KVM: introduce KVM_PFN_ERR_HWPOISON
[linux-3.10.git] / virt / kvm / assigned-dev.c
1 /*
2  * Kernel-based Virtual Machine - device assignment support
3  *
4  * Copyright (C) 2010 Red Hat, Inc. and/or its affiliates.
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 <linux/namei.h>
21 #include <linux/fs.h>
22 #include "irq.h"
23
24 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
25                                                       int assigned_dev_id)
26 {
27         struct list_head *ptr;
28         struct kvm_assigned_dev_kernel *match;
29
30         list_for_each(ptr, head) {
31                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
32                 if (match->assigned_dev_id == assigned_dev_id)
33                         return match;
34         }
35         return NULL;
36 }
37
38 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
39                                     *assigned_dev, int irq)
40 {
41         int i, index;
42         struct msix_entry *host_msix_entries;
43
44         host_msix_entries = assigned_dev->host_msix_entries;
45
46         index = -1;
47         for (i = 0; i < assigned_dev->entries_nr; i++)
48                 if (irq == host_msix_entries[i].vector) {
49                         index = i;
50                         break;
51                 }
52         if (index < 0)
53                 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
54
55         return index;
56 }
57
58 static irqreturn_t kvm_assigned_dev_intx(int irq, void *dev_id)
59 {
60         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
61         int ret;
62
63         spin_lock(&assigned_dev->intx_lock);
64         if (pci_check_and_mask_intx(assigned_dev->dev)) {
65                 assigned_dev->host_irq_disabled = true;
66                 ret = IRQ_WAKE_THREAD;
67         } else
68                 ret = IRQ_NONE;
69         spin_unlock(&assigned_dev->intx_lock);
70
71         return ret;
72 }
73
74 static void
75 kvm_assigned_dev_raise_guest_irq(struct kvm_assigned_dev_kernel *assigned_dev,
76                                  int vector)
77 {
78         if (unlikely(assigned_dev->irq_requested_type &
79                      KVM_DEV_IRQ_GUEST_INTX)) {
80                 spin_lock(&assigned_dev->intx_mask_lock);
81                 if (!(assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX))
82                         kvm_set_irq(assigned_dev->kvm,
83                                     assigned_dev->irq_source_id, vector, 1);
84                 spin_unlock(&assigned_dev->intx_mask_lock);
85         } else
86                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
87                             vector, 1);
88 }
89
90 static irqreturn_t kvm_assigned_dev_thread_intx(int irq, void *dev_id)
91 {
92         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
93
94         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
95                 spin_lock_irq(&assigned_dev->intx_lock);
96                 disable_irq_nosync(irq);
97                 assigned_dev->host_irq_disabled = true;
98                 spin_unlock_irq(&assigned_dev->intx_lock);
99         }
100
101         kvm_assigned_dev_raise_guest_irq(assigned_dev,
102                                          assigned_dev->guest_irq);
103
104         return IRQ_HANDLED;
105 }
106
107 #ifdef __KVM_HAVE_MSI
108 static irqreturn_t kvm_assigned_dev_thread_msi(int irq, void *dev_id)
109 {
110         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
111
112         kvm_assigned_dev_raise_guest_irq(assigned_dev,
113                                          assigned_dev->guest_irq);
114
115         return IRQ_HANDLED;
116 }
117 #endif
118
119 #ifdef __KVM_HAVE_MSIX
120 static irqreturn_t kvm_assigned_dev_thread_msix(int irq, void *dev_id)
121 {
122         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
123         int index = find_index_from_host_irq(assigned_dev, irq);
124         u32 vector;
125
126         if (index >= 0) {
127                 vector = assigned_dev->guest_msix_entries[index].vector;
128                 kvm_assigned_dev_raise_guest_irq(assigned_dev, vector);
129         }
130
131         return IRQ_HANDLED;
132 }
133 #endif
134
135 /* Ack the irq line for an assigned device */
136 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
137 {
138         struct kvm_assigned_dev_kernel *dev =
139                 container_of(kian, struct kvm_assigned_dev_kernel,
140                              ack_notifier);
141
142         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
143
144         spin_lock(&dev->intx_mask_lock);
145
146         if (!(dev->flags & KVM_DEV_ASSIGN_MASK_INTX)) {
147                 bool reassert = false;
148
149                 spin_lock_irq(&dev->intx_lock);
150                 /*
151                  * The guest IRQ may be shared so this ack can come from an
152                  * IRQ for another guest device.
