hw/virtio.c

   1 /*
   2  * Virtio Support
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include <inttypes.h>
  15 #include <err.h>
  16
  17 #include "virtio.h"
  18 #include "sysemu.h"
  19
  20 /* from Linux's linux/virtio_pci.h */
  21
  22 /* A 32-bit r/o bitmask of the features supported by the host */
  23 #define VIRTIO_PCI_HOST_FEATURES        0
  24
  25 /* A 32-bit r/w bitmask of features activated by the guest */
  26 #define VIRTIO_PCI_GUEST_FEATURES       4
  27
  28 /* A 32-bit r/w PFN for the currently selected queue */
  29 #define VIRTIO_PCI_QUEUE_PFN            8
  30
  31 /* A 16-bit r/o queue size for the currently selected queue */
  32 #define VIRTIO_PCI_QUEUE_NUM            12
  33
  34 /* A 16-bit r/w queue selector */
  35 #define VIRTIO_PCI_QUEUE_SEL            14
  36
  37 /* A 16-bit r/w queue notifier */
  38 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  39
  40 /* An 8-bit device status register.  */
  41 #define VIRTIO_PCI_STATUS               18
  42
  43 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  44  * current contents of the ISR and will also clear it.  This is effectively
  45  * a read-and-acknowledge. */
  46 #define VIRTIO_PCI_ISR                  19
  47
  48 #define VIRTIO_PCI_CONFIG               20
  49
  50 /* QEMU doesn't strictly need write barriers since everything runs in
  51  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  52  * KVM or if kqemu gets SMP support.
  53  */
  54 #define wmb() do { } while (0)
  55
  56 /* virt queue functions */
  57
  58 static void virtqueue_init(VirtQueue *vq, void *p)
  59 {
  60     vq->vring.desc = p;
  61     vq->vring.avail = p + vq->vring.num * sizeof(VRingDesc);
  62     vq->vring.used = (void *)TARGET_PAGE_ALIGN((unsigned long)&vq->vring.avail->ring[vq->vring.num]);
  63 }
  64
  65 static unsigned virtqueue_next_desc(VirtQueue *vq, unsigned int i)
  66 {
  67     unsigned int next;
  68
  69     /* If this descriptor says it doesn't chain, we're done. */
  70     if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
  71         return vq->vring.num;
  72
  73     /* Check they're not leading us off end of descriptors. */
  74     next = vq->vring.desc[i].next;
  75     /* Make sure compiler knows to grab that: we don't want it changing! */
  76     wmb();
  77
  78     if (next >= vq->vring.num)
  79         errx(1, "Desc next is %u", next);
  80
  81     return next;
  82 }
  83
  84 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
  85                     unsigned int len)
  86 {
  87     VRingUsedElem *used;
  88
  89     /* Get a pointer to the next entry in the used ring. */
  90     used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
  91     used->id = elem->index;
  92     used->len = len;
  93     /* Make sure buffer is written before we update index. */
  94     wmb();
  95     vq->vring.used->idx++;
  96 }
  97
  98 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
  99 {
 100     unsigned int i, head;
 101     unsigned int position;
 102
 103     /* Check it isn't doing very strange things with descriptor numbers. */
 104     if ((uint16_t)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num)
 105         errx(1, "Guest moved used index from %u to %u",
 106              vq->last_avail_idx, vq->vring.avail->idx);
 107
 108     /* If there's nothing new since last we looked, return invalid. */
 109     if (vq->vring.avail->idx == vq->last_avail_idx)
 110         return 0;
 111
 112     /* Grab the next descriptor number they're advertising, and increment
 113      * the index we've seen. */
