hw/virtio-pci.c

   1 /*
   2  * Virtio PCI Bindings
   3  *
   4  * Copyright IBM, Corp. 2007
   5  * Copyright (c) 2009 CodeSourcery
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *  Paul Brook        <paul@codesourcery.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.  See
  12  * the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include <inttypes.h>
  17
  18 #include "virtio.h"
  19 #include "pci.h"
  20 #include "sysemu.h"
  21 #include "msix.h"
  22 #include "net.h"
  23
  24 /* from Linux's linux/virtio_pci.h */
  25
  26 /* A 32-bit r/o bitmask of the features supported by the host */
  27 #define VIRTIO_PCI_HOST_FEATURES        0
  28
  29 /* A 32-bit r/w bitmask of features activated by the guest */
  30 #define VIRTIO_PCI_GUEST_FEATURES       4
  31
  32 /* A 32-bit r/w PFN for the currently selected queue */
  33 #define VIRTIO_PCI_QUEUE_PFN            8
  34
  35 /* A 16-bit r/o queue size for the currently selected queue */
  36 #define VIRTIO_PCI_QUEUE_NUM            12
  37
  38 /* A 16-bit r/w queue selector */
  39 #define VIRTIO_PCI_QUEUE_SEL            14
  40
  41 /* A 16-bit r/w queue notifier */
  42 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  43
  44 /* An 8-bit device status register.  */
  45 #define VIRTIO_PCI_STATUS               18
  46
  47 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  48  * current contents of the ISR and will also clear it.  This is effectively
  49  * a read-and-acknowledge. */
  50 #define VIRTIO_PCI_ISR                  19
  51
  52 /* MSI-X registers: only enabled if MSI-X is enabled. */
  53 /* A 16-bit vector for configuration changes. */
  54 #define VIRTIO_MSI_CONFIG_VECTOR        20
  55 /* A 16-bit vector for selected queue notifications. */
  56 #define VIRTIO_MSI_QUEUE_VECTOR         22
  57
  58 /* Config space size */
  59 #define VIRTIO_PCI_CONFIG_NOMSI         20
  60 #define VIRTIO_PCI_CONFIG_MSI           24
  61 #define VIRTIO_PCI_REGION_SIZE(dev)     (msix_present(dev) ? \
  62                                          VIRTIO_PCI_CONFIG_MSI : \
  63                                          VIRTIO_PCI_CONFIG_NOMSI)
  64
  65 /* The remaining space is defined by each driver as the per-driver
  66  * configuration space */
  67 #define VIRTIO_PCI_CONFIG(dev)          (msix_enabled(dev) ? \
  68                                          VIRTIO_PCI_CONFIG_MSI : \
  69                                          VIRTIO_PCI_CONFIG_NOMSI)
  70
  71 /* Virtio ABI version, if we increment this, we break the guest driver. */
  72 #define VIRTIO_PCI_ABI_VERSION          0
  73
  74 /* How many bits to shift physical queue address written to QUEUE_PFN.
  75  * 12 is historical, and due to x86 page size. */
  76 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT    12
  77
  78 /* QEMU doesn't strictly need write barriers since everything runs in
  79  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  80  * KVM or if kqemu gets SMP support.
