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_msi(int irq, void *dev_id)
 109 {
 110         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
 111         int ret = kvm_set_irq_inatomic(assigned_dev->kvm,
 112                                        assigned_dev->irq_source_id,
 113                                        assigned_dev->guest_irq, 1);
 114         return unlikely(ret == -EWOULDBLOCK) ? IRQ_WAKE_THREAD : IRQ_HANDLED;
 115 }
 116
 117 static irqreturn_t kvm_assigned_dev_thread_msi(int irq, void *dev_id)
 118 {
 119         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
 120
 121         kvm_assigned_dev_raise_guest_irq(assigned_dev,
 122                                          assigned_dev->guest_irq);
 123
 124         return IRQ_HANDLED;
 125 }
 126 #endif
 127
 128 #ifdef __KVM_HAVE_MSIX
 129 static irqreturn_t kvm_assigned_dev_msix(int irq, void *dev_id)
 130 {
 131         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
 132         int index = find_index_from_host_irq(assigned_dev, irq);
 133         u32 vector;
 134         int ret = 0;
 135
 136         if (index >= 0) {
 137                 vector = assigned_dev->guest_msix_entries[index].vector;
 138                 ret = kvm_set_irq_inatomic(assigned_dev->kvm,
 139                                            assigned_dev->irq_source_id,
 140                                            vector, 1);
 141         }
 142
 143         return unlikely(ret == -EWOULDBLOCK) ? IRQ_WAKE_THREAD : IRQ_HANDLED;
 144 }
 145
 146 static irqreturn_t kvm_assigned_dev_thread_msix(int irq, void *dev_id)
 147 {
 148         struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
 149         int index = find_index_from_host_irq(assigned_dev, irq);
 150         u32 vector;
 151
 152         if (index >= 0) {
 153                 vector = assigned_dev->guest_msix_entries[index].vector;
 154                 kvm_assigned_dev_raise_guest_irq(assigned_dev, vector);
 155         }
 156
 157         return IRQ_HANDLED;
 158 }
 159 #endif
 160
 161 /* Ack the irq line for an assigned device */
 162 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 163 {
 164         struct kvm_assigned_dev_kernel *dev =
 165                 container_of(kian, struct kvm_assigned_dev_kernel,
 166                              ack_notifier);
 167
 168         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 169
 170         spin_lock(&dev->intx_mask_lock);
 171
 172         if (!(dev->flags & KVM_DEV_ASSIGN_MASK_INTX)) {
 173                 bool reassert = false;
 174
 175                 spin_lock_irq(&dev->intx_lock);
 176                 /*
 177                  * The guest IRQ may be shared so this ack can come from an
 178                  * IRQ for another guest device.
 179                  */
 180                 if (dev->host_irq_disabled) {
 181                         if (!(dev->flags & KVM_DEV_ASSIGN_PCI_2_3))
 182                                 enable_irq(dev->host_irq);
 183                         else if (!pci_check_and_unmask_intx(dev->dev))
 184                                 reassert = true;
 185                         dev->host_irq_disabled = reassert;
 186                 }
 187                 spin_unlock_irq(&dev->intx_lock);
 188
 189                 if (reassert)
 190                         kvm_set_irq(dev->kvm, dev->irq_source_id,
 191                                     dev->guest_irq, 1);
 192         }
 193
 194         spin_unlock(&dev->intx_mask_lock);
 195 }
 196
 197 static void deassign_guest_irq(struct kvm *kvm,
 198                                struct kvm_assigned_dev_kernel *assigned_dev)
 199 {
 200         if (assigned_dev->ack_notifier.gsi != -1)
 201                 kvm_unregister_irq_ack_notifier(kvm,
 202                                                 &assigned_dev->ack_notifier);
 203
 204         kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 205                     assigned_dev->guest_irq, 0);
 206
 207         if (assigned_dev->irq_source_id != -1)
 208                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 209         assigned_dev->irq_source_id = -1;
 210         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 211 }
 212
 213 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 214 static void deassign_host_irq(struct kvm *kvm,
 215                               struct kvm_assigned_dev_kernel *assigned_dev)
 216 {
 217         /*
 218          * We disable irq here to prevent further events.
