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
  16 #include "virtio.h"
  17 #include "sysemu.h"
  18
  19 /* from Linux's linux/virtio_pci.h */
  20
  21 /* A 32-bit r/o bitmask of the features supported by the host */
  22 #define VIRTIO_PCI_HOST_FEATURES        0
  23
  24 /* A 32-bit r/w bitmask of features activated by the guest */
  25 #define VIRTIO_PCI_GUEST_FEATURES       4
  26
  27 /* A 32-bit r/w PFN for the currently selected queue */
  28 #define VIRTIO_PCI_QUEUE_PFN            8
  29
  30 /* A 16-bit r/o queue size for the currently selected queue */
  31 #define VIRTIO_PCI_QUEUE_NUM            12
  32
  33 /* A 16-bit r/w queue selector */
  34 #define VIRTIO_PCI_QUEUE_SEL            14
  35
  36 /* A 16-bit r/w queue notifier */
  37 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  38
  39 /* An 8-bit device status register.  */
  40 #define VIRTIO_PCI_STATUS               18
  41
  42 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  43  * current contents of the ISR and will also clear it.  This is effectively
  44  * a read-and-acknowledge. */
  45 #define VIRTIO_PCI_ISR                  19
  46
  47 #define VIRTIO_PCI_CONFIG               20
  48
  49 /* Virtio ABI version, if we increment this, we break the guest driver. */
  50 #define VIRTIO_PCI_ABI_VERSION          0
  51
  52 /* How many bits to shift physical queue address written to QUEUE_PFN.
  53  * 12 is historical, and due to x86 page size. */
  54 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT    12
  55
  56 /* The alignment to use between consumer and producer parts of vring.
  57  * x86 pagesize again. */
  58 #define VIRTIO_PCI_VRING_ALIGN         4096
  59
  60 /* QEMU doesn't strictly need write barriers since everything runs in
  61  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  62  * KVM or if kqemu gets SMP support.
  63  */
  64 #define wmb() do { } while (0)
  65
  66 typedef struct VRingDesc
  67 {
  68     uint64_t addr;
  69     uint32_t len;
  70     uint16_t flags;
  71     uint16_t next;
  72 } VRingDesc;
  73
  74 typedef struct VRingAvail
  75 {
  76     uint16_t flags;
  77     uint16_t idx;
  78     uint16_t ring[0];
  79 } VRingAvail;
  80
  81 typedef struct VRingUsedElem
  82 {
  83     uint32_t id;
  84     uint32_t len;
  85 } VRingUsedElem;
  86
  87 typedef struct VRingUsed
  88 {
  89     uint16_t flags;
  90     uint16_t idx;
  91     VRingUsedElem ring[0];
  92 } VRingUsed;
  93
  94 typedef struct VRing
  95 {
  96     unsigned int num;
  97     target_phys_addr_t desc;
  98     target_phys_addr_t avail;
  99     target_phys_addr_t used;
 100 } VRing;
 101
 102 struct VirtQueue
 103 {
 104     VRing vring;
 105     uint32_t pfn;
 106     uint16_t last_avail_idx;
 107     int inuse;
 108     void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
 109 };
 110
 111 #define VIRTIO_PCI_QUEUE_MAX        16
 112
 113 /* virt queue functions */
 114 static void virtqueue_init(VirtQueue *vq, target_phys_addr_t pa)
 115 {
 116     vq->vring.desc = pa;
 117     vq->vring.avail = pa + vq->vring.num * sizeof(VRingDesc);
 118     vq->vring.used = vring_align(vq->vring.avail +
 119                                  offsetof(VRingAvail, ring[vq->vring.num]),
 120                                  VIRTIO_PCI_VRING_ALIGN);
 121 }
 122
 123 static inline uint64_t vring_desc_addr(VirtQueue *vq, int i)
 124 {
 125     target_phys_addr_t pa;
 126     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
 127     return ldq_phys(pa);
 128 }
 129
 130 static inline uint32_t vring_desc_len(VirtQueue *vq, int i)
 131 {
 132     target_phys_addr_t pa;
