irq_disabled = false;
135 }
136 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
137}
138
139static void deassign_guest_irq(struct kvm *kvm,
140 struct kvm_assigned_dev_kernel *assigned_dev)
141{
142 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
143 assigned_dev->ack_notifier.gsi = -1;
144
145 if (assigned_dev->irq_source_id != -1)
146 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
147 assigned_dev->irq_source_id = -1;
148 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
149}
150
151/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
152static void deassign_host_irq(struct kvm *kvm,
153 struct kvm_assigned_dev_kernel *assigned_dev)
154{
155 /*
156 * In kvm_free_device_irq, cancel_work_sync return true if:
157 * 1. work is scheduled, and then cancelled.
158 * 2. work callback is executed.
159 *
160 * The first one ensured that the irq is disabled and no more events
161 * would happen. But for the second one, the irq may be enabled (e.g.
162 * for MSI). So we disable irq here to prevent further events.
163 *
164 * Notice this maybe result in nested disable if the interrupt type is
165 * INTx, but it's OK for we are going to free it.
166 *
167 * If this function is a part of VM destroy, please ensure that till
168 * now, the kvm state is still legal for probably we also have to wait
169 * interrupt_work done.
170 */
171 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
172 int i;
173 for (i = 0; i < assigned_dev->entries_nr; i++)
174 disable_irq_nosync(assigned_dev->
175 host_msix_entries[i].vector);
176
177 cancel_work_sync(&assigned_dev->interrupt_work);
178
179 for (i = 0; i < assigned_dev->entries_nr; i++)
180 free_irq(assigned_dev->host_msix_entries[i].vector,
181 (void *)assigned_dev);
182
183 assigned_dev->entries_nr = 0;
184 kfree(assigned_dev->host_msix_entries);
185 kfree(assigned_dev->guest_msix_entries);
186 pci_disable_msix(assigned_dev->dev);
187 } else {
188 /* Deal with MSI and INTx */
189 disable_irq_nosync(assigned_dev->host_irq);
190 cancel_work_sync(&assigned_dev->interrupt_work);
191
192 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
193
194 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
195 pci_disable_msi(assigned_dev->dev);
196 }
197
198 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
199}
200
201static int kvm_deassign_irq(struct kvm *kvm,
202 struct kvm_assigned_dev_kernel *assigned_dev,
203 unsigned long irq_requested_type)
204{
205 unsigned long guest_irq_type, host_irq_type;
206
207 if (!irqchip_in_kernel(kvm))
208 return -EINVAL;
209 /* no irq assignment to deassign */
210 if (!assigned_dev->irq_requested_type)
211 return -ENXIO;
212
213 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
214 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
215
216 if (host_irq_type)
217 deassign_host_irq(kvm, assigned_dev);
218 if (guest_irq_type)
219 deassign_guest_irq(kvm, assigned_dev);
220
221 return 0;
222}
223
224static void kvm_free_assigned_irq(struct kvm *kvm,
225 struct kvm_assigned_dev_kernel *assigned_dev)
226{
227 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
228}
229
230static void kvm_free_assigned_device(struct kvm *kvm,
231 struct kvm_assigned_dev_kernel
232 *assigned_dev)
233{
234 kvm_free_assigned_irq(kvm, assigned_dev);
235
236 pci_reset_function(assigned_dev->dev);
237
238 pci_release_regions(assigned_dev->dev);
239 pci_disable_device(assigned_dev->dev);
240 pci_dev_put(assigned_dev->dev);
241
242 list_del(&assigned_dev->list);
243 kfree(assigned_dev);
244}
245
246void kvm_free_all_assigned_devices(struct kvm *kvm)
247{
248 struct list_head *ptr, *ptr2;
249 struct kvm_assigned_dev_kernel *assigned_dev;
250
251 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
252 assigned_dev = list_entry(ptr,
253 struct kvm_assigned_dev_kernel,
254 list);
255
256 kvm_free_assigned_device(kvm, assigned_dev);
257 }
258}
259
260static int assigned_device_enable_host_intx(struct kvm *kvm,
261 struct kvm_assigned_dev_kernel *dev)
262{
263 dev->host_irq = dev->dev->irq;
264 /* Even though this is PCI, we don't want to use shared
265 * interrupts. Sharing host devices with guest-assigned devices
266 * on the same interrupt line is not a happy situation: there
267 * are going to be long delays in accepting, acking, etc.