153                  */
154                 if (dev->host_irq_disabled) {
155                         if (!(dev->flags & KVM_DEV_ASSIGN_PCI_2_3))
156                                 enable_irq(dev->host_irq);
157                         else if (!pci_check_and_unmask_intx(dev->dev))
158                                 reassert = true;
159                         dev->host_irq_disabled = reassert;
160                 }
161                 spin_unlock_irq(&dev->intx_lock);
162
163                 if (reassert)
164                         kvm_set_irq(dev->kvm, dev->irq_source_id,
165                                     dev->guest_irq, 1);
166         }
167
168         spin_unlock(&dev->intx_mask_lock);
169 }
170
171 static void deassign_guest_irq(struct kvm *kvm,
172                                struct kvm_assigned_dev_kernel *assigned_dev)
173 {
174         if (assigned_dev->ack_notifier.gsi != -1)
175                 kvm_unregister_irq_ack_notifier(kvm,
176                                                 &assigned_dev->ack_notifier);
177
178         kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
179                     assigned_dev->guest_irq, 0);
180
181         if (assigned_dev->irq_source_id != -1)
182                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
183         assigned_dev->irq_source_id = -1;
184         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
185 }
186
187 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
188 static void deassign_host_irq(struct kvm *kvm,
189                               struct kvm_assigned_dev_kernel *assigned_dev)
190 {
191         /*
192          * We disable irq here to prevent further events.
193          *
194          * Notice this maybe result in nested disable if the interrupt type is
195          * INTx, but it's OK for we are going to free it.
196          *
197          * If this function is a part of VM destroy, please ensure that till
198          * now, the kvm state is still legal for probably we also have to wait
199          * on a currently running IRQ handler.
200          */
201         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
202                 int i;
203                 for (i = 0; i < assigned_dev->entries_nr; i++)
204                         disable_irq(assigned_dev->host_msix_entries[i].vector);
205
206                 for (i = 0; i < assigned_dev->entries_nr; i++)
207                         free_irq(assigned_dev->host_msix_entries[i].vector,
208                                  assigned_dev);
209
210                 assigned_dev->entries_nr = 0;
211                 kfree(assigned_dev->host_msix_entries);
212                 kfree(assigned_dev->guest_msix_entries);
213                 pci_disable_msix(assigned_dev->dev);
214         } else {
215                 /* Deal with MSI and INTx */
216                 if ((assigned_dev->irq_requested_type &
217                      KVM_DEV_IRQ_HOST_INTX) &&
218                     (assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
219                         spin_lock_irq(&assigned_dev->intx_lock);
220                         pci_intx(assigned_dev->dev, false);
221                         spin_unlock_irq(&assigned_dev->intx_lock);
222                         synchronize_irq(assigned_dev->host_irq);
223                 } else
224                         disable_irq(assigned_dev->host_irq);
225
226                 free_irq(assigned_dev->host_irq, assigned_dev);
227
228                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
229                         pci_disable_msi(assigned_dev->dev);
230         }
231
232         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
233 }
234
235 static int kvm_deassign_irq(struct kvm *kvm,
236                             struct kvm_assigned_dev_kernel *assigned_dev,
237                             unsigned long irq_requested_type)
238 {
239         unsigned long guest_irq_type, host_irq_type;
240
241         if (!irqchip_in_kernel(kvm))
242                 return -EINVAL;
243         /* no irq assignment to deassign */
244         if (!assigned_dev->irq_requested_type)
245                 return -ENXIO;
246
247         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
248         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
249
250         if (host_irq_type)
251                 deassign_host_irq(kvm, assigned_dev);
252         if (guest_irq_type)
253                 deassign_guest_irq(kvm, assigned_dev);
254
255         return 0;
256 }
257
