KVM: x86: Clear apic tsc-deadline after deadline
[sfrench/cifs-2.6.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                                     false);
85                 spin_unlock(&assigned_dev->intx_mask_lock);
86         } else
87                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
88                             vector, 1, false);
89 }
90
91 static irqreturn_t kvm_assigned_dev_thread_intx(int irq, void *dev_id)
92 {
93         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
94
95         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
96                 spin_lock_irq(&assigned_dev->intx_lock);
97                 disable_irq_nosync(irq);
98                 assigned_dev->host_irq_disabled = true;
99                 spin_unlock_irq(&assigned_dev->intx_lock);
100         }
101
102         kvm_assigned_dev_raise_guest_irq(assigned_dev,
103                                          assigned_dev->guest_irq);
104
105         return IRQ_HANDLED;
106 }
107
108 #ifdef __KVM_HAVE_MSI
109 static irqreturn_t kvm_assigned_dev_msi(int irq, void *dev_id)
110 {
111         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
112         int ret = kvm_set_irq_inatomic(assigned_dev->kvm,
113                                        assigned_dev->irq_source_id,
114                                        assigned_dev->guest_irq, 1);
115         return unlikely(ret == -EWOULDBLOCK) ? IRQ_WAKE_THREAD : IRQ_HANDLED;
116 }
117
118 static irqreturn_t kvm_assigned_dev_thread_msi(int irq, void *dev_id)
119 {
120         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
121
122         kvm_assigned_dev_raise_guest_irq(assigned_dev,
123                                          assigned_dev->guest_irq);
124
125         return IRQ_HANDLED;
126 }
127 #endif
128
129 #ifdef __KVM_HAVE_MSIX
130 static irqreturn_t kvm_assigned_dev_msix(int irq, void *dev_id)
131 {
132         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
133         int index = find_index_from_host_irq(assigned_dev, irq);
134         u32 vector;
135         int ret = 0;
136
137         if (index >= 0) {
138                 vector = assigned_dev->guest_msix_entries[index].vector;
139                 ret = kvm_set_irq_inatomic(assigned_dev->kvm,
140                                            assigned_dev->irq_source_id,
141                                            vector, 1);
142         }
143
144         return unlikely(ret == -EWOULDBLOCK) ? IRQ_WAKE_THREAD : IRQ_HANDLED;
145 }
146
147 static irqreturn_t kvm_assigned_dev_thread_msix(int irq, void *dev_id)
148 {
149         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
150         int index = find_index_from_host_irq(assigned_dev, irq);
151         u32 vector;
152
153         if (index >= 0) {
154                 vector = assigned_dev->guest_msix_entries[index].vector;
155                 kvm_assigned_dev_raise_guest_irq(assigned_dev, vector);
156         }
157
158         return IRQ_HANDLED;
159 }
160 #endif
161
162 /* Ack the irq line for an assigned device */
163 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
164 {
165         struct kvm_assigned_dev_kernel *dev =
166                 container_of(kian, struct kvm_assigned_dev_kernel,
167                              ack_notifier);
168
169         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0, false);
170
171         spin_lock(&dev->intx_mask_lock);
172
173         if (!(dev->flags & KVM_DEV_ASSIGN_MASK_INTX)) {
174                 bool reassert = false;
175
176                 spin_lock_irq(&dev->intx_lock);
177                 /*
178                  * The guest IRQ may be shared so this ack can come from an
179                  * IRQ for another guest device.
180                  */
181                 if (dev->host_irq_disabled) {
182                         if (!(dev->flags & KVM_DEV_ASSIGN_PCI_2_3))
183                                 enable_irq(dev->host_irq);
184                         else if (!pci_check_and_unmask_intx(dev->dev))
185                                 reassert = true;
186                         dev->host_irq_disabled = reassert;
187                 }
188                 spin_unlock_irq(&dev->intx_lock);
189
190                 if (reassert)
191                         kvm_set_irq(dev->kvm, dev->irq_source_id,
192                                     dev->guest_irq, 1, false);
193         }
194
195         spin_unlock(&dev->intx_mask_lock);
196 }
197
198 static void deassign_guest_irq(struct kvm *kvm,
199                                struct kvm_assigned_dev_kernel *assigned_dev)
200 {
201         if (assigned_dev->ack_notifier.gsi != -1)
202                 kvm_unregister_irq_ack_notifier(kvm,
203                                                 &assigned_dev->ack_notifier);
204
205         kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
206                     assigned_dev->guest_irq, 0, false);
207
208         if (assigned_dev->irq_source_id != -1)
209                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
210         assigned_dev->irq_source_id = -1;
211         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
212 }
213
214 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
215 static void deassign_host_irq(struct kvm *kvm,
216                               struct kvm_assigned_dev_kernel *assigned_dev)
217 {
218         /*
219          * We disable irq here to prevent further events.
