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 //#define VIRTIO_ZERO_COPY
  20
  21 /* from Linux's linux/virtio_pci.h */
  22
  23 /* A 32-bit r/o bitmask of the features supported by the host */
  24 #define VIRTIO_PCI_HOST_FEATURES        0
  25
  26 /* A 32-bit r/w bitmask of features activated by the guest */
  27 #define VIRTIO_PCI_GUEST_FEATURES       4
  28
  29 /* A 32-bit r/w PFN for the currently selected queue */
  30 #define VIRTIO_PCI_QUEUE_PFN            8
  31
  32 /* A 16-bit r/o queue size for the currently selected queue */
  33 #define VIRTIO_PCI_QUEUE_NUM            12
  34
  35 /* A 16-bit r/w queue selector */
  36 #define VIRTIO_PCI_QUEUE_SEL            14
  37
  38 /* A 16-bit r/w queue notifier */
  39 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  40
  41 /* An 8-bit device status register.  */
  42 #define VIRTIO_PCI_STATUS               18
  43
  44 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  45  * current contents of the ISR and will also clear it.  This is effectively
  46  * a read-and-acknowledge. */
  47 #define VIRTIO_PCI_ISR                  19
  48
  49 #define VIRTIO_PCI_CONFIG               20
  50
  51 /* Virtio ABI version, if we increment this, we break the guest driver. */
  52 #define VIRTIO_PCI_ABI_VERSION          0
  53
  54 /* How many bits to shift physical queue address written to QUEUE_PFN.
  55  * 12 is historical, and due to x86 page size. */
  56 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT    12
  57
  58 /* The alignment to use between consumer and producer parts of vring.
  59  * x86 pagesize again. */
  60 #define VIRTIO_PCI_VRING_ALIGN         4096
  61
  62 /* QEMU doesn't strictly need write barriers since everything runs in
  63  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  64  * KVM or if kqemu gets SMP support.
  65  */
  66 #define wmb() do { } while (0)
  67
  68 typedef struct VRingDesc
  69 {
  70     uint64_t addr;
  71     uint32_t len;
  72     uint16_t flags;
  73     uint16_t next;
  74 } VRingDesc;
  75
  76 typedef struct VRingAvail
  77 {
  78     uint16_t flags;
  79     uint16_t idx;
  80     uint16_t ring[0];
  81 } VRingAvail;
  82
  83 typedef struct VRingUsedElem
  84 {
  85     uint32_t id;
  86     uint32_t len;
  87 } VRingUsedElem;
  88
  89 typedef struct VRingUsed
  90 {
  91     uint16_t flags;
  92     uint16_t idx;
  93     VRingUsedElem ring[0];
  94 } VRingUsed;
  95
  96 typedef struct VRing
  97 {
  98     unsigned int num;
  99     target_phys_addr_t desc;
 100     target_phys_addr_t avail;
 101     target_phys_addr_t used;
 102 } VRing;
 103
 104 struct VirtQueue
 105 {
 106     VRing vring;
 107     uint32_t pfn;
 108     uint16_t last_avail_idx;
 109     int inuse;
 110     void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
 111 };
 112
 113 #define VIRTIO_PCI_QUEUE_MAX        16
 114
 115 /* virt queue functions */
 116 #ifdef VIRTIO_ZERO_COPY
 117 static void *virtio_map_gpa(target_phys_addr_t addr, size_t size)
 118 {
 119     ram_addr_t off;
 120     target_phys_addr_t addr1;
 121
 122     off = cpu_get_physical_page_desc(addr);
 123     if ((off & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
 124         fprintf(stderr, "virtio DMA to IO ram\n");
 125         exit(1);
 126     }
 127
 128     off = (off & TARGET_PAGE_MASK) | (addr & ~TARGET_PAGE_MASK);
 129
 130     for (addr1 = addr + TARGET_PAGE_SIZE;
 131          addr1 < TARGET_PAGE_ALIGN(addr + size);
 132          addr1 += TARGET_PAGE_SIZE) {
 133         ram_addr_t off1;
 134
 135         off1 = cpu_get_physical_page_desc(addr1);
 136         if ((off1 & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
 137             fprintf(stderr, "virtio DMA to IO ram\n");
 138             exit(1);
 139         }
 140
 141         off1 = (off1 & TARGET_PAGE_MASK) | (addr1 & ~TARGET_PAGE_MASK);
