virt/kvm/assigned-dev.c

   1 /*
   2  * Kernel-based Virtual Machine - device assignment support
   3  *
   4  * Copyright (C) 2010 Red Hat, Inc. and/or its affiliates.
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2.  See
   7  * the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include <linux/kvm_host.h>
  12 #include <linux/kvm.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/errno.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/pci.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/slab.h>
  20 #include <linux/namei.h>
  21 #include <linux/fs.h>
  22 #include "irq.h"
  23
  24 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  25                                                       int assigned_dev_id)
  26 {
  27         struct list_head *ptr;
  28         struct kvm_assigned_dev_kernel *match;
  29
  30         list_for_each(ptr, head) {
  31                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
  32                 if (match->assigned_dev_id == assigned_dev_id)
  33                         return match;
  34         }
  35         return NULL;
  36 }
  37
  38 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
  39                                     *assigned_dev, int irq)
  40 {
  41         int i, index;
  42         struct msix_entry *host_msix_entries;
  43
  44         host_msix_entries = assigned_dev->host_msix_entries;
  45
  46         index = -1;
  47         for (i = 0; i < assigned_dev->entries_nr; i++)
  48                 if (irq == host_msix_entries[i].vector) {
  49                         index = i;
  50                         break;
  51                 }
  52         if (index < 0)
  53                 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
  54
  55         return index;
  56 }
  57
  58 static irqreturn_t kvm_assigned_dev_intx(int irq, void *dev_id)
  59 {
  60         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
  61         int ret;
  62
  63         spin_lock(&assigned_dev->intx_lock);
  64         if (pci_check_and_mask_intx(assigned_dev->dev)) {
  65                 assigned_dev->host_irq_disabled = true;
  66                 ret = IRQ_WAKE_THREAD;
  67         } else
  68                 ret = IRQ_NONE;
  69         spin_unlock(&assigned_dev->intx_lock);
  70
  71         return ret;
  72 }
  73
  74 static void
  75 kvm_assigned_dev_raise_guest_irq(struct kvm_assigned_dev_kernel *assigned_dev,
  76                                  int vector)
  77 {
  78         if (unlikely(assigned_dev->irq_requested_type &
  79                      KVM_DEV_IRQ_GUEST_INTX)) {
  80                 spin_lock(&assigned_dev->intx_mask_lock);
  81                 if (!(assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX))
  82                         kvm_set_irq(assigned_dev->kvm,
  83                                     assigned_dev->irq_source_id, vector, 1);
  84                 spin_unlock(&assigned_dev->intx_mask_lock);
  85         } else
  86                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
  87                             vector, 1);
  88 }
  89
  90 static irqreturn_t kvm_assigned_dev_thread_intx(int irq, void *dev_id)
  91 {
  92         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
  93
  94         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
  95                 spin_lock_irq(&assigned_dev->intx_lock);
  96                 disable_irq_nosync(irq);
  97                 assigned_dev->host_irq_disabled = true;
  98                 spin_unlock_irq(&assigned_dev->intx_lock);
  99         }
 100
 101         kvm_assigned_dev_raise_guest_irq(assigned_dev,
 102                                          assigned_dev->guest_irq);
 103
 104         return IRQ_HANDLED;
 105 }
 106
 107 #ifdef __KVM_HAVE_MSI
 108 static irqreturn_t kvm_assigned_dev_thread_msi(int irq, void *dev_id)
 109 {
 110         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
 111
 112         kvm_assigned_dev_raise_guest_irq(assigned_dev,
 113                                          assigned_dev->guest_irq);
 114
 115         return IRQ_HANDLED;
 116 }
 117 #endif
 118
 119 #ifdef __KVM_HAVE_MSIX
 120 static irqreturn_t kvm_assigned_dev_thread_msix(int irq, void *dev_id)
 121 {
 122         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
 123         int index = find_index_from_host_irq(assigned_dev, irq);
 124         u32 vector;
 125
 126         if (index >= 0) {
 127                 vector = assigned_dev->guest_msix_entries[index].vector;
 128                 kvm_assigned_dev_raise_guest_irq(assigned_dev, vector);
 129         }
 130
 131         return IRQ_HANDLED;
 132 }
 133 #endif
 134
 135 /* Ack the irq line for an assigned device */
 136 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 137 {
 138         struct kvm_assigned_dev_kernel *dev =
 139                 container_of(kian, struct kvm_assigned_dev_kernel,
 140                              ack_notifier);
 141
 142         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 143
 144         spin_lock(&dev->intx_mask_lock);
 145
 146         if (!(dev->flags & KVM_DEV_ASSIGN_MASK_INTX)) {
 147                 bool reassert = false;
 148
 149                 spin_lock_irq(&dev->intx_lock);
 150                 /*
 151                  * The guest IRQ may be shared so this ack can come from an
 152                  * IRQ for another guest device.