  81  */
  82 #define wmb() do { } while (0)
  83
  84 /* PCI bindings.  */
  85
  86 typedef struct {
  87     PCIDevice pci_dev;
  88     VirtIODevice *vdev;
  89     uint32_t addr;
  90     uint32_t class_code;
  91     uint32_t nvectors;
  92     DriveInfo *dinfo;
  93 } VirtIOPCIProxy;
  94
  95 /* virtio device */
  96
  97 static void virtio_pci_notify(void *opaque, uint16_t vector)
  98 {
  99     VirtIOPCIProxy *proxy = opaque;
 100     if (msix_enabled(&proxy->pci_dev))
 101         msix_notify(&proxy->pci_dev, vector);
 102     else
 103         qemu_set_irq(proxy->pci_dev.irq[0], proxy->vdev->isr & 1);
 104 }
 105
 106 static void virtio_pci_save_config(void * opaque, QEMUFile *f)
 107 {
 108     VirtIOPCIProxy *proxy = opaque;
 109     pci_device_save(&proxy->pci_dev, f);
 110     msix_save(&proxy->pci_dev, f);
 111     if (msix_present(&proxy->pci_dev))
 112         qemu_put_be16(f, proxy->vdev->config_vector);
 113 }
 114
 115 static void virtio_pci_save_queue(void * opaque, int n, QEMUFile *f)
 116 {
 117     VirtIOPCIProxy *proxy = opaque;
 118     if (msix_present(&proxy->pci_dev))
 119         qemu_put_be16(f, virtio_queue_vector(proxy->vdev, n));
 120 }
 121
 122 static int virtio_pci_load_config(void * opaque, QEMUFile *f)
 123 {
 124     VirtIOPCIProxy *proxy = opaque;
 125     int ret;
 126     ret = pci_device_load(&proxy->pci_dev, f);
 127     if (ret) {
 128         return ret;
 129     }
 130     msix_load(&proxy->pci_dev, f);
 131     if (msix_present(&proxy->pci_dev)) {
 132         qemu_get_be16s(f, &proxy->vdev->config_vector);
 133     } else {
 134         proxy->vdev->config_vector = VIRTIO_NO_VECTOR;
 135     }
 136     if (proxy->vdev->config_vector != VIRTIO_NO_VECTOR) {
 137         return msix_vector_use(&proxy->pci_dev, proxy->vdev->config_vector);
 138     }
 139     return 0;
 140 }
 141
 142 static int virtio_pci_load_queue(void * opaque, int n, QEMUFile *f)
 143 {
 144     VirtIOPCIProxy *proxy = opaque;
 145     uint16_t vector;
 146     if (msix_present(&proxy->pci_dev)) {
 147         qemu_get_be16s(f, &vector);
 148     } else {
 149         vector = VIRTIO_NO_VECTOR;
 150     }
 151     virtio_queue_set_vector(proxy->vdev, n, vector);
 152     if (vector != VIRTIO_NO_VECTOR) {
 153         return msix_vector_use(&proxy->pci_dev, vector);
 154     }
 155     return 0;
 156 }
 157
 158 static void virtio_pci_reset(void *opaque)
 159 {
 160     VirtIOPCIProxy *proxy = opaque;
 161     virtio_reset(proxy->vdev);
 162     msix_reset(&proxy->pci_dev);
 163 }
 164
 165 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 166 {
 167     VirtIOPCIProxy *proxy = opaque;
 168     VirtIODevice *vdev = proxy->vdev;
 169     target_phys_addr_t pa;
 170
 171     switch (addr) {
 172     case VIRTIO_PCI_GUEST_FEATURES:
 173         /* Guest does not negotiate properly?  We have to assume nothing. */
 174         if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 175             if (vdev->bad_features)
 176                 val = vdev->bad_features(vdev);
 177             else
 178                 val = 0;
 179         }
 180         if (vdev->set_features)
 181             vdev->set_features(vdev, val);
 182         vdev->features = val;
 183         break;
 184     case VIRTIO_PCI_QUEUE_PFN:
 185         pa = (target_phys_addr_t)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 186         if (pa == 0)
 187             virtio_pci_reset(proxy);
 188         else
 189             virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
 190         break;
 191     case VIRTIO_PCI_QUEUE_SEL:
 192         if (val < VIRTIO_PCI_QUEUE_MAX)
 193             vdev->queue_sel = val;
 194         break;
 195     case VIRTIO_PCI_QUEUE_NOTIFY:
 196         virtio_queue_notify(vdev, val);
 197         break;
 198     case VIRTIO_PCI_STATUS:
 199         vdev->status = val & 0xFF;
 200         if (vdev->status == 0)
 201             virtio_pci_reset(proxy);
 202         break;
 203     case VIRTIO_MSI_CONFIG_VECTOR:
 204         msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
 205         /* Make it possible for guest to discover an error took place. */
 206         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 207             val = VIRTIO_NO_VECTOR;
