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 "trace.h"
  17 #include "qemu/error-report.h"
  18 #include "virtio.h"
  19 #include "qemu/atomic.h"
  20 #include "virtio-bus.h"
  21
  22 /* The alignment to use between consumer and producer parts of vring.
  23  * x86 pagesize again. */
  24 #define VIRTIO_PCI_VRING_ALIGN         4096
  25
  26 typedef struct VRingDesc
  27 {
  28     uint64_t addr;
  29     uint32_t len;
  30     uint16_t flags;
  31     uint16_t next;
  32 } VRingDesc;
  33
  34 typedef struct VRingAvail
  35 {
  36     uint16_t flags;
  37     uint16_t idx;
  38     uint16_t ring[0];
  39 } VRingAvail;
  40
  41 typedef struct VRingUsedElem
  42 {
  43     uint32_t id;
  44     uint32_t len;
  45 } VRingUsedElem;
  46
  47 typedef struct VRingUsed
  48 {
  49     uint16_t flags;
  50     uint16_t idx;
  51     VRingUsedElem ring[0];
  52 } VRingUsed;
  53
  54 typedef struct VRing
  55 {
  56     unsigned int num;
  57     hwaddr desc;
  58     hwaddr avail;
  59     hwaddr used;
  60 } VRing;
  61
  62 struct VirtQueue
  63 {
  64     VRing vring;
  65     hwaddr pa;
  66     uint16_t last_avail_idx;
  67     /* Last used index value we have signalled on */
  68     uint16_t signalled_used;
  69
  70     /* Last used index value we have signalled on */
  71     bool signalled_used_valid;
  72
  73     /* Notification enabled? */
  74     bool notification;
  75
  76     uint16_t queue_index;
  77
  78     int inuse;
  79
  80     uint16_t vector;
  81     void (*handle_output)(VirtIODevice *vdev, VirtQueue *vq);
  82     VirtIODevice *vdev;
  83     EventNotifier guest_notifier;
  84     EventNotifier host_notifier;
  85 };
  86
  87 /* virt queue functions */
  88 static void virtqueue_init(VirtQueue *vq)
  89 {
  90     hwaddr pa = vq->pa;
  91
  92     vq->vring.desc = pa;
  93     vq->vring.avail = pa + vq->vring.num * sizeof(VRingDesc);
  94     vq->vring.used = vring_align(vq->vring.avail +
  95                                  offsetof(VRingAvail, ring[vq->vring.num]),
  96                                  VIRTIO_PCI_VRING_ALIGN);
  97 }
  98
  99 static inline uint64_t vring_desc_addr(hwaddr desc_pa, int i)
 100 {
 101     hwaddr pa;
 102     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
 103     return ldq_phys(pa);
 104 }
 105
 106 static inline uint32_t vring_desc_len(hwaddr desc_pa, int i)
 107 {
 108     hwaddr pa;
 109     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
 110     return ldl_phys(pa);
 111 }
 112
 113 static inline uint16_t vring_desc_flags(hwaddr desc_pa, int i)
 114 {
 115     hwaddr pa;
 116     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
 117     return lduw_phys(pa);
 118 }
 119
 120 static inline uint16_t vring_desc_next(hwaddr desc_pa, int i)
 121 {
 122     hwaddr pa;
 123     pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
 124     return lduw_phys(pa);
 125 }
 126
 127 static inline uint16_t vring_avail_flags(VirtQueue *vq)
 128 {
 129     hwaddr pa;
 130     pa = vq->vring.avail + offsetof(VRingAvail, flags);
 131     return lduw_phys(pa);
 132 }
 133
 134 static inline uint16_t vring_avail_idx(VirtQueue *vq)
 135 {
 136     hwaddr pa;
 137     pa = vq->vring.avail + offsetof(VRingAvail, idx);
 138     return lduw_phys(pa);
 139 }
 140
 141 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
 142 {
 143     hwaddr pa;
 144     pa = vq->vring.avail + offsetof(VRingAvail, ring[i]);
 145     return lduw_phys(pa);
 146 }
 147
 148 static inline uint16_t vring_used_event(VirtQueue *vq)
 149 {
 150     return vring_avail_ring(vq, vq->vring.num);
 151 }
 152
 153 static inline void vring_used_ring_id(VirtQueue *vq, int i, uint32_t val)
 154 {
 155     hwaddr pa;
 156     pa = vq->vring.used + offsetof(VRingUsed, ring[i].id);
 157     stl_phys(pa, val);
 158 }
 159
 160 static inline void vring_used_ring_len(VirtQueue *vq, int i, uint32_t val)
 161 {
 162     hwaddr pa;
 163     pa = vq->vring.used + offsetof(VRingUsed, ring[i].len);
 164     stl_phys(pa, val);
 165 }
 166