 142
 143         if (off1 != (off + (addr1 - addr))) {
 144             fprintf(stderr, "discontigous virtio memory\n");
 145             exit(1);
 146         }
 147     }
 148
 149     return phys_ram_base + off;
 150 }
 151 #endif
 152
 153 static void virtqueue_init(VirtQueue *vq, target_phys_addr_t pa)
 154 {
 155     vq->vring.desc = pa;
 156     vq->vring.avail = pa + vq->vring.num * sizeof(VRingDesc);
 157     vq->vring.used = vring_align(vq->vring.avail +
 158                                  offsetof(VRingAvail, ring[vq->vring.num]),
 159                                  VIRTIO_PCI_VRING_ALIGN);
 160 }
 161
 162 static inline uint64_t vring_desc_addr(VirtQueue *vq, int i)
 163 {
 164     target_phys_addr_t pa;
 165     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
 166     return ldq_phys(pa);
 167 }
 168
 169 static inline uint32_t vring_desc_len(VirtQueue *vq, int i)
 170 {
 171     target_phys_addr_t pa;
 172     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
 173     return ldl_phys(pa);
 174 }
 175
 176 static inline uint16_t vring_desc_flags(VirtQueue *vq, int i)
 177 {
 178     target_phys_addr_t pa;
 179     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
 180     return lduw_phys(pa);
 181 }
 182
 183 static inline uint16_t vring_desc_next(VirtQueue *vq, int i)
 184 {
 185     target_phys_addr_t pa;
 186     pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
 187     return lduw_phys(pa);
 188 }
 189
 190 static inline uint16_t vring_avail_flags(VirtQueue *vq)
 191 {
 192     target_phys_addr_t pa;
 193     pa = vq->vring.avail + offsetof(VRingAvail, flags);
 194     return lduw_phys(pa);
 195 }
 196
 197 static inline uint16_t vring_avail_idx(VirtQueue *vq)
 198 {
 199     target_phys_addr_t pa;
 200     pa = vq->vring.avail + offsetof(VRingAvail, idx);
 201     return lduw_phys(pa);
 202 }
 203
 204 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
 205 {
 206     target_phys_addr_t pa;
 207     pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
 208     return lduw_phys(pa);
 209 }
 210
 211 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
 212 {
 213     target_phys_addr_t pa;
 214     pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
 215     stl_phys(pa, val);
 216 }
 217
 218 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
 219 {
 220     target_phys_addr_t pa;
 221     pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
 222     stl_phys(pa, val);
 223 }
 224
 225 static uint16_t vring_used_idx(VirtQueue *vq)
 226 {
 227     target_phys_addr_t pa;
 228     pa = vq->vring.used + offsetof(VRingUsed, idx);
 229     return lduw_phys(pa);
 230 }
 231
 232 static inline void vring_used_idx_increment(VirtQueue *vq, uint16_t val)
 233 {
 234     target_phys_addr_t pa;
 235     pa = vq->vring.used + offsetof(VRingUsed, idx);
 236     stw_phys(pa, vring_used_idx(vq) + val);
 237 }
 238
 239 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
 240 {
 241     target_phys_addr_t pa;
 242     pa = vq->vring.used + offsetof(VRingUsed, flags);
 243     stw_phys(pa, lduw_phys(pa) | mask);
 244 }
 245
 246 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
 247 {
 248     target_phys_addr_t pa;
 249     pa = vq->vring.used + offsetof(VRingUsed, flags);
 250     stw_phys(pa, lduw_phys(pa) & ~mask);
 251 }
 252
 253 void virtio_queue_set_notification(VirtQueue *vq, int enable)
 254 {
 255     if (enable)
 256         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
 257     else
 258         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
 259 }
 260
 261 int virtio_queue_ready(VirtQueue *vq)
 262 {
 263     return vq->vring.avail != 0;
 264 }
 265
 266 int virtio_queue_empty(VirtQueue *vq)
 267 {
 268     return vring_avail_idx(vq) == vq->last_avail_idx;
 269 }
 270
 271 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