 153                  */
 154                 if (dev->host_irq_disabled) {
 155                         if (!(dev->flags & KVM_DEV_ASSIGN_PCI_2_3))
 156                                 enable_irq(dev->host_irq);
 157                         else if (!pci_check_and_unmask_intx(dev->dev))
 158                                 reassert = true;
 159                         dev->host_irq_disabled = reassert;
 160                 }
 161                 spin_unlock_irq(&dev->intx_lock);
 162
 163                 if (reassert)
 164                         kvm_set_irq(dev->kvm, dev->irq_source_id,
 165                                     dev->guest_irq, 1);
 166         }
 167
 168         spin_unlock(&dev->intx_mask_lock);
 169 }
 170
 171 static void deassign_guest_irq(struct kvm *kvm,
 172                                struct kvm_assigned_dev_kernel *assigned_dev)
 173 {
 174         if (assigned_dev->ack_notifier.gsi != -1)
 175                 kvm_unregister_irq_ack_notifier(kvm,
 176                                                 &assigned_dev->ack_notifier);
 177
 178         kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 179                     assigned_dev->guest_irq, 0);
 180
 181         if (assigned_dev->irq_source_id != -1)
 182                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 183         assigned_dev->irq_source_id = -1;
 184         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 185 }
 186
 187 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 188 static void deassign_host_irq(struct kvm *kvm,
 189                               struct kvm_assigned_dev_kernel *assigned_dev)
 190 {
 191         /*
 192          * We disable irq here to prevent further events.
 193          *
 194          * Notice this maybe result in nested disable if the interrupt type is
 195          * INTx, but it's OK for we are going to free it.
 196          *
 197          * If this function is a part of VM destroy, please ensure that till
 198          * now, the kvm state is still legal for probably we also have to wait
 199          * on a currently running IRQ handler.
 200          */
 201         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 202                 int i;
 203                 for (i = 0; i < assigned_dev->entries_nr; i++)
 204                         disable_irq(assigned_dev->host_msix_entries[i].vector);
 205
 206                 for (i = 0; i < assigned_dev->entries_nr; i++)
 207                         free_irq(assigned_dev->host_msix_entries[i].vector,
 208                                  assigned_dev);
 209
 210                 assigned_dev->entries_nr = 0;
 211                 kfree(assigned_dev->host_msix_entries);
 212                 kfree(assigned_dev->guest_msix_entries);
 213                 pci_disable_msix(assigned_dev->dev);
 214         } else {
 215                 /* Deal with MSI and INTx */
 216                 if ((assigned_dev->irq_requested_type &
 217                      KVM_DEV_IRQ_HOST_INTX) &&
 218                     (assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
 219                         spin_lock_irq(&assigned_dev->intx_lock);
 220                         pci_intx(assigned_dev->dev, false);
 221                         spin_unlock_irq(&assigned_dev->intx_lock);
 222                         synchronize_irq(assigned_dev->host_irq);
 223                 } else
 224                         disable_irq(assigned_dev->host_irq);
 225
 226                 free_irq(assigned_dev->host_irq, assigned_dev);
 227
 228                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 229                         pci_disable_msi(assigned_dev->dev);
 230         }
 231
 232         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 233 }
 234
 235 static int kvm_deassign_irq(struct kvm *kvm,
 236                             struct kvm_assigned_dev_kernel *assigned_dev,
 237                             unsigned long irq_requested_type)
 238 {
 239         unsigned long guest_irq_type, host_irq_type;
 240
 241         if (!irqchip_in_kernel(kvm))
 242                 return -EINVAL;
 243         /* no irq assignment to deassign */
 244         if (!assigned_dev->irq_requested_type)
 245                 return -ENXIO;
 246
 247         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 248         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 249
 250         if (host_irq_type)
 251                 deassign_host_irq(kvm, assigned_dev);
 252         if (guest_irq_type)
 253                 deassign_guest_irq(kvm, assigned_dev);
 254
 255         return 0;
 256 }
 257
