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 "qemu/error-report.h" 19 #include "qemu/log.h" 20 #include "qemu/main-loop.h" 21 #include "qemu/module.h" 22 #include "hw/virtio/virtio.h" 23 #include "migration/qemu-file-types.h" 24 #include "qemu/atomic.h" 25 #include "hw/virtio/virtio-bus.h" 26 #include "hw/qdev-properties.h" 27 #include "hw/virtio/virtio-access.h" 28 #include "sysemu/dma.h" 29 #include "sysemu/runstate.h" 30 #include "standard-headers/linux/virtio_ids.h" 31 32 /* 33 * The alignment to use between consumer and producer parts of vring. 34 * x86 pagesize again. This is the default, used by transports like PCI 35 * which don't provide a means for the guest to tell the host the alignment. 36 */ 37 #define VIRTIO_PCI_VRING_ALIGN 4096 38 39 typedef struct VRingDesc 40 { 41 uint64_t addr; 42 uint32_t len; 43 uint16_t flags; 44 uint16_t next; 45 } VRingDesc; 46 47 typedef struct VRingPackedDesc { 48 uint64_t addr; 49 uint32_t len; 50 uint16_t id; 51 uint16_t flags; 52 } VRingPackedDesc; 53 54 typedef struct VRingAvail 55 { 56 uint16_t flags; 57 uint16_t idx; 58 uint16_t ring[]; 59 } VRingAvail; 60 61 typedef struct VRingUsedElem 62 { 63 uint32_t id; 64 uint32_t len; 65 } VRingUsedElem; 66 67 typedef struct VRingUsed 68 { 69 uint16_t flags; 70 uint16_t idx; 71 VRingUsedElem ring[]; 72 } VRingUsed; 73 74 typedef struct VRingMemoryRegionCaches { 75 struct rcu_head rcu; 76 MemoryRegionCache desc; 77 MemoryRegionCache avail; 78 MemoryRegionCache used; 79 } VRingMemoryRegionCaches; 80 81 typedef struct VRing 82 { 83 unsigned int num; 84 unsigned int num_default; 85 unsigned int align; 86 hwaddr desc; 87 hwaddr avail; 88 hwaddr used; 89 VRingMemoryRegionCaches *caches; 90 } VRing; 91 92 typedef struct VRingPackedDescEvent { 93 uint16_t off_wrap; 94 uint16_t flags; 95 } VRingPackedDescEvent ; 96 97 struct VirtQueue 98 { 99 VRing vring; 100 VirtQueueElement *used_elems; 101 102 /* Next head to pop */ 103 uint16_t last_avail_idx; 104 bool last_avail_wrap_counter; 105 106 /* Last avail_idx read from VQ. */ 107 uint16_t shadow_avail_idx; 108 bool shadow_avail_wrap_counter; 109 110 uint16_t used_idx; 111 bool used_wrap_counter; 112 113 /* Last used index value we have signalled on */ 114 uint16_t signalled_used; 115 116 /* Last used index value we have signalled on */ 117 bool signalled_used_valid; 118 119 /* Notification enabled? */ 120 bool notification; 121 122 uint16_t queue_index; 123 124 unsigned int inuse; 125 126 uint16_t vector; 127 VirtIOHandleOutput handle_output; 128 VirtIOHandleOutput handle_aio_output; 129 VirtIODevice *vdev; 130 EventNotifier guest_notifier; 131 EventNotifier host_notifier; 132 bool host_notifier_enabled; 133 QLIST_ENTRY(VirtQueue) node; 134 }; 135 136 /* Called within call_rcu(). */ 137 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches) 138 { 139 assert(caches != NULL); 140 address_space_cache_destroy(&caches->desc); 141 address_space_cache_destroy(&caches->avail); 142 address_space_cache_destroy(&caches->used); 143 g_free(caches); 144 } 145 146 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq) 147 { 148 VRingMemoryRegionCaches *caches; 149 150 caches = qatomic_read(&vq->vring.caches); 151 qatomic_rcu_set(&vq->vring.caches, NULL); 152 if (caches) { 153 call_rcu(caches, virtio_free_region_cache, rcu); 154 } 155 } 156 157 static void virtio_init_region_cache(VirtIODevice *vdev, int n) 158 { 159 VirtQueue *vq = &vdev->vq[n]; 160 VRingMemoryRegionCaches *old = vq->vring.caches; 161 VRingMemoryRegionCaches *new = NULL; 162 hwaddr addr, size; 163 int64_t len; 164 bool packed; 165 166 167 addr = vq->vring.desc; 168 if (!addr) { 169 goto out_no_cache; 170 } 171 new = g_new0(VRingMemoryRegionCaches, 1); 172 size = virtio_queue_get_desc_size(vdev, n); 173 packed = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ? 174 true : false; 175 len = address_space_cache_init(&new->desc, vdev->dma_as, 176 addr, size, packed); 177 if (len < size) { 178 virtio_error(vdev, "Cannot map desc"); 179 goto err_desc; 180 } 181 182 size = virtio_queue_get_used_size(vdev, n); 183 len = address_space_cache_init(&new->used, vdev->dma_as, 184 vq->vring.used, size, true); 185 if (len < size) { 186 virtio_error(vdev, "Cannot map used"); 187 goto err_used; 188 } 189 190 size = virtio_queue_get_avail_size(vdev, n); 191 len = address_space_cache_init(&new->avail, vdev->dma_as, 192 vq->vring.avail, size, false); 193 if (len < size) { 194 virtio_error(vdev, "Cannot map avail"); 195 goto err_avail; 196 } 197 198 qatomic_rcu_set(&vq->vring.caches, new); 199 if (old) { 200 call_rcu(old, virtio_free_region_cache, rcu); 201 } 202 return; 203 204 err_avail: 205 address_space_cache_destroy(&new->avail); 206 err_used: 207 address_space_cache_destroy(&new->used); 208 err_desc: 209 address_space_cache_destroy(&new->desc); 210 out_no_cache: 211 g_free(new); 212 virtio_virtqueue_reset_region_cache(vq); 213 } 214 215 /* virt queue functions */ 216 void virtio_queue_update_rings(VirtIODevice *vdev, int n) 217 { 218 VRing *vring = &vdev->vq[n].vring; 219 220 if (!vring->num || !vring->desc || !vring->align) { 221 /* not yet setup -> nothing to do */ 222 return; 223 } 224 vring->avail = vring->desc + vring->num * sizeof(VRingDesc); 225 vring->used = vring_align(vring->avail + 226 offsetof(VRingAvail, ring[vring->num]), 227 vring->align); 228 virtio_init_region_cache(vdev, n); 229 } 230 231 /* Called within rcu_read_lock(). */ 232 static void vring_split_desc_read(VirtIODevice *vdev, VRingDesc *desc, 233 MemoryRegionCache *cache, int i) 234 { 235 address_space_read_cached(cache, i * sizeof(VRingDesc), 236 desc, sizeof(VRingDesc)); 237 virtio_tswap64s(vdev, &desc->addr); 238 virtio_tswap32s(vdev, &desc->len); 239 virtio_tswap16s(vdev, &desc->flags); 240 virtio_tswap16s(vdev, &desc->next); 241 } 242 243 static void vring_packed_event_read(VirtIODevice *vdev, 244 MemoryRegionCache *cache, 245 VRingPackedDescEvent *e) 246 { 247 hwaddr off_off = offsetof(VRingPackedDescEvent, off_wrap); 248 hwaddr off_flags = offsetof(VRingPackedDescEvent, flags); 249 250 e->flags = virtio_lduw_phys_cached(vdev, cache, off_flags); 251 /* Make sure flags is seen before off_wrap */ 252 smp_rmb(); 253 e->off_wrap = virtio_lduw_phys_cached(vdev, cache, off_off); 254 virtio_tswap16s(vdev, &e->flags); 255 } 256 257 static void vring_packed_off_wrap_write(VirtIODevice *vdev, 258 MemoryRegionCache *cache, 259 uint16_t off_wrap) 260 { 261 hwaddr off = offsetof(VRingPackedDescEvent, off_wrap); 262 263 virtio_stw_phys_cached(vdev, cache, off, off_wrap); 264 address_space_cache_invalidate(cache, off, sizeof(off_wrap)); 265 } 266 267 static void vring_packed_flags_write(VirtIODevice *vdev, 268 MemoryRegionCache *cache, uint16_t flags) 269 { 270 hwaddr off = offsetof(VRingPackedDescEvent, flags); 271 272 virtio_stw_phys_cached(vdev, cache, off, flags); 273 address_space_cache_invalidate(cache, off, sizeof(flags)); 274 } 275 276 /* Called within rcu_read_lock(). */ 277 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq) 278 { 279 return qatomic_rcu_read(&vq->vring.caches); 280 } 281 282 /* Called within rcu_read_lock(). */ 283 static inline uint16_t vring_avail_flags(VirtQueue *vq) 284 { 285 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 286 hwaddr pa = offsetof(VRingAvail, flags); 287 288 if (!caches) { 289 return 0; 290 } 291 292 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa); 293 } 294 295 /* Called within rcu_read_lock(). */ 296 static inline uint16_t vring_avail_idx(VirtQueue *vq) 297 { 298 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 299 hwaddr pa = offsetof(VRingAvail, idx); 300 301 if (!caches) { 302 return 0; 303 } 304 305 vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa); 306 return vq->shadow_avail_idx; 307 } 308 309 /* Called within rcu_read_lock(). */ 310 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i) 311 { 312 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 313 hwaddr pa = offsetof(VRingAvail, ring[i]); 314 315 if (!caches) { 316 return 0; 317 } 318 319 return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa); 320 } 321 322 /* Called within rcu_read_lock(). */ 323 static inline uint16_t vring_get_used_event(VirtQueue *vq) 324 { 325 return vring_avail_ring(vq, vq->vring.num); 326 } 327 328 /* Called within rcu_read_lock(). */ 329 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem, 330 int i) 331 { 332 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 333 hwaddr pa = offsetof(VRingUsed, ring[i]); 334 335 if (!caches) { 336 return; 337 } 338 339 virtio_tswap32s(vq->vdev, &uelem->id); 340 virtio_tswap32s(vq->vdev, &uelem->len); 341 address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem)); 342 address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem)); 343 } 344 345 /* Called within rcu_read_lock(). */ 346 static uint16_t vring_used_idx(VirtQueue *vq) 347 { 348 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 349 hwaddr pa = offsetof(VRingUsed, idx); 350 351 if (!caches) { 352 return 0; 353 } 354 355 return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa); 356 } 357 358 /* Called within rcu_read_lock(). */ 359 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val) 360 { 361 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 362 hwaddr pa = offsetof(VRingUsed, idx); 363 364 if (caches) { 365 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val); 366 address_space_cache_invalidate(&caches->used, pa, sizeof(val)); 367 } 368 369 vq->used_idx = val; 370 } 371 372 /* Called within rcu_read_lock(). */ 373 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask) 374 { 375 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 376 VirtIODevice *vdev = vq->vdev; 377 hwaddr pa = offsetof(VRingUsed, flags); 378 uint16_t flags; 379 380 if (!caches) { 381 return; 382 } 383 384 flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa); 385 virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask); 386 address_space_cache_invalidate(&caches->used, pa, sizeof(flags)); 387 } 388 389 /* Called within rcu_read_lock(). */ 390 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask) 391 { 392 VRingMemoryRegionCaches *caches = vring_get_region_caches(vq); 393 VirtIODevice *vdev = vq->vdev; 394 hwaddr pa = offsetof(VRingUsed, flags); 395 uint16_t flags; 396 397 if (!caches) { 398 return; 399 } 400 401 flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa); 402 virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask); 403 address_space_cache_invalidate(&caches->used, pa, sizeof(flags)); 404 } 405 406 /* Called within rcu_read_lock(). */ 407 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val) 408 { 409 VRingMemoryRegionCaches *caches; 410 hwaddr pa; 411 if (!vq->notification) { 412 return; 413 } 414 415 caches = vring_get_region_caches(vq); 416 if (!caches) { 417 return; 418 } 419 420 pa = offsetof(VRingUsed, ring[vq->vring.num]); 421 virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val); 422 address_space_cache_invalidate(&caches->used, pa, sizeof(val)); 423 } 424 425 static void virtio_queue_split_set_notification(VirtQueue *vq, int enable) 426 { 427 RCU_READ_LOCK_GUARD(); 428 429 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) { 430 vring_set_avail_event(vq, vring_avail_idx(vq)); 431 } else if (enable) { 432 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY); 433 } else { 434 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY); 435 } 436 if (enable) { 437 /* Expose avail event/used flags before caller checks the avail idx. */ 438 smp_mb(); 439 } 440 } 441 442 static void virtio_queue_packed_set_notification(VirtQueue *vq, int enable) 443 { 444 uint16_t off_wrap; 445 VRingPackedDescEvent e; 446 VRingMemoryRegionCaches *caches; 447 448 RCU_READ_LOCK_GUARD(); 449 caches = vring_get_region_caches(vq); 450 if (!caches) { 451 return; 452 } 453 454 vring_packed_event_read(vq->vdev, &caches->used, &e); 455 456 if (!enable) { 457 e.flags = VRING_PACKED_EVENT_FLAG_DISABLE; 458 } else if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) { 459 off_wrap = vq->shadow_avail_idx | vq->shadow_avail_wrap_counter << 15; 460 vring_packed_off_wrap_write(vq->vdev, &caches->used, off_wrap); 461 /* Make sure off_wrap is wrote before flags */ 462 smp_wmb(); 463 e.flags = VRING_PACKED_EVENT_FLAG_DESC; 464 } else { 465 e.flags = VRING_PACKED_EVENT_FLAG_ENABLE; 466 } 467 468 vring_packed_flags_write(vq->vdev, &caches->used, e.flags); 469 if (enable) { 470 /* Expose avail event/used flags before caller checks the avail idx. */ 471 smp_mb(); 472 } 473 } 474 475 bool virtio_queue_get_notification(VirtQueue *vq) 476 { 477 return vq->notification; 478 } 479 480 void virtio_queue_set_notification(VirtQueue *vq, int enable) 481 { 482 vq->notification = enable; 483 484 if (!vq->vring.desc) { 485 return; 486 } 487 488 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 489 virtio_queue_packed_set_notification(vq, enable); 490 } else { 491 virtio_queue_split_set_notification(vq, enable); 492 } 493 } 494 495 int virtio_queue_ready(VirtQueue *vq) 496 { 497 return vq->vring.avail != 0; 498 } 499 500 static void vring_packed_desc_read_flags(VirtIODevice *vdev, 501 uint16_t *flags, 502 MemoryRegionCache *cache, 503 int i) 504 { 505 hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags); 506 507 *flags = virtio_lduw_phys_cached(vdev, cache, off); 508 } 509 510 static void vring_packed_desc_read(VirtIODevice *vdev, 511 VRingPackedDesc *desc, 512 MemoryRegionCache *cache, 513 int i, bool strict_order) 514 { 515 hwaddr off = i * sizeof(VRingPackedDesc); 516 517 vring_packed_desc_read_flags(vdev, &desc->flags, cache, i); 518 519 if (strict_order) { 520 /* Make sure flags is read before the rest fields. */ 521 smp_rmb(); 522 } 523 524 address_space_read_cached(cache, off + offsetof(VRingPackedDesc, addr), 525 &desc->addr, sizeof(desc->addr)); 526 address_space_read_cached(cache, off + offsetof(VRingPackedDesc, id), 527 &desc->id, sizeof(desc->id)); 528 address_space_read_cached(cache, off + offsetof(VRingPackedDesc, len), 529 &desc->len, sizeof(desc->len)); 530 virtio_tswap64s(vdev, &desc->addr); 531 virtio_tswap16s(vdev, &desc->id); 532 virtio_tswap32s(vdev, &desc->len); 533 } 534 535 static void vring_packed_desc_write_data(VirtIODevice *vdev, 536 VRingPackedDesc *desc, 537 MemoryRegionCache *cache, 538 int i) 539 { 540 hwaddr off_id = i * sizeof(VRingPackedDesc) + 541 offsetof(VRingPackedDesc, id); 542 hwaddr off_len = i * sizeof(VRingPackedDesc) + 543 offsetof(VRingPackedDesc, len); 544 545 virtio_tswap32s(vdev, &desc->len); 546 virtio_tswap16s(vdev, &desc->id); 547 address_space_write_cached(cache, off_id, &desc->id, sizeof(desc->id)); 548 address_space_cache_invalidate(cache, off_id, sizeof(desc->id)); 549 address_space_write_cached(cache, off_len, &desc->len, sizeof(desc->len)); 550 address_space_cache_invalidate(cache, off_len, sizeof(desc->len)); 551 } 552 553 static void vring_packed_desc_write_flags(VirtIODevice *vdev, 554 VRingPackedDesc *desc, 555 MemoryRegionCache *cache, 556 int i) 557 { 558 hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags); 559 560 virtio_stw_phys_cached(vdev, cache, off, desc->flags); 561 address_space_cache_invalidate(cache, off, sizeof(desc->flags)); 562 } 563 564 static void vring_packed_desc_write(VirtIODevice *vdev, 565 VRingPackedDesc *desc, 566 MemoryRegionCache *cache, 567 int i, bool strict_order) 568 { 569 vring_packed_desc_write_data(vdev, desc, cache, i); 570 if (strict_order) { 571 /* Make sure data is wrote before flags. */ 572 smp_wmb(); 573 } 574 vring_packed_desc_write_flags(vdev, desc, cache, i); 575 } 576 577 static inline bool is_desc_avail(uint16_t flags, bool wrap_counter) 578 { 579 bool avail, used; 580 581 avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL)); 582 used = !!(flags & (1 << VRING_PACKED_DESC_F_USED)); 583 return (avail != used) && (avail == wrap_counter); 584 } 585 586 /* Fetch avail_idx from VQ memory only when we really need to know if 587 * guest has added some buffers. 588 * Called within rcu_read_lock(). */ 589 static int virtio_queue_empty_rcu(VirtQueue *vq) 590 { 591 if (virtio_device_disabled(vq->vdev)) { 592 return 1; 593 } 594 595 if (unlikely(!vq->vring.avail)) { 596 return 1; 597 } 598 599 if (vq->shadow_avail_idx != vq->last_avail_idx) { 600 return 0; 601 } 602 603 return vring_avail_idx(vq) == vq->last_avail_idx; 604 } 605 606 static int virtio_queue_split_empty(VirtQueue *vq) 607 { 608 bool empty; 609 610 if (virtio_device_disabled(vq->vdev)) { 611 return 1; 612 } 613 614 if (unlikely(!vq->vring.avail)) { 615 return 1; 616 } 617 618 if (vq->shadow_avail_idx != vq->last_avail_idx) { 619 return 0; 620 } 621 622 RCU_READ_LOCK_GUARD(); 623 empty = vring_avail_idx(vq) == vq->last_avail_idx; 624 return empty; 625 } 626 627 /* Called within rcu_read_lock(). */ 628 static int virtio_queue_packed_empty_rcu(VirtQueue *vq) 629 { 630 struct VRingPackedDesc desc; 631 VRingMemoryRegionCaches *cache; 632 633 if (unlikely(!vq->vring.desc)) { 634 return 1; 635 } 636 637 cache = vring_get_region_caches(vq); 638 if (!cache) { 639 return 1; 640 } 641 642 vring_packed_desc_read_flags(vq->vdev, &desc.flags, &cache->desc, 643 vq->last_avail_idx); 644 645 return !is_desc_avail(desc.flags, vq->last_avail_wrap_counter); 646 } 647 648 static int virtio_queue_packed_empty(VirtQueue *vq) 649 { 650 RCU_READ_LOCK_GUARD(); 651 return virtio_queue_packed_empty_rcu(vq); 652 } 653 654 int virtio_queue_empty(VirtQueue *vq) 655 { 656 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 657 return virtio_queue_packed_empty(vq); 658 } else { 659 return virtio_queue_split_empty(vq); 660 } 661 } 662 663 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem, 664 unsigned int len) 665 { 666 AddressSpace *dma_as = vq->vdev->dma_as; 667 unsigned int offset; 668 int i; 669 670 offset = 0; 671 for (i = 0; i < elem->in_num; i++) { 672 size_t size = MIN(len - offset, elem->in_sg[i].iov_len); 673 674 dma_memory_unmap(dma_as, elem->in_sg[i].iov_base, 675 elem->in_sg[i].iov_len, 676 DMA_DIRECTION_FROM_DEVICE, size); 677 678 offset += size; 679 } 680 681 for (i = 0; i < elem->out_num; i++) 682 dma_memory_unmap(dma_as, elem->out_sg[i].iov_base, 683 elem->out_sg[i].iov_len, 684 DMA_DIRECTION_TO_DEVICE, 685 elem->out_sg[i].iov_len); 686 } 687 688 /* virtqueue_detach_element: 689 * @vq: The #VirtQueue 690 * @elem: The #VirtQueueElement 691 * @len: number of bytes written 692 * 693 * Detach the element from the virtqueue. This function is suitable for device 694 * reset or other situations where a #VirtQueueElement is simply freed and will 695 * not be pushed or discarded. 696 */ 697 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem, 698 unsigned int len) 699 { 700 vq->inuse -= elem->ndescs; 701 virtqueue_unmap_sg(vq, elem, len); 702 } 703 704 static void virtqueue_split_rewind(VirtQueue *vq, unsigned int num) 705 { 706 vq->last_avail_idx -= num; 707 } 708 709 static void virtqueue_packed_rewind(VirtQueue *vq, unsigned int num) 710 { 711 if (vq->last_avail_idx < num) { 712 vq->last_avail_idx = vq->vring.num + vq->last_avail_idx - num; 713 vq->last_avail_wrap_counter ^= 1; 714 } else { 715 vq->last_avail_idx -= num; 716 } 717 } 718 719 /* virtqueue_unpop: 720 * @vq: The #VirtQueue 721 * @elem: The #VirtQueueElement 722 * @len: number of bytes written 723 * 724 * Pretend the most recent element wasn't popped from the virtqueue. The next 725 * call to virtqueue_pop() will refetch the element. 726 */ 727 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem, 728 unsigned int len) 729 { 730 731 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 732 virtqueue_packed_rewind(vq, 1); 733 } else { 734 virtqueue_split_rewind(vq, 1); 735 } 736 737 virtqueue_detach_element(vq, elem, len); 738 } 739 740 /* virtqueue_rewind: 741 * @vq: The #VirtQueue 742 * @num: Number of elements to push back 743 * 744 * Pretend that elements weren't popped from the virtqueue. The next 745 * virtqueue_pop() will refetch the oldest element. 746 * 747 * Use virtqueue_unpop() instead if you have a VirtQueueElement. 748 * 749 * Returns: true on success, false if @num is greater than the number of in use 750 * elements. 751 */ 752 bool virtqueue_rewind(VirtQueue *vq, unsigned int num) 753 { 754 if (num > vq->inuse) { 755 return false; 756 } 757 758 vq->inuse -= num; 759 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 760 virtqueue_packed_rewind(vq, num); 761 } else { 762 virtqueue_split_rewind(vq, num); 763 } 764 return true; 765 } 766 767 static void virtqueue_split_fill(VirtQueue *vq, const VirtQueueElement *elem, 768 unsigned int len, unsigned int idx) 769 { 770 VRingUsedElem uelem; 771 772 if (unlikely(!vq->vring.used)) { 773 return; 774 } 775 776 idx = (idx + vq->used_idx) % vq->vring.num; 777 778 uelem.id = elem->index; 779 uelem.len = len; 780 vring_used_write(vq, &uelem, idx); 781 } 782 783 static void virtqueue_packed_fill(VirtQueue *vq, const VirtQueueElement *elem, 784 unsigned int len, unsigned int idx) 785 { 786 vq->used_elems[idx].index = elem->index; 787 vq->used_elems[idx].len = len; 788 vq->used_elems[idx].ndescs = elem->ndescs; 789 } 790 791 static void virtqueue_packed_fill_desc(VirtQueue *vq, 792 const VirtQueueElement *elem, 793 unsigned int idx, 794 bool strict_order) 795 { 796 uint16_t head; 797 VRingMemoryRegionCaches *caches; 798 VRingPackedDesc desc = { 799 .id = elem->index, 800 .len = elem->len, 801 }; 802 bool wrap_counter = vq->used_wrap_counter; 803 804 if (unlikely(!vq->vring.desc)) { 805 return; 806 } 807 808 head = vq->used_idx + idx; 809 if (head >= vq->vring.num) { 810 head -= vq->vring.num; 811 wrap_counter ^= 1; 812 } 813 if (wrap_counter) { 814 desc.flags |= (1 << VRING_PACKED_DESC_F_AVAIL); 815 desc.flags |= (1 << VRING_PACKED_DESC_F_USED); 816 } else { 817 desc.flags &= ~(1 << VRING_PACKED_DESC_F_AVAIL); 818 desc.flags &= ~(1 << VRING_PACKED_DESC_F_USED); 819 } 820 821 caches = vring_get_region_caches(vq); 822 if (!caches) { 823 return; 824 } 825 826 vring_packed_desc_write(vq->vdev, &desc, &caches->desc, head, strict_order); 827 } 828 829 /* Called within rcu_read_lock(). */ 830 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem, 831 unsigned int len, unsigned int idx) 832 { 833 trace_virtqueue_fill(vq, elem, len, idx); 834 835 virtqueue_unmap_sg(vq, elem, len); 836 837 if (virtio_device_disabled(vq->vdev)) { 838 return; 839 } 840 841 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 842 virtqueue_packed_fill(vq, elem, len, idx); 843 } else { 844 virtqueue_split_fill(vq, elem, len, idx); 845 } 846 } 847 848 /* Called within rcu_read_lock(). */ 849 static void virtqueue_split_flush(VirtQueue *vq, unsigned int count) 850 { 851 uint16_t old, new; 852 853 if (unlikely(!vq->vring.used)) { 854 return; 855 } 856 857 /* Make sure buffer is written before we update index. */ 858 smp_wmb(); 859 trace_virtqueue_flush(vq, count); 860 old = vq->used_idx; 861 new = old + count; 862 vring_used_idx_set(vq, new); 863 vq->inuse -= count; 864 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old))) 865 vq->signalled_used_valid = false; 866 } 867 868 static void virtqueue_packed_flush(VirtQueue *vq, unsigned int count) 869 { 870 unsigned int i, ndescs = 0; 871 872 if (unlikely(!vq->vring.desc)) { 873 return; 874 } 875 876 for (i = 1; i < count; i++) { 877 virtqueue_packed_fill_desc(vq, &vq->used_elems[i], i, false); 878 ndescs += vq->used_elems[i].ndescs; 879 } 880 virtqueue_packed_fill_desc(vq, &vq->used_elems[0], 0, true); 881 ndescs += vq->used_elems[0].ndescs; 882 883 vq->inuse -= ndescs; 884 vq->used_idx += ndescs; 885 if (vq->used_idx >= vq->vring.num) { 886 vq->used_idx -= vq->vring.num; 887 vq->used_wrap_counter ^= 1; 888 vq->signalled_used_valid = false; 889 } 890 } 891 892 void virtqueue_flush(VirtQueue *vq, unsigned int count) 893 { 894 if (virtio_device_disabled(vq->vdev)) { 895 vq->inuse -= count; 896 return; 897 } 898 899 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 900 virtqueue_packed_flush(vq, count); 901 } else { 902 virtqueue_split_flush(vq, count); 903 } 904 } 905 906 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem, 907 unsigned int len) 908 { 909 RCU_READ_LOCK_GUARD(); 910 virtqueue_fill(vq, elem, len, 0); 911 virtqueue_flush(vq, 1); 912 } 913 914 /* Called within rcu_read_lock(). */ 915 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx) 916 { 917 uint16_t num_heads = vring_avail_idx(vq) - idx; 918 919 /* Check it isn't doing very strange things with descriptor numbers. */ 920 if (num_heads > vq->vring.num) { 921 virtio_error(vq->vdev, "Guest moved used index from %u to %u", 922 idx, vq->shadow_avail_idx); 923 return -EINVAL; 924 } 925 /* On success, callers read a descriptor at vq->last_avail_idx. 926 * Make sure descriptor read does not bypass avail index read. */ 927 if (num_heads) { 928 smp_rmb(); 929 } 930 931 return num_heads; 932 } 933 934 /* Called within rcu_read_lock(). */ 935 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx, 936 unsigned int *head) 937 { 938 /* Grab the next descriptor number they're advertising, and increment 939 * the index we've seen. */ 940 *head = vring_avail_ring(vq, idx % vq->vring.num); 941 942 /* If their number is silly, that's a fatal mistake. */ 943 if (*head >= vq->vring.num) { 944 virtio_error(vq->vdev, "Guest says index %u is available", *head); 945 return false; 946 } 947 948 return true; 949 } 950 951 enum { 952 VIRTQUEUE_READ_DESC_ERROR = -1, 953 VIRTQUEUE_READ_DESC_DONE = 0, /* end of chain */ 954 VIRTQUEUE_READ_DESC_MORE = 1, /* more buffers in chain */ 955 }; 956 957 static int virtqueue_split_read_next_desc(VirtIODevice *vdev, VRingDesc *desc, 958 MemoryRegionCache *desc_cache, 959 unsigned int max, unsigned int *next) 960 { 961 /* If this descriptor says it doesn't chain, we're done. */ 962 if (!