 272                     unsigned int len, unsigned int idx)
 273 {
 274     unsigned int offset;
 275     int i;
 276
 277 #ifndef VIRTIO_ZERO_COPY
 278     for (i = 0; i < elem->out_num; i++)
 279         qemu_free(elem->out_sg[i].iov_base);
 280 #endif
 281
 282     offset = 0;
 283     for (i = 0; i < elem->in_num; i++) {
 284         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
 285
 286 #ifdef VIRTIO_ZERO_COPY
 287         if (size) {
 288             ram_addr_t addr = (uint8_t *)elem->in_sg[i].iov_base - phys_ram_base;
 289             ram_addr_t off;
 290
 291             for (off = 0; off < size; off += TARGET_PAGE_SIZE)
 292                 cpu_physical_memory_set_dirty(addr + off);
 293         }
 294 #else
 295         if (size)
 296             cpu_physical_memory_write(elem->in_addr[i],
 297                                       elem->in_sg[i].iov_base,
 298                                       size);
 299
 300         qemu_free(elem->in_sg[i].iov_base);
 301 #endif
 302
 303         offset += size;
 304     }
 305
 306     idx = (idx + vring_used_idx(vq)) % vq->vring.num;
 307
 308     /* Get a pointer to the next entry in the used ring. */
 309     vring_used_ring_id(vq, idx, elem->index);
 310     vring_used_ring_len(vq, idx, len);
 311 }
 312
 313 void virtqueue_flush(VirtQueue *vq, unsigned int count)
 314 {
 315     /* Make sure buffer is written before we update index. */
 316     wmb();
 317     vring_used_idx_increment(vq, count);
 318     vq->inuse -= count;
 319 }
 320
 321 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
 322                     unsigned int len)
 323 {
 324     virtqueue_fill(vq, elem, len, 0);
 325     virtqueue_flush(vq, 1);
 326 }
 327
 328 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
 329 {
 330     uint16_t num_heads = vring_avail_idx(vq) - idx;
 331
 332     /* Check it isn't doing very strange things with descriptor numbers. */
 333     if (num_heads > vq->vring.num) {
 334         fprintf(stderr, "Guest moved used index from %u to %u",
 335                 idx, vring_avail_idx(vq));
 336         exit(1);
 337     }
 338
 339     return num_heads;
 340 }
 341
 342 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
 343 {
 344     unsigned int head;
 345
 346     /* Grab the next descriptor number they're advertising, and increment
 347      * the index we've seen. */
 348     head = vring_avail_ring(vq, idx % vq->vring.num);
 349
 350     /* If their number is silly, that's a fatal mistake. */
 351     if (head >= vq->vring.num) {
 352         fprintf(stderr, "Guest says index %u is available", head);
 353         exit(1);
 354     }
 355
 356     return head;
 357 }
 358
 359 static unsigned virtqueue_next_desc(VirtQueue *vq, unsigned int i)
 360 {
 361     unsigned int next;
 362
 363     /* If this descriptor says it doesn't chain, we're done. */
 364     if (!(vring_desc_flags(vq, i) & VRING_DESC_F_NEXT))
 365         return vq->vring.num;
 366
 367     /* Check they're not leading us off end of descriptors. */
 368     next = vring_desc_next(vq, i);
 369     /* Make sure compiler knows to grab that: we don't want it changing! */
 370     wmb();
 371
 372     if (next >= vq->vring.num) {
 373         fprintf(stderr, "Desc next is %u", next);
 374         exit(1);
 375     }
 376
 377     return next;
 378 }
 379
 380 int virtqueue_avail_bytes(VirtQueue *vq, int in_bytes, int out_bytes)
 381 {
 382     unsigned int idx;
 383     int num_bufs, in_total, out_total;
 384
 385     idx = vq->last_avail_idx;
 386
 387     num_bufs = in_total = out_total = 0;
 388     while (virtqueue_num_heads(vq, idx)) {
 389         int i;
 390
 391         i = virtqueue_get_head(vq, idx++);
 392         do {
 393             /* If we've got too many, that implies a descriptor loop. */
 394             if (++num_bufs > vq->vring.num) {