 219          *
 220          * Notice this maybe result in nested disable if the interrupt type is
 221          * INTx, but it's OK for we are going to free it.
 222          *
 223          * If this function is a part of VM destroy, please ensure that till
 224          * now, the kvm state is still legal for probably we also have to wait
 225          * on a currently running IRQ handler.
 226          */
 227         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 228                 int i;
 229                 for (i = 0; i < assigned_dev->entries_nr; i++)
 230                         disable_irq(assigned_dev->host_msix_entries[i].vector);
 231
 232                 for (i = 0; i < assigned_dev->entries_nr; i++)
 233                         free_irq(assigned_dev->host_msix_entries[i].vector,
 234                                  assigned_dev);
 235
 236                 assigned_dev->entries_nr = 0;
 237                 kfree(assigned_dev->host_msix_entries);
 238                 kfree(assigned_dev->guest_msix_entries);
 239                 pci_disable_msix(assigned_dev->dev);
 240         } else {
 241                 /* Deal with MSI and INTx */
 242                 if ((assigned_dev->irq_requested_type &
 243                      KVM_DEV_IRQ_HOST_INTX) &&
 244                     (assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
 245                         spin_lock_irq(&assigned_dev->intx_lock);
 246                         pci_intx(assigned_dev->dev, false);
 247                         spin_unlock_irq(&assigned_dev->intx_lock);
 248                         synchronize_irq(assigned_dev->host_irq);
 249                 } else
 250                         disable_irq(assigned_dev->host_irq);
 251
 252                 free_irq(assigned_dev->host_irq, assigned_dev);
 253
 254                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 255                         pci_disable_msi(assigned_dev->dev);
 256         }
 257
 258         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 259 }
 260
 261 static int kvm_deassign_irq(struct kvm *kvm,
 262                             struct kvm_assigned_dev_kernel *assigned_dev,
 263                             unsigned long irq_requested_type)
 264 {
 265         unsigned long guest_irq_type, host_irq_type;
 266
 267         if (!irqchip_in_kernel(kvm))
 268                 return -EINVAL;
 269         /* no irq assignment to deassign */
 270         if (!assigned_dev->irq_requested_type)
 271                 return -ENXIO;
 272
 273         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 274         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 275
 276         if (host_irq_type)
 277                 deassign_host_irq(kvm, assigned_dev);
 278         if (guest_irq_type)
 279                 deassign_guest_irq(kvm, assigned_dev);
 280
 281         return 0;
 282 }
 283
 284 static void kvm_free_assigned_irq(struct kvm *kvm,
 285                                   struct kvm_assigned_dev_kernel *assigned_dev)
 286 {
 287         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 288 }
 289
 290 static void kvm_free_assigned_device(struct kvm *kvm,
 291                                      struct kvm_assigned_dev_kernel
 292                                      *assigned_dev)
 293 {
 294         kvm_free_assigned_irq(kvm, assigned_dev);
 295
 296         pci_reset_function(assigned_dev->dev);
 297         if (pci_load_and_free_saved_state(assigned_dev->dev,
 298                                           &assigned_dev->pci_saved_state))
 299                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
 300                        __func__, dev_name(&assigned_dev->dev->dev));
 301         else
 302                 pci_restore_state(assigned_dev->dev);
 303
 304         assigned_dev->dev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
 305
 306         pci_release_regions(assigned_dev->dev);
 307         pci_disable_device(assigned_dev->dev);
 308         pci_dev_put(assigned_dev->dev);
 309
 310         list_del(&assigned_dev->list);
 311         kfree(assigned_dev);
 312 }
 313
 314 void kvm_free_all_assigned_devices(struct kvm *kvm)
 315 {
 316         struct list_head *ptr, *ptr2;
 317         struct kvm_assigned_dev_kernel *assigned_dev;
 318
 319         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 320                 assigned_dev = list_entry(ptr,
 321                                           struct kvm_assigned_dev_kernel,
 322                                           list);
 323
 324                 kvm_free_assigned_device(kvm, assigned_dev);
 325         }
 326 }
 327
 328 static int assigned_device_enable_host_intx(struct kvm *kvm,
 329                                             struct kvm_assigned_dev_kernel *dev)
 330 {
 331         irq_handler_t irq_handler;
 332         unsigned long flags;
 333
 334         dev->host_irq = dev->dev->irq;
 335
 336         /*
 337          * We can only share the IRQ line with other host devices if we are
 338          * able to disable the IRQ source at device-level - independently of
 339          * the guest driver. Otherwise host devices may suffer from unbounded
 340          * IRQ latencies when the guest keeps the line asserted.