 114     head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num];
 115
 116     /* If their number is silly, that's a fatal mistake. */
 117     if (head >= vq->vring.num)
 118         errx(1, "Guest says index %u is available", head);
 119
 120     /* When we start there are none of either input nor output. */
 121     position = elem->out_num = elem->in_num = 0;
 122
 123     i = head;
 124     do {
 125         struct iovec *sg;
 126
 127         if ((vq->vring.desc[i].addr + vq->vring.desc[i].len) > ram_size)
 128             errx(1, "Guest sent invalid pointer");
 129
 130         if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
 131             sg = &elem->in_sg[elem->in_num++];
 132         else
 133             sg = &elem->out_sg[elem->out_num++];
 134
 135         /* Grab the first descriptor, and check it's OK. */
 136         sg->iov_len = vq->vring.desc[i].len;
 137         sg->iov_base = phys_ram_base + vq->vring.desc[i].addr;
 138
 139         /* If we've got too many, that implies a descriptor loop. */
 140         if ((elem->in_num + elem->out_num) > vq->vring.num)
 141             errx(1, "Looped descriptor");
 142     } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
 143
 144     elem->index = head;
 145
 146     return elem->in_num + elem->out_num;
 147 }
 148
 149 /* virtio device */
 150
 151 static VirtIODevice *to_virtio_device(PCIDevice *pci_dev)
 152 {
 153     return (VirtIODevice *)pci_dev;
 154 }
 155
 156 static void virtio_update_irq(VirtIODevice *vdev)
 157 {
 158     qemu_set_irq(vdev->pci_dev.irq[0], vdev->isr & 1);
 159 }
 160
 161 void virtio_reset(void *opaque)
 162 {
 163     VirtIODevice *vdev = opaque;
 164     int i;
 165
 166     vdev->features = 0;
 167     vdev->queue_sel = 0;
 168     vdev->status = 0;
 169     vdev->isr = 0;
 170
 171     for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 172         vdev->vq[i].vring.desc = NULL;
 173         vdev->vq[i].vring.avail = NULL;
 174         vdev->vq[i].vring.used = NULL;
 175         vdev->vq[i].last_avail_idx = 0;
 176         vdev->vq[i].pfn = 0;
 177     }
 178 }
 179
 180 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 181 {
 182     VirtIODevice *vdev = to_virtio_device(opaque);
 183     ram_addr_t pa;
 184
 185     addr -= vdev->addr;
 186
 187     switch (addr) {
 188     case VIRTIO_PCI_GUEST_FEATURES:
 189         if (vdev->set_features)
 190             vdev->set_features(vdev, val);
 191         vdev->features = val;
 192         break;
 193     case VIRTIO_PCI_QUEUE_PFN:
 194         pa = (ram_addr_t)val << TARGET_PAGE_BITS;
 195         vdev->vq[vdev->queue_sel].pfn = val;
 196         if (pa == 0) {
 197             virtio_reset(vdev);
 198         } else if (pa < (ram_size - TARGET_PAGE_SIZE)) {
 199             virtqueue_init(&vdev->vq[vdev->queue_sel], phys_ram_base + pa);
 200             /* FIXME if pa == 0, deal with device tear down */
 201         }
 202         break;
 203     case VIRTIO_PCI_QUEUE_SEL:
 204         if (val < VIRTIO_PCI_QUEUE_MAX)
 205             vdev->queue_sel = val;
 206         break;
 207     case VIRTIO_PCI_QUEUE_NOTIFY:
 208         if (val < VIRTIO_PCI_QUEUE_MAX && vdev->vq[val].vring.desc)
 209             vdev->vq[val].handle_output(vdev, &vdev->vq[val]);
 210         break;
 211     case VIRTIO_PCI_STATUS:
 212         vdev->status = val & 0xFF;
 213         break;
 214     }
 215 }
 216
 217 static uint32_t virtio_ioport_read(void *opaque, uint32_t addr)
 218 {
 219     VirtIODevice *vdev = to_virtio_device(opaque);
 220     uint32_t ret = 0xFFFFFFFF;
 221
 222     addr -= vdev->addr;
 223