 395                 fprintf(stderr, "Looped descriptor");
 396                 exit(1);
 397             }
 398
 399             if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
 400                 if (in_bytes > 0 &&
 401                     (in_total += vring_desc_len(vq, i)) >= in_bytes)
 402                     return 1;
 403             } else {
 404                 if (out_bytes > 0 &&
 405                     (out_total += vring_desc_len(vq, i)) >= out_bytes)
 406                     return 1;
 407             }
 408         } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
 409     }
 410
 411     return 0;
 412 }
 413
 414 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
 415 {
 416     unsigned int i, head;
 417
 418     if (!virtqueue_num_heads(vq, vq->last_avail_idx))
 419         return 0;
 420
 421     /* When we start there are none of either input nor output. */
 422     elem->out_num = elem->in_num = 0;
 423
 424     i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
 425     do {
 426         struct iovec *sg;
 427
 428         if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
 429             elem->in_addr[elem->in_num] = vring_desc_addr(vq, i);
 430             sg = &elem->in_sg[elem->in_num++];
 431         } else
 432             sg = &elem->out_sg[elem->out_num++];
 433
 434         /* Grab the first descriptor, and check it's OK. */
 435         sg->iov_len = vring_desc_len(vq, i);
 436
 437 #ifdef VIRTIO_ZERO_COPY
 438         sg->iov_base = virtio_map_gpa(vring_desc_addr(vq, i), sg->iov_len);
 439 #else
 440         /* cap individual scatter element size to prevent unbounded allocations
 441            of memory from the guest.  Practically speaking, no virtio driver
 442            will ever pass more than a page in each element.  We set the cap to
 443            be 2MB in case for some reason a large page makes it way into the
 444            sg list.  When we implement a zero copy API, this limitation will
 445            disappear */
 446         if (sg->iov_len > (2 << 20))
 447             sg->iov_len = 2 << 20;
 448
 449         sg->iov_base = qemu_malloc(sg->iov_len);
 450         if (sg->iov_base &&
 451             !(vring_desc_flags(vq, i) & VRING_DESC_F_WRITE)) {
 452             cpu_physical_memory_read(vring_desc_addr(vq, i),
 453                                      sg->iov_base,
 454                                      sg->iov_len);
 455         }
 456 #endif
 457         if (sg->iov_base == NULL) {
 458             fprintf(stderr, "Invalid mapping\n");
 459             exit(1);
 460         }
 461
 462         /* If we've got too many, that implies a descriptor loop. */
 463         if ((elem->in_num + elem->out_num) > vq->vring.num) {
 464             fprintf(stderr, "Looped descriptor");
 465             exit(1);
 466         }
 467     } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
 468
 469     elem->index = head;
 470
 471     vq->inuse++;
 472
 473     return elem->in_num + elem->out_num;
 474 }
 475
 476 /* virtio device */
 477
 478 static VirtIODevice *to_virtio_device(PCIDevice *pci_dev)
 479 {
 480     return (VirtIODevice *)pci_dev;
 481 }
 482
 483 static void virtio_update_irq(VirtIODevice *vdev)
 484 {
 485     qemu_set_irq(vdev->pci_dev.irq[0], vdev->isr & 1);
 486 }
 487
 488 static void virtio_reset(void *opaque)
 489 {
 490     VirtIODevice *vdev = opaque;
 491     int i;
 492
 493     if (vdev->reset)
 494         vdev->reset(vdev);
 495
 496     vdev->features = 0;
 497     vdev->queue_sel = 0;
 498     vdev->status = 0;
 499     vdev->isr = 0;
 500     virtio_update_irq(vdev);
 501
 502     for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 503         vdev->vq[i].vring.desc = 0;
 504         vdev->vq[i].vring.avail = 0;
 505         vdev->vq[i].vring.used = 0;
 506         vdev->vq[i].last_avail_idx = 0;