 341          */
 342         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
 343                 irq_handler = kvm_assigned_dev_intx;
 344                 flags = IRQF_SHARED;
 345         } else {
 346                 irq_handler = NULL;
 347                 flags = IRQF_ONESHOT;
 348         }
 349         if (request_threaded_irq(dev->host_irq, irq_handler,
 350                                  kvm_assigned_dev_thread_intx, flags,
 351                                  dev->irq_name, dev))
 352                 return -EIO;
 353
 354         if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
 355                 spin_lock_irq(&dev->intx_lock);
 356                 pci_intx(dev->dev, true);
 357                 spin_unlock_irq(&dev->intx_lock);
 358         }
 359         return 0;
 360 }
 361
 362 #ifdef __KVM_HAVE_MSI
 363 static int assigned_device_enable_host_msi(struct kvm *kvm,
 364                                            struct kvm_assigned_dev_kernel *dev)
 365 {
 366         int r;
 367
 368         if (!dev->dev->msi_enabled) {
 369                 r = pci_enable_msi(dev->dev);
 370                 if (r)
 371                         return r;
 372         }
 373
 374         dev->host_irq = dev->dev->irq;
 375         if (request_threaded_irq(dev->host_irq, kvm_assigned_dev_msi,
 376                                  kvm_assigned_dev_thread_msi, 0,
 377                                  dev->irq_name, dev)) {
 378                 pci_disable_msi(dev->dev);
 379                 return -EIO;
 380         }
 381
 382         return 0;
 383 }
 384 #endif
 385
 386 #ifdef __KVM_HAVE_MSIX
 387 static int assigned_device_enable_host_msix(struct kvm *kvm,
 388                                             struct kvm_assigned_dev_kernel *dev)
 389 {
 390         int i, r = -EINVAL;
 391
 392         /* host_msix_entries and guest_msix_entries should have been
 393          * initialized */
 394         if (dev->entries_nr == 0)
 395                 return r;
 396
 397         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 398         if (r)
 399                 return r;
 400
 401         for (i = 0; i < dev->entries_nr; i++) {
 402                 r = request_threaded_irq(dev->host_msix_entries[i].vector,
 403                                          kvm_assigned_dev_msix,
 404                                          kvm_assigned_dev_thread_msix,
 405                                          0, dev->irq_name, dev);
 406                 if (r)
 407                         goto err;
 408         }
 409
 410         return 0;
 411 err:
 412         for (i -= 1; i >= 0; i--)
 413                 free_irq(dev->host_msix_entries[i].vector, dev);
 414         pci_disable_msix(dev->dev);
 415         return r;
 416 }
 417
 418 #endif
 419
 420 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 421                                 struct kvm_assigned_dev_kernel *dev,
 422                                 struct kvm_assigned_irq *irq)
 423 {
 424         dev->guest_irq = irq->guest_irq;
 425         dev->ack_notifier.gsi = irq->guest_irq;
 426         return 0;
 427 }
 428
 429 #ifdef __KVM_HAVE_MSI
 430 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 431                         struct kvm_assigned_dev_kernel *dev,
 432                         struct kvm_assigned_irq *irq)
 433 {
 434         dev->guest_irq = irq->guest_irq;
 435         dev->ack_notifier.gsi = -1;
 436         return 0;
 437 }
 438 #endif
 439
 440 #ifdef __KVM_HAVE_MSIX
 441 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 442                         struct kvm_assigned_dev_kernel *dev,
 443                         struct kvm_assigned_irq *irq)
 444 {
 445         dev->guest_irq = irq->guest_irq;
 446         dev->ack_notifier.gsi = -1;
 447         return 0;
 448 }
 449 #endif
 450
 451 static int assign_host_irq(struct kvm *kvm,
 452                            struct kvm_assigned_dev_kernel *dev,
 453                            __u32 host_irq_type)
 454 {
 455         int r = -EEXIST;
 456
 457         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 458                 return r;
 459
 460         snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