258 static void kvm_free_assigned_irq(struct kvm *kvm,
259                                   struct kvm_assigned_dev_kernel *assigned_dev)
260 {
261         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
262 }
263
264 static void kvm_free_assigned_device(struct kvm *kvm,
265                                      struct kvm_assigned_dev_kernel
266                                      *assigned_dev)
267 {
268         kvm_free_assigned_irq(kvm, assigned_dev);
269
270         pci_reset_function(assigned_dev->dev);
271         if (pci_load_and_free_saved_state(assigned_dev->dev,
272                                           &assigned_dev->pci_saved_state))
273                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
274                        __func__, dev_name(&assigned_dev->dev->dev));
275         else
276                 pci_restore_state(assigned_dev->dev);
277
278         assigned_dev->dev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
279
280         pci_release_regions(assigned_dev->dev);
281         pci_disable_device(assigned_dev->dev);
282         pci_dev_put(assigned_dev->dev);
283
284         list_del(&assigned_dev->list);
285         kfree(assigned_dev);
286 }
287
288 void kvm_free_all_assigned_devices(struct kvm *kvm)
289 {
290         struct list_head *ptr, *ptr2;
291         struct kvm_assigned_dev_kernel *assigned_dev;
292
293         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
294                 assigned_dev = list_entry(ptr,
295                                           struct kvm_assigned_dev_kernel,
296                                           list);
297
298                 kvm_free_assigned_device(kvm, assigned_dev);
299         }
300 }
301
302 static int assigned_device_enable_host_intx(struct kvm *kvm,
303                                             struct kvm_assigned_dev_kernel *dev)
304 {
305         irq_handler_t irq_handler;
306         unsigned long flags;
307
308         dev->host_irq = dev->dev->irq;
309
310         /*
311          * We can only share the IRQ line with other host devices if we are
312          * able to disable the IRQ source at device-level - independently of
313          * the guest driver. Otherwise host devices may suffer from unbounded
314          * IRQ latencies when the guest keeps the line asserted.
315          */
316         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
317                 irq_handler = kvm_assigned_dev_intx;
318                 flags = IRQF_SHARED;
319         } else {
320                 irq_handler = NULL;
321                 flags = IRQF_ONESHOT;
322         }
323         if (request_threaded_irq(dev->host_irq, irq_handler,
324                                  kvm_assigned_dev_thread_intx, flags,
325                                  dev->irq_name, dev))
326                 return -EIO;
327
328         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
329                 spin_lock_irq(&dev->intx_lock);
330                 pci_intx(dev->dev, true);
331                 spin_unlock_irq(&dev->intx_lock);
332         }
333         return 0;
334 }
335
336 #ifdef __KVM_HAVE_MSI
337 static irqreturn_t kvm_assigned_dev_msi(int irq, void *dev_id)
338 {
339         return IRQ_WAKE_THREAD;
340 }
341
342 static int assigned_device_enable_host_msi(struct kvm *kvm,
343                                            struct kvm_assigned_dev_kernel *dev)
344 {
345         int r;
346
347         if (!dev->dev->msi_enabled) {
348                 r = pci_enable_msi(dev->dev);
349                 if (r)
350                         return r;
351         }
352
353         dev->host_irq = dev->dev->irq;
354         if (request_threaded_irq(dev->host_irq, kvm_assigned_dev_msi,
355                                  kvm_assigned_dev_thread_msi, 0,
356                                  dev->irq_name, dev)) {
357                 pci_disable_msi(dev->dev);
358                 return -EIO;
359         }
360
361         return 0;
362 }
363 #endif
364
365 #ifdef __KVM_HAVE_MSIX
366 static irqreturn_t kvm_assigned_dev_msix(int irq, void *dev_id)
367 {
368         return IRQ_WAKE_THREAD;
369 }
370
371 static int assigned_device_enable_host_msix(struct kvm *kvm,
372                                             struct kvm_assigned_dev_kernel *dev)
373 {
374         int i, r = -EINVAL;
375
376         /* host_msix_entries and guest_msix_entries should have been
377          * initialized */
378         if (dev->entries_nr == 0)
379                 return r;
380
381         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