 133     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
 134     return ldl_phys(pa);
 135 }
 136
 137 static inline uint16_t vring_desc_flags(VirtQueue *vq, int i)
 138 {
 139     target_phys_addr_t pa;
 140     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
 141     return lduw_phys(pa);
 142 }
 143
 144 static inline uint16_t vring_desc_next(VirtQueue *vq, int i)
 145 {
 146     target_phys_addr_t pa;
 147     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
 148     return lduw_phys(pa);
 149 }
 150
 151 static inline uint16_t vring_avail_flags(VirtQueue *vq)
 152 {
 153     target_phys_addr_t pa;
 154     pa = vq->vring.avail + offsetof(VRingAvail, flags);
 155     return lduw_phys(pa);
 156 }
 157
 158 static inline uint16_t vring_avail_idx(VirtQueue *vq)
 159 {
 160     target_phys_addr_t pa;
 161     pa = vq->vring.avail + offsetof(VRingAvail, idx);
 162     return lduw_phys(pa);
 163 }
 164
 165 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
 166 {
 167     target_phys_addr_t pa;
 168     pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
 169     return lduw_phys(pa);
 170 }
 171
 172 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
 173 {
 174     target_phys_addr_t pa;
 175     pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
 176     stl_phys(pa, val);
 177 }
 178
 179 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
 180 {
 181     target_phys_addr_t pa;
 182     pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
 183     stl_phys(pa, val);
 184 }
 185
 186 static uint16_t vring_used_idx(VirtQueue *vq)
 187 {
 188     target_phys_addr_t pa;
 189     pa = vq->vring.used + offsetof(VRingUsed, idx);
 190     return lduw_phys(pa);
 191 }
 192
 193 static inline void vring_used_idx_increment(VirtQueue *vq, uint16_t val)
 194 {
 195     target_phys_addr_t pa;
 196     pa = vq->vring.used + offsetof(VRingUsed, idx);
 197     stw_phys(pa, vring_used_idx(vq) + val);
 198 }
 199
 200 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
 201 {
 202     target_phys_addr_t pa;
 203     pa = vq->vring.used + offsetof(VRingUsed, flags);
 204     stw_phys(pa, lduw_phys(pa) | mask);
 205 }
 206
 207 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
 208 {
 209     target_phys_addr_t pa;
 210     pa = vq->vring.used + offsetof(VRingUsed, flags);
 211     stw_phys(pa, lduw_phys(pa) & ~mask);
 212 }
 213
 214 void virtio_queue_set_notification(VirtQueue *vq, int enable)
 215 {
 216     if (enable)
 217         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
 218     else
 219         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
 220 }
 221
 222 int virtio_queue_ready(VirtQueue *vq)
 223 {
 224     return vq->vring.avail != 0;
 225 }
 226
 227 int virtio_queue_empty(VirtQueue *vq)
 228 {
 229     return vring_avail_idx(vq) == vq->last_avail_idx;
 230 }
 231
 232 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
 233                     unsigned int len, unsigned int idx)
 234 {
 235     unsigned int offset;
 236     int i;
 237
 238     offset = 0;
 239     for (i = 0; i < elem->in_num; i++) {
 240         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
 241
 242         cpu_physical_memory_unmap(elem->in_sg[i].iov_base,
 243                                   elem->in_sg[i].iov_len,
 244                                   1, size);
 245
 246         offset += elem->in_sg[i].iov_len;
 247     }
 248
 249     for (i = 0; i < elem->out_num; i++)
 250         cpu_physical_memory_unmap(elem->out_sg[i].iov_base,
 251                                   elem->out_sg[i].iov_len,