268 */
269 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
270 0, "kvm_assigned_intx_device", (void *)dev))
271 return -EIO;
272 return 0;
273}
274
275#ifdef __KVM_HAVE_MSI
276static int assigned_device_enable_host_msi(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *dev)
278{
279 int r;
280
281 if (!dev->dev->msi_enabled) {
282 r = pci_enable_msi(dev->dev);
283 if (r)
284 return r;
285 }
286
287 dev->host_irq = dev->dev->irq;
288 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
289 "kvm_assigned_msi_device", (void *)dev)) {
290 pci_disable_msi(dev->dev);
291 return -EIO;
292 }
293
294 return 0;
295}
296#endif
297
298#ifdef __KVM_HAVE_MSIX
299static int assigned_device_enable_host_msix(struct kvm *kvm,
300 struct kvm_assigned_dev_kernel *dev)
301{
302 int i, r = -EINVAL;
303
304 /* host_msix_entries and guest_msix_entries should have been
305 * initialized */
306 if (dev->entries_nr == 0)
307 return r;
308
309 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
310 if (r)
311 return r;
312
313 for (i = 0; i < dev->entries_nr; i++) {
314 r = request_irq(dev->host_msix_entries[i].vector,
315 kvm_assigned_dev_intr, 0,
316 "kvm_assigned_msix_device",
317 (void *)dev);
318 /* FIXME: free requested_irq's on failure */
319 if (r)
320 return r;
321 }
322
323 return 0;
324}
325
326#endif
327
328static int assigned_device_enable_guest_intx(struct kvm *kvm,
329 struct kvm_assigned_dev_kernel *dev,
330 struct kvm_assigned_irq *irq)
331{
332 dev->guest_irq = irq->guest_irq;
333 dev->ack_notifier.gsi = irq->guest_irq;
334 return 0;
335}
336
337#ifdef __KVM_HAVE_MSI
338static int assigned_device_enable_guest_msi(struct kvm *kvm,
339 struct kvm_assigned_dev_kernel *dev,
340 struct kvm_assigned_irq *irq)
341{
342 dev->guest_irq = irq->guest_irq;
343 dev->ack_notifier.gsi = -1;
344 dev->host_irq_disabled = false;
345 return 0;
346}
347#endif
348
349#ifdef __KVM_HAVE_MSIX
350static int assigned_device_enable_guest_msix(struct kvm *kvm,
351 struct kvm_assigned_dev_kernel *dev,
352 struct kvm_assigned_irq *irq)
353{
354 dev->guest_irq = irq->guest_irq;
355 dev->ack_notifier.gsi = -1;
356 dev->host_irq_disabled = false;
357 return 0;
358}
359#endif
360
361static int assign_host_irq(struct kvm *kvm,
362 struct kvm_assigned_dev_kernel *dev,
363 __u32 host_irq_type)
364{
365 int r = -EEXIST;
366
367 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
368 return r;
369
370 switch (host_irq_type) {
371 case KVM_DEV_IRQ_HOST_INTX:
372 r = assigned_device_enable_host_intx(kvm, dev);
373 break;
374#ifdef __KVM_HAVE_MSI
375 case KVM_DEV_IRQ_HOST_MSI:
376 r = assigned_device_enable_host_msi(kvm, dev);
377 break;
378#endif
379#ifdef __KVM_HAVE_MSIX
380 case KVM_DEV_IRQ_HOST_MSIX:
381 r = assigned_device_enable_host_msix(kvm, dev);
382 break;
383#endif
384 default:
385 r = -EINVAL;
386 }
387
388 if (!r)
389 dev->irq_requested_type |= host_irq_type;
390
391 return r;
392}
393
394static int assign_guest_irq(struct kvm *kvm,
395 struct kvm_assigned_dev_kernel *dev,
396 struct kvm_assigned_irq *irq,
397 unsigned long guest_irq_type)
398{
399 int id;
400 int r = -EEXIST;
401
402 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
403 return r;
404
405 id = kvm_request_irq_source_id(kvm);
406 if (id < 0)
407 return id;
408
409 dev->irq_source_id = id;
410
411 switch (guest_irq_type) {
412 case KVM_DEV_IRQ_GUEST_INTX:
413 r = assigned_device_enable_guest_intx(kvm, dev, irq);
414 break;
415#ifdef __KVM_HAVE_MSI
416 case KVM_DEV_IRQ_GUEST_MSI:
417 r = assigned_device_enable_guest_msi(kvm, dev, irq);
418 break;
419#endif
420#ifdef __KVM_HAVE_MSIX
421 case KVM_DEV_IRQ_GUEST_MSIX:
422 r = assigned_device_enable_guest_msix(kvm, dev, irq);
423 break;
424#endif
425 default:
426 r = -EINVAL;
427 }
428
429 if (!r) {
430 dev->irq_requested_type |= guest_irq_type;
431 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
432 } else
433 kvm_free_irq_source_id(kvm, dev->irq_source_id);
434
435 return r;
436}
437
438/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
439static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
440 struct kvm_assigned_irq *assigned_irq)
441{
442 int r = -EINVAL;
443 struct kvm_assigned_dev_kernel *match;
444 unsigned long host_irq_type, guest_irq_type;
445
446 if (!capable(CAP_SYS_RAWIO))
447 return -EPERM;
448
449 if (!irqchip_in_kernel(kvm))
450 return r;
451
452 mutex_lock(&kvm->lock);
453 r = -ENODEV;
454 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
455 assigned_irq->assigned_dev_id);
456 if (!match)
457 goto out;
458
459 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
460 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
461
462 r = -EINVAL;
463 /* can only assign one type at a time */
464 if (hweight_long(host_irq_type) > 1)
465 goto out;
466 if (hweight_long(guest_irq_type) > 1)
467 goto out;
468 if (host_irq_type == 0 && guest_irq_type == 0)
469 goto out;
470
471 r = 0;
472 if (host_irq_type)
473 r = assign_host_irq(kvm, match, host_irq_type);
474 if (r)
475 goto out;
476
477 if (guest_irq_type)
478 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
479out:
480 mutex_unlock(&kvm->lock);
481 return r;
482}
483
484static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
485 struct kvm_assigned_irq
486 *assigned_irq)
487{
488 int r = -ENODEV;
489 struct kvm_assigned_dev_kernel *match;
490
491 mutex_lock(&kvm->lock);
492