 461                  pci_name(dev->dev));
 462
 463         switch (host_irq_type) {
 464         case KVM_DEV_IRQ_HOST_INTX:
 465                 r = assigned_device_enable_host_intx(kvm, dev);
 466                 break;
 467 #ifdef __KVM_HAVE_MSI
 468         case KVM_DEV_IRQ_HOST_MSI:
 469                 r = assigned_device_enable_host_msi(kvm, dev);
 470                 break;
 471 #endif
 472 #ifdef __KVM_HAVE_MSIX
 473         case KVM_DEV_IRQ_HOST_MSIX:
 474                 r = assigned_device_enable_host_msix(kvm, dev);
 475                 break;
 476 #endif
 477         default:
 478                 r = -EINVAL;
 479         }
 480         dev->host_irq_disabled = false;
 481
 482         if (!r)
 483                 dev->irq_requested_type |= host_irq_type;
 484
 485         return r;
 486 }
 487
 488 static int assign_guest_irq(struct kvm *kvm,
 489                             struct kvm_assigned_dev_kernel *dev,
 490                             struct kvm_assigned_irq *irq,
 491                             unsigned long guest_irq_type)
 492 {
 493         int id;
 494         int r = -EEXIST;
 495
 496         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 497                 return r;
 498
 499         id = kvm_request_irq_source_id(kvm);
 500         if (id < 0)
 501                 return id;
 502
 503         dev->irq_source_id = id;
 504
 505         switch (guest_irq_type) {
 506         case KVM_DEV_IRQ_GUEST_INTX:
 507                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 508                 break;
 509 #ifdef __KVM_HAVE_MSI
 510         case KVM_DEV_IRQ_GUEST_MSI:
 511                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 512                 break;
 513 #endif
 514 #ifdef __KVM_HAVE_MSIX
 515         case KVM_DEV_IRQ_GUEST_MSIX:
 516                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 517                 break;
 518 #endif
 519         default:
 520                 r = -EINVAL;
 521         }
 522
 523         if (!r) {
 524                 dev->irq_requested_type |= guest_irq_type;
 525                 if (dev->ack_notifier.gsi != -1)
 526                         kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 527         } else
 528                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 529
 530         return r;
 531 }
 532
 533 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 534 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 535                                    struct kvm_assigned_irq *assigned_irq)
 536 {
 537         int r = -EINVAL;
 538         struct kvm_assigned_dev_kernel *match;
 539         unsigned long host_irq_type, guest_irq_type;
 540
 541         if (!irqchip_in_kernel(kvm))
 542                 return r;
 543
 544         mutex_lock(&kvm->lock);
 545         r = -ENODEV;
 546         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 547                                       assigned_irq->assigned_dev_id);
 548         if (!match)
 549                 goto out;
 550
 551         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 552         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 553
 554         r = -EINVAL;
 555         /* can only assign one type at a time */
 556         if (hweight_long(host_irq_type) > 1)
 557                 goto out;
 558         if (hweight_long(guest_irq_type) > 1)
 559                 goto out;
 560         if (host_irq_type == 0 && guest_irq_type == 0)
 561                 goto out;
 562
 563         r = 0;
 564         if (host_irq_type)
 565                 r = assign_host_irq(kvm, match, host_irq_type);
 566         if (r)
 567                 goto out;
 568
 569         if (guest_irq_type)
 570                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 571 out:
 572         mutex_unlock(&kvm->lock);
 573         return r;
 574 }
 575
 576 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 577                                          struct kvm_assigned_irq
 578                                          *assigned_irq)
 579 {
 580         int r = -ENODEV;
 581         struct kvm_assigned_dev_kernel *match;