 252                                   0, elem->out_sg[i].iov_len);
 253
 254     idx = (idx + vring_used_idx(vq)) % vq->vring.num;
 255
 256     /* Get a pointer to the next entry in the used ring. */
 257     vring_used_ring_id(vq, idx, elem->index);
 258     vring_used_ring_len(vq, idx, len);
 259 }
 260
 261 void virtqueue_flush(VirtQueue *vq, unsigned int count)
 262 {
 263     /* Make sure buffer is written before we update index. */
 264     wmb();
 265     vring_used_idx_increment(vq, count);
 266     vq->inuse -= count;
 267 }
 268
 269 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
 270                     unsigned int len)
 271 {
 272     virtqueue_fill(vq, elem, len, 0);
 273     virtqueue_flush(vq, 1);
 274 }
 275
 276 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
 277 {
 278     uint16_t num_heads = vring_avail_idx(vq) - idx;
 279
 280     /* Check it isn't doing very strange things with descriptor numbers. */
 281     if (num_heads > vq->vring.num) {
 282         fprintf(stderr, "Guest moved used index from %u to %u",
 283                 idx, vring_avail_idx(vq));
 284         exit(1);
 285     }
 286
 287     return num_heads;
 288 }
 289
 290 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
 291 {
 292     unsigned int head;
 293
 294     /* Grab the next descriptor number they're advertising, and increment
 295      * the index we've seen. */
 296     head = vring_avail_ring(vq, idx % vq->vring.num);
 297
 298     /* If their number is silly, that's a fatal mistake. */
 299     if (head >= vq->vring.num) {
 300         fprintf(stderr, "Guest says index %u is available", head);
 301         exit(1);
 302     }
 303
 304     return head;
 305 }
 306
 307 static unsigned virtqueue_next_desc(VirtQueue *vq, unsigned int i)
 308 {
 309     unsigned int next;
 310
 311     /* If this descriptor says it doesn't chain, we're done. */
 312     if (!(vring_desc_flags(vq, i) & VRING_DESC_F_NEXT))
 313         return vq->vring.num;
 314
 315     /* Check they're not leading us off end of descriptors. */
 316     next = vring_desc_next(vq, i);
 317     /* Make sure compiler knows to grab that: we don't want it changing! */
 318     wmb();
 319
 320     if (next >= vq->vring.num) {
 321         fprintf(stderr, "Desc next is %u", next);
 322         exit(1);
 323     }
 324
 325     return next;
 326 }
 327
 328 int virtqueue_avail_bytes(VirtQueue *vq, int in_bytes, int out_bytes)
 329 {
 330     unsigned int idx;
 331     int num_bufs, in_total, out_total;
 332
 333     idx = vq->last_avail_idx;
 334
 335     num_bufs = in_total = out_total = 0;
 336     while (virtqueue_num_heads(vq, idx)) {
 337         int i;
 338
 339         i = virtqueue_get_head(vq, idx++);
 340         do {
 341             /* If we've got too many, that implies a descriptor loop. */
 342             if (++num_bufs > vq->vring.num) {
 343                 fprintf(stderr, "Looped descriptor");
 344                 exit(1);
 345             }
 346
 347             if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
 348                 if (in_bytes > 0 &&
 349                     (in_total += vring_desc_len(vq, i)) >= in_bytes)
 350                     return 1;
 351             } else {
 352                 if (out_bytes > 0 &&
 353                     (out_total += vring_desc_len(vq, i)) >= out_bytes)
 354                     return 1;
 355             }
 356         } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
 357     }
 358
 359     return 0;
 360 }
 361
 362 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
 363 {
 364     unsigned int i, head;
 365     target_phys_addr_t len;
 366