493 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
494 assigned_irq->assigned_dev_id);
495 if (!match)
496 goto out;
497
498 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
499out:
500 mutex_unlock(&kvm->lock);
501 return r;
502}
503
504static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
505 struct kvm_assigned_pci_dev *assigned_dev)
506{
507 int r = 0;
508 struct kvm_assigned_dev_kernel *match;
509 struct pci_dev *dev;
510
511 down_read(&kvm->slots_lock);
512 mutex_lock(&kvm->lock);
513
514 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
515 assigned_dev->assigned_dev_id);
516 if (match) {
517 /* device already assigned */
518 r = -EEXIST;
519 goto out;
520 }
521
522 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
523 if (match == NULL) {
524 printk(KERN_INFO "%s: Couldn't allocate memory\n",
525 __func__);
526 r = -ENOMEM;
527 goto out;
528 }
529 dev = pci_get_bus_and_slot(assigned_dev->busnr,
530 assigned_dev->devfn);
531 if (!dev) {
532 printk(KERN_INFO "%s: host device not found\n", __func__);
533 r = -EINVAL;
534 goto out_free;
535 }
536 if (pci_enable_device(dev)) {
537 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
538 r = -EBUSY;
539 goto out_put;
540 }
541 r = pci_request_regions(dev, "kvm_assigned_device");
542 if (r) {
543 printk(KERN_INFO "%s: Could not get access to device regions\n",
544 __func__);
545 goto out_disable;
546 }
547
548 pci_reset_function(dev);
549
550 match->assigned_dev_id = assigned_dev->assigned_dev_id;
551 match->host_busnr = assigned_dev->busnr;
552 match->host_devfn = assigned_dev->devfn;
553 match->flags = assigned_dev->flags;
554 match->dev = dev;
555 spin_lock_init(&match->assigned_dev_lock);
556 match->irq_source_id = -1;
557 match->kvm = kvm;
558 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
559 INIT_WORK(&match->interrupt_work,
560 kvm_assigned_dev_interrupt_work_handler);
561
562 list_add(&match->list, &kvm->arch.assigned_dev_head);
563
564 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
565 if (!kvm->arch.iommu_domain) {
566 r = kvm_iommu_map_guest(kvm);
567 if (r)
568 goto out_list_del;
569 }
570 r = kvm_assign_device(kvm, match);
571 if (r)
572 goto out_list_del;
573 }
574
575out:
576 mutex_unlock(&kvm->lock);
577 up_read(&kvm->slots_lock);
578 return r;
579out_list_del:
580 list_del(&match->list);
581 pci_release_regions(dev);
582out_disable:
583 pci_disable_device(dev);
584out_put:
585 pci_dev_put(dev);
586out_free:
587 kfree(match);
588 mutex_unlock(&kvm->lock);
589 up_read(&kvm->slots_lock);
590 return r;
591}
592
593static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
594 struct kvm_assigned_pci_dev *assigned_dev)
595{
596 int r = 0;
597 struct kvm_assigned_dev_kernel *match;
598
599 mutex_lock(&kvm->lock);
600
601 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
602 assigned_dev->assigned_dev_id);
603 if (!match) {
604 printk(KERN_INFO "%s: device hasn't been assigned before, "
605 "so cannot be deassigned\n", __func__);
606 r = -EINVAL;
607 goto out;
608 }
609
610 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
611 kvm_deassign_device(kvm, match);
612
613 kvm_free_assigned_device(kvm, match);
614
615out:
616 mutex_unlock(&kvm->lock);
617 return r;
618}
619
620
621#ifdef __KVM_HAVE_MSIX
622static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
623 struct kvm_assigned_msix_nr *entry_nr)
624{
625 int r = 0;
626 struct kvm_assigned_dev_kernel *adev;
627
628 mutex_lock(&kvm->lock);
629
630 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
631 entry_nr->assigned_dev_id);
632 if (!adev) {
633 r = -EINVAL;
634 goto msix_nr_out;
635 }
636
637 if (adev->entries_nr == 0) {
638 adev->entries_nr = entry_nr->entry_nr;
639 if (adev->entries_nr == 0 ||
640 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
641 r = -EINVAL;
642 goto msix_nr_out;
643 }
644
645 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
646 entry_nr->entry_nr,
647 GFP_KERNEL);
648 if (!adev->host_msix_entries) {
649 r = -ENOMEM;
650 goto msix_nr_out;
651 }
652 adev->guest_msix_entries = kzalloc(
653 sizeof(struct kvm_guest_msix_entry) *
654 entry_nr->entry_nr, GFP_KERNEL);
655 if (!adev->guest_msix_entries) {
656 kfree(adev->host_msix_entries);
657 r = -ENOMEM;
658 goto msix_nr_out;
659 }
660 } else /* Not allowed set MSI-X number twice */
661 r = -EINVAL;
662msix_nr_out:
663 mutex_unlock(&kvm->lock);
664 return r;
665}
666
667static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
668 struct kvm_assigned_msix_entry *entry)
669{
670 int r = 0, i;
671 struct kvm_assigned_dev_kernel *adev;
672
673 mutex_lock(&kvm->lock);
674
675 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
676 entry->assigned_dev_id);
677
678 if (!adev) {
679 r = -EINVAL;
680 goto msix_entry_out;
681 }
682
683 for (i = 0; i < adev->entries_nr; i++)
684 if (adev->guest_msix_entries[i].vector == 0 ||
685 adev->guest_msix_entries[i].entry == entry->entry) {
686 adev->guest_msix_entries[i].entry = entry->entry;
687 adev->guest_msix_entries[i].vector = entry->gsi;
688 adev->host_msix_entries[i].entry = entry->entry;
689 break;
690 }
691 if (i == adev->entries_nr) {
692 r = -ENOSPC;
693 goto msix_entry_out;
694 }
695
696msix_entry_out:
697 mutex_unlock(&kvm->lock);
698
699 return r;
700}
701#endif
702
703long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
704 unsigned long arg)
705{
