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 "irq.h"
  21
  22 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  23                                                       int assigned_dev_id)
  24 {
  25         struct list_head *ptr;
  26         struct kvm_assigned_dev_kernel *match;
  27
  28         list_for_each(ptr, head) {
  29                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
  30                 if (match->assigned_dev_id == assigned_dev_id)
  31                         return match;
  32         }
  33         return NULL;
  34 }
  35
  36 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
  37                                     *assigned_dev, int irq)
  38 {
  39         int i, index;
  40         struct msix_entry *host_msix_entries;
  41
  42         host_msix_entries = assigned_dev->host_msix_entries;
  43
  44         index = -1;
  45         for (i = 0; i < assigned_dev->entries_nr; i++)
  46                 if (irq == host_msix_entries[i].vector) {
  47                         index = i;
  48                         break;
  49                 }
  50         if (index < 0) {
  51                 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
  52                 return 0;
  53         }
  54
  55         return index;
  56 }
  57
  58 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
  59 {
  60         struct kvm_assigned_dev_kernel *assigned_dev;
  61         int i;
  62
  63         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
  64                                     interrupt_work);
  65
  66         spin_lock_irq(&assigned_dev->assigned_dev_lock);
  67         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
  68                 struct kvm_guest_msix_entry *guest_entries =
  69                         assigned_dev->guest_msix_entries;
  70                 for (i = 0; i < assigned_dev->entries_nr; i++) {
  71                         if (!(guest_entries[i].flags &
  72                                         KVM_ASSIGNED_MSIX_PENDING))
  73                                 continue;
  74                         guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
  75                         kvm_set_irq(assigned_dev->kvm,
  76                                     assigned_dev->irq_source_id,
  77                                     guest_entries[i].vector, 1);
  78                 }
  79         } else
  80                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
  81                             assigned_dev->guest_irq, 1);
  82
  83         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
  84 }
  85
  86 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
  87 {
  88         unsigned long flags;
  89         struct kvm_assigned_dev_kernel *assigned_dev =
  90                 (struct kvm_assigned_dev_kernel *) dev_id;
  91
  92         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
  93         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
  94                 int index = find_index_from_host_irq(assigned_dev, irq);
  95                 if (index < 0)
  96                         goto out;
  97                 assigned_dev->guest_msix_entries[index].flags |=
  98                         KVM_ASSIGNED_MSIX_PENDING;
  99         }
 100
 101         schedule_work(&assigned_dev->interrupt_work);
 102
 103         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 104                 disable_irq_nosync(irq);
 105                 assigned_dev->host_irq_disabled = true;
 106         }
 107
 108 out:
 109         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
 110         return IRQ_HANDLED;
 111 }
 112
 113 /* Ack the irq line for an assigned device */
 114 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 115 {
 116         struct kvm_assigned_dev_kernel *dev;
 117         unsigned long flags;
 118
 119         if (kian->gsi == -1)
 120                 return;
 121
 122         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 123                            ack_notifier);
 124
 125         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 126
 127         /* The guest irq may be shared so this ack may be
 128          * from another device.
 129          */
 130         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
 131         if (dev->host_irq_disabled) {
 132                 enable_irq(dev->host_irq);
 133                 dev->host_irq_disabled = false;
 134         }
 135         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
 136 }
 137
 138 static void deassign_guest_irq(struct kvm *kvm,
 139                                struct kvm_assigned_dev_kernel *assigned_dev)
 140 {
 141         kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
 142         assigned_dev->ack_notifier.gsi = -1;
 143
 144         if (assigned_dev->irq_source_id != -1)
 145                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 146         assigned_dev->irq_source_id = -1;
 147         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 148 }
 149
 150 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 151 static void deassign_host_irq(struct kvm *kvm,
 152                               struct kvm_assigned_dev_kernel *assigned_dev)
 153 {
 154         /*
 155          * In kvm_free_device_irq, cancel_work_sync return true if:
 156          * 1. work is scheduled, and then cancelled.