220          *
221          * Notice this maybe result in nested disable if the interrupt type is
222          * INTx, but it's OK for we are going to free it.
223          *
224          * If this function is a part of VM destroy, please ensure that till
225          * now, the kvm state is still legal for probably we also have to wait
226          * on a currently running IRQ handler.
227          */
228         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
229                 int i;
230                 for (i = 0; i < assigned_dev->entries_nr; i++)
231                         disable_irq(assigned_dev->host_msix_entries[i].vector);
232
233                 for (i = 0; i < assigned_dev->entries_nr; i++)
234                         free_irq(assigned_dev->host_msix_entries[i].vector,
235                                  assigned_dev);
236
237                 assigned_dev->entries_nr = 0;
238                 kfree(assigned_dev->host_msix_entries);
239                 kfree(assigned_dev->guest_msix_entries);
240                 pci_disable_msix(assigned_dev->dev);
241         } else {
242                 /* Deal with MSI and INTx */
243                 if ((assigned_dev->irq_requested_type &
244                      KVM_DEV_IRQ_HOST_INTX) &&
245                     (assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
246                         spin_lock_irq(&assigned_dev->intx_lock);
247                         pci_intx(assigned_dev->dev, false);
248                         spin_unlock_irq(&assigned_dev->intx_lock);
249                         synchronize_irq(assigned_dev->host_irq);
250                 } else
251                         disable_irq(assigned_dev->host_irq);
252
253                 free_irq(assigned_dev->host_irq, assigned_dev);
254
255                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
256                         pci_disable_msi(assigned_dev->dev);
257         }
258
259         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
260 }
261
262 static int kvm_deassign_irq(struct kvm *kvm,
263                             struct kvm_assigned_dev_kernel *assigned_dev,
264                             unsigned long irq_requested_type)
265 {
266         unsigned long guest_irq_type, host_irq_type;
267
268         if (!irqchip_in_kernel(kvm))
269                 return -EINVAL;
270         /* no irq assignment to deassign */
271         if (!assigned_dev->irq_requested_type)
272                 return -ENXIO;
273
274         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
275         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
276
277         if (host_irq_type)
278                 deassign_host_irq(kvm, assigned_dev);
279         if (guest_irq_type)
280                 deassign_guest_irq(kvm, assigned_dev);
281
282         return 0;
283 }
284
285 static void kvm_free_assigned_irq(struct kvm *kvm,
286                                   struct kvm_assigned_dev_kernel *assigned_dev)
287 {
288         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
289 }
290
291 static void kvm_free_assigned_device(struct kvm *kvm,
292                                      struct kvm_assigned_dev_kernel
293                                      *assigned_dev)
294 {
295         kvm_free_assigned_irq(kvm, assigned_dev);
296
297         pci_reset_function(assigned_dev->dev);
298         if (pci_load_and_free_saved_state(assigned_dev->dev,
299                                           &assigned_dev->pci_saved_state))
300                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
301                        __func__, dev_name(&assigned_dev->dev->dev));
302         else
303                 pci_restore_state(assigned_dev->dev);
304
305         assigned_dev->dev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
306
307         pci_release_regions(assigned_dev->dev);
308         pci_disable_device(assigned_dev->dev);
309         pci_dev_put(assigned_dev->dev);
310
311         list_del(&assigned_dev->list);
312         kfree(assigned_dev);
313 }
314
315 void kvm_free_all_assigned_devices(struct kvm *kvm)
316 {
317         struct list_head *ptr, *ptr2;
318         struct kvm_assigned_dev_kernel *assigned_dev;
319
320         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
321                 assigned_dev = list_entry(ptr,
322                                           struct kvm_assigned_dev_kernel,
323                                           list);
324
325                 kvm_free_assigned_device(kvm, assigned_dev);
326         }
327 }
328
329 static int assigned_device_enable_host_intx(struct kvm *kvm,
330                                             struct kvm_assigned_dev_kernel *dev)
331 {
332         irq_handler_t irq_handler;
333         unsigned long flags;
334
335         dev->host_irq = dev->dev->irq;
336
337         /*
338          * We can only share the IRQ line with other host devices if we are
339          * able to disable the IRQ source at device-level - independently of
340          * the guest driver. Otherwise host devices may suffer from unbounded
341          * IRQ latencies when the guest keeps the line asserted.
