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 "qemu/osdep.h" 15 #include "qapi/error.h" 16 #include "cpu.h" 17 #include "trace.h" 18 #include "exec/address-spaces.h" 19 #include "qemu/error-report.h" 20 #include "qemu/module.h" 21 #include "hw/virtio/virtio.h" 22 #include "migration/qemu-file-types.h" 23 #include "qemu/atomic.h" 24 #include "hw/virtio/virtio-bus.h" 25 #include "hw/virtio/virtio-access.h" 26 #include "sysemu/dma.h" 27 28 /* 29 * The alignment to use between consumer and producer parts of vring. 30 * x86 pagesize again. This is the default, used by transports like PCI 31 * which don't provide a means for the guest to tell the host the alignment. 32 */ 33 #define VIRTIO_PCI_VRING_ALIGN 4096 34 35 typedef struct VRingDesc 36 { 37 uint64_t addr; 38 uint32_t len; 39 uint16_t flags; 40 uint16_t next; 41 } VRingDesc; 42 43 typedef struct VRingAvail 44 { 45 uint16_t flags; 46 uint16_t idx; 47 uint16_t ring[0]; 48 } VRingAvail; 49 50 typedef struct VRingUsedElem 51 { 52 uint32_t id; 53 uint32_t len; 54 } VRingUsedElem; 55 56 typedef struct VRingUsed 57 { 58 uint16_t flags; 59 uint16_t idx; 60 VRingUsedElem ring[0]; 61 } VRingUsed; 62 63 typedef struct VRingMemoryRegionCaches { 64 struct rcu_head rcu; 65 MemoryRegionCache desc; 66 MemoryRegionCache avail; 67 MemoryRegionCache used; 68 } VRingMemoryRegionCaches; 69 70 typedef struct VRing 71 { 72 unsigned int num; 73 unsigned int num_default; 74 unsigned int align; 75 hwaddr desc; 76 hwaddr avail; 77 hwaddr used; 78 VRingMemoryRegionCaches *caches; 79 } VRing; 80 81 struct VirtQueue 82 { 83 VRing vring; 84 85 /* Next head to pop */ 86 uint16_t last_avail_idx; 87 88 /* Last avail_idx read from VQ. */ 89 uint16_t shadow_avail_idx; 90 91 uint16_t used_idx; 92 93 /* Last used index value we have signalled on */ 94 uint16_t signalled_used; 95 96 /* Last used index value we have signalled on */ 97 bool signalled_used_valid; 98 99 /* Notification enabled? */ 100 bool notification; 101 102 uint16_t queue_index; 103 104 unsigned int inuse; 105 106 uint16_t vector; 107 VirtIOHandleOutput handle_output; 108 VirtIOHandleAIOOutput handle_aio_output; 109 VirtIODevice *vdev; 110 EventNotifier guest_notifier; 111 EventNotifier host_notifier; 112 QLIST_ENTRY(VirtQueue) node; 113 }; 114 115 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches) 116 { 117 if (!caches) { 118 return; 119 } 120 121 address_space_cache_destroy(&caches->desc); 122 address_space_cache_destroy(&caches->avail); 123 address_space_cache_destroy(&caches->used); 124 g_free(caches); 125 } 126 127 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq) 128 { 129 VRingMemoryRegionCaches *caches; 130 131 caches = atomic_read(&vq->vring.caches); 132 atomic_rcu_set(&vq->vring.caches, NULL); 133 if (caches) { 134 call_rcu(caches, virtio_free_region_cache, rcu); 135 } 136 } 137 138 static void virtio_init_region_cache(VirtIODevice *vdev, int n) 139 { 140 VirtQueue *vq = &vdev->vq[n]; 141 VRingMemoryRegionCaches *old = vq->vring.caches; 142 VRingMemoryRegionCaches *new = NULL; 143 hwaddr addr, size; 144 int event_size; 145 int64_t len; 146 147 event_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; 148 149 addr = vq->vring.desc; 150 if (!addr) { 151 goto out_no_cache; 152 } 153 new = g_new0(VRingMemoryRegionCaches, 1); 154 size = virtio_queue_get_desc_size(vdev, n); 155 len = address_space_cache_init(&new->desc, vdev->dma_as, 156 addr, size, false); 157 if (len < size) { 158 virtio_error(vdev, "Cannot map desc"); 159 goto err_desc; 160 } 161 162 size = virtio_queue_get_used_size(vdev, n) + event_size; 163 len = address_space_cache_init(&new->used, vdev->dma_as, 164 vq->vring.used, size, true); 165 if (len < size) { 166 virtio_error(vdev, "Cannot map used"); 167 goto err_used; 168 } 169 170 size = virtio_queue_get_avail_size(vdev, n) + event_size; 171 len = address_space_cache_init(&new->avail, vdev->dma_as, 172 vq->vring.avail, size, false); 173 if (len < size) { 174 virtio_error(vdev, "Cannot map avail"); 175 goto err_avail; 176 } 177 178 atomic_rcu_set(&vq->vring.caches, new); 179 if (old) { 180 call_rcu(old, virtio_free_region_cache, rcu); 181 } 182 return; 183 184 err_avail: 185 address_space_cache_destroy(&new->avail); 186 err_used: 187 address_space_cache_destroy(&new->used); 188 err_desc: 189 address_space_cache_destroy(&new->desc); 190 out_no_cache: 191 g_free(new); 192 virtio_virtqueue_reset_region_cache(vq); 193 } 194 195 /* virt queue functions */ 196 void virtio_queue_update_rings(VirtIODevice *vdev, int n) 197 { 198 VRing *vring = &vdev->vq[n].vring; 199 200 if (!vring->num || !vring->desc || !vring->align) { 201 /* not yet setup -> nothing to do */ 202 return; 203 } 204 vring->avail = vring->desc + vring->num * sizeof(VRingDesc); 205 vring->used = vring_align(vring->avail + 206 offsetof(VRingAvail, ring[vring->num]), 207 vring->align); 208 virtio_init_region_cache(vdev, n); 209 } 210 211 /* Called within rcu_read_lock(). */ 212 static void vring_desc_read(VirtIODevice *vdev, VRingDesc *desc, 213 MemoryRegionCache *cache, int i) 214 { 215 address_space_read_cached(cache, i * sizeof(VRingDesc), 216 desc, sizeof(VRingDesc)); 217 virtio_tswap64s(vdev, &desc->addr); 218 virtio_tswap32s(vdev, &desc->len); 219 virtio_tswap16s(vdev, &desc->flags); 220 virtio_tswap16s(vdev, &desc->next); 221 } 222 223 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq) 224 { 225 VRingMemoryRegionCaches *caches = atomic_rcu_read(&vq->vring.caches); 226 assert(caches != NULL); 227 return caches; 228 } 229 /* Called within rcu_read_lock(). */ 230 static inline uint16_t vring_avail_flags(VirtQueue *vq) 231 { 232 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 233 hwaddr pa = offsetof(VRingAvail, flags); 234 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa); 235 } 236 237 /* Called within rcu_read_lock(). */ 238 static inline uint16_t vring_avail_idx(VirtQueue *vq) 239 { 240 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 241 hwaddr pa = offsetof(VRingAvail, idx); 242 vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa); 243 return vq->shadow_avail_idx; 244 } 245 246 /* Called within rcu_read_lock(). */ 247 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i) 248 { 249 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 250 hwaddr pa = offsetof(VRingAvail, ring[i]); 251 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa); 252 } 253 254 /* Called within rcu_read_lock(). */ 255 static inline uint16_t vring_get_used_event(VirtQueue *vq) 256 { 257 return vring_avail_ring(vq, vq->vring.num); 258 } 259 260 /* Called within rcu_read_lock(). */ 261 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem, 262 int i) 263 { 264 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 265 hwaddr pa = offsetof(VRingUsed, ring[i]); 266 virtio_tswap32s(vq->vdev, &uelem->id); 267 virtio_tswap32s(vq->vdev, &uelem->len); 268 address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem)); 269 address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem)); 270 } 271 272 /* Called within rcu_read_lock(). */ 273 static uint16_t vring_used_idx(VirtQueue *vq) 274 { 275 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 276 hwaddr pa = offsetof(VRingUsed, idx); 277 return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa); 278 } 279 280 /* Called within rcu_read_lock(). */ 281 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val) 282 { 283 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 284 hwaddr pa = offsetof(VRingUsed, idx); 285 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val); 286 address_space_cache_invalidate(&caches->used, pa, sizeof(val)); 287 vq->used_idx = val; 288 } 289 290 /* Called within rcu_read_lock(). */ 291 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask) 292 { 293 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 294 VirtIODevice *vdev = vq->vdev; 295 hwaddr pa = offsetof(VRingUsed, flags); 296 uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa); 297 298 virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask); 299 address_space_cache_invalidate(&caches->used, pa, sizeof(flags)); 300 } 301 302 /* Called within rcu_read_lock(). */ 303 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask) 304 { 305 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 306 VirtIODevice *vdev = vq->vdev; 307 hwaddr pa = offsetof(VRingUsed, flags); 308 uint16_t flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa); 309 310 virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask); 311 address_space_cache_invalidate(&caches->used, pa, sizeof(flags)); 312 } 313 314 /* Called within rcu_read_lock(). */ 315 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val) 316 { 317 VRingMemoryRegionCaches *caches; 318 hwaddr pa; 319 if (!vq->notification) { 320 return; 321 } 322 323 caches = vring_get_region_caches(vq); 324 pa = offsetof(VRingUsed, ring[vq->vring.num]); 325 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val); 326 address_space_cache_invalidate(&caches->used, pa, sizeof(val)); 327 } 328 329 void virtio_queue_set_notification(VirtQueue *vq, int enable) 330 { 331 vq->notification = enable; 332 333 if (!vq->vring.desc) { 334 return; 335 } 336 337 rcu_read_lock(); 338 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) { 339 vring_set_avail_event(vq, vring_avail_idx(vq)); 340 } else if (enable) { 341 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY); 342 } else { 343 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY); 344 } 345 if (enable) { 346 /* Expose avail event/used flags before caller checks the avail idx. */ 347 smp_mb(); 348 } 349 rcu_read_unlock(); 350 } 351 352 int virtio_queue_ready(VirtQueue *vq) 353 { 354 return vq->vring.avail != 0; 355 } 356 357 /* Fetch avail_idx from VQ memory only when we really need to know if 358 * guest has added some buffers. 