382         if (r)
383                 return r;
384
385         for (i = 0; i < dev->entries_nr; i++) {
386                 r = request_threaded_irq(dev->host_msix_entries[i].vector,
387                                          kvm_assigned_dev_msix,
388                                          kvm_assigned_dev_thread_msix,
389                                          0, dev->irq_name, dev);
390                 if (r)
391                         goto err;
392         }
393
394         return 0;
395 err:
396         for (i -= 1; i >= 0; i--)
397                 free_irq(dev->host_msix_entries[i].vector, dev);
398         pci_disable_msix(dev->dev);
399         return r;
400 }
401
402 #endif
403
404 static int assigned_device_enable_guest_intx(struct kvm *kvm,
405                                 struct kvm_assigned_dev_kernel *dev,
406                                 struct kvm_assigned_irq *irq)
407 {
408         dev->guest_irq = irq->guest_irq;
409         dev->ack_notifier.gsi = irq->guest_irq;
410         return 0;
411 }
412
413 #ifdef __KVM_HAVE_MSI
414 static int assigned_device_enable_guest_msi(struct kvm *kvm,
415                         struct kvm_assigned_dev_kernel *dev,
416                         struct kvm_assigned_irq *irq)
417 {
418         dev->guest_irq = irq->guest_irq;
419         dev->ack_notifier.gsi = -1;
420         return 0;
421 }
422 #endif
423
424 #ifdef __KVM_HAVE_MSIX
425 static int assigned_device_enable_guest_msix(struct kvm *kvm,
426                         struct kvm_assigned_dev_kernel *dev,
427                         struct kvm_assigned_irq *irq)
428 {
429         dev->guest_irq = irq->guest_irq;
430         dev->ack_notifier.gsi = -1;
431         return 0;
432 }
433 #endif
434
435 static 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         snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
445                  pci_name(dev->dev));
446
447         switch (host_irq_type) {
448         case KVM_DEV_IRQ_HOST_INTX:
449                 r = assigned_device_enable_host_intx(kvm, dev);
450                 break;
451 #ifdef __KVM_HAVE_MSI
452         case KVM_DEV_IRQ_HOST_MSI:
453                 r = assigned_device_enable_host_msi(kvm, dev);
454                 break;
455 #endif
456 #ifdef __KVM_HAVE_MSIX
457         case KVM_DEV_IRQ_HOST_MSIX:
458                 r = assigned_device_enable_host_msix(kvm, dev);
459                 break;
460 #endif
461         default:
462                 r = -EINVAL;
463         }
464         dev->host_irq_disabled = false;
465
466         if (!r)
467                 dev->irq_requested_type |= host_irq_type;
468
469         return r;
470 }
471
472 static int assign_guest_irq(struct kvm *kvm,
473                             struct kvm_assigned_dev_kernel *dev,
474                             struct kvm_assigned_irq *irq,
475                             unsigned long guest_irq_type)
476 {
477         int id;
478         int r = -EEXIST;
479
480         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
481                 return r;
482
483         id = kvm_request_irq_source_id(kvm);
484         if (id < 0)
485                 return id;
486
487         dev->irq_source_id = id;
488
489         switch (guest_irq_type) {
490         case KVM_DEV_IRQ_GUEST_INTX:
491                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
492                 break;
493 #ifdef __KVM_HAVE_MSI
494         case KVM_DEV_IRQ_GUEST_MSI:
495                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
496                 break;
497 #endif
498 #ifdef __KVM_HAVE_MSIX
499         case KVM_DEV_IRQ_GUEST_MSIX:
500                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
501                 break;
502 #endif
503         default:
504                 r = -EINVAL;
505         }
506
507         if (!r) {
508                 dev->irq_requested_type |= guest_irq_type;
509                 if (dev->ack_notifier.gsi != -1)
510                         kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
511         } else
512                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
513
514         return r;
515 }
516
517 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
518 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
519                                    struct kvm_assigned_irq *assigned_irq)
520 {
521         int r = -EINVAL;
522         struct kvm_assigned_dev_kernel *match;
523         unsigned long host_irq_type, guest_irq_type;
524
525         if (!irqchip_in_kernel(kvm))