 507         vdev->vq[i].pfn = 0;
 508     }
 509 }
 510
 511 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 512 {
 513     VirtIODevice *vdev = to_virtio_device(opaque);
 514     ram_addr_t pa;
 515
 516     addr -= vdev->addr;
 517
 518     switch (addr) {
 519     case VIRTIO_PCI_GUEST_FEATURES:
 520         if (vdev->set_features)
 521             vdev->set_features(vdev, val);
 522         vdev->features = val;
 523         break;
 524     case VIRTIO_PCI_QUEUE_PFN:
 525         pa = (ram_addr_t)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 526         vdev->vq[vdev->queue_sel].pfn = val;
 527         if (pa == 0) {
 528             virtio_reset(vdev);
 529         } else {
 530             virtqueue_init(&vdev->vq[vdev->queue_sel], pa);
 531         }
 532         break;
 533     case VIRTIO_PCI_QUEUE_SEL:
 534         if (val < VIRTIO_PCI_QUEUE_MAX)
 535             vdev->queue_sel = val;
 536         break;
 537     case VIRTIO_PCI_QUEUE_NOTIFY:
 538         if (val < VIRTIO_PCI_QUEUE_MAX && vdev->vq[val].vring.desc)
 539             vdev->vq[val].handle_output(vdev, &vdev->vq[val]);
 540         break;
 541     case VIRTIO_PCI_STATUS:
 542         vdev->status = val & 0xFF;
 543         if (vdev->status == 0)
 544             virtio_reset(vdev);
 545         break;
 546     }
 547 }
 548
 549 static uint32_t virtio_ioport_read(void *opaque, uint32_t addr)
 550 {
 551     VirtIODevice *vdev = to_virtio_device(opaque);
 552     uint32_t ret = 0xFFFFFFFF;
 553
 554     addr -= vdev->addr;
 555
 556     switch (addr) {
 557     case VIRTIO_PCI_HOST_FEATURES:
 558         ret = vdev->get_features(vdev);
 559         ret |= (1 << VIRTIO_F_NOTIFY_ON_EMPTY);
 560         break;
 561     case VIRTIO_PCI_GUEST_FEATURES:
 562         ret = vdev->features;
 563         break;
 564     case VIRTIO_PCI_QUEUE_PFN:
 565         ret = vdev->vq[vdev->queue_sel].pfn;
 566         break;
 567     case VIRTIO_PCI_QUEUE_NUM:
 568         ret = vdev->vq[vdev->queue_sel].vring.num;
 569         break;
 570     case VIRTIO_PCI_QUEUE_SEL:
 571         ret = vdev->queue_sel;
 572         break;
 573     case VIRTIO_PCI_STATUS:
 574         ret = vdev->status;
 575         break;
 576     case VIRTIO_PCI_ISR:
 577         /* reading from the ISR also clears it. */
 578         ret = vdev->isr;
 579         vdev->isr = 0;
 580         virtio_update_irq(vdev);
 581         break;
 582     default:
 583         break;
 584     }
 585
 586     return ret;
 587 }
 588
 589 static uint32_t virtio_config_readb(void *opaque, uint32_t addr)
 590 {
 591     VirtIODevice *vdev = opaque;
 592     uint8_t val;
 593
 594     vdev->get_config(vdev, vdev->config);
 595
 596     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 597     if (addr > (vdev->config_len - sizeof(val)))
 598         return (uint32_t)-1;
 599
 600     memcpy(&val, vdev->config + addr, sizeof(val));
 601     return val;
 602 }
 603
 604 static uint32_t virtio_config_readw(void *opaque, uint32_t addr)
 605 {
 606     VirtIODevice *vdev = opaque;
 607     uint16_t val;
 608
 609     vdev->get_config(vdev, vdev->config);
 610
 611     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 612     if (addr > (vdev->config_len - sizeof(val)))
 613         return (uint32_t)-1;
 614
 615     memcpy(&val, vdev->config + addr, sizeof(val));
 616     return val;
 617 }
 618
 619 static uint32_t virtio_config_readl(void *opaque, uint32_t addr)
 620 {
 621     VirtIODevice *vdev = opaque;
 622     uint32_t val;
 623
 624     vdev->get_config(vdev, vdev->config);
 625
 626     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 627     if (addr > (vdev->config_len - sizeof(val)))
 628         return (uint32_t)-1;
 629
 630     memcpy(&val, vdev->config + addr, sizeof(val));
 631     return val;
 632 }
 633
 634 static void virtio_config_writeb(void *opaque, uint32_t addr, uint32_t data)