 258 static void kvm_free_assigned_irq(struct kvm *kvm,
 259                                   struct kvm_assigned_dev_kernel *assigned_dev)
 260 {
 261         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 262 }
 263
 264 static void kvm_free_assigned_device(struct kvm *kvm,
 265                                      struct kvm_assigned_dev_kernel
 266                                      *assigned_dev)
 267 {
 268         kvm_free_assigned_irq(kvm, assigned_dev);
 269
 270         pci_reset_function(assigned_dev->dev);
 271         if (pci_load_and_free_saved_state(assigned_dev->dev,
 272                                           &assigned_dev->pci_saved_state))
 273                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
 274                        __func__, dev_name(&assigned_dev->dev->dev));
 275         else
 276                 pci_restore_state(assigned_dev->dev);
 277
 278         assigned_dev->dev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
 279
 280         pci_release_regions(assigned_dev->dev);
 281         pci_disable_device(assigned_dev->dev);
 282         pci_dev_put(assigned_dev->dev);
 283
 284         list_del(&assigned_dev->list);
 285         kfree(assigned_dev);
 286 }
 287
 288 void kvm_free_all_assigned_devices(struct kvm *kvm)
 289 {
 290         struct list_head *ptr, *ptr2;
 291         struct kvm_assigned_dev_kernel *assigned_dev;
 292
 293         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 294                 assigned_dev = list_entry(ptr,
 295                                           struct kvm_assigned_dev_kernel,
 296                                           list);
 297
 298                 kvm_free_assigned_device(kvm, assigned_dev);
 299         }
 300 }
 301
 302 static int assigned_device_enable_host_intx(struct kvm *kvm,
 303                                             struct kvm_assigned_dev_kernel *dev)
 304 {
 305         irq_handler_t irq_handler;
 306         unsigned long flags;
 307
 308         dev->host_irq = dev->dev->irq;
 309
 310         /*
 311          * We can only share the IRQ line with other host devices if we are
 312          * able to disable the IRQ source at device-level - independently of
 313          * the guest driver. Otherwise host devices may suffer from unbounded
 314          * IRQ latencies when the guest keeps the line asserted.
 315          */
 316         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
 317                 irq_handler = kvm_assigned_dev_intx;
 318                 flags = IRQF_SHARED;
 319         } else {
 320                 irq_handler = NULL;
 321                 flags = IRQF_ONESHOT;
 322         }
 323         if (request_threaded_irq(dev->host_irq, irq_handler,
 324                                  kvm_assigned_dev_thread_intx, flags,
 325                                  dev->irq_name, dev))
 326                 return -EIO;
 327
 328         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
 329                 spin_lock_irq(&dev->intx_lock);
 330                 pci_intx(dev->dev, true);
 331                 spin_unlock_irq(&dev->intx_lock);
 332         }
 333         return 0;
 334 }
 335
 336 #ifdef __KVM_HAVE_MSI
 337 static int assigned_device_enable_host_msi(struct kvm *kvm,
 338                                            struct kvm_assigned_dev_kernel *dev)
 339 {
 340         int r;
 341
 342         if (!dev->dev->msi_enabled) {
 343                 r = pci_enable_msi(dev->dev);
 344                 if (r)
 345                         return r;
 346         }
 347
 348         dev->host_irq = dev->dev->irq;
 349         if (request_threaded_irq(dev->host_irq, NULL,
 350                                  kvm_assigned_dev_thread_msi, 0,
 351                                  dev->irq_name, dev)) {
 352                 pci_disable_msi(dev->dev);
 353                 return -EIO;
 354         }
 355
 356         return 0;
 357 }
 358 #endif
 359
 360 #ifdef __KVM_HAVE_MSIX
 361 static int assigned_device_enable_host_msix(struct kvm *kvm,
 362                                             struct kvm_assigned_dev_kernel *dev)
 363 {
 364         int i, r = -EINVAL;
 365
 366         /* host_msix_entries and guest_msix_entries should have been
 367          * initialized */
 368         if (dev->entries_nr == 0)
 369                 return r;
 370
 371         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 372         if (r)
 373                 return r;
 374
 375         for (i = 0; i < dev->entries_nr; i++) {
 376                 r = request_threaded_irq(dev->host_msix_entries[i].vector,
 377                                          NULL, kvm_assigned_dev_thread_msix,