706 void __user *argp = (void __user *)arg;
707 int r = -ENOTTY;
708
709 switch (ioctl) {
710 case KVM_ASSIGN_PCI_DEVICE: {
711 struct kvm_assigned_pci_dev assigned_dev;
712
713 r = -EFAULT;
714 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
715 goto out;
716 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
717 if (r)
718 goto out;
719 break;
720 }
721 case KVM_ASSIGN_IRQ: {
722 r = -EOPNOTSUPP;
723 break;
724 }
725#ifdef KVM_CAP_ASSIGN_DEV_IRQ
726 case KVM_ASSIGN_DEV_IRQ: {
727 struct kvm_assigned_irq assigned_irq;
728
729 r = -EFAULT;
730 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
731 goto out;
732 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
733 if (r)
734 goto out;
735 break;
736 }
737 case KVM_DEASSIGN_DEV_IRQ: {
738 struct kvm_assigned_irq assigned_irq;
739
740 r = -EFAULT;
741 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
742 goto out;
743 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
744 if (r)
745 goto out;
746 break;
747 }
748#endif
749#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
750 case KVM_DEASSIGN_PCI_DEVICE: {
751 struct kvm_assigned_pci_dev assigned_dev;
752
753 r = -EFAULT;
754 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
755 goto out;
756 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
757 if (r)
758 goto out;
759 break;
760 }
761#endif
762#ifdef KVM_CAP_IRQ_ROUTING
763 case KVM_SET_GSI_ROUTING: {
764 struct kvm_irq_routing routing;
765 struct kvm_irq_routing __user *urouting;
766 struct kvm_irq_routing_entry *entries;
767
768 r = -EFAULT;
769 if (copy_from_user(&routing, argp, sizeof(routing)))
770 goto out;
771 r = -EINVAL;
772 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
773 goto out;
774 if (routing.flags)
775 goto out;
776 r = -ENOMEM;
777 entries = vmalloc(routing.nr * sizeof(*entries));
778 if (!entries)
779 goto out;
780 r = -EFAULT;
781 urouting = argp;
782 if (copy_from_user(entries, urouting->entries,
783 routing.nr * sizeof(*entries)))
784 goto out_free_irq_routing;
785 r = kvm_set_irq_routing(kvm, entries, routing.nr,
786 routing.flags);
787 out_free_irq_routing:
788 vfree(entries);
789 break;
790 }
791#endif /* KVM_CAP_IRQ_ROUTING */
792#ifdef __KVM_HAVE_MSIX
793 case KVM_ASSIGN_SET_MSIX_NR: {
794 struct kvm_assigned_msix_nr entry_nr;
795 r = -EFAULT;
796 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
797 goto out;
798 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
799 if (r)
800 goto out;
801 break;
802 }
803 case KVM_ASSIGN_SET_MSIX_ENTRY: {
804 struct kvm_assigned_msix_entry entry;
805 r = -EFAULT;
806 if (copy_from_user(&entry, argp, sizeof entry))
807 goto out;
808 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
809 if (r)
810 goto out;
811 break;
812 }
813#endif
814 }
815out:
816 return r;
817}
818
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index c12c95b1b641..38e4d2c34ac1 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -53,12 +53,6 @@
53#include "coalesced_mmio.h" 53#include "coalesced_mmio.h"
54#endif 54#endif
55 55
56#ifdef KVM_CAP_DEVICE_ASSIGNMENT
57#include <linux/pci.h>
58#include <linux/interrupt.h>
59#include "irq.h"
60#endif
61
62#define CREATE_TRACE_POINTS 56#define CREATE_TRACE_POINTS
63#include <trace/events/kvm.h> 57#include <trace/events/kvm.h>
64 58
@@ -90,608 +84,6 @@ static bool kvm_rebooting;
90 84
91static bool largepages_enabled = true; 85static bool largepages_enabled = true;
92 86
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 spin_lock_irq(&assigned_dev->assigned_dev_lock);
141 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
142 struct kvm_guest_msix_entry *guest_entries =
143 assigned_dev->guest_msix_entries;
144 for (i = 0; i < assigned_dev->entries_nr; i++) {
145 if (!(guest_entries[i].flags &
146 KVM_ASSIGNED_MSIX_PENDING))
147 continue;
148 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
149 kvm_set_irq(assigned_dev->kvm,
150 assigned_dev->irq_source_id,
151 guest_entries[i].vector, 1);
152 }
153 } else
154 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
155 assigned_dev->guest_irq, 1);
156
157 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
158}
159
160static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
161{
162 unsigned long flags;
163 struct kvm_assigned_dev_kernel *assigned_dev =
164 (struct kvm_assigned_dev_kernel *) dev_id;
165
166 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
167 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
168 int index = find_index_from_host_irq(assigned_dev, irq);
169 if (index < 0)
170 goto out;
171 assigned_dev->guest_msix_entries[index].flags |=
172 KVM_ASSIGNED_MSIX_PENDING;
173 }
174
175 schedule_work(&assigned_dev->interrupt_work);
176
177 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
178 disable_irq_nosync(irq);
179 assigned_dev->host_irq_disabled = true;
180 }
181
182out:
183 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
184 return IRQ_HANDLED;
185}
186
187/* Ack the irq line for an assigned device */
188static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
189{
190 struct kvm_assigned_dev_kernel *dev;
191 unsigned long flags;
192
193 if (kian->gsi == -1)
194 return;
195
196 dev = container_of(kian, struct kvm_assigned_dev_kernel,
197 ack_notifier);
198
199 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
200
201 /* The guest irq may be shared so this ack may be
202 * from another device.
