1 /*-
2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.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
27 /*
28 * Implements the virtqueue interface as basically described
29 * in the original VirtIO paper.
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: stable/10/sys/dev/virtio/virtqueue.c 270270 2014-08-21 13:27:05Z bryanv $");
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/sglist.h>
40 #include <vm/vm.h>
41 #include <vm/pmap.h>
42
43 #include <machine/cpu.h>
44 #include <machine/bus.h>
45 #include <machine/atomic.h>
46 #include <machine/resource.h>
47 #include <sys/bus.h>
48 #include <sys/rman.h>
49
50 #include <dev/virtio/virtio.h>
51 #include <dev/virtio/virtqueue.h>
52 #include <dev/virtio/virtio_ring.h>
53
54 #include "virtio_bus_if.h"
55
56 struct virtqueue {
57 device_t vq_dev;
58 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
59 uint16_t vq_queue_index;
60 uint16_t vq_nentries;
61 uint32_t vq_flags;
62 #define VIRTQUEUE_FLAG_INDIRECT 0x0001
63 #define VIRTQUEUE_FLAG_EVENT_IDX 0x0002
64
65 int vq_alignment;
66 int vq_ring_size;
67 void *vq_ring_mem;
68 int vq_max_indirect_size;
69 int vq_indirect_mem_size;
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_enable_interrupt(struct virtqueue *, uint16_t);
131 static int vq_ring_must_notify_host(struct virtqueue *);
132 static void vq_ring_notify_host(struct virtqueue *);
133 static void vq_ring_free_chain(struct virtqueue *, uint16_t);
134
135 uint64_t
virtqueue_filter_features(uint64_t features)136 virtqueue_filter_features(uint64_t features)
137 {
138 uint64_t mask;
139
140 mask = (1 << VIRTIO_TRANSPORT_F_START) - 1;
141 mask |= VIRTIO_RING_F_INDIRECT_DESC;
142 mask |= VIRTIO_RING_F_EVENT_IDX;
143
144 return (features & mask);
145 }
146
147 int
virtqueue_alloc(device_t dev,uint16_t queue,uint16_t size,int align,vm_paddr_t highaddr,struct vq_alloc_info * info,struct virtqueue ** vqp)148 virtqueue_alloc(device_t dev, uint16_t queue, uint16_t size, int align,
149 vm_paddr_t highaddr, struct vq_alloc_info *info, struct virtqueue **vqp)
150 {
151 struct virtqueue *vq;
152 int error;
153
154 *vqp = NULL;
155 error = 0;
156
157 if (size == 0) {
158 device_printf(dev,
159 "virtqueue %d (%s) does not exist (size is zero)\n",
160 queue, info->vqai_name);
161 return (ENODEV);
162 } else if (!powerof2(size)) {
163 device_printf(dev,
164 "virtqueue %d (%s) size is not a power of 2: %d\n",
165 queue, info->vqai_name, size);
166 return (ENXIO);
167 } else if (info->vqai_maxindirsz > VIRTIO_MAX_INDIRECT) {
168 device_printf(dev, "virtqueue %d (%s) requested too many "
169 "indirect descriptors: %d, max %d\n",
170 queue, info->vqai_name, info->vqai_maxindirsz,
171 VIRTIO_MAX_INDIRECT);
172 return (EINVAL);
173 }
174
175 vq = malloc(sizeof(struct virtqueue) +
176 size * sizeof(struct vq_desc_extra), M_DEVBUF, M_NOWAIT | M_ZERO);
177 if (vq == NULL) {
178 device_printf(dev, "cannot allocate virtqueue\n");
179 return (ENOMEM);
180 }
181
182 vq->vq_dev = dev;
183 strlcpy(vq->vq_name, info->vqai_name, sizeof(vq->vq_name));
184 vq->vq_queue_index = queue;
185 vq->vq_alignment = align;
186 vq->vq_nentries = size;
187 vq->vq_free_cnt = size;
188 vq->vq_intrhand = info->vqai_intr;
189 vq->vq_intrhand_arg = info->vqai_intr_arg;