 378                                          0, dev->irq_name, dev);
 379                 if (r)
 380                         goto err;
 381         }
 382
 383         return 0;
 384 err:
 385         for (i -= 1; i >= 0; i--)
 386                 free_irq(dev->host_msix_entries[i].vector, dev);
 387         pci_disable_msix(dev->dev);
 388         return r;
 389 }
 390
 391 #endif
 392
 393 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 394                                 struct kvm_assigned_dev_kernel *dev,
 395                                 struct kvm_assigned_irq *irq)
 396 {
 397         dev->guest_irq = irq->guest_irq;
 398         dev->ack_notifier.gsi = irq->guest_irq;
 399         return 0;
 400 }
 401
 402 #ifdef __KVM_HAVE_MSI
 403 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 404                         struct kvm_assigned_dev_kernel *dev,
 405                         struct kvm_assigned_irq *irq)
 406 {
 407         dev->guest_irq = irq->guest_irq;
 408         dev->ack_notifier.gsi = -1;
 409         return 0;
 410 }
 411 #endif
 412
 413 #ifdef __KVM_HAVE_MSIX
 414 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 415                         struct kvm_assigned_dev_kernel *dev,
 416                         struct kvm_assigned_irq *irq)
 417 {
 418         dev->guest_irq = irq->guest_irq;
 419         dev->ack_notifier.gsi = -1;
 420         return 0;
 421 }
 422 #endif
 423
 424 static int assign_host_irq(struct kvm *kvm,
 425                            struct kvm_assigned_dev_kernel *dev,
 426                            __u32 host_irq_type)
 427 {
 428         int r = -EEXIST;
 429
 430         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 431                 return r;
 432
 433         snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
 434                  pci_name(dev->dev));
 435
 436         switch (host_irq_type) {
 437         case KVM_DEV_IRQ_HOST_INTX:
 438                 r = assigned_device_enable_host_intx(kvm, dev);
 439                 break;
 440 #ifdef __KVM_HAVE_MSI
 441         case KVM_DEV_IRQ_HOST_MSI:
 442                 r = assigned_device_enable_host_msi(kvm, dev);
 443                 break;
 444 #endif
 445 #ifdef __KVM_HAVE_MSIX
 446         case KVM_DEV_IRQ_HOST_MSIX:
 447                 r = assigned_device_enable_host_msix(kvm, dev);
 448                 break;
 449 #endif
 450         default:
 451                 r = -EINVAL;
 452         }
 453         dev->host_irq_disabled = false;
 454
 455         if (!r)
 456                 dev->irq_requested_type |= host_irq_type;
 457
 458         return r;
 459 }
 460
 461 static int assign_guest_irq(struct kvm *kvm,
 462                             struct kvm_assigned_dev_kernel *dev,
 463                             struct kvm_assigned_irq *irq,
 464                             unsigned long guest_irq_type)
 465 {
 466         int id;
 467         int r = -EEXIST;
 468
 469         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 470                 return r;
 471
 472         id = kvm_request_irq_source_id(kvm);
 473         if (id < 0)
 474                 return id;
 475
 476         dev->irq_source_id = id;
 477
 478         switch (guest_irq_type) {
 479         case KVM_DEV_IRQ_GUEST_INTX:
 480                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 481                 break;
 482 #ifdef __KVM_HAVE_MSI
 483         case KVM_DEV_IRQ_GUEST_MSI:
 484                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 485                 break;
 486 #endif
 487 #ifdef __KVM_HAVE_MSIX
 488         case KVM_DEV_IRQ_GUEST_MSIX:
 489                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 490                 break;
 491 #endif
 492         default:
 493                 r = -EINVAL;
 494         }
 495
 496         if (!r) {
 497                 dev->irq_requested_type |= guest_irq_type;
 498                 if (dev->ack_notifier.gsi != -1)
 499                         kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 500         } else
 501                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 502
 503         return r;
 504 }
 505
 506 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 507 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 508                                    struct kvm_assigned_irq *assigned_irq)
 509 {
 510         int r = -EINVAL;
 511         struct kvm_assigned_dev_kernel *match;
 512         unsigned long host_irq_type, guest_irq_type;
 513
 514         if (!irqchip_in_kernel(kvm))
 515                 return r;
 516
 517         mutex_lock(&kvm->lock);
