drivers/virtio/virtio_pci_common.c

   1 /*
   2  * Virtio PCI driver - common functionality for all device versions
   3  *
   4  * This module allows virtio devices to be used over a virtual PCI device.
   5  * This can be used with QEMU based VMMs like KVM or Xen.
   6  *
   7  * Copyright IBM Corp. 2007
   8  * Copyright Red Hat, Inc. 2014
   9  *
  10  * Authors:
  11  *  Anthony Liguori  <aliguori@us.ibm.com>
  12  *  Rusty Russell <rusty@rustcorp.com.au>
  13  *  Michael S. Tsirkin <mst@redhat.com>
  14  *
  15  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  16  * See the COPYING file in the top-level directory.
  17  *
  18  */
  19
  20 #include "virtio_pci_common.h"
  21
  22 static bool force_legacy = false;
  23
  24 #if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
  25 module_param(force_legacy, bool, 0444);
  26 MODULE_PARM_DESC(force_legacy,
  27                  "Force legacy mode for transitional virtio 1 devices");
  28 #endif
  29
  30 /* wait for pending irq handlers */
  31 void vp_synchronize_vectors(struct virtio_device *vdev)
  32 {
  33         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
  34         int i;
  35
  36         synchronize_irq(pci_irq_vector(vp_dev->pci_dev, 0));
  37         for (i = 1; i < vp_dev->msix_vectors; i++)
  38                 synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
  39 }
  40
  41 /* the notify function used when creating a virt queue */
  42 bool vp_notify(struct virtqueue *vq)
  43 {
  44         /* we write the queue's selector into the notification register to
  45          * signal the other end */
  46         iowrite16(vq->index, (void __iomem *)vq->priv);
  47         return true;
  48 }
  49
  50 /* Handle a configuration change: Tell driver if it wants to know. */
  51 static irqreturn_t vp_config_changed(int irq, void *opaque)
  52 {
  53         struct virtio_pci_device *vp_dev = opaque;
  54
  55         virtio_config_changed(&vp_dev->vdev);
  56         return IRQ_HANDLED;
  57 }
  58
  59 /* Notify all virtqueues on an interrupt. */
  60 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
  61 {
  62         struct virtio_pci_device *vp_dev = opaque;
  63         irqreturn_t ret = IRQ_NONE;
  64         struct virtqueue *vq;
  65
  66         list_for_each_entry(vq, &vp_dev->vdev.vqs, list) {
  67                 if (vq->callback && vring_interrupt(irq, vq) == IRQ_HANDLED)
  68                         ret = IRQ_HANDLED;
  69         }
  70
  71         return ret;
  72 }
  73
  74 /* A small wrapper to also acknowledge the interrupt when it's handled.
  75  * I really need an EIO hook for the vring so I can ack the interrupt once we
  76  * know that we'll be handling the IRQ but before we invoke the callback since
  77  * the callback may notify the host which results in the host attempting to
  78  * raise an interrupt that we would then mask once we acknowledged the
  79  * interrupt. */
  80 static irqreturn_t vp_interrupt(int irq, void *opaque)
  81 {
  82         struct virtio_pci_device *vp_dev = opaque;
  83         u8 isr;
  84
  85         /* reading the ISR has the effect of also clearing it so it's very
  86          * important to save off the value. */
  87         isr = ioread8(vp_dev->isr);
  88
  89         /* It's definitely not us if the ISR was not high */
  90         if (!isr)
  91                 return IRQ_NONE;
  92
  93         /* Configuration change?  Tell driver if it wants to know. */
  94         if (isr & VIRTIO_PCI_ISR_CONFIG)
  95                 vp_config_changed(irq, opaque);
  96
  97         return vp_vring_interrupt(irq, opaque);
  98 }
  99
 100 static void vp_remove_vqs(struct virtio_device *vdev)
 101 {
 102         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 103         struct virtqueue *vq, *n;
 104
 105         list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 106                 if (vp_dev->msix_vector_map) {
 107                         int v = vp_dev->msix_vector_map[vq->index];
 108
 109                         if (v != VIRTIO_MSI_NO_VECTOR)
 110                                 free_irq(pci_irq_vector(vp_dev->pci_dev, v),
 111                                         vq);
 112                 }
 113                 vp_dev->del_vq(vq);
 114         }
 115 }
 116
 117 /* the config->del_vqs() implementation */
 118 void vp_del_vqs(struct virtio_device *vdev)
 119 {
 120         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 121         int i;
 122
 123         if (WARN_ON_ONCE(list_empty_careful(&vdev->vqs)))
 124                 return;
 125
 126         vp_remove_vqs(vdev);
 127
 128         if (vp_dev->pci_dev->msix_enabled) {
 129                 for (i = 0; i < vp_dev->msix_vectors; i++)
 130                         free_cpumask_var(vp_dev->msix_affinity_masks[i]);
 131
 132                 /* Disable the vector used for configuration */
 133                 vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
 134
 135                 kfree(vp_dev->msix_affinity_masks);
 136                 kfree(vp_dev->msix_names);
 137                 kfree(vp_dev->msix_vector_map);