 157          * 2. work callback is executed.
 158          *
 159          * The first one ensured that the irq is disabled and no more events
 160          * would happen. But for the second one, the irq may be enabled (e.g.
 161          * for MSI). So we disable irq here to prevent further events.
 162          *
 163          * Notice this maybe result in nested disable if the interrupt type is
 164          * INTx, but it's OK for we are going to free it.
 165          *
 166          * If this function is a part of VM destroy, please ensure that till
 167          * now, the kvm state is still legal for probably we also have to wait
 168          * interrupt_work done.
 169          */
 170         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 171                 int i;
 172                 for (i = 0; i < assigned_dev->entries_nr; i++)
 173                         disable_irq_nosync(assigned_dev->
 174                                            host_msix_entries[i].vector);
 175
 176                 cancel_work_sync(&assigned_dev->interrupt_work);
 177
 178                 for (i = 0; i < assigned_dev->entries_nr; i++)
 179                         free_irq(assigned_dev->host_msix_entries[i].vector,
 180                                  (void *)assigned_dev);
 181
 182                 assigned_dev->entries_nr = 0;
 183                 kfree(assigned_dev->host_msix_entries);
 184                 kfree(assigned_dev->guest_msix_entries);
 185                 pci_disable_msix(assigned_dev->dev);
 186         } else {
 187                 /* Deal with MSI and INTx */
 188                 disable_irq_nosync(assigned_dev->host_irq);
 189                 cancel_work_sync(&assigned_dev->interrupt_work);
 190
 191                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 192
 193                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 194                         pci_disable_msi(assigned_dev->dev);
 195         }
 196
 197         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 198 }
 199
 200 static int kvm_deassign_irq(struct kvm *kvm,
 201                             struct kvm_assigned_dev_kernel *assigned_dev,
 202                             unsigned long irq_requested_type)
 203 {
 204         unsigned long guest_irq_type, host_irq_type;
 205
 206         if (!irqchip_in_kernel(kvm))
 207                 return -EINVAL;
 208         /* no irq assignment to deassign */
 209         if (!assigned_dev->irq_requested_type)
 210                 return -ENXIO;
 211
 212         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 213         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 214
 215         if (host_irq_type)
 216                 deassign_host_irq(kvm, assigned_dev);
 217         if (guest_irq_type)
 218                 deassign_guest_irq(kvm, assigned_dev);
 219
 220         return 0;
 221 }
 222
 223 static void kvm_free_assigned_irq(struct kvm *kvm,
 224                                   struct kvm_assigned_dev_kernel *assigned_dev)
 225 {
 226         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 227 }
 228
 229 static void kvm_free_assigned_device(struct kvm *kvm,
 230                                      struct kvm_assigned_dev_kernel
 231                                      *assigned_dev)
 232 {
 233         kvm_free_assigned_irq(kvm, assigned_dev);
 234
 235         pci_reset_function(assigned_dev->dev);
 236
 237         pci_release_regions(assigned_dev->dev);
 238         pci_disable_device(assigned_dev->dev);
 239         pci_dev_put(assigned_dev->dev);
 240
 241         list_del(&assigned_dev->list);
 242         kfree(assigned_dev);
 243 }
 244
 245 void kvm_free_all_assigned_devices(struct kvm *kvm)
 246 {
 247         struct list_head *ptr, *ptr2;
 248         struct kvm_assigned_dev_kernel *assigned_dev;
 249
 250         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 251                 assigned_dev = list_entry(ptr,
 252                                           struct kvm_assigned_dev_kernel,
 253                                           list);
 254
 255                 kvm_free_assigned_device(kvm, assigned_dev);
 256         }
 257 }
 258
 259 static int assigned_device_enable_host_intx(struct kvm *kvm,
 260                                             struct kvm_assigned_dev_kernel *dev)
 261 {
 262         dev->host_irq = dev->dev->irq;
 263         /* Even though this is PCI, we don't want to use shared
 264          * interrupts. Sharing host devices with guest-assigned devices
 265          * on the same interrupt line is not a happy situation: there
 266          * are going to be long delays in accepting, acking, etc.
