1 /*- 2 * Copyright (c) 2011, Bryan Venteicher <bryanv@daemoninthecloset.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $FreeBSD: src/sys/dev/virtio/virtqueue.c,v 1.2 2012/04/14 05:48:04 grehan Exp $ 27 */ 28 29 /* 30 * Implements the virtqueue interface as basically described 31 * in the original VirtIO paper. 32 */ 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/malloc.h> 38 #include <sys/sglist.h> 39 #include <sys/serialize.h> 40 #include <vm/vm.h> 41 #include <vm/pmap.h> 42 43 #include <machine/cpu.h> 44 #include <machine/atomic.h> 45 #include <sys/bus.h> 46 #include <sys/rman.h> 47 48 #include "virtio.h" 49 #include "virtqueue.h" 50 #include "virtio_ring.h" 51 52 #include "virtio_bus_if.h" 53 54 struct virtqueue { 55 device_t vq_dev; 56 char vq_name[VIRTQUEUE_MAX_NAME_SZ]; 57 uint16_t vq_queue_index; 58 uint16_t vq_nentries; 59 uint32_t vq_flags; 60 61 #define VIRTQUEUE_FLAG_INDIRECT 0x0001 62 #define VIRTQUEUE_FLAG_EVENT_IDX 0x0002 63 64 int vq_alignment; 65 int vq_ring_size; 66 void *vq_ring_mem; 67 int vq_max_indirect_size; 68 int vq_indirect_mem_size; 69 70 virtqueue_intr_t *vq_intrhand; 71 void *vq_intrhand_arg; 72 73 struct vring vq_ring; 74 uint16_t vq_free_cnt; 75 uint16_t vq_queued_cnt; 76 /* 77 * Head of the free chain in the descriptor table. If 78 * there are no free descriptors, this will be set to 79 * VQ_RING_DESC_CHAIN_END. 80 */ 81 uint16_t vq_desc_head_idx; 82 /* 83 * Last consumed descriptor in the used table, 84 * trails vq_ring.used->idx. 85 */ 86 uint16_t vq_used_cons_idx; 87 88 struct vq_desc_extra { 89 void *cookie; 90 struct vring_desc *indirect; 91 vm_paddr_t indirect_paddr; 92 uint16_t ndescs; 93 } vq_descx[0]; 94 }; 95 96 /* 97 * The maximum virtqueue size is 2^15. Use that value as the end of 98 * descriptor chain terminator since it will never be a valid index 99 * in the descriptor table. This is used to verify we are correctly 100 * handling vq_free_cnt. 101 */ 102 #define VQ_RING_DESC_CHAIN_END 32768 103 104 #define VQASSERT(_vq, _exp, _msg, ...) \ 105 KASSERT((_exp),("%s: %s - "_msg, __func__, (_vq)->vq_name, \ 106 ##__VA_ARGS__)) 107 108 #define VQ_RING_ASSERT_VALID_IDX(_vq, _idx) \ 109 VQASSERT((_vq), (_idx) < (_vq)->vq_nentries, \ 110 "invalid ring index: %d, max: %d", (_idx), \ 111 (_vq)->vq_nentries) 112 113 #define VQ_RING_ASSERT_CHAIN_TERM(_vq) \ 114 VQASSERT((_vq), (_vq)->vq_desc_head_idx == \ 115 VQ_RING_DESC_CHAIN_END, "full ring terminated " \ 116 "incorrectly: head idx: %d", (_vq)->vq_desc_head_idx) 117 118 static int virtqueue_init_indirect(struct virtqueue *vq, int); 119 static void virtqueue_free_indirect(struct virtqueue *vq); 120 static void virtqueue_init_indirect_list(struct virtqueue *, 121 struct vring_desc *); 122 123 static void vq_ring_init(struct virtqueue *); 124 static void vq_ring_update_avail(struct virtqueue *, uint16_t); 125 static uint16_t vq_ring_enqueue_segments(struct virtqueue *, 126 struct vring_desc *, uint16_t, struct sglist *, int, int); 127 static int