359 * Called within rcu_read_lock(). */ 360 static int virtio_queue_empty_rcu(VirtQueue *vq) 361 { 362 if (unlikely(vq->vdev->broken)) { 363 return 1; 364 } 365 366 if (unlikely(!vq->vring.avail)) { 367 return 1; 368 } 369 370 if (vq->shadow_avail_idx != vq->last_avail_idx) { 371 return 0; 372 } 373 374 return vring_avail_idx(vq) == vq->last_avail_idx; 375 } 376 377 int virtio_queue_empty(VirtQueue *vq) 378 { 379 bool empty; 380 381 if (unlikely(vq->vdev->broken)) { 382 return 1; 383 } 384 385 if (unlikely(!vq->vring.avail)) { 386 return 1; 387 } 388 389 if (vq->shadow_avail_idx != vq->last_avail_idx) { 390 return 0; 391 } 392 393 rcu_read_lock(); 394 empty = vring_avail_idx(vq) == vq->last_avail_idx; 395 rcu_read_unlock(); 396 return empty; 397 } 398 399 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem, 400 unsigned int len) 401 { 402 AddressSpace *dma_as = vq->vdev->dma_as; 403 unsigned int offset; 404 int i; 405 406 offset = 0; 407 for (i = 0; i < elem->in_num; i++) { 408 size_t size = MIN(len - offset, elem->in_sg[i].iov_len); 409 410 dma_memory_unmap(dma_as, elem->in_sg[i].iov_base, 411 elem->in_sg[i].iov_len, 412 DMA_DIRECTION_FROM_DEVICE, size); 413 414 offset += size; 415 } 416 417 for (i = 0; i < elem->out_num; i++) 418 dma_memory_unmap(dma_as, elem->out_sg[i].iov_base, 419 elem->out_sg[i].iov_len, 420 DMA_DIRECTION_TO_DEVICE, 421 elem->out_sg[i].iov_len); 422 } 423 424 /* virtqueue_detach_element: 425 * @vq: The #VirtQueue 426 * @elem: The #VirtQueueElement 427 * @len: number of bytes written 428 * 429 * Detach the element from the virtqueue. This function is suitable for device 430 * reset or other situations where a #VirtQueueElement is simply freed and will 431 * not be pushed or discarded. 432 */ 433 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem, 434 unsigned int len) 435 { 436 vq->inuse--; 437 virtqueue_unmap_sg(vq, elem, len); 438 } 439 440 /* virtqueue_unpop: 441 * @vq: The #VirtQueue 442 * @elem: The #VirtQueueElement 443 * @len: number of bytes written 444 * 445 * Pretend the most recent element wasn't popped from the virtqueue. The next 446 * call to virtqueue_pop() will refetch the element. 447 */ 448 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem, 449 unsigned int len) 450 { 451 vq->last_avail_idx--; 452 virtqueue_detach_element(vq, elem, len); 453 } 454 455 /* virtqueue_rewind: 456 * @vq: The #VirtQueue 457 * @num: Number of elements to push back 458 * 459 * Pretend that elements weren't popped from the virtqueue. The next 460 * virtqueue_pop() will refetch the oldest element. 461 * 462 * Use virtqueue_unpop() instead if you have a VirtQueueElement. 463 * 464 * Returns: true on success, false if @num is greater than the number of in use 465 * elements. 466 */ 467 bool virtqueue_rewind(VirtQueue *vq, unsigned int num) 468 { 469 if (num > vq->inuse) { 470 return false; 471 } 472 vq->last_avail_idx -= num; 473 vq->inuse -= num; 474 return true; 475 } 476 477 /* Called within rcu_read_lock(). */ 478 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem, 479 unsigned int len, unsigned int idx) 480 { 481 VRingUsedElem uelem; 482 483 trace_virtqueue_fill(vq, elem, len, idx); 484 485 virtqueue_unmap_sg(vq, elem, len); 486 487 if (unlikely(vq->vdev->broken)) { 488 return; 489 } 490 491 if (unlikely(!vq->vring.used)) { 492 return; 493 } 494 495 idx = (idx + vq->used_idx) % vq->vring.num; 496 497 uelem.id = elem->index; 498 uelem.len = len; 499 vring_used_write(vq, &uelem, idx); 500 } 501 502 /* Called within rcu_read_lock(). */ 503 void virtqueue_flush(VirtQueue *vq, unsigned int count) 504 { 505 uint16_t old, new; 506 507 if (unlikely(vq->vdev->broken)) { 508 vq->inuse -= count; 509 return; 510 } 511 512 if (unlikely(!vq->vring.used)) { 513 return; 514 } 515 516 /* Make sure buffer is written before we update index. */ 517 smp_wmb(); 518 trace_virtqueue_flush(vq, count); 519 old = vq->used_idx; 520 new = old + count; 521 vring_used_idx_set(vq, new); 522 vq->inuse -= count; 523 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old))) 524 vq->signalled_used_valid = false; 525 } 526 527 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem, 528 unsigned int len) 529 { 530 rcu_read_lock(); 531 virtqueue_fill(vq, elem, len, 0); 532 virtqueue_flush(vq, 1); 533 rcu_read_unlock(); 534 } 535 536 /* Called within rcu_read_lock(). */ 537 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx) 538 { 539 uint16_t num_heads = vring_avail_idx(vq) - idx; 540 541 /* Check it isn't doing very strange things with descriptor numbers. */ 542 if (num_heads > vq->vring.num) { 543 virtio_error(vq->vdev, "Guest moved used index from %u to %u", 544 idx, vq->shadow_avail_idx); 545 return -EINVAL; 546 } 547 /* On success, callers read a descriptor at vq->last_avail_idx. 548 * Make sure descriptor read does not bypass avail index read. */ 549 if (num_heads) { 550 smp_rmb(); 551 } 552 553 return num_heads; 554 } 555 556 /* Called within rcu_read_lock(). */ 557 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx, 558 unsigned int *head) 559 { 560 /* Grab the next descriptor number they're advertising, and increment 561 * the index we've seen. */ 562 *head = vring_avail_ring(vq, idx % vq->vring.num); 563 564 /* If their number is silly, that's a fatal mistake. */ 565 if (*head >= vq->vring.num) { 566 virtio_error(vq->vdev, "Guest says index %u is available", *head); 567 return false; 568 } 569 570 return true; 571 } 572 573 enum { 574 VIRTQUEUE_READ_DESC_ERROR = -1, 575 VIRTQUEUE_READ_DESC_DONE = 0, /* end of chain */ 576 VIRTQUEUE_READ_DESC_MORE = 1, /* more buffers in chain */ 577 }; 578 579 static int virtqueue_read_next_desc(VirtIODevice *vdev, VRingDesc *desc, 580 MemoryRegionCache *desc_cache, unsigned int max, 581 unsigned int *next) 582 { 583 /* If this descriptor says it doesn't chain, we're done. */ 584 if (!(desc->flags & VRING_DESC_F_NEXT)) { 585 return VIRTQUEUE_READ_DESC_DONE; 586 } 587 588 /* Check they're not leading us off end of descriptors. */ 589 *next = desc->next; 590 /* Make sure compiler knows to grab that: we don't want it changing! */ 591 smp_wmb(); 592 593 if (*next >= max) { 594 virtio_error(vdev, "Desc next is %u", *next); 595 return VIRTQUEUE_READ_DESC_ERROR; 596 } 597 598 vring_desc_read(vdev, desc, desc_cache, *next); 599 return VIRTQUEUE_READ_DESC_MORE; 600 } 601 602 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes, 603 unsigned int *out_bytes, 604 unsigned max_in_bytes, unsigned max_out_bytes) 605 { 606 VirtIODevice *vdev = vq->vdev; 607 unsigned int max, idx; 608 unsigned int total_bufs, in_total, out_total; 609 VRingMemoryRegionCaches *caches; 610 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; 611 int64_t len = 0; 612 int rc; 613 614 if (unlikely(!vq->vring.desc)) { 615 if (in_bytes) { 616 *in_bytes = 0; 617 } 618 if (out_bytes) { 619 *out_bytes = 0; 620 } 621 return; 622 } 623 624 rcu_read_lock(); 625 idx = vq->last_avail_idx; 626 total_bufs = in_total = out_total = 0; 627 628 max = vq->vring.num; 629 caches = vring_get_region_caches(vq); 630 if (caches->desc.len < max * sizeof(VRingDesc)) { 631 virtio_error(vdev, "Cannot map descriptor ring"); 632 goto err; 633 } 634 635 while ((rc = virtqueue_num_heads(vq, idx)) > 0) { 636 MemoryRegionCache *desc_cache = &caches->desc; 637 unsigned int num_bufs; 638 VRingDesc desc; 639 unsigned int i; 640 641 num_bufs = total_bufs; 642 643 if (!virtqueue_get_head(vq, idx++, &i)) { 644 goto err; 645 } 646 647 vring_desc_read(vdev, &desc, desc_cache, i); 648 649 if (desc.flags & VRING_DESC_F_INDIRECT) { 650 if (!desc.len || (desc.len % sizeof(VRingDesc))) { 651 virtio_error(vdev, "Invalid size for indirect buffer table"); 652 goto err; 653 } 654 655 /* If we've got too many, that implies a descriptor loop. */ 656 if (num_bufs >= max) { 657 virtio_error(vdev, "Looped descriptor"); 658 goto err; 659 } 660 661 /* loop over the indirect descriptor table */ 662 len = address_space_cache_init(&indirect_desc_cache, 663 vdev->dma_as, 664 desc.addr, desc.len, false); 665 desc_cache = &indirect_desc_cache; 666 if (len < desc.len) { 667 virtio_error(vdev, "Cannot map indirect buffer"); 668 goto err; 669 } 670 671 max = desc.len / sizeof(VRingDesc); 672 num_bufs = i = 0; 673 vring_desc_read(vdev, &desc, desc_cache, i); 674 } 675 676 do { 677 /* If we've got too many, that implies a descriptor loop. */ 678 if (++num_bufs > max) { 679 virtio_error(vdev, "Looped descriptor"); 680 goto err; 681 } 682 683 if (desc.flags & VRING_DESC_F_WRITE) { 684 in_total += desc.len; 685 } else { 686 out_total += desc.len; 687 } 688 if (in_total >= max_in_bytes && out_total >= max_out_bytes) { 689 goto done; 690 } 691 692 rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i); 693 } while (rc == VIRTQUEUE_READ_DESC_MORE); 694 695 if (rc == VIRTQUEUE_READ_DESC_ERROR) { 696 goto err; 697 } 698 699 if (desc_cache == &indirect_desc_cache) { 700 address_space_cache_destroy(&indirect_desc_cache); 701 total_bufs++; 702 } else { 703 total_bufs = num_bufs; 704 } 705 } 706 707 if (rc < 0) { 708 goto err; 709 } 710 711 done: 712 address_space_cache_destroy(&indirect_desc_cache); 713 if (in_bytes) { 714 *in_bytes = in_total; 715 } 716 if (out_bytes) { 717 *out_bytes = out_total; 718 } 719 rcu_read_unlock(); 720 return; 721 722 err: 723 in_total = out_total = 0; 724 goto done; 725 } 726 727 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes, 728 unsigned int out_bytes) 729 { 730 unsigned int in_total, out_total; 731 732 virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes); 733 return in_bytes <= in_total && out_bytes <= out_total; 734 } 735 736 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg, 737 hwaddr *addr, struct iovec *iov, 738 unsigned int max_num_sg, bool is_write, 739 hwaddr pa, size_t sz) 740 { 741 bool ok = false; 742 unsigned num_sg = *p_num_sg; 743 assert(num_sg <= max_num_sg); 744 745 if (!sz) { 746 virtio_error(vdev, "virtio: zero sized buffers are not allowed"); 747 goto out; 748 } 749 750 while (sz) { 751 hwaddr len = sz; 752 753 if (num_sg == max_num_sg) { 754 virtio_error(vdev, "virtio: too many write descriptors in " 755 "indirect table"); 756 goto out; 757 } 758 759 iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len, 760 is_write ? 