 267          */
 268         if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
 269                         0, "kvm_assigned_intx_device", (void *)dev))
 270                 return -EIO;
 271         return 0;
 272 }
 273
 274 #ifdef __KVM_HAVE_MSI
 275 static int assigned_device_enable_host_msi(struct kvm *kvm,
 276                                            struct kvm_assigned_dev_kernel *dev)
 277 {
 278         int r;
 279
 280         if (!dev->dev->msi_enabled) {
 281                 r = pci_enable_msi(dev->dev);
 282                 if (r)
 283                         return r;
 284         }
 285
 286         dev->host_irq = dev->dev->irq;
 287         if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
 288                         "kvm_assigned_msi_device", (void *)dev)) {
 289                 pci_disable_msi(dev->dev);
 290                 return -EIO;
 291         }
 292
 293         return 0;
 294 }
 295 #endif
 296
 297 #ifdef __KVM_HAVE_MSIX
 298 static int assigned_device_enable_host_msix(struct kvm *kvm,
 299                                             struct kvm_assigned_dev_kernel *dev)
 300 {
 301         int i, r = -EINVAL;
 302
 303         /* host_msix_entries and guest_msix_entries should have been
 304          * initialized */
 305         if (dev->entries_nr == 0)
 306                 return r;
 307
 308         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 309         if (r)
 310                 return r;
 311
 312         for (i = 0; i < dev->entries_nr; i++) {
 313                 r = request_irq(dev->host_msix_entries[i].vector,
 314                                 kvm_assigned_dev_intr, 0,
 315                                 "kvm_assigned_msix_device",
 316                                 (void *)dev);
 317                 if (r)
 318                         goto err;
 319         }
 320
 321         return 0;
 322 err:
 323         for (i -= 1; i >= 0; i--)
 324                 free_irq(dev->host_msix_entries[i].vector, (void *)dev);
 325         pci_disable_msix(dev->dev);
 326         return r;
 327 }
 328
 329 #endif
 330
 331 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 332                                 struct kvm_assigned_dev_kernel *dev,
 333                                 struct kvm_assigned_irq *irq)
 334 {
 335         dev->guest_irq = irq->guest_irq;
 336         dev->ack_notifier.gsi = irq->guest_irq;
 337         return 0;
 338 }
 339
 340 #ifdef __KVM_HAVE_MSI
 341 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 342                         struct kvm_assigned_dev_kernel *dev,
 343                         struct kvm_assigned_irq *irq)
 344 {
 345         dev->guest_irq = irq->guest_irq;
 346         dev->ack_notifier.gsi = -1;
 347         dev->host_irq_disabled = false;
 348         return 0;
 349 }
 350 #endif
 351
 352 #ifdef __KVM_HAVE_MSIX
 353 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 354                         struct kvm_assigned_dev_kernel *dev,
 355                         struct kvm_assigned_irq *irq)
 356 {
 357         dev->guest_irq = irq->guest_irq;
 358         dev->ack_notifier.gsi = -1;
 359         dev->host_irq_disabled = false;
 360         return 0;
 361 }
 362 #endif
 363
 364 static int assign_host_irq(struct kvm *kvm,
 365                            struct kvm_assigned_dev_kernel *dev,
 366                            __u32 host_irq_type)
 367 {
 368         int r = -EEXIST;
 369
 370         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 371                 return r;
 372
 373         switch (host_irq_type) {
 374         case KVM_DEV_IRQ_HOST_INTX:
 375                 r = assigned_device_enable_host_intx(kvm, dev);
 376                 break;
 377 #ifdef __KVM_HAVE_MSI
 378         case KVM_DEV_IRQ_HOST_MSI:
 379                 r = assigned_device_enable_host_msi(kvm, dev);
 380                 break;
 381 #endif
 382 #ifdef __KVM_HAVE_MSIX
 383         case KVM_DEV_IRQ_HOST_MSIX:
 384                 r = assigned_device_enable_host_msix(kvm, dev);
 385                 break;
 386 #endif
 387         default:
 388                 r = -EINVAL;
 389         }
 390
 391         if (!r)
 392                 dev->irq_requested_type |= host_irq_type;
 393
 394         return r;
 395 }
 396
 397 static int assign_guest_irq(struct kvm *kvm,
 398                             struct kvm_assigned_dev_kernel *dev,
 399                             struct kvm_assigned_irq *irq,