190
191 if (VIRTIO_BUS_WITH_FEATURE(dev, VIRTIO_RING_F_EVENT_IDX) != 0)
192 vq->vq_flags |= VIRTQUEUE_FLAG_EVENT_IDX;
193
194 if (info->vqai_maxindirsz > 1) {
195 error = virtqueue_init_indirect(vq, info->vqai_maxindirsz);
196 if (error)
197 goto fail;
198 }
199
200 vq->vq_ring_size = round_page(vring_size(size, align));
201 vq->vq_ring_mem = contigmalloc(vq->vq_ring_size, M_DEVBUF,
202 M_NOWAIT | M_ZERO, 0, highaddr, PAGE_SIZE, 0);
203 if (vq->vq_ring_mem == NULL) {
204 device_printf(dev,
205 "cannot allocate memory for virtqueue ring\n");
206 error = ENOMEM;
207 goto fail;
208 }
209
210 vq_ring_init(vq);
211 virtqueue_disable_intr(vq);
212
213 *vqp = vq;
214
215 fail:
216 if (error)
217 virtqueue_free(vq);
218
219 return (error);
220 }
221
222 static int
virtqueue_init_indirect(struct virtqueue * vq,int indirect_size)223 virtqueue_init_indirect(struct virtqueue *vq, int indirect_size)
224 {
225 device_t dev;
226 struct vq_desc_extra *dxp;
227 int i, size;
228
229 dev = vq->vq_dev;
230
231 if (VIRTIO_BUS_WITH_FEATURE(dev, VIRTIO_RING_F_INDIRECT_DESC) == 0) {
232 /*
233 * Indirect descriptors requested by the driver but not
234 * negotiated. Return zero to keep the initialization
235 * going: we'll run fine without.
236 */
237 if (bootverbose)
238 device_printf(dev, "virtqueue %d (%s) requested "
239 "indirect descriptors but not negotiated\n",
240 vq->vq_queue_index, vq->vq_name);
241 return (0);
242 }
243
244 size = indirect_size * sizeof(struct vring_desc);
245 vq->vq_max_indirect_size = indirect_size;
246 vq->vq_indirect_mem_size = size;
247 vq->vq_flags |= VIRTQUEUE_FLAG_INDIRECT;
248
249 for (i = 0; i < vq->vq_nentries; i++) {
250 dxp = &vq->vq_descx[i];
251
252 dxp->indirect = malloc(size, M_DEVBUF, M_NOWAIT);
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
virtqueue_free_indirect(struct virtqueue * vq)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 free(dxp->indirect, 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
virtqueue_init_indirect_list(struct virtqueue * vq,struct vring_desc * indirect)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
virtqueue_reinit(struct virtqueue * vq,uint16_t size)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
virtqueue_free(struct virtqueue * vq)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 free(vq, M_DEVBUF);
361 }
362
363 vm_paddr_t
virtqueue_paddr(struct virtqueue * vq)364 virtqueue_paddr(struct virtqueue *vq)
365 {
366
367 return (vtophys(vq->vq_ring_mem));
368 }
369
370 int
virtqueue_size(struct virtqueue * vq)371 virtqueue_size(struct virtqueue *vq)
372 {
373
374 return (vq->vq_nentries);
375 }
376
377 int
virtqueue_nfree(struct virtqueue * vq)378 virtqueue_nfree(struct virtqueue *vq)
379 {
380
381 return (vq->vq_free_cnt);
382 }
383
384 int
virtqueue_empty(struct virtqueue * vq)385 virtqueue_empty(struct virtqueue *vq)
386 {
387
388 return (vq->vq_nentries == vq->vq_free_cnt);
389 }
390
391 int
virtqueue_full(struct virtqueue * vq)392 virtqueue_full(struct virtqueue *vq)
393 {
394
395 return (vq->vq_free_cnt == 0);
396 }
397
398 void
virtqueue_notify(struct virtqueue * vq)399 virtqueue_notify(struct virtqueue *vq)
400 {
401
402 /* Ensure updated avail->idx is visible to host. */
403 mb();
404
405 if (vq_ring_must_notify_host(vq))