 635 {
 636     VirtIODevice *vdev = opaque;
 637     uint8_t val = data;
 638
 639     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 640     if (addr > (vdev->config_len - sizeof(val)))
 641         return;
 642
 643     memcpy(vdev->config + addr, &val, sizeof(val));
 644
 645     if (vdev->set_config)
 646         vdev->set_config(vdev, vdev->config);
 647 }
 648
 649 static void virtio_config_writew(void *opaque, uint32_t addr, uint32_t data)
 650 {
 651     VirtIODevice *vdev = opaque;
 652     uint16_t val = data;
 653
 654     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 655     if (addr > (vdev->config_len - sizeof(val)))
 656         return;
 657
 658     memcpy(vdev->config + addr, &val, sizeof(val));
 659
 660     if (vdev->set_config)
 661         vdev->set_config(vdev, vdev->config);
 662 }
 663
 664 static void virtio_config_writel(void *opaque, uint32_t addr, uint32_t data)
 665 {
 666     VirtIODevice *vdev = opaque;
 667     uint32_t val = data;
 668
 669     addr -= vdev->addr + VIRTIO_PCI_CONFIG;
 670     if (addr > (vdev->config_len - sizeof(val)))
 671         return;
 672
 673     memcpy(vdev->config + addr, &val, sizeof(val));
 674
 675     if (vdev->set_config)
 676         vdev->set_config(vdev, vdev->config);
 677 }
 678
 679 static void virtio_map(PCIDevice *pci_dev, int region_num,
 680                        uint32_t addr, uint32_t size, int type)
 681 {
 682     VirtIODevice *vdev = to_virtio_device(pci_dev);
 683     int i;
 684
 685     vdev->addr = addr;
 686     for (i = 0; i < 3; i++) {
 687         register_ioport_write(addr, 20, 1 << i, virtio_ioport_write, vdev);
 688         register_ioport_read(addr, 20, 1 << i, virtio_ioport_read, vdev);
 689     }
 690
 691     if (vdev->config_len) {
 692         register_ioport_write(addr + 20, vdev->config_len, 1,
 693                               virtio_config_writeb, vdev);
 694         register_ioport_write(addr + 20, vdev->config_len, 2,
 695                               virtio_config_writew, vdev);
 696         register_ioport_write(addr + 20, vdev->config_len, 4,
 697                               virtio_config_writel, vdev);
 698         register_ioport_read(addr + 20, vdev->config_len, 1,
 699                              virtio_config_readb, vdev);
 700         register_ioport_read(addr + 20, vdev->config_len, 2,
 701                              virtio_config_readw, vdev);
 702         register_ioport_read(addr + 20, vdev->config_len, 4,
 703                              virtio_config_readl, vdev);
 704
 705         vdev->get_config(vdev, vdev->config);
 706     }
 707 }
 708
 709 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
 710                             void (*handle_output)(VirtIODevice *, VirtQueue *))
 711 {
 712     int i;
 713
 714     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 715         if (vdev->vq[i].vring.num == 0)
 716             break;
 717     }
 718
 719     if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
 720         abort();
 721
 722     vdev->vq[i].vring.num = queue_size;
 723     vdev->vq[i].handle_output = handle_output;
 724
 725     return &vdev->vq[i];
 726 }
 727
 728 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
 729 {
 730     /* Always notify when queue is empty */
 731     if ((vq->inuse || vring_avail_idx(vq) != vq->last_avail_idx) &&
 732         (vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT))
 733         return;
 734
 735     vdev->isr |= 0x01;
 736     virtio_update_irq(vdev);
 737 }
 738
 739 void virtio_notify_config(VirtIODevice *vdev)
 740 {
 741     vdev->isr |= 0x03;
 742     virtio_update_irq(vdev);
 743 }
 744
 745 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
 746 {
 747     int i;
 748
 749     pci_device_save(&vdev->pci_dev, f);
 750
 751     qemu_put_be32s(f, &vdev->addr);