761 DMA_DIRECTION_FROM_DEVICE : 762 DMA_DIRECTION_TO_DEVICE); 763 if (!iov[num_sg].iov_base) { 764 virtio_error(vdev, "virtio: bogus descriptor or out of resources"); 765 goto out; 766 } 767 768 iov[num_sg].iov_len = len; 769 addr[num_sg] = pa; 770 771 sz -= len; 772 pa += len; 773 num_sg++; 774 } 775 ok = true; 776 777 out: 778 *p_num_sg = num_sg; 779 return ok; 780 } 781 782 /* Only used by error code paths before we have a VirtQueueElement (therefore 783 * virtqueue_unmap_sg() can't be used). Assumes buffers weren't written to 784 * yet. 785 */ 786 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num, 787 struct iovec *iov) 788 { 789 unsigned int i; 790 791 for (i = 0; i < out_num + in_num; i++) { 792 int is_write = i >= out_num; 793 794 cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0); 795 iov++; 796 } 797 } 798 799 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg, 800 hwaddr *addr, unsigned int num_sg, 801 int is_write) 802 { 803 unsigned int i; 804 hwaddr len; 805 806 for (i = 0; i < num_sg; i++) { 807 len = sg[i].iov_len; 808 sg[i].iov_base = dma_memory_map(vdev->dma_as, 809 addr[i], &len, is_write ? 810 DMA_DIRECTION_FROM_DEVICE : 811 DMA_DIRECTION_TO_DEVICE); 812 if (!sg[i].iov_base) { 813 error_report("virtio: error trying to map MMIO memory"); 814 exit(1); 815 } 816 if (len != sg[i].iov_len) { 817 error_report("virtio: unexpected memory split"); 818 exit(1); 819 } 820 } 821 } 822 823 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem) 824 { 825 virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, 1); 826 virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num, 0); 827 } 828 829 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num) 830 { 831 VirtQueueElement *elem; 832 size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0])); 833 size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]); 834 size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]); 835 size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0])); 836 size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]); 837 size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]); 838 839 assert(sz >= sizeof(VirtQueueElement)); 840 elem = g_malloc(out_sg_end); 841 trace_virtqueue_alloc_element(elem, sz, in_num, out_num); 842 elem->out_num = out_num; 843 elem->in_num = in_num; 844 elem->in_addr = (void *)elem + in_addr_ofs; 845 elem->out_addr = (void *)elem + out_addr_ofs; 846 elem->in_sg = (void *)elem + in_sg_ofs; 847 elem->out_sg = (void *)elem + out_sg_ofs; 848 return elem; 849 } 850 851 void *virtqueue_pop(VirtQueue *vq, size_t sz) 852 { 853 unsigned int i, head, max; 854 VRingMemoryRegionCaches *caches; 855 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; 856 MemoryRegionCache *desc_cache; 857 int64_t len; 858 VirtIODevice *vdev = vq->vdev; 859 VirtQueueElement *elem = NULL; 860 unsigned out_num, in_num, elem_entries; 861 hwaddr addr[VIRTQUEUE_MAX_SIZE]; 862 struct iovec iov[VIRTQUEUE_MAX_SIZE]; 863 VRingDesc desc; 864 int rc; 865 866 if (unlikely(vdev->broken)) { 867 return NULL; 868 } 869 rcu_read_lock(); 870 if (virtio_queue_empty_rcu(vq)) { 871 goto done; 872 } 873 /* Needed after virtio_queue_empty(), see comment in 874 * virtqueue_num_heads(). */ 875 smp_rmb(); 876 877 /* When we start there are none of either input nor output. */ 878 out_num = in_num = elem_entries = 0; 879 880 max = vq->vring.num; 881 882 if (vq->inuse >= vq->vring.num) { 883 virtio_error(vdev, "Virtqueue size exceeded"); 884 goto done; 885 } 886 887 if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) { 888 goto done; 889 } 890 891 if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { 892 vring_set_avail_event(vq, vq->last_avail_idx); 893 } 894 895 i = head; 896 897 caches = vring_get_region_caches(vq); 898 if (caches->desc.len < max * sizeof(VRingDesc)) { 899 virtio_error(vdev, "Cannot map descriptor ring"); 900 goto done; 901 } 902 903 desc_cache = &caches->desc; 904 vring_desc_read(vdev, &desc, desc_cache, i); 905 if (desc.flags & VRING_DESC_F_INDIRECT) { 906 if (!desc.len || (desc.len % sizeof(VRingDesc))) { 907 virtio_error(vdev, "Invalid size for indirect buffer table"); 908 goto done; 909 } 910 911 /* loop over the indirect descriptor table */ 912 len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as, 913 desc.addr, desc.len, false); 914 desc_cache = &indirect_desc_cache; 915 if (len < desc.len) { 916 virtio_error(vdev, "Cannot map indirect buffer"); 917 goto done; 918 } 919 920 max = desc.len / sizeof(VRingDesc); 921 i = 0; 922 vring_desc_read(vdev, &desc, desc_cache, i); 923 } 924 925 /* Collect all the descriptors */ 926 do { 927 bool map_ok; 928 929 if (desc.flags & VRING_DESC_F_WRITE) { 930 map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num, 931 iov + out_num, 932 VIRTQUEUE_MAX_SIZE - out_num, true, 933 desc.addr, desc.len); 934 } else { 935 if (in_num) { 936 virtio_error(vdev, "Incorrect order for descriptors"); 937 goto err_undo_map; 938 } 939 map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov, 940 VIRTQUEUE_MAX_SIZE, false, 941 desc.addr, desc.len); 942 } 943 if (!map_ok) { 944 goto err_undo_map; 945 } 946 947 /* If we've got too many, that implies a descriptor loop. */ 948 if (++elem_entries > max) { 949 virtio_error(vdev, "Looped descriptor"); 950 goto err_undo_map; 951 } 952 953 rc = virtqueue_read_next_desc(vdev, &desc, desc_cache, max, &i); 954 } while (rc == VIRTQUEUE_READ_DESC_MORE); 955 956 if (rc == VIRTQUEUE_READ_DESC_ERROR) { 957 goto err_undo_map; 958 } 959 960 /* Now copy what we have collected and mapped */ 961 elem = virtqueue_alloc_element(sz, out_num, in_num); 962 elem->index = head; 963 for (i = 0; i < out_num; i++) { 964 elem->out_addr[i] = addr[i]; 965 elem->out_sg[i] = iov[i]; 966 } 967 for (i = 0; i < in_num; i++) { 968 elem->in_addr[i] = addr[out_num + i]; 969 elem->in_sg[i] = iov[out_num + i]; 970 } 971 972 vq->inuse++; 973 974 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num); 975 done: 976 address_space_cache_destroy(&indirect_desc_cache); 977 rcu_read_unlock(); 978 979 return elem; 980 981 err_undo_map: 982 virtqueue_undo_map_desc(out_num, in_num, iov); 983 goto done; 984 } 985 986 /* virtqueue_drop_all: 987 * @vq: The #VirtQueue 988 * Drops all queued buffers and indicates them to the guest 989 * as if they are done. Useful when buffers can not be 990 * processed but must be returned to the guest. 991 */ 992 unsigned int virtqueue_drop_all(VirtQueue *vq) 993 { 994 unsigned int dropped = 0; 995 VirtQueueElement elem = {}; 996 VirtIODevice *vdev = vq->vdev; 997 bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); 998 999 if (unlikely(vdev->broken)) { 1000 return 0; 1001 } 1002 1003 while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) { 1004 /* works similar to virtqueue_pop but does not map buffers 1005 * and does not allocate any memory */ 1006 smp_rmb(); 1007 if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) { 1008 break; 1009 } 1010 vq->inuse++; 1011 vq->last_avail_idx++; 1012 if (fEventIdx) { 1013 vring_set_avail_event(vq, vq->last_avail_idx); 1014 } 1015 /* immediately push the element, nothing to unmap 1016 * as both in_num and out_num are set to 0 */ 1017 virtqueue_push(vq, &elem, 0); 1018 dropped++; 1019 } 1020 1021 return dropped; 1022 } 1023 1024 /* Reading and writing a structure directly to QEMUFile is *awful*, but 1025 * it is what QEMU has always done by mistake. We can change it sooner 1026 * or later by bumping the version number of the affected vm states. 1027 * In the meanwhile, since the in-memory layout of VirtQueueElement 1028 * has changed, we need to marshal to and from the layout that was 1029 * used before the change. 1030 */ 1031 typedef struct VirtQueueElementOld { 1032 unsigned int index; 1033 unsigned int out_num; 1034 unsigned int in_num; 1035 hwaddr in_addr[VIRTQUEUE_MAX_SIZE]; 1036 hwaddr out_addr[VIRTQUEUE_MAX_SIZE]; 1037 struct iovec in_sg[VIRTQUEUE_MAX_SIZE]; 1038 struct iovec out_sg[VIRTQUEUE_MAX_SIZE]; 1039 } VirtQueueElementOld; 1040 1041 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz) 1042 { 1043 VirtQueueElement *elem; 1044 VirtQueueElementOld data; 1045 int i; 1046 1047 qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld)); 1048 1049 /* TODO: teach all callers that this can fail, and return failure instead 1050 * of asserting here. 1051 * This is just one thing (there are probably more) that must be 1052 * fixed before we can allow NDEBUG compilation. 