342          */
343         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
344                 irq_handler = kvm_assigned_dev_intx;
345                 flags = IRQF_SHARED;
346         } else {
347                 irq_handler = NULL;
348                 flags = IRQF_ONESHOT;
349         }
350         if (request_threaded_irq(dev->host_irq, irq_handler,
351                                  kvm_assigned_dev_thread_intx, flags,
352                                  dev->irq_name, dev))
353                 return -EIO;
354
355         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
356                 spin_lock_irq(&dev->intx_lock);
357                 pci_intx(dev->dev, true);
358                 spin_unlock_irq(&dev->intx_lock);
359         }
360         return 0;
361 }
362
363 #ifdef __KVM_HAVE_MSI
364 static int assigned_device_enable_host_msi(struct kvm *kvm,
365                                            struct kvm_assigned_dev_kernel *dev)
366 {
367         int r;
368
369         if (!dev->dev->msi_enabled) {
370                 r = pci_enable_msi(dev->dev);
371                 if (r)
372                         return r;
373         }
374
375         dev->host_irq = dev->dev->irq;
376         if (request_threaded_irq(dev->host_irq, kvm_assigned_dev_msi,
377                                  kvm_assigned_dev_thread_msi, 0,
378                                  dev->irq_name, dev)) {
379                 pci_disable_msi(dev->dev);
380                 return -EIO;
381         }
382
383         return 0;
384 }
385 #endif
386
387 #ifdef __KVM_HAVE_MSIX
388 static int assigned_device_enable_host_msix(struct kvm *kvm,
389                                             struct kvm_assigned_dev_kernel *dev)
390 {
391         int i, r = -EINVAL;
392
393         /* host_msix_entries and guest_msix_entries should have been
394          * initialized */
395         if (dev->entries_nr == 0)
396                 return r;
397
398         r = pci_enable_msix_exact(dev->dev,
399                                   dev->host_msix_entries, dev->entries_nr);
400         if (r)
401                 return r;
402
403         for (i = 0; i < dev->entries_nr; i++) {
404                 r = request_threaded_irq(dev->host_msix_entries[i].vector,
405                                          kvm_assigned_dev_msix,
406                                          kvm_assigned_dev_thread_msix,
407                                          0, dev->irq_name, dev);
408                 if (r)
409                         goto err;
410         }
411
412         return 0;
413 err:
414         for (i -= 1; i >= 0; i--)
415                 free_irq(dev->host_msix_entries[i].vector, dev);
416         pci_disable_msix(dev->dev);
417         return r;
418 }
419
420 #endif
421
422 static int assigned_device_enable_guest_intx(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 = irq->guest_irq;
428         return 0;
429 }
430
431 #ifdef __KVM_HAVE_MSI
432 static int assigned_device_enable_guest_msi(struct kvm *kvm,
433                         struct kvm_assigned_dev_kernel *dev,
434                         struct kvm_assigned_irq *irq)
435 {
436         dev->guest_irq = irq->guest_irq;
437         dev->ack_notifier.gsi = -1;
438         return 0;
439 }
440 #endif
441
442 #ifdef __KVM_HAVE_MSIX
443 static int assigned_device_enable_guest_msix(struct kvm *kvm,
444                         struct kvm_assigned_dev_kernel *dev,
445                         struct kvm_assigned_irq *irq)
446 {
447         dev->guest_irq = irq->guest_irq;
448         dev->ack_notifier.gsi = -1;
449         return 0;
450 }
451 #endif
452
453 static int assign_host_irq(struct kvm *kvm,
454                            struct kvm_assigned_dev_kernel *dev,
455                            __u32 host_irq_type)
456 {
457         int r = -EEXIST;
458
459         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
460                 return r;
461
462         snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
463                  pci_name(dev->dev));
464
465         switch (host_irq_type) {
466         case KVM_DEV_IRQ_HOST_INTX:
467                 r = assigned_device_enable_host_intx(kvm, dev);
468                 break;
469 #ifdef __KVM_HAVE_MSI
470         case KVM_DEV_IRQ_HOST_MSI:
471                 r = assigned_device_enable_host_msi(kvm, dev);
472                 break;
473 #endif
474 #ifdef __KVM_HAVE_MSIX
475         case KVM_DEV_IRQ_HOST_MSIX:
476                 r = assigned_device_enable_host_msix(kvm, dev);
477                 break;
478 #endif
479         default:
480                 r = -EINVAL;
481         }
482         dev->host_irq_disabled = false;
483
484         if (!r)
485                 dev->irq_requested_type |= host_irq_type;
486
487         return r;
488 }
489
490 static int assign_guest_irq(struct kvm *kvm,
491                             struct kvm_assigned_dev_kernel *dev,
492                             struct kvm_assigned_irq *irq,
493                             unsigned long guest_irq_type)
494 {
495         int id;
496         int r = -EEXIST;
497
498         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
499                 return r;
500
501         id = kvm_request_irq_source_id(kvm);
502         if (id < 0)
503                 return id;
504
505         dev->irq_source_id = id;
506
507         switch (guest_irq_type) {
508         case KVM_DEV_IRQ_GUEST_INTX:
509                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
510                 break;
511 #ifdef __KVM_HAVE_MSI
512         case KVM_DEV_IRQ_GUEST_MSI:
513                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
514                 break;
515 #endif
516 #ifdef __KVM_HAVE_MSIX
517         case KVM_DEV_IRQ_GUEST_MSIX:
518                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
519                 break;
520 #endif
521         default:
522                 r = -EINVAL;
523         }
524
525         if (!r) {
526                 dev->irq_requested_type |= guest_irq_type;
527                 if (dev->ack_notifier.gsi != -1)
528                         kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
529         } else {
530                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
531                 dev->irq_source_id = -1;
532         }
533
534         return r;
535 }
536
537 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
538 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
539                                    struct kvm_assigned_irq *assigned_irq)
540 {
541         int r = -EINVAL;
542         struct kvm_assigned_dev_kernel *match;
543         unsigned long host_irq_type, guest_irq_type;
544
545         if (!irqchip_in_kernel(kvm))
546                 return r;
547
548         mutex_lock(&kvm->lock);
549         r = -ENODEV;
550         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
551                                       assigned_irq->assigned_dev_id);
552         if (!match)
553                 goto out;
554
555         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
556         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
557
558         r = -EINVAL;
559         /* can only assign one type at a time */
560         if (hweight_long(host_irq_type) > 1)
561                 goto out;
562         if (hweight_long(guest_irq_type) > 1)
563                 goto out;
564         if (host_irq_type == 0 && guest_irq_type == 0)
565                 goto out;
566
567         r = 0;
568         if (host_irq_type)
569                 r = assign_host_irq(kvm, match, host_irq_type);
570         if (r)
571                 goto out;
572
573         if (guest_irq_type)
574                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
575 out:
576         mutex_unlock(&kvm->lock);
577         return r;
578 }
579
580 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
581                                          struct kvm_assigned_irq
582                                          *assigned_irq)
583 {
584         int r = -ENODEV;
585         struct kvm_assigned_dev_kernel *match;
586         unsigned long irq_type;
587
588         mutex_lock(&kvm->lock);
589
590         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
591                                       assigned_irq->assigned_dev_id);
592         if (!match)
593                 goto out;
594
595         irq_type = assigned_irq->flags & (KVM_DEV_IRQ_HOST_MASK |
596                                           KVM_DEV_IRQ_GUEST_MASK);
597         r = kvm_deassign_irq(kvm, match, irq_type);
598 out:
599         mutex_unlock(&kvm->lock);
600         return r;
601 }
602
603 /*
604  * We want to test whether the caller has been granted permissions to
605  * use this device.  To be able to configure and control the device,
606  * the user needs access to PCI configuration space and BAR resources.