vq_ring_use_indirect(struct virtqueue *, int); 128 static void vq_ring_enqueue_indirect(struct virtqueue *, void *, 129 struct sglist *, int, int); 130 static int vq_ring_must_notify_host(struct virtqueue *); 131 static void vq_ring_notify_host(struct virtqueue *); 132 static void vq_ring_free_chain(struct virtqueue *, uint16_t); 133 134 uint64_t 135 virtqueue_filter_features(uint64_t features) 136 { 137 uint64_t mask; 138 139 mask = (1 << VIRTIO_TRANSPORT_F_START) - 1; 140 mask |= VIRTIO_RING_F_INDIRECT_DESC; 141 mask |= VIRTIO_RING_F_EVENT_IDX; 142 143 return (features & mask); 144 } 145 146 int 147 virtqueue_alloc(device_t dev, uint16_t queue, uint16_t size, int align, 148 vm_paddr_t highaddr, struct vq_alloc_info *info, struct virtqueue **vqp) 149 { 150 struct virtqueue *vq; 151 int error; 152 153 *vqp = NULL; 154 error = 0; 155 156 if (size == 0) { 157 device_printf(dev, 158 "virtqueue %d (%s) does not exist (size is zero)\n", 159 queue, info->vqai_name); 160 return (ENODEV); 161 } else if (!powerof2(size)) { 162 device_printf(dev, 163 "virtqueue %d (%s) size is not a power of 2: %d\n", 164 queue, info->vqai_name, size); 165 return (ENXIO); 166 } else if (info->vqai_maxindirsz > VIRTIO_MAX_INDIRECT) { 167 device_printf(dev, "virtqueue %d (%s) requested too many " 168 "indirect descriptors: %d, max %d\n", 169 queue, info->vqai_name, info->vqai_maxindirsz, 170 VIRTIO_MAX_INDIRECT); 171 return (EINVAL); 172 } 173 174 vq = kmalloc(sizeof(struct virtqueue) + 175 size * sizeof(struct vq_desc_extra), M_DEVBUF, M_NOWAIT | M_ZERO); 176 if (vq == NULL) { 177 device_printf(dev, "cannot allocate virtqueue\n"); 178 return (ENOMEM); 179 } 180 181 vq->vq_dev = dev; 182 strlcpy(vq->vq_name, info->vqai_name, sizeof(vq->vq_name)); 183 vq->vq_queue_index = queue; 184 vq->vq_alignment = align; 185 vq->vq_nentries = size; 186 vq->vq_free_cnt = size; 187 vq->vq_intrhand = info->vqai_intr; 188 vq->vq_intrhand_arg = info->vqai_intr_arg; 189 190 if (VIRTIO_BUS_WITH_FEATURE(dev, VIRTIO_RING_F_EVENT_IDX) != 0) 191 vq->vq_flags |= VIRTQUEUE_FLAG_EVENT_IDX; 192 193 if (info->vqai_maxindirsz > 1) { 194 error = virtqueue_init_indirect(vq, info->vqai_maxindirsz); 195 if (error) 196 goto fail; 197 } 198 199 vq->vq_ring_size = round_page(vring_size(size, align)); 200 vq->vq_ring_mem = contigmalloc(vq->vq_ring_size, M_DEVBUF, 201 M_NOWAIT | M_ZERO, 0, highaddr, PAGE_SIZE, 0); 202 if (vq->vq_ring_mem == NULL) { 203 device_printf(dev, 204 "cannot allocate memory for virtqueue ring\n"); 205 error = ENOMEM; 206 goto fail; 207 } 208 209 vq_ring_init(vq); 210 virtqueue_disable_intr(vq); 211 212 *vqp = vq; 213 214 fail: 215 if (error) 216 virtqueue_free(vq); 217 218 return (error); 219 } 220 221 static int 222 virtqueue_init_indirect(struct virtqueue *vq, int indirect_size) 223 { 224 device_t dev; 225 struct vq_desc_extra *dxp; 226 int i, size; 227 228 dev = vq->vq_dev; 229 230 if (VIRTIO_BUS_WITH_FEATURE(dev, VIRTIO_RING_F_INDIRECT_DESC) == 0) { 231 /* 232 * Indirect descriptors requested by the driver but not 233 * negotiated. Return zero to keep the initialization 234 * going: we'll run fine without. 