 367     if (!virtqueue_num_heads(vq, vq->last_avail_idx))
 368         return 0;
 369
 370     /* When we start there are none of either input nor output. */
 371     elem->out_num = elem->in_num = 0;
 372
 373     i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
 374     do {
 375         struct iovec *sg;
 376         int is_write = 0;
 377
 378         if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
 379             elem->in_addr[elem->in_num] = vring_desc_addr(vq, i);
 380             sg = &elem->in_sg[elem->in_num++];
 381             is_write = 1;
 382         } else
 383             sg = &elem->out_sg[elem->out_num++];
 384
 385         /* Grab the first descriptor, and check it's OK. */
 386         sg->iov_len = vring_desc_len(vq, i);
 387         len = sg->iov_len;
 388
 389         sg->iov_base = cpu_physical_memory_map(vring_desc_addr(vq, i), &len, is_write);
 390
 391         if (sg->iov_base == NULL || len != sg->iov_len) {
 392             fprintf(stderr, "virtio: trying to map MMIO memory\n");
 393             exit(1);
 394         }
 395
 396         /* If we've got too many, that implies a descriptor loop. */
 397         if ((elem->in_num + elem->out_num) > vq->vring.num) {
 398             fprintf(stderr, "Looped descriptor");
 399             exit(1);
 400         }
 401     } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
 402
 403     elem->index = head;
 404
 405     vq->inuse++;
 406
 407     return elem->in_num + elem->out_num;
 408 }
 409
 410 /* virtio device */
 411
 412 static VirtIODevice *to_virtio_device(PCIDevice *pci_dev)
 413 {
 414     return (VirtIODevice *)pci_dev;
 415 }
 416
 417 static void virtio_update_irq(VirtIODevice *vdev)
 418 {
 419     qemu_set_irq(vdev->pci_dev.irq[0], vdev->isr & 1);
 420 }
 421
 422 static void virtio_reset(void *opaque)
 423 {
 424     VirtIODevice *vdev = opaque;
 425     int i;
 426
 427     if (vdev->reset)
 428         vdev->reset(vdev);
 429
 430     vdev->features = 0;
 431     vdev->queue_sel = 0;
 432     vdev->status = 0;
 433     vdev->isr = 0;
 434     virtio_update_irq(vdev);
 435
 436     for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 437         vdev->vq[i].vring.desc = 0;
 438         vdev->vq[i].vring.avail = 0;
 439         vdev->vq[i].vring.used = 0;
 440         vdev->vq[i].last_avail_idx = 0;
 441         vdev->vq[i].pfn = 0;
 442     }
 443 }
 444
 445 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 446 {
 447     VirtIODevice *vdev = to_virtio_device(opaque);
 448     ram_addr_t pa;
 449
 450     addr -= vdev->addr;
 451
 452     switch (addr) {
 453     case VIRTIO_PCI_GUEST_FEATURES:
 454         /* Guest does not negotiate properly?  We have to assume nothing. */
 455         if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 456             if (vdev->bad_features)
 457                 val = vdev->bad_features(vdev);
 458             else
 459                 val = 0;
 460         }
 461         if (vdev->set_features)
 462             vdev->set_features(vdev, val);
 463         vdev->features = val;
 464         break;
 465     case VIRTIO_PCI_QUEUE_PFN:
 466         pa = (ram_addr_t)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 467         vdev->vq[vdev->queue_sel].pfn = val;
 468         if (pa == 0) {
 469             virtio_reset(vdev);
 470         } else {
 471             virtqueue_init(&vdev->vq[vdev->queue_sel], pa);
 472         }
 473         break;
 474     case VIRTIO_PCI_QUEUE_SEL:
 475         if (val < VIRTIO_PCI_QUEUE_MAX)
 476             vdev->queue_sel = val;
 477         break;
 478     case VIRTIO_PCI_QUEUE_NOTIFY:
 479         if (val < VIRTIO_PCI_QUEUE_MAX && vdev->vq[val].vring.desc)
 480             vdev->vq[val].handle_output(vdev, &vdev->vq[val]);
 481         break;
 482     case VIRTIO_PCI_STATUS:
 483         vdev->status = val & 0xFF;
 484         if (vdev->status == 0)
 485             virtio_reset(vdev);
 486         break;
 487     }
 488 }
 489
 490 static uint32_t virtio_ioport_read(void *opaque, uint32_t addr)
 491 {
 492     VirtIODevice *vdev = to_virtio_device(opaque);
 493     uint32_t ret = 0xFFFFFFFF;
 494
 495     addr -= vdev->addr;
 496
 497     switch (addr) {
 498     case VIRTIO_PCI_HOST_FEATURES:
 499         ret = vdev->get_features(vdev);
 500         ret |= (1 << VIRTIO_F_NOTIFY_ON_EMPTY) | (1 << VIRTIO_F_BAD_FEATURE);
 501         break;
 502     case VIRTIO_PCI_GUEST_FEATURES:
 503         ret = vdev->features;
 504         break;
 505     case VIRTIO_PCI_QUEUE_PFN:
 506         ret = vdev->vq[vdev->queue_sel].pfn;
 507         break;
 508     case VIRTIO_PCI_QUEUE_NUM:
 509         ret = vdev->vq[vdev->queue_sel].vring.num;
 510         break;
 511     case VIRTIO_PCI_QUEUE_SEL:
 512         ret = vdev->queue_sel;
 513         break;
 514     case VIRTIO_PCI_STATUS:
 515         ret = vdev->status;
 516         break;
 517     case VIRTIO_PCI_ISR:
 518         /* reading from the ISR also clears it. */
 519         ret = vdev->isr;
 520         vdev->isr = 0;
 521         virtio_update_irq(vdev);
 522         break;
 523     default:
 524         break;
 525     }
 526
 527     return ret;
 528 }
 529
 530 static uint32_t virtio_config_readb(void *opaque, uint32_t addr)
 531 {
 532     VirtIODevice *vdev = opaque;
 533     uint8_t val;
 534
 535     vdev->get_config(vdev, vdev->config);
 536
 537     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 538     if (addr > (vdev->config_len - sizeof(val)))
 539         return (uint32_t)-1;
 540
 541     memcpy(&val, vdev->config + addr, sizeof(val));
 542     return val;
 543 }
 544
 545 static uint32_t virtio_config_readw(void *opaque, uint32_t addr)
 546 {
 547     VirtIODevice *vdev = opaque;
 548     uint16_t val;
 549
 550     vdev->get_config(vdev, vdev->config);
 551
 552     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 553     if (addr > (vdev->config_len - sizeof(val)))
 554         return (uint32_t)-1;
 555
 556     memcpy(&val, vdev->config + addr, sizeof(val));
 557     return val;
 558 }
 559
 560 static uint32_t virtio_config_readl(void *opaque, uint32_t addr)
 561 {
 562     VirtIODevice *vdev = opaque;
 563     uint32_t val;
 564
 565     vdev->get_config(vdev, vdev->config);
 566
 567     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 568     if (addr > (vdev->config_len - sizeof(val)))
 569         return (uint32_t)-1;
 570
 571     memcpy(&val, vdev->config + addr, sizeof(val));
 572     return val;
 573 }
 574
 575 static void virtio_config_writeb(void *opaque, uint32_t addr, uint32_t data)
 576 {
 577     VirtIODevice *vdev = opaque;
 578     uint8_t val = data;
 579
 580     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 581     if (addr > (vdev->config_len - sizeof(val)))
 582         return;
 583
 584     memcpy(vdev->config + addr, &val, sizeof(val));
 585
 586     if (vdev->set_config)
 587         vdev->set_config(vdev, vdev->config);
 588 }
 589
 590 static void virtio_config_writew(void *opaque, uint32_t addr, uint32_t data)
 591 {
 592     VirtIODevice *vdev = opaque;
 593     uint16_t val = data;
 594
 595     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 596     if (addr > (vdev->config_len - sizeof(val)))
 597         return;
 598