526                 return r;
527
528         mutex_lock(&kvm->lock);
529         r = -ENODEV;
530         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
531                                       assigned_irq->assigned_dev_id);
532         if (!match)
533                 goto out;
534
535         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
536         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
537
538         r = -EINVAL;
539         /* can only assign one type at a time */
540         if (hweight_long(host_irq_type) > 1)
541                 goto out;
542         if (hweight_long(guest_irq_type) > 1)
543                 goto out;
544         if (host_irq_type == 0 && guest_irq_type == 0)
545                 goto out;
546
547         r = 0;
548         if (host_irq_type)
549                 r = assign_host_irq(kvm, match, host_irq_type);
550         if (r)
551                 goto out;
552
553         if (guest_irq_type)
554                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
555 out:
556         mutex_unlock(&kvm->lock);
557         return r;
558 }
559
560 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
561                                          struct kvm_assigned_irq
562                                          *assigned_irq)
563 {
564         int r = -ENODEV;
565         struct kvm_assigned_dev_kernel *match;
566         unsigned long irq_type;
567
568         mutex_lock(&kvm->lock);
569
570         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
571                                       assigned_irq->assigned_dev_id);
572         if (!match)
573                 goto out;
574
575         irq_type = assigned_irq->flags & (KVM_DEV_IRQ_HOST_MASK |
576                                           KVM_DEV_IRQ_GUEST_MASK);
577         r = kvm_deassign_irq(kvm, match, irq_type);
578 out:
579         mutex_unlock(&kvm->lock);
580         return r;
581 }
582
583 /*
584  * We want to test whether the caller has been granted permissions to
585  * use this device.  To be able to configure and control the device,
586  * the user needs access to PCI configuration space and BAR resources.
587  * These are accessed through PCI sysfs.  PCI config space is often
588  * passed to the process calling this ioctl via file descriptor, so we
589  * can't rely on access to that file.  We can check for permissions
590  * on each of the BAR resource files, which is a pretty clear
591  * indicator that the user has been granted access to the device.
592  */
593 static int probe_sysfs_permissions(struct pci_dev *dev)
594 {
595 #ifdef CONFIG_SYSFS
596         int i;
597         bool bar_found = false;
598
599         for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++) {
600                 char *kpath, *syspath;
601                 struct path path;
602                 struct inode *inode;
603                 int r;
604
605                 if (!pci_resource_len(dev, i))
606                         continue;
607
608                 kpath = kobject_get_path(&dev->dev.kobj, GFP_KERNEL);
609                 if (!kpath)
610                         return -ENOMEM;
611
612                 /* Per sysfs-rules, sysfs is always at /sys */
613                 syspath = kasprintf(GFP_KERNEL, "/sys%s/resource%d", kpath, i);
614                 kfree(kpath);
615                 if (!syspath)
616                         return -ENOMEM;
617
618                 r = kern_path(syspath, LOOKUP_FOLLOW, &path);
619                 kfree(syspath);
620                 if (r)
621                         return r;
622
623                 inode = path.dentry->d_inode;
624
625                 r = inode_permission(inode, MAY_READ | MAY_WRITE | MAY_ACCESS);
626                 path_put(&path);
627                 if (r)
628                         return r;
629
630                 bar_found = true;
631         }
632
633         /* If no resources, probably something special */
634         if (!bar_found)
635                 return -EPERM;
636
637         return 0;
638 #else
639         return -EINVAL; /* No way to control the device without sysfs */
640 #endif
641 }
642
643 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
644                                       struct kvm_assigned_pci_dev *assigned_dev)
645 {
646         int r = 0, idx;
647         struct kvm_assigned_dev_kernel *match;
648         struct pci_dev *dev;
649
650         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU))
651                 return -EINVAL;
652
653         mutex_lock(&kvm->lock);
