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(dev->dev, dev->host_msix_entries, dev->entries_nr);
 399         if (r)
 400                 return r;
 401
 402         for (i = 0; i < dev->entries_nr; i++) {
 403                 r = request_threaded_irq(dev->host_msix_entries[i].vector,
 404                                          kvm_assigned_dev_msix,
 405                                          kvm_assigned_dev_thread_msix,
 406                                          0, dev->irq_name, dev);
 407                 if (r)
 408                         goto err;
 409         }
 410
 411         return 0;
 412 err:
 413         for (i -= 1; i >= 0; i--)
 414                 free_irq(dev->host_msix_entries[i].vector, dev);
 415         pci_disable_msix(dev->dev);
 416         return r;
 417 }
 418
 419 #endif
 420
 421 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 422                                 struct kvm_assigned_dev_kernel *dev,
 423                                 struct kvm_assigned_irq *irq)
 424 {
 425         dev->guest_irq = irq->guest_irq;
 426         dev->ack_notifier.gsi = irq->guest_irq;
 427         return 0;
 428 }
 429
 430 #ifdef __KVM_HAVE_MSI
 431 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 432                         struct kvm_assigned_dev_kernel *dev,
 433                         struct kvm_assigned_irq *irq)
 434 {
 435         dev->guest_irq = irq->guest_irq;
 436         dev->ack_notifier.gsi = -1;
 437         return 0;
 438 }
 439 #endif
 440
 441 #ifdef __KVM_HAVE_MSIX
 442 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 443                         struct kvm_assigned_dev_kernel *dev,
 444                         struct kvm_assigned_irq *irq)
 445 {
 446         dev->guest_irq = irq->guest_irq;
 447         dev->ack_notifier.gsi = -1;
 448         return 0;
 449 }
 450 #endif
 451
 452 static int assign_host_irq(struct kvm *kvm,
 453                            struct kvm_assigned_dev_kernel *dev,
 454                            __u32 host_irq_type)
 455 {
 456         int r = -EEXIST;
 457
 458         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 459                 return r;
 460
 461         snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
 462                  pci_name(dev->dev));
 463
 464         switch (host_irq_type) {
 465         case KVM_DEV_IRQ_HOST_INTX:
 466                 r = assigned_device_enable_host_intx(kvm, dev);
 467                 break;
 468 #ifdef __KVM_HAVE_MSI
 469         case KVM_DEV_IRQ_HOST_MSI:
 470                 r = assigned_device_enable_host_msi(kvm, dev);
 471                 break;
 472 #endif
 473 #ifdef __KVM_HAVE_MSIX
 474         case KVM_DEV_IRQ_HOST_MSIX:
 475                 r = assigned_device_enable_host_msix(kvm, dev);
 476                 break;
 477 #endif
 478         default:
 479                 r = -EINVAL;
 480         }
 481         dev->host_irq_disabled = false;
 482
 483         if (!r)
 484                 dev->irq_requested_type |= host_irq_type;
 485
 486         return r;
 487 }
 488
 489 static int assign_guest_irq(struct kvm *kvm,
 490                             struct kvm_assigned_dev_kernel *dev,
 491                             struct kvm_assigned_irq *irq,
 492                             unsigned long guest_irq_type)
 493 {
 494         int id;
 495         int r = -EEXIST;
 496
 497         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 498                 return r;
 499
 500         id = kvm_request_irq_source_id(kvm);
 501         if (id < 0)
 502                 return id;
 503
 504         dev->irq_source_id = id;
 505
 506         switch (guest_irq_type) {
 507         case KVM_DEV_IRQ_GUEST_INTX:
 508                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 509                 break;