 167 static uint16_t vring_used_idx(VirtQueue *vq)
 168 {
 169     hwaddr pa;
 170     pa = vq->vring.used + offsetof(VRingUsed, idx);
 171     return lduw_phys(pa);
 172 }
 173
 174 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
 175 {
 176     hwaddr pa;
 177     pa = vq->vring.used + offsetof(VRingUsed, idx);
 178     stw_phys(pa, val);
 179 }
 180
 181 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
 182 {
 183     hwaddr pa;
 184     pa = vq->vring.used + offsetof(VRingUsed, flags);
 185     stw_phys(pa, lduw_phys(pa) | mask);
 186 }
 187
 188 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
 189 {
 190     hwaddr pa;
 191     pa = vq->vring.used + offsetof(VRingUsed, flags);
 192     stw_phys(pa, lduw_phys(pa) & ~mask);
 193 }
 194
 195 static inline void vring_avail_event(VirtQueue *vq, uint16_t val)
 196 {
 197     hwaddr pa;
 198     if (!vq->notification) {
 199         return;
 200     }
 201     pa = vq->vring.used + offsetof(VRingUsed, ring[vq->vring.num]);
 202     stw_phys(pa, val);
 203 }
 204
 205 void virtio_queue_set_notification(VirtQueue *vq, int enable)
 206 {
 207     vq->notification = enable;
 208     if (vq->vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX)) {
 209         vring_avail_event(vq, vring_avail_idx(vq));
 210     } else if (enable) {
 211         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
 212     } else {
 213         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
 214     }
 215     if (enable) {
 216         /* Expose avail event/used flags before caller checks the avail idx. */
 217         smp_mb();
 218     }
 219 }
 220
 221 int virtio_queue_ready(VirtQueue *vq)
 222 {
 223     return vq->vring.avail != 0;
 224 }
 225
 226 int virtio_queue_empty(VirtQueue *vq)
 227 {
 228     return vring_avail_idx(vq) == vq->last_avail_idx;
 229 }
 230
 231 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
 232                     unsigned int len, unsigned int idx)
 233 {
 234     unsigned int offset;
 235     int i;
 236
 237     trace_virtqueue_fill(vq, elem, len, idx);
 238
 239     offset = 0;
 240     for (i = 0; i < elem->in_num; i++) {
 241         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
 242
 243         cpu_physical_memory_unmap(elem->in_sg[i].iov_base,
 244                                   elem->in_sg[i].iov_len,
 245                                   1, size);
 246
 247         offset += size;
 248     }
 249
 250     for (i = 0; i < elem->out_num; i++)
 251         cpu_physical_memory_unmap(elem->out_sg[i].iov_base,
 252                                   elem->out_sg[i].iov_len,
 253                                   0, elem->out_sg[i].iov_len);
 254
 255     idx = (idx + vring_used_idx(vq)) % vq->vring.num;
 256
 257     /* Get a pointer to the next entry in the used ring. */
 258     vring_used_ring_id(vq, idx, elem->index);
 259     vring_used_ring_len(vq, idx, len);
 260 }
 261
 262 void virtqueue_flush(VirtQueue *vq, unsigned int count)
 263 {
 264     uint16_t old, new;
 265     /* Make sure buffer is written before we update index. */
 266     smp_wmb();
 267     trace_virtqueue_flush(vq, count);
 268     old = vring_used_idx(vq);
 269     new = old + count;
 270     vring_used_idx_set(vq, new);
 271     vq->inuse -= count;
 272     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
 273         vq->signalled_used_valid = false;
 274 }
 275
 276 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
 277                     unsigned int len)
 278 {
 279     virtqueue_fill(vq, elem, len, 0);
 280     virtqueue_flush(vq, 1);
 281 }
 282
 283 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
 284 {
 285     uint16_t num_heads = vring_avail_idx(vq) - idx;
 286
 287     /* Check it isn't doing very strange things with descriptor numbers. */
 288     if (num_heads > vq->vring.num) {
 289         error_report("Guest moved used index from %u to %u",
 290                      idx, vring_avail_idx(vq));
 291         exit(1);
 292     }
 293     /* On success, callers read a descriptor at vq->last_avail_idx.