654         idx = srcu_read_lock(&kvm->srcu);
655
656         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
657                                       assigned_dev->assigned_dev_id);
658         if (match) {
659                 /* device already assigned */
660                 r = -EEXIST;
661                 goto out;
662         }
663
664         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
665         if (match == NULL) {
666                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
667                        __func__);
668                 r = -ENOMEM;
669                 goto out;
670         }
671         dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
672                                    assigned_dev->busnr,
673                                    assigned_dev->devfn);
674         if (!dev) {
675                 printk(KERN_INFO "%s: host device not found\n", __func__);
676                 r = -EINVAL;
677                 goto out_free;
678         }
679
680         /* Don't allow bridges to be assigned */
681         if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) {
682                 r = -EPERM;
683                 goto out_put;
684         }
685
686         r = probe_sysfs_permissions(dev);
687         if (r)
688                 goto out_put;
689
690         if (pci_enable_device(dev)) {
691                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
692                 r = -EBUSY;
693                 goto out_put;
694         }
695         r = pci_request_regions(dev, "kvm_assigned_device");
696         if (r) {
697                 printk(KERN_INFO "%s: Could not get access to device regions\n",
698                        __func__);
699                 goto out_disable;
700         }
701
702         pci_reset_function(dev);
703         pci_save_state(dev);
704         match->pci_saved_state = pci_store_saved_state(dev);
705         if (!match->pci_saved_state)
706                 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
707                        __func__, dev_name(&dev->dev));
708
709         if (!pci_intx_mask_supported(dev))
710                 assigned_dev->flags &= ~KVM_DEV_ASSIGN_PCI_2_3;
711
712         match->assigned_dev_id = assigned_dev->assigned_dev_id;
713         match->host_segnr = assigned_dev->segnr;
714         match->host_busnr = assigned_dev->busnr;
715         match->host_devfn = assigned_dev->devfn;
716         match->flags = assigned_dev->flags;
717         match->dev = dev;
718         spin_lock_init(&match->intx_lock);
719         spin_lock_init(&match->intx_mask_lock);
720         match->irq_source_id = -1;
721         match->kvm = kvm;
722         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
723
724         list_add(&match->list, &kvm->arch.assigned_dev_head);
725
726         if (!kvm->arch.iommu_domain) {
727                 r = kvm_iommu_map_guest(kvm);
728                 if (r)
729                         goto out_list_del;
730         }
731         r = kvm_assign_device(kvm, match);
732         if (r)
733                 goto out_list_del;
734
735 out:
736         srcu_read_unlock(&kvm->srcu, idx);
737         mutex_unlock(&kvm->lock);
738         return r;
739 out_list_del:
740         if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
741                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
742                        __func__, dev_name(&dev->dev));
743         list_del(&match->list);
744         pci_release_regions(dev);
745 out_disable:
746         pci_disable_device(dev);
747 out_put:
748         pci_dev_put(dev);
749 out_free:
750         kfree(match);
751         srcu_read_unlock(&kvm->srcu, idx);
752         mutex_unlock(&kvm->lock);
753         return r;
754 }
755
756 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
757                 struct kvm_assigned_pci_dev *assigned_dev)
758 {
759         int r = 0;
760         struct kvm_assigned_dev_kernel *match;
761
762         mutex_lock(&kvm->lock);
763
764         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
765                                       assigned_dev->assigned_dev_id);
766         if (!match) {
767                 printk(KERN_INFO "%s: device hasn't been assigned before, "
768                   "so cannot be deassigned\n", __func__);
769                 r = -EINVAL;
770                 goto out;
771         }
772
773         kvm_deassign_device(kvm, match);
774
775         kvm_free_assigned_device(kvm, match);
776
777 out:
778         mutex_unlock(&kvm->lock);
779         return r;