 582         unsigned long irq_type;
 583
 584         mutex_lock(&kvm->lock);
 585
 586         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 587                                       assigned_irq->assigned_dev_id);
 588         if (!match)
 589                 goto out;
 590
 591         irq_type = assigned_irq->flags & (KVM_DEV_IRQ_HOST_MASK |
 592                                           KVM_DEV_IRQ_GUEST_MASK);
 593         r = kvm_deassign_irq(kvm, match, irq_type);
 594 out:
 595         mutex_unlock(&kvm->lock);
 596         return r;
 597 }
 598
 599 /*
 600  * We want to test whether the caller has been granted permissions to
 601  * use this device.  To be able to configure and control the device,
 602  * the user needs access to PCI configuration space and BAR resources.
 603  * These are accessed through PCI sysfs.  PCI config space is often
 604  * passed to the process calling this ioctl via file descriptor, so we
 605  * can't rely on access to that file.  We can check for permissions
 606  * on each of the BAR resource files, which is a pretty clear
 607  * indicator that the user has been granted access to the device.
 608  */
 609 static int probe_sysfs_permissions(struct pci_dev *dev)
 610 {
 611 #ifdef CONFIG_SYSFS
 612         int i;
 613         bool bar_found = false;
 614
 615         for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++) {
 616                 char *kpath, *syspath;
 617                 struct path path;
 618                 struct inode *inode;
 619                 int r;
 620
 621                 if (!pci_resource_len(dev, i))
 622                         continue;
 623
 624                 kpath = kobject_get_path(&dev->dev.kobj, GFP_KERNEL);
 625                 if (!kpath)
 626                         return -ENOMEM;
 627
 628                 /* Per sysfs-rules, sysfs is always at /sys */
 629                 syspath = kasprintf(GFP_KERNEL, "/sys%s/resource%d", kpath, i);
 630                 kfree(kpath);
 631                 if (!syspath)
 632                         return -ENOMEM;
 633
 634                 r = kern_path(syspath, LOOKUP_FOLLOW, &path);
 635                 kfree(syspath);
 636                 if (r)
 637                         return r;
 638
 639                 inode = path.dentry->d_inode;
 640
 641                 r = inode_permission(inode, MAY_READ | MAY_WRITE | MAY_ACCESS);
 642                 path_put(&path);
 643                 if (r)
 644                         return r;
 645
 646                 bar_found = true;
 647         }
 648
 649         /* If no resources, probably something special */
 650         if (!bar_found)
 651                 return -EPERM;
 652
 653         return 0;
 654 #else
 655         return -EINVAL; /* No way to control the device without sysfs */
 656 #endif
 657 }
 658
 659 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 660                                       struct kvm_assigned_pci_dev *assigned_dev)
 661 {
 662         int r = 0, idx;
 663         struct kvm_assigned_dev_kernel *match;
 664         struct pci_dev *dev;
 665
 666         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU))
 667                 return -EINVAL;
 668
 669         mutex_lock(&kvm->lock);
 670         idx = srcu_read_lock(&kvm->srcu);
 671
 672         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 673                                       assigned_dev->assigned_dev_id);
 674         if (match) {
 675                 /* device already assigned */
 676                 r = -EEXIST;
 677                 goto out;
 678         }
 679
 680         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 681         if (match == NULL) {
 682                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 683                        __func__);
 684                 r = -ENOMEM;
 685                 goto out;
 686         }
 687         dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
 688                                    assigned_dev->busnr,
 689                                    assigned_dev->devfn);
 690         if (!dev) {
 691                 printk(KERN_INFO "%s: host device not found\n", __func__);