 138         }
 139
 140         free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
 141         pci_free_irq_vectors(vp_dev->pci_dev);
 142 }
 143
 144 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
 145                 struct virtqueue *vqs[], vq_callback_t *callbacks[],
 146                 const char * const names[], struct irq_affinity *desc)
 147 {
 148         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 149         const char *name = dev_name(&vp_dev->vdev.dev);
 150         int i, err = -ENOMEM, allocated_vectors, nvectors;
 151         unsigned flags = PCI_IRQ_MSIX;
 152         bool shared = false;
 153         u16 msix_vec;
 154
 155         if (desc) {
 156                 flags |= PCI_IRQ_AFFINITY;
 157                 desc->pre_vectors++; /* virtio config vector */
 158         }
 159
 160         nvectors = 1;
 161         for (i = 0; i < nvqs; i++)
 162                 if (callbacks[i])
 163                         nvectors++;
 164
 165         /* Try one vector per queue first. */
 166         err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
 167                         nvectors, flags, desc);
 168         if (err < 0) {
 169                 /* Fallback to one vector for config, one shared for queues. */
 170                 shared = true;
 171                 err = pci_alloc_irq_vectors(vp_dev->pci_dev, 2, 2,
 172                                 PCI_IRQ_MSIX);
 173                 if (err < 0)
 174                         return err;
 175         }
 176         if (err < 0)
 177                 return err;
 178
 179         vp_dev->msix_vectors = nvectors;
 180         vp_dev->msix_names = kmalloc_array(nvectors,
 181                         sizeof(*vp_dev->msix_names), GFP_KERNEL);
 182         if (!vp_dev->msix_names)
 183                 goto out_free_irq_vectors;
 184
 185         vp_dev->msix_affinity_masks = kcalloc(nvectors,
 186                         sizeof(*vp_dev->msix_affinity_masks), GFP_KERNEL);
 187         if (!vp_dev->msix_affinity_masks)
 188                 goto out_free_msix_names;
 189
 190         for (i = 0; i < nvectors; ++i) {
 191                 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
 192                                 GFP_KERNEL))
 193                         goto out_free_msix_affinity_masks;
 194         }
 195
 196         /* Set the vector used for configuration */
 197         snprintf(vp_dev->msix_names[0], sizeof(*vp_dev->msix_names),
 198                  "%s-config", name);
 199         err = request_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_config_changed,
 200                         0, vp_dev->msix_names[0], vp_dev);
 201         if (err)
 202                 goto out_free_msix_affinity_masks;
 203
 204         /* Verify we had enough resources to assign the vector */
 205         if (vp_dev->config_vector(vp_dev, 0) == VIRTIO_MSI_NO_VECTOR) {
 206                 err = -EBUSY;
 207                 goto out_free_config_irq;
 208         }
 209
 210         vp_dev->msix_vector_map = kmalloc_array(nvqs,
 211                         sizeof(*vp_dev->msix_vector_map), GFP_KERNEL);
 212         if (!vp_dev->msix_vector_map)
 213                 goto out_disable_config_irq;
 214
 215         allocated_vectors = 1; /* vector 0 is the config interrupt */
 216         for (i = 0; i < nvqs; ++i) {
 217                 if (!names[i]) {
 218                         vqs[i] = NULL;
 219                         continue;
 220                 }
 221
 222                 if (callbacks[i])
 223                         msix_vec = allocated_vectors;
 224                 else
 225                         msix_vec = VIRTIO_MSI_NO_VECTOR;
 226
 227                 vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
 228                                 msix_vec);
 229                 if (IS_ERR(vqs[i])) {
 230                         err = PTR_ERR(vqs[i]);
 231                         goto out_remove_vqs;
 232                 }
 233
 234                 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 235                         vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
 236                         continue;
 237                 }
 238
 239                 snprintf(vp_dev->msix_names[i + 1],
 240                          sizeof(*vp_dev->msix_names), "%s-%s",
 241                          dev_name(&vp_dev->vdev.dev), names[i]);
 242                 err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
 243                                   vring_interrupt, IRQF_SHARED,
 244                                   vp_dev->msix_names[i + 1], vqs[i]);
 245                 if (err) {
 246                         /* don't free this irq on error */
 247                         vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
 248                         goto out_remove_vqs;
 249                 }
 250                 vp_dev->msix_vector_map[i] = msix_vec;
 251
 252                 /*
 253                  * Use a different vector for each queue if they are available,
 254                  * else share the same vector for all VQs.