(desc->flags & VRING_DESC_F_NEXT)) { 963 return VIRTQUEUE_READ_DESC_DONE; 964 } 965 966 /* Check they're not leading us off end of descriptors. */ 967 *next = desc->next; 968 /* Make sure compiler knows to grab that: we don't want it changing! */ 969 smp_wmb(); 970 971 if (*next >= max) { 972 virtio_error(vdev, "Desc next is %u", *next); 973 return VIRTQUEUE_READ_DESC_ERROR; 974 } 975 976 vring_split_desc_read(vdev, desc, desc_cache, *next); 977 return VIRTQUEUE_READ_DESC_MORE; 978 } 979 980 /* Called within rcu_read_lock(). */ 981 static void virtqueue_split_get_avail_bytes(VirtQueue *vq, 982 unsigned int *in_bytes, unsigned int *out_bytes, 983 unsigned max_in_bytes, unsigned max_out_bytes, 984 VRingMemoryRegionCaches *caches) 985 { 986 VirtIODevice *vdev = vq->vdev; 987 unsigned int max, idx; 988 unsigned int total_bufs, in_total, out_total; 989 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; 990 int64_t len = 0; 991 int rc; 992 993 idx = vq->last_avail_idx; 994 total_bufs = in_total = out_total = 0; 995 996 max = vq->vring.num; 997 998 while ((rc = virtqueue_num_heads(vq, idx)) > 0) { 999 MemoryRegionCache *desc_cache = &caches->desc; 1000 unsigned int num_bufs; 1001 VRingDesc desc; 1002 unsigned int i; 1003 1004 num_bufs = total_bufs; 1005 1006 if (!virtqueue_get_head(vq, idx++, &i)) { 1007 goto err; 1008 } 1009 1010 vring_split_desc_read(vdev, &desc, desc_cache, i); 1011 1012 if (desc.flags & VRING_DESC_F_INDIRECT) { 1013 if (!desc.len || (desc.len % sizeof(VRingDesc))) { 1014 virtio_error(vdev, "Invalid size for indirect buffer table"); 1015 goto err; 1016 } 1017 1018 /* If we've got too many, that implies a descriptor loop. */ 1019 if (num_bufs >= max) { 1020 virtio_error(vdev, "Looped descriptor"); 1021 goto err; 1022 } 1023 1024 /* loop over the indirect descriptor table */ 1025 len = address_space_cache_init(&indirect_desc_cache, 1026 vdev->dma_as, 1027 desc.addr, desc.len, false); 1028 desc_cache = &indirect_desc_cache; 1029 if (len < desc.len) { 1030 virtio_error(vdev, "Cannot map indirect buffer"); 1031 goto err; 1032 } 1033 1034 max = desc.len / sizeof(VRingDesc); 1035 num_bufs = i = 0; 1036 vring_split_desc_read(vdev, &desc, desc_cache, i); 1037 } 1038 1039 do { 1040 /* If we've got too many, that implies a descriptor loop. */ 1041 if (++num_bufs > max) { 1042 virtio_error(vdev, "Looped descriptor"); 1043 goto err; 1044 } 1045 1046 if (desc.flags & VRING_DESC_F_WRITE) { 1047 in_total += desc.len; 1048 } else { 1049 out_total += desc.len; 1050 } 1051 if (in_total >= max_in_bytes && out_total >= max_out_bytes) { 1052 goto done; 1053 } 1054 1055 rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i); 1056 } while (rc == VIRTQUEUE_READ_DESC_MORE); 1057 1058 if (rc == VIRTQUEUE_READ_DESC_ERROR) { 1059 goto err; 1060 } 1061 1062 if (desc_cache == &indirect_desc_cache) { 1063 address_space_cache_destroy(&indirect_desc_cache); 1064 total_bufs++; 1065 } else { 1066 total_bufs = num_bufs; 1067 } 1068 } 1069 1070 if (rc < 0) { 1071 goto err; 1072 } 1073 1074 done: 1075 address_space_cache_destroy(&indirect_desc_cache); 1076 if (in_bytes) { 1077 *in_bytes = in_total; 1078 } 1079 if (out_bytes) { 1080 *out_bytes = out_total; 1081 } 1082 return; 1083 1084 err: 1085 in_total = out_total = 0; 1086 goto done; 1087 } 1088 1089 static int virtqueue_packed_read_next_desc(VirtQueue *vq, 1090 VRingPackedDesc *desc, 1091 MemoryRegionCache 1092 *desc_cache, 1093 unsigned int max, 1094 unsigned int *next, 1095 bool indirect) 1096 { 1097 /* If this descriptor says it doesn't chain, we're done. */ 1098 if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) { 1099 return VIRTQUEUE_READ_DESC_DONE; 1100 } 1101 1102 ++*next; 1103 if (*next == max) { 1104 if (indirect) { 1105 return VIRTQUEUE_READ_DESC_DONE; 1106 } else { 1107 (*next) -= vq->vring.num; 1108 } 1109 } 1110 1111 vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false); 1112 return VIRTQUEUE_READ_DESC_MORE; 1113 } 1114 1115 /* Called within rcu_read_lock(). */ 1116 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq, 1117 unsigned int *in_bytes, 1118 unsigned int *out_bytes, 1119 unsigned max_in_bytes, 1120 unsigned max_out_bytes, 1121 VRingMemoryRegionCaches *caches) 1122 { 1123 VirtIODevice *vdev = vq->vdev; 1124 unsigned int max, idx; 1125 unsigned int total_bufs, in_total, out_total; 1126 MemoryRegionCache *desc_cache; 1127 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; 1128 int64_t len = 0; 1129 VRingPackedDesc desc; 1130 bool wrap_counter; 1131 1132 idx = vq->last_avail_idx; 1133 wrap_counter = vq->last_avail_wrap_counter; 1134 total_bufs = in_total = out_total = 0; 1135 1136 max = vq->vring.num; 1137 1138 for (;;) { 1139 unsigned int num_bufs = total_bufs; 1140 unsigned int i = idx; 1141 int rc; 1142 1143 desc_cache = &caches->desc; 1144 vring_packed_desc_read(vdev, &desc, desc_cache, idx, true); 1145 if (!is_desc_avail(desc.flags, wrap_counter)) { 1146 break; 1147 } 1148 1149 if (desc.flags & VRING_DESC_F_INDIRECT) { 1150 if (desc.len % sizeof(VRingPackedDesc)) { 1151 virtio_error(vdev, "Invalid size for indirect buffer table"); 1152 goto err; 1153 } 1154 1155 /* If we've got too many, that implies a descriptor loop. */ 1156 if (num_bufs >= max) { 1157 virtio_error(vdev, "Looped descriptor"); 1158 goto err; 1159 } 1160 1161 /* loop over the indirect descriptor table */ 1162 len = address_space_cache_init(&indirect_desc_cache, 1163 vdev->dma_as, 1164 desc.addr, desc.len, false); 1165 desc_cache = &indirect_desc_cache; 1166 if (len < desc.len) { 1167 virtio_error(vdev, "Cannot map indirect buffer"); 1168 goto err; 1169 } 1170 1171 max = desc.len / sizeof(VRingPackedDesc); 1172 num_bufs = i = 0; 1173 vring_packed_desc_read(vdev, &desc, desc_cache, i, false); 1174 } 1175 1176 do { 1177 /* If we've got too many, that implies a descriptor loop. */ 1178 if (++num_bufs > max) { 1179 virtio_error(vdev, "Looped descriptor"); 1180 goto err; 1181 } 1182 1183 if (desc.flags & VRING_DESC_F_WRITE) { 1184 in_total += desc.len; 1185 } else { 1186 out_total += desc.len; 1187 } 1188 if (in_total >= max_in_bytes && out_total >= max_out_bytes) { 1189 goto done; 1190 } 1191 1192 rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, 1193 &i, desc_cache == 1194 &indirect_desc_cache); 1195 } while (rc == VIRTQUEUE_READ_DESC_MORE); 1196 1197 if (desc_cache == &indirect_desc_cache) { 1198 address_space_cache_destroy(&indirect_desc_cache); 1199 total_bufs++; 1200 idx++; 1201 } else { 1202 idx += num_bufs - total_bufs; 1203 total_bufs = num_bufs; 1204 } 1205 1206 if (idx >= vq->vring.num) { 1207 idx -= vq->vring.num; 1208 wrap_counter ^= 1; 1209 } 1210 } 1211 1212 /* Record the index and wrap counter for a kick we want */ 1213 vq->shadow_avail_idx = idx; 1214 vq->shadow_avail_wrap_counter = wrap_counter; 1215 done: 1216 address_space_cache_destroy(&indirect_desc_cache); 1217 if (in_bytes) { 1218 *in_bytes = in_total; 1219 } 1220 if (out_bytes) { 1221 *out_bytes = out_total; 1222 } 1223 return; 1224 1225 err: 1226 in_total = out_total = 0; 1227 goto done; 1228 } 1229 1230 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes, 1231 unsigned int *out_bytes, 1232 unsigned max_in_bytes, unsigned max_out_bytes) 1233 { 1234 uint16_t desc_size; 1235 VRingMemoryRegionCaches *caches; 1236 1237 RCU_READ_LOCK_GUARD(); 1238 1239 if (unlikely(!vq->vring.desc)) { 1240 goto err; 1241 } 1242 1243 caches = vring_get_region_caches(vq); 1244 if (!caches) { 1245 goto err; 1246 } 1247 1248 desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ? 1249 sizeof(VRingPackedDesc) : sizeof(VRingDesc); 1250 if (caches->desc.len < vq->vring.num * desc_size) { 1251 virtio_error(vq->vdev, "Cannot map descriptor ring"); 1252 goto err; 1253 } 1254 1255 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 1256 virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes, 1257 max_in_bytes, max_out_bytes, 1258 caches); 1259 } else { 1260 virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes, 1261 max_in_bytes, max_out_bytes, 1262 caches); 1263 } 1264 1265 return; 1266 err: 1267 if (in_bytes) { 1268 *in_bytes = 0; 1269 } 1270 if (out_bytes) { 1271 *out_bytes = 0; 1272 } 1273 } 1274 1275 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes, 1276 unsigned int out_bytes) 1277 { 1278 unsigned int in_total, out_total; 1279 1280 virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes); 1281 return in_bytes <= in_total && out_bytes <= out_total; 1282 } 1283 1284 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg, 1285 hwaddr *addr, struct iovec *iov, 1286 unsigned int max_num_sg, bool is_write, 1287 hwaddr pa, size_t sz) 1288 { 1289 bool ok = false; 1290 unsigned num_sg = *p_num_sg; 1291 assert(num_sg <= max_num_sg); 1292 1293 if (!sz) { 1294 virtio_error(vdev, "virtio: zero sized buffers are not allowed"); 1295 goto out; 1296 } 1297 1298 while (sz) { 1299 hwaddr len = sz; 1300 1301 if (num_sg == max_num_sg) { 1302 virtio_error(vdev, "virtio: too many write descriptors in " 1303 "indirect table"); 1304 goto out; 1305 } 1306 1307 iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len, 1308 is_write ? 1309 DMA_DIRECTION_FROM_DEVICE : 1310 DMA_DIRECTION_TO_DEVICE, 1311 MEMTXATTRS_UNSPECIFIED); 1312 if (!iov[num_sg].iov_base) { 1313 virtio_error(vdev, "virtio: bogus descriptor or out of resources"); 1314 goto out; 1315 } 1316 1317 iov[num_sg].iov_len = len; 1318 addr[num_sg] = pa; 1319 1320 sz -= len; 1321 pa += len; 1322 num_sg++; 1323 } 1324 ok = true; 1325 1326 out: 1327 *p_num_sg = num_sg; 1328 return ok; 1329 } 1330 1331 /* Only used by error code paths before we have a VirtQueueElement (therefore 1332 * virtqueue_unmap_sg() can't be used). Assumes buffers weren't written to 1333 * yet. 1334 */ 1335 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num, 1336 struct iovec *iov) 1337 { 1338 unsigned int i; 1339 1340 for (i = 0; i < out_num + in_num; i++) { 1341 int is_write = i >= out_num; 1342 1343 cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0); 1344 iov++; 1345 } 1346 } 1347 1348 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg, 1349 hwaddr *addr, unsigned int num_sg, 1350 bool is_write) 1351 { 1352 unsigned int i; 1353 hwaddr len; 1354 1355 for (i = 0; i < num_sg; i++) { 1356 len = sg[i].iov_len; 1357 sg[i].iov_base = dma_memory_map(vdev->dma_as, 1358 addr[i], &len, is_write ? 1359 DMA_DIRECTION_FROM_DEVICE : 1360 DMA_DIRECTION_TO_DEVICE, 1361 MEMTXATTRS_UNSPECIFIED); 1362 if (!sg[i].iov_base) { 1363 error_report("virtio: error trying to map MMIO memory"); 1364 exit(1); 1365 } 1366 if (len != sg[i].iov_len) { 1367 error_report("virtio: unexpected memory split"); 1368 exit(1); 1369 } 1370 } 1371 } 1372 1373 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem) 1374 { 1375 virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, true); 1376 virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num, 1377 false); 1378 } 1379 1380 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num) 1381 { 1382 VirtQueueElement *elem; 1383 size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0])); 1384 size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]); 1385 size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]); 1386 size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0])); 1387 size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]); 1388 size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]); 1389 1390 assert(sz >= sizeof(VirtQueueElement)); 1391 elem = g_malloc(out_sg_end); 1392 trace_virtqueue_alloc_element(elem, sz, in_num, out_num); 1393 elem->out_num = out_num; 1394 elem->in_num = in_num; 1395 elem->in_addr = (void *)elem + in_addr_ofs; 1396 elem->out_addr = (void *)elem + out_addr_ofs; 1397 elem->in_sg = (void *)elem + in_sg_ofs; 1398 elem->out_sg = (void *)elem + out_sg_ofs; 1399 return elem; 1400 } 1401 1402 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz) 1403 { 1404 unsigned int i, head, max; 1405 VRingMemoryRegionCaches *caches; 1406 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; 1407 MemoryRegionCache *desc_cache; 1408 int64_t len; 1409 VirtIODevice *vdev = vq->vdev; 1410 VirtQueueElement *elem = NULL; 1411 unsigned out_num, in_num, elem_entries; 1412 hwaddr addr[VIRTQUEUE_MAX_SIZE]; 1413 struct iovec iov[VIRTQUEUE_MAX_SIZE]; 1414 VRingDesc desc; 1415 int rc; 1416 1417 RCU_READ_LOCK_GUARD(); 1418 if (virtio_queue_empty_rcu(vq)) { 1419 goto done; 1420 } 1421 /* Needed after virtio_queue_empty(), see comment in 1422 * virtqueue_num_heads(). */ 1423 smp_rmb(); 1424 1425 /* When we start there are none of either input nor output. */ 1426 out_num = in_num = elem_entries = 0; 1427 1428 max = vq->vring.num; 1429 1430 if (vq->inuse >= vq->vring.num) { 1431 virtio_error(vdev, "Virtqueue size exceeded"); 1432 goto done; 1433 } 1434 1435 if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) { 1436 goto done; 1437 } 1438 1439 if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { 1440 vring_set_avail_event(vq, vq->last_avail_idx); 1441 } 1442 1443 i = head; 1444 1445 caches = vring_get_region_caches(vq); 1446 if (!caches) { 1447 virtio_error(vdev, "Region caches not initialized"); 1448 goto done; 1449 } 1450 1451 if (caches->desc.len < max * sizeof(VRingDesc)) { 1452 virtio_error(vdev, "Cannot map descriptor ring"); 1453 goto done; 1454 } 1455 1456 desc_cache = &caches->desc; 1457 vring_split_desc_read(vdev, &desc, desc_cache, i); 1458 if (desc.flags & VRING_DESC_F_INDIRECT) { 1459 if (!desc.len || (desc.len % sizeof(VRingDesc))) { 1460 virtio_error(vdev, "Invalid size for indirect buffer table"); 1461 goto done; 1462 } 1463 1464 /* loop over the indirect descriptor table */ 1465 len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as, 1466 desc.addr, desc.len, false); 1467 desc_cache = &indirect_desc_cache; 1468 if (len < desc.len) { 1469 virtio_error(vdev, "Cannot map indirect buffer"); 1470 goto done; 1471 } 1472 1473 max = desc.len / sizeof(VRingDesc); 1474 i = 0; 1475 vring_split_desc_read(vdev, &desc, desc_cache, i); 1476 } 1477 1478 /* Collect all the descriptors */ 1479 do { 1480 bool map_ok; 1481 1482 if (desc.flags & VRING_DESC_F_WRITE) { 1483 map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num, 1484 iov + out_num, 1485 VIRTQUEUE_MAX_SIZE - out_num, true, 1486 desc.addr, desc.len); 1487 } else { 1488 if (in_num) { 1489 virtio_error(vdev, "Incorrect order for descriptors"); 1490 goto err_undo_map; 1491 } 1492 map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov, 1493 VIRTQUEUE_MAX_SIZE, false, 1494 desc.addr, desc.len); 1495 } 1496 if (!map_ok) { 1497 goto err_undo_map; 1498 } 1499 1500 /* If