 294      * Make sure descriptor read does not bypass avail index read. */
 295     if (num_heads) {
 296         smp_rmb();
 297     }
 298
 299     return num_heads;
 300 }
 301
 302 static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
 303 {
 304     unsigned int head;
 305
 306     /* Grab the next descriptor number they're advertising, and increment
 307      * the index we've seen. */
 308     head = vring_avail_ring(vq, idx % vq->vring.num);
 309
 310     /* If their number is silly, that's a fatal mistake. */
 311     if (head >= vq->vring.num) {
 312         error_report("Guest says index %u is available", head);
 313         exit(1);
 314     }
 315
 316     return head;
 317 }
 318
 319 static unsigned virtqueue_next_desc(hwaddr desc_pa,
 320                                     unsigned int i, unsigned int max)
 321 {
 322     unsigned int next;
 323
 324     /* If this descriptor says it doesn't chain, we're done. */
 325     if (!(vring_desc_flags(desc_pa, i) & VRING_DESC_F_NEXT))
 326         return max;
 327
 328     /* Check they're not leading us off end of descriptors. */
 329     next = vring_desc_next(desc_pa, i);
 330     /* Make sure compiler knows to grab that: we don't want it changing! */
 331     smp_wmb();
 332
 333     if (next >= max) {
 334         error_report("Desc next is %u", next);
 335         exit(1);
 336     }
 337
 338     return next;
 339 }
 340
 341 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
 342                                unsigned int *out_bytes,
 343                                unsigned max_in_bytes, unsigned max_out_bytes)
 344 {
 345     unsigned int idx;
 346     unsigned int total_bufs, in_total, out_total;
 347
 348     idx = vq->last_avail_idx;
 349
 350     total_bufs = in_total = out_total = 0;
 351     while (virtqueue_num_heads(vq, idx)) {
 352         unsigned int max, num_bufs, indirect = 0;
 353         hwaddr desc_pa;
 354         int i;
 355
 356         max = vq->vring.num;
 357         num_bufs = total_bufs;
 358         i = virtqueue_get_head(vq, idx++);
 359         desc_pa = vq->vring.desc;
 360
 361         if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_INDIRECT) {
 362             if (vring_desc_len(desc_pa, i) % sizeof(VRingDesc)) {
 363                 error_report("Invalid size for indirect buffer table");
 364                 exit(1);
 365             }
 366
 367             /* If we've got too many, that implies a descriptor loop. */
 368             if (num_bufs >= max) {
 369                 error_report("Looped descriptor");
 370                 exit(1);
 371             }
 372
 373             /* loop over the indirect descriptor table */
 374             indirect = 1;
 375             max = vring_desc_len(desc_pa, i) / sizeof(VRingDesc);
 376             num_bufs = i = 0;
 377             desc_pa = vring_desc_addr(desc_pa, i);
 378         }
 379
 380         do {
 381             /* If we've got too many, that implies a descriptor loop. */
 382             if (++num_bufs > max) {
 383                 error_report("Looped descriptor");
 384                 exit(1);
 385             }
 386
 387             if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_WRITE) {
 388                 in_total += vring_desc_len(desc_pa, i);
 389             } else {
 390                 out_total += vring_desc_len(desc_pa, i);
 391             }
 392             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
 393                 goto done;
 394             }
 395         } while ((i = virtqueue_next_desc(desc_pa, i, max)) != max);
 396
 397         if (!indirect)
 398             total_bufs = num_bufs;
 399         else
 400             total_bufs++;
 401     }
 402 done:
 403     if (in_bytes) {
 404         *in_bytes = in_total;
 405     }
 406     if (out_bytes) {
 407         *out_bytes = out_total;
 408     }
 409 }
 410
 411 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
 412                           unsigned int out_bytes)
 413 {
 414     unsigned int in_total, out_total;
 415
 416     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
 417     return in_bytes <= in_total && out_bytes <= out_total;
 418 }
 419
 420 void virtqueue_map_sg(struct iovec *sg, hwaddr *addr,
 421     size_t num_sg, int is_write)
 422 {
 423     unsigned int i;
 424     hwaddr len;
 425
 426     for (i = 0; i < num_sg; i++) {
 427         len = sg[i].iov_len;
 428         sg[i].iov_base = cpu_physical_memory_map(addr[i], &len, is_write);