 208         vdev->config_vector = val;
 209         break;
 210     case VIRTIO_MSI_QUEUE_VECTOR:
 211         msix_vector_unuse(&proxy->pci_dev,
 212                           virtio_queue_vector(vdev, vdev->queue_sel));
 213         /* Make it possible for guest to discover an error took place. */
 214         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 215             val = VIRTIO_NO_VECTOR;
 216         virtio_queue_set_vector(vdev, vdev->queue_sel, val);
 217         break;
 218     default:
 219         fprintf(stderr, "%s: unexpected address 0x%x value 0x%x\n",
 220                 __func__, addr, val);
 221         break;
 222     }
 223 }
 224
 225 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr)
 226 {
 227     VirtIODevice *vdev = proxy->vdev;
 228     uint32_t ret = 0xFFFFFFFF;
 229
 230     switch (addr) {
 231     case VIRTIO_PCI_HOST_FEATURES:
 232         ret = vdev->get_features(vdev);
 233         ret |= (1 << VIRTIO_F_NOTIFY_ON_EMPTY);
 234         ret |= (1 << VIRTIO_RING_F_INDIRECT_DESC);
 235         ret |= (1 << VIRTIO_F_BAD_FEATURE);
 236         break;
 237     case VIRTIO_PCI_GUEST_FEATURES:
 238         ret = vdev->features;
 239         break;
 240     case VIRTIO_PCI_QUEUE_PFN:
 241         ret = virtio_queue_get_addr(vdev, vdev->queue_sel)
 242               >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 243         break;
 244     case VIRTIO_PCI_QUEUE_NUM:
 245         ret = virtio_queue_get_num(vdev, vdev->queue_sel);
 246         break;
 247     case VIRTIO_PCI_QUEUE_SEL:
 248         ret = vdev->queue_sel;
 249         break;
 250     case VIRTIO_PCI_STATUS:
 251         ret = vdev->status;
 252         break;
 253     case VIRTIO_PCI_ISR:
 254         /* reading from the ISR also clears it. */
 255         ret = vdev->isr;
 256         vdev->isr = 0;
 257         qemu_set_irq(proxy->pci_dev.irq[0], 0);
 258         break;
 259     case VIRTIO_MSI_CONFIG_VECTOR:
 260         ret = vdev->config_vector;
 261         break;
 262     case VIRTIO_MSI_QUEUE_VECTOR:
 263         ret = virtio_queue_vector(vdev, vdev->queue_sel);
 264         break;
 265     default:
 266         break;
 267     }
 268
 269     return ret;
 270 }
 271
 272 static uint32_t virtio_pci_config_readb(void *opaque, uint32_t addr)
 273 {
 274     VirtIOPCIProxy *proxy = opaque;
 275     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 276     addr -= proxy->addr;
 277     if (addr < config)
 278         return virtio_ioport_read(proxy, addr);
 279     addr -= config;
 280     return virtio_config_readb(proxy->vdev, addr);
 281 }
 282
 283 static uint32_t virtio_pci_config_readw(void *opaque, uint32_t addr)
 284 {
 285     VirtIOPCIProxy *proxy = opaque;
 286     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 287     addr -= proxy->addr;
 288     if (addr < config)
 289         return virtio_ioport_read(proxy, addr);
 290     addr -= config;
 291     return virtio_config_readw(proxy->vdev, addr);
 292 }
 293
 294 static uint32_t virtio_pci_config_readl(void *opaque, uint32_t addr)
 295 {
 296     VirtIOPCIProxy *proxy = opaque;
 297     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 298     addr -= proxy->addr;
 299     if (addr < config)
 300         return virtio_ioport_read(proxy, addr);
 301     addr -= config;
 302     return virtio_config_readl(proxy->vdev, addr);
 303 }
 304
 305 static void virtio_pci_config_writeb(void *opaque, uint32_t addr, uint32_t val)
 306 {
 307     VirtIOPCIProxy *proxy = opaque;
 308     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 309     addr -= proxy->addr;
 310     if (addr < config) {
 311         virtio_ioport_write(proxy, addr, val);
 312         return;
 313     }
 314     addr -= config;
 315     virtio_config_writeb(proxy->vdev, addr, val);
 316 }
 317
 318 static void virtio_pci_config_writew(void *opaque, uint32_t addr, uint32_t val)
 319 {
 320     VirtIOPCIProxy *proxy = opaque;
 321     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 322     addr -= proxy->addr;
 323     if (addr < config) {
 324         virtio_ioport_write(proxy, addr, val);
 325         return;
 326     }
