virt/kvm/assigned-dev.c

   1 /*
   2  * Kernel-based Virtual Machine - device assignment support
   3  *
   4  * Copyright (C) 2006-9 Red Hat, Inc
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2.  See
   7  * the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include <linux/kvm_host.h>
  12 #include <linux/kvm.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/errno.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/pci.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/slab.h>
  20 #include "irq.h"
  21
  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         struct kvm *kvm;
  62         int i;
  63
  64         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
  65                                     interrupt_work);
  66         kvm = assigned_dev->kvm;
  67
  68         spin_lock_irq(&assigned_dev->assigned_dev_lock);
  69         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
  70                 struct kvm_guest_msix_entry *guest_entries =
  71                         assigned_dev->guest_msix_entries;
  72                 for (i = 0; i < assigned_dev->entries_nr; i++) {
  73                         if (!(guest_entries[i].flags &
  74                                         KVM_ASSIGNED_MSIX_PENDING))
  75                                 continue;
  76                         guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
  77                         kvm_set_irq(assigned_dev->kvm,
  78                                     assigned_dev->irq_source_id,
  79                                     guest_entries[i].vector, 1);
  80                 }
  81         } else
  82                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
  83                             assigned_dev->guest_irq, 1);
  84
  85         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
  86 }
  87
  88 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
  89 {
  90         unsigned long flags;
  91         struct kvm_assigned_dev_kernel *assigned_dev =
  92                 (struct kvm_assigned_dev_kernel *) dev_id;
  93
  94         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
  95         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
  96                 int index = find_index_from_host_irq(assigned_dev, irq);
  97                 if (index < 0)
  98                         goto out;
  99                 assigned_dev->guest_msix_entries[index].flags |=
 100                         KVM_ASSIGNED_MSIX_PENDING;
 101         }
 102
 103         schedule_work(&assigned_dev->interrupt_work);
 104
 105         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 106                 disable_irq_nosync(irq);
 107                 assigned_dev->host_irq_disabled = true;
 108         }
 109
 110 out:
 111         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
 112         return IRQ_HANDLED;
 113 }
 114
 115 /* Ack the irq line for an assigned device */
 116 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 117 {
 118         struct kvm_assigned_dev_kernel *dev;
 119         unsigned long flags;
 120
 121         if (kian->gsi == -1)
 122                 return;
 123
 124         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 125                            ack_notifier);
 126
 127         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 128
 129         /* The guest irq may be shared so this ack may be
 130          * from another device.
 131          */
 132         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
 133         if (dev->host_irq_disabled) {
 134                 enable_irq(dev->host_irq);
 135                 dev->host_irq_disabled = false;
 136         }
 137         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
 138 }
 139
 140 static void deassign_guest_irq(struct kvm *kvm,
 141                                struct kvm_assigned_dev_kernel *assigned_dev)
 142 {
 143         kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
 144         assigned_dev->ack_notifier.gsi = -1;
 145
 146         if (assigned_dev->irq_source_id != -1)
 147                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 148         assigned_dev->irq_source_id = -1;
 149         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 150 }
 151
 152 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 153 static void deassign_host_irq(struct kvm *kvm,
 154                               struct kvm_assigned_dev_kernel *assigned_dev)
 155 {
 156         /*
 157          * In kvm_free_device_irq, cancel_work_sync return true if:
 158          * 1. work is scheduled, and then cancelled.
 159          * 2. work callback is executed.
 160          *
 161          * The first one ensured that the irq is disabled and no more events
 162          * would happen. But for the second one, the irq may be enabled (e.g.
 163          * for MSI). So we disable irq here to prevent further events.
 164          *
 165          * Notice this maybe result in nested disable if the interrupt type is
 166          * INTx, but it's OK for we are going to free it.
 167          *
 168          * If this function is a part of VM destroy, please ensure that till
 169          * now, the kvm state is still legal for probably we also have to wait
 170          * interrupt_work done.