 400                             unsigned long guest_irq_type)
 401 {
 402         int id;
 403         int r = -EEXIST;
 404
 405         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 406                 return r;
 407
 408         id = kvm_request_irq_source_id(kvm);
 409         if (id < 0)
 410                 return id;
 411
 412         dev->irq_source_id = id;
 413
 414         switch (guest_irq_type) {
 415         case KVM_DEV_IRQ_GUEST_INTX:
 416                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 417                 break;
 418 #ifdef __KVM_HAVE_MSI
 419         case KVM_DEV_IRQ_GUEST_MSI:
 420                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 421                 break;
 422 #endif
 423 #ifdef __KVM_HAVE_MSIX
 424         case KVM_DEV_IRQ_GUEST_MSIX:
 425                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 426                 break;
 427 #endif
 428         default:
 429                 r = -EINVAL;
 430         }
 431
 432         if (!r) {
 433                 dev->irq_requested_type |= guest_irq_type;
 434                 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 435         } else
 436                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 437
 438         return r;
 439 }
 440
 441 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 442 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 443                                    struct kvm_assigned_irq *assigned_irq)
 444 {
 445         int r = -EINVAL;
 446         struct kvm_assigned_dev_kernel *match;
 447         unsigned long host_irq_type, guest_irq_type;
 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);
 479 out:
 480         mutex_unlock(&kvm->lock);
 481         return r;
 482 }
 483
 484 static 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);
 499 out:
 500         mutex_unlock(&kvm->lock);
 501         return r;
 502 }
 503
 504 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 505                                       struct kvm_assigned_pci_dev *assigned_dev)
 506 {
 507         int r = 0, idx;
 508         struct kvm_assigned_dev_kernel *match;
 509         struct pci_dev *dev;
 510
 511         mutex_lock(&kvm->lock);
 512         idx = srcu_read_lock(&kvm->srcu);
 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_domain_bus_and_slot(assigned_dev->segnr,
 530                                    assigned_dev->busnr,
 531                                    assigned_dev->devfn);
 532         if (!dev) {
 533                 printk(KERN_INFO "%s: host device not found\n", __func__);
 534                 r = -EINVAL;
 535                 goto out_free;
 536         }
 537         if (pci_enable_device(dev)) {
 538                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 539                 r = -EBUSY;
 540                 goto out_put;
 541         }
 542         r = pci_request_regions(dev, "kvm_assigned_device");
 543         if (r) {
 544                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 545                        __func__);
 546                 goto out_disable;
 547         }
 548
 549         pci_reset_function(dev);
 550
 551         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 552         match->host_segnr = assigned_dev->segnr;
 553         match->host_busnr = assigned_dev->busnr;
 554         match->host_devfn = assigned_dev->devfn;
 555         match->flags = assigned_dev->flags;
 556         match->dev = dev;
 557         spin_lock_init(&match->assigned_dev_lock);
 558         match->irq_source_id = -1;
 559         match->kvm = kvm;
 560         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 561         INIT_WORK(&match->interrupt_work,
 562                   kvm_assigned_dev_interrupt_work_handler);
 563
 564         list_add(&match->list, &kvm->arch.assigned_dev_head);
 565
 566         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 567                 if (!kvm->arch.iommu_domain) {
 568                         r = kvm_iommu_map_guest(kvm);
 569                         if (r)
 570                                 goto out_list_del;
 571                 }
 572                 r = kvm_assign_device(kvm, match);
 573                 if (r)
 574                         goto out_list_del;
 575         }
 576
 577 out:
 578         srcu_read_unlock(&kvm->srcu, idx);
 579         mutex_unlock(&kvm->lock);
 580         return r;
 581 out_list_del:
 582         list_del(&match->list);