 224     switch (addr) {
 225     case VIRTIO_PCI_HOST_FEATURES:
 226         ret = vdev->get_features(vdev);
 227         break;
 228     case VIRTIO_PCI_GUEST_FEATURES:
 229         ret = vdev->features;
 230         break;
 231     case VIRTIO_PCI_QUEUE_PFN:
 232         ret = vdev->vq[vdev->queue_sel].pfn;
 233         break;
 234     case VIRTIO_PCI_QUEUE_NUM:
 235         ret = vdev->vq[vdev->queue_sel].vring.num;
 236         break;
 237     case VIRTIO_PCI_QUEUE_SEL:
 238         ret = vdev->queue_sel;
 239         break;
 240     case VIRTIO_PCI_STATUS:
 241         ret = vdev->status;
 242         break;
 243     case VIRTIO_PCI_ISR:
 244         /* reading from the ISR also clears it. */
 245         ret = vdev->isr;
 246         vdev->isr = 0;
 247         virtio_update_irq(vdev);
 248         break;
 249     default:
 250         break;
 251     }
 252
 253     return ret;
 254 }
 255
 256 static uint32_t virtio_config_readb(void *opaque, uint32_t addr)
 257 {
 258     VirtIODevice *vdev = opaque;
 259     uint8_t val;
 260
 261     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 262     if (addr > (vdev->config_len - sizeof(val)))
 263         return (uint32_t)-1;
 264
 265     memcpy(&val, vdev->config + addr, sizeof(val));
 266     return val;
 267 }
 268
 269 static uint32_t virtio_config_readw(void *opaque, uint32_t addr)
 270 {
 271     VirtIODevice *vdev = opaque;
 272     uint16_t val;
 273
 274     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 275     if (addr > (vdev->config_len - sizeof(val)))
 276         return (uint32_t)-1;
 277
 278     memcpy(&val, vdev->config + addr, sizeof(val));
 279     return val;
 280 }
 281
 282 static uint32_t virtio_config_readl(void *opaque, uint32_t addr)
 283 {
 284     VirtIODevice *vdev = opaque;
 285     uint32_t val;
 286
 287     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 288     if (addr > (vdev->config_len - sizeof(val)))
 289         return (uint32_t)-1;
 290
 291     memcpy(&val, vdev->config + addr, sizeof(val));
 292     return val;
 293 }
 294
 295 static void virtio_config_writeb(void *opaque, uint32_t addr, uint32_t data)
 296 {
 297     VirtIODevice *vdev = opaque;
 298     uint8_t val = data;
 299
 300     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 301     if (addr > (vdev->config_len - sizeof(val)))
 302         return;
 303
 304     memcpy(vdev->config + addr, &val, sizeof(val));
 305 }
 306
 307 static void virtio_config_writew(void *opaque, uint32_t addr, uint32_t data)
 308 {
 309     VirtIODevice *vdev = opaque;
 310     uint16_t val = data;
 311
 312     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 313     if (addr > (vdev->config_len - sizeof(val)))
 314         return;
 315
 316     memcpy(vdev->config + addr, &val, sizeof(val));
 317 }
 318
 319 static void virtio_config_writel(void *opaque, uint32_t addr, uint32_t data)
 320 {
 321     VirtIODevice *vdev = opaque;
 322     uint32_t val = data;
 323
 324     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 325     if (addr > (vdev->config_len - sizeof(val)))
 326         return;
 327
 328     memcpy(vdev->config + addr, &val, sizeof(val));
 329 }
 330
 331 static void virtio_map(PCIDevice *pci_dev, int region_num,
 332                        uint32_t addr, uint32_t size, int type)
 333 {
 334     VirtIODevice *vdev = to_virtio_device(pci_dev);
 335     int i;
 336
 337     vdev->addr = addr;
 338     for (i = 0; i < 3; i++) {
 339         register_ioport_write(addr, 20, 1 << i, virtio_ioport_write, vdev);
 340         register_ioport_read(addr, 20, 1 << i, virtio_ioport_read, vdev);