235 */ 236 if (bootverbose) 237 device_printf(dev, "virtqueue %d (%s) requested " 238 "indirect descriptors but not negotiated\n", 239 vq->vq_queue_index, vq->vq_name); 240 return (0); 241 } 242 243 size = indirect_size * sizeof(struct vring_desc); 244 vq->vq_max_indirect_size = indirect_size; 245 vq->vq_indirect_mem_size = size; 246 vq->vq_flags |= VIRTQUEUE_FLAG_INDIRECT; 247 248 for (i = 0; i < vq->vq_nentries; i++) { 249 dxp = &vq->vq_descx[i]; 250 251 dxp->indirect = contigmalloc(size, M_DEVBUF, M_NOWAIT, 252 0, BUS_SPACE_MAXADDR, 4, 0); 253 if (dxp->indirect == NULL) { 254 device_printf(dev, "cannot allocate indirect list\n"); 255 return (ENOMEM); 256 } 257 258 dxp->indirect_paddr = vtophys(dxp->indirect); 259 virtqueue_init_indirect_list(vq, dxp->indirect); 260 } 261 262 return (0); 263 } 264 265 static void 266 virtqueue_free_indirect(struct virtqueue *vq) 267 { 268 struct vq_desc_extra *dxp; 269 int i; 270 271 for (i = 0; i < vq->vq_nentries; i++) { 272 dxp = &vq->vq_descx[i]; 273 274 if (dxp->indirect == NULL) 275 break; 276 277 contigfree(dxp->indirect, vq->vq_indirect_mem_size, M_DEVBUF); 278 dxp->indirect = NULL; 279 dxp->indirect_paddr = 0; 280 } 281 282 vq->vq_flags &= ~VIRTQUEUE_FLAG_INDIRECT; 283 vq->vq_indirect_mem_size = 0; 284 } 285 286 static void 287 virtqueue_init_indirect_list(struct virtqueue *vq, 288 struct vring_desc *indirect) 289 { 290 int i; 291 292 bzero(indirect, vq->vq_indirect_mem_size); 293 294 for (i = 0; i < vq->vq_max_indirect_size - 1; i++) 295 indirect[i].next = i + 1; 296 indirect[i].next = VQ_RING_DESC_CHAIN_END; 297 } 298 299 int 300 virtqueue_reinit(struct virtqueue *vq, uint16_t size) 301 { 302 struct vq_desc_extra *dxp; 303 int i; 304 305 if (vq->vq_nentries != size) { 306 device_printf(vq->vq_dev, 307 "%s: '%s' changed size; old=%hu, new=%hu\n", 308 __func__, vq->vq_name, vq->vq_nentries, size); 309 return (EINVAL); 310 } 311 312 /* Warn if the virtqueue was not properly cleaned up. */ 313 if (vq->vq_free_cnt != vq->vq_nentries) { 314 device_printf(vq->vq_dev, 315 "%s: warning, '%s' virtqueue not empty, " 316 "leaking %d entries\n", __func__, vq->vq_name, 317 vq->vq_nentries - vq->vq_free_cnt); 318 } 319 320 vq->vq_desc_head_idx = 0; 321 vq->vq_used_cons_idx = 0; 322 vq->vq_queued_cnt = 0; 323 vq->vq_free_cnt = vq->vq_nentries; 324 325 /* To be safe, reset all our allocated memory. */ 326 bzero(vq->vq_ring_mem, vq->vq_ring_size); 327 for (i = 0; i < vq->vq_nentries; i++) { 328 dxp = &vq->vq_descx[i]; 329 dxp->cookie = NULL; 330 dxp->ndescs = 0; 331 if (vq->vq_flags & VIRTQUEUE_FLAG_INDIRECT) 332 virtqueue_init_indirect_list(vq, dxp->indirect); 333 } 334 335 vq_ring_init(vq); 336 virtqueue_disable_intr(vq); 337 338 return (0); 339 } 340 341 void 342 virtqueue_free(struct virtqueue *vq) 343 { 344 345 if (vq->vq_free_cnt != vq->vq_nentries) { 346 device_printf(vq->vq_dev, "%s: freeing non-empty