203 */
204 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
205 if (dev->host_irq_disabled) {
206 enable_irq(dev->host_irq);
207 dev->host_irq_disabled = false;
208 }
209 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
210}
211
212static void deassign_guest_irq(struct kvm *kvm,
213 struct kvm_assigned_dev_kernel *assigned_dev)
214{
215 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
216 assigned_dev->ack_notifier.gsi = -1;
217
218 if (assigned_dev->irq_source_id != -1)
219 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
220 assigned_dev->irq_source_id = -1;
221 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
222}
223
224/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
225static void deassign_host_irq(struct kvm *kvm,
226 struct kvm_assigned_dev_kernel *assigned_dev)
227{
228 /*
229 * In kvm_free_device_irq, cancel_work_sync return true if:
230 * 1. work is scheduled, and then cancelled.
231 * 2. work callback is executed.
232 *
233 * The first one ensured that the irq is disabled and no more events
234 * would happen. But for the second one, the irq may be enabled (e.g.
235 * for MSI). So we disable irq here to prevent further events.
236 *
237 * Notice this maybe result in nested disable if the interrupt type is
238 * INTx, but it's OK for we are going to free it.
239 *
240 * If this function is a part of VM destroy, please ensure that till
241 * now, the kvm state is still legal for probably we also have to wait
242 * interrupt_work done.
243 */
244 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
245 int i;
246 for (i = 0; i < assigned_dev->entries_nr; i++)
247 disable_irq_nosync(assigned_dev->
248 host_msix_entries[i].vector);
249
250 cancel_work_sync(&assigned_dev->interrupt_work);
251
252 for (i = 0; i < assigned_dev->entries_nr; i++)
253 free_irq(assigned_dev->host_msix_entries[i].vector,
254 (void *)assigned_dev);
255
256 assigned_dev->entries_nr = 0;
257 kfree(assigned_dev->host_msix_entries);
258 kfree(assigned_dev->guest_msix_entries);
259 pci_disable_msix(assigned_dev->dev);
260 } else {
261 /* Deal with MSI and INTx */
262 disable_irq_nosync(assigned_dev->host_irq);
263 cancel_work_sync(&assigned_dev->interrupt_work);
264
265 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
266
267 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
268 pci_disable_msi(assigned_dev->dev);
269 }
270
271 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
272}
273
274static int kvm_deassign_irq(struct kvm *kvm,
275 struct kvm_assigned_dev_kernel *assigned_dev,
276 unsigned long irq_requested_type)
277{
278 unsigned long guest_irq_type, host_irq_type;
279
280 if (!irqchip_in_kernel(kvm))
281 return -EINVAL;
282 /* no irq assignment to deassign */
283 if (!assigned_dev->irq_requested_type)
284 return -ENXIO;
285
286 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
287 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
288
289 if (host_irq_type)
290 deassign_host_irq(kvm, assigned_dev);
291 if (guest_irq_type)
292 deassign_guest_irq(kvm, assigned_dev);
293
294 return 0;
295}
296
297static void kvm_free_assigned_irq(struct kvm *kvm,
298 struct kvm_assigned_dev_kernel *assigned_dev)
299{
300 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
301}
302
303static void kvm_free_assigned_device(struct kvm *kvm,
304 struct kvm_assigned_dev_kernel
305 *assigned_dev)
306{
307 kvm_free_assigned_irq(kvm, assigned_dev);
308
309 pci_reset_function(assigned_dev->dev);
310
311 pci_release_regions(assigned_dev->dev);
312 pci_disable_device(assigned_dev->dev);
313 pci_dev_put(assigned_dev->dev);
314
315 list_del(&assigned_dev->list);
316 kfree(assigned_dev);
317}
318
319void kvm_free_all_assigned_devices(struct kvm *kvm)
320{
321 struct list_head *ptr, *ptr2;
322 struct kvm_assigned_dev_kernel *assigned_dev;
323
324 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
325 assigned_dev = list_entry(ptr,
326 struct kvm_assigned_dev_kernel,
327 list);
328
329 kvm_free_assigned_device(kvm, assigned_dev);
330 }
331}
332
333static int assigned_device_enable_host_intx(struct kvm *kvm,
334 struct kvm_assigned_dev_kernel *dev)
335{
336 dev->host_irq = dev->dev->irq;
337 /* Even though this is PCI, we don't want to use shared
338 * interrupts. Sharing host devices with guest-assigned devices
339 * on the same interrupt line is not a happy situation: there
340 * are going to be long delays in accepting, acking, etc.