 327     addr -= config;
 328     virtio_config_writew(proxy->vdev, addr, val);
 329 }
 330
 331 static void virtio_pci_config_writel(void *opaque, uint32_t addr, uint32_t val)
 332 {
 333     VirtIOPCIProxy *proxy = opaque;
 334     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 335     addr -= proxy->addr;
 336     if (addr < config) {
 337         virtio_ioport_write(proxy, addr, val);
 338         return;
 339     }
 340     addr -= config;
 341     virtio_config_writel(proxy->vdev, addr, val);
 342 }
 343
 344 static void virtio_map(PCIDevice *pci_dev, int region_num,
 345                        uint32_t addr, uint32_t size, int type)
 346 {
 347     VirtIOPCIProxy *proxy = container_of(pci_dev, VirtIOPCIProxy, pci_dev);
 348     VirtIODevice *vdev = proxy->vdev;
 349     unsigned config_len = VIRTIO_PCI_REGION_SIZE(pci_dev) + vdev->config_len;
 350
 351     proxy->addr = addr;
 352
 353     register_ioport_write(addr, config_len, 1, virtio_pci_config_writeb, proxy);
 354     register_ioport_write(addr, config_len, 2, virtio_pci_config_writew, proxy);
 355     register_ioport_write(addr, config_len, 4, virtio_pci_config_writel, proxy);
 356     register_ioport_read(addr, config_len, 1, virtio_pci_config_readb, proxy);
 357     register_ioport_read(addr, config_len, 2, virtio_pci_config_readw, proxy);
 358     register_ioport_read(addr, config_len, 4, virtio_pci_config_readl, proxy);
 359
 360     if (vdev->config_len)
 361         vdev->get_config(vdev, vdev->config);
 362 }
 363
 364 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address,
 365                                 uint32_t val, int len)
 366 {
 367     pci_default_write_config(pci_dev, address, val, len);
 368     msix_write_config(pci_dev, address, val, len);
 369 }
 370
 371 static const VirtIOBindings virtio_pci_bindings = {
 372     .notify = virtio_pci_notify,
 373     .save_config = virtio_pci_save_config,
 374     .load_config = virtio_pci_load_config,
 375     .save_queue = virtio_pci_save_queue,
 376     .load_queue = virtio_pci_load_queue,
 377 };
 378
 379 static void virtio_init_pci(VirtIOPCIProxy *proxy, VirtIODevice *vdev,
 380                             uint16_t vendor, uint16_t device,
 381                             uint16_t class_code, uint8_t pif)
 382 {
 383     uint8_t *config;
 384     uint32_t size;
 385
 386     proxy->vdev = vdev;
 387
 388     config = proxy->pci_dev.config;
 389     pci_config_set_vendor_id(config, vendor);
 390     pci_config_set_device_id(config, device);
 391
 392     config[0x08] = VIRTIO_PCI_ABI_VERSION;
 393
 394     config[0x09] = pif;
 395     pci_config_set_class(config, class_code);
 396     config[PCI_HEADER_TYPE] = PCI_HEADER_TYPE_NORMAL;
 397
 398     config[0x2c] = vendor & 0xFF;
 399     config[0x2d] = (vendor >> 8) & 0xFF;
 400     config[0x2e] = vdev->device_id & 0xFF;
 401     config[0x2f] = (vdev->device_id >> 8) & 0xFF;
 402
 403     config[0x3d] = 1;
 404
 405     if (vdev->nvectors && !msix_init(&proxy->pci_dev, vdev->nvectors, 1, 0)) {
 406         pci_register_bar(&proxy->pci_dev, 1,
 407                          msix_bar_size(&proxy->pci_dev),
 408                          PCI_ADDRESS_SPACE_MEM,
 409                          msix_mmio_map);
 410         proxy->pci_dev.config_write = virtio_write_config;
 411         proxy->pci_dev.unregister = msix_uninit;
 412     } else
 413         vdev->nvectors = 0;
 414
 415     size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev) + vdev->config_len;
 416     if (size & (size-1))
 417         size = 1 << qemu_fls(size);
 418
 419     pci_register_bar(&proxy->pci_dev, 0, size, PCI_ADDRESS_SPACE_IO,
 420                            virtio_map);
 421
 422     qemu_register_reset(virtio_pci_reset, proxy);
 423
 424     virtio_bind_device(vdev, &virtio_pci_bindings, proxy);
 425 }
 426
 427 static void virtio_blk_init_pci(PCIDevice *pci_dev)
 428 {
 429     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 430     VirtIODevice *vdev;
 431
 432     if (proxy->class_code != PCI_CLASS_STORAGE_SCSI &&
 433         proxy->class_code != PCI_CLASS_STORAGE_OTHER)
 434         proxy->class_code = PCI_CLASS_STORAGE_SCSI;
 435
 436     if (!proxy->dinfo) {