 518         r = -ENODEV;
 519         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 520                                       assigned_irq->assigned_dev_id);
 521         if (!match)
 522                 goto out;
 523
 524         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 525         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 526
 527         r = -EINVAL;
 528         /* can only assign one type at a time */
 529         if (hweight_long(host_irq_type) > 1)
 530                 goto out;
 531         if (hweight_long(guest_irq_type) > 1)
 532                 goto out;
 533         if (host_irq_type == 0 && guest_irq_type == 0)
 534                 goto out;
 535
 536         r = 0;
 537         if (host_irq_type)
 538                 r = assign_host_irq(kvm, match, host_irq_type);
 539         if (r)
 540                 goto out;
 541
 542         if (guest_irq_type)
 543                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 544 out:
 545         mutex_unlock(&kvm->lock);
 546         return r;
 547 }
 548
 549 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 550                                          struct kvm_assigned_irq
 551                                          *assigned_irq)
 552 {
 553         int r = -ENODEV;
 554         struct kvm_assigned_dev_kernel *match;
 555         unsigned long irq_type;
 556
 557         mutex_lock(&kvm->lock);
 558
 559         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 560                                       assigned_irq->assigned_dev_id);
 561         if (!match)
 562                 goto out;
 563
 564         irq_type = assigned_irq->flags & (KVM_DEV_IRQ_HOST_MASK |
 565                                           KVM_DEV_IRQ_GUEST_MASK);
 566         r = kvm_deassign_irq(kvm, match, irq_type);
 567 out:
 568         mutex_unlock(&kvm->lock);
 569         return r;
 570 }
 571
 572 /*
 573  * We want to test whether the caller has been granted permissions to
 574  * use this device.  To be able to configure and control the device,
 575  * the user needs access to PCI configuration space and BAR resources.
 576  * These are accessed through PCI sysfs.  PCI config space is often
 577  * passed to the process calling this ioctl via file descriptor, so we
 578  * can't rely on access to that file.  We can check for permissions
 579  * on each of the BAR resource files, which is a pretty clear
 580  * indicator that the user has been granted access to the device.
 581  */
 582 static int probe_sysfs_permissions(struct pci_dev *dev)
 583 {
 584 #ifdef CONFIG_SYSFS
 585         int i;
 586         bool bar_found = false;
 587
 588         for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++) {
 589                 char *kpath, *syspath;
 590                 struct path path;
 591                 struct inode *inode;
 592                 int r;
 593
 594                 if (!pci_resource_len(dev, i))
 595                         continue;
 596
 597                 kpath = kobject_get_path(&dev->dev.kobj, GFP_KERNEL);
 598                 if (!kpath)
 599                         return -ENOMEM;
 600
 601                 /* Per sysfs-rules, sysfs is always at /sys */
 602                 syspath = kasprintf(GFP_KERNEL, "/sys%s/resource%d", kpath, i);
 603                 kfree(kpath);
 604                 if (!syspath)
 605                         return -ENOMEM;
 606
 607                 r = kern_path(syspath, LOOKUP_FOLLOW, &path);
 608                 kfree(syspath);
 609                 if (r)
 610                         return r;
 611
 612                 inode = path.dentry->d_inode;
 613
 614                 r = inode_permission(inode, MAY_READ | MAY_WRITE | MAY_ACCESS);
 615                 path_put(&path);
 616                 if (r)
 617                         return r;
 618
 619                 bar_found = true;
 620         }
 621
 622         /* If no resources, probably something special */
 623         if (!bar_found)
 624                 return -EPERM;
 625
 626         return 0;
 627 #else
 628         return -EINVAL; /* No way to control the device without sysfs */
 629 #endif
 630 }
 631
 632 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 633                                       struct kvm_assigned_pci_dev *assigned_dev)
 634 {
 635         int r = 0, idx;
 636         struct kvm_assigned_dev_kernel *match;
 637         struct pci_dev *dev;
 638
 639         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU))
 640                 return -EINVAL;
 641
 642         mutex_lock(&kvm->lock);
 643         idx = srcu_read_lock(&kvm->srcu);
