xref: /netbsd-src/sys/dev/pci/virtio.c (revision e7ac2a8b5bd66fa2e050809de09a075c36a7014d)
1 /*	$NetBSD: virtio.c,v 1.41 2020/05/25 07:52:16 yamaguchi Exp $	*/
2 
3 /*
4  * Copyright (c) 2010 Minoura Makoto.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __KERNEL_RCSID(0, "$NetBSD: virtio.c,v 1.41 2020/05/25 07:52:16 yamaguchi Exp $");
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/atomic.h>
35 #include <sys/bus.h>
36 #include <sys/device.h>
37 #include <sys/kmem.h>
38 #include <sys/module.h>
39 
40 #define VIRTIO_PRIVATE
41 
42 #include <dev/pci/virtioreg.h> /* XXX: move to non-pci */
43 #include <dev/pci/virtiovar.h> /* XXX: move to non-pci */
44 
45 #define MINSEG_INDIRECT		2 /* use indirect if nsegs >= this value */
46 
47 static void	virtio_init_vq(struct virtio_softc *,
48 		    struct virtqueue *, const bool);
49 
50 void
51 virtio_set_status(struct virtio_softc *sc, int status)
52 {
53 	sc->sc_ops->set_status(sc, status);
54 }
55 
56 /*
57  * Reset the device.
58  */
59 /*
60  * To reset the device to a known state, do following:
61  *	virtio_reset(sc);	     // this will stop the device activity
62  *	<dequeue finished requests>; // virtio_dequeue() still can be called
63  *	<revoke pending requests in the vqs if any>;
64  *	virtio_reinit_begin(sc);     // dequeue prohibitted
65  *	newfeatures = virtio_negotiate_features(sc, requestedfeatures);
66  *	<some other initialization>;
67  *	virtio_reinit_end(sc);	     // device activated; enqueue allowed
68  * Once attached, feature negotiation can only be allowed after virtio_reset.
69  */
70 void
71 virtio_reset(struct virtio_softc *sc)
72 {
73 	virtio_device_reset(sc);
74 }
75 
76 void
77 virtio_reinit_start(struct virtio_softc *sc)
78 {
79 	int i;
80 
81 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
82 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
83 	for (i = 0; i < sc->sc_nvqs; i++) {
84 		int n;
85 		struct virtqueue *vq = &sc->sc_vqs[i];
86 		n = sc->sc_ops->read_queue_size(sc, vq->vq_index);
87 		if (n == 0)	/* vq disappeared */
88 			continue;
89 		if (n != vq->vq_num) {
90 			panic("%s: virtqueue size changed, vq index %d\n",
91 			      device_xname(sc->sc_dev),
92 			      vq->vq_index);
93 		}
94 		virtio_init_vq(sc, vq, true);
95 		sc->sc_ops->setup_queue(sc, vq->vq_index,
96 		    vq->vq_dmamap->dm_segs[0].ds_addr / VIRTIO_PAGE_SIZE);
97 	}
98 }
99 
100 void
101 virtio_reinit_end(struct virtio_softc *sc)
102 {
103 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
104 }
105 
106 /*
107  * Feature negotiation.
108  */
109 uint32_t
110 virtio_negotiate_features(struct virtio_softc *sc, uint32_t guest_features)
111 {
112 	uint32_t r;
113 
114 	if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) &&
115 	    !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */
116 		guest_features |= VIRTIO_F_RING_INDIRECT_DESC;
117 	r = sc->sc_ops->neg_features(sc, guest_features);
118 	sc->sc_features = r;
119 	if (r & VIRTIO_F_RING_INDIRECT_DESC)
120 		sc->sc_indirect = true;
121 	else
122 		sc->sc_indirect = false;
123 
124 	return r;
125 }
126 
127 /*
128  * Device configuration registers.
129  */
130 uint8_t
131 virtio_read_device_config_1(struct virtio_softc *sc, int index)
132 {
133 	return sc->sc_ops->read_dev_cfg_1(sc, index);
134 }
135 
136 uint16_t
137 virtio_read_device_config_2(struct virtio_softc *sc, int index)
138 {
139 	return sc->sc_ops->read_dev_cfg_2(sc, index);
140 }
141 
142 uint32_t
143 virtio_read_device_config_4(struct virtio_softc *sc, int index)
144 {
145 	return sc->sc_ops->read_dev_cfg_4(sc, index);
146 }
147 
148 uint64_t
149 virtio_read_device_config_8(struct virtio_softc *sc, int index)
150 {
151 	return sc->sc_ops->read_dev_cfg_8(sc, index);
152 }
153 
154 void
155 virtio_write_device_config_1(struct virtio_softc *sc,
156 			     int index, uint8_t value)
157 {
158 	return sc->sc_ops->write_dev_cfg_1(sc, index, value);
159 }
160 
161 void
162 virtio_write_device_config_2(struct virtio_softc *sc,
163 			     int index, uint16_t value)
164 {
165 	return sc->sc_ops->write_dev_cfg_2(sc, index, value);
166 }
167 
168 void
169 virtio_write_device_config_4(struct virtio_softc *sc,
170 			     int index, uint32_t value)
171 {
172 	return sc->sc_ops->write_dev_cfg_4(sc, index, value);
173 }
174 
175 void
176 virtio_write_device_config_8(struct virtio_softc *sc,
177 			     int index, uint64_t value)
178 {
179 	return sc->sc_ops->write_dev_cfg_8(sc, index, value);
180 }
181 
182 /*
183  * Interrupt handler.