1053 */ 1054 assert(ARRAY_SIZE(data.in_addr) >= data.in_num); 1055 assert(ARRAY_SIZE(data.out_addr) >= data.out_num); 1056 1057 elem = virtqueue_alloc_element(sz, data.out_num, data.in_num); 1058 elem->index = data.index; 1059 1060 for (i = 0; i < elem->in_num; i++) { 1061 elem->in_addr[i] = data.in_addr[i]; 1062 } 1063 1064 for (i = 0; i < elem->out_num; i++) { 1065 elem->out_addr[i] = data.out_addr[i]; 1066 } 1067 1068 for (i = 0; i < elem->in_num; i++) { 1069 /* Base is overwritten by virtqueue_map. */ 1070 elem->in_sg[i].iov_base = 0; 1071 elem->in_sg[i].iov_len = data.in_sg[i].iov_len; 1072 } 1073 1074 for (i = 0; i < elem->out_num; i++) { 1075 /* Base is overwritten by virtqueue_map. */ 1076 elem->out_sg[i].iov_base = 0; 1077 elem->out_sg[i].iov_len = data.out_sg[i].iov_len; 1078 } 1079 1080 virtqueue_map(vdev, elem); 1081 return elem; 1082 } 1083 1084 void qemu_put_virtqueue_element(QEMUFile *f, VirtQueueElement *elem) 1085 { 1086 VirtQueueElementOld data; 1087 int i; 1088 1089 memset(&data, 0, sizeof(data)); 1090 data.index = elem->index; 1091 data.in_num = elem->in_num; 1092 data.out_num = elem->out_num; 1093 1094 for (i = 0; i < elem->in_num; i++) { 1095 data.in_addr[i] = elem->in_addr[i]; 1096 } 1097 1098 for (i = 0; i < elem->out_num; i++) { 1099 data.out_addr[i] = elem->out_addr[i]; 1100 } 1101 1102 for (i = 0; i < elem->in_num; i++) { 1103 /* Base is overwritten by virtqueue_map when loading. Do not 1104 * save it, as it would leak the QEMU address space layout. */ 1105 data.in_sg[i].iov_len = elem->in_sg[i].iov_len; 1106 } 1107 1108 for (i = 0; i < elem->out_num; i++) { 1109 /* Do not save iov_base as above. */ 1110 data.out_sg[i].iov_len = elem->out_sg[i].iov_len; 1111 } 1112 qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld)); 1113 } 1114 1115 /* virtio device */ 1116 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector) 1117 { 1118 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 1119 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 1120 1121 if (unlikely(vdev->broken)) { 1122 return; 1123 } 1124 1125 if (k->notify) { 1126 k->notify(qbus->parent, vector); 1127 } 1128 } 1129 1130 void virtio_update_irq(VirtIODevice *vdev) 1131 { 1132 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR); 1133 } 1134 1135 static int virtio_validate_features(VirtIODevice *vdev) 1136 { 1137 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1138 1139 if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) && 1140 !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) { 1141 return -EFAULT; 1142 } 1143 1144 if (k->validate_features) { 1145 return k->validate_features(vdev); 1146 } else { 1147 return 0; 1148 } 1149 } 1150 1151 int virtio_set_status(VirtIODevice *vdev, uint8_t val) 1152 { 1153 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1154 trace_virtio_set_status(vdev, val); 1155 1156 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 1157 if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) && 1158 val & VIRTIO_CONFIG_S_FEATURES_OK) { 1159 int ret = virtio_validate_features(vdev); 1160 1161 if (ret) { 1162 return ret; 1163 } 1164 } 1165 } 1166 1167 if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) != 1168 (val & VIRTIO_CONFIG_S_DRIVER_OK)) { 1169 virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK); 1170 } 1171 1172 if (k->set_status) { 1173 k->set_status(vdev, val); 1174 } 1175 vdev->status = val; 1176 1177 return 0; 1178 } 1179 1180 static enum virtio_device_endian virtio_default_endian(void) 1181 { 1182 if (target_words_bigendian()) { 1183 return VIRTIO_DEVICE_ENDIAN_BIG; 1184 } else { 1185 return VIRTIO_DEVICE_ENDIAN_LITTLE; 1186 } 1187 } 1188 1189 static enum virtio_device_endian virtio_current_cpu_endian(void) 1190 { 1191 CPUClass *cc = CPU_GET_CLASS(current_cpu); 1192 1193 if (cc->virtio_is_big_endian(current_cpu)) { 1194 return VIRTIO_DEVICE_ENDIAN_BIG; 1195 } else { 1196 return VIRTIO_DEVICE_ENDIAN_LITTLE; 1197 } 1198 } 1199 1200 void virtio_reset(void *opaque) 1201 { 1202 VirtIODevice *vdev = opaque; 1203 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1204 int i; 1205 1206 virtio_set_status(vdev, 0); 1207 if (current_cpu) { 1208 /* Guest initiated reset */ 1209 vdev->device_endian = virtio_current_cpu_endian(); 1210 } else { 1211 /* System reset */ 1212 vdev->device_endian = virtio_default_endian(); 1213 } 1214 1215 if (k->reset) { 1216 k->reset(vdev); 1217 } 1218 1219 vdev->start_on_kick = false; 1220 vdev->started = false; 1221 vdev->broken = false; 1222 vdev->guest_features = 0; 1223 vdev->queue_sel = 0; 1224 vdev->status = 0; 1225 atomic_set(&vdev->isr, 0); 1226 vdev->config_vector = VIRTIO_NO_VECTOR; 1227 virtio_notify_vector(vdev, vdev->config_vector); 1228 1229 for(i = 0; i < VIRTIO_QUEUE_MAX; i++) { 1230 vdev->vq[i].vring.desc = 0; 1231 vdev->vq[i].vring.avail = 0; 1232 vdev->vq[i].vring.used = 0; 1233 vdev->vq[i].last_avail_idx = 0; 1234 vdev->vq[i].shadow_avail_idx = 0; 1235 vdev->vq[i].used_idx = 0; 1236 virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR); 1237 vdev->vq[i].signalled_used = 0; 1238 vdev->vq[i].signalled_used_valid = false; 1239 vdev->vq[i].notification = true; 1240 vdev->vq[i].vring.num = vdev->vq[i].vring.num_default; 1241 vdev->vq[i].inuse = 0; 1242 virtio_virtqueue_reset_region_cache(&vdev->vq[i]); 1243 } 1244 } 1245 1246 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr) 1247 { 1248 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1249 uint8_t val; 1250 1251 if (addr + sizeof(val) > vdev->config_len) { 1252 return (uint32_t)-1; 1253 } 1254 1255 k->get_config(vdev, vdev->config); 1256 1257 val = ldub_p(vdev->config + addr); 1258 return val; 1259 } 1260 1261 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr) 1262 { 1263 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1264 uint16_t val; 1265 1266 if (addr + sizeof(val) > vdev->config_len) { 1267 return (uint32_t)-1; 1268 } 1269 1270 k->get_config(vdev, vdev->config); 1271 1272 val = lduw_p(vdev->config + addr); 1273 return val; 1274 } 1275 1276 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr) 1277 { 1278 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1279 uint32_t val; 1280 1281 if (addr + sizeof(val) > vdev->config_len) { 1282 return (uint32_t)-1; 1283 } 1284 1285 k->get_config(vdev, vdev->config); 1286 1287 val = ldl_p(vdev->config + addr); 1288 return val; 1289 } 1290 1291 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data) 1292 { 1293 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1294 uint8_t val = data; 1295 1296 if (addr + sizeof(val) > vdev->config_len) { 1297 return; 1298 } 1299 1300 stb_p(vdev->config + addr, val); 1301 1302 if (k->set_config) { 1303 k->set_config(vdev, vdev->config); 1304 } 1305 } 1306 1307 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data) 1308 { 1309 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1310 uint16_t val = data; 1311 1312 if (addr + sizeof(val) > vdev->config_len) { 1313 return; 1314 } 1315 1316 stw_p(vdev->config + addr, val); 1317 1318 if (k->set_config) { 1319 k->set_config(vdev, vdev->config); 1320 } 1321 } 1322 1323 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data) 1324 { 1325 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1326 uint32_t val = data; 1327 1328 if (addr + sizeof(val) > vdev->config_len) { 1329 return; 1330 } 1331 1332 stl_p(vdev->config + addr, val); 1333 1334 if (k->set_config) { 1335 k->set_config(vdev, vdev->config); 1336 } 1337 } 1338 1339 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr) 1340 { 1341 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1342 uint8_t val; 1343 1344 if (addr + sizeof(val) > vdev->config_len) { 1345 return (uint32_t)-1; 1346 } 1347 1348 k->get_config(vdev, vdev->config); 1349 1350 val = ldub_p(vdev->config + addr); 1351 return val; 1352 } 1353 1354 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr) 1355 { 1356 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1357 uint16_t val; 1358 1359 if (addr + sizeof(val) > vdev->config_len) { 1360 return (uint32_t)-1; 1361 } 1362 1363 k->get_config(vdev, vdev->config); 1364 1365 val = lduw_le_p(vdev->config + addr); 1366 return val; 1367 } 1368 1369 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr) 1370 { 1371 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1372 uint32_t val; 1373 1374 if (addr + sizeof(val) > vdev->config_len) { 1375 return (uint32_t)-1; 1376 } 1377 1378 k->get_config(vdev, vdev->config); 1379 1380 val = ldl_le_p(vdev->config + addr); 1381 return val; 1382 } 1383 1384 void virtio_config_modern_writeb(VirtIODevice *vdev, 1385 uint32_t addr, uint32_t data) 1386 { 1387 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1388 uint8_t val = data; 1389 1390 if (addr + sizeof(val) > vdev->config_len) { 1391 return; 1392 } 1393 1394 stb_p(vdev->config + addr, val); 1395 1396 if (k->set_config) { 1397 k->set_config(vdev, vdev->config); 1398 } 1399 } 1400 1401 void virtio_config_modern_writew(VirtIODevice *vdev, 1402 uint32_t addr, uint32_t data) 1403 { 1404 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1405 uint16_t val = data; 1406 1407 if (addr + sizeof(val) > vdev->config_len) { 1408 return; 1409 } 1410 1411 stw_le_p(vdev->config + addr, val); 1412 1413 if (k->set_config) { 1414 k->set_config(vdev, vdev->config); 1415 } 1416 } 1417 1418 void virtio_config_modern_writel(VirtIODevice *vdev, 1419 uint32_t addr, uint32_t data) 1420 { 1421 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1422 uint32_t val = data; 1423 1424 if (addr + sizeof(val) > vdev->config_len) { 1425 return; 1426 } 1427 1428 stl_le_p(vdev->config + addr, val); 1429 1430 if (k->set_config) { 1431 k->set_config(vdev, vdev->config); 1432 } 1433 } 1434 1435 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr) 1436 { 1437 if (!vdev->vq[n].vring.num) { 1438 return; 1439 } 1440 vdev->vq[n].vring.desc = addr; 1441 virtio_queue_update_rings(vdev, n); 1442 } 1443 1444 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n) 1445 { 1446 return vdev->vq[n].vring.desc; 1447 } 1448 1449 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc, 1450 hwaddr avail, hwaddr used) 1451 { 1452 if (!vdev->vq[n].vring.num) { 1453 return; 1454 } 1455 vdev->vq[n].vring.desc = desc; 1456 vdev->vq[n].vring.avail = avail; 1457 vdev->vq[n].vring.used = used; 1458 virtio_init_region_cache(vdev, n); 1459 } 1460 1461 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num) 1462 { 1463 /* Don't allow guest to flip queue between existent and 1464 * nonexistent states, or to set it to an invalid