 171          */
 172         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 173                 int i;
 174                 for (i = 0; i < assigned_dev->entries_nr; i++)
 175                         disable_irq_nosync(assigned_dev->
 176                                            host_msix_entries[i].vector);
 177
 178                 cancel_work_sync(&assigned_dev->interrupt_work);
 179
 180                 for (i = 0; i < assigned_dev->entries_nr; i++)
 181                         free_irq(assigned_dev->host_msix_entries[i].vector,
 182                                  (void *)assigned_dev);
 183
 184                 assigned_dev->entries_nr = 0;
 185                 kfree(assigned_dev->host_msix_entries);
 186                 kfree(assigned_dev->guest_msix_entries);
 187                 pci_disable_msix(assigned_dev->dev);
 188         } else {
 189                 /* Deal with MSI and INTx */
 190                 disable_irq_nosync(assigned_dev->host_irq);
 191                 cancel_work_sync(&assigned_dev->interrupt_work);
 192
 193                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 194
 195                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 196                         pci_disable_msi(assigned_dev->dev);
 197         }
 198
 199         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 200 }
 201
 202 static int kvm_deassign_irq(struct kvm *kvm,
 203                             struct kvm_assigned_dev_kernel *assigned_dev,
 204                             unsigned long irq_requested_type)
 205 {
 206         unsigned long guest_irq_type, host_irq_type;
 207
 208         if (!irqchip_in_kernel(kvm))
 209                 return -EINVAL;
 210         /* no irq assignment to deassign */
 211         if (!assigned_dev->irq_requested_type)
 212                 return -ENXIO;
 213
 214         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 215         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 216
 217         if (host_irq_type)
 218                 deassign_host_irq(kvm, assigned_dev);
 219         if (guest_irq_type)
 220                 deassign_guest_irq(kvm, assigned_dev);
 221
 222         return 0;
 223 }
 224
 225 static void kvm_free_assigned_irq(struct kvm *kvm,
 226                                   struct kvm_assigned_dev_kernel *assigned_dev)
 227 {
 228         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 229 }
 230
 231 static void kvm_free_assigned_device(struct kvm *kvm,
 232                                      struct kvm_assigned_dev_kernel
 233                                      *assigned_dev)
 234 {
 235         kvm_free_assigned_irq(kvm, assigned_dev);
 236
 237         pci_reset_function(assigned_dev->dev);
 238
 239         pci_release_regions(assigned_dev->dev);
 240         pci_disable_device(assigned_dev->dev);
 241         pci_dev_put(assigned_dev->dev);
 242
 243         list_del(&assigned_dev->list);
 244         kfree(assigned_dev);
 245 }
 246
 247 void kvm_free_all_assigned_devices(struct kvm *kvm)
 248 {
 249         struct list_head *ptr, *ptr2;
 250         struct kvm_assigned_dev_kernel *assigned_dev;
 251
 252         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 253                 assigned_dev = list_entry(ptr,
 254                                           struct kvm_assigned_dev_kernel,
 255                                           list);
 256
 257                 kvm_free_assigned_device(kvm, assigned_dev);
 258         }
 259 }
 260
 261 static int assigned_device_enable_host_intx(struct kvm *kvm,
 262                                             struct kvm_assigned_dev_kernel *dev)
 263 {
 264         dev->host_irq = dev->dev->irq;
 265         /* Even though this is PCI, we don't want to use shared
 266          * interrupts. Sharing host devices with guest-assigned devices
 267          * on the same interrupt line is not a happy situation: there
 268          * are going to be long delays in accepting, acking, etc.