 429         if (sg[i].iov_base == NULL || len != sg[i].iov_len) {
 430             error_report("virtio: trying to map MMIO memory");
 431             exit(1);
 432         }
 433     }
 434 }
 435
 436 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
 437 {
 438     unsigned int i, head, max;
 439     hwaddr desc_pa = vq->vring.desc;
 440
 441     if (!virtqueue_num_heads(vq, vq->last_avail_idx))
 442         return 0;
 443
 444     /* When we start there are none of either input nor output. */
 445     elem->out_num = elem->in_num = 0;
 446
 447     max = vq->vring.num;
 448
 449     i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
 450     if (vq->vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX)) {
 451         vring_avail_event(vq, vring_avail_idx(vq));
 452     }
 453
 454     if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_INDIRECT) {
 455         if (vring_desc_len(desc_pa, i) % sizeof(VRingDesc)) {
 456             error_report("Invalid size for indirect buffer table");
 457             exit(1);
 458         }
 459
 460         /* loop over the indirect descriptor table */
 461         max = vring_desc_len(desc_pa, i) / sizeof(VRingDesc);
 462         desc_pa = vring_desc_addr(desc_pa, i);
 463         i = 0;
 464     }
 465
 466     /* Collect all the descriptors */
 467     do {
 468         struct iovec *sg;
 469
 470         if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_WRITE) {
 471             if (elem->in_num >= ARRAY_SIZE(elem->in_sg)) {
 472                 error_report("Too many write descriptors in indirect table");
 473                 exit(1);
 474             }
 475             elem->in_addr[elem->in_num] = vring_desc_addr(desc_pa, i);
 476             sg = &elem->in_sg[elem->in_num++];
 477         } else {
 478             if (elem->out_num >= ARRAY_SIZE(elem->out_sg)) {
 479                 error_report("Too many read descriptors in indirect table");
 480                 exit(1);
 481             }
 482             elem->out_addr[elem->out_num] = vring_desc_addr(desc_pa, i);
 483             sg = &elem->out_sg[elem->out_num++];
 484         }
 485
 486         sg->iov_len = vring_desc_len(desc_pa, i);
 487
 488         /* If we've got too many, that implies a descriptor loop. */
 489         if ((elem->in_num + elem->out_num) > max) {
 490             error_report("Looped descriptor");
 491             exit(1);
 492         }
 493     } while ((i = virtqueue_next_desc(desc_pa, i, max)) != max);
 494
 495     /* Now map what we have collected */
 496     virtqueue_map_sg(elem->in_sg, elem->in_addr, elem->in_num, 1);
 497     virtqueue_map_sg(elem->out_sg, elem->out_addr, elem->out_num, 0);
 498
 499     elem->index = head;
 500
 501     vq->inuse++;
 502
 503     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
 504     return elem->in_num + elem->out_num;
 505 }
 506
 507 /* virtio device */
 508 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
 509 {
 510     if (vdev->binding->notify) {
 511         vdev->binding->notify(vdev->binding_opaque, vector);
 512     }
 513 }
 514
 515 void virtio_update_irq(VirtIODevice *vdev)
 516 {
 517     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
 518 }
 519
 520 void virtio_set_status(VirtIODevice *vdev, uint8_t val)
 521 {
 522     trace_virtio_set_status(vdev, val);
 523
 524     if (vdev->set_status) {
 525         vdev->set_status(vdev, val);
 526     }
 527     vdev->status = val;
 528 }
 529
 530 void virtio_reset(void *opaque)
 531 {
 532     VirtIODevice *vdev = opaque;
 533     int i;
 534
 535     virtio_set_status(vdev, 0);
 536
 537     if (vdev->reset)
 538         vdev->reset(vdev);
 539
 540     vdev->guest_features = 0;
 541     vdev->queue_sel = 0;
 542     vdev->status = 0;
 543     vdev->isr = 0;
 544     vdev->config_vector = VIRTIO_NO_VECTOR;
 545     virtio_notify_vector(vdev, vdev->config_vector);
 546
 547     for(i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 548         vdev->vq[i].vring.desc = 0;
 549         vdev->vq[i].vring.avail = 0;
 550         vdev->vq[i].vring.used = 0;
 551         vdev->vq[i].last_avail_idx = 0;
 552         vdev->vq[i].pa = 0;
 553         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
 554         vdev->vq[i].signalled_used = 0;
 555         vdev->vq[i].signalled_used_valid = false;
 556         vdev->vq[i].notification = true;
 557     }
 558 }
 559
 560 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
 561 {
 562     uint8_t val;
 563