607  * These are accessed through PCI sysfs.  PCI config space is often
608  * passed to the process calling this ioctl via file descriptor, so we
609  * can't rely on access to that file.  We can check for permissions
610  * on each of the BAR resource files, which is a pretty clear
611  * indicator that the user has been granted access to the device.
612  */
613 static int probe_sysfs_permissions(struct pci_dev *dev)
614 {
615 #ifdef CONFIG_SYSFS
616         int i;
617         bool bar_found = false;
618
619         for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++) {
620                 char *kpath, *syspath;
621                 struct path path;
622                 struct inode *inode;
623                 int r;
624
625                 if (!pci_resource_len(dev, i))
626                         continue;
627
628                 kpath = kobject_get_path(&dev->dev.kobj, GFP_KERNEL);
629                 if (!kpath)
630                         return -ENOMEM;
631
632                 /* Per sysfs-rules, sysfs is always at /sys */
633                 syspath = kasprintf(GFP_KERNEL, "/sys%s/resource%d", kpath, i);
634                 kfree(kpath);
635                 if (!syspath)
636                         return -ENOMEM;
637
638                 r = kern_path(syspath, LOOKUP_FOLLOW, &path);
639                 kfree(syspath);
640                 if (r)
641                         return r;
642
643                 inode = path.dentry->d_inode;
644
645                 r = inode_permission(inode, MAY_READ | MAY_WRITE | MAY_ACCESS);
646                 path_put(&path);
647                 if (r)
648                         return r;
649
650                 bar_found = true;
651         }
652
653         /* If no resources, probably something special */
654         if (!bar_found)
655                 return -EPERM;
656
657         return 0;
658 #else
659         return -EINVAL; /* No way to control the device without sysfs */
660 #endif
661 }
662
663 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
664                                       struct kvm_assigned_pci_dev *assigned_dev)
665 {
666         int r = 0, idx;
667         struct kvm_assigned_dev_kernel *match;
668         struct pci_dev *dev;
669
670         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU))
671                 return -EINVAL;
672
673         mutex_lock(&kvm->lock);
674         idx = srcu_read_lock(&kvm->srcu);
675
676         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
677                                       assigned_dev->assigned_dev_id);
678         if (match) {
679                 /* device already assigned */
680                 r = -EEXIST;
681                 goto out;
682         }
683
684         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
685         if (match == NULL) {
686                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
687                        __func__);
688                 r = -ENOMEM;
689                 goto out;
690         }
691         dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
692                                    assigned_dev->busnr,
693                                    assigned_dev->devfn);
694         if (!dev) {
695                 printk(KERN_INFO "%s: host device not found\n", __func__);
696                 r = -EINVAL;
697                 goto out_free;
698         }
699
700         /* Don't allow bridges to be assigned */
701         if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) {
702                 r = -EPERM;
703                 goto out_put;
704         }
705
706         r = probe_sysfs_permissions(dev);
707         if (r)
708                 goto out_put;
709
710         if (pci_enable_device(dev)) {
711                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
712                 r = -EBUSY;
713                 goto out_put;
714         }
715         r = pci_request_regions(dev, "kvm_assigned_device");
716         if (r) {
717                 printk(KERN_INFO "%s: Could not get access to device regions\n",
718                        __func__);
719                 goto out_disable;
720         }
721
722         pci_reset_function(dev);
723         pci_save_state(dev);
724         match->pci_saved_state = pci_store_saved_state(dev);
725         if (!match->pci_saved_state)
726                 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
727                        __func__, dev_name(&dev->dev));
728
729         if (!pci_intx_mask_supported(dev))
730                 assigned_dev->flags &= ~KVM_DEV_ASSIGN_PCI_2_3;
731
732         match->assigned_dev_id = assigned_dev->assigned_dev_id;
733         match->host_segnr = assigned_dev->segnr;
734         match->host_busnr = assigned_dev->busnr;
735         match->host_devfn = assigned_dev->devfn;
736         match->flags = assigned_dev->flags;
737         match->dev = dev;
738         spin_lock_init(&match->intx_lock);
739         spin_lock_init(&match->intx_mask_lock);
740         match->irq_source_id = -1;
741         match->kvm = kvm;
742         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
743
744         list_add(&match->list, &kvm->arch.assigned_dev_head);
745
746         if (!kvm->arch.iommu_domain) {
747                 r = kvm_iommu_map_guest(kvm);
748                 if (r)
749                         goto out_list_del;
750         }
751         r = kvm_assign_device(kvm, match);
752         if (r)
753                 goto out_list_del;
754
755 out:
756         srcu_read_unlock(&kvm->srcu, idx);
757         mutex_unlock(&kvm->lock);
758         return r;