780 }
781
782
783 #ifdef __KVM_HAVE_MSIX
784 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
785                                     struct kvm_assigned_msix_nr *entry_nr)
786 {
787         int r = 0;
788         struct kvm_assigned_dev_kernel *adev;
789
790         mutex_lock(&kvm->lock);
791
792         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
793                                       entry_nr->assigned_dev_id);
794         if (!adev) {
795                 r = -EINVAL;
796                 goto msix_nr_out;
797         }
798
799         if (adev->entries_nr == 0) {
800                 adev->entries_nr = entry_nr->entry_nr;
801                 if (adev->entries_nr == 0 ||
802                     adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
803                         r = -EINVAL;
804                         goto msix_nr_out;
805                 }
806
807                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
808                                                 entry_nr->entry_nr,
809                                                 GFP_KERNEL);
810                 if (!adev->host_msix_entries) {
811                         r = -ENOMEM;
812                         goto msix_nr_out;
813                 }
814                 adev->guest_msix_entries =
815                         kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
816                                 GFP_KERNEL);
817                 if (!adev->guest_msix_entries) {
818                         kfree(adev->host_msix_entries);
819                         r = -ENOMEM;
820                         goto msix_nr_out;
821                 }
822         } else /* Not allowed set MSI-X number twice */
823                 r = -EINVAL;
824 msix_nr_out:
825         mutex_unlock(&kvm->lock);
826         return r;
827 }
828
829 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
830                                        struct kvm_assigned_msix_entry *entry)
831 {
832         int r = 0, i;
833         struct kvm_assigned_dev_kernel *adev;
834
835         mutex_lock(&kvm->lock);
836
837         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
838                                       entry->assigned_dev_id);
839
840         if (!adev) {
841                 r = -EINVAL;
842                 goto msix_entry_out;
843         }
844
845         for (i = 0; i < adev->entries_nr; i++)
846                 if (adev->guest_msix_entries[i].vector == 0 ||
847                     adev->guest_msix_entries[i].entry == entry->entry) {
848                         adev->guest_msix_entries[i].entry = entry->entry;
849                         adev->guest_msix_entries[i].vector = entry->gsi;
850                         adev->host_msix_entries[i].entry = entry->entry;
851                         break;
852                 }
853         if (i == adev->entries_nr) {
854                 r = -ENOSPC;
855                 goto msix_entry_out;
856         }
857
858 msix_entry_out:
859         mutex_unlock(&kvm->lock);
860
861         return r;
862 }
863 #endif
864
865 static int kvm_vm_ioctl_set_pci_irq_mask(struct kvm *kvm,
866                 struct kvm_assigned_pci_dev *assigned_dev)
867 {
868         int r = 0;
869         struct kvm_assigned_dev_kernel *match;
870
871         mutex_lock(&kvm->lock);
872
873         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
874                                       assigned_dev->assigned_dev_id);
875         if (!match) {
876                 r = -ENODEV;
877                 goto out;
878         }
879
880         spin_lock(&match->intx_mask_lock);
881
882         match->flags &= ~KVM_DEV_ASSIGN_MASK_INTX;
883         match->flags |= assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX;
884
885         if (match->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
886                 if (assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX) {
887                         kvm_set_irq(match->kvm, match->irq_source_id,
888                                     match->guest_irq, 0);
889                         /*
890                          * Masking at hardware-level is performed on demand,
891                          * i.e. when an IRQ actually arrives at the host.
892                          */
893                 } else if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
894                         /*
895                          * Unmask the IRQ line if required. Unmasking at
896                          * device level will be performed by user space.