size. 1465 */ 1466 if (!!num != !!vdev->vq[n].vring.num || 1467 num > VIRTQUEUE_MAX_SIZE || 1468 num < 0) { 1469 return; 1470 } 1471 vdev->vq[n].vring.num = num; 1472 } 1473 1474 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector) 1475 { 1476 return QLIST_FIRST(&vdev->vector_queues[vector]); 1477 } 1478 1479 VirtQueue *virtio_vector_next_queue(VirtQueue *vq) 1480 { 1481 return QLIST_NEXT(vq, node); 1482 } 1483 1484 int virtio_queue_get_num(VirtIODevice *vdev, int n) 1485 { 1486 return vdev->vq[n].vring.num; 1487 } 1488 1489 int virtio_queue_get_max_num(VirtIODevice *vdev, int n) 1490 { 1491 return vdev->vq[n].vring.num_default; 1492 } 1493 1494 int virtio_get_num_queues(VirtIODevice *vdev) 1495 { 1496 int i; 1497 1498 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 1499 if (!virtio_queue_get_num(vdev, i)) { 1500 break; 1501 } 1502 } 1503 1504 return i; 1505 } 1506 1507 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align) 1508 { 1509 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 1510 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 1511 1512 /* virtio-1 compliant devices cannot change the alignment */ 1513 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 1514 error_report("tried to modify queue alignment for virtio-1 device"); 1515 return; 1516 } 1517 /* Check that the transport told us it was going to do this 1518 * (so a buggy transport will immediately assert rather than 1519 * silently failing to migrate this state) 1520 */ 1521 assert(k->has_variable_vring_alignment); 1522 1523 if (align) { 1524 vdev->vq[n].vring.align = align; 1525 virtio_queue_update_rings(vdev, n); 1526 } 1527 } 1528 1529 static bool virtio_queue_notify_aio_vq(VirtQueue *vq) 1530 { 1531 bool ret = false; 1532 1533 if (vq->vring.desc && vq->handle_aio_output) { 1534 VirtIODevice *vdev = vq->vdev; 1535 1536 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); 1537 ret = vq->handle_aio_output(vdev, vq); 1538 1539 if (unlikely(vdev->start_on_kick)) { 1540 virtio_set_started(vdev, true); 1541 } 1542 } 1543 1544 return ret; 1545 } 1546 1547 static void virtio_queue_notify_vq(VirtQueue *vq) 1548 { 1549 if (vq->vring.desc && vq->handle_output) { 1550 VirtIODevice *vdev = vq->vdev; 1551 1552 if (unlikely(vdev->broken)) { 1553 return; 1554 } 1555 1556 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); 1557 vq->handle_output(vdev, vq); 1558 1559 if (unlikely(vdev->start_on_kick)) { 1560 virtio_set_started(vdev, true); 1561 } 1562 } 1563 } 1564 1565 void virtio_queue_notify(VirtIODevice *vdev, int n) 1566 { 1567 VirtQueue *vq = &vdev->vq[n]; 1568 1569 if (unlikely(!vq->vring.desc || vdev->broken)) { 1570 return; 1571 } 1572 1573 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); 1574 if (vq->handle_aio_output) { 1575 event_notifier_set(&vq->host_notifier); 1576 } else if (vq->handle_output) { 1577 vq->handle_output(vdev, vq); 1578 1579 if (unlikely(vdev->start_on_kick)) { 1580 virtio_set_started(vdev, true); 1581 } 1582 } 1583 } 1584 1585 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n) 1586 { 1587 return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector : 1588 VIRTIO_NO_VECTOR; 1589 } 1590 1591 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector) 1592 { 1593 VirtQueue *vq = &vdev->vq[n]; 1594 1595 if (n < VIRTIO_QUEUE_MAX) { 1596 if (vdev->vector_queues && 1597 vdev->vq[n].vector != VIRTIO_NO_VECTOR) { 1598 QLIST_REMOVE(vq, node); 1599 } 1600 vdev->vq[n].vector = vector; 1601 if (vdev->vector_queues && 1602 vector != VIRTIO_NO_VECTOR) { 1603 QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node); 1604 } 1605 } 1606 } 1607 1608 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size, 1609 VirtIOHandleOutput handle_output) 1610 { 1611 int i; 1612 1613 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 1614 if (vdev->vq[i].vring.num == 0) 1615 break; 1616 } 1617 1618 if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE) 1619 abort(); 1620 1621 vdev->vq[i].vring.num = queue_size; 1622 vdev->vq[i].vring.num_default = queue_size; 1623 vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN; 1624 vdev->vq[i].handle_output = handle_output; 1625 vdev->vq[i].handle_aio_output = NULL; 1626 1627 return &vdev->vq[i]; 1628 } 1629 1630 void virtio_del_queue(VirtIODevice *vdev, int n) 1631 { 1632 if (n < 0 || n >= VIRTIO_QUEUE_MAX) { 1633 abort(); 1634 } 1635 1636 vdev->vq[n].vring.num = 0; 1637 vdev->vq[n].vring.num_default = 0; 1638 vdev->vq[n].handle_output = NULL; 1639 vdev->vq[n].handle_aio_output = NULL; 1640 } 1641 1642 static void virtio_set_isr(VirtIODevice *vdev, int value) 1643 { 1644 uint8_t old = atomic_read(&vdev->isr); 1645 1646 /* Do not write ISR if it does not change, so that its cacheline remains 1647 * shared in the common case where the guest does not read it. 1648 */ 1649 if ((old & value) != value) { 1650 atomic_or(&vdev->isr, value); 1651 } 1652 } 1653 1654 /* Called within rcu_read_lock(). */ 1655 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq) 1656 { 1657 uint16_t old, new; 1658 bool v; 1659 /* We need to expose used array entries before checking used event. */ 1660 smp_mb(); 1661 /* Always notify when queue is empty (when feature acknowledge) */ 1662 if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) && 1663 !vq->inuse && virtio_queue_empty(vq)) { 1664 return true; 1665 } 1666 1667 if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { 1668 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT); 1669 } 1670 1671 v = vq->signalled_used_valid; 1672 vq->signalled_used_valid = true; 1673 old = vq->signalled_used; 1674 new = vq->signalled_used = vq->used_idx; 1675 return !v || vring_need_event(vring_get_used_event(vq), new, old); 1676 } 1677 1678 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq) 1679 { 1680 bool should_notify; 1681 rcu_read_lock(); 1682 should_notify = virtio_should_notify(vdev, vq); 1683 rcu_read_unlock(); 1684 1685 if (!should_notify) { 1686 return; 1687 } 1688 1689 trace_virtio_notify_irqfd(vdev, vq); 1690 1691 /* 1692 * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but 1693 * windows drivers included in virtio-win 1.8.0 (circa 2015) are 1694 * incorrectly polling this bit during crashdump and hibernation 1695 * in MSI mode, causing a hang if this bit is never updated. 1696 * Recent releases of Windows do not really shut down, but rather 1697 * log out and hibernate to make the next startup faster. Hence, 1698 * this manifested as a more serious hang during shutdown with 1699 * 1700 * Next driver release from 2016 fixed this problem, so working around it 1701 * is not a must, but it's easy to do so let's do it here. 1702 * 1703 * Note: it's safe to update ISR from any thread as it was switched 1704 * to an atomic operation. 1705 */ 1706 virtio_set_isr(vq->vdev, 0x1); 1707 event_notifier_set(&vq->guest_notifier); 1708 } 1709 1710 static void virtio_irq(VirtQueue *vq) 1711 { 1712 virtio_set_isr(vq->vdev, 0x1); 1713 virtio_notify_vector(vq->vdev, vq->vector); 1714 } 1715 1716 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq) 1717 { 1718 bool should_notify; 1719 rcu_read_lock(); 1720 should_notify = virtio_should_notify(vdev, vq); 1721 rcu_read_unlock(); 1722 1723 if (!should_notify) { 1724 return; 1725 } 1726 1727 trace_virtio_notify(vdev, vq); 1728 virtio_irq(vq); 1729 } 1730 1731 void virtio_notify_config(VirtIODevice *vdev) 1732 { 1733 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) 1734 return; 1735 1736 virtio_set_isr(vdev, 0x3); 1737 vdev->generation++; 1738 virtio_notify_vector(vdev, vdev->config_vector); 1739 } 1740 1741 static bool virtio_device_endian_needed(void *opaque) 1742 { 1743 VirtIODevice *vdev = opaque; 1744 1745 assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN); 1746 if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 1747 return vdev->device_endian != virtio_default_endian(); 1748 } 1749 /* Devices conforming to VIRTIO 1.0 or later are always LE. */ 1750 return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE; 1751 } 1752 1753 static bool virtio_64bit_features_needed(void *opaque) 1754 { 1755 VirtIODevice *vdev = opaque; 1756 1757 return (vdev->host_features >> 32) != 0; 1758 } 1759 1760 static bool virtio_virtqueue_needed(void *opaque) 1761 { 1762 VirtIODevice *vdev = opaque; 1763 1764 return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1); 1765 } 1766 1767 static bool virtio_ringsize_needed(void *opaque) 1768 { 1769 VirtIODevice *vdev = opaque; 1770 int i; 1771 1772 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 1773 if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) { 1774 return true; 1775 } 1776 } 1777 return false; 1778 } 1779 1780 static bool virtio_extra_state_needed(void *opaque) 1781 { 1782 VirtIODevice *vdev = opaque; 1783 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 1784 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 1785 1786 return k->has_extra_state && 1787 k->has_extra_state(qbus->parent); 1788 } 1789 1790 static bool virtio_broken_needed(void *opaque) 1791 { 1792 VirtIODevice *vdev = opaque; 1793 1794 return vdev->broken; 1795 } 1796 1797 static bool virtio_started_needed(void *opaque) 1798 { 1799 VirtIODevice *vdev = opaque; 1800 1801 return vdev->started; 1802 } 1803 1804 static const VMStateDescription vmstate_virtqueue = { 1805 .name = "virtqueue_state", 1806 .version_id = 1, 1807 .minimum_version_id = 1, 1808 .fields = (VMStateField[]) { 1809 VMSTATE_UINT64(vring.avail, struct VirtQueue), 1810 VMSTATE_UINT64(vring.used, struct VirtQueue), 1811 VMSTATE_END_OF_LIST() 1812 } 1813 }; 1814 1815 static const VMStateDescription vmstate_virtio_virtqueues = { 1816 .name = "virtio/virtqueues", 1817 .version_id = 1, 1818 .minimum_version_id = 1, 1819 .needed = &virtio_virtqueue_needed, 