we've got too many, that implies a descriptor loop. */ 1501 if (++elem_entries > max) { 1502 virtio_error(vdev, "Looped descriptor"); 1503 goto err_undo_map; 1504 } 1505 1506 rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i); 1507 } while (rc == VIRTQUEUE_READ_DESC_MORE); 1508 1509 if (rc == VIRTQUEUE_READ_DESC_ERROR) { 1510 goto err_undo_map; 1511 } 1512 1513 /* Now copy what we have collected and mapped */ 1514 elem = virtqueue_alloc_element(sz, out_num, in_num); 1515 elem->index = head; 1516 elem->ndescs = 1; 1517 for (i = 0; i < out_num; i++) { 1518 elem->out_addr[i] = addr[i]; 1519 elem->out_sg[i] = iov[i]; 1520 } 1521 for (i = 0; i < in_num; i++) { 1522 elem->in_addr[i] = addr[out_num + i]; 1523 elem->in_sg[i] = iov[out_num + i]; 1524 } 1525 1526 vq->inuse++; 1527 1528 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num); 1529 done: 1530 address_space_cache_destroy(&indirect_desc_cache); 1531 1532 return elem; 1533 1534 err_undo_map: 1535 virtqueue_undo_map_desc(out_num, in_num, iov); 1536 goto done; 1537 } 1538 1539 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz) 1540 { 1541 unsigned int i, max; 1542 VRingMemoryRegionCaches *caches; 1543 MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID; 1544 MemoryRegionCache *desc_cache; 1545 int64_t len; 1546 VirtIODevice *vdev = vq->vdev; 1547 VirtQueueElement *elem = NULL; 1548 unsigned out_num, in_num, elem_entries; 1549 hwaddr addr[VIRTQUEUE_MAX_SIZE]; 1550 struct iovec iov[VIRTQUEUE_MAX_SIZE]; 1551 VRingPackedDesc desc; 1552 uint16_t id; 1553 int rc; 1554 1555 RCU_READ_LOCK_GUARD(); 1556 if (virtio_queue_packed_empty_rcu(vq)) { 1557 goto done; 1558 } 1559 1560 /* When we start there are none of either input nor output. */ 1561 out_num = in_num = elem_entries = 0; 1562 1563 max = vq->vring.num; 1564 1565 if (vq->inuse >= vq->vring.num) { 1566 virtio_error(vdev, "Virtqueue size exceeded"); 1567 goto done; 1568 } 1569 1570 i = vq->last_avail_idx; 1571 1572 caches = vring_get_region_caches(vq); 1573 if (!caches) { 1574 virtio_error(vdev, "Region caches not initialized"); 1575 goto done; 1576 } 1577 1578 if (caches->desc.len < max * sizeof(VRingDesc)) { 1579 virtio_error(vdev, "Cannot map descriptor ring"); 1580 goto done; 1581 } 1582 1583 desc_cache = &caches->desc; 1584 vring_packed_desc_read(vdev, &desc, desc_cache, i, true); 1585 id = desc.id; 1586 if (desc.flags & VRING_DESC_F_INDIRECT) { 1587 if (desc.len % sizeof(VRingPackedDesc)) { 1588 virtio_error(vdev, "Invalid size for indirect buffer table"); 1589 goto done; 1590 } 1591 1592 /* loop over the indirect descriptor table */ 1593 len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as, 1594 desc.addr, desc.len, false); 1595 desc_cache = &indirect_desc_cache; 1596 if (len < desc.len) { 1597 virtio_error(vdev, "Cannot map indirect buffer"); 1598 goto done; 1599 } 1600 1601 max = desc.len / sizeof(VRingPackedDesc); 1602 i = 0; 1603 vring_packed_desc_read(vdev, &desc, desc_cache, i, false); 1604 } 1605 1606 /* Collect all the descriptors */ 1607 do { 1608 bool map_ok; 1609 1610 if (desc.flags & VRING_DESC_F_WRITE) { 1611 map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num, 1612 iov + out_num, 1613 VIRTQUEUE_MAX_SIZE - out_num, true, 1614 desc.addr, desc.len); 1615 } else { 1616 if (in_num) { 1617 virtio_error(vdev, "Incorrect order for descriptors"); 1618 goto err_undo_map; 1619 } 1620 map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov, 1621 VIRTQUEUE_MAX_SIZE, false, 1622 desc.addr, desc.len); 1623 } 1624 if (!map_ok) { 1625 goto err_undo_map; 1626 } 1627 1628 /* If we've got too many, that implies a descriptor loop. */ 1629 if (++elem_entries > max) { 1630 virtio_error(vdev, "Looped descriptor"); 1631 goto err_undo_map; 1632 } 1633 1634 rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i, 1635 desc_cache == 1636 &indirect_desc_cache); 1637 } while (rc == VIRTQUEUE_READ_DESC_MORE); 1638 1639 /* Now copy what we have collected and mapped */ 1640 elem = virtqueue_alloc_element(sz, out_num, in_num); 1641 for (i = 0; i < out_num; i++) { 1642 elem->out_addr[i] = addr[i]; 1643 elem->out_sg[i] = iov[i]; 1644 } 1645 for (i = 0; i < in_num; i++) { 1646 elem->in_addr[i] = addr[out_num + i]; 1647 elem->in_sg[i] = iov[out_num + i]; 1648 } 1649 1650 elem->index = id; 1651 elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries; 1652 vq->last_avail_idx += elem->ndescs; 1653 vq->inuse += elem->ndescs; 1654 1655 if (vq->last_avail_idx >= vq->vring.num) { 1656 vq->last_avail_idx -= vq->vring.num; 1657 vq->last_avail_wrap_counter ^= 1; 1658 } 1659 1660 vq->shadow_avail_idx = vq->last_avail_idx; 1661 vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter; 1662 1663 trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num); 1664 done: 1665 address_space_cache_destroy(&indirect_desc_cache); 1666 1667 return elem; 1668 1669 err_undo_map: 1670 virtqueue_undo_map_desc(out_num, in_num, iov); 1671 goto done; 1672 } 1673 1674 void *virtqueue_pop(VirtQueue *vq, size_t sz) 1675 { 1676 if (virtio_device_disabled(vq->vdev)) { 1677 return NULL; 1678 } 1679 1680 if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) { 1681 return virtqueue_packed_pop(vq, sz); 1682 } else { 1683 return virtqueue_split_pop(vq, sz); 1684 } 1685 } 1686 1687 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq) 1688 { 1689 VRingMemoryRegionCaches *caches; 1690 MemoryRegionCache *desc_cache; 1691 unsigned int dropped = 0; 1692 VirtQueueElement elem = {}; 1693 VirtIODevice *vdev = vq->vdev; 1694 VRingPackedDesc desc; 1695 1696 RCU_READ_LOCK_GUARD(); 1697 1698 caches = vring_get_region_caches(vq); 1699 if (!caches) { 1700 return 0; 1701 } 1702 1703 desc_cache = &caches->desc; 1704 1705 virtio_queue_set_notification(vq, 0); 1706 1707 while (vq->inuse < vq->vring.num) { 1708 unsigned int idx = vq->last_avail_idx; 1709 /* 1710 * works similar to virtqueue_pop but does not map buffers 1711 * and does not allocate any memory. 1712 */ 1713 vring_packed_desc_read(vdev, &desc, desc_cache, 1714 vq->last_avail_idx , true); 1715 if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) { 1716 break; 1717 } 1718 elem.index = desc.id; 1719 elem.ndescs = 1; 1720 while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache, 1721 vq->vring.num, &idx, false)) { 1722 ++elem.ndescs; 1723 } 1724 /* 1725 * immediately push the element, nothing to unmap 1726 * as both in_num and out_num are set to 0. 1727 */ 1728 virtqueue_push(vq, &elem, 0); 1729 dropped++; 1730 vq->last_avail_idx += elem.ndescs; 1731 if (vq->last_avail_idx >= vq->vring.num) { 1732 vq->last_avail_idx -= vq->vring.num; 1733 vq->last_avail_wrap_counter ^= 1; 1734 } 1735 } 1736 1737 return dropped; 1738 } 1739 1740 static unsigned int virtqueue_split_drop_all(VirtQueue *vq) 1741 { 1742 unsigned int dropped = 0; 1743 VirtQueueElement elem = {}; 1744 VirtIODevice *vdev = vq->vdev; 1745 bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); 1746 1747 while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) { 1748 /* works similar to virtqueue_pop but does not map buffers 1749 * and does not allocate any memory */ 1750 smp_rmb(); 1751 if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) { 1752 break; 1753 } 1754 vq->inuse++; 1755 vq->last_avail_idx++; 1756 if (fEventIdx) { 1757 vring_set_avail_event(vq, vq->last_avail_idx); 1758 } 1759 /* immediately push the element, nothing to unmap 1760 * as both in_num and out_num are set to 0 */ 1761 virtqueue_push(vq, &elem, 0); 1762 dropped++; 1763 } 1764 1765 return dropped; 1766 } 1767 1768 /* virtqueue_drop_all: 1769 * @vq: The #VirtQueue 1770 * Drops all queued buffers and indicates them to the guest 1771 * as if they are done. Useful when buffers can not be 1772 * processed but must be returned to the guest. 1773 */ 1774 unsigned int virtqueue_drop_all(VirtQueue *vq) 1775 { 1776 struct VirtIODevice *vdev = vq->vdev; 1777 1778 if (virtio_device_disabled(vq->vdev)) { 1779 return 0; 1780 } 1781 1782 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 1783 return virtqueue_packed_drop_all(vq); 1784 } else { 1785 return virtqueue_split_drop_all(vq); 1786 } 1787 } 1788 1789 /* Reading and writing a structure directly to QEMUFile is *awful*, but 1790 * it is what QEMU has always done by mistake. We can change it sooner 1791 * or later by bumping the version number of the affected vm states. 1792 * In the meanwhile, since the in-memory layout of VirtQueueElement 1793 * has changed, we need to marshal to and from the layout that was 1794 * used before the change. 1795 */ 1796 typedef struct VirtQueueElementOld { 1797 unsigned int index; 1798 unsigned int out_num; 1799 unsigned int in_num; 1800 hwaddr in_addr[VIRTQUEUE_MAX_SIZE]; 1801 hwaddr out_addr[VIRTQUEUE_MAX_SIZE]; 1802 struct iovec in_sg[VIRTQUEUE_MAX_SIZE]; 1803 struct iovec out_sg[VIRTQUEUE_MAX_SIZE]; 1804 } VirtQueueElementOld; 1805 1806 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz) 1807 { 1808 VirtQueueElement *elem; 1809 VirtQueueElementOld data; 1810 int i; 1811 1812 qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld)); 1813 1814 /* TODO: teach all callers that this can fail, and return failure instead 1815 * of asserting here. 1816 * This is just one thing (there are probably more) that must be 1817 * fixed before we can allow NDEBUG compilation. 1818 */ 1819 assert(ARRAY_SIZE(data.in_addr) >= data.in_num); 1820 assert(ARRAY_SIZE(data.out_addr) >= data.out_num); 1821 1822 elem = virtqueue_alloc_element(sz, data.out_num, data.in_num); 1823 elem->index = data.index; 1824 1825 for (i = 0; i < elem->in_num; i++) { 1826 elem->in_addr[i] = data.in_addr[i]; 1827 } 1828 1829 for (i = 0; i < elem->out_num; i++) { 1830 elem->out_addr[i] = data.out_addr[i]; 1831 } 1832 1833 for (i = 0; i < elem->in_num; i++) { 1834 /* Base is overwritten by virtqueue_map. */ 1835 elem->in_sg[i].iov_base = 0; 1836 elem->in_sg[i].iov_len = data.in_sg[i].iov_len; 1837 } 1838 1839 for (i = 0; i < elem->out_num; i++) { 1840 /* Base is overwritten by virtqueue_map. */ 1841 elem->out_sg[i].iov_base = 0; 1842 elem->out_sg[i].iov_len = data.out_sg[i].iov_len; 1843 } 1844 1845 if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 1846 qemu_get_be32s(f, &elem->ndescs); 1847 } 1848 1849 virtqueue_map(vdev, elem); 1850 return elem; 1851 } 1852 1853 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, 1854 VirtQueueElement *elem) 1855 { 1856 VirtQueueElementOld data; 1857 int i; 1858 1859 memset(&data, 0, sizeof(data)); 1860 data.index = elem->index; 1861 data.in_num = elem->in_num; 1862 data.out_num = elem->out_num; 1863 1864 for (i = 0; i < elem->in_num; i++) { 1865 data.in_addr[i] = elem->in_addr[i]; 1866 } 1867 1868 for (i = 0; i < elem->out_num; i++) { 1869 data.out_addr[i] = elem->out_addr[i]; 1870 } 1871 1872 for (i = 0; i < elem->in_num; i++) { 1873 /* Base is overwritten by virtqueue_map when loading. Do not 1874 * save it, as it would leak the QEMU address space layout. */ 1875 data.in_sg[i].iov_len = elem->in_sg[i].iov_len; 1876 } 1877 1878 for (i = 0; i < elem->out_num; i++) { 1879 /* Do not save iov_base as above. */ 1880 data.out_sg[i].iov_len = elem->out_sg[i].iov_len; 1881 } 1882 1883 if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 1884 qemu_put_be32s(f, &elem->ndescs); 1885 } 1886 1887 qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld)); 1888 } 1889 1890 /* virtio device */ 1891 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector) 1892 { 1893 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 1894 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 1895 1896 if (virtio_device_disabled(vdev)) { 1897 return; 1898 } 1899 1900 if (k->notify) { 1901 k->notify(qbus->parent, vector); 1902 } 1903 } 1904 1905 void virtio_update_irq(VirtIODevice *vdev) 1906 { 1907 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR); 1908 } 1909 1910 static int virtio_validate_features(VirtIODevice *vdev) 1911 { 1912 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1913 1914 if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) && 1915 !