406 vq_ring_notify_host(vq);
407 vq->vq_queued_cnt = 0;
408 }
409
410 int
virtqueue_nused(struct virtqueue * vq)411 virtqueue_nused(struct virtqueue *vq)
412 {
413 uint16_t used_idx, nused;
414
415 used_idx = vq->vq_ring.used->idx;
416
417 nused = (uint16_t)(used_idx - vq->vq_used_cons_idx);
418 VQASSERT(vq, nused <= vq->vq_nentries, "used more than available");
419
420 return (nused);
421 }
422
423 int
virtqueue_intr_filter(struct virtqueue * vq)424 virtqueue_intr_filter(struct virtqueue *vq)
425 {
426
427 if (vq->vq_used_cons_idx == vq->vq_ring.used->idx)
428 return (0);
429
430 virtqueue_disable_intr(vq);
431
432 return (1);
433 }
434
435 void
virtqueue_intr(struct virtqueue * vq)436 virtqueue_intr(struct virtqueue *vq)
437 {
438
439 vq->vq_intrhand(vq->vq_intrhand_arg);
440 }
441
442 int
virtqueue_enable_intr(struct virtqueue * vq)443 virtqueue_enable_intr(struct virtqueue *vq)
444 {
445
446 return (vq_ring_enable_interrupt(vq, 0));
447 }
448
449 int
virtqueue_postpone_intr(struct virtqueue * vq,vq_postpone_t hint)450 virtqueue_postpone_intr(struct virtqueue *vq, vq_postpone_t hint)
451 {
452 uint16_t ndesc, avail_idx;
453
454 avail_idx = vq->vq_ring.avail->idx;
455 ndesc = (uint16_t)(avail_idx - vq->vq_used_cons_idx);
456
457 switch (hint) {
458 case VQ_POSTPONE_SHORT:
459 ndesc = ndesc / 4;
460 break;
461 case VQ_POSTPONE_LONG:
462 ndesc = (ndesc * 3) / 4;
463 break;
464 case VQ_POSTPONE_EMPTIED:
465 break;
466 }
467
468 return (vq_ring_enable_interrupt(vq, ndesc));
469 }
470
471 /*
472 * Note this is only considered a hint to the host.
473 */
474 void
virtqueue_disable_intr(struct virtqueue * vq)475 virtqueue_disable_intr(struct virtqueue *vq)
476 {
477
478 if (vq->vq_flags & VIRTQUEUE_FLAG_EVENT_IDX) {
479 vring_used_event(&vq->vq_ring) = vq->vq_used_cons_idx -
480 vq->vq_nentries - 1;
481 } else
482 vq->vq_ring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
483 }
484
485 int
virtqueue_enqueue(struct virtqueue * vq,void * cookie,struct sglist * sg,int readable,int writable)486 virtqueue_enqueue(struct virtqueue *vq, void *cookie, struct sglist *sg,
487 int readable, int writable)
488 {
489 struct vq_desc_extra *dxp;
490 int needed;
491 uint16_t head_idx, idx;
492
493 needed = readable + writable;
494
495 VQASSERT(vq, cookie != NULL, "enqueuing with no cookie");
496 VQASSERT(vq, needed == sg->sg_nseg,
497 "segment count mismatch, %d, %d", needed, sg->sg_nseg);
498 VQASSERT(vq,
499 needed <= vq->vq_nentries || needed <= vq->vq_max_indirect_size,
500 "too many segments to enqueue: %d, %d/%d", needed,
501 vq->vq_nentries, vq->vq_max_indirect_size);
502
503 if (needed < 1)
504 return (EINVAL);
505 if (vq->vq_free_cnt == 0)
506 return (ENOSPC);
507
508 if (vq_ring_use_indirect(vq, needed)) {
509 vq_ring_enqueue_indirect(vq, cookie, sg, readable, writable);
510 return (0);
511 } else if (vq->vq_free_cnt < needed)
512 return (EMSGSIZE);
513
514 head_idx = vq->vq_desc_head_idx;
515 VQ_RING_ASSERT_VALID_IDX(vq, head_idx);
516 dxp = &vq->vq_descx[head_idx];
517
518 VQASSERT(vq, dxp->cookie == NULL,
519 "cookie already exists for index %d", head_idx);
520 dxp->cookie = cookie;
521 dxp->ndescs = needed;
522
523 idx = vq_ring_enqueue_segments(vq, vq->vq_ring.desc, head_idx,