 269          */
 270         if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
 271                         0, "kvm_assigned_intx_device", (void *)dev))
 272                 return -EIO;
 273         return 0;
 274 }
 275
 276 #ifdef __KVM_HAVE_MSI
 277 static int assigned_device_enable_host_msi(struct kvm *kvm,
 278                                            struct kvm_assigned_dev_kernel *dev)
 279 {
 280         int r;
 281
 282         if (!dev->dev->msi_enabled) {
 283                 r = pci_enable_msi(dev->dev);
 284                 if (r)
 285                         return r;
 286         }
 287
 288         dev->host_irq = dev->dev->irq;
 289         if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
 290                         "kvm_assigned_msi_device", (void *)dev)) {
 291                 pci_disable_msi(dev->dev);
 292                 return -EIO;
 293         }
 294
 295         return 0;
 296 }
 297 #endif
 298
 299 #ifdef __KVM_HAVE_MSIX
 300 static int assigned_device_enable_host_msix(struct kvm *kvm,
 301                                             struct kvm_assigned_dev_kernel *dev)
 302 {
 303         int i, r = -EINVAL;
 304
 305         /* host_msix_entries and guest_msix_entries should have been
 306          * initialized */
 307         if (dev->entries_nr == 0)
 308                 return r;
 309
 310         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 311         if (r)
 312                 return r;
 313
 314         for (i = 0; i < dev->entries_nr; i++) {
 315                 r = request_irq(dev->host_msix_entries[i].vector,
 316                                 kvm_assigned_dev_intr, 0,
 317                                 "kvm_assigned_msix_device",
 318                                 (void *)dev);
 319                 /* FIXME: free requested_irq's on failure */
 320                 if (r)
 321                         return r;
 322         }
 323
 324         return 0;
 325 }
 326
 327 #endif
 328
 329 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 330                                 struct kvm_assigned_dev_kernel *dev,
 331                                 struct kvm_assigned_irq *irq)
 332 {
 333         dev->guest_irq = irq->guest_irq;
 334         dev->ack_notifier.gsi = irq->guest_irq;
 335         return 0;
 336 }
 337
 338 #ifdef __KVM_HAVE_MSI
 339 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 340                         struct kvm_assigned_dev_kernel *dev,
 341                         struct kvm_assigned_irq *irq)
 342 {
 343         dev->guest_irq = irq->guest_irq;
 344         dev->ack_notifier.gsi = -1;
 345         dev->host_irq_disabled = false;
 346         return 0;
 347 }
 348 #endif
 349
 350 #ifdef __KVM_HAVE_MSIX
 351 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 352                         struct kvm_assigned_dev_kernel *dev,
 353                         struct kvm_assigned_irq *irq)
 354 {
 355         dev->guest_irq = irq->guest_irq;
 356         dev->ack_notifier.gsi = -1;
 357         dev->host_irq_disabled = false;
 358         return 0;
 359 }
 360 #endif
 361
 362 static int assign_host_irq(struct kvm *kvm,
 363                            struct kvm_assigned_dev_kernel *dev,
 364                            __u32 host_irq_type)
 365 {
 366         int r = -EEXIST;
 367
 368         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 369                 return r;
 370
 371         switch (host_irq_type) {
 372         case KVM_DEV_IRQ_HOST_INTX:
 373                 r = assigned_device_enable_host_intx(kvm, dev);
 374                 break;
 375 #ifdef __KVM_HAVE_MSI
 376         case KVM_DEV_IRQ_HOST_MSI:
 377                 r = assigned_device_enable_host_msi(kvm, dev);
 378                 break;
 379 #endif
 380 #ifdef __KVM_HAVE_MSIX
 381         case KVM_DEV_IRQ_HOST_MSIX:
 382                 r = assigned_device_enable_host_msix(kvm, dev);
 383                 break;
 384 #endif
 385         default:
 386                 r = -EINVAL;
 387         }
 388
 389         if (!r)
 390                 dev->irq_requested_type |= host_irq_type;
 391
 392         return r;
 393 }
 394
 395 static int assign_guest_irq(struct kvm *kvm,
 396                             struct kvm_assigned_dev_kernel *dev,
 397                             struct kvm_assigned_irq *irq,
 398                             unsigned long guest_irq_type)
 399 {
 400         int id;
 401         int r = -EEXIST;
 402
 403         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 404                 return r;
 405
 406         id = kvm_request_irq_source_id(kvm);
 407         if (id < 0)
 408                 return id;
 409
 410         dev->irq_source_id = id;
 411
 412         switch (guest_irq_type) {
 413         case KVM_DEV_IRQ_GUEST_INTX:
 414                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 415                 break;
 416 #ifdef __KVM_HAVE_MSI
 417         case KVM_DEV_IRQ_GUEST_MSI:
 418                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 419                 break;
 420 #endif
 421 #ifdef __KVM_HAVE_MSIX
 422         case KVM_DEV_IRQ_GUEST_MSIX:
 423                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 424                 break;
 425 #endif
 426         default:
 427                 r = -EINVAL;
 428         }
 429
 430         if (!r) {
 431                 dev->irq_requested_type |= guest_irq_type;
 432                 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 433         } else
 434                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 435
 436         return r;
 437 }
 438
 439 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 440 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 441                                    struct kvm_assigned_irq *assigned_irq)
 442 {
 443         int r = -EINVAL;
 444         struct kvm_assigned_dev_kernel *match;
 445         unsigned long host_irq_type, guest_irq_type;
 446
 447         if (!capable(CAP_SYS_RAWIO))
 448                 return -EPERM;
 449
 450         if (!irqchip_in_kernel(kvm))
 451                 return r;
 452
 453         mutex_lock(&kvm->lock);
 454         r = -ENODEV;
 455         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 456                                       assigned_irq->assigned_dev_id);
 457         if (!match)
 458                 goto out;
 459
 460         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 461         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 462
 463         r = -EINVAL;
 464         /* can only assign one type at a time */
 465         if (hweight_long(host_irq_type) > 1)
 466                 goto out;
 467         if (hweight_long(guest_irq_type) > 1)
 468                 goto out;
 469         if (host_irq_type == 0 && guest_irq_type == 0)
 470                 goto out;
 471
 472         r = 0;
 473         if (host_irq_type)
 474                 r = assign_host_irq(kvm, match, host_irq_type);
 475         if (r)
 476                 goto out;
 477
 478         if (guest_irq_type)
 479                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 480 out:
 481         mutex_unlock(&kvm->lock);
 482         return r;
 483 }
 484
 485 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 486                                          struct kvm_assigned_irq
 487                                          *assigned_irq)
 488 {
 489         int r = -ENODEV;
 490         struct kvm_assigned_dev_kernel *match;
 491
 492         mutex_lock(&kvm->lock);
 493
 494         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 495                                       assigned_irq->assigned_dev_id);
 496         if (!match)
 497                 goto out;
 498
 499         r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
 500 out:
 501         mutex_unlock(&kvm->lock);
 502         return r;
 503 }
 504
 505 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 506                                       struct kvm_assigned_pci_dev *assigned_dev)
 507 {
 508         int r = 0, idx;
 509         struct kvm_assigned_dev_kernel *match;
 510         struct pci_dev *dev;
 511
 512         mutex_lock(&kvm->lock);
 513         idx = srcu_read_lock(&kvm->srcu);
 514
 515         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 516                                       assigned_dev->assigned_dev_id);
 517         if (match) {
 518                 /* device already assigned */
 519                 r = -EEXIST;
 520                 goto out;
 521         }
 522
 523         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 524         if (match == NULL) {
 525                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 526                        __func__);
 527                 r = -ENOMEM;
 528                 goto out;
 529         }
 530         dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
 531                                    assigned_dev->busnr,
 532                                    assigned_dev->devfn);
 533         if (!dev) {
 534                 printk(KERN_INFO "%s: host device not found\n", __func__);
 535                 r = -EINVAL;
 536                 goto out_free;
 537         }
 538         if (pci_enable_device(dev)) {
 539                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 540                 r = -EBUSY;
 541                 goto out_put;
 542         }
 543         r = pci_request_regions(dev, "kvm_assigned_device");
 544         if (r) {
 545                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 546                        __func__);
 547                 goto out_disable;
 548         }
 549
 550         pci_reset_function(dev);
 551
 552         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 553         match->host_segnr = assigned_dev->segnr;
 554         match->host_busnr = assigned_dev->busnr;
 555         match->host_devfn = assigned_dev->devfn;