 599     memcpy(vdev->config + addr, &val, sizeof(val));
 600
 601     if (vdev->set_config)
 602         vdev->set_config(vdev, vdev->config);
 603 }
 604
 605 static void virtio_config_writel(void *opaque, uint32_t addr, uint32_t data)
 606 {
 607     VirtIODevice *vdev = opaque;
 608     uint32_t val = data;
 609
 610     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 611     if (addr > (vdev->config_len - sizeof(val)))
 612         return;
 613
 614     memcpy(vdev->config + addr, &val, sizeof(val));
 615
 616     if (vdev->set_config)
 617         vdev->set_config(vdev, vdev->config);
 618 }
 619
 620 static void virtio_map(PCIDevice *pci_dev, int region_num,
 621                        uint32_t addr, uint32_t size, int type)
 622 {
 623     VirtIODevice *vdev = to_virtio_device(pci_dev);
 624     int i;
 625
 626     vdev->addr = addr;
 627     for (i = 0; i < 3; i++) {
 628         register_ioport_write(addr, 20, 1 << i, virtio_ioport_write, vdev);
 629         register_ioport_read(addr, 20, 1 << i, virtio_ioport_read, vdev);
 630     }
 631
 632     if (vdev->config_len) {
 633         register_ioport_write(addr + 20, vdev->config_len, 1,
 634                               virtio_config_writeb, vdev);
 635         register_ioport_write(addr + 20, vdev->config_len, 2,
 636                               virtio_config_writew, vdev);
 637         register_ioport_write(addr + 20, vdev->config_len, 4,
 638                               virtio_config_writel, vdev);
 639         register_ioport_read(addr + 20, vdev->config_len, 1,
 640                              virtio_config_readb, vdev);
 641         register_ioport_read(addr + 20, vdev->config_len, 2,
 642                              virtio_config_readw, vdev);
 643         register_ioport_read(addr + 20, vdev->config_len, 4,
 644                              virtio_config_readl, vdev);
 645
 646         vdev->get_config(vdev, vdev->config);
 647     }
 648 }
 649
 650 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
 651                             void (*handle_output)(VirtIODevice *, VirtQueue *))
 652 {
 653     int i;
 654
 655     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 656         if (vdev->vq[i].vring.num == 0)
 657             break;
 658     }
 659
 660     if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
 661         abort();
 662
 663     vdev->vq[i].vring.num = queue_size;
 664     vdev->vq[i].handle_output = handle_output;
 665
 666     return &vdev->vq[i];
 667 }
 668
 669 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
 670 {
 671     /* Always notify when queue is empty (when feature acknowledge) */
 672     if ((vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT) &&
 673         (!(vdev->features & (1 << VIRTIO_F_NOTIFY_ON_EMPTY)) ||
 674          (vq->inuse || vring_avail_idx(vq) != vq->last_avail_idx)))
 675         return;
 676
 677     vdev->isr |= 0x01;
 678     virtio_update_irq(vdev);
 679 }
 680
 681 void virtio_notify_config(VirtIODevice *vdev)
 682 {
 683     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
 684         return;
 685
 686     vdev->isr |= 0x03;
 687     virtio_update_irq(vdev);
 688 }
 689
 690 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
 691 {
 692     int i;
 693
 694     pci_device_save(&vdev->pci_dev, f);
 695
 696     qemu_put_be32s(f, &vdev->addr);
 697     qemu_put_8s(f, &vdev->status);
 698     qemu_put_8s(f, &vdev->isr);
 699     qemu_put_be16s(f, &vdev->queue_sel);
 700     qemu_put_be32s(f, &vdev->features);
 701     qemu_put_be32(f, vdev->config_len);
 702     qemu_put_buffer(f, vdev->config, vdev->config_len);