 255                  */
 256                 if (!shared)
 257                         allocated_vectors++;
 258         }
 259
 260         return 0;
 261
 262 out_remove_vqs:
 263         vp_remove_vqs(vdev);
 264         kfree(vp_dev->msix_vector_map);
 265 out_disable_config_irq:
 266         vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
 267 out_free_config_irq:
 268         free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
 269 out_free_msix_affinity_masks:
 270         for (i = 0; i < nvectors; i++) {
 271                 if (vp_dev->msix_affinity_masks[i])
 272                         free_cpumask_var(vp_dev->msix_affinity_masks[i]);
 273         }
 274         kfree(vp_dev->msix_affinity_masks);
 275 out_free_msix_names:
 276         kfree(vp_dev->msix_names);
 277 out_free_irq_vectors:
 278         pci_free_irq_vectors(vp_dev->pci_dev);
 279         return err;
 280 }
 281
 282 static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned nvqs,
 283                 struct virtqueue *vqs[], vq_callback_t *callbacks[],
 284                 const char * const names[])
 285 {
 286         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 287         int i, err;
 288
 289         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
 290                         dev_name(&vdev->dev), vp_dev);
 291         if (err)
 292                 return err;
 293
 294         for (i = 0; i < nvqs; ++i) {
 295                 if (!names[i]) {
 296                         vqs[i] = NULL;
 297                         continue;
 298                 }
 299                 vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
 300                                 VIRTIO_MSI_NO_VECTOR);
 301                 if (IS_ERR(vqs[i])) {
 302                         err = PTR_ERR(vqs[i]);
 303                         goto out_remove_vqs;
 304                 }
 305         }
 306
 307         return 0;
 308
 309 out_remove_vqs:
 310         vp_remove_vqs(vdev);
 311         free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
 312         return err;
 313 }
 314
 315 /* the config->find_vqs() implementation */
 316 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 317                 struct virtqueue *vqs[], vq_callback_t *callbacks[],
 318                 const char * const names[], struct irq_affinity *desc)
 319 {
 320         int err;
 321
 322         err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, desc);
 323         if (!err)
 324                 return 0;
 325         return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names);
 326 }
 327
 328 const char *vp_bus_name(struct virtio_device *vdev)
 329 {
 330         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 331
 332         return pci_name(vp_dev->pci_dev);
 333 }
 334
 335 /* Setup the affinity for a virtqueue:
 336  * - force the affinity for per vq vector
 337  * - OR over all affinities for shared MSI
 338  * - ignore the affinity request if we're using INTX
 339  */
 340 int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
 341 {
 342         struct virtio_device *vdev = vq->vdev;
 343         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 344
 345         if (!vq->callback)
 346                 return -EINVAL;
 347
 348         if (vp_dev->pci_dev->msix_enabled) {
 349                 int vec = vp_dev->msix_vector_map[vq->index];
 350                 struct cpumask *mask = vp_dev->msix_affinity_masks[vec];
 351                 unsigned int irq = pci_irq_vector(vp_dev->pci_dev, vec);
 352
 353                 if (cpu == -1)
 354                         irq_set_affinity_hint(irq, NULL);
 355                 else {
 356                         cpumask_clear(mask);
 357                         cpumask_set_cpu(cpu, mask);
 358                         irq_set_affinity_hint(irq, mask);
 359                 }
 360         }
 361         return 0;
 362 }
 363
 364 const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
 365 {
 366         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 367         unsigned int *map = vp_dev->msix_vector_map;
 368
 369         if (!map || map[index] == VIRTIO_MSI_NO_VECTOR)
 370                 return NULL;
 371
 372         return pci_irq_get_affinity(vp_dev->pci_dev, map[index]);
 373 }
 374
 375 #ifdef CONFIG_PM_SLEEP
 376 static int virtio_pci_freeze(struct device *dev)
 377 {
 378         struct pci_dev *pci_dev = to_pci_dev(dev);
 379         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 380         int ret;