 564     vdev->get_config(vdev, vdev->config);
 565
 566     if (addr > (vdev->config_len - sizeof(val)))
 567         return (uint32_t)-1;
 568
 569     val = ldub_p(vdev->config + addr);
 570     return val;
 571 }
 572
 573 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
 574 {
 575     uint16_t val;
 576
 577     vdev->get_config(vdev, vdev->config);
 578
 579     if (addr > (vdev->config_len - sizeof(val)))
 580         return (uint32_t)-1;
 581
 582     val = lduw_p(vdev->config + addr);
 583     return val;
 584 }
 585
 586 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
 587 {
 588     uint32_t val;
 589
 590     vdev->get_config(vdev, vdev->config);
 591
 592     if (addr > (vdev->config_len - sizeof(val)))
 593         return (uint32_t)-1;
 594
 595     val = ldl_p(vdev->config + addr);
 596     return val;
 597 }
 598
 599 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
 600 {
 601     uint8_t val = data;
 602
 603     if (addr > (vdev->config_len - sizeof(val)))
 604         return;
 605
 606     stb_p(vdev->config + addr, val);
 607
 608     if (vdev->set_config)
 609         vdev->set_config(vdev, vdev->config);
 610 }
 611
 612 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
 613 {
 614     uint16_t val = data;
 615
 616     if (addr > (vdev->config_len - sizeof(val)))
 617         return;
 618
 619     stw_p(vdev->config + addr, val);
 620
 621     if (vdev->set_config)
 622         vdev->set_config(vdev, vdev->config);
 623 }
 624
 625 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
 626 {
 627     uint32_t val = data;
 628
 629     if (addr > (vdev->config_len - sizeof(val)))
 630         return;
 631
 632     stl_p(vdev->config + addr, val);
 633
 634     if (vdev->set_config)
 635         vdev->set_config(vdev, vdev->config);
 636 }
 637
 638 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
 639 {
 640     vdev->vq[n].pa = addr;
 641     virtqueue_init(&vdev->vq[n]);
 642 }
 643
 644 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
 645 {
 646     return vdev->vq[n].pa;
 647 }
 648
 649 int virtio_queue_get_num(VirtIODevice *vdev, int n)
 650 {
 651     return vdev->vq[n].vring.num;
 652 }
 653
 654 int virtio_queue_get_id(VirtQueue *vq)
 655 {
 656     VirtIODevice *vdev = vq->vdev;
 657     assert(vq >= &vdev->vq[0] && vq < &vdev->vq[VIRTIO_PCI_QUEUE_MAX]);
 658     return vq - &vdev->vq[0];
 659 }
 660
 661 void virtio_queue_notify_vq(VirtQueue *vq)
 662 {
 663     if (vq->vring.desc) {
 664         VirtIODevice *vdev = vq->vdev;
 665         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
 666         vq->handle_output(vdev, vq);
 667     }
 668 }
 669
 670 void virtio_queue_notify(VirtIODevice *vdev, int n)
 671 {
 672     virtio_queue_notify_vq(&vdev->vq[n]);
 673 }
 674
 675 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
 676 {
 677     return n < VIRTIO_PCI_QUEUE_MAX ? vdev->vq[n].vector :
 678         VIRTIO_NO_VECTOR;
 679 }
 680
 681 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
 682 {
 683     if (n < VIRTIO_PCI_QUEUE_MAX)
 684         vdev->vq[n].vector = vector;
 685 }
 686
 687 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
 688                             void (*handle_output)(VirtIODevice *, VirtQueue *))
 689 {
 690     int i;
 691
 692     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 693         if (vdev->vq[i].vring.num == 0)
 694             break;
 695     }
 696
 697     if (i == VIRTIO_PCI_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
 698         abort();
 699
 700     vdev->vq[i].vring.num = queue_size;
 701     vdev->vq[i].handle_output = handle_output;
 702
 703     return &vdev->vq[i];
 704 }
 705
 706 void virtio_del_queue(VirtIODevice *vdev, int n)
 707 {
 708     if (n < 0 || n >= VIRTIO_PCI_QUEUE_MAX) {
 709         abort();
 710     }
 711
 712     vdev->vq[n].vring.num = 0;
 713 }
 714
 715 void virtio_irq(VirtQueue *vq)
 716 {
 717     trace_virtio_irq(vq);
 718     vq->vdev->isr |= 0x01;
 719     virtio_notify_vector(vq->vdev, vq->vector);
 720 }
 721
 722 /* Assuming a given event_idx value from the other size, if
 723  * we have just incremented index from old to new_idx,
 724  * should we trigger an event? */
 725 static inline int vring_need_event(uint16_t event, uint16_t new, uint16_t old)
 726 {
 727         /* Note: Xen has similar logic for notification hold-off
 728          * in include/xen/interface/io/ring.h with req_event and req_prod
 729          * corresponding to event_idx + 1 and new respectively.