 583         pci_release_regions(dev);
 584 out_disable:
 585         pci_disable_device(dev);
 586 out_put:
 587         pci_dev_put(dev);
 588 out_free:
 589         kfree(match);
 590         srcu_read_unlock(&kvm->srcu, idx);
 591         mutex_unlock(&kvm->lock);
 592         return r;
 593 }
 594
 595 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 596                 struct kvm_assigned_pci_dev *assigned_dev)
 597 {
 598         int r = 0;
 599         struct kvm_assigned_dev_kernel *match;
 600
 601         mutex_lock(&kvm->lock);
 602
 603         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 604                                       assigned_dev->assigned_dev_id);
 605         if (!match) {
 606                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 607                   "so cannot be deassigned\n", __func__);
 608                 r = -EINVAL;
 609                 goto out;
 610         }
 611
 612         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 613                 kvm_deassign_device(kvm, match);
 614
 615         kvm_free_assigned_device(kvm, match);
 616
 617 out:
 618         mutex_unlock(&kvm->lock);
 619         return r;
 620 }
 621
 622
 623 #ifdef __KVM_HAVE_MSIX
 624 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
 625                                     struct kvm_assigned_msix_nr *entry_nr)
 626 {
 627         int r = 0;
 628         struct kvm_assigned_dev_kernel *adev;
 629
 630         mutex_lock(&kvm->lock);
 631
 632         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 633                                       entry_nr->assigned_dev_id);
 634         if (!adev) {
 635                 r = -EINVAL;
 636                 goto msix_nr_out;
 637         }
 638
 639         if (adev->entries_nr == 0) {
 640                 adev->entries_nr = entry_nr->entry_nr;
 641                 if (adev->entries_nr == 0 ||
 642                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
 643                         r = -EINVAL;
 644                         goto msix_nr_out;
 645                 }
 646
 647                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
 648                                                 entry_nr->entry_nr,
 649                                                 GFP_KERNEL);
 650                 if (!adev->host_msix_entries) {
 651                         r = -ENOMEM;
 652                         goto msix_nr_out;
 653                 }
 654                 adev->guest_msix_entries = kzalloc(
 655                                 sizeof(struct kvm_guest_msix_entry) *
 656                                 entry_nr->entry_nr, GFP_KERNEL);
 657                 if (!adev->guest_msix_entries) {
 658                         kfree(adev->host_msix_entries);
 659                         r = -ENOMEM;
 660                         goto msix_nr_out;
 661                 }
 662         } else /* Not allowed set MSI-X number twice */
 663                 r = -EINVAL;
 664 msix_nr_out:
 665         mutex_unlock(&kvm->lock);
 666         return r;
 667 }
 668
 669 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
 670                                        struct kvm_assigned_msix_entry *entry)
 671 {
 672         int r = 0, i;
 673         struct kvm_assigned_dev_kernel *adev;
 674
 675         mutex_lock(&kvm->lock);
 676
 677         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 678                                       entry->assigned_dev_id);
 679
 680         if (!adev) {
 681                 r = -EINVAL;
 682                 goto msix_entry_out;
 683         }
 684
 685         for (i = 0; i < adev->entries_nr; i++)
 686                 if (adev->guest_msix_entries[i].vector == 0 ||
 687                     adev->guest_msix_entries[i].entry == entry->entry) {
 688                         adev->guest_msix_entries[i].entry = entry->entry;
 689                         adev->guest_msix_entries[i].vector = entry->gsi;
 690                         adev->host_msix_entries[i].entry = entry->entry;
 691                         break;
 692                 }
 693         if (i == adev->entries_nr) {
 694                 r = -ENOSPC;
 695                 goto msix_entry_out;
 696         }
 697
 698 msix_entry_out:
 699         mutex_unlock(&kvm->lock);
 700
 701         return r;
 702 }
 703 #endif
 704
 705 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
 706                                   unsigned long arg)