 703
 704     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 705         if (vdev->vq[i].vring.num == 0)
 706             break;
 707     }
 708
 709     qemu_put_be32(f, i);
 710
 711     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 712         if (vdev->vq[i].vring.num == 0)
 713             break;
 714
 715         qemu_put_be32(f, vdev->vq[i].vring.num);
 716         qemu_put_be32s(f, &vdev->vq[i].pfn);
 717         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
 718     }
 719 }
 720
 721 void virtio_load(VirtIODevice *vdev, QEMUFile *f)
 722 {
 723     int num, i;
 724
 725     pci_device_load(&vdev->pci_dev, f);
 726
 727     qemu_get_be32s(f, &vdev->addr);
 728     qemu_get_8s(f, &vdev->status);
 729     qemu_get_8s(f, &vdev->isr);
 730     qemu_get_be16s(f, &vdev->queue_sel);
 731     qemu_get_be32s(f, &vdev->features);
 732     vdev->config_len = qemu_get_be32(f);
 733     qemu_get_buffer(f, vdev->config, vdev->config_len);
 734
 735     num = qemu_get_be32(f);
 736
 737     for (i = 0; i < num; i++) {
 738         vdev->vq[i].vring.num = qemu_get_be32(f);
 739         qemu_get_be32s(f, &vdev->vq[i].pfn);
 740         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
 741
 742         if (vdev->vq[i].pfn) {
 743             target_phys_addr_t pa;
 744
 745             pa = (ram_addr_t)vdev->vq[i].pfn << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 746             virtqueue_init(&vdev->vq[i], pa);
 747         }
 748     }
 749
 750     virtio_update_irq(vdev);
 751 }
 752
 753 void virtio_cleanup(VirtIODevice *vdev)
 754 {
 755     if (vdev->config)
 756         qemu_free(vdev->config);
 757     qemu_free(vdev->vq);
 758 }
 759
 760 VirtIODevice *virtio_init_pci(PCIBus *bus, const char *name,
 761                               uint16_t vendor, uint16_t device,
 762                               uint16_t subvendor, uint16_t subdevice,
 763                               uint16_t class_code, uint8_t pif,
 764                               size_t config_size, size_t struct_size)
 765 {
 766     VirtIODevice *vdev;
 767     PCIDevice *pci_dev;
 768     uint8_t *config;
 769     uint32_t size;
 770
 771     pci_dev = pci_register_device(bus, name, struct_size,
 772                                   -1, NULL, NULL);
 773     if (!pci_dev)
 774         return NULL;
 775
 776     vdev = to_virtio_device(pci_dev);
 777
 778     vdev->status = 0;
 779     vdev->isr = 0;
 780     vdev->queue_sel = 0;
 781     vdev->vq = qemu_mallocz(sizeof(VirtQueue) * VIRTIO_PCI_QUEUE_MAX);
 782
 783     config = pci_dev->config;
 784     pci_config_set_vendor_id(config, vendor);
 785     pci_config_set_device_id(config, device);
 786
 787     config[0x08] = VIRTIO_PCI_ABI_VERSION;
 788
 789     config[0x09] = pif;
 790     pci_config_set_class(config, class_code);
 791     config[PCI_HEADER_TYPE] = PCI_HEADER_TYPE_NORMAL;
 792
 793     config[0x2c] = subvendor & 0xFF;
 794     config[0x2d] = (subvendor >> 8) & 0xFF;
 795     config[0x2e] = subdevice & 0xFF;
 796     config[0x2f] = (subdevice >> 8) & 0xFF;
 797
 798     config[0x3d] = 1;
 799
 800     vdev->name = name;
 801     vdev->config_len = config_size;
 802     if (vdev->config_len)
 803         vdev->config = qemu_mallocz(config_size);
 804     else
 805         vdev->config = NULL;
 806
 807     size = 20 + config_size;
 808     if (size & (size-1))
 809         size = 1 << qemu_fls(size);
 810
 811     pci_register_io_region(pci_dev, 0, size, PCI_ADDRESS_SPACE_IO,
 812                            virtio_map);
 813     qemu_register_reset(virtio_reset, vdev);
 814
 815     return vdev;
 816 }