341 */
342 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
343 0, "kvm_assigned_intx_device", (void *)dev))
344 return -EIO;
345 return 0;
346}
347
348#ifdef __KVM_HAVE_MSI
349static int assigned_device_enable_host_msi(struct kvm *kvm,
350 struct kvm_assigned_dev_kernel *dev)
351{
352 int r;
353
354 if (!dev->dev->msi_enabled) {
355 r = pci_enable_msi(dev->dev);
356 if (r)
357 return r;
358 }
359
360 dev->host_irq = dev->dev->irq;
361 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
362 "kvm_assigned_msi_device", (void *)dev)) {
363 pci_disable_msi(dev->dev);
364 return -EIO;
365 }
366
367 return 0;
368}
369#endif
370
371#ifdef __KVM_HAVE_MSIX
372static int assigned_device_enable_host_msix(struct kvm *kvm,
373 struct kvm_assigned_dev_kernel *dev)
374{
375 int i, r = -EINVAL;
376
377 /* host_msix_entries and guest_msix_entries should have been
378 * initialized */
379 if (dev->entries_nr == 0)
380 return r;
381
382 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
383 if (r)
384 return r;
385
386 for (i = 0; i < dev->entries_nr; i++) {
387 r = request_irq(dev->host_msix_entries[i].vector,
388 kvm_assigned_dev_intr, 0,
389 "kvm_assigned_msix_device",
390 (void *)dev);
391 /* FIXME: free requested_irq's on failure */
392 if (r)
393 return r;
394 }
395
396 return 0;
397}
398
399#endif
400
401static int assigned_device_enable_guest_intx(struct kvm *kvm,
402 struct kvm_assigned_dev_kernel *dev,
403 struct kvm_assigned_irq *irq)
404{
405 dev->guest_irq = irq->guest_irq;
406 dev->ack_notifier.gsi = irq->guest_irq;
407 return 0;
408}
409
410#ifdef __KVM_HAVE_MSI
411static int assigned_device_enable_guest_msi(struct kvm *kvm,
412 struct kvm_assigned_dev_kernel *dev,
413 struct kvm_assigned_irq *irq)
414{
415 dev->guest_irq = irq->guest_irq;
416 dev->ack_notifier.gsi = -1;
417 dev->host_irq_disabled = false;
418 return 0;
419}
420#endif
421#ifdef __KVM_HAVE_MSIX
422static int assigned_device_enable_guest_msix(struct kvm *kvm,
423 struct kvm_assigned_dev_kernel *dev,
424 struct kvm_assigned_irq *irq)
425{
426 dev->guest_irq = irq->guest_irq;
427 dev->ack_notifier.gsi = -1;
428 dev->host_irq_disabled = false;
429 return 0;
430}
431#endif
432
433static int assign_host_irq(struct kvm *kvm,
434 struct kvm_assigned_dev_kernel *dev,
435 __u32 host_irq_type)
436{
437 int r = -EEXIST;
438
439 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
440 return r;
441
442 switch (host_irq_type) {
443 case KVM_DEV_IRQ_HOST_INTX:
444 r = assigned_device_enable_host_intx(kvm, dev);
445 break;
446#ifdef __KVM_HAVE_MSI
447 case KVM_DEV_IRQ_HOST_MSI:
448 r = assigned_device_enable_host_msi(kvm, dev);
449 break;
450#endif
451#ifdef __KVM_HAVE_MSIX
452 case KVM_DEV_IRQ_HOST_MSIX:
453 r = assigned_device_enable_host_msix(kvm, dev);
454 break;
455#endif
456 default:
457 r = -EINVAL;
458 }
459
460 if (!r)
461 dev->irq_requested_type |= host_irq_type;
462
463 return r;
464}
465
466static int assign_guest_irq(struct kvm *kvm,
467 struct kvm_assigned_dev_kernel *dev,
468 struct kvm_assigned_irq *irq,
469 unsigned long guest_irq_type)
470{
471 int id;
472 int r = -EEXIST;
473
474 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
475 return r;
476
477 id = kvm_request_irq_source_id(kvm);
478 if (id < 0)
479 return id;
480
481 dev->irq_source_id = id;
482
483 switch (guest_irq_type) {
484 case KVM_DEV_IRQ_GUEST_INTX:
485 r = assigned_device_enable_guest_intx(kvm, dev, irq);
486 break;
487#ifdef __KVM_HAVE_MSI
488 case KVM_DEV_IRQ_GUEST_MSI:
489 r = assigned_device_enable_guest_msi(kvm, dev, irq);
490 break;
491#endif
492#ifdef __KVM_HAVE_MSIX
493 case KVM_DEV_IRQ_GUEST_MSIX:
494 r = assigned_device_enable_guest_msix(kvm, dev, irq);
495 break;
496#endif
497 default:
498 r = -EINVAL;
499 }
500
501 if (!r) {
502 dev->irq_requested_type |= guest_irq_type;
503 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
504 } else
505 kvm_free_irq_source_id(kvm, dev->irq_source_id);
506
507 return r;
508}
509
510/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
511static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
512 struct kvm_assigned_irq *assigned_irq)
513{
514 int r = -EINVAL;
515 struct kvm_assigned_dev_kernel *match;
516 unsigned long host_irq_type, guest_irq_type;
517
518 if (!capable(CAP_SYS_RAWIO))
519 return -EPERM;
520
521 if (!irqchip_in_kernel(kvm))
522 return r;
523
524 mutex_lock(&kvm->lock);
525 r = -ENODEV;
526 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
527 assigned_irq->assigned_dev_id);
528 if (!match)
529 goto out;
530
531 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
532 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
533
534 r = -EINVAL;
535 /* can only assign one type at a time */
536 if (hweight_long(host_irq_type) > 1)
537 goto out;
538 if (hweight_long(guest_irq_type) > 1)
539 goto out;
540 if (host_irq_type == 0 && guest_irq_type == 0)
541 goto out;
542
543 r = 0;