524 sg, readable, writable);
525
526 vq->vq_desc_head_idx = idx;
527 vq->vq_free_cnt -= needed;
528 if (vq->vq_free_cnt == 0)
529 VQ_RING_ASSERT_CHAIN_TERM(vq);
530 else
531 VQ_RING_ASSERT_VALID_IDX(vq, idx);
532
533 vq_ring_update_avail(vq, head_idx);
534
535 return (0);
536 }
537
538 void *
virtqueue_dequeue(struct virtqueue * vq,uint32_t * len)539 virtqueue_dequeue(struct virtqueue *vq, uint32_t *len)
540 {
541 struct vring_used_elem *uep;
542 void *cookie;
543 uint16_t used_idx, desc_idx;
544
545 if (vq->vq_used_cons_idx == vq->vq_ring.used->idx)
546 return (NULL);
547
548 used_idx = vq->vq_used_cons_idx++ & (vq->vq_nentries - 1);
549 uep = &vq->vq_ring.used->ring[used_idx];
550
551 rmb();
552 desc_idx = (uint16_t) uep->id;
553 if (len != NULL)
554 *len = uep->len;
555
556 vq_ring_free_chain(vq, desc_idx);
557
558 cookie = vq->vq_descx[desc_idx].cookie;
559 VQASSERT(vq, cookie != NULL, "no cookie for index %d", desc_idx);
560 vq->vq_descx[desc_idx].cookie = NULL;
561
562 return (cookie);
563 }
564
565 void *
virtqueue_poll(struct virtqueue * vq,uint32_t * len)566 virtqueue_poll(struct virtqueue *vq, uint32_t *len)
567 {
568 void *cookie;
569
570 while ((cookie = virtqueue_dequeue(vq, len)) == NULL)
571 cpu_spinwait();
572
573 return (cookie);
574 }
575
576 void *
virtqueue_drain(struct virtqueue * vq,int * last)577 virtqueue_drain(struct virtqueue *vq, int *last)
578 {
579 void *cookie;
580 int idx;
581
582 cookie = NULL;
583 idx = *last;
584
585 while (idx < vq->vq_nentries && cookie == NULL) {
586 if ((cookie = vq->vq_descx[idx].cookie) != NULL) {
587 vq->vq_descx[idx].cookie = NULL;
588 /* Free chain to keep free count consistent. */
589 vq_ring_free_chain(vq, idx);
590 }
591 idx++;
592 }
593
594 *last = idx;
595
596 return (cookie);
597 }
598
599 void
virtqueue_dump(struct virtqueue * vq)600 virtqueue_dump(struct virtqueue *vq)
601 {
602
603 if (vq == NULL)
604 return;
605
606 printf("VQ: %s - size=%d; free=%d; used=%d; queued=%d; "
607 "desc_head_idx=%d; avail.idx=%d; used_cons_idx=%d; "
608 "used.idx=%d; avail.flags=0x%x; used.flags=0x%x\n",
609 vq->vq_name, vq->vq_nentries, vq->vq_free_cnt,
610 virtqueue_nused(vq), vq->vq_queued_cnt, vq->vq_desc_head_idx,
611 vq->vq_ring.avail->idx, vq->vq_used_cons_idx,
612 vq->vq_ring.used->idx, vq->vq_ring.avail->flags,
613 vq->vq_ring.used->flags);
614 }
615
616 static void
vq_ring_init(struct virtqueue * vq)617 vq_ring_init(struct virtqueue *vq)
618 {
619 struct vring *vr;
620 char *ring_mem;
621 int i, size;
622
623 ring_mem = vq->vq_ring_mem;
624 size = vq->vq_nentries;
625 vr = &vq->vq_ring;
626
627 vring_init(vr, size, ring_mem, vq->vq_alignment);
628
629 for (i = 0; i < size - 1; i++)
630 vr->desc[i].next = i + 1;
631 vr->desc[i].next = VQ_RING_DESC_CHAIN_END;
632 }
633
634 static void
vq_ring_update_avail(struct virtqueue * vq,uint16_t desc_idx)635 vq_ring_update_avail(struct virtqueue *vq, uint16_t desc_idx)
636 {
637 uint16_t avail_idx;
638
639 /*
640 * Place the head of the descriptor chain into the next slot and make
641 * it usable to the host. The chain is made available now rather than
642 * deferring to virtqueue_notify() in the hopes that if the host is
643 * currently running on another CPU, we can keep it processing the new
644 * descriptor.