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) { 1916 return -EFAULT; 1917 } 1918 1919 if (k->validate_features) { 1920 return k->validate_features(vdev); 1921 } else { 1922 return 0; 1923 } 1924 } 1925 1926 int virtio_set_status(VirtIODevice *vdev, uint8_t val) 1927 { 1928 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1929 trace_virtio_set_status(vdev, val); 1930 1931 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 1932 if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) && 1933 val & VIRTIO_CONFIG_S_FEATURES_OK) { 1934 int ret = virtio_validate_features(vdev); 1935 1936 if (ret) { 1937 return ret; 1938 } 1939 } 1940 } 1941 1942 if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) != 1943 (val & VIRTIO_CONFIG_S_DRIVER_OK)) { 1944 virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK); 1945 } 1946 1947 if (k->set_status) { 1948 k->set_status(vdev, val); 1949 } 1950 vdev->status = val; 1951 1952 return 0; 1953 } 1954 1955 static enum virtio_device_endian virtio_default_endian(void) 1956 { 1957 if (target_words_bigendian()) { 1958 return VIRTIO_DEVICE_ENDIAN_BIG; 1959 } else { 1960 return VIRTIO_DEVICE_ENDIAN_LITTLE; 1961 } 1962 } 1963 1964 static enum virtio_device_endian virtio_current_cpu_endian(void) 1965 { 1966 if (cpu_virtio_is_big_endian(current_cpu)) { 1967 return VIRTIO_DEVICE_ENDIAN_BIG; 1968 } else { 1969 return VIRTIO_DEVICE_ENDIAN_LITTLE; 1970 } 1971 } 1972 1973 void virtio_reset(void *opaque) 1974 { 1975 VirtIODevice *vdev = opaque; 1976 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1977 int i; 1978 1979 virtio_set_status(vdev, 0); 1980 if (current_cpu) { 1981 /* Guest initiated reset */ 1982 vdev->device_endian = virtio_current_cpu_endian(); 1983 } else { 1984 /* System reset */ 1985 vdev->device_endian = virtio_default_endian(); 1986 } 1987 1988 if (k->reset) { 1989 k->reset(vdev); 1990 } 1991 1992 vdev->start_on_kick = false; 1993 vdev->started = false; 1994 vdev->broken = false; 1995 vdev->guest_features = 0; 1996 vdev->queue_sel = 0; 1997 vdev->status = 0; 1998 vdev->disabled = false; 1999 qatomic_set(&vdev->isr, 0); 2000 vdev->config_vector = VIRTIO_NO_VECTOR; 2001 virtio_notify_vector(vdev, vdev->config_vector); 2002 2003 for(i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2004 vdev->vq[i].vring.desc = 0; 2005 vdev->vq[i].vring.avail = 0; 2006 vdev->vq[i].vring.used = 0; 2007 vdev->vq[i].last_avail_idx = 0; 2008 vdev->vq[i].shadow_avail_idx = 0; 2009 vdev->vq[i].used_idx = 0; 2010 vdev->vq[i].last_avail_wrap_counter = true; 2011 vdev->vq[i].shadow_avail_wrap_counter = true; 2012 vdev->vq[i].used_wrap_counter = true; 2013 virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR); 2014 vdev->vq[i].signalled_used = 0; 2015 vdev->vq[i].signalled_used_valid = false; 2016 vdev->vq[i].notification = true; 2017 vdev->vq[i].vring.num = vdev->vq[i].vring.num_default; 2018 vdev->vq[i].inuse = 0; 2019 virtio_virtqueue_reset_region_cache(&vdev->vq[i]); 2020 } 2021 } 2022 2023 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr) 2024 { 2025 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2026 uint8_t val; 2027 2028 if (addr + sizeof(val) > vdev->config_len) { 2029 return (uint32_t)-1; 2030 } 2031 2032 k->get_config(vdev, vdev->config); 2033 2034 val = ldub_p(vdev->config + addr); 2035 return val; 2036 } 2037 2038 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr) 2039 { 2040 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2041 uint16_t val; 2042 2043 if (addr + sizeof(val) > vdev->config_len) { 2044 return (uint32_t)-1; 2045 } 2046 2047 k->get_config(vdev, vdev->config); 2048 2049 val = lduw_p(vdev->config + addr); 2050 return val; 2051 } 2052 2053 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr) 2054 { 2055 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2056 uint32_t val; 2057 2058 if (addr + sizeof(val) > vdev->config_len) { 2059 return (uint32_t)-1; 2060 } 2061 2062 k->get_config(vdev, vdev->config); 2063 2064 val = ldl_p(vdev->config + addr); 2065 return val; 2066 } 2067 2068 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data) 2069 { 2070 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2071 uint8_t val = data; 2072 2073 if (addr + sizeof(val) > vdev->config_len) { 2074 return; 2075 } 2076 2077 stb_p(vdev->config + addr, val); 2078 2079 if (k->set_config) { 2080 k->set_config(vdev, vdev->config); 2081 } 2082 } 2083 2084 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data) 2085 { 2086 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2087 uint16_t val = data; 2088 2089 if (addr + sizeof(val) > vdev->config_len) { 2090 return; 2091 } 2092 2093 stw_p(vdev->config + addr, val); 2094 2095 if (k->set_config) { 2096 k->set_config(vdev, vdev->config); 2097 } 2098 } 2099 2100 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data) 2101 { 2102 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2103 uint32_t val = data; 2104 2105 if (addr + sizeof(val) > vdev->config_len) { 2106 return; 2107 } 2108 2109 stl_p(vdev->config + addr, val); 2110 2111 if (k->set_config) { 2112 k->set_config(vdev, vdev->config); 2113 } 2114 } 2115 2116 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr) 2117 { 2118 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2119 uint8_t val; 2120 2121 if (addr + sizeof(val) > vdev->config_len) { 2122 return (uint32_t)-1; 2123 } 2124 2125 k->get_config(vdev, vdev->config); 2126 2127 val = ldub_p(vdev->config + addr); 2128 return val; 2129 } 2130 2131 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr) 2132 { 2133 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2134 uint16_t val; 2135 2136 if (addr + sizeof(val) > vdev->config_len) { 2137 return (uint32_t)-1; 2138 } 2139 2140 k->get_config(vdev, vdev->config); 2141 2142 val = lduw_le_p(vdev->config + addr); 2143 return val; 2144 } 2145 2146 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr) 2147 { 2148 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2149 uint32_t val; 2150 2151 if (addr + sizeof(val) > vdev->config_len) { 2152 return (uint32_t)-1; 2153 } 2154 2155 k->get_config(vdev, vdev->config); 2156 2157 val = ldl_le_p(vdev->config + addr); 2158 return val; 2159 } 2160 2161 void virtio_config_modern_writeb(VirtIODevice *vdev, 2162 uint32_t addr, uint32_t data) 2163 { 2164 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2165 uint8_t val = data; 2166 2167 if (addr + sizeof(val) > vdev->config_len) { 2168 return; 2169 } 2170 2171 stb_p(vdev->config + addr, val); 2172 2173 if (k->set_config) { 2174 k->set_config(vdev, vdev->config); 2175 } 2176 } 2177 2178 void virtio_config_modern_writew(VirtIODevice *vdev, 2179 uint32_t addr, uint32_t data) 2180 { 2181 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2182 uint16_t val = data; 2183 2184 if (addr + sizeof(val) > vdev->config_len) { 2185 return; 2186 } 2187 2188 stw_le_p(vdev->config + addr, val); 2189 2190 if (k->set_config) { 2191 k->set_config(vdev, vdev->config); 2192 } 2193 } 2194 2195 void virtio_config_modern_writel(VirtIODevice *vdev, 2196 uint32_t addr, uint32_t data) 2197 { 2198 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2199 uint32_t val = data; 2200 2201 if (addr + sizeof(val) > vdev->config_len) { 2202 return; 2203 } 2204 2205 stl_le_p(vdev->config + addr, val); 2206 2207 if (k->set_config) { 2208 k->set_config(vdev, vdev->config); 2209 } 2210 } 2211 2212 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr) 2213 { 2214 if (!vdev->vq[n].vring.num) { 2215 return; 2216 } 2217 vdev->vq[n].vring.desc = addr; 2218 virtio_queue_update_rings(vdev, n); 2219 } 2220 2221 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n) 2222 { 2223 return vdev->vq[n].vring.desc; 2224 } 2225 2226 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc, 2227 hwaddr avail, hwaddr used) 2228 { 2229 if (!vdev->vq[n].vring.num) { 2230 return; 2231 } 2232 vdev->vq[n].vring.desc = desc; 2233 vdev->vq[n].vring.avail = avail; 2234 vdev->vq[n].vring.used = used; 2235 virtio_init_region_cache(vdev, n); 2236 } 2237 2238 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num) 2239 { 2240 /* Don't allow guest to flip queue between existent and 2241 * nonexistent states, or to set it to an invalid size. 2242 */ 2243 if (!!num != !!vdev->vq[n].vring.num || 2244 num > VIRTQUEUE_MAX_SIZE || 2245 num < 0) { 2246 return; 2247 } 2248 vdev->vq[n].vring.num = num; 2249 } 2250 2251 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector) 2252 { 2253 return QLIST_FIRST(&vdev->vector_queues[vector]); 2254 } 2255 2256 VirtQueue *virtio_vector_next_queue(VirtQueue *vq) 2257 { 2258 return QLIST_NEXT(vq, node); 2259 } 2260 2261 int virtio_queue_get_num(VirtIODevice *vdev, int n) 2262 { 2263 return vdev->vq[n].vring.num; 2264 } 2265 2266 int virtio_queue_get_max_num(VirtIODevice *vdev, int n) 2267 { 2268 return vdev->vq[n].vring.num_default; 2269 } 2270 2271 int virtio_get_num_queues(VirtIODevice *vdev) 2272 { 2273 int i; 2274 2275 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2276 if (!virtio_queue_get_num(vdev, i)) { 2277 break; 2278 } 2279 } 2280 2281 return i; 2282 } 2283 2284 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align) 2285 { 2286 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2287 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2288 2289 /* virtio-1 compliant devices cannot change the alignment */ 2290 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 2291 error_report("tried to modify queue alignment for virtio-1 device"); 2292 return; 2293 } 2294 /* Check that the transport told us it was going to do this 2295 * (so a buggy transport will immediately assert rather than 2296 * silently failing to migrate this state) 2297 */ 2298 assert(k->has_variable_vring_alignment); 2299 2300 if (align) { 2301 vdev->vq[n].vring.align = align; 2302 virtio_queue_update_rings(vdev, n); 2303 } 2304 } 2305 2306 static void virtio_queue_notify_aio_vq(VirtQueue *vq) 2307 { 2308 if (vq->vring.desc && vq->handle_aio_output) { 2309 VirtIODevice *vdev = vq->vdev; 2310 2311 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); 2312 vq->handle_aio_output(vdev, vq); 2313 2314 if (unlikely(vdev->start_on_kick)) { 2315 virtio_set_started(vdev, true); 2316 } 2317 } 2318 } 2319 2320 static void virtio_queue_notify_vq(VirtQueue *vq) 2321 { 2322 if (vq->vring.desc && vq->handle_output) { 2323 VirtIODevice *vdev = vq->vdev; 2324 2325 if (unlikely(vdev->broken)) { 2326 return; 2327 } 2328 2329 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); 2330 vq->handle_output(vdev, vq); 2331 2332 if (unlikely(vdev->start_on_kick)) { 2333 virtio_set_started(vdev, true); 2334 } 2335 } 2336 } 2337 2338 void virtio_queue_notify(VirtIODevice *vdev, int n) 2339 { 2340 VirtQueue *vq = &vdev->vq[n]; 2341 2342 if (unlikely(!vq->vring.desc || vdev->broken)) { 2343 return; 2344 } 2345 2346 trace_virtio_queue_notify(vdev, vq - vdev->vq, vq); 2347 if (vq->host_notifier_enabled) { 2348 event_notifier_set(&vq->host_notifier); 2349 } else if (vq->handle_output) { 2350 vq->handle_output(vdev, vq); 2351 2352 if (unlikely(vdev->start_on_kick)) { 2353 virtio_set_started(vdev, true); 2354 } 2355 } 2356 } 2357 2358 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n) 2359 { 2360 return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector : 2361 VIRTIO_NO_VECTOR; 2362 } 2363 2364 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector) 2365 { 2366 VirtQueue *vq = &vdev->vq[n]; 2367 2368 if (n < VIRTIO_QUEUE_MAX) { 2369 if (vdev->vector_queues && 2370 vdev->vq[n].vector != VIRTIO_NO_VECTOR) { 2371 QLIST_REMOVE(vq, node); 2372 } 2373 vdev->vq[n].vector = vector; 2374 if (vdev->vector_queues && 2375 vector != VIRTIO_NO_VECTOR) { 2376 QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node); 2377 } 2378 } 2379 } 2380 2381 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size, 2382 VirtIOHandleOutput handle_output) 2383 { 2384 int i; 2385 2386 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2387 if (vdev->vq[i].vring.num == 0) 2388 break; 2389 } 2390 2391 if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE) 2392 abort(); 2393 2394 vdev->vq[i].vring.num = queue_size; 2395 vdev->vq[i].vring.num_default = queue_size; 2396 vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN; 2397 vdev->vq[i].handle_output = handle_output; 2398 vdev->vq[i].handle_aio_output = NULL; 2399 vdev->vq[i].used_elems = g_malloc0(sizeof(VirtQueueElement) * 2400 queue_size); 2401 2402 return &vdev->vq[i]; 2403 } 2404 2405 void virtio_delete_queue(VirtQueue *vq) 2406 { 2407 vq->vring.num = 0; 2408 vq->vring.num_default = 0; 2409 vq->handle_output = NULL; 2410 vq->handle_aio_output = NULL; 2411 g_free(vq->used_elems); 2412 vq->used_elems = NULL; 2413 virtio_virtqueue_reset_region_cache(vq); 2414 } 2415 2416 void virtio_del_queue(VirtIODevice *vdev, int n) 2417 { 2418 if (n < 0 || n >= VIRTIO_QUEUE_MAX) { 2419 abort(); 2420 } 2421 2422 virtio_delete_queue(&vdev->vq[n]); 2423 } 2424 2425 static void virtio_set_isr(VirtIODevice *vdev, int value) 2426 { 2427 uint8_t old = qatomic_read(&vdev->isr); 2428 2429 /* Do not write ISR if it does not change, so that its cacheline remains 2430 * shared in the common case where the guest does not read it. 2431 */ 2432 if ((old & value) != value) { 2433 qatomic_or(&vdev->isr, value); 2434 } 2435 } 2436 2437 /* Called within rcu_read_lock(). */ 2438 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq) 2439 { 2440 uint16_t old, new; 2441 bool v; 2442 /* We need to expose used array entries before checking used event. */ 2443 smp_mb(); 2444 /* Always notify when queue is empty (when feature acknowledge) */ 2445 if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) && 2446 !vq->inuse && virtio_queue_empty(vq)) { 2447 return true; 2448 } 2449 2450 if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { 2451 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT); 2452 } 2453 2454 v = vq->signalled_used_valid; 2455 vq->signalled_used_valid = true; 2456 old = vq->signalled_used; 2457 new = vq->signalled_used = vq->used_idx; 2458 return !v || vring_need_event(vring_get_used_event(vq), new, old); 2459 } 2460 2461 static bool vring_packed_need_event(VirtQueue *vq, bool wrap, 2462 uint16_t off_wrap, uint16_t new, 2463 uint16_t old) 2464 { 2465 int off = off_wrap & ~(1 << 15); 2466 2467 if (wrap != off_wrap >> 15) { 2468 off -= vq->vring.num; 2469 } 2470 2471 return vring_need_event(off, new, old); 2472 } 2473 2474 /* Called within rcu_read_lock(). */ 2475 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq) 2476 { 2477 VRingPackedDescEvent e; 2478 uint16_t old, new; 2479 bool v; 2480 VRingMemoryRegionCaches *caches; 2481 2482 caches = vring_get_region_caches(vq); 2483 if (!caches) { 2484 return false; 2485 } 2486 2487 vring_packed_event_read(vdev, &caches->avail, &e); 2488 2489 old = vq->signalled_used; 2490 new = vq->signalled_used = vq->used_idx; 2491 v = vq->signalled_used_valid; 2492 vq->signalled_used_valid = true; 2493 2494 if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) { 2495 return false; 2496 } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) { 2497 return true; 2498 } 2499 2500 return !v || vring_packed_need_event(vq, vq->used_wrap_counter, 2501 e.off_wrap, new, old); 2502 } 2503 2504 /* Called within rcu_read_lock(). */ 2505 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq) 2506 { 2507 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 2508 return virtio_packed_should_notify(vdev, vq); 2509 } else { 2510 return virtio_split_should_notify(vdev, vq); 2511 } 2512 } 2513 2514 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq) 2515 { 2516 WITH_RCU_READ_LOCK_GUARD() { 2517 if (!virtio_should_notify(vdev, vq)) { 2518 return; 2519 } 2520 } 2521 2522 trace_virtio_notify_irqfd(vdev, vq); 2523 2524 /* 2525 * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but 2526 * windows drivers included in virtio-win 1.8.0 (circa 2015) are 2527 * incorrectly polling this bit during crashdump and hibernation 2528 * in MSI mode, causing a hang if this bit is never updated. 