544 if (host_irq_type)
545 r = assign_host_irq(kvm, match, host_irq_type);
546 if (r)
547 goto out;
548
549 if (guest_irq_type)
550 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
551out:
552 mutex_unlock(&kvm->lock);
553 return r;
554}
555
556static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
557 struct kvm_assigned_irq
558 *assigned_irq)
559{
560 int r = -ENODEV;
561 struct kvm_assigned_dev_kernel *match;
562
563 mutex_lock(&kvm->lock);
564
565 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
566 assigned_irq->assigned_dev_id);
567 if (!match)
568 goto out;
569
570 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
571out:
572 mutex_unlock(&kvm->lock);
573 return r;
574}
575
576static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
577 struct kvm_assigned_pci_dev *assigned_dev)
578{
579 int r = 0;
580 struct kvm_assigned_dev_kernel *match;
581 struct pci_dev *dev;
582
583 down_read(&kvm->slots_lock);
584 mutex_lock(&kvm->lock);
585
586 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
587 assigned_dev->assigned_dev_id);
588 if (match) {
589 /* device already assigned */
590 r = -EEXIST;
591 goto out;
592 }
593
594 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
595 if (match == NULL) {
596 printk(KERN_INFO "%s: Couldn't allocate memory\n",
597 __func__);
598 r = -ENOMEM;
599 goto out;
600 }
601 dev = pci_get_bus_and_slot(assigned_dev->busnr,
602 assigned_dev->devfn);
603 if (!dev) {
604 printk(KERN_INFO "%s: host device not found\n", __func__);
605 r = -EINVAL;
606 goto out_free;
607 }
608 if (pci_enable_device(dev)) {
609 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
610 r = -EBUSY;
611 goto out_put;
612 }
613 r = pci_request_regions(dev, "kvm_assigned_device");
614 if (r) {
615 printk(KERN_INFO "%s: Could not get access to device regions\n",
616 __func__);
617 goto out_disable;
618 }
619
620 pci_reset_function(dev);
621
622 match->assigned_dev_id = assigned_dev->assigned_dev_id;
623 match->host_busnr = assigned_dev->busnr;
624 match->host_devfn = assigned_dev->devfn;
625 match->flags = assigned_dev->flags;
626 match->dev = dev;
627 spin_lock_init(&match->assigned_dev_lock);
628 match->irq_source_id = -1;
629 match->kvm = kvm;
630 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
631 INIT_WORK(&match->interrupt_work,
632 kvm_assigned_dev_interrupt_work_handler);
633
634 list_add(&match->list, &kvm->arch.assigned_dev_head);
635
636 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
637 if (!kvm->arch.iommu_domain) {
638 r = kvm_iommu_map_guest(kvm);
639 if (r)
640 goto out_list_del;
641 }
642 r = kvm_assign_device(kvm, match);
643 if (r)
644 goto out_list_del;
645 }
646
647out:
648 mutex_unlock(&kvm->lock);
649 up_read(&kvm->slots_lock);
650 return r;
651out_list_del:
652 list_del(&match->list);
653 pci_release_regions(dev);
654out_disable:
655 pci_disable_device(dev);
656out_put:
657 pci_dev_put(dev);
658out_free:
659 kfree(match);
660 mutex_unlock(&kvm->lock);
661 up_read(&kvm->slots_lock);
662 return r;
663}
664#endif
665
666#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
667static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
668 struct kvm_assigned_pci_dev *assigned_dev)
669{
670 int r = 0;
671 struct kvm_assigned_dev_kernel *match;
672
673 mutex_lock(&kvm->lock);
674
675 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
676 assigned_dev->assigned_dev_id);
677 if (!match) {
678 printk(KERN_INFO "%s: device hasn't been assigned before, "
679 "so cannot be deassigned\n", __func__);
680 r = -EINVAL;
681 goto out;
682 }
683
684 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
685 kvm_deassign_device(kvm, match);
686
687 kvm_free_assigned_device(kvm, match);
688
689out:
690 mutex_unlock(&kvm->lock);
691 return r;
692}
693#endif
694
695inline int kvm_is_mmio_pfn(pfn_t pfn) 87inline int kvm_is_mmio_pfn(pfn_t pfn)
696{ 88{
697 if (pfn_valid(pfn)) { 89 if (pfn_valid(pfn)) {
@@ -1824,88 +1216,6 @@ static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1824 return 0; 1216 return 0;
1825} 1217}
1826 1218
1827#ifdef __KVM_HAVE_MSIX
1828static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1829 struct kvm_assigned_msix_nr *entry_nr)
1830{
1831 int r = 0;
1832 struct kvm_assigned_dev_kernel *adev;
1833
1834 mutex_lock(&kvm->lock);
1835
1836 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1837 entry_nr->assigned_dev_id);
1838 if (!adev) {
1839 r = -EINVAL;
1840 goto msix_nr_out;
1841 }
1842
1843 if (adev->entries_nr == 0) {
1844 adev->entries_nr = entry_nr->entry_nr;
1845 if (adev->entries_nr == 0 ||
1846 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1847 r = -EINVAL;
1848 goto msix_nr_out;
1849 }
1850
1851 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1852 entry_nr->entry_nr,
1853 GFP_KERNEL);
1854 if (!adev->host_msix_entries) {
1855 r = -ENOMEM;
1856 goto msix_nr_out;
1857 }
1858 adev->guest_msix_entries = kzalloc(