 692                 r = -EINVAL;
 693                 goto out_free;
 694         }
 695
 696         /* Don't allow bridges to be assigned */
 697         if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) {
 698                 r = -EPERM;
 699                 goto out_put;
 700         }
 701
 702         r = probe_sysfs_permissions(dev);
 703         if (r)
 704                 goto out_put;
 705
 706         if (pci_enable_device(dev)) {
 707                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 708                 r = -EBUSY;
 709                 goto out_put;
 710         }
 711         r = pci_request_regions(dev, "kvm_assigned_device");
 712         if (r) {
 713                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 714                        __func__);
 715                 goto out_disable;
 716         }
 717
 718         pci_reset_function(dev);
 719         pci_save_state(dev);
 720         match->pci_saved_state = pci_store_saved_state(dev);
 721         if (!match->pci_saved_state)
 722                 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
 723                        __func__, dev_name(&dev->dev));
 724
 725         if (!pci_intx_mask_supported(dev))
 726                 assigned_dev->flags &= ~KVM_DEV_ASSIGN_PCI_2_3;
 727
 728         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 729         match->host_segnr = assigned_dev->segnr;
 730         match->host_busnr = assigned_dev->busnr;
 731         match->host_devfn = assigned_dev->devfn;
 732         match->flags = assigned_dev->flags;
 733         match->dev = dev;
 734         spin_lock_init(&match->intx_lock);
 735         spin_lock_init(&match->intx_mask_lock);
 736         match->irq_source_id = -1;
 737         match->kvm = kvm;
 738         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 739
 740         list_add(&match->list, &kvm->arch.assigned_dev_head);
 741
 742         if (!kvm->arch.iommu_domain) {
 743                 r = kvm_iommu_map_guest(kvm);
 744                 if (r)
 745                         goto out_list_del;
 746         }
 747         r = kvm_assign_device(kvm, match);
 748         if (r)
 749                 goto out_list_del;
 750
 751 out:
 752         srcu_read_unlock(&kvm->srcu, idx);
 753         mutex_unlock(&kvm->lock);
 754         return r;
 755 out_list_del:
 756         if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
 757                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
 758                        __func__, dev_name(&dev->dev));
 759         list_del(&match->list);
 760         pci_release_regions(dev);
 761 out_disable:
 762         pci_disable_device(dev);
 763 out_put:
 764         pci_dev_put(dev);
 765 out_free:
 766         kfree(match);
 767         srcu_read_unlock(&kvm->srcu, idx);
 768         mutex_unlock(&kvm->lock);
 769         return r;
 770 }
 771
 772 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 773                 struct kvm_assigned_pci_dev *assigned_dev)
 774 {
 775         int r = 0;
 776         struct kvm_assigned_dev_kernel *match;
 777
 778         mutex_lock(&kvm->lock);
 779
 780         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 781                                       assigned_dev->assigned_dev_id);
 782         if (!match) {
 783                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 784                   "so cannot be deassigned\n", __func__);
 785                 r = -EINVAL;
 786                 goto out;
 787         }
 788
 789         kvm_deassign_device(kvm, match);
 790
 791         kvm_free_assigned_device(kvm, match);
 792
 793 out:
 794         mutex_unlock(&kvm->lock);
 795         return r;
 796 }
 797
 798
 799 #ifdef __KVM_HAVE_MSIX
 800 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
 801                                     struct kvm_assigned_msix_nr *entry_nr)
 802 {
 803         int r = 0;
 804         struct kvm_assigned_dev_kernel *adev;
 805
 806         mutex_lock(&kvm->lock);
 807
 808         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 809                                       entry_nr->assigned_dev_id);
 810         if (!adev) {