virtqueue, " 347 "leaking %d entries\n", vq->vq_name, 348 vq->vq_nentries - vq->vq_free_cnt); 349 } 350 351 if (vq->vq_flags & VIRTQUEUE_FLAG_INDIRECT) 352 virtqueue_free_indirect(vq); 353 354 if (vq->vq_ring_mem != NULL) { 355 contigfree(vq->vq_ring_mem, vq->vq_ring_size, M_DEVBUF); 356 vq->vq_ring_size = 0; 357 vq->vq_ring_mem = NULL; 358 } 359 360 kfree(vq, M_DEVBUF); 361 } 362 363 vm_paddr_t 364 virtqueue_paddr(struct virtqueue *vq) 365 { 366 return (vtophys(vq->vq_ring_mem)); 367 } 368 369 int 370 virtqueue_size(struct virtqueue *vq) 371 { 372 return (vq->vq_nentries); 373 } 374 375 int 376 virtqueue_empty(struct virtqueue *vq) 377 { 378 379 return (vq->vq_nentries == vq->vq_free_cnt); 380 } 381 382 int 383 virtqueue_full(struct virtqueue *vq) 384 { 385 386 return (vq->vq_free_cnt == 0); 387 } 388 389 void 390 virtqueue_notify(struct virtqueue *vq, lwkt_serialize_t interlock) 391 { 392 /* Ensure updated avail->idx is visible to host. */ 393 cpu_mfence(); 394 395 if (vq_ring_must_notify_host(vq)) { 396 if (interlock != NULL) 397 lwkt_serialize_exit(interlock); 398 vq_ring_notify_host(vq); 399 if (interlock != NULL) 400 lwkt_serialize_enter(interlock); 401 } 402 vq->vq_queued_cnt = 0; 403 } 404 405 int 406 virtqueue_nused(struct virtqueue *vq) 407 { 408 uint16_t used_idx, nused; 409 410 used_idx = vq->vq_ring.used->idx; 411 nused = (uint16_t)(used_idx - vq->vq_used_cons_idx); 412 VQASSERT(vq, nused <= vq->vq_nentries, "used more than available"); 413 414 return (nused); 415 } 416 417 int 418 virtqueue_intr(struct virtqueue *vq) 419 { 420 421 if (vq->vq_intrhand == NULL || 422 vq->vq_used_cons_idx == vq->vq_ring.used->idx) 423 return (0); 424 425 vq->vq_intrhand(vq->vq_intrhand_arg); 426 427 return (1); 428 } 429 430 /* 431 * Enable interrupts on a given virtqueue. Returns 1 if there are 432 * additional entries to process on the virtqueue after we return. 433 */ 434 int 435 virtqueue_enable_intr(struct virtqueue *vq) 436 { 437 /* 438 * Enable interrupts, making sure we get the latest 439 * index of what's already been consumed. 440 */ 441 vq->vq_ring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; 442 if (vq->vq_flags & VIRTQUEUE_FLAG_EVENT_IDX) { 443 vring_used_event(&vq->vq_ring) = vq->vq_used_cons_idx; 444 } else { 445 vq->vq_ring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; 446 } 447 448 cpu_mfence(); 449 450 /* 451 * Additional items may have been consumed in the time between 452 * since we last checked and enabled interrupts above. Let our 453 * caller know so it processes the new entries. 454 */ 455 if (vq->vq_used_cons_idx != vq->vq_ring.used->idx) 456 return (1); 457 458 return (0); 459 } 460 461 int 462 virtqueue_postpone_intr(struct virtqueue *vq) 463 { 464 uint16_t ndesc; 465 466 /* 467 * Postpone until at least half of the available descriptors 468 * have been consumed. 