 556         match->flags = assigned_dev->flags;
 557         match->dev = dev;
 558         spin_lock_init(&match->assigned_dev_lock);
 559         match->irq_source_id = -1;
 560         match->kvm = kvm;
 561         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 562         INIT_WORK(&match->interrupt_work,
 563                   kvm_assigned_dev_interrupt_work_handler);
 564
 565         list_add(&match->list, &kvm->arch.assigned_dev_head);
 566
 567         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 568                 if (!kvm->arch.iommu_domain) {
 569                         r = kvm_iommu_map_guest(kvm);
 570                         if (r)
 571                                 goto out_list_del;
 572                 }
 573                 r = kvm_assign_device(kvm, match);
 574                 if (r)
 575                         goto out_list_del;
 576         }
 577
 578 out:
 579         srcu_read_unlock(&kvm->srcu, idx);
 580         mutex_unlock(&kvm->lock);
 581         return r;
 582 out_list_del:
 583         list_del(&match->list);
 584         pci_release_regions(dev);
 585 out_disable:
 586         pci_disable_device(dev);
 587 out_put:
 588         pci_dev_put(dev);
 589 out_free:
 590         kfree(match);
 591         srcu_read_unlock(&kvm->srcu, idx);
 592         mutex_unlock(&kvm->lock);
 593         return r;
 594 }
 595
 596 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 597                 struct kvm_assigned_pci_dev *assigned_dev)
 598 {
 599         int r = 0;
 600         struct kvm_assigned_dev_kernel *match;
 601
 602         mutex_lock(&kvm->lock);
 603
 604         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 605                                       assigned_dev->assigned_dev_id);
 606         if (!match) {
 607                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 608                   "so cannot be deassigned\n", __func__);
 609                 r = -EINVAL;
 610                 goto out;
 611         }
 612
 613         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 614                 kvm_deassign_device(kvm, match);
 615
 616         kvm_free_assigned_device(kvm, match);
 617
 618 out:
 619         mutex_unlock(&kvm->lock);
 620         return r;
 621 }
 622
 623
 624 #ifdef __KVM_HAVE_MSIX
 625 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
 626                                     struct kvm_assigned_msix_nr *entry_nr)
 627 {
 628         int r = 0;
 629         struct kvm_assigned_dev_kernel *adev;
 630
 631         mutex_lock(&kvm->lock);
 632
 633         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 634                                       entry_nr->assigned_dev_id);
 635         if (!adev) {
 636                 r = -EINVAL;
 637                 goto msix_nr_out;
 638         }
 639
 640         if (adev->entries_nr == 0) {
 641                 adev->entries_nr = entry_nr->entry_nr;
 642                 if (adev->entries_nr == 0 ||
 643                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
 644                         r = -EINVAL;
 645                         goto msix_nr_out;
 646                 }
 647
 648                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
 649                                                 entry_nr->entry_nr,
 650                                                 GFP_KERNEL);
 651                 if (!adev->host_msix_entries) {
 652                         r = -ENOMEM;
 653                         goto msix_nr_out;
 654                 }
 655                 adev->guest_msix_entries = kzalloc(
 656                                 sizeof(struct kvm_guest_msix_entry) *
 657                                 entry_nr->entry_nr, GFP_KERNEL);
 658                 if (!adev->guest_msix_entries) {
 659                         kfree(adev->host_msix_entries);
 660                         r = -ENOMEM;
 661                         goto msix_nr_out;
 662                 }
 663         } else /* Not allowed set MSI-X number twice */
 664                 r = -EINVAL;
 665 msix_nr_out:
 666         mutex_unlock(&kvm->lock);
 667         return r;
 668 }
 669
 670 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
 671                                        struct kvm_assigned_msix_entry *entry)
 672 {
 673         int r = 0, i;
 674         struct kvm_assigned_dev_kernel *adev;
 675
 676         mutex_lock(&kvm->lock);
 677
 678         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 679                                       entry->assigned_dev_id);
 680
 681         if (!adev) {
 682                 r = -EINVAL;
 683                 goto msix_entry_out;
 684         }
 685
 686         for (i = 0; i < adev->entries_nr; i++)
 687                 if (adev->guest_msix_entries[i].vector == 0 ||
 688                     adev->guest_msix_entries[i].entry == entry->entry) {
 689                         adev->guest_msix_entries[i].entry = entry->entry;