759 out_list_del:
760         if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
761                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
762                        __func__, dev_name(&dev->dev));
763         list_del(&match->list);
764         pci_release_regions(dev);
765 out_disable:
766         pci_disable_device(dev);
767 out_put:
768         pci_dev_put(dev);
769 out_free:
770         kfree(match);
771         srcu_read_unlock(&kvm->srcu, idx);
772         mutex_unlock(&kvm->lock);
773         return r;
774 }
775
776 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
777                 struct kvm_assigned_pci_dev *assigned_dev)
778 {
779         int r = 0;
780         struct kvm_assigned_dev_kernel *match;
781
782         mutex_lock(&kvm->lock);
783
784         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
785                                       assigned_dev->assigned_dev_id);
786         if (!match) {
787                 printk(KERN_INFO "%s: device hasn't been assigned before, "
788                   "so cannot be deassigned\n", __func__);
789                 r = -EINVAL;
790                 goto out;
791         }
792
793         kvm_deassign_device(kvm, match);
794
795         kvm_free_assigned_device(kvm, match);
796
797 out:
798         mutex_unlock(&kvm->lock);
799         return r;
800 }
801
802
803 #ifdef __KVM_HAVE_MSIX
804 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
805                                     struct kvm_assigned_msix_nr *entry_nr)
806 {
807         int r = 0;
808         struct kvm_assigned_dev_kernel *adev;
809
810         mutex_lock(&kvm->lock);
811
812         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
813                                       entry_nr->assigned_dev_id);
814         if (!adev) {
815                 r = -EINVAL;
816                 goto msix_nr_out;
817         }
818
819         if (adev->entries_nr == 0) {
820                 adev->entries_nr = entry_nr->entry_nr;
821                 if (adev->entries_nr == 0 ||
822                     adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
823                         r = -EINVAL;
824                         goto msix_nr_out;
825                 }
826
827                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
828                                                 entry_nr->entry_nr,
829                                                 GFP_KERNEL);
830                 if (!adev->host_msix_entries) {
831                         r = -ENOMEM;
832                         goto msix_nr_out;
833                 }
834                 adev->guest_msix_entries =
835                         kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
836                                 GFP_KERNEL);
837                 if (!adev->guest_msix_entries) {
838                         kfree(adev->host_msix_entries);
839                         r = -ENOMEM;
840                         goto msix_nr_out;
841                 }
842         } else /* Not allowed set MSI-X number twice */
843                 r = -EINVAL;
844 msix_nr_out:
845         mutex_unlock(&kvm->lock);
846         return r;
847 }
848
849 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
850                                        struct kvm_assigned_msix_entry *entry)
851 {
852         int r = 0, i;
853         struct kvm_assigned_dev_kernel *adev;
854
855         mutex_lock(&kvm->lock);
856
857         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
858                                       entry->assigned_dev_id);
859
860         if (!adev) {
861                 r = -EINVAL;
862                 goto msix_entry_out;
863         }
864
865         for (i = 0; i < adev->entries_nr; i++)
866                 if (adev->guest_msix_entries[i].vector == 0 ||
867                     adev->guest_msix_entries[i].entry == entry->entry) {
868                         adev->guest_msix_entries[i].entry = entry->entry;
869                         adev->guest_msix_entries[i].vector = entry->gsi;
870                         adev->host_msix_entries[i].entry = entry->entry;
871                         break;
872                 }
873         if (i == adev->entries_nr) {
874                 r = -ENOSPC;
875                 goto msix_entry_out;
876         }
877
878 msix_entry_out:
879         mutex_unlock(&kvm->lock);
880
881         return r;
882 }
883 #endif
884
885 static int kvm_vm_ioctl_set_pci_irq_mask(struct kvm *kvm,
886                 struct kvm_assigned_pci_dev *assigned_dev)
887 {
888         int r = 0;
889         struct kvm_assigned_dev_kernel *match;
890
891         mutex_lock(&kvm->lock);
892
893         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
894                                       assigned_dev->assigned_dev_id);
895         if (!match) {
896                 r = -ENODEV;
897                 goto out;
898         }
899
900         spin_lock(&match->intx_mask_lock);
901
902         match->flags &= ~KVM_DEV_ASSIGN_MASK_INTX;
903         match->flags |= assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX;
904
905         if (match->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
906                 if (assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX) {
907                         kvm_set_irq(match->kvm, match->irq_source_id,
908                                     match->guest_irq, 0, false);
909                         /*
910                          * Masking at hardware-level is performed on demand,
911                          * i.e. when an IRQ actually arrives at the host.