1859 sizeof(struct kvm_guest_msix_entry) *
1860 entry_nr->entry_nr, GFP_KERNEL);
1861 if (!adev->guest_msix_entries) {
1862 kfree(adev->host_msix_entries);
1863 r = -ENOMEM;
1864 goto msix_nr_out;
1865 }
1866 } else /* Not allowed set MSI-X number twice */
1867 r = -EINVAL;
1868msix_nr_out:
1869 mutex_unlock(&kvm->lock);
1870 return r;
1871}
1872
1873static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1874 struct kvm_assigned_msix_entry *entry)
1875{
1876 int r = 0, i;
1877 struct kvm_assigned_dev_kernel *adev;
1878
1879 mutex_lock(&kvm->lock);
1880
1881 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1882 entry->assigned_dev_id);
1883
1884 if (!adev) {
1885 r = -EINVAL;
1886 goto msix_entry_out;
1887 }
1888
1889 for (i = 0; i < adev->entries_nr; i++)
1890 if (adev->guest_msix_entries[i].vector == 0 ||
1891 adev->guest_msix_entries[i].entry == entry->entry) {
1892 adev->guest_msix_entries[i].entry = entry->entry;
1893 adev->guest_msix_entries[i].vector = entry->gsi;
1894 adev->host_msix_entries[i].entry = entry->entry;
1895 break;
1896 }
1897 if (i == adev->entries_nr) {
1898 r = -ENOSPC;
1899 goto msix_entry_out;
1900 }
1901
1902msix_entry_out:
1903 mutex_unlock(&kvm->lock);
1904
1905 return r;
1906}
1907#endif
1908
1909static long kvm_vcpu_ioctl(struct file *filp, 1219static long kvm_vcpu_ioctl(struct file *filp,
1910 unsigned int ioctl, unsigned long arg) 1220 unsigned int ioctl, unsigned long arg)
1911{ 1221{
@@ -2164,112 +1474,6 @@ static long kvm_vm_ioctl(struct file *filp,
2164 break; 1474 break;
2165 } 1475 }
2166#endif 1476#endif
2167#ifdef KVM_CAP_DEVICE_ASSIGNMENT
2168 case KVM_ASSIGN_PCI_DEVICE: {
2169 struct kvm_assigned_pci_dev assigned_dev;
2170
2171 r = -EFAULT;
2172 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2173 goto out;
2174 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2175 if (r)
2176 goto out;
2177 break;
2178 }
2179 case KVM_ASSIGN_IRQ: {
2180 r = -EOPNOTSUPP;
2181 break;
2182 }
2183#ifdef KVM_CAP_ASSIGN_DEV_IRQ
2184 case KVM_ASSIGN_DEV_IRQ: {
2185 struct kvm_assigned_irq assigned_irq;
2186
2187 r = -EFAULT;
2188 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2189 goto out;
2190 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2191 if (r)
2192 goto out;
2193 break;
2194 }
2195 case KVM_DEASSIGN_DEV_IRQ: {
2196 struct kvm_assigned_irq assigned_irq;
2197
2198 r = -EFAULT;
2199 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2200 goto out;
2201 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2202 if (r)
2203 goto out;
2204 break;
2205 }
2206#endif
2207#endif
2208#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2209 case KVM_DEASSIGN_PCI_DEVICE: {
2210 struct kvm_assigned_pci_dev assigned_dev;
2211
2212 r = -EFAULT;
2213 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2214 goto out;
2215 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2216 if (r)
2217 goto out;
2218 break;
2219 }
2220#endif
2221#ifdef KVM_CAP_IRQ_ROUTING
2222 case KVM_SET_GSI_ROUTING: {
2223 struct kvm_irq_routing routing;
2224 struct kvm_irq_routing __user *urouting;
2225 struct kvm_irq_routing_entry *entries;
2226
2227 r = -EFAULT;
2228 if (copy_from_user(&routing, argp, sizeof(routing)))
2229 goto out;
2230 r = -EINVAL;
2231 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2232 goto out;
2233 if (routing.flags)
2234 goto out;
2235 r = -ENOMEM;
2236 entries = vmalloc(routing.nr * sizeof(*entries));
2237 if (!entries)
2238 goto out;
2239 r = -EFAULT;
2240 urouting = argp;
2241 if (copy_from_user(entries, urouting->entries,
2242 routing.nr * sizeof(*entries)))
2243 goto out_free_irq_routing;
2244 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2245 routing.flags);
2246 out_free_irq_routing:
2247 vfree(entries);
2248 break;
2249 }
2250#endif /* KVM_CAP_IRQ_ROUTING */
2251#ifdef __KVM_HAVE_MSIX
2252 case KVM_ASSIGN_SET_MSIX_NR: {
2253 struct kvm_assigned_msix_nr entry_nr;
2254 r = -EFAULT;
2255 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2256 goto out;
2257 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2258 if (r)
2259 goto out;
2260 break;
2261 }
2262 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2263 struct kvm_assigned_msix_entry entry;
2264 r = -EFAULT;
2265 if (copy_from_user(&entry, argp, sizeof entry))
2266 goto out;
2267 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2268 if (r)
2269 goto out;
2270 break;
2271 }
2272#endif
2273 case KVM_IRQFD: { 1477 case KVM_IRQFD: {
2274 struct kvm_irqfd data; 1478 struct kvm_irqfd data;
2275 1479
@@ -2301,6 +1505,8 @@ static long kvm_vm_ioctl(struct file *filp,
2301#endif 1505#endif
2302 default: 1506 default:
2303 r = kvm_arch_vm_ioctl(filp, ioctl, arg); 1507 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1508 if (r == -ENOTTY)
1509 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2304 } 1510 }
2305out: 1511out:
2306 return r; 1512 return r;