 510 #ifdef __KVM_HAVE_MSI
 511         case KVM_DEV_IRQ_GUEST_MSI:
 512                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 513                 break;
 514 #endif
 515 #ifdef __KVM_HAVE_MSIX
 516         case KVM_DEV_IRQ_GUEST_MSIX:
 517                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 518                 break;
 519 #endif
 520         default:
 521                 r = -EINVAL;
 522         }
 523
 524         if (!r) {
 525                 dev->irq_requested_type |= guest_irq_type;
 526                 if (dev->ack_notifier.gsi != -1)
 527                         kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 528         } else
 529                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 530
 531         return r;
 532 }
 533
 534 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 535 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 536                                    struct kvm_assigned_irq *assigned_irq)
 537 {
 538         int r = -EINVAL;
 539         struct kvm_assigned_dev_kernel *match;
 540         unsigned long host_irq_type, guest_irq_type;
 541
 542         if (!irqchip_in_kernel(kvm))
 543                 return r;
 544
 545         mutex_lock(&kvm->lock);
 546         r = -ENODEV;
 547         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 548                                       assigned_irq->assigned_dev_id);
 549         if (!match)
 550                 goto out;
 551
 552         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 553         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 554
 555         r = -EINVAL;
 556         /* can only assign one type at a time */
 557         if (hweight_long(host_irq_type) > 1)
 558                 goto out;
 559         if (hweight_long(guest_irq_type) > 1)
 560                 goto out;
 561         if (host_irq_type == 0 && guest_irq_type == 0)
 562                 goto out;
 563
 564         r = 0;
 565         if (host_irq_type)
 566                 r = assign_host_irq(kvm, match, host_irq_type);
 567         if (r)
 568                 goto out;
 569
 570         if (guest_irq_type)
 571                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 572 out:
 573         mutex_unlock(&kvm->lock);
 574         return r;
 575 }
 576
 577 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 578                                          struct kvm_assigned_irq
 579                                          *assigned_irq)
 580 {
 581         int r = -ENODEV;
 582         struct kvm_assigned_dev_kernel *match;
 583         unsigned long irq_type;
 584
 585         mutex_lock(&kvm->lock);
 586
 587         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 588                                       assigned_irq->assigned_dev_id);
 589         if (!match)
 590                 goto out;
 591
 592         irq_type = assigned_irq->flags & (KVM_DEV_IRQ_HOST_MASK |
 593                                           KVM_DEV_IRQ_GUEST_MASK);
 594         r = kvm_deassign_irq(kvm, match, irq_type);
 595 out:
 596         mutex_unlock(&kvm->lock);
 597         return r;
 598 }
 599
 600 /*
 601  * We want to test whether the caller has been granted permissions to
 602  * use this device.  To be able to configure and control the device,
 603  * the user needs access to PCI configuration space and BAR resources.
 604  * These are accessed through PCI sysfs.  PCI config space is often
 605  * passed to the process calling this ioctl via file descriptor, so we
 606  * can't rely on access to that file.  We can check for permissions
 607  * on each of the BAR resource files, which is a pretty clear
 608  * indicator that the user has been granted access to the device.
