virt/kvm/assigned-dev.c

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