 730          * Note also that req_event and req_prod in Xen start at 1,
 731          * event indexes in virtio start at 0. */
 732         return (uint16_t)(new - event - 1) < (uint16_t)(new - old);
 733 }
 734
 735 static bool vring_notify(VirtIODevice *vdev, VirtQueue *vq)
 736 {
 737     uint16_t old, new;
 738     bool v;
 739     /* We need to expose used array entries before checking used event. */
 740     smp_mb();
 741     /* Always notify when queue is empty (when feature acknowledge) */
 742     if (((vdev->guest_features & (1 << VIRTIO_F_NOTIFY_ON_EMPTY)) &&
 743          !vq->inuse && vring_avail_idx(vq) == vq->last_avail_idx)) {
 744         return true;
 745     }
 746
 747     if (!(vdev->guest_features & (1 << VIRTIO_RING_F_EVENT_IDX))) {
 748         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
 749     }
 750
 751     v = vq->signalled_used_valid;
 752     vq->signalled_used_valid = true;
 753     old = vq->signalled_used;
 754     new = vq->signalled_used = vring_used_idx(vq);
 755     return !v || vring_need_event(vring_used_event(vq), new, old);
 756 }
 757
 758 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
 759 {
 760     if (!vring_notify(vdev, vq)) {
 761         return;
 762     }
 763
 764     trace_virtio_notify(vdev, vq);
 765     vdev->isr |= 0x01;
 766     virtio_notify_vector(vdev, vq->vector);
 767 }
 768
 769 void virtio_notify_config(VirtIODevice *vdev)
 770 {
 771     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
 772         return;
 773
 774     vdev->isr |= 0x03;
 775     virtio_notify_vector(vdev, vdev->config_vector);
 776 }
 777
 778 void virtio_save(VirtIODevice *vdev, QEMUFile *f)
 779 {
 780     int i;
 781
 782     if (vdev->binding->save_config)
 783         vdev->binding->save_config(vdev->binding_opaque, f);
 784
 785     qemu_put_8s(f, &vdev->status);
 786     qemu_put_8s(f, &vdev->isr);
 787     qemu_put_be16s(f, &vdev->queue_sel);
 788     qemu_put_be32s(f, &vdev->guest_features);
 789     qemu_put_be32(f, vdev->config_len);
 790     qemu_put_buffer(f, vdev->config, vdev->config_len);
 791
 792     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 793         if (vdev->vq[i].vring.num == 0)
 794             break;
 795     }
 796
 797     qemu_put_be32(f, i);
 798
 799     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 800         if (vdev->vq[i].vring.num == 0)
 801             break;
 802
 803         qemu_put_be32(f, vdev->vq[i].vring.num);
 804         qemu_put_be64(f, vdev->vq[i].pa);
 805         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
 806         if (vdev->binding->save_queue)
 807             vdev->binding->save_queue(vdev->binding_opaque, i, f);
 808     }
 809 }
 810
 811 int virtio_set_features(VirtIODevice *vdev, uint32_t val)
 812 {
 813     uint32_t supported_features =
 814         vdev->binding->get_features(vdev->binding_opaque);
 815     bool bad = (val & ~supported_features) != 0;
 816
 817     val &= supported_features;
 818     if (vdev->set_features) {
 819         vdev->set_features(vdev, val);
 820     }
 821     vdev->guest_features = val;
 822     return bad ? -1 : 0;
 823 }
 824
 825 int virtio_load(VirtIODevice *vdev, QEMUFile *f)
 826 {
 827     int num, i, ret;
 828     uint32_t features;
 829     uint32_t supported_features;
 830
 831     if (vdev->binding->load_config) {
 832         ret = vdev->binding->load_config(vdev->binding_opaque, f);
 833         if (ret)
 834             return ret;
 835     }
 836
 837     qemu_get_8s(f, &vdev->status);
 838     qemu_get_8s(f, &vdev->isr);
 839     qemu_get_be16s(f, &vdev->queue_sel);
 840     qemu_get_be32s(f, &features);
 841
 842     if (virtio_set_features(vdev, features) < 0) {
 843         supported_features = vdev->binding->get_features(vdev->binding_opaque);
 844         error_report("Features 0x%x unsupported. Allowed features: 0x%x",
 845                      features, supported_features);
 846         return -1;
 847     }
 848     vdev->config_len = qemu_get_be32(f);
 849     qemu_get_buffer(f, vdev->config, vdev->config_len);
 850
 851     num = qemu_get_be32(f);
 852
 853     for (i = 0; i < num; i++) {
 854         vdev->vq[i].vring.num = qemu_get_be32(f);
 855         vdev->vq[i].pa = qemu_get_be64(f);
 856         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