 609  */
 610 static int probe_sysfs_permissions(struct pci_dev *dev)
 611 {
 612 #ifdef CONFIG_SYSFS
 613         int i;
 614         bool bar_found = false;
 615
 616         for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++) {
 617                 char *kpath, *syspath;
 618                 struct path path;
 619                 struct inode *inode;
 620                 int r;
 621
 622                 if (!pci_resource_len(dev, i))
 623                         continue;
 624
 625                 kpath = kobject_get_path(&dev->dev.kobj, GFP_KERNEL);
 626                 if (!kpath)
 627                         return -ENOMEM;
 628
 629                 /* Per sysfs-rules, sysfs is always at /sys */
 630                 syspath = kasprintf(GFP_KERNEL, "/sys%s/resource%d", kpath, i);
 631                 kfree(kpath);
 632                 if (!syspath)
 633                         return -ENOMEM;
 634
 635                 r = kern_path(syspath, LOOKUP_FOLLOW, &path);
 636                 kfree(syspath);
 637                 if (r)
 638                         return r;
 639
 640                 inode = path.dentry->d_inode;
 641
 642                 r = inode_permission(inode, MAY_READ | MAY_WRITE | MAY_ACCESS);
 643                 path_put(&path);
 644                 if (r)
 645                         return r;
 646
 647                 bar_found = true;
 648         }
 649
 650         /* If no resources, probably something special */
 651         if (!bar_found)
 652                 return -EPERM;
 653
 654         return 0;
 655 #else
 656         return -EINVAL; /* No way to control the device without sysfs */
 657 #endif
 658 }
 659
 660 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 661                                       struct kvm_assigned_pci_dev *assigned_dev)
 662 {
 663         int r = 0, idx;
 664         struct kvm_assigned_dev_kernel *match;
 665         struct pci_dev *dev;
 666
 667         if (!(assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU))
 668                 return -EINVAL;
 669
 670         mutex_lock(&kvm->lock);
 671         idx = srcu_read_lock(&kvm->srcu);
 672
 673         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 674                                       assigned_dev->assigned_dev_id);
 675         if (match) {
 676                 /* device already assigned */
 677                 r = -EEXIST;
 678                 goto out;
 679         }
 680
 681         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 682         if (match == NULL) {
 683                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 684                        __func__);
 685                 r = -ENOMEM;
 686                 goto out;
 687         }
 688         dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
 689                                    assigned_dev->busnr,
 690                                    assigned_dev->devfn);
 691         if (!dev) {
 692                 printk(KERN_INFO "%s: host device not found\n", __func__);
 693                 r = -EINVAL;
 694                 goto out_free;
 695         }
 696
 697         /* Don't allow bridges to be assigned */
 698         if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) {
 699                 r = -EPERM;
 700                 goto out_put;
 701         }
 702
 703         r = probe_sysfs_permissions(dev);
 704         if (r)
 705                 goto out_put;
 706
 707         if (pci_enable_device(dev)) {
 708                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 709                 r = -EBUSY;
 710                 goto out_put;
 711         }
 712         r = pci_request_regions(dev, "kvm_assigned_device");
 713         if (r) {
 714                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 715                        __func__);
 716                 goto out_disable;
 717         }
 718
 719         pci_reset_function(dev);
 720         pci_save_state(dev);
 721         match->pci_saved_state = pci_store_saved_state(dev);
 722         if (!match->pci_saved_state)
 723                 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
 724                        __func__, dev_name(&dev->dev));
 725
 726         if (!pci_intx_mask_supported(dev))
 727                 assigned_dev->flags &= ~KVM_DEV_ASSIGN_PCI_2_3;
 728
 729         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 730         match->host_segnr = assigned_dev->segnr;
 731         match->host_busnr = assigned_dev->busnr;
 732         match->host_devfn = assigned_dev->devfn;
 733         match->flags = assigned_dev->flags;
 734         match->dev = dev;
 735         spin_lock_init(&match->intx_lock);
 736         spin_lock_init(&match->intx_mask_lock);
 737         match->irq_source_id = -1;
 738         match->kvm = kvm;
 739         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 740
 741         list_add(&match->list, &kvm->arch.assigned_dev_head);
 742
 743         if (!kvm->arch.iommu_domain) {
 744                 r = kvm_iommu_map_guest(kvm);
 745                 if (r)
 746                         goto out_list_del;
 747         }
 748         r = kvm_assign_device(kvm, match);
 749         if (r)
 750                 goto out_list_del;
 751
 752 out:
 753         srcu_read_unlock(&kvm->srcu, idx);
 754         mutex_unlock(&kvm->lock);
 755         return r;
 756 out_list_del:
 757         if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
 758                 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