1820 .fields = (VMStateField[]) { 1821 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice, 1822 VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue), 1823 VMSTATE_END_OF_LIST() 1824 } 1825 }; 1826 1827 static const VMStateDescription vmstate_ringsize = { 1828 .name = "ringsize_state", 1829 .version_id = 1, 1830 .minimum_version_id = 1, 1831 .fields = (VMStateField[]) { 1832 VMSTATE_UINT32(vring.num_default, struct VirtQueue), 1833 VMSTATE_END_OF_LIST() 1834 } 1835 }; 1836 1837 static const VMStateDescription vmstate_virtio_ringsize = { 1838 .name = "virtio/ringsize", 1839 .version_id = 1, 1840 .minimum_version_id = 1, 1841 .needed = &virtio_ringsize_needed, 1842 .fields = (VMStateField[]) { 1843 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice, 1844 VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue), 1845 VMSTATE_END_OF_LIST() 1846 } 1847 }; 1848 1849 static int get_extra_state(QEMUFile *f, void *pv, size_t size, 1850 const VMStateField *field) 1851 { 1852 VirtIODevice *vdev = pv; 1853 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 1854 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 1855 1856 if (!k->load_extra_state) { 1857 return -1; 1858 } else { 1859 return k->load_extra_state(qbus->parent, f); 1860 } 1861 } 1862 1863 static int put_extra_state(QEMUFile *f, void *pv, size_t size, 1864 const VMStateField *field, QJSON *vmdesc) 1865 { 1866 VirtIODevice *vdev = pv; 1867 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 1868 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 1869 1870 k->save_extra_state(qbus->parent, f); 1871 return 0; 1872 } 1873 1874 static const VMStateInfo vmstate_info_extra_state = { 1875 .name = "virtqueue_extra_state", 1876 .get = get_extra_state, 1877 .put = put_extra_state, 1878 }; 1879 1880 static const VMStateDescription vmstate_virtio_extra_state = { 1881 .name = "virtio/extra_state", 1882 .version_id = 1, 1883 .minimum_version_id = 1, 1884 .needed = &virtio_extra_state_needed, 1885 .fields = (VMStateField[]) { 1886 { 1887 .name = "extra_state", 1888 .version_id = 0, 1889 .field_exists = NULL, 1890 .size = 0, 1891 .info = &vmstate_info_extra_state, 1892 .flags = VMS_SINGLE, 1893 .offset = 0, 1894 }, 1895 VMSTATE_END_OF_LIST() 1896 } 1897 }; 1898 1899 static const VMStateDescription vmstate_virtio_device_endian = { 1900 .name = "virtio/device_endian", 1901 .version_id = 1, 1902 .minimum_version_id = 1, 1903 .needed = &virtio_device_endian_needed, 1904 .fields = (VMStateField[]) { 1905 VMSTATE_UINT8(device_endian, VirtIODevice), 1906 VMSTATE_END_OF_LIST() 1907 } 1908 }; 1909 1910 static const VMStateDescription vmstate_virtio_64bit_features = { 1911 .name = "virtio/64bit_features", 1912 .version_id = 1, 1913 .minimum_version_id = 1, 1914 .needed = &virtio_64bit_features_needed, 1915 .fields = (VMStateField[]) { 1916 VMSTATE_UINT64(guest_features, VirtIODevice), 1917 VMSTATE_END_OF_LIST() 1918 } 1919 }; 1920 1921 static const VMStateDescription vmstate_virtio_broken = { 1922 .name = "virtio/broken", 1923 .version_id = 1, 1924 .minimum_version_id = 1, 1925 .needed = &virtio_broken_needed, 1926 .fields = (VMStateField[]) { 1927 VMSTATE_BOOL(broken, VirtIODevice), 1928 VMSTATE_END_OF_LIST() 1929 } 1930 }; 1931 1932 static const VMStateDescription vmstate_virtio_started = { 1933 .name = "virtio/started", 1934 .version_id = 1, 1935 .minimum_version_id = 1, 1936 .needed = &virtio_started_needed, 1937 .fields = (VMStateField[]) { 1938 VMSTATE_BOOL(started, VirtIODevice), 1939 VMSTATE_END_OF_LIST() 1940 } 1941 }; 1942 1943 static const VMStateDescription vmstate_virtio = { 1944 .name = "virtio", 1945 .version_id = 1, 1946 .minimum_version_id = 1, 1947 .minimum_version_id_old = 1, 1948 .fields = (VMStateField[]) { 1949 VMSTATE_END_OF_LIST() 1950 }, 1951 .subsections = (const VMStateDescription*[]) { 1952 &vmstate_virtio_device_endian, 1953 &vmstate_virtio_64bit_features, 1954 &vmstate_virtio_virtqueues, 1955 &vmstate_virtio_ringsize, 1956 &vmstate_virtio_broken, 1957 &vmstate_virtio_extra_state, 1958 &vmstate_virtio_started, 1959 NULL 1960 } 1961 }; 1962 1963 int virtio_save(VirtIODevice *vdev, QEMUFile *f) 1964 { 1965 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 1966 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 1967 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 1968 uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff); 1969 int i; 1970 1971 if (k->save_config) { 1972 k->save_config(qbus->parent, f); 1973 } 1974 1975 qemu_put_8s(f, &vdev->status); 1976 qemu_put_8s(f, &vdev->isr); 1977 qemu_put_be16s(f, &vdev->queue_sel); 1978 qemu_put_be32s(f, &guest_features_lo); 1979 qemu_put_be32(f, vdev->config_len); 1980 qemu_put_buffer(f, vdev->config, vdev->config_len); 1981 1982 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 1983 if (vdev->vq[i].vring.num == 0) 1984 break; 1985 } 1986 1987 qemu_put_be32(f, i); 1988 1989 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 1990 if (vdev->vq[i].vring.num == 0) 1991 break; 1992 1993 qemu_put_be32(f, vdev->vq[i].vring.num); 1994 if (k->has_variable_vring_alignment) { 1995 qemu_put_be32(f, vdev->vq[i].vring.align); 1996 } 1997 /* 1998 * Save desc now, the rest of the ring addresses are saved in 1999 * subsections for VIRTIO-1 devices. 2000 */ 2001 qemu_put_be64(f, vdev->vq[i].vring.desc); 2002 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx); 2003 if (k->save_queue) { 2004 k->save_queue(qbus->parent, i, f); 2005 } 2006 } 2007 2008 if (vdc->save != NULL) { 2009 vdc->save(vdev, f); 2010 } 2011 2012 if (vdc->vmsd) { 2013 int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL); 2014 if (ret) { 2015 return ret; 2016 } 2017 } 2018 2019 /* Subsections */ 2020 return vmstate_save_state(f, &vmstate_virtio, vdev, NULL); 2021 } 2022 2023 /* A wrapper for use as a VMState .put function */ 2024 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size, 2025 const VMStateField *field, QJSON *vmdesc) 2026 { 2027 return virtio_save(VIRTIO_DEVICE(opaque), f); 2028 } 2029 2030 /* A wrapper for use as a VMState .get function */ 2031 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size, 2032 const VMStateField *field) 2033 { 2034 VirtIODevice *vdev = VIRTIO_DEVICE(opaque); 2035 DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev)); 2036 2037 return virtio_load(vdev, f, dc->vmsd->version_id); 2038 } 2039 2040 const VMStateInfo virtio_vmstate_info = { 2041 .name = "virtio", 2042 .get = virtio_device_get, 2043 .put = virtio_device_put, 2044 }; 2045 2046 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val) 2047 { 2048 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2049 bool bad = (val & ~(vdev->host_features)) != 0; 2050 2051 val &= vdev->host_features; 2052 if (k->set_features) { 2053 k->set_features(vdev, val); 2054 } 2055 vdev->guest_features = val; 2056 return bad ? -1 : 0; 2057 } 2058 2059 int virtio_set_features(VirtIODevice *vdev, uint64_t val) 2060 { 2061 int ret; 2062 /* 2063 * The driver must not attempt to set features after feature negotiation 2064 * has finished. 2065 */ 2066 if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) { 2067 return -EINVAL; 2068 } 2069 ret = virtio_set_features_nocheck(vdev, val); 2070 if (!ret) { 2071 if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { 2072 /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches. */ 2073 int i; 2074 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2075 if (vdev->vq[i].vring.num != 0) { 2076 virtio_init_region_cache(vdev, i); 2077 } 2078 } 2079 } 2080 2081 if (!virtio_device_started(vdev, vdev->status) && 2082 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 2083 vdev->start_on_kick = true; 2084 } 2085 } 2086 return ret; 2087 } 2088 2089 size_t virtio_feature_get_config_size(VirtIOFeature *feature_sizes, 2090 uint64_t host_features) 2091 { 2092 size_t config_size = 0; 2093 int i; 2094 2095 for (i = 0; feature_sizes[i].flags != 0; i++) { 2096 if (host_features & feature_sizes[i].flags) { 2097 config_size = MAX(feature_sizes[i].end, config_size); 2098 } 2099 } 2100 2101 return config_size; 2102 } 2103 2104 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) 2105 { 2106 int i, ret; 2107 int32_t config_len; 2108 uint32_t num; 2109 uint32_t features; 2110 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2111 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2112 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 2113 2114 /* 2115 * We poison the endianness to ensure it does not get used before 2116 * subsections have been loaded. 2117 */ 2118 vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN; 2119 2120 if (k->load_config) { 2121 ret = k->load_config(qbus->parent, f); 2122 if (ret) 2123 return ret; 2124 } 2125 2126 qemu_get_8s(f, &vdev->status); 2127 qemu_get_8s(f, &vdev->isr); 2128 qemu_get_be16s(f, &vdev->queue_sel); 2129 if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) { 2130 return -1; 2131 } 2132 qemu_get_be32s(f, &features); 2133 2134 /* 2135 * Temporarily set guest_features low bits - needed by 2136 * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS 2137 * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ. 2138 * 2139 * Note: devices should always test host features in future - don't create 2140 * new dependencies like this. 2141 */ 2142 vdev->guest_features = features; 2143 2144 config_len = qemu_get_be32(f); 2145 2146 /* 2147 * There are cases where the incoming config can be bigger or smaller 2148 * than what we have; so load what we have space for, and skip 2149 * any excess that's in the stream. 