 857         vdev->vq[i].signalled_used_valid = false;
 858         vdev->vq[i].notification = true;
 859
 860         if (vdev->vq[i].pa) {
 861             uint16_t nheads;
 862             virtqueue_init(&vdev->vq[i]);
 863             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
 864             /* Check it isn't doing very strange things with descriptor numbers. */
 865             if (nheads > vdev->vq[i].vring.num) {
 866                 error_report("VQ %d size 0x%x Guest index 0x%x "
 867                              "inconsistent with Host index 0x%x: delta 0x%x",
 868                              i, vdev->vq[i].vring.num,
 869                              vring_avail_idx(&vdev->vq[i]),
 870                              vdev->vq[i].last_avail_idx, nheads);
 871                 return -1;
 872             }
 873         } else if (vdev->vq[i].last_avail_idx) {
 874             error_report("VQ %d address 0x0 "
 875                          "inconsistent with Host index 0x%x",
 876                          i, vdev->vq[i].last_avail_idx);
 877                 return -1;
 878         }
 879         if (vdev->binding->load_queue) {
 880             ret = vdev->binding->load_queue(vdev->binding_opaque, i, f);
 881             if (ret)
 882                 return ret;
 883         }
 884     }
 885
 886     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
 887     return 0;
 888 }
 889
 890 void virtio_common_cleanup(VirtIODevice *vdev)
 891 {
 892     qemu_del_vm_change_state_handler(vdev->vmstate);
 893     g_free(vdev->config);
 894     g_free(vdev->vq);
 895 }
 896
 897 void virtio_cleanup(VirtIODevice *vdev)
 898 {
 899     virtio_common_cleanup(vdev);
 900     g_free(vdev);
 901 }
 902
 903 static void virtio_vmstate_change(void *opaque, int running, RunState state)
 904 {
 905     VirtIODevice *vdev = opaque;
 906     bool backend_run = running && (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK);
 907     vdev->vm_running = running;
 908
 909     if (backend_run) {
 910         virtio_set_status(vdev, vdev->status);
 911     }
 912
 913     if (vdev->binding->vmstate_change) {
 914         vdev->binding->vmstate_change(vdev->binding_opaque, backend_run);
 915     }
 916
 917     if (!backend_run) {
 918         virtio_set_status(vdev, vdev->status);
 919     }
 920 }
 921
 922 void virtio_init(VirtIODevice *vdev, const char *name,
 923                  uint16_t device_id, size_t config_size)
 924 {
 925     int i;
 926     vdev->device_id = device_id;
 927     vdev->status = 0;
 928     vdev->isr = 0;
 929     vdev->queue_sel = 0;
 930     vdev->config_vector = VIRTIO_NO_VECTOR;
 931     vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_PCI_QUEUE_MAX);
 932     vdev->vm_running = runstate_is_running();
 933     for (i = 0; i < VIRTIO_PCI_QUEUE_MAX; i++) {
 934         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
 935         vdev->vq[i].vdev = vdev;
 936         vdev->vq[i].queue_index = i;
 937     }
 938
 939     vdev->name = name;
 940     vdev->config_len = config_size;
 941     if (vdev->config_len) {
 942         vdev->config = g_malloc0(config_size);
 943     } else {
 944         vdev->config = NULL;
 945     }
 946     vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
 947                                                      vdev);
 948 }
 949
 950 VirtIODevice *virtio_common_init(const char *name, uint16_t device_id,
 951                                  size_t config_size, size_t struct_size)
 952 {
 953     VirtIODevice *vdev;
 954     vdev = g_malloc0(struct_size);
 955     virtio_init(vdev, name, device_id, config_size);
 956     return vdev;
 957 }
 958
 959 void virtio_bind_device(VirtIODevice *vdev, const VirtIOBindings *binding,
 960                         DeviceState *opaque)
 961 {
 962     vdev->binding = binding;
 963     vdev->binding_opaque = opaque;
 964 }
 965
 966 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
 967 {
 968     return vdev->vq[n].vring.desc;
 969 }
 970
 971 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
 972 {
 973     return vdev->vq[n].vring.avail;
 974 }
 975
 976 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
 977 {
 978     return vdev->vq[n].vring.used;
 979 }
 980
 981 hwaddr virtio_queue_get_ring_addr(VirtIODevice *vdev, int n)