897                          */
898                         spin_lock_irq(&match->intx_lock);
899                         if (match->host_irq_disabled) {
900                                 enable_irq(match->host_irq);
901                                 match->host_irq_disabled = false;
902                         }
903                         spin_unlock_irq(&match->intx_lock);
904                 }
905         }
906
907         spin_unlock(&match->intx_mask_lock);
908
909 out:
910         mutex_unlock(&kvm->lock);
911         return r;
912 }
913
914 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
915                                   unsigned long arg)
916 {
917         void __user *argp = (void __user *)arg;
918         int r;
919
920         switch (ioctl) {
921         case KVM_ASSIGN_PCI_DEVICE: {
922                 struct kvm_assigned_pci_dev assigned_dev;
923
924                 r = -EFAULT;
925                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
926                         goto out;
927                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
928                 if (r)
929                         goto out;
930                 break;
931         }
932         case KVM_ASSIGN_IRQ: {
933                 r = -EOPNOTSUPP;
934                 break;
935         }
936         case KVM_ASSIGN_DEV_IRQ: {
937                 struct kvm_assigned_irq assigned_irq;
938
939                 r = -EFAULT;
940                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
941                         goto out;
942                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
943                 if (r)
944                         goto out;
945                 break;
946         }
947         case KVM_DEASSIGN_DEV_IRQ: {
948                 struct kvm_assigned_irq assigned_irq;
949
950                 r = -EFAULT;
951                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
952                         goto out;
953                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
954                 if (r)
955                         goto out;
956                 break;
957         }
958         case KVM_DEASSIGN_PCI_DEVICE: {
959                 struct kvm_assigned_pci_dev assigned_dev;
960
961                 r = -EFAULT;
962                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
963                         goto out;
964                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
965                 if (r)
966                         goto out;
967                 break;
968         }
969 #ifdef KVM_CAP_IRQ_ROUTING
970         case KVM_SET_GSI_ROUTING: {
971                 struct kvm_irq_routing routing;
972                 struct kvm_irq_routing __user *urouting;
973                 struct kvm_irq_routing_entry *entries;
974
975                 r = -EFAULT;
976                 if (copy_from_user(&routing, argp, sizeof(routing)))
977                         goto out;
978                 r = -EINVAL;
979                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
980                         goto out;
981                 if (routing.flags)
982                         goto out;
983                 r = -ENOMEM;
984                 entries = vmalloc(routing.nr * sizeof(*entries));
985                 if (!entries)
986                         goto out;
987                 r = -EFAULT;
988                 urouting = argp;
989                 if (copy_from_user(entries, urouting->entries,
990                                    routing.nr * sizeof(*entries)))
991                         goto out_free_irq_routing;
992                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
993                                         routing.flags);
994         out_free_irq_routing:
995                 vfree(entries);
996                 break;
997         }
998 #endif /* KVM_CAP_IRQ_ROUTING */
999 #ifdef __KVM_HAVE_MSIX
1000         case KVM_ASSIGN_SET_MSIX_NR: {
1001                 struct kvm_assigned_msix_nr entry_nr;
1002                 r = -EFAULT;
1003                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
1004                         goto out;
1005                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
1006                 if (r)
1007                         goto out;
1008                 break;
1009         }
1010         case KVM_ASSIGN_SET_MSIX_ENTRY: {
1011                 struct kvm_assigned_msix_entry entry;
1012                 r = -EFAULT;
1013                 if (copy_from_user(&entry, argp, sizeof entry))
1014                         goto out;
1015                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
1016                 if (r)
1017                         goto out;
1018                 break;
1019         }
1020 #endif
1021         case KVM_ASSIGN_SET_INTX_MASK: {
1022                 struct kvm_assigned_pci_dev assigned_dev;
1023
1024                 r = -EFAULT;
1025                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1026                         goto out;
1027                 r = kvm_vm_ioctl_set_pci_irq_mask(kvm, &assigned_dev);
1028                 break;
1029         }
1030         default:
1031                 r = -ENOTTY;
1032                 break;
1033         }
1034 out:
1035         return r;
1036 }