469 * 470 * XXX Adaptive factor? (Linux uses 3/4) 471 */ 472 ndesc = (uint16_t)(vq->vq_ring.avail->idx - vq->vq_used_cons_idx) / 2; 473 474 if (vq->vq_flags & VIRTQUEUE_FLAG_EVENT_IDX) 475 vring_used_event(&vq->vq_ring) = vq->vq_used_cons_idx + ndesc; 476 else 477 vq->vq_ring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; 478 479 cpu_mfence(); 480 481 /* 482 * Enough items may have already been consumed to meet our 483 * threshold since we last checked. Let our caller know so 484 * it processes the new entries. 485 */ 486 if (virtqueue_nused(vq) > ndesc) 487 return (1); 488 489 return (0); 490 } 491 492 void 493 virtqueue_disable_intr(struct virtqueue *vq) 494 { 495 /* 496 * Note this is only considered a hint to the host. 497 */ 498 if ((vq->vq_flags & VIRTQUEUE_FLAG_EVENT_IDX) == 0) 499 vq->vq_ring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; 500 } 501 502 int 503 virtqueue_enqueue(struct virtqueue *vq, void *cookie, struct sglist *sg, 504 int readable, int writable) 505 { 506 struct vq_desc_extra *dxp; 507 int needed; 508 uint16_t head_idx, idx; 509 510 needed = readable + writable; 511 512 VQASSERT(vq, cookie != NULL, "enqueuing with no cookie"); 513 VQASSERT(vq, needed == sg->sg_nseg, 514 "segment count mismatch, %d, %d", needed, sg->sg_nseg); 515 VQASSERT(vq, 516 needed <= vq->vq_nentries || needed <= vq->vq_max_indirect_size, 517 "too many segments to enqueue: %d, %d/%d", needed, 518 vq->vq_nentries, vq->vq_max_indirect_size); 519 520 if (needed < 1) 521 return (EINVAL); 522 if (vq->vq_free_cnt == 0) 523 return (ENOSPC); 524 525 if (vq_ring_use_indirect(vq, needed)) { 526 vq_ring_enqueue_indirect(vq, cookie, sg, readable, writable); 527 return (0); 528 } else if (vq->vq_free_cnt < needed) 529 return (EMSGSIZE); 530 531 head_idx = vq->vq_desc_head_idx; 532 VQ_RING_ASSERT_VALID_IDX(vq, head_idx); 533 dxp = &vq->vq_descx[head_idx]; 534 535 VQASSERT(vq, dxp->cookie == NULL, 536 "cookie already exists for index %d", head_idx); 537 dxp->cookie = cookie; 538 dxp->ndescs = needed; 539 540 idx = vq_ring_enqueue_segments(vq, vq->vq_ring.desc, head_idx, 541 sg, readable, writable); 542 543 vq->vq_desc_head_idx = idx; 544 vq->vq_free_cnt -= needed; 545 if (vq->vq_free_cnt == 0) 546 VQ_RING_ASSERT_CHAIN_TERM(vq); 547 else 548 VQ_RING_ASSERT_VALID_IDX(vq, idx); 549 550 vq_ring_update_avail(vq, head_idx); 551 552 return (0); 553 } 554 555 void * 556 virtqueue_dequeue(struct virtqueue *vq, uint32_t *len) 557 { 558 struct vring_used_elem *uep; 559 void *cookie; 560 uint16_t used_idx, desc_idx; 561 562 if (vq->vq_used_cons_idx == vq->vq_ring.used->idx) 563 return (NULL); 564 565 used_idx = vq->vq_used_cons_idx++ & (vq->vq_nentries - 1); 566 uep = &vq->vq_ring.used->ring[used_idx]; 567 568 cpu_mfence(); 569 desc_idx = (uint16_t) uep->id; 570 if (len != NULL) 571 *len = uep->len; 572 573 vq_ring_free_chain(vq, desc_idx); 574 575 cookie = vq->vq_descx[desc_idx].cookie; 576 VQASSERT(vq, cookie != NULL, "no cookie for index %d", desc_idx); 577 vq->vq_descx[desc_idx].cookie = NULL; 578 579 return (cookie); 580 } 581 582 void * 583 virtqueue_poll(struct virtqueue *vq, uint32_t *len) 584 { 585 void *cookie; 586 