645 */
646 avail_idx = vq->vq_ring.avail->idx & (vq->vq_nentries - 1);
647 vq->vq_ring.avail->ring[avail_idx] = desc_idx;
648
649 wmb();
650 vq->vq_ring.avail->idx++;
651
652 /* Keep pending count until virtqueue_notify(). */
653 vq->vq_queued_cnt++;
654 }
655
656 static uint16_t
vq_ring_enqueue_segments(struct virtqueue * vq,struct vring_desc * desc,uint16_t head_idx,struct sglist * sg,int readable,int writable)657 vq_ring_enqueue_segments(struct virtqueue *vq, struct vring_desc *desc,
658 uint16_t head_idx, struct sglist *sg, int readable, int writable)
659 {
660 struct sglist_seg *seg;
661 struct vring_desc *dp;
662 int i, needed;
663 uint16_t idx;
664
665 needed = readable + writable;
666
667 for (i = 0, idx = head_idx, seg = sg->sg_segs;
668 i < needed;
669 i++, idx = dp->next, seg++) {
670 VQASSERT(vq, idx != VQ_RING_DESC_CHAIN_END,
671 "premature end of free desc chain");
672
673 dp = &desc[idx];
674 dp->addr = seg->ss_paddr;
675 dp->len = seg->ss_len;
676 dp->flags = 0;
677
678 if (i < needed - 1)
679 dp->flags |= VRING_DESC_F_NEXT;
680 if (i >= readable)
681 dp->flags |= VRING_DESC_F_WRITE;
682 }
683
684 return (idx);
685 }
686
687 static int
vq_ring_use_indirect(struct virtqueue * vq,int needed)688 vq_ring_use_indirect(struct virtqueue *vq, int needed)
689 {
690
691 if ((vq->vq_flags & VIRTQUEUE_FLAG_INDIRECT) == 0)
692 return (0);
693
694 if (vq->vq_max_indirect_size < needed)
695 return (0);
696
697 if (needed < 2)
698 return (0);
699
700 return (1);
701 }
702
703 static void
vq_ring_enqueue_indirect(struct virtqueue * vq,void * cookie,struct sglist * sg,int readable,int writable)704 vq_ring_enqueue_indirect(struct virtqueue *vq, void *cookie,
705 struct sglist *sg, int readable, int writable)
706 {
707 struct vring_desc *dp;
708 struct vq_desc_extra *dxp;
709 int needed;
710 uint16_t head_idx;
711
712 needed = readable + writable;
713 VQASSERT(vq, needed <= vq->vq_max_indirect_size,
714 "enqueuing too many indirect descriptors");
715
716 head_idx = vq->vq_desc_head_idx;
717 VQ_RING_ASSERT_VALID_IDX(vq, head_idx);
718 dp = &vq->vq_ring.desc[head_idx];
719 dxp = &vq->vq_descx[head_idx];
720
721 VQASSERT(vq, dxp->cookie == NULL,
722 "cookie already exists for index %d", head_idx);
723 dxp->cookie = cookie;
724 dxp->ndescs = 1;
725
726 dp->addr = dxp->indirect_paddr;
727 dp->len = needed * sizeof(struct vring_desc);
728 dp->flags = VRING_DESC_F_INDIRECT;
729
730 vq_ring_enqueue_segments(vq, dxp->indirect, 0,
731 sg, readable, writable);
732
733 vq->vq_desc_head_idx = dp->next;
734 vq->vq_free_cnt--;
735 if (vq->vq_free_cnt == 0)
736 VQ_RING_ASSERT_CHAIN_TERM(vq);
737 else
738 VQ_RING_ASSERT_VALID_IDX(vq, vq->vq_desc_head_idx);
739
740 vq_ring_update_avail(vq, head_idx);
741 }
742
743 static int
vq_ring_enable_interrupt(struct virtqueue * vq,uint16_t ndesc)744 vq_ring_enable_interrupt(struct virtqueue *vq, uint16_t ndesc)
745 {
746
747 /*
748 * Enable interrupts, making sure we get the latest index of
749 * what's already been consumed.