 759                        __func__, dev_name(&dev->dev));
 760         list_del(&match->list);
 761         pci_release_regions(dev);
 762 out_disable:
 763         pci_disable_device(dev);
 764 out_put:
 765         pci_dev_put(dev);
 766 out_free:
 767         kfree(match);
 768         srcu_read_unlock(&kvm->srcu, idx);
 769         mutex_unlock(&kvm->lock);
 770         return r;
 771 }
 772
 773 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 774                 struct kvm_assigned_pci_dev *assigned_dev)
 775 {
 776         int r = 0;
 777         struct kvm_assigned_dev_kernel *match;
 778
 779         mutex_lock(&kvm->lock);
 780
 781         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 782                                       assigned_dev->assigned_dev_id);
 783         if (!match) {
 784                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 785                   "so cannot be deassigned\n", __func__);
 786                 r = -EINVAL;
 787                 goto out;
 788         }
 789
 790         kvm_deassign_device(kvm, match);
 791
 792         kvm_free_assigned_device(kvm, match);
 793
 794 out:
 795         mutex_unlock(&kvm->lock);
 796         return r;
 797 }
 798
 799
 800 #ifdef __KVM_HAVE_MSIX
 801 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
 802                                     struct kvm_assigned_msix_nr *entry_nr)
 803 {
 804         int r = 0;
 805         struct kvm_assigned_dev_kernel *adev;
 806
 807         mutex_lock(&kvm->lock);
 808
 809         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 810                                       entry_nr->assigned_dev_id);
 811         if (!adev) {
 812                 r = -EINVAL;
 813                 goto msix_nr_out;
 814         }
 815
 816         if (adev->entries_nr == 0) {
 817                 adev->entries_nr = entry_nr->entry_nr;
 818                 if (adev->entries_nr == 0 ||
 819                     adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
 820                         r = -EINVAL;
 821                         goto msix_nr_out;
 822                 }
 823
 824                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
 825                                                 entry_nr->entry_nr,
 826                                                 GFP_KERNEL);
 827                 if (!adev->host_msix_entries) {
 828                         r = -ENOMEM;
 829                         goto msix_nr_out;
 830                 }
 831                 adev->guest_msix_entries =
 832                         kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
 833                                 GFP_KERNEL);
 834                 if (!adev->guest_msix_entries) {
 835                         kfree(adev->host_msix_entries);
 836                         r = -ENOMEM;
 837                         goto msix_nr_out;
 838                 }
 839         } else /* Not allowed set MSI-X number twice */
 840                 r = -EINVAL;
 841 msix_nr_out:
 842         mutex_unlock(&kvm->lock);
 843         return r;
 844 }
 845
 846 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
 847                                        struct kvm_assigned_msix_entry *entry)
 848 {
 849         int r = 0, i;
 850         struct kvm_assigned_dev_kernel *adev;
 851
 852         mutex_lock(&kvm->lock);
 853
 854         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 855                                       entry->assigned_dev_id);
 856
 857         if (!adev) {
 858                 r = -EINVAL;
 859                 goto msix_entry_out;
 860         }
 861
 862         for (i = 0; i < adev->entries_nr; i++)
 863                 if (adev->guest_msix_entries[i].vector == 0 ||
 864                     adev->guest_msix_entries[i].entry == entry->entry) {
 865                         adev->guest_msix_entries[i].entry = entry->entry;
 866                         adev->guest_msix_entries[i].vector = entry->gsi;
 867                         adev->host_msix_entries[i].entry = entry->entry;
 868                         break;
 869                 }
 870         if (i == adev->entries_nr) {
 871                 r = -ENOSPC;
 872                 goto msix_entry_out;
 873         }
 874
 875 msix_entry_out:
 876         mutex_unlock(&kvm->lock);
 877
 878         return r;
 879 }
 880 #endif
 881
 882 static int kvm_vm_ioctl_set_pci_irq_mask(struct kvm *kvm,
 883                 struct kvm_assigned_pci_dev *assigned_dev)
 884 {
 885         int r = 0;
 886         struct kvm_assigned_dev_kernel *match;
 887
 888         mutex_lock(&kvm->lock);
 889
 890         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 891                                       assigned_dev->assigned_dev_id);
 892         if (!match) {
 893                 r = -ENODEV;
 894                 goto out;
 895         }
 896
 897         spin_lock(&match->intx_mask_lock);
 898
 899         match->flags &= ~KVM_DEV_ASSIGN_MASK_INTX;
 900         match->flags |= assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX;
 901
 902         if (match->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 903                 if (assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX) {
 904                         kvm_set_irq(match->kvm, match->irq_source_id,
 905                                     match->guest_irq, 0, false);
 906                         /*
 907                          * Masking at hardware-level is performed on demand,
 908                          * i.e. when an IRQ actually arrives at the host.