 644
 645         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 646                                       assigned_dev->assigned_dev_id);
 647         if (match) {
 648                 /* device already assigned */
 649                 r = -EEXIST;
 650                 goto out;
 651         }
 652
 653         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 654         if (match == NULL) {
 655                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 656                        __func__);
 657                 r = -ENOMEM;
 658                 goto out;
 659         }
 660         dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
 661                                    assigned_dev->busnr,
 662                                    assigned_dev->devfn);
 663         if (!dev) {
 664                 printk(KERN_INFO "%s: host device not found\n", __func__);
 665                 r = -EINVAL;
 666                 goto out_free;
 667         }
 668
 669         /* Don't allow bridges to be assigned */
 670         if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) {
 671                 r = -EPERM;
 672                 goto out_put;
 673         }
 674
 675         r = probe_sysfs_permissions(dev);
 676         if (r)
 677                 goto out_put;
 678
 679         if (pci_enable_device(dev)) {
 680                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 681                 r = -EBUSY;
 682                 goto out_put;
 683         }
 684         r = pci_request_regions(dev, "kvm_assigned_device");
 685         if (r) {
 686                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 687                        __func__);
 688                 goto out_disable;
 689         }
 690
 691         pci_reset_function(dev);
 692         pci_save_state(dev);
 693         match->pci_saved_state = pci_store_saved_state(dev);
 694         if (!match->pci_saved_state)
 695                 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
 696                        __func__, dev_name(&dev->dev));
 697
 698         if (!pci_intx_mask_supported(dev))
 699                 assigned_dev->flags &= ~KVM_DEV_ASSIGN_PCI_2_3;
 700
 701         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 702         match->host_segnr = assigned_dev->segnr;
 703         match->host_busnr = assigned_dev->busnr;
 704         match->host_devfn = assigned_dev->devfn;
 705         match->flags = assigned_dev->flags;
 706         match->dev = dev;
 707         spin_lock_init(&match->intx_lock);
 708         spin_lock_init(&match->intx_mask_lock);
 709         match->irq_source_id = -1;
 710         match->kvm = kvm;
 711         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 712
 713         list_add(&match->list, &kvm->arch.assigned_dev_head);
 714
 715         if (!kvm->arch.iommu_domain) {
 716                 r = kvm_iommu_map_guest(kvm);
 717                 if (r)
 718                         goto out_list_del;
 719         }
 720         r = kvm_assign_device(kvm, match);
 721         if (r)
 722                 goto out_list_del;
 723
 724 out:
 725         srcu_read_unlock(&kvm->srcu, idx);
 726         mutex_unlock(&kvm->lock);
 727         return r;
 728 out_list_del:
 729         if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
 730                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
 731                        __func__, dev_name(&dev->dev));
 732         list_del(&match->list);
 733         pci_release_regions(dev);
 734 out_disable:
 735         pci_disable_device(dev);
 736 out_put:
 737         pci_dev_put(dev);
 738 out_free:
 739         kfree(match);
 740         srcu_read_unlock(&kvm->srcu, idx);
 741         mutex_unlock(&kvm->lock);
 742         return r;
 743 }
 744
 745 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 746                 struct kvm_assigned_pci_dev *assigned_dev)
 747 {
 748         int r = 0;
 749         struct kvm_assigned_dev_kernel *match;
 750
 751         mutex_lock(&kvm->lock);
 752
 753         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 754                                       assigned_dev->assigned_dev_id);
 755         if (!match) {
 756                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 757                   "so cannot be deassigned\n", __func__);
 758                 r = -EINVAL;
 759                 goto out;
 760         }
 761
 762         kvm_deassign_device(kvm, match);
 763
 764         kvm_free_assigned_device(kvm, match);
 765
 766 out:
 767         mutex_unlock(&kvm->lock);
 768         return r;
 769 }
 770
 771