 811                 r = -EINVAL;
 812                 goto msix_nr_out;
 813         }
 814
 815         if (adev->entries_nr == 0) {
 816                 adev->entries_nr = entry_nr->entry_nr;
 817                 if (adev->entries_nr == 0 ||
 818                     adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
 819                         r = -EINVAL;
 820                         goto msix_nr_out;
 821                 }
 822
 823                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
 824                                                 entry_nr->entry_nr,
 825                                                 GFP_KERNEL);
 826                 if (!adev->host_msix_entries) {
 827                         r = -ENOMEM;
 828                         goto msix_nr_out;
 829                 }
 830                 adev->guest_msix_entries =
 831                         kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
 832                                 GFP_KERNEL);
 833                 if (!adev->guest_msix_entries) {
 834                         kfree(adev->host_msix_entries);
 835                         r = -ENOMEM;
 836                         goto msix_nr_out;
 837                 }
 838         } else /* Not allowed set MSI-X number twice */
 839                 r = -EINVAL;
 840 msix_nr_out:
 841         mutex_unlock(&kvm->lock);
 842         return r;
 843 }
 844
 845 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
 846                                        struct kvm_assigned_msix_entry *entry)
 847 {
 848         int r = 0, i;
 849         struct kvm_assigned_dev_kernel *adev;
 850
 851         mutex_lock(&kvm->lock);
 852
 853         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 854                                       entry->assigned_dev_id);
 855
 856         if (!adev) {
 857                 r = -EINVAL;
 858                 goto msix_entry_out;
 859         }
 860
 861         for (i = 0; i < adev->entries_nr; i++)
 862                 if (adev->guest_msix_entries[i].vector == 0 ||
 863                     adev->guest_msix_entries[i].entry == entry->entry) {
 864                         adev->guest_msix_entries[i].entry = entry->entry;
 865                         adev->guest_msix_entries[i].vector = entry->gsi;
 866                         adev->host_msix_entries[i].entry = entry->entry;
 867                         break;
 868                 }
 869         if (i == adev->entries_nr) {
 870                 r = -ENOSPC;
 871                 goto msix_entry_out;
 872         }
 873
 874 msix_entry_out:
 875         mutex_unlock(&kvm->lock);
 876
 877         return r;
 878 }
 879 #endif
 880
 881 static int kvm_vm_ioctl_set_pci_irq_mask(struct kvm *kvm,
 882                 struct kvm_assigned_pci_dev *assigned_dev)
 883 {
 884         int r = 0;
 885         struct kvm_assigned_dev_kernel *match;
 886
 887         mutex_lock(&kvm->lock);
 888
 889         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 890                                       assigned_dev->assigned_dev_id);
 891         if (!match) {
 892                 r = -ENODEV;
 893                 goto out;
 894         }
 895
 896         spin_lock(&match->intx_mask_lock);
 897
 898         match->flags &= ~KVM_DEV_ASSIGN_MASK_INTX;
 899         match->flags |= assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX;
 900
 901         if (match->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 902                 if (assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX) {
 903                         kvm_set_irq(match->kvm, match->irq_source_id,
 904                                     match->guest_irq, 0);
 905                         /*
 906                          * Masking at hardware-level is performed on demand,
 907                          * i.e. when an IRQ actually arrives at the host.
 908                          */
 909                 } else if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
 910                         /*
 911                          * Unmask the IRQ line if required. Unmasking at
 912                          * device level will be performed by user space.