 341     }
 342
 343     if (vdev->config_len) {
 344         register_ioport_write(addr + 20, vdev->config_len, 1,
 345                               virtio_config_writeb, vdev);
 346         register_ioport_write(addr + 20, vdev->config_len, 2,
 347                               virtio_config_writew, vdev);
 348         register_ioport_write(addr + 20, vdev->config_len, 4,
 349                               virtio_config_writel, vdev);
 350         register_ioport_read(addr + 20, vdev->config_len, 1,
 351                              virtio_config_readb, vdev);
 352         register_ioport_read(addr + 20, vdev->config_len, 2,
 353                              virtio_config_readw, vdev);
 354         register_ioport_read(addr + 20, vdev->config_len, 4,
 355                              virtio_config_readl, vdev);
 356
 357         vdev->update_config(vdev, vdev->config);
 358     }
 359 }
 360
 361 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
 362                             void (*handle_output)(VirtIODevice *, VirtQueue *))
 363 {
 364     int i;
 365
 366     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 367         if (vdev->vq[i].vring.num == 0)
 368             break;
 369     }
 370
 371     if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
 372         abort();
 373
 374     vdev->vq[i].vring.num = queue_size;
 375     vdev->vq[i].handle_output = handle_output;
 376     vdev->vq[i].index = i;
 377
 378     return &vdev->vq[i];
 379 }
 380
 381 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
 382 {
 383     /* Always notify when queue is empty */
 384     if (vq->vring.avail->idx != vq->last_avail_idx &&
 385         (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT))
 386         return;
 387
 388     vdev->isr = 1;
 389     virtio_update_irq(vdev);
 390 }
 391
 392 VirtIODevice *virtio_init_pci(PCIBus *bus, const char *name,
 393                               uint16_t vendor, uint16_t device,
 394                               uint16_t subvendor, uint16_t subdevice,
 395                               uint8_t class_code, uint8_t subclass_code,
 396                               uint8_t pif, size_t config_size,
 397                               size_t struct_size)
 398 {
 399     VirtIODevice *vdev;
 400     PCIDevice *pci_dev;
 401     uint8_t *config;
 402
 403     pci_dev = pci_register_device(bus, name, struct_size,
 404                                   -1, NULL, NULL);
 405     vdev = to_virtio_device(pci_dev);
 406
 407     vdev->status = 0;
 408     vdev->isr = 0;
 409     vdev->queue_sel = 0;
 410     memset(vdev->vq, 0, sizeof(vdev->vq));
 411
 412     config = pci_dev->config;
 413     config[0x00] = vendor & 0xFF;
 414     config[0x01] = (vendor >> 8) & 0xFF;
 415     config[0x02] = device & 0xFF;
 416     config[0x03] = (device >> 8) & 0xFF;
 417
 418     config[0x09] = pif;
 419     config[0x0a] = subclass_code;
 420     config[0x0b] = class_code;
 421     config[0x0e] = 0x00;
 422
 423     config[0x2c] = subvendor & 0xFF;
 424     config[0x2d] = (subvendor >> 8) & 0xFF;
 425     config[0x2e] = subdevice & 0xFF;
 426     config[0x2f] = (subdevice >> 8) & 0xFF;
 427
 428     config[0x3d] = 1;
 429
 430     vdev->name = name;
 431     vdev->config_len = config_size;
 432     if (vdev->config_len)
 433         vdev->config = qemu_mallocz(config_size);
 434     else
 435         vdev->config = NULL;
 436
 437     pci_register_io_region(pci_dev, 0, 20 + config_size, PCI_ADDRESS_SPACE_IO,
 438                            virtio_map);
 439     qemu_register_reset(virtio_reset, vdev);
 440
 441     return vdev;
 442 }