587 /* We only poll the virtqueue when dumping to virtio-blk */ 588 while ((cookie = virtqueue_dequeue(vq, len)) == NULL) 589 ; 590 591 return (cookie); 592 } 593 594 void * 595 virtqueue_drain(struct virtqueue *vq, int *last) 596 { 597 void *cookie; 598 int idx; 599 600 cookie = NULL; 601 idx = *last; 602 603 while (idx < vq->vq_nentries && cookie == NULL) { 604 if ((cookie = vq->vq_descx[idx].cookie) != NULL) { 605 vq->vq_descx[idx].cookie = NULL; 606 /* Free chain to keep free count consistent. */ 607 vq_ring_free_chain(vq, idx); 608 } 609 idx++; 610 } 611 612 *last = idx; 613 614 return (cookie); 615 } 616 617 void 618 virtqueue_dump(struct virtqueue *vq) 619 { 620 621 if (vq == NULL) 622 return; 623 624 kprintf("VQ: %s - size=%d; free=%d; used=%d; queued=%d; " 625 "desc_head_idx=%d; avail.idx=%d; used_cons_idx=%d; " 626 "used.idx=%d; avail.flags=0x%x; used.flags=0x%x\n", 627 vq->vq_name, vq->vq_nentries, vq->vq_free_cnt, 628 virtqueue_nused(vq), vq->vq_queued_cnt, vq->vq_desc_head_idx, 629 vq->vq_ring.avail->idx, vq->vq_used_cons_idx, 630 vq->vq_ring.used->idx, vq->vq_ring.avail->flags, 631 vq->vq_ring.used->flags); 632 } 633 634 static void 635 vq_ring_init(struct virtqueue *vq) 636 { 637 struct vring *vr; 638 char *ring_mem; 639 int i, size; 640 641 ring_mem = vq->vq_ring_mem; 642 size = vq->vq_nentries; 643 vr = &vq->vq_ring; 644 645 vring_init(vr, size, ring_mem, vq->vq_alignment); 646 647 for (i = 0; i < size - 1; i++) 648 vr->desc[i].next = i + 1; 649 vr->desc[i].next = VQ_RING_DESC_CHAIN_END; 650 } 651 652 static void 653 vq_ring_update_avail(struct virtqueue *vq, uint16_t desc_idx) 654 { 655 uint16_t avail_idx; 656 657 /* 658 * Place the head of the descriptor chain into the next slot and make 659 * it usable to the host. The chain is made available now rather than 660 * deferring to virtqueue_notify() in the hopes that if the host is 661 * currently running on another CPU, we can keep it processing the new 662 * descriptor. 663 */ 664 avail_idx = vq->vq_ring.avail->idx & (vq->vq_nentries - 1); 665 vq->vq_ring.avail->ring[avail_idx] = desc_idx; 666 667 cpu_mfence(); 668 vq->vq_ring.avail->idx++; 669 670 /* Keep pending count until virtqueue_notify() for debugging. */ 671 vq->vq_queued_cnt++; 672 } 673 674 static uint16_t 675 vq_ring_enqueue_segments(struct virtqueue *vq, struct vring_desc *desc, 676 uint16_t head_idx, struct sglist *sg, int readable, int writable) 677 { 678 struct sglist_seg *seg; 679 struct vring_desc *dp; 680 int i, needed; 681 uint16_t idx; 682 683 needed = readable + writable; 684 685 for (i = 0, idx = head_idx, seg = sg->sg_segs; 686 i < needed; 687 i++, idx = dp->next, seg++) { 688 VQASSERT(vq, idx != VQ_RING_DESC_CHAIN_END, 689 "premature end of free desc chain"); 690 691 dp = &desc[idx]; 692 dp->addr = seg->ss_paddr; 693 dp->len = seg->ss_len; 694 dp->flags = 0; 695 696 if (i < needed - 1) 697 dp->flags |= VRING_DESC_F_NEXT; 698 if (i >= readable) 699 dp->flags |= VRING_DESC_F_WRITE; 700 } 701 702 return (idx); 