 381
 382         ret = virtio_device_freeze(&vp_dev->vdev);
 383
 384         if (!ret)
 385                 pci_disable_device(pci_dev);
 386         return ret;
 387 }
 388
 389 static int virtio_pci_restore(struct device *dev)
 390 {
 391         struct pci_dev *pci_dev = to_pci_dev(dev);
 392         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 393         int ret;
 394
 395         ret = pci_enable_device(pci_dev);
 396         if (ret)
 397                 return ret;
 398
 399         pci_set_master(pci_dev);
 400         return virtio_device_restore(&vp_dev->vdev);
 401 }
 402
 403 static const struct dev_pm_ops virtio_pci_pm_ops = {
 404         SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
 405 };
 406 #endif
 407
 408
 409 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
 410 static const struct pci_device_id virtio_pci_id_table[] = {
 411         { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
 412         { 0 }
 413 };
 414
 415 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
 416
 417 static void virtio_pci_release_dev(struct device *_d)
 418 {
 419         struct virtio_device *vdev = dev_to_virtio(_d);
 420         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 421
 422         /* As struct device is a kobject, it's not safe to
 423          * free the memory (including the reference counter itself)
 424          * until it's release callback. */
 425         kfree(vp_dev);
 426 }
 427
 428 static int virtio_pci_probe(struct pci_dev *pci_dev,
 429                             const struct pci_device_id *id)
 430 {
 431         struct virtio_pci_device *vp_dev;
 432         int rc;
 433
 434         /* allocate our structure and fill it out */
 435         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 436         if (!vp_dev)
 437                 return -ENOMEM;
 438
 439         pci_set_drvdata(pci_dev, vp_dev);
 440         vp_dev->vdev.dev.parent = &pci_dev->dev;
 441         vp_dev->vdev.dev.release = virtio_pci_release_dev;
 442         vp_dev->pci_dev = pci_dev;
 443
 444         /* enable the device */
 445         rc = pci_enable_device(pci_dev);
 446         if (rc)
 447                 goto err_enable_device;
 448
 449         if (force_legacy) {
 450                 rc = virtio_pci_legacy_probe(vp_dev);
 451                 /* Also try modern mode if we can't map BAR0 (no IO space). */
 452                 if (rc == -ENODEV || rc == -ENOMEM)
 453                         rc = virtio_pci_modern_probe(vp_dev);
 454                 if (rc)
 455                         goto err_probe;
 456         } else {
 457                 rc = virtio_pci_modern_probe(vp_dev);
 458                 if (rc == -ENODEV)
 459                         rc = virtio_pci_legacy_probe(vp_dev);
 460                 if (rc)
 461                         goto err_probe;
 462         }
 463
 464         pci_set_master(pci_dev);
 465
 466         rc = register_virtio_device(&vp_dev->vdev);
 467         if (rc)
 468                 goto err_register;
 469
 470         return 0;
 471
 472 err_register:
 473         if (vp_dev->ioaddr)
 474              virtio_pci_legacy_remove(vp_dev);
 475         else
 476              virtio_pci_modern_remove(vp_dev);
 477 err_probe:
 478         pci_disable_device(pci_dev);
 479 err_enable_device:
 480         kfree(vp_dev);
 481         return rc;
 482 }
 483
 484 static void virtio_pci_remove(struct pci_dev *pci_dev)
 485 {
 486         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 487         struct device *dev = get_device(&vp_dev->vdev.dev);
 488
 489         unregister_virtio_device(&vp_dev->vdev);
 490
 491         if (vp_dev->ioaddr)
 492                 virtio_pci_legacy_remove(vp_dev);
 493         else
 494                 virtio_pci_modern_remove(vp_dev);
 495
 496         pci_disable_device(pci_dev);
 497         put_device(dev);
 498 }
 499
 500 static struct pci_driver virtio_pci_driver = {
 501         .name           = "virtio-pci",
 502         .id_table       = virtio_pci_id_table,
 503         .probe          = virtio_pci_probe,
 504         .remove         = virtio_pci_remove,
 505 #ifdef CONFIG_PM_SLEEP
 506         .driver.pm      = &virtio_pci_pm_ops,
 507 #endif
 508 };
 509
 510 module_pci_driver(virtio_pci_driver);
 511
 512 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
 513 MODULE_DESCRIPTION("virtio-pci");
 514 MODULE_LICENSE("GPL");
 515 MODULE_VERSION("1");