 913                          */
 914                         spin_lock_irq(&match->intx_lock);
 915                         if (match->host_irq_disabled) {
 916                                 enable_irq(match->host_irq);
 917                                 match->host_irq_disabled = false;
 918                         }
 919                         spin_unlock_irq(&match->intx_lock);
 920                 }
 921         }
 922
 923         spin_unlock(&match->intx_mask_lock);
 924
 925 out:
 926         mutex_unlock(&kvm->lock);
 927         return r;
 928 }
 929
 930 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
 931                                   unsigned long arg)
 932 {
 933         void __user *argp = (void __user *)arg;
 934         int r;
 935
 936         switch (ioctl) {
 937         case KVM_ASSIGN_PCI_DEVICE: {
 938                 struct kvm_assigned_pci_dev assigned_dev;
 939
 940                 r = -EFAULT;
 941                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 942                         goto out;
 943                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
 944                 if (r)
 945                         goto out;
 946                 break;
 947         }
 948         case KVM_ASSIGN_IRQ: {
 949                 r = -EOPNOTSUPP;
 950                 break;
 951         }
 952         case KVM_ASSIGN_DEV_IRQ: {
 953                 struct kvm_assigned_irq assigned_irq;
 954
 955                 r = -EFAULT;
 956                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 957                         goto out;
 958                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
 959                 if (r)
 960                         goto out;
 961                 break;
 962         }
 963         case KVM_DEASSIGN_DEV_IRQ: {
 964                 struct kvm_assigned_irq assigned_irq;
 965
 966                 r = -EFAULT;
 967                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 968                         goto out;
 969                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
 970                 if (r)
 971                         goto out;
 972                 break;
 973         }
 974         case KVM_DEASSIGN_PCI_DEVICE: {
 975                 struct kvm_assigned_pci_dev assigned_dev;
 976
 977                 r = -EFAULT;
 978                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 979                         goto out;
 980                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
 981                 if (r)
 982                         goto out;
 983                 break;
 984         }
 985 #ifdef KVM_CAP_IRQ_ROUTING
 986         case KVM_SET_GSI_ROUTING: {
 987                 struct kvm_irq_routing routing;
 988                 struct kvm_irq_routing __user *urouting;
 989                 struct kvm_irq_routing_entry *entries;
 990
 991                 r = -EFAULT;
 992                 if (copy_from_user(&routing, argp, sizeof(routing)))
 993                         goto out;
 994                 r = -EINVAL;
 995                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
 996                         goto out;
 997                 if (routing.flags)
 998                         goto out;
 999                 r = -ENOMEM;
1000                 entries = vmalloc(routing.nr * sizeof(*entries));
1001                 if (!entries)
1002                         goto out;
1003                 r = -EFAULT;
1004                 urouting = argp;
1005                 if (copy_from_user(entries, urouting->entries,
1006                                    routing.nr * sizeof(*entries)))
1007                         goto out_free_irq_routing;
1008                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
1009                                         routing.flags);
1010         out_free_irq_routing:
1011                 vfree(entries);
1012                 break;
1013         }
1014 #endif /* KVM_CAP_IRQ_ROUTING */
1015 #ifdef __KVM_HAVE_MSIX
1016         case KVM_ASSIGN_SET_MSIX_NR: {
1017                 struct kvm_assigned_msix_nr entry_nr;
1018                 r = -EFAULT;
1019                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
1020                         goto out;
1021                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
1022                 if (r)
1023                         goto out;
1024                 break;
1025         }
1026         case KVM_ASSIGN_SET_MSIX_ENTRY: {
1027                 struct kvm_assigned_msix_entry entry;
1028                 r = -EFAULT;
1029                 if (copy_from_user(&entry, argp, sizeof entry))
1030                         goto out;
1031                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
1032                 if (r)
1033                         goto out;
1034                 break;
1035         }
1036 #endif
1037         case KVM_ASSIGN_SET_INTX_MASK: {
1038                 struct kvm_assigned_pci_dev assigned_dev;
1039
1040                 r = -EFAULT;
1041                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1042                         goto out;
1043                 r = kvm_vm_ioctl_set_pci_irq_mask(kvm, &assigned_dev);
1044                 break;
1045         }
1046         default:
1047                 r = -ENOTTY;
1048                 break;
1049         }
1050 out:
1051         return r;
1052 }