 982 {
 983     return vdev->vq[n].vring.desc;
 984 }
 985
 986 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
 987 {
 988     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
 989 }
 990
 991 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
 992 {
 993     return offsetof(VRingAvail, ring) +
 994         sizeof(uint64_t) * vdev->vq[n].vring.num;
 995 }
 996
 997 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
 998 {
 999     return offsetof(VRingUsed, ring) +
1000         sizeof(VRingUsedElem) * vdev->vq[n].vring.num;
1001 }
1002
1003 hwaddr virtio_queue_get_ring_size(VirtIODevice *vdev, int n)
1004 {
1005     return vdev->vq[n].vring.used - vdev->vq[n].vring.desc +
1006             virtio_queue_get_used_size(vdev, n);
1007 }
1008
1009 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
1010 {
1011     return vdev->vq[n].last_avail_idx;
1012 }
1013
1014 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx)
1015 {
1016     vdev->vq[n].last_avail_idx = idx;
1017 }
1018
1019 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
1020 {
1021     return vdev->vq + n;
1022 }
1023
1024 uint16_t virtio_get_queue_index(VirtQueue *vq)
1025 {
1026     return vq->queue_index;
1027 }
1028
1029 static void virtio_queue_guest_notifier_read(EventNotifier *n)
1030 {
1031     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
1032     if (event_notifier_test_and_clear(n)) {
1033         virtio_irq(vq);
1034     }
1035 }
1036
1037 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
1038                                                 bool with_irqfd)
1039 {
1040     if (assign && !with_irqfd) {
1041         event_notifier_set_handler(&vq->guest_notifier,
1042                                    virtio_queue_guest_notifier_read);
1043     } else {
1044         event_notifier_set_handler(&vq->guest_notifier, NULL);
1045     }
1046     if (!assign) {
1047         /* Test and clear notifier before closing it,
1048          * in case poll callback didn't have time to run. */
1049         virtio_queue_guest_notifier_read(&vq->guest_notifier);
1050     }
1051 }
1052
1053 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
1054 {
1055     return &vq->guest_notifier;
1056 }
1057
1058 static void virtio_queue_host_notifier_read(EventNotifier *n)
1059 {
1060     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
1061     if (event_notifier_test_and_clear(n)) {
1062         virtio_queue_notify_vq(vq);
1063     }
1064 }
1065
1066 void virtio_queue_set_host_notifier_fd_handler(VirtQueue *vq, bool assign,
1067                                                bool set_handler)
1068 {
1069     if (assign && set_handler) {
1070         event_notifier_set_handler(&vq->host_notifier,
1071                                    virtio_queue_host_notifier_read);
1072     } else {
1073         event_notifier_set_handler(&vq->host_notifier, NULL);
1074     }
1075     if (!assign) {
1076         /* Test and clear notifier before after disabling event,
1077          * in case poll callback didn't have time to run. */
1078         virtio_queue_host_notifier_read(&vq->host_notifier);
1079     }
1080 }
1081
1082 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
1083 {
1084     return &vq->host_notifier;
1085 }
1086
1087 static int virtio_device_init(DeviceState *qdev)
1088 {
1089     VirtIODevice *vdev = VIRTIO_DEVICE(qdev);
1090     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(qdev);
1091     assert(k->init != NULL);
1092     if (k->init(vdev) < 0) {
1093         return -1;
1094     }
1095     virtio_bus_plug_device(vdev);
1096     return 0;
1097 }
1098
1099 static void virtio_device_class_init(ObjectClass *klass, void *data)
1100 {
1101     /* Set the default value here. */
1102     DeviceClass *dc = DEVICE_CLASS(klass);
1103     dc->init = virtio_device_init;
1104     dc->bus_type = TYPE_VIRTIO_BUS;
1105 }
1106
1107 static const TypeInfo virtio_device_info = {
1108     .name = TYPE_VIRTIO_DEVICE,
1109     .parent = TYPE_DEVICE,
1110     .instance_size = sizeof(VirtIODevice),
1111     .class_init = virtio_device_class_init,
1112     .abstract = true,
1113     .class_size = sizeof(VirtioDeviceClass),
1114 };
1115
1116 static void virtio_register_types(void)
1117 {
1118     type_register_static(&virtio_device_info);
1119 }
1120
1121 type_init(virtio_register_types)