 707 {
 708         void __user *argp = (void __user *)arg;
 709         int r = -ENOTTY;
 710
 711         switch (ioctl) {
 712         case KVM_ASSIGN_PCI_DEVICE: {
 713                 struct kvm_assigned_pci_dev assigned_dev;
 714
 715                 r = -EFAULT;
 716                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 717                         goto out;
 718                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
 719                 if (r)
 720                         goto out;
 721                 break;
 722         }
 723         case KVM_ASSIGN_IRQ: {
 724                 r = -EOPNOTSUPP;
 725                 break;
 726         }
 727 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
 728         case KVM_ASSIGN_DEV_IRQ: {
 729                 struct kvm_assigned_irq assigned_irq;
 730
 731                 r = -EFAULT;
 732                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 733                         goto out;
 734                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
 735                 if (r)
 736                         goto out;
 737                 break;
 738         }
 739         case KVM_DEASSIGN_DEV_IRQ: {
 740                 struct kvm_assigned_irq assigned_irq;
 741
 742                 r = -EFAULT;
 743                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 744                         goto out;
 745                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
 746                 if (r)
 747                         goto out;
 748                 break;
 749         }
 750 #endif
 751 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 752         case KVM_DEASSIGN_PCI_DEVICE: {
 753                 struct kvm_assigned_pci_dev assigned_dev;
 754
 755                 r = -EFAULT;
 756                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 757                         goto out;
 758                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
 759                 if (r)
 760                         goto out;
 761                 break;
 762         }
 763 #endif
 764 #ifdef KVM_CAP_IRQ_ROUTING
 765         case KVM_SET_GSI_ROUTING: {
 766                 struct kvm_irq_routing routing;
 767                 struct kvm_irq_routing __user *urouting;
 768                 struct kvm_irq_routing_entry *entries;
 769
 770                 r = -EFAULT;
 771                 if (copy_from_user(&routing, argp, sizeof(routing)))
 772                         goto out;
 773                 r = -EINVAL;
 774                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
 775                         goto out;
 776                 if (routing.flags)
 777                         goto out;
 778                 r = -ENOMEM;
 779                 entries = vmalloc(routing.nr * sizeof(*entries));
 780                 if (!entries)
 781                         goto out;
 782                 r = -EFAULT;
 783                 urouting = argp;
 784                 if (copy_from_user(entries, urouting->entries,
 785                                    routing.nr * sizeof(*entries)))
 786                         goto out_free_irq_routing;
 787                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
 788                                         routing.flags);
 789         out_free_irq_routing:
 790                 vfree(entries);
 791                 break;
 792         }
 793 #endif /* KVM_CAP_IRQ_ROUTING */
 794 #ifdef __KVM_HAVE_MSIX
 795         case KVM_ASSIGN_SET_MSIX_NR: {
 796                 struct kvm_assigned_msix_nr entry_nr;
 797                 r = -EFAULT;
 798                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
 799                         goto out;
 800                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
 801                 if (r)
 802                         goto out;
 803                 break;
 804         }
 805         case KVM_ASSIGN_SET_MSIX_ENTRY: {
 806                 struct kvm_assigned_msix_entry entry;
 807                 r = -EFAULT;
 808                 if (copy_from_user(&entry, argp, sizeof entry))
 809                         goto out;
 810                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
 811                 if (r)
 812                         goto out;
 813                 break;
 814         }
 815 #endif
 816         }
 817 out:
 818         return r;
 819 }
 820