 772 #ifdef __KVM_HAVE_MSIX
 773 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
 774                                     struct kvm_assigned_msix_nr *entry_nr)
 775 {
 776         int r = 0;
 777         struct kvm_assigned_dev_kernel *adev;
 778
 779         mutex_lock(&kvm->lock);
 780
 781         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 782                                       entry_nr->assigned_dev_id);
 783         if (!adev) {
 784                 r = -EINVAL;
 785                 goto msix_nr_out;
 786         }
 787
 788         if (adev->entries_nr == 0) {
 789                 adev->entries_nr = entry_nr->entry_nr;
 790                 if (adev->entries_nr == 0 ||
 791                     adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
 792                         r = -EINVAL;
 793                         goto msix_nr_out;
 794                 }
 795
 796                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
 797                                                 entry_nr->entry_nr,
 798                                                 GFP_KERNEL);
 799                 if (!adev->host_msix_entries) {
 800                         r = -ENOMEM;
 801                         goto msix_nr_out;
 802                 }
 803                 adev->guest_msix_entries =
 804                         kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
 805                                 GFP_KERNEL);
 806                 if (!adev->guest_msix_entries) {
 807                         kfree(adev->host_msix_entries);
 808                         r = -ENOMEM;
 809                         goto msix_nr_out;
 810                 }
 811         } else /* Not allowed set MSI-X number twice */
 812                 r = -EINVAL;
 813 msix_nr_out:
 814         mutex_unlock(&kvm->lock);
 815         return r;
 816 }
 817
 818 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
 819                                        struct kvm_assigned_msix_entry *entry)
 820 {
 821         int r = 0, i;
 822         struct kvm_assigned_dev_kernel *adev;
 823
 824         mutex_lock(&kvm->lock);
 825
 826         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 827                                       entry->assigned_dev_id);
 828
 829         if (!adev) {
 830                 r = -EINVAL;
 831                 goto msix_entry_out;
 832         }
 833
 834         for (i = 0; i < adev->entries_nr; i++)
 835                 if (adev->guest_msix_entries[i].vector == 0 ||
 836                     adev->guest_msix_entries[i].entry == entry->entry) {
 837                         adev->guest_msix_entries[i].entry = entry->entry;
 838                         adev->guest_msix_entries[i].vector = entry->gsi;
 839                         adev->host_msix_entries[i].entry = entry->entry;
 840                         break;
 841                 }
 842         if (i == adev->entries_nr) {
 843                 r = -ENOSPC;
 844                 goto msix_entry_out;
 845         }
 846
 847 msix_entry_out:
 848         mutex_unlock(&kvm->lock);
 849
 850         return r;
 851 }
 852 #endif
 853
 854 static int kvm_vm_ioctl_set_pci_irq_mask(struct kvm *kvm,
 855                 struct kvm_assigned_pci_dev *assigned_dev)
 856 {
 857         int r = 0;
 858         struct kvm_assigned_dev_kernel *match;
 859
 860         mutex_lock(&kvm->lock);
 861
 862         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 863                                       assigned_dev->assigned_dev_id);
 864         if (!match) {
 865                 r = -ENODEV;
 866                 goto out;
 867         }
 868
 869         spin_lock(&match->intx_mask_lock);
 870
 871         match->flags &= ~KVM_DEV_ASSIGN_MASK_INTX;
 872         match->flags |= assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX;
 873
 874         if (match->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 875                 if (assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX) {
 876                         kvm_set_irq(match->kvm, match->irq_source_id,
 877                                     match->guest_irq, 0);
 878                         /*
 879                          * Masking at hardware-level is performed on demand,
 880                          * i.e. when an IRQ actually arrives at the host.
 881                          */
 882                 } else if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
 883                         /*
 884                          * Unmask the IRQ line if required. Unmasking at
 885                          * device level will be performed by user space.