2150 */ 2151 qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len)); 2152 2153 while (config_len > vdev->config_len) { 2154 qemu_get_byte(f); 2155 config_len--; 2156 } 2157 2158 num = qemu_get_be32(f); 2159 2160 if (num > VIRTIO_QUEUE_MAX) { 2161 error_report("Invalid number of virtqueues: 0x%x", num); 2162 return -1; 2163 } 2164 2165 for (i = 0; i < num; i++) { 2166 vdev->vq[i].vring.num = qemu_get_be32(f); 2167 if (k->has_variable_vring_alignment) { 2168 vdev->vq[i].vring.align = qemu_get_be32(f); 2169 } 2170 vdev->vq[i].vring.desc = qemu_get_be64(f); 2171 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx); 2172 vdev->vq[i].signalled_used_valid = false; 2173 vdev->vq[i].notification = true; 2174 2175 if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) { 2176 error_report("VQ %d address 0x0 " 2177 "inconsistent with Host index 0x%x", 2178 i, vdev->vq[i].last_avail_idx); 2179 return -1; 2180 } 2181 if (k->load_queue) { 2182 ret = k->load_queue(qbus->parent, i, f); 2183 if (ret) 2184 return ret; 2185 } 2186 } 2187 2188 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR); 2189 2190 if (vdc->load != NULL) { 2191 ret = vdc->load(vdev, f, version_id); 2192 if (ret) { 2193 return ret; 2194 } 2195 } 2196 2197 if (vdc->vmsd) { 2198 ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id); 2199 if (ret) { 2200 return ret; 2201 } 2202 } 2203 2204 /* Subsections */ 2205 ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1); 2206 if (ret) { 2207 return ret; 2208 } 2209 2210 if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) { 2211 vdev->device_endian = virtio_default_endian(); 2212 } 2213 2214 if (virtio_64bit_features_needed(vdev)) { 2215 /* 2216 * Subsection load filled vdev->guest_features. Run them 2217 * through virtio_set_features to sanity-check them against 2218 * host_features. 2219 */ 2220 uint64_t features64 = vdev->guest_features; 2221 if (virtio_set_features_nocheck(vdev, features64) < 0) { 2222 error_report("Features 0x%" PRIx64 " unsupported. " 2223 "Allowed features: 0x%" PRIx64, 2224 features64, vdev->host_features); 2225 return -1; 2226 } 2227 } else { 2228 if (virtio_set_features_nocheck(vdev, features) < 0) { 2229 error_report("Features 0x%x unsupported. " 2230 "Allowed features: 0x%" PRIx64, 2231 features, vdev->host_features); 2232 return -1; 2233 } 2234 } 2235 2236 if (!virtio_device_started(vdev, vdev->status) && 2237 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 2238 vdev->start_on_kick = true; 2239 } 2240 2241 rcu_read_lock(); 2242 for (i = 0; i < num; i++) { 2243 if (vdev->vq[i].vring.desc) { 2244 uint16_t nheads; 2245 2246 /* 2247 * VIRTIO-1 devices migrate desc, used, and avail ring addresses so 2248 * only the region cache needs to be set up. Legacy devices need 2249 * to calculate used and avail ring addresses based on the desc 2250 * address. 2251 */ 2252 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 2253 virtio_init_region_cache(vdev, i); 2254 } else { 2255 virtio_queue_update_rings(vdev, i); 2256 } 2257 2258 nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx; 2259 /* Check it isn't doing strange things with descriptor numbers. */ 2260 if (nheads > vdev->vq[i].vring.num) { 2261 error_report("VQ %d size 0x%x Guest index 0x%x " 2262 "inconsistent with Host index 0x%x: delta 0x%x", 2263 i, vdev->vq[i].vring.num, 2264 vring_avail_idx(&vdev->vq[i]), 2265 vdev->vq[i].last_avail_idx, nheads); 2266 return -1; 2267 } 2268 vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]); 2269 vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]); 2270 2271 /* 2272 * Some devices migrate VirtQueueElements that have been popped 2273 * from the avail ring but not yet returned to the used ring. 2274 * Since max ring size < UINT16_MAX it's safe to use modulo 2275 * UINT16_MAX + 1 subtraction. 2276 */ 2277 vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx - 2278 vdev->vq[i].used_idx); 2279 if (vdev->vq[i].inuse > vdev->vq[i].vring.num) { 2280 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - " 2281 "used_idx 0x%x", 2282 i, vdev->vq[i].vring.num, 2283 vdev->vq[i].last_avail_idx, 2284 vdev->vq[i].used_idx); 2285 return -1; 2286 } 2287 } 2288 } 2289 rcu_read_unlock(); 2290 2291 return 0; 2292 } 2293 2294 void virtio_cleanup(VirtIODevice *vdev) 2295 { 2296 qemu_del_vm_change_state_handler(vdev->vmstate); 2297 } 2298 2299 static void virtio_vmstate_change(void *opaque, int running, RunState state) 2300 { 2301 VirtIODevice *vdev = opaque; 2302 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2303 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2304 bool backend_run = running && virtio_device_started(vdev, vdev->status); 2305 vdev->vm_running = running; 2306 2307 if (backend_run) { 2308 virtio_set_status(vdev, vdev->status); 2309 } 2310 2311 if (k->vmstate_change) { 2312 k->vmstate_change(qbus->parent, backend_run); 2313 } 2314 2315 if (!backend_run) { 2316 virtio_set_status(vdev, vdev->status); 2317 } 2318 } 2319 2320 void virtio_instance_init_common(Object *proxy_obj, void *data, 2321 size_t vdev_size, const char *vdev_name) 2322 { 2323 DeviceState *vdev = data; 2324 2325 object_initialize_child(proxy_obj, "virtio-backend", vdev, vdev_size, 2326 vdev_name, &error_abort, NULL); 2327 qdev_alias_all_properties(vdev, proxy_obj); 2328 } 2329 2330 void virtio_init(VirtIODevice *vdev, const char *name, 2331 uint16_t device_id, size_t config_size) 2332 { 2333 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2334 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2335 int i; 2336 int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0; 2337 2338 if (nvectors) { 2339 vdev->vector_queues = 2340 g_malloc0(sizeof(*vdev->vector_queues) * nvectors); 2341 } 2342 2343 vdev->start_on_kick = false; 2344 vdev->started = false; 2345 vdev->device_id = device_id; 2346 vdev->status = 0; 2347 atomic_set(&vdev->isr, 0); 2348 vdev->queue_sel = 0; 2349 vdev->config_vector = VIRTIO_NO_VECTOR; 2350 vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX); 2351 vdev->vm_running = runstate_is_running(); 2352 vdev->broken = false; 2353 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2354 vdev->vq[i].vector = VIRTIO_NO_VECTOR; 2355 vdev->vq[i].vdev = vdev; 2356 vdev->vq[i].queue_index = i; 2357 } 2358 2359 vdev->name = name; 2360 vdev->config_len = config_size; 2361 if (vdev->config_len) { 2362 vdev->config = g_malloc0(config_size); 2363 } else { 2364 vdev->config = NULL; 2365 } 2366 vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev), 2367 virtio_vmstate_change, vdev); 2368 vdev->device_endian = virtio_default_endian(); 2369 vdev->use_guest_notifier_mask = true; 2370 } 2371 2372 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n) 2373 { 2374 return vdev->vq[n].vring.desc; 2375 } 2376 2377 bool virtio_queue_enabled(VirtIODevice *vdev, int n) 2378 { 2379 return virtio_queue_get_desc_addr(vdev, n) != 0; 2380 } 2381 2382 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n) 2383 { 2384 return vdev->vq[n].vring.avail; 2385 } 2386 2387 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n) 2388 { 2389 return vdev->vq[n].vring.used; 2390 } 2391 2392 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n) 2393 { 2394 return sizeof(VRingDesc) * vdev->vq[n].vring.num; 2395 } 2396 2397 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n) 2398 { 2399 return offsetof(VRingAvail, ring) + 2400 sizeof(uint16_t) * vdev->vq[n].vring.num; 2401 } 2402 2403 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n) 2404 { 2405 return offsetof(VRingUsed, ring) + 2406 sizeof(VRingUsedElem) * vdev->vq[n].vring.num; 2407 } 2408 2409 uint16_t virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n) 2410 { 2411 return vdev->vq[n].last_avail_idx; 2412 } 2413 2414 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, uint16_t idx) 2415 { 2416 vdev->vq[n].last_avail_idx = idx; 2417 vdev->vq[n].shadow_avail_idx = idx; 2418 } 2419 2420 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n) 2421 { 2422 rcu_read_lock(); 2423 if (vdev->vq[n].vring.desc) { 2424 vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]); 2425 vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx; 2426 } 2427 rcu_read_unlock(); 2428 } 2429 2430 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n) 2431 { 2432 rcu_read_lock(); 2433 if (vdev->vq[n].vring.desc) { 2434 vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]); 2435 } 2436 rcu_read_unlock(); 2437 } 2438 2439 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n) 2440 { 2441 vdev->vq[n].signalled_used_valid = false; 2442 } 2443 2444 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n) 2445 { 2446 return vdev->vq + n; 2447 } 2448 2449 uint16_t virtio_get_queue_index(VirtQueue *vq) 2450 { 2451 return vq->queue_index; 2452 } 2453 2454 static void virtio_queue_guest_notifier_read(EventNotifier *n) 2455 { 2456 VirtQueue *vq = container_of(n, VirtQueue, guest_notifier); 2457 if (event_notifier_test_and_clear(n)) { 2458 virtio_irq(vq); 2459 } 2460 } 2461 2462 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign, 2463 bool with_irqfd) 2464 { 2465 if (assign && !with_irqfd) { 2466 event_notifier_set_handler(&vq->guest_notifier, 2467 virtio_queue_guest_notifier_read); 2468 } else { 2469 event_notifier_set_handler(&vq->guest_notifier, NULL); 2470 } 2471 if (!assign) { 2472 /* Test and clear notifier before closing it, 2473 * in case poll callback didn't have time to run. */ 2474 virtio_queue_guest_notifier_read(&vq->guest_notifier); 2475 } 2476 } 2477 2478 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq) 2479 { 2480 return &vq->guest_notifier; 2481 } 2482 2483 static void virtio_queue_host_notifier_aio_read(EventNotifier *n) 2484 { 2485 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 2486 if (event_notifier_test_and_clear(n)) { 2487 virtio_queue_notify_aio_vq(vq); 2488 } 2489 } 2490 2491 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n) 2492 { 2493 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 2494 