 437         fprintf(stderr, "drive property not set\n");
 438     }
 439     vdev = virtio_blk_init(&pci_dev->qdev, proxy->dinfo);
 440     vdev->nvectors = proxy->nvectors;
 441     virtio_init_pci(proxy, vdev,
 442                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 443                     PCI_DEVICE_ID_VIRTIO_BLOCK,
 444                     proxy->class_code, 0x00);
 445     /* make the actual value visible */
 446     proxy->nvectors = vdev->nvectors;
 447 }
 448
 449 static void virtio_console_init_pci(PCIDevice *pci_dev)
 450 {
 451     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 452     VirtIODevice *vdev;
 453
 454     if (proxy->class_code != PCI_CLASS_COMMUNICATION_OTHER &&
 455         proxy->class_code != PCI_CLASS_DISPLAY_OTHER && /* qemu 0.10 */
 456         proxy->class_code != PCI_CLASS_OTHERS)          /* qemu-kvm  */
 457         proxy->class_code = PCI_CLASS_COMMUNICATION_OTHER;
 458
 459     vdev = virtio_console_init(&pci_dev->qdev);
 460     virtio_init_pci(proxy, vdev,
 461                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 462                     PCI_DEVICE_ID_VIRTIO_CONSOLE,
 463                     proxy->class_code, 0x00);
 464 }
 465
 466 static void virtio_net_init_pci(PCIDevice *pci_dev)
 467 {
 468     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 469     VirtIODevice *vdev;
 470
 471     vdev = virtio_net_init(&pci_dev->qdev);
 472
 473     /* set nvectors from property, unless the user specified something
 474      * via -net nic,model=virtio,vectors=n command line option */
 475     if (pci_dev->qdev.nd->nvectors == NIC_NVECTORS_UNSPECIFIED)
 476         if (proxy->nvectors != NIC_NVECTORS_UNSPECIFIED)
 477             vdev->nvectors = proxy->nvectors;
 478
 479     virtio_init_pci(proxy, vdev,
 480                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 481                     PCI_DEVICE_ID_VIRTIO_NET,
 482                     PCI_CLASS_NETWORK_ETHERNET,
 483                     0x00);
 484
 485     /* make the actual value visible */
 486     proxy->nvectors = vdev->nvectors;
 487 }
 488
 489 static void virtio_balloon_init_pci(PCIDevice *pci_dev)
 490 {
 491     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 492     VirtIODevice *vdev;
 493
 494     vdev = virtio_balloon_init(&pci_dev->qdev);
 495     virtio_init_pci(proxy, vdev,
 496                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 497                     PCI_DEVICE_ID_VIRTIO_BALLOON,
 498                     PCI_CLASS_MEMORY_RAM,
 499                     0x00);
 500 }
 501
 502 static PCIDeviceInfo virtio_info[] = {
 503     {
 504         .qdev.name = "virtio-blk-pci",
 505         .qdev.size = sizeof(VirtIOPCIProxy),
 506         .init      = virtio_blk_init_pci,
 507         .qdev.props = (Property[]) {
 508             DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 509             DEFINE_PROP_DRIVE("drive", VirtIOPCIProxy, dinfo),
 510             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 2),
 511             DEFINE_PROP_END_OF_LIST(),
 512         },
 513     },{
 514         .qdev.name  = "virtio-net-pci",
 515         .qdev.size  = sizeof(VirtIOPCIProxy),
 516         .init       = virtio_net_init_pci,
 517         .qdev.props = (Property[]) {
 518             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
 519                                NIC_NVECTORS_UNSPECIFIED),
 520             DEFINE_PROP_END_OF_LIST(),
 521         },
 522     },{
 523         .qdev.name = "virtio-console-pci",
 524         .qdev.size = sizeof(VirtIOPCIProxy),
 525         .init      = virtio_console_init_pci,
 526         .qdev.props = (Property[]) {
 527             DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 528             DEFINE_PROP_END_OF_LIST(),
 529         },
 530     },{
 531         .qdev.name = "virtio-balloon-pci",
 532         .qdev.size = sizeof(VirtIOPCIProxy),
 533         .init      = virtio_balloon_init_pci,
 534     },{
 535         /* end of list */
 536     }
 537 };
 538
 539 static void virtio_pci_register_devices(void)
 540 {
 541     pci_qdev_register_many(virtio_info);
 542 }
 543
 544 device_init(virtio_pci_register_devices)