750 */
751 if (vq->vq_flags & VIRTQUEUE_FLAG_EVENT_IDX)
752 vring_used_event(&vq->vq_ring) = vq->vq_used_cons_idx + ndesc;
753 else
754 vq->vq_ring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
755
756 mb();
757
758 /*
759 * Enough items may have already been consumed to meet our threshold
760 * since we last checked. Let our caller know so it processes the new
761 * entries.
762 */
763 if (virtqueue_nused(vq) > ndesc)
764 return (1);
765
766 return (0);
767 }
768
769 static int
vq_ring_must_notify_host(struct virtqueue * vq)770 vq_ring_must_notify_host(struct virtqueue *vq)
771 {
772 uint16_t new_idx, prev_idx, event_idx;
773
774 if (vq->vq_flags & VIRTQUEUE_FLAG_EVENT_IDX) {
775 new_idx = vq->vq_ring.avail->idx;
776 prev_idx = new_idx - vq->vq_queued_cnt;
777 event_idx = vring_avail_event(&vq->vq_ring);
778
779 return (vring_need_event(event_idx, new_idx, prev_idx) != 0);
780 }
781
782 return ((vq->vq_ring.used->flags & VRING_USED_F_NO_NOTIFY) == 0);
783 }
784
785 static void
vq_ring_notify_host(struct virtqueue * vq)786 vq_ring_notify_host(struct virtqueue *vq)
787 {
788
789 VIRTIO_BUS_NOTIFY_VQ(vq->vq_dev, vq->vq_queue_index);
790 }
791
792 static void
vq_ring_free_chain(struct virtqueue * vq,uint16_t desc_idx)793 vq_ring_free_chain(struct virtqueue *vq, uint16_t desc_idx)
794 {
795 struct vring_desc *dp;
796 struct vq_desc_extra *dxp;
797
798 VQ_RING_ASSERT_VALID_IDX(vq, desc_idx);
799 dp = &vq->vq_ring.desc[desc_idx];
800 dxp = &vq->vq_descx[desc_idx];
801
802 if (vq->vq_free_cnt == 0)
803 VQ_RING_ASSERT_CHAIN_TERM(vq);
804
805 vq->vq_free_cnt += dxp->ndescs;
806 dxp->ndescs--;
807
808 if ((dp->flags & VRING_DESC_F_INDIRECT) == 0) {
809 while (dp->flags & VRING_DESC_F_NEXT) {
810 VQ_RING_ASSERT_VALID_IDX(vq, dp->next);
811 dp = &vq->vq_ring.desc[dp->next];
812 dxp->ndescs--;
813 }
814 }
815
816 VQASSERT(vq, dxp->ndescs == 0,
817 "failed to free entire desc chain, remaining: %d", dxp->ndescs);
818
819 /*
820 * We must append the existing free chain, if any, to the end of
821 * newly freed chain. If the virtqueue was completely used, then
822 * head would be VQ_RING_DESC_CHAIN_END (ASSERTed above).
823 */
824 dp->next = vq->vq_desc_head_idx;
825 vq->vq_desc_head_idx = desc_idx;
826 }
827