2529 * Recent releases of Windows do not really shut down, but rather 2530 * log out and hibernate to make the next startup faster. Hence, 2531 * this manifested as a more serious hang during shutdown with 2532 * 2533 * Next driver release from 2016 fixed this problem, so working around it 2534 * is not a must, but it's easy to do so let's do it here. 2535 * 2536 * Note: it's safe to update ISR from any thread as it was switched 2537 * to an atomic operation. 2538 */ 2539 virtio_set_isr(vq->vdev, 0x1); 2540 event_notifier_set(&vq->guest_notifier); 2541 } 2542 2543 static void virtio_irq(VirtQueue *vq) 2544 { 2545 virtio_set_isr(vq->vdev, 0x1); 2546 virtio_notify_vector(vq->vdev, vq->vector); 2547 } 2548 2549 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq) 2550 { 2551 WITH_RCU_READ_LOCK_GUARD() { 2552 if (!virtio_should_notify(vdev, vq)) { 2553 return; 2554 } 2555 } 2556 2557 trace_virtio_notify(vdev, vq); 2558 virtio_irq(vq); 2559 } 2560 2561 void virtio_notify_config(VirtIODevice *vdev) 2562 { 2563 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) 2564 return; 2565 2566 virtio_set_isr(vdev, 0x3); 2567 vdev->generation++; 2568 virtio_notify_vector(vdev, vdev->config_vector); 2569 } 2570 2571 static bool virtio_device_endian_needed(void *opaque) 2572 { 2573 VirtIODevice *vdev = opaque; 2574 2575 assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN); 2576 if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 2577 return vdev->device_endian != virtio_default_endian(); 2578 } 2579 /* Devices conforming to VIRTIO 1.0 or later are always LE. */ 2580 return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE; 2581 } 2582 2583 static bool virtio_64bit_features_needed(void *opaque) 2584 { 2585 VirtIODevice *vdev = opaque; 2586 2587 return (vdev->host_features >> 32) != 0; 2588 } 2589 2590 static bool virtio_virtqueue_needed(void *opaque) 2591 { 2592 VirtIODevice *vdev = opaque; 2593 2594 return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1); 2595 } 2596 2597 static bool virtio_packed_virtqueue_needed(void *opaque) 2598 { 2599 VirtIODevice *vdev = opaque; 2600 2601 return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED); 2602 } 2603 2604 static bool virtio_ringsize_needed(void *opaque) 2605 { 2606 VirtIODevice *vdev = opaque; 2607 int i; 2608 2609 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2610 if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) { 2611 return true; 2612 } 2613 } 2614 return false; 2615 } 2616 2617 static bool virtio_extra_state_needed(void *opaque) 2618 { 2619 VirtIODevice *vdev = opaque; 2620 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2621 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2622 2623 return k->has_extra_state && 2624 k->has_extra_state(qbus->parent); 2625 } 2626 2627 static bool virtio_broken_needed(void *opaque) 2628 { 2629 VirtIODevice *vdev = opaque; 2630 2631 return vdev->broken; 2632 } 2633 2634 static bool virtio_started_needed(void *opaque) 2635 { 2636 VirtIODevice *vdev = opaque; 2637 2638 return vdev->started; 2639 } 2640 2641 static bool virtio_disabled_needed(void *opaque) 2642 { 2643 VirtIODevice *vdev = opaque; 2644 2645 return vdev->disabled; 2646 } 2647 2648 static const VMStateDescription vmstate_virtqueue = { 2649 .name = "virtqueue_state", 2650 .version_id = 1, 2651 .minimum_version_id = 1, 2652 .fields = (VMStateField[]) { 2653 VMSTATE_UINT64(vring.avail, struct VirtQueue), 2654 VMSTATE_UINT64(vring.used, struct VirtQueue), 2655 VMSTATE_END_OF_LIST() 2656 } 2657 }; 2658 2659 static const VMStateDescription vmstate_packed_virtqueue = { 2660 .name = "packed_virtqueue_state", 2661 .version_id = 1, 2662 .minimum_version_id = 1, 2663 .fields = (VMStateField[]) { 2664 VMSTATE_UINT16(last_avail_idx, struct VirtQueue), 2665 VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue), 2666 VMSTATE_UINT16(used_idx, struct VirtQueue), 2667 VMSTATE_BOOL(used_wrap_counter, struct VirtQueue), 2668 VMSTATE_UINT32(inuse, struct VirtQueue), 2669 VMSTATE_END_OF_LIST() 2670 } 2671 }; 2672 2673 static const VMStateDescription vmstate_virtio_virtqueues = { 2674 .name = "virtio/virtqueues", 2675 .version_id = 1, 2676 .minimum_version_id = 1, 2677 .needed = &virtio_virtqueue_needed, 2678 .fields = (VMStateField[]) { 2679 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice, 2680 VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue), 2681 VMSTATE_END_OF_LIST() 2682 } 2683 }; 2684 2685 static const VMStateDescription vmstate_virtio_packed_virtqueues = { 2686 .name = "virtio/packed_virtqueues", 2687 .version_id = 1, 2688 .minimum_version_id = 1, 2689 .needed = &virtio_packed_virtqueue_needed, 2690 .fields = (VMStateField[]) { 2691 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice, 2692 VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue), 2693 VMSTATE_END_OF_LIST() 2694 } 2695 }; 2696 2697 static const VMStateDescription vmstate_ringsize = { 2698 .name = "ringsize_state", 2699 .version_id = 1, 2700 .minimum_version_id = 1, 2701 .fields = (VMStateField[]) { 2702 VMSTATE_UINT32(vring.num_default, struct VirtQueue), 2703 VMSTATE_END_OF_LIST() 2704 } 2705 }; 2706 2707 static const VMStateDescription vmstate_virtio_ringsize = { 2708 .name = "virtio/ringsize", 2709 .version_id = 1, 2710 .minimum_version_id = 1, 2711 .needed = &virtio_ringsize_needed, 2712 .fields = (VMStateField[]) { 2713 VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice, 2714 VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue), 2715 VMSTATE_END_OF_LIST() 2716 } 2717 }; 2718 2719 static int get_extra_state(QEMUFile *f, void *pv, size_t size, 2720 const VMStateField *field) 2721 { 2722 VirtIODevice *vdev = pv; 2723 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2724 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2725 2726 if (!k->load_extra_state) { 2727 return -1; 2728 } else { 2729 return k->load_extra_state(qbus->parent, f); 2730 } 2731 } 2732 2733 static int put_extra_state(QEMUFile *f, void *pv, size_t size, 2734 const VMStateField *field, JSONWriter *vmdesc) 2735 { 2736 VirtIODevice *vdev = pv; 2737 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2738 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2739 2740 k->save_extra_state(qbus->parent, f); 2741 return 0; 2742 } 2743 2744 static const VMStateInfo vmstate_info_extra_state = { 2745 .name = "virtqueue_extra_state", 2746 .get = get_extra_state, 2747 .put = put_extra_state, 2748 }; 2749 2750 static const VMStateDescription vmstate_virtio_extra_state = { 2751 .name = "virtio/extra_state", 2752 .version_id = 1, 2753 .minimum_version_id = 1, 2754 .needed = &virtio_extra_state_needed, 2755 .fields = (VMStateField[]) { 2756 { 2757 .name = "extra_state", 2758 .version_id = 0, 2759 .field_exists = NULL, 2760 .size = 0, 2761 .info = &vmstate_info_extra_state, 2762 .flags = VMS_SINGLE, 2763 .offset = 0, 2764 }, 2765 VMSTATE_END_OF_LIST() 2766 } 2767 }; 2768 2769 static const VMStateDescription vmstate_virtio_device_endian = { 2770 .name = "virtio/device_endian", 2771 .version_id = 1, 2772 .minimum_version_id = 1, 2773 .needed = &virtio_device_endian_needed, 2774 .fields = (VMStateField[]) { 2775 VMSTATE_UINT8(device_endian, VirtIODevice), 2776 VMSTATE_END_OF_LIST() 2777 } 2778 }; 2779 2780 static const VMStateDescription vmstate_virtio_64bit_features = { 2781 .name = "virtio/64bit_features", 2782 .version_id = 1, 2783 .minimum_version_id = 1, 2784 .needed = &virtio_64bit_features_needed, 2785 .fields = (VMStateField[]) { 2786 VMSTATE_UINT64(guest_features, VirtIODevice), 2787 VMSTATE_END_OF_LIST() 2788 } 2789 }; 2790 2791 static const VMStateDescription vmstate_virtio_broken = { 2792 .name = "virtio/broken", 2793 .version_id = 1, 2794 .minimum_version_id = 1, 2795 .needed = &virtio_broken_needed, 2796 .fields = (VMStateField[]) { 2797 VMSTATE_BOOL(broken, VirtIODevice), 2798 VMSTATE_END_OF_LIST() 2799 } 2800 }; 2801 2802 static const VMStateDescription vmstate_virtio_started = { 2803 .name = "virtio/started", 2804 .version_id = 1, 2805 .minimum_version_id = 1, 2806 .needed = &virtio_started_needed, 2807 .fields = (VMStateField[]) { 2808 VMSTATE_BOOL(started, VirtIODevice), 2809 VMSTATE_END_OF_LIST() 2810 } 2811 }; 2812 2813 static const VMStateDescription vmstate_virtio_disabled = { 2814 .name = "virtio/disabled", 2815 .version_id = 1, 2816 .minimum_version_id = 1, 2817 .needed = &virtio_disabled_needed, 2818 .fields = (VMStateField[]) { 2819 VMSTATE_BOOL(disabled, VirtIODevice), 2820 VMSTATE_END_OF_LIST() 2821 } 2822 }; 2823 2824 static const VMStateDescription vmstate_virtio = { 2825 .name = "virtio", 2826 .version_id = 1, 2827 .minimum_version_id = 1, 2828 .minimum_version_id_old = 1, 2829 .fields = (VMStateField[]) { 2830 VMSTATE_END_OF_LIST() 2831 }, 2832 .subsections = (const VMStateDescription*[]) { 2833 &vmstate_virtio_device_endian, 2834 &vmstate_virtio_64bit_features, 2835 &vmstate_virtio_virtqueues, 2836 &vmstate_virtio_ringsize, 2837 &vmstate_virtio_broken, 2838 &vmstate_virtio_extra_state, 2839 &vmstate_virtio_started, 2840 &vmstate_virtio_packed_virtqueues, 2841 &vmstate_virtio_disabled, 2842 NULL 2843 } 2844 }; 2845 2846 int virtio_save(VirtIODevice *vdev, QEMUFile *f) 2847 { 2848 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2849 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2850 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 2851 uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff); 2852 int i; 2853 2854 if (k->save_config) { 2855 k->save_config(qbus->parent, f); 2856 } 2857 2858 qemu_put_8s(f, &vdev->status); 2859 qemu_put_8s(f, &vdev->isr); 2860 qemu_put_be16s(f, &vdev->queue_sel); 2861 qemu_put_be32s(f, &guest_features_lo); 2862 qemu_put_be32(f, vdev->config_len); 2863 qemu_put_buffer(f, vdev->config, vdev->config_len); 2864 2865 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2866 if (vdev->vq[i].vring.num == 0) 2867 break; 2868 } 2869 2870 qemu_put_be32(f, i); 2871 2872 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2873 if (vdev->vq[i].vring.num == 0) 2874 break; 2875 2876 qemu_put_be32(f, vdev->vq[i].vring.num); 2877 if (k->has_variable_vring_alignment) { 2878 qemu_put_be32(f, vdev->vq[i].vring.align); 2879 } 2880 /* 2881 * Save desc now, the rest of the ring addresses are saved in 2882 * subsections for VIRTIO-1 devices. 2883 */ 2884 qemu_put_be64(f, vdev->vq[i].vring.desc); 2885 qemu_put_be16s(f, &vdev->vq[i].last_avail_idx); 2886 if (k->save_queue) { 2887 k->save_queue(qbus->parent, i, f); 2888 } 2889 } 2890 2891 if (vdc->save != NULL) { 2892 vdc->save(vdev, f); 2893 } 2894 2895 if (vdc->vmsd) { 2896 int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL); 2897 if (ret) { 2898 return ret; 2899 } 2900 } 2901 2902 /* Subsections */ 2903 return vmstate_save_state(f, &vmstate_virtio, vdev, NULL); 2904 } 2905 2906 /* A wrapper for use as a VMState .put function */ 2907 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size, 2908 const VMStateField *field, JSONWriter *vmdesc) 2909 { 2910 return virtio_save(VIRTIO_DEVICE(opaque), f); 2911 } 2912 2913 /* A wrapper for use as a VMState .get function */ 2914 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size, 2915 const VMStateField *field) 2916 { 2917 VirtIODevice *vdev = VIRTIO_DEVICE(opaque); 2918 DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev)); 2919 2920 return virtio_load(vdev, f, dc->vmsd->version_id); 2921 } 2922 2923 const VMStateInfo virtio_vmstate_info = { 2924 .name = "virtio", 2925 .get = virtio_device_get, 2926 .put = virtio_device_put, 2927 }; 2928 2929 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val) 2930 { 2931 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 2932 bool bad = (val & ~(vdev->host_features)) != 0; 2933 2934 val &= vdev->host_features; 2935 if (k->set_features) { 2936 k->set_features(vdev, val); 2937 } 2938 vdev->guest_features = val; 2939 return bad ? -1 : 0; 2940 } 2941 2942 int virtio_set_features(VirtIODevice *vdev, uint64_t val) 2943 { 2944 int ret; 2945 /* 2946 * The driver must not attempt to set features after feature negotiation 2947 * has finished. 