 909                          */
 910                 } else if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
 911                         /*
 912                          * Unmask the IRQ line if required. Unmasking at
 913                          * device level will be performed by user space.
 914                          */
 915                         spin_lock_irq(&match->intx_lock);
 916                         if (match->host_irq_disabled) {
 917                                 enable_irq(match->host_irq);
 918                                 match->host_irq_disabled = false;
 919                         }
 920                         spin_unlock_irq(&match->intx_lock);
 921                 }
 922         }
 923
 924         spin_unlock(&match->intx_mask_lock);
 925
 926 out:
 927         mutex_unlock(&kvm->lock);
 928         return r;
 929 }
 930
 931 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
 932                                   unsigned long arg)
 933 {
 934         void __user *argp = (void __user *)arg;
 935         int r;
 936
 937         switch (ioctl) {
 938         case KVM_ASSIGN_PCI_DEVICE: {
 939                 struct kvm_assigned_pci_dev assigned_dev;
 940
 941                 r = -EFAULT;
 942                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 943                         goto out;
 944                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
 945                 if (r)
 946                         goto out;
 947                 break;
 948         }
 949         case KVM_ASSIGN_IRQ: {
 950                 r = -EOPNOTSUPP;
 951                 break;
 952         }
 953         case KVM_ASSIGN_DEV_IRQ: {
 954                 struct kvm_assigned_irq assigned_irq;
 955
 956                 r = -EFAULT;
 957                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 958                         goto out;
 959                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
 960                 if (r)
 961                         goto out;
 962                 break;
 963         }
 964         case KVM_DEASSIGN_DEV_IRQ: {
 965                 struct kvm_assigned_irq assigned_irq;
 966
 967                 r = -EFAULT;
 968                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
 969                         goto out;
 970                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
 971                 if (r)
 972                         goto out;
 973                 break;
 974         }
 975         case KVM_DEASSIGN_PCI_DEVICE: {
 976                 struct kvm_assigned_pci_dev assigned_dev;
 977
 978                 r = -EFAULT;
 979                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
 980                         goto out;
 981                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
 982                 if (r)
 983                         goto out;
 984                 break;
 985         }
 986 #ifdef __KVM_HAVE_MSIX
 987         case KVM_ASSIGN_SET_MSIX_NR: {
 988                 struct kvm_assigned_msix_nr entry_nr;
 989                 r = -EFAULT;
 990                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
 991                         goto out;
 992                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
 993                 if (r)
 994                         goto out;
 995                 break;
 996         }
 997         case KVM_ASSIGN_SET_MSIX_ENTRY: {
 998                 struct kvm_assigned_msix_entry entry;
 999                 r = -EFAULT;
1000                 if (copy_from_user(&entry, argp, sizeof entry))
1001                         goto out;
1002                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
1003                 if (r)
1004                         goto out;
1005                 break;
1006         }
1007 #endif
1008         case KVM_ASSIGN_SET_INTX_MASK: {
1009                 struct kvm_assigned_pci_dev assigned_dev;
1010
1011                 r = -EFAULT;
1012                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1013                         goto out;
1014                 r = kvm_vm_ioctl_set_pci_irq_mask(kvm, &assigned_dev);
1015                 break;
1016         }
1017         default:
1018                 r = -ENOTTY;
1019                 break;
1020         }
1021 out:
1022         return r;
1023 }