912                          */
913                 } else if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
914                         /*
915                          * Unmask the IRQ line if required. Unmasking at
916                          * device level will be performed by user space.
917                          */
918                         spin_lock_irq(&match->intx_lock);
919                         if (match->host_irq_disabled) {
920                                 enable_irq(match->host_irq);
921                                 match->host_irq_disabled = false;
922                         }
923                         spin_unlock_irq(&match->intx_lock);
924                 }
925         }
926
927         spin_unlock(&match->intx_mask_lock);
928
929 out:
930         mutex_unlock(&kvm->lock);
931         return r;
932 }
933
934 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
935                                   unsigned long arg)
936 {
937         void __user *argp = (void __user *)arg;
938         int r;
939
940         switch (ioctl) {
941         case KVM_ASSIGN_PCI_DEVICE: {
942                 struct kvm_assigned_pci_dev assigned_dev;
943
944                 r = -EFAULT;
945                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
946                         goto out;
947                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
948                 if (r)
949                         goto out;
950                 break;
951         }
952         case KVM_ASSIGN_IRQ: {
953                 r = -EOPNOTSUPP;
954                 break;
955         }
956         case KVM_ASSIGN_DEV_IRQ: {
957                 struct kvm_assigned_irq assigned_irq;
958
959                 r = -EFAULT;
960                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
961                         goto out;
962                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
963                 if (r)
964                         goto out;
965                 break;
966         }
967         case KVM_DEASSIGN_DEV_IRQ: {
968                 struct kvm_assigned_irq assigned_irq;
969
970                 r = -EFAULT;
971                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
972                         goto out;
973                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
974                 if (r)
975                         goto out;
976                 break;
977         }
978         case KVM_DEASSIGN_PCI_DEVICE: {
979                 struct kvm_assigned_pci_dev assigned_dev;
980
981                 r = -EFAULT;
982                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
983                         goto out;
984                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
985                 if (r)
986                         goto out;
987                 break;
988         }
989 #ifdef __KVM_HAVE_MSIX
990         case KVM_ASSIGN_SET_MSIX_NR: {
991                 struct kvm_assigned_msix_nr entry_nr;
992                 r = -EFAULT;
993                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
994                         goto out;
995                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
996                 if (r)
997                         goto out;
998                 break;
999         }
1000         case KVM_ASSIGN_SET_MSIX_ENTRY: {
1001                 struct kvm_assigned_msix_entry entry;
1002                 r = -EFAULT;
1003                 if (copy_from_user(&entry, argp, sizeof entry))
1004                         goto out;
1005                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
1006                 if (r)
1007                         goto out;
1008                 break;
1009         }
1010 #endif
1011         case KVM_ASSIGN_SET_INTX_MASK: {
1012                 struct kvm_assigned_pci_dev assigned_dev;
1013
1014                 r = -EFAULT;
1015                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1016                         goto out;
1017                 r = kvm_vm_ioctl_set_pci_irq_mask(kvm, &assigned_dev);
1018                 break;
1019         }
1020         default:
1021                 r = -ENOTTY;
1022                 break;
1023         }
1024 out:
1025         return r;
1026 }