2948 */ 2949 if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) { 2950 return -EINVAL; 2951 } 2952 ret = virtio_set_features_nocheck(vdev, val); 2953 if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) { 2954 /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches. */ 2955 int i; 2956 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2957 if (vdev->vq[i].vring.num != 0) { 2958 virtio_init_region_cache(vdev, i); 2959 } 2960 } 2961 } 2962 if (!ret) { 2963 if (!virtio_device_started(vdev, vdev->status) && 2964 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 2965 vdev->start_on_kick = true; 2966 } 2967 } 2968 return ret; 2969 } 2970 2971 size_t virtio_feature_get_config_size(const VirtIOFeature *feature_sizes, 2972 uint64_t host_features) 2973 { 2974 size_t config_size = 0; 2975 int i; 2976 2977 for (i = 0; feature_sizes[i].flags != 0; i++) { 2978 if (host_features & feature_sizes[i].flags) { 2979 config_size = MAX(feature_sizes[i].end, config_size); 2980 } 2981 } 2982 2983 return config_size; 2984 } 2985 2986 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) 2987 { 2988 int i, ret; 2989 int32_t config_len; 2990 uint32_t num; 2991 uint32_t features; 2992 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 2993 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 2994 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 2995 2996 /* 2997 * We poison the endianness to ensure it does not get used before 2998 * subsections have been loaded. 2999 */ 3000 vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN; 3001 3002 if (k->load_config) { 3003 ret = k->load_config(qbus->parent, f); 3004 if (ret) 3005 return ret; 3006 } 3007 3008 qemu_get_8s(f, &vdev->status); 3009 qemu_get_8s(f, &vdev->isr); 3010 qemu_get_be16s(f, &vdev->queue_sel); 3011 if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) { 3012 return -1; 3013 } 3014 qemu_get_be32s(f, &features); 3015 3016 /* 3017 * Temporarily set guest_features low bits - needed by 3018 * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS 3019 * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ. 3020 * 3021 * Note: devices should always test host features in future - don't create 3022 * new dependencies like this. 3023 */ 3024 vdev->guest_features = features; 3025 3026 config_len = qemu_get_be32(f); 3027 3028 /* 3029 * There are cases where the incoming config can be bigger or smaller 3030 * than what we have; so load what we have space for, and skip 3031 * any excess that's in the stream. 3032 */ 3033 qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len)); 3034 3035 while (config_len > vdev->config_len) { 3036 qemu_get_byte(f); 3037 config_len--; 3038 } 3039 3040 num = qemu_get_be32(f); 3041 3042 if (num > VIRTIO_QUEUE_MAX) { 3043 error_report("Invalid number of virtqueues: 0x%x", num); 3044 return -1; 3045 } 3046 3047 for (i = 0; i < num; i++) { 3048 vdev->vq[i].vring.num = qemu_get_be32(f); 3049 if (k->has_variable_vring_alignment) { 3050 vdev->vq[i].vring.align = qemu_get_be32(f); 3051 } 3052 vdev->vq[i].vring.desc = qemu_get_be64(f); 3053 qemu_get_be16s(f, &vdev->vq[i].last_avail_idx); 3054 vdev->vq[i].signalled_used_valid = false; 3055 vdev->vq[i].notification = true; 3056 3057 if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) { 3058 error_report("VQ %d address 0x0 " 3059 "inconsistent with Host index 0x%x", 3060 i, vdev->vq[i].last_avail_idx); 3061 return -1; 3062 } 3063 if (k->load_queue) { 3064 ret = k->load_queue(qbus->parent, i, f); 3065 if (ret) 3066 return ret; 3067 } 3068 } 3069 3070 virtio_notify_vector(vdev, VIRTIO_NO_VECTOR); 3071 3072 if (vdc->load != NULL) { 3073 ret = vdc->load(vdev, f, version_id); 3074 if (ret) { 3075 return ret; 3076 } 3077 } 3078 3079 if (vdc->vmsd) { 3080 ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id); 3081 if (ret) { 3082 return ret; 3083 } 3084 } 3085 3086 /* Subsections */ 3087 ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1); 3088 if (ret) { 3089 return ret; 3090 } 3091 3092 if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) { 3093 vdev->device_endian = virtio_default_endian(); 3094 } 3095 3096 if (virtio_64bit_features_needed(vdev)) { 3097 /* 3098 * Subsection load filled vdev->guest_features. Run them 3099 * through virtio_set_features to sanity-check them against 3100 * host_features. 3101 */ 3102 uint64_t features64 = vdev->guest_features; 3103 if (virtio_set_features_nocheck(vdev, features64) < 0) { 3104 error_report("Features 0x%" PRIx64 " unsupported. " 3105 "Allowed features: 0x%" PRIx64, 3106 features64, vdev->host_features); 3107 return -1; 3108 } 3109 } else { 3110 if (virtio_set_features_nocheck(vdev, features) < 0) { 3111 error_report("Features 0x%x unsupported. " 3112 "Allowed features: 0x%" PRIx64, 3113 features, vdev->host_features); 3114 return -1; 3115 } 3116 } 3117 3118 if (!virtio_device_started(vdev, vdev->status) && 3119 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 3120 vdev->start_on_kick = true; 3121 } 3122 3123 RCU_READ_LOCK_GUARD(); 3124 for (i = 0; i < num; i++) { 3125 if (vdev->vq[i].vring.desc) { 3126 uint16_t nheads; 3127 3128 /* 3129 * VIRTIO-1 devices migrate desc, used, and avail ring addresses so 3130 * only the region cache needs to be set up. Legacy devices need 3131 * to calculate used and avail ring addresses based on the desc 3132 * address. 3133 */ 3134 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 3135 virtio_init_region_cache(vdev, i); 3136 } else { 3137 virtio_queue_update_rings(vdev, i); 3138 } 3139 3140 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 3141 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx; 3142 vdev->vq[i].shadow_avail_wrap_counter = 3143 vdev->vq[i].last_avail_wrap_counter; 3144 continue; 3145 } 3146 3147 nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx; 3148 /* Check it isn't doing strange things with descriptor numbers. */ 3149 if (nheads > vdev->vq[i].vring.num) { 3150 virtio_error(vdev, "VQ %d size 0x%x Guest index 0x%x " 3151 "inconsistent with Host index 0x%x: delta 0x%x", 3152 i, vdev->vq[i].vring.num, 3153 vring_avail_idx(&vdev->vq[i]), 3154 vdev->vq[i].last_avail_idx, nheads); 3155 vdev->vq[i].used_idx = 0; 3156 vdev->vq[i].shadow_avail_idx = 0; 3157 vdev->vq[i].inuse = 0; 3158 continue; 3159 } 3160 vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]); 3161 vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]); 3162 3163 /* 3164 * Some devices migrate VirtQueueElements that have been popped 3165 * from the avail ring but not yet returned to the used ring. 3166 * Since max ring size < UINT16_MAX it's safe to use modulo 3167 * UINT16_MAX + 1 subtraction. 3168 */ 3169 vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx - 3170 vdev->vq[i].used_idx); 3171 if (vdev->vq[i].inuse > vdev->vq[i].vring.num) { 3172 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - " 3173 "used_idx 0x%x", 3174 i, vdev->vq[i].vring.num, 3175 vdev->vq[i].last_avail_idx, 3176 vdev->vq[i].used_idx); 3177 return -1; 3178 } 3179 } 3180 } 3181 3182 if (vdc->post_load) { 3183 ret = vdc->post_load(vdev); 3184 if (ret) { 3185 return ret; 3186 } 3187 } 3188 3189 return 0; 3190 } 3191 3192 void virtio_cleanup(VirtIODevice *vdev) 3193 { 3194 qemu_del_vm_change_state_handler(vdev->vmstate); 3195 } 3196 3197 static void virtio_vmstate_change(void *opaque, bool running, RunState state) 3198 { 3199 VirtIODevice *vdev = opaque; 3200 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3201 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 3202 bool backend_run = running && virtio_device_started(vdev, vdev->status); 3203 vdev->vm_running = running; 3204 3205 if (backend_run) { 3206 virtio_set_status(vdev, vdev->status); 3207 } 3208 3209 if (k->vmstate_change) { 3210 k->vmstate_change(qbus->parent, backend_run); 3211 } 3212 3213 if (!backend_run) { 3214 virtio_set_status(vdev, vdev->status); 3215 } 3216 } 3217 3218 void virtio_instance_init_common(Object *proxy_obj, void *data, 3219 size_t vdev_size, const char *vdev_name) 3220 { 3221 DeviceState *vdev = data; 3222 3223 object_initialize_child_with_props(proxy_obj, "virtio-backend", vdev, 3224 vdev_size, vdev_name, &error_abort, 3225 NULL); 3226 qdev_alias_all_properties(vdev, proxy_obj); 3227 } 3228 3229 void virtio_init(VirtIODevice *vdev, const char *name, 3230 uint16_t device_id, size_t config_size) 3231 { 3232 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3233 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 3234 int i; 3235 int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0; 3236 3237 if (nvectors) { 3238 vdev->vector_queues = 3239 g_malloc0(sizeof(*vdev->vector_queues) * nvectors); 3240 } 3241 3242 vdev->start_on_kick = false; 3243 vdev->started = false; 3244 vdev->device_id = device_id; 3245 vdev->status = 0; 3246 qatomic_set(&vdev->isr, 0); 3247 vdev->queue_sel = 0; 3248 vdev->config_vector = VIRTIO_NO_VECTOR; 3249 vdev->vq = g_malloc0(sizeof(VirtQueue) * VIRTIO_QUEUE_MAX); 3250 vdev->vm_running = runstate_is_running(); 3251 vdev->broken = false; 3252 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 3253 vdev->vq[i].vector = VIRTIO_NO_VECTOR; 3254 vdev->vq[i].vdev = vdev; 3255 vdev->vq[i].queue_index = i; 3256 vdev->vq[i].host_notifier_enabled = false; 3257 } 3258 3259 vdev->name = name; 3260 vdev->config_len = config_size; 3261 if (vdev->config_len) { 3262 vdev->config = g_malloc0(config_size); 3263 } else { 3264 vdev->config = NULL; 3265 } 3266 vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev), 3267 virtio_vmstate_change, vdev); 3268 vdev->device_endian = virtio_default_endian(); 3269 vdev->use_guest_notifier_mask = true; 3270 } 3271 3272 /* 3273 * Only devices that have already been around prior to defining the virtio 3274 * standard support legacy mode; this includes devices not specified in the 3275 * standard. All newer devices conform to the virtio standard only. 3276 */ 3277 bool virtio_legacy_allowed(VirtIODevice *vdev) 3278 { 3279 switch (vdev->device_id) { 3280 case VIRTIO_ID_NET: 3281 case VIRTIO_ID_BLOCK: 3282 case VIRTIO_ID_CONSOLE: 3283 case VIRTIO_ID_RNG: 3284 case VIRTIO_ID_BALLOON: 3285 case VIRTIO_ID_RPMSG: 3286 case VIRTIO_ID_SCSI: 3287 case VIRTIO_ID_9P: 3288 case VIRTIO_ID_RPROC_SERIAL: 3289 case VIRTIO_ID_CAIF: 3290 return true; 3291 default: 3292 return false; 3293 } 3294 } 3295 3296 bool virtio_legacy_check_disabled(VirtIODevice *vdev) 3297 { 3298 return vdev->disable_legacy_check; 3299 } 3300 3301 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n) 3302 { 3303 return vdev->vq[n].vring.desc; 3304 } 3305 3306 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n) 3307 { 3308 return virtio_queue_get_desc_addr(vdev, n) != 0; 3309 } 3310 3311 bool virtio_queue_enabled(VirtIODevice *vdev, int n) 3312 { 3313 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3314 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 3315 3316 if (k->queue_enabled) { 3317 return k->queue_enabled(qbus->parent, n); 3318 } 3319 return virtio_queue_enabled_legacy(vdev, n); 3320 } 3321 3322 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n) 3323 { 3324 return vdev->vq[n].vring.avail; 3325 } 3326 3327 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n) 3328 { 3329 return vdev->vq[n].vring.used; 3330 } 3331 3332 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n) 3333 { 3334 return sizeof(VRingDesc) * vdev->vq[n].vring.num; 3335 } 3336 3337 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n) 3338 { 3339 int s; 3340 3341 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 3342 return sizeof(struct VRingPackedDescEvent); 3343 } 3344 3345 s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; 3346 return offsetof(VRingAvail, ring) + 3347 sizeof(uint16_t) * vdev->vq[n].vring.num + s; 3348 } 3349 3350 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n) 3351 { 3352 int s; 3353 3354 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 3355 return sizeof(struct VRingPackedDescEvent); 3356 } 3357 3358 s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; 3359 return offsetof(VRingUsed, ring) + 3360 sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s; 3361 } 3362 3363 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev, 3364 int n) 3365 { 3366 unsigned int avail, used; 3367 3368 avail = vdev->vq[n].last_avail_idx; 3369 avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15; 3370 3371 used = vdev->vq[n].used_idx; 3372 used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15; 3373 3374 return avail | used << 16; 3375 } 3376 3377 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev, 3378 int n) 3379 { 3380 return vdev->vq[n].last_avail_idx; 3381 } 3382 3383 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n) 3384 { 3385 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 3386 return virtio_queue_packed_get_last_avail_idx(vdev, n); 3387 } else { 3388 return virtio_queue_split_get_last_avail_idx(vdev, n); 3389 } 3390 } 3391 3392 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev, 3393 int n, unsigned int idx) 3394 { 3395 struct VirtQueue *vq = &vdev->vq[n]; 3396 3397 vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff; 3398 vq->last_avail_wrap_counter = 3399 vq->shadow_avail_wrap_counter = !!(idx & 0x8000); 3400 idx >>= 16; 3401 vq->used_idx = idx & 0x7ffff; 3402 vq->used_wrap_counter = !!