2495 virtio_queue_set_notification(vq, 0); 2496 } 2497 2498 static bool virtio_queue_host_notifier_aio_poll(void *opaque) 2499 { 2500 EventNotifier *n = opaque; 2501 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 2502 bool progress; 2503 2504 if (!vq->vring.desc || virtio_queue_empty(vq)) { 2505 return false; 2506 } 2507 2508 progress = virtio_queue_notify_aio_vq(vq); 2509 2510 /* In case the handler function re-enabled notifications */ 2511 virtio_queue_set_notification(vq, 0); 2512 return progress; 2513 } 2514 2515 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n) 2516 { 2517 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 2518 2519 /* Caller polls once more after this to catch requests that race with us */ 2520 virtio_queue_set_notification(vq, 1); 2521 } 2522 2523 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx, 2524 VirtIOHandleAIOOutput handle_output) 2525 { 2526 if (handle_output) { 2527 vq->handle_aio_output = handle_output; 2528 aio_set_event_notifier(ctx, &vq->host_notifier, true, 2529 virtio_queue_host_notifier_aio_read, 2530 virtio_queue_host_notifier_aio_poll); 2531 aio_set_event_notifier_poll(ctx, &vq->host_notifier, 2532 virtio_queue_host_notifier_aio_poll_begin, 2533 virtio_queue_host_notifier_aio_poll_end); 2534 } else { 2535 aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL); 2536 /* Test and clear notifier before after disabling event, 2537 * in case poll callback didn't have time to run. */ 2538 virtio_queue_host_notifier_aio_read(&vq->host_notifier); 2539 vq->handle_aio_output = NULL; 2540 } 2541 } 2542 2543 void virtio_queue_host_notifier_read(EventNotifier *n) 2544 { 2545 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 2546 if (event_notifier_test_and_clear(n)) { 2547 virtio_queue_notify_vq(vq); 2548 } 2549 } 2550 2551 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq) 2552 { 2553 return &vq->host_notifier; 2554 } 2555 2556 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n, 2557 MemoryRegion *mr, bool assign) 2558 { 2559 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2560 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2561 2562 if (k->set_host_notifier_mr) { 2563 return k->set_host_notifier_mr(qbus->parent, n, mr, assign); 2564 } 2565 2566 return -1; 2567 } 2568 2569 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name) 2570 { 2571 g_free(vdev->bus_name); 2572 vdev->bus_name = g_strdup(bus_name); 2573 } 2574 2575 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...) 2576 { 2577 va_list ap; 2578 2579 va_start(ap, fmt); 2580 error_vreport(fmt, ap); 2581 va_end(ap); 2582 2583 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 2584 vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET; 2585 virtio_notify_config(vdev); 2586 } 2587 2588 vdev->broken = true; 2589 } 2590 2591 static void virtio_memory_listener_commit(MemoryListener *listener) 2592 { 2593 VirtIODevice *vdev = container_of(listener, VirtIODevice, listener); 2594 int i; 2595 2596 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2597 if (vdev->vq[i].vring.num == 0) { 2598 break; 2599 } 2600 virtio_init_region_cache(vdev, i); 2601 } 2602 } 2603 2604 static void virtio_device_realize(DeviceState *dev, Error **errp) 2605 { 2606 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 2607 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 2608 Error *err = NULL; 2609 2610 /* Devices should either use vmsd or the load/save methods */ 2611 assert(!vdc->vmsd || !vdc->load); 2612 2613 if (vdc->realize != NULL) { 2614 vdc->realize(dev, &err); 2615 if (err != NULL) { 2616 error_propagate(errp, err); 2617 return; 2618 } 2619 } 2620 2621 virtio_bus_device_plugged(vdev, &err); 2622 if (err != NULL) { 2623 error_propagate(errp, err); 2624 vdc->unrealize(dev, NULL); 2625 return; 2626 } 2627 2628 vdev->listener.commit = virtio_memory_listener_commit; 2629 memory_listener_register(&vdev->listener, vdev->dma_as); 2630 } 2631 2632 static void virtio_device_unrealize(DeviceState *dev, Error **errp) 2633 { 2634 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 2635 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 2636 Error *err = NULL; 2637 2638 virtio_bus_device_unplugged(vdev); 2639 2640 if (vdc->unrealize != NULL) { 2641 vdc->unrealize(dev, &err); 2642 if (err != NULL) { 2643 error_propagate(errp, err); 2644 return; 2645 } 2646 } 2647 2648 g_free(vdev->bus_name); 2649 vdev->bus_name = NULL; 2650 } 2651 2652 static void virtio_device_free_virtqueues(VirtIODevice *vdev) 2653 { 2654 int i; 2655 if (!vdev->vq) { 2656 return; 2657 } 2658 2659 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2660 if (vdev->vq[i].vring.num == 0) { 2661 break; 2662 } 2663 virtio_virtqueue_reset_region_cache(&vdev->vq[i]); 2664 } 2665 g_free(vdev->vq); 2666 } 2667 2668 static void virtio_device_instance_finalize(Object *obj) 2669 { 2670 VirtIODevice *vdev = VIRTIO_DEVICE(obj); 2671 2672 memory_listener_unregister(&vdev->listener); 2673 virtio_device_free_virtqueues(vdev); 2674 2675 g_free(vdev->config); 2676 g_free(vdev->vector_queues); 2677 } 2678 2679 static Property virtio_properties[] = { 2680 DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features), 2681 DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true), 2682 DEFINE_PROP_END_OF_LIST(), 2683 }; 2684 2685 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev) 2686 { 2687 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev))); 2688 int i, n, r, err; 2689 2690 memory_region_transaction_begin(); 2691 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 2692 VirtQueue *vq = &vdev->vq[n]; 2693 if (!virtio_queue_get_num(vdev, n)) { 2694 continue; 2695 } 2696 r = virtio_bus_set_host_notifier(qbus, n, true); 2697 if (r < 0) { 2698 err = r; 2699 goto assign_error; 2700 } 2701 event_notifier_set_handler(&vq->host_notifier, 2702 virtio_queue_host_notifier_read); 2703 } 2704 2705 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 2706 /* Kick right away to begin processing requests already in vring */ 2707 VirtQueue *vq = &vdev->vq[n]; 2708 if (!vq->vring.num) { 2709 continue; 2710 } 2711 event_notifier_set(&vq->host_notifier); 2712 } 2713 memory_region_transaction_commit(); 2714 return 0; 2715 2716 assign_error: 2717 i = n; /* save n for a second iteration after transaction is committed. */ 2718 while (--n >= 0) { 2719 VirtQueue *vq = &vdev->vq[n]; 2720 if (!virtio_queue_get_num(vdev, n)) { 2721 continue; 2722 } 2723 2724 event_notifier_set_handler(&vq->host_notifier, NULL); 2725 r = virtio_bus_set_host_notifier(qbus, n, false); 2726 assert(r >= 0); 2727 } 2728 memory_region_transaction_commit(); 2729 2730 while (--i >= 0) { 2731 if (!virtio_queue_get_num(vdev, i)) { 2732 continue; 2733 } 2734 virtio_bus_cleanup_host_notifier(qbus, i); 2735 } 2736 return err; 2737 } 2738 2739 int virtio_device_start_ioeventfd(VirtIODevice *vdev) 2740 { 2741 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2742 VirtioBusState *vbus = VIRTIO_BUS(qbus); 2743 2744 return virtio_bus_start_ioeventfd(vbus); 2745 } 2746 2747 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev) 2748 { 2749 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev))); 2750 int n, r; 2751 2752 memory_region_transaction_begin(); 2753 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 2754 VirtQueue *vq = &vdev->vq[n]; 2755 2756 if (!virtio_queue_get_num(vdev, n)) { 2757 continue; 2758 } 2759 event_notifier_set_handler(&vq->host_notifier, NULL); 2760 r = virtio_bus_set_host_notifier(qbus, n, false); 2761 assert(r >= 0); 2762 } 2763 memory_region_transaction_commit(); 2764 2765 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 2766 if (!virtio_queue_get_num(vdev, n)) { 2767 continue; 2768 } 2769 virtio_bus_cleanup_host_notifier(qbus, n); 2770 } 2771 } 2772 2773 void virtio_device_stop_ioeventfd(VirtIODevice *vdev) 2774 { 2775 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2776 VirtioBusState *vbus = VIRTIO_BUS(qbus); 2777 2778 virtio_bus_stop_ioeventfd(vbus); 2779 } 2780 2781 int virtio_device_grab_ioeventfd(VirtIODevice *vdev) 2782 { 2783 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2784 VirtioBusState *vbus = VIRTIO_BUS(qbus); 2785 2786 return virtio_bus_grab_ioeventfd(vbus); 2787 } 2788 2789 void virtio_device_release_ioeventfd(VirtIODevice *vdev) 2790 { 2791 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2792 VirtioBusState *vbus = VIRTIO_BUS(qbus); 2793 2794 virtio_bus_release_ioeventfd(vbus); 2795 } 2796 2797 static void virtio_device_class_init(ObjectClass *klass, void *data) 2798 { 2799 /* Set the default value here. */ 2800 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 2801 DeviceClass *dc = DEVICE_CLASS(klass); 2802 2803 dc->realize = virtio_device_realize; 2804 dc->unrealize = virtio_device_unrealize; 2805 dc->bus_type = TYPE_VIRTIO_BUS; 2806 dc->props = virtio_properties; 2807 vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl; 2808 vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl; 2809 2810 vdc->legacy_features |= VIRTIO_LEGACY_FEATURES; 2811 } 2812 2813 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev) 2814 { 2815 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2816 VirtioBusState *vbus = VIRTIO_BUS(qbus); 2817 2818 return virtio_bus_ioeventfd_enabled(vbus); 2819 } 2820 2821 static const TypeInfo virtio_device_info = { 2822 .name = TYPE_VIRTIO_DEVICE, 2823 .parent = TYPE_DEVICE, 2824 .instance_size = sizeof(VirtIODevice), 2825 .class_init = virtio_device_class_init, 2826 .instance_finalize = virtio_device_instance_finalize, 2827 .abstract = true, 2828 .class_size = sizeof(VirtioDeviceClass), 2829 }; 2830 2831 static void virtio_register_types(void) 2832 { 2833 type_register_static(&virtio_device_info); 2834 } 2835 2836 type_init(virtio_register_types) 2837