 690                         adev->guest_msix_entries[i].vector = entry->gsi;
 691                         adev->host_msix_entries[i].entry = entry->entry;
 692                         break;
 693                 }
 694         if (i == adev->entries_nr) {
 695                 r = -ENOSPC;
 696                 goto msix_entry_out;
 697         }
 698
 699 msix_entry_out:
 700         mutex_unlock(&kvm->lock);
 701
 702         return r;
 703 }
 704 #endif
 705
 706 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
 707                                   unsigned long arg)
 708 {
 709         void __user *argp = (void __user *)arg;
 710         int r = -ENOTTY;
 711
 712         switch (ioctl) {
 713         case KVM_ASSIGN_PCI_DEVICE: {
 714                 struct kvm_assigned_pci_dev assigned_dev;
 715
 716                 r = -EFAULT;
 717                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 718                         goto out;
 719                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
 720                 if (r)
 721                         goto out;
 722                 break;
 723         }
 724         case KVM_ASSIGN_IRQ: {
 725                 r = -EOPNOTSUPP;
 726                 break;
 727         }
 728 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
 729         case KVM_ASSIGN_DEV_IRQ: {
 730                 struct kvm_assigned_irq assigned_irq;
 731
 732                 r = -EFAULT;
 733                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 734                         goto out;
 735                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
 736                 if (r)
 737                         goto out;
 738                 break;
 739         }
 740         case KVM_DEASSIGN_DEV_IRQ: {
 741                 struct kvm_assigned_irq assigned_irq;
 742
 743                 r = -EFAULT;
 744                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 745                         goto out;
 746                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
 747                 if (r)
 748                         goto out;
 749                 break;
 750         }
 751 #endif
 752 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 753         case KVM_DEASSIGN_PCI_DEVICE: {
 754                 struct kvm_assigned_pci_dev assigned_dev;
 755
 756                 r = -EFAULT;
 757                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 758                         goto out;
 759                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
 760                 if (r)
 761                         goto out;
 762                 break;
 763         }
 764 #endif
 765 #ifdef KVM_CAP_IRQ_ROUTING
 766         case KVM_SET_GSI_ROUTING: {
 767                 struct kvm_irq_routing routing;
 768                 struct kvm_irq_routing __user *urouting;
 769                 struct kvm_irq_routing_entry *entries;
 770
 771                 r = -EFAULT;
 772                 if (copy_from_user(&routing, argp, sizeof(routing)))
 773                         goto out;
 774                 r = -EINVAL;
 775                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
 776                         goto out;
 777                 if (routing.flags)
 778                         goto out;
 779                 r = -ENOMEM;
 780                 entries = vmalloc(routing.nr * sizeof(*entries));
 781                 if (!entries)
 782                         goto out;
 783                 r = -EFAULT;
 784                 urouting = argp;
 785                 if (copy_from_user(entries, urouting->entries,
 786                                    routing.nr * sizeof(*entries)))
 787                         goto out_free_irq_routing;
 788                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
 789                                         routing.flags);
 790         out_free_irq_routing:
 791                 vfree(entries);
 792                 break;
 793         }
 794 #endif /* KVM_CAP_IRQ_ROUTING */
 795 #ifdef __KVM_HAVE_MSIX
 796         case KVM_ASSIGN_SET_MSIX_NR: {
 797                 struct kvm_assigned_msix_nr entry_nr;
 798                 r = -EFAULT;
 799                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
 800                         goto out;
 801                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
 802                 if (r)
 803                         goto out;
 804                 break;
 805         }
 806         case KVM_ASSIGN_SET_MSIX_ENTRY: {
 807                 struct kvm_assigned_msix_entry entry;
 808                 r = -EFAULT;
 809                 if (copy_from_user(&entry, argp, sizeof entry))
 810                         goto out;
 811                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
 812                 if (r)
 813                         goto out;
 814                 break;
 815         }
 816 #endif
 817         }
 818 out:
 819         return r;
 820 }
 821