(idx & 0x8000); 3403 } 3404 3405 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev, 3406 int n, unsigned int idx) 3407 { 3408 vdev->vq[n].last_avail_idx = idx; 3409 vdev->vq[n].shadow_avail_idx = idx; 3410 } 3411 3412 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n, 3413 unsigned int idx) 3414 { 3415 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 3416 virtio_queue_packed_set_last_avail_idx(vdev, n, idx); 3417 } else { 3418 virtio_queue_split_set_last_avail_idx(vdev, n, idx); 3419 } 3420 } 3421 3422 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev, 3423 int n) 3424 { 3425 /* We don't have a reference like avail idx in shared memory */ 3426 return; 3427 } 3428 3429 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev, 3430 int n) 3431 { 3432 RCU_READ_LOCK_GUARD(); 3433 if (vdev->vq[n].vring.desc) { 3434 vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]); 3435 vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx; 3436 } 3437 } 3438 3439 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n) 3440 { 3441 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 3442 virtio_queue_packed_restore_last_avail_idx(vdev, n); 3443 } else { 3444 virtio_queue_split_restore_last_avail_idx(vdev, n); 3445 } 3446 } 3447 3448 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n) 3449 { 3450 /* used idx was updated through set_last_avail_idx() */ 3451 return; 3452 } 3453 3454 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n) 3455 { 3456 RCU_READ_LOCK_GUARD(); 3457 if (vdev->vq[n].vring.desc) { 3458 vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]); 3459 } 3460 } 3461 3462 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n) 3463 { 3464 if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) { 3465 return virtio_queue_packed_update_used_idx(vdev, n); 3466 } else { 3467 return virtio_split_packed_update_used_idx(vdev, n); 3468 } 3469 } 3470 3471 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n) 3472 { 3473 vdev->vq[n].signalled_used_valid = false; 3474 } 3475 3476 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n) 3477 { 3478 return vdev->vq + n; 3479 } 3480 3481 uint16_t virtio_get_queue_index(VirtQueue *vq) 3482 { 3483 return vq->queue_index; 3484 } 3485 3486 static void virtio_queue_guest_notifier_read(EventNotifier *n) 3487 { 3488 VirtQueue *vq = container_of(n, VirtQueue, guest_notifier); 3489 if (event_notifier_test_and_clear(n)) { 3490 virtio_irq(vq); 3491 } 3492 } 3493 3494 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign, 3495 bool with_irqfd) 3496 { 3497 if (assign && !with_irqfd) { 3498 event_notifier_set_handler(&vq->guest_notifier, 3499 virtio_queue_guest_notifier_read); 3500 } else { 3501 event_notifier_set_handler(&vq->guest_notifier, NULL); 3502 } 3503 if (!assign) { 3504 /* Test and clear notifier before closing it, 3505 * in case poll callback didn't have time to run. */ 3506 virtio_queue_guest_notifier_read(&vq->guest_notifier); 3507 } 3508 } 3509 3510 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq) 3511 { 3512 return &vq->guest_notifier; 3513 } 3514 3515 static void virtio_queue_host_notifier_aio_read(EventNotifier *n) 3516 { 3517 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 3518 if (event_notifier_test_and_clear(n)) { 3519 virtio_queue_notify_aio_vq(vq); 3520 } 3521 } 3522 3523 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n) 3524 { 3525 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 3526 3527 virtio_queue_set_notification(vq, 0); 3528 } 3529 3530 static bool virtio_queue_host_notifier_aio_poll(void *opaque) 3531 { 3532 EventNotifier *n = opaque; 3533 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 3534 3535 return vq->vring.desc && !virtio_queue_empty(vq); 3536 } 3537 3538 static void virtio_queue_host_notifier_aio_poll_ready(EventNotifier *n) 3539 { 3540 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 3541 3542 virtio_queue_notify_aio_vq(vq); 3543 } 3544 3545 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n) 3546 { 3547 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 3548 3549 /* Caller polls once more after this to catch requests that race with us */ 3550 virtio_queue_set_notification(vq, 1); 3551 } 3552 3553 void virtio_queue_aio_set_host_notifier_handler(VirtQueue *vq, AioContext *ctx, 3554 VirtIOHandleOutput handle_output) 3555 { 3556 if (handle_output) { 3557 vq->handle_aio_output = handle_output; 3558 aio_set_event_notifier(ctx, &vq->host_notifier, true, 3559 virtio_queue_host_notifier_aio_read, 3560 virtio_queue_host_notifier_aio_poll, 3561 virtio_queue_host_notifier_aio_poll_ready); 3562 aio_set_event_notifier_poll(ctx, &vq->host_notifier, 3563 virtio_queue_host_notifier_aio_poll_begin, 3564 virtio_queue_host_notifier_aio_poll_end); 3565 } else { 3566 aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL, NULL); 3567 /* Test and clear notifier before after disabling event, 3568 * in case poll callback didn't have time to run. */ 3569 virtio_queue_host_notifier_aio_read(&vq->host_notifier); 3570 vq->handle_aio_output = NULL; 3571 } 3572 } 3573 3574 void virtio_queue_host_notifier_read(EventNotifier *n) 3575 { 3576 VirtQueue *vq = container_of(n, VirtQueue, host_notifier); 3577 if (event_notifier_test_and_clear(n)) { 3578 virtio_queue_notify_vq(vq); 3579 } 3580 } 3581 3582 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq) 3583 { 3584 return &vq->host_notifier; 3585 } 3586 3587 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled) 3588 { 3589 vq->host_notifier_enabled = enabled; 3590 } 3591 3592 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n, 3593 MemoryRegion *mr, bool assign) 3594 { 3595 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3596 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 3597 3598 if (k->set_host_notifier_mr) { 3599 return k->set_host_notifier_mr(qbus->parent, n, mr, assign); 3600 } 3601 3602 return -1; 3603 } 3604 3605 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name) 3606 { 3607 g_free(vdev->bus_name); 3608 vdev->bus_name = g_strdup(bus_name); 3609 } 3610 3611 void GCC_FMT_ATTR(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...) 3612 { 3613 va_list ap; 3614 3615 va_start(ap, fmt); 3616 error_vreport(fmt, ap); 3617 va_end(ap); 3618 3619 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 3620 vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET; 3621 virtio_notify_config(vdev); 3622 } 3623 3624 vdev->broken = true; 3625 } 3626 3627 static void virtio_memory_listener_commit(MemoryListener *listener) 3628 { 3629 VirtIODevice *vdev = container_of(listener, VirtIODevice, listener); 3630 int i; 3631 3632 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 3633 if (vdev->vq[i].vring.num == 0) { 3634 break; 3635 } 3636 virtio_init_region_cache(vdev, i); 3637 } 3638 } 3639 3640 static void virtio_device_realize(DeviceState *dev, Error **errp) 3641 { 3642 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3643 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 3644 Error *err = NULL; 3645 3646 /* Devices should either use vmsd or the load/save methods */ 3647 assert(!vdc->vmsd || !vdc->load); 3648 3649 if (vdc->realize != NULL) { 3650 vdc->realize(dev, &err); 3651 if (err != NULL) { 3652 error_propagate(errp, err); 3653 return; 3654 } 3655 } 3656 3657 virtio_bus_device_plugged(vdev, &err); 3658 if (err != NULL) { 3659 error_propagate(errp, err); 3660 vdc->unrealize(dev); 3661 return; 3662 } 3663 3664 vdev->listener.commit = virtio_memory_listener_commit; 3665 vdev->listener.name = "virtio"; 3666 memory_listener_register(&vdev->listener, vdev->dma_as); 3667 } 3668 3669 static void virtio_device_unrealize(DeviceState *dev) 3670 { 3671 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3672 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 3673 3674 memory_listener_unregister(&vdev->listener); 3675 virtio_bus_device_unplugged(vdev); 3676 3677 if (vdc->unrealize != NULL) { 3678 vdc->unrealize(dev); 3679 } 3680 3681 g_free(vdev->bus_name); 3682 vdev->bus_name = NULL; 3683 } 3684 3685 static void virtio_device_free_virtqueues(VirtIODevice *vdev) 3686 { 3687 int i; 3688 if (!vdev->vq) { 3689 return; 3690 } 3691 3692 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 3693 if (vdev->vq[i].vring.num == 0) { 3694 break; 3695 } 3696 virtio_virtqueue_reset_region_cache(&vdev->vq[i]); 3697 } 3698 g_free(vdev->vq); 3699 } 3700 3701 static void virtio_device_instance_finalize(Object *obj) 3702 { 3703 VirtIODevice *vdev = VIRTIO_DEVICE(obj); 3704 3705 virtio_device_free_virtqueues(vdev); 3706 3707 g_free(vdev->config); 3708 g_free(vdev->vector_queues); 3709 } 3710 3711 static Property virtio_properties[] = { 3712 DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features), 3713 DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true), 3714 DEFINE_PROP_BOOL("use-disabled-flag", VirtIODevice, use_disabled_flag, true), 3715 DEFINE_PROP_BOOL("x-disable-legacy-check", VirtIODevice, 3716 disable_legacy_check, false), 3717 DEFINE_PROP_END_OF_LIST(), 3718 }; 3719 3720 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev) 3721 { 3722 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev))); 3723 int i, n, r, err; 3724 3725 /* 3726 * Batch all the host notifiers in a single transaction to avoid 3727 * quadratic time complexity in address_space_update_ioeventfds(). 3728 */ 3729 memory_region_transaction_begin(); 3730 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 3731 VirtQueue *vq = &vdev->vq[n]; 3732 if (!virtio_queue_get_num(vdev, n)) { 3733 continue; 3734 } 3735 r = virtio_bus_set_host_notifier(qbus, n, true); 3736 if (r < 0) { 3737 err = r; 3738 goto assign_error; 3739 } 3740 event_notifier_set_handler(&vq->host_notifier, 3741 virtio_queue_host_notifier_read); 3742 } 3743 3744 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 3745 /* Kick right away to begin processing requests already in vring */ 3746 VirtQueue *vq = &vdev->vq[n]; 3747 if (!vq->vring.num) { 3748 continue; 3749 } 3750 event_notifier_set(&vq->host_notifier); 3751 } 3752 memory_region_transaction_commit(); 3753 return 0; 3754 3755 assign_error: 3756 i = n; /* save n for a second iteration after transaction is committed. */ 3757 while (--n >= 0) { 3758 VirtQueue *vq = &vdev->vq[n]; 3759 if (!virtio_queue_get_num(vdev, n)) { 3760 continue; 3761 } 3762 3763 event_notifier_set_handler(&vq->host_notifier, NULL); 3764 r = virtio_bus_set_host_notifier(qbus, n, false); 3765 assert(r >= 0); 3766 } 3767 /* 3768 * The transaction expects the ioeventfds to be open when it 3769 * commits. Do it now, before the cleanup loop. 3770 */ 3771 memory_region_transaction_commit(); 3772 3773 while (--i >= 0) { 3774 if (!virtio_queue_get_num(vdev, i)) { 3775 continue; 3776 } 3777 virtio_bus_cleanup_host_notifier(qbus, i); 3778 } 3779 return err; 3780 } 3781 3782 int virtio_device_start_ioeventfd(VirtIODevice *vdev) 3783 { 3784 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3785 VirtioBusState *vbus = VIRTIO_BUS(qbus); 3786 3787 return virtio_bus_start_ioeventfd(vbus); 3788 } 3789 3790 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev) 3791 { 3792 VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev))); 3793 int n, r; 3794 3795 /* 3796 * Batch all the host notifiers in a single transaction to avoid 3797 * quadratic time complexity in address_space_update_ioeventfds(). 3798 */ 3799 memory_region_transaction_begin(); 3800 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 3801 VirtQueue *vq = &vdev->vq[n]; 3802 3803 if (!virtio_queue_get_num(vdev, n)) { 3804 continue; 3805 } 3806 event_notifier_set_handler(&vq->host_notifier, NULL); 3807 r = virtio_bus_set_host_notifier(qbus, n, false); 3808 assert(r >= 0); 3809 } 3810 /* 3811 * The transaction expects the ioeventfds to be open when it 3812 * commits. Do it now, before the cleanup loop. 3813 */ 3814 memory_region_transaction_commit(); 3815 3816 for (n = 0; n < VIRTIO_QUEUE_MAX; n++) { 3817 if (!virtio_queue_get_num(vdev, n)) { 3818 continue; 3819 } 3820 virtio_bus_cleanup_host_notifier(qbus, n); 3821 } 3822 } 3823 3824 int virtio_device_grab_ioeventfd(VirtIODevice *vdev) 3825 { 3826 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3827 VirtioBusState *vbus = VIRTIO_BUS(qbus); 3828 3829 return virtio_bus_grab_ioeventfd(vbus); 3830 } 3831 3832 void virtio_device_release_ioeventfd(VirtIODevice *vdev) 3833 { 3834 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3835 VirtioBusState *vbus = VIRTIO_BUS(qbus); 3836 3837 virtio_bus_release_ioeventfd(vbus); 3838 } 3839 3840 static void virtio_device_class_init(ObjectClass *klass, void *data) 3841 { 3842 /* Set the default value here. */ 3843 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 3844 DeviceClass *dc = DEVICE_CLASS(klass); 3845 3846 dc->realize = virtio_device_realize; 3847 dc->unrealize = virtio_device_unrealize; 3848 dc->bus_type = TYPE_VIRTIO_BUS; 3849 device_class_set_props(dc, virtio_properties); 3850 vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl; 3851 vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl; 3852 3853 vdc->legacy_features |= VIRTIO_LEGACY_FEATURES; 3854 } 3855 3856 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev) 3857 { 3858 BusState *qbus = qdev_get_parent_bus(DEVICE(vdev)); 3859 VirtioBusState *vbus = VIRTIO_BUS(qbus); 3860 3861 return virtio_bus_ioeventfd_enabled(vbus); 3862 } 3863 3864 static const TypeInfo virtio_device_info = { 3865 .name = TYPE_VIRTIO_DEVICE, 3866 .parent = TYPE_DEVICE, 3867 .instance_size = sizeof(VirtIODevice), 3868 .class_init = virtio_device_class_init, 3869 .instance_finalize = virtio_device_instance_finalize, 3870 .abstract = true, 3871 .class_size = sizeof(VirtioDeviceClass), 3872 }; 3873 3874 static void virtio_register_types(void) 3875 { 3876 type_register_static(&virtio_device_info); 3877 } 3878 3879 type_init(virtio_register_types) 3880