 886                          */
 887                         spin_lock_irq(&match->intx_lock);
 888                         if (match->host_irq_disabled) {
 889                                 enable_irq(match->host_irq);
 890                                 match->host_irq_disabled = false;
 891                         }
 892                         spin_unlock_irq(&match->intx_lock);
 893                 }
 894         }
 895
 896         spin_unlock(&match->intx_mask_lock);
 897
 898 out:
 899         mutex_unlock(&kvm->lock);
 900         return r;
 901 }
 902
 903 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
 904                                   unsigned long arg)
 905 {
 906         void __user *argp = (void __user *)arg;
 907         int r;
 908
 909         switch (ioctl) {
 910         case KVM_ASSIGN_PCI_DEVICE: {
 911                 struct kvm_assigned_pci_dev assigned_dev;
 912
 913                 r = -EFAULT;
 914                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 915                         goto out;
 916                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
 917                 if (r)
 918                         goto out;
 919                 break;
 920         }
 921         case KVM_ASSIGN_IRQ: {
 922                 r = -EOPNOTSUPP;
 923                 break;
 924         }
 925         case KVM_ASSIGN_DEV_IRQ: {
 926                 struct kvm_assigned_irq assigned_irq;
 927
 928                 r = -EFAULT;
 929                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 930                         goto out;
 931                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
 932                 if (r)
 933                         goto out;
 934                 break;
 935         }
 936         case KVM_DEASSIGN_DEV_IRQ: {
 937                 struct kvm_assigned_irq assigned_irq;
 938
 939                 r = -EFAULT;
 940                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 941                         goto out;
 942                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
 943                 if (r)
 944                         goto out;
 945                 break;
 946         }
 947         case KVM_DEASSIGN_PCI_DEVICE: {
 948                 struct kvm_assigned_pci_dev assigned_dev;
 949
 950                 r = -EFAULT;
 951                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 952                         goto out;
 953                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
 954                 if (r)
 955                         goto out;
 956                 break;
 957         }
 958 #ifdef KVM_CAP_IRQ_ROUTING
 959         case KVM_SET_GSI_ROUTING: {
 960                 struct kvm_irq_routing routing;
 961                 struct kvm_irq_routing __user *urouting;
 962                 struct kvm_irq_routing_entry *entries;
 963
 964                 r = -EFAULT;
 965                 if (copy_from_user(&routing, argp, sizeof(routing)))
 966                         goto out;
 967                 r = -EINVAL;
 968                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
 969                         goto out;
 970                 if (routing.flags)
 971                         goto out;
 972                 r = -ENOMEM;
 973                 entries = vmalloc(routing.nr * sizeof(*entries));
 974                 if (!entries)
 975                         goto out;
 976                 r = -EFAULT;
 977                 urouting = argp;
 978                 if (copy_from_user(entries, urouting->entries,
 979                                    routing.nr * sizeof(*entries)))
 980                         goto out_free_irq_routing;
 981                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
 982                                         routing.flags);
 983         out_free_irq_routing:
 984                 vfree(entries);
 985                 break;
 986         }
 987 #endif /* KVM_CAP_IRQ_ROUTING */
 988 #ifdef __KVM_HAVE_MSIX
 989         case KVM_ASSIGN_SET_MSIX_NR: {
 990                 struct kvm_assigned_msix_nr entry_nr;
 991                 r = -EFAULT;
 992                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
 993                         goto out;
 994                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
 995                 if (r)
 996                         goto out;
 997                 break;
 998         }
 999         case KVM_ASSIGN_SET_MSIX_ENTRY: {
1000                 struct kvm_assigned_msix_entry entry;
1001                 r = -EFAULT;
1002                 if (copy_from_user(&entry, argp, sizeof entry))
1003                         goto out;
1004                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
1005                 if (r)
1006                         goto out;
1007                 break;
1008         }
1009 #endif
1010         case KVM_ASSIGN_SET_INTX_MASK: {
1011                 struct kvm_assigned_pci_dev assigned_dev;
1012
1013                 r = -EFAULT;
1014                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1015                         goto out;
1016                 r = kvm_vm_ioctl_set_pci_irq_mask(kvm, &assigned_dev);
1017                 break;
1018         }
1019         default:
1020                 r = -ENOTTY;
1021                 break;
1022         }
1023 out:
1024         return r;
1025 }