703 } 704 705 static int 706 vq_ring_use_indirect(struct virtqueue *vq, int needed) 707 { 708 709 if ((vq->vq_flags & VIRTQUEUE_FLAG_INDIRECT) == 0) 710 return (0); 711 712 if (vq->vq_max_indirect_size < needed) 713 return (0); 714 715 if (needed < 2) 716 return (0); 717 718 return (1); 719 } 720 721 static void 722 vq_ring_enqueue_indirect(struct virtqueue *vq, void *cookie, 723 struct sglist *sg, int readable, int writable) 724 { 725 struct vring_desc *dp; 726 struct vq_desc_extra *dxp; 727 int needed; 728 uint16_t head_idx; 729 730 needed = readable + writable; 731 VQASSERT(vq, needed <= vq->vq_max_indirect_size, 732 "enqueuing too many indirect descriptors"); 733 734 head_idx = vq->vq_desc_head_idx; 735 VQ_RING_ASSERT_VALID_IDX(vq, head_idx); 736 dp = &vq->vq_ring.desc[head_idx]; 737 dxp = &vq->vq_descx[head_idx]; 738 739 VQASSERT(vq, dxp->cookie == NULL, 740 "cookie already exists for index %d", head_idx); 741 dxp->cookie = cookie; 742 dxp->ndescs = 1; 743 744 dp->addr = dxp->indirect_paddr; 745 dp->len = needed * sizeof(struct vring_desc); 746 dp->flags = VRING_DESC_F_INDIRECT; 747 748 vq_ring_enqueue_segments(vq, dxp->indirect, 0, 749 sg, readable, writable); 750 751 vq->vq_desc_head_idx = dp->next; 752 vq->vq_free_cnt--; 753 if (vq->vq_free_cnt == 0) 754 VQ_RING_ASSERT_CHAIN_TERM(vq); 755 else 756 VQ_RING_ASSERT_VALID_IDX(vq, vq->vq_desc_head_idx); 757 758 vq_ring_update_avail(vq, head_idx); 759 } 760 761 static int 762 vq_ring_must_notify_host(struct virtqueue *vq) 763 { 764 uint16_t new_idx, prev_idx, event_idx; 765 766 if (vq->vq_flags & VIRTQUEUE_FLAG_EVENT_IDX) { 767 new_idx = vq->vq_ring.avail->idx; 768 prev_idx = new_idx - vq->vq_queued_cnt; 769 event_idx = vring_avail_event(&vq->vq_ring); 770 771 return (vring_need_event(event_idx, new_idx, prev_idx) != 0); 772 } 773 774 return ((vq->vq_ring.used->flags & VRING_USED_F_NO_NOTIFY) == 0); 775 } 776 777 static void 778 vq_ring_notify_host(struct virtqueue *vq) 779 { 780 VIRTIO_BUS_NOTIFY_VQ(vq->vq_dev, vq->vq_queue_index); 781 } 782 783 static void 784 vq_ring_free_chain(struct virtqueue *vq, uint16_t desc_idx) 785 { 786 struct vring_desc *dp; 787 struct vq_desc_extra *dxp; 788 789 VQ_RING_ASSERT_VALID_IDX(vq, desc_idx); 790 dp = &vq->vq_ring.desc[desc_idx]; 791 dxp = &vq->vq_descx[desc_idx]; 792 793 if (vq->vq_free_cnt == 0) 794 VQ_RING_ASSERT_CHAIN_TERM(vq); 795 796 vq->vq_free_cnt += dxp->ndescs; 797 dxp->ndescs--; 798 799 if ((dp->flags & VRING_DESC_F_INDIRECT) == 0) { 800 while (dp->flags & VRING_DESC_F_NEXT) { 801 VQ_RING_ASSERT_VALID_IDX(vq, dp->next); 802 dp = &vq->vq_ring.desc[dp->next]; 803 dxp->ndescs--; 804 } 805 } 806 VQASSERT(vq, dxp->ndescs == 0, "failed to free entire desc chain"); 807 808 /* 809 * We must append the existing free chain, if any, to the end of 810 * newly freed chain. If the virtqueue was completely used, then 811 * head would be VQ_RING_DESC_CHAIN_END (ASSERTed above). 812 */ 813 dp->next = vq->vq_desc_head_idx; 814 vq->vq_desc_head_idx = desc_idx; 815 } 816