 752     qemu_put_8s(f, &vdev->status);
 753     qemu_put_8s(f, &vdev->isr);
 754     qemu_put_be16s(f, &vdev->queue_sel);
 755     qemu_put_be32s(f, &vdev->features);
 756     qemu_put_be32(f, vdev->config_len);
 757     qemu_put_buffer(f, vdev->config, vdev->config_len);
 758
 759     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 760         if (vdev->vq[i].vring.num == 0)
 761             break;
 762     }
 763
 764     qemu_put_be32(f, i);
 765
 766     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 767         if (vdev->vq[i].vring.num == 0)
 768             break;
 769
 770         qemu_put_be32(f, vdev->vq[i].vring.num);
 771         qemu_put_be32s(f, &vdev->vq[i].pfn);
 772         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
 773     }
 774 }
 775
 776 void virtio_load(VirtIODevice *vdev, QEMUFile *f)
 777 {
 778     int num, i;
 779
 780     pci_device_load(&vdev->pci_dev, f);
 781
 782     qemu_get_be32s(f, &vdev->addr);
 783     qemu_get_8s(f, &vdev->status);
 784     qemu_get_8s(f, &vdev->isr);
 785     qemu_get_be16s(f, &vdev->queue_sel);
 786     qemu_get_be32s(f, &vdev->features);
 787     vdev->config_len = qemu_get_be32(f);
 788     qemu_get_buffer(f, vdev->config, vdev->config_len);
 789
 790     num = qemu_get_be32(f);
 791
 792     for (i = 0; i < num; i++) {
 793         vdev->vq[i].vring.num = qemu_get_be32(f);
 794         qemu_get_be32s(f, &vdev->vq[i].pfn);
 795         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
 796
 797         if (vdev->vq[i].pfn) {
 798             target_phys_addr_t pa;
 799
 800             pa = (ram_addr_t)vdev->vq[i].pfn << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 801             virtqueue_init(&vdev->vq[i], pa);
 802         }
 803     }
 804
 805     virtio_update_irq(vdev);
 806 }
 807
 808 VirtIODevice *virtio_init_pci(PCIBus *bus, const char *name,
 809                               uint16_t vendor, uint16_t device,
 810                               uint16_t subvendor, uint16_t subdevice,
 811                               uint8_t class_code, uint8_t subclass_code,
 812                               uint8_t pif, size_t config_size,
 813                               size_t struct_size)
 814 {
 815     VirtIODevice *vdev;
 816     PCIDevice *pci_dev;
 817     uint8_t *config;
 818     uint32_t size;
 819
 820     pci_dev = pci_register_device(bus, name, struct_size,
 821                                   -1, NULL, NULL);
 822     if (!pci_dev)
 823         return NULL;
 824
 825     vdev = to_virtio_device(pci_dev);
 826
 827     vdev->status = 0;
 828     vdev->isr = 0;
 829     vdev->queue_sel = 0;
 830     vdev->vq = qemu_mallocz(sizeof(VirtQueue) * VIRTIO_PCI_QUEUE_MAX);
 831
 832     config = pci_dev->config;
 833     config[0x00] = vendor & 0xFF;
 834     config[0x01] = (vendor >> 8) & 0xFF;
 835     config[0x02] = device & 0xFF;
 836     config[0x03] = (device >> 8) & 0xFF;
 837
 838     config[0x08] = VIRTIO_PCI_ABI_VERSION;
 839
 840     config[0x09] = pif;
 841     config[0x0a] = subclass_code;
 842     config[0x0b] = class_code;
 843     config[0x0e] = 0x00;
 844
 845     config[0x2c] = subvendor & 0xFF;
 846     config[0x2d] = (subvendor >> 8) & 0xFF;
 847     config[0x2e] = subdevice & 0xFF;
 848     config[0x2f] = (subdevice >> 8) & 0xFF;
 849
 850     config[0x3d] = 1;
 851
 852     vdev->name = name;
 853     vdev->config_len = config_size;
 854     if (vdev->config_len)
 855         vdev->config = qemu_mallocz(config_size);
 856     else
 857         vdev->config = NULL;
 858
 859     size = 20 + config_size;
 860     if (size & (size-1))
 861         size = 1 << fls(size);
 862
 863     pci_register_io_region(pci_dev, 0, size, PCI_ADDRESS_SPACE_IO,
 864                            virtio_map);
 865     qemu_register_reset(virtio_reset, vdev);
 866
 867     return vdev;
 868 }