184  */
185 static void
186 virtio_soft_intr(void *arg)
187 {
188 	struct virtio_softc *sc = arg;
189 
190 	KASSERT(sc->sc_intrhand != NULL);
191 
192 	(sc->sc_intrhand)(sc);
193 }
194 
195 /*
196  * dmamap sync operations for a virtqueue.
197  */
198 static inline void
199 vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops)
200 {
201 	/* availoffset == sizeof(vring_desc)*vq_num */
202 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset,
203 			ops);
204 }
205 
206 static inline void
207 vq_sync_aring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
208 {
209 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
210 			vq->vq_availoffset,
211 			offsetof(struct vring_avail, ring)
212 			 + vq->vq_num * sizeof(uint16_t),
213 			ops);
214 }
215 
216 static inline void
217 vq_sync_uring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
218 {
219 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
220 			vq->vq_usedoffset,
221 			offsetof(struct vring_used, ring)
222 			 + vq->vq_num * sizeof(struct vring_used_elem),
223 			ops);
224 }
225 
226 static inline void
227 vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot,
228 		     int ops)
229 {
230 	int offset = vq->vq_indirectoffset
231 		      + sizeof(struct vring_desc) * vq->vq_maxnsegs * slot;
232 
233 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
234 			offset, sizeof(struct vring_desc) * vq->vq_maxnsegs,
235 			ops);
236 }
237 
238 /*
239  * Can be used as sc_intrhand.
240  */
241 /*
242  * Scan vq, bus_dmamap_sync for the vqs (not for the payload),
243  * and calls (*vq_done)() if some entries are consumed.
244  */
245 bool
246 virtio_vq_is_enqueued(struct virtio_softc *sc, struct virtqueue *vq)
247 {
248 
249 	if (vq->vq_queued) {
250 		vq->vq_queued = 0;
251 		vq_sync_aring(sc, vq, BUS_DMASYNC_POSTWRITE);
252 	}
253 	vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
254 	membar_consumer();
255 
256 	return (vq->vq_used_idx != vq->vq_used->idx) ? 1 : 0;
257 }
258 
259 int
260 virtio_vq_intr(struct virtio_softc *sc)
261 {
262 	struct virtqueue *vq;
263 	int i, r = 0;
264 
265 	for (i = 0; i < sc->sc_nvqs; i++) {
266 		vq = &sc->sc_vqs[i];
267 		if (virtio_vq_is_enqueued(sc, vq) == 1) {
268 			if (vq->vq_done)
269 				r |= (vq->vq_done)(vq);
270 		}
271 	}
272 
273 	return r;
274 }
275 
276 int
277 virtio_vq_intrhand(struct virtio_softc *sc)
278 {
279 	struct virtqueue *vq;
280 	int i, r = 0;
281 
282 	for (i = 0; i < sc->sc_nvqs; i++) {
283 		vq = &sc->sc_vqs[i];
284 		r |= (vq->vq_intrhand)(vq->vq_intrhand_arg);
285 	}
286 
287 	return r;
288 }
289 
290 /*
291  * Start/stop vq interrupt.  No guarantee.
292  */
293 void
294 virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
295 {
296 	vq->vq_avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
297 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
298 	vq->vq_queued++;
299 }
300 
301 void
302 virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
303 {
304 	vq->vq_avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
305 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
306 	vq->vq_queued++;
307 }
308 
309 /*
310  * Initialize vq structure.
311  */
312 static void
313 virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq,
314     const bool reinit)
315 {
316 	int i, j;
317 	int vq_size = vq->vq_num;
318 
319 	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
320 
321 	/* build the indirect descriptor chain */
322 	if (vq->vq_indirect != NULL) {
323 		struct vring_desc *vd;
324 
325 		for (i = 0; i < vq_size; i++) {
326 			vd = vq->vq_indirect;
327 			vd += vq->vq_maxnsegs * i;
328 			for (j = 0; j < vq->vq_maxnsegs-1; j++) {
329 				vd[j].next = j + 1;
330 			}
331 		}
332 	}
333 
334 	/* free slot management */
335 	SIMPLEQ_INIT(&vq->vq_freelist);
336 	for (i = 0; i < vq_size; i++) {
337 		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
338 				    &vq->vq_entries[i], qe_list);
339 		vq->vq_entries[i].qe_index = i;
340 	}
341 	if (!reinit)
342 		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
343 
344 	/* enqueue/dequeue status */
345 	vq->vq_avail_idx = 0;
346 	vq->vq_used_idx = 0;
347 	vq->vq_queued = 0;
348 	if (!reinit) {
349 		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
350 		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
351 	}
352 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
353 	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
354 	vq->vq_queued++;
355 }
356 
357 /*
358  * Allocate/free a vq.
359  */
360 int
361 virtio_alloc_vq(struct virtio_softc *sc, struct virtqueue *vq, int index,
362     int maxsegsize, int maxnsegs, const char *name)
363 {
364 	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
365 	int rsegs, r;
366 #define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
367 				 ~(VIRTIO_PAGE_SIZE-1))
368 
369 	/* Make sure callers allocate vqs in order */
370 	KASSERT(sc->sc_nvqs == index);
371 
372 	memset(vq, 0, sizeof(*vq));
373 
374 	vq_size = sc->sc_ops->read_queue_size(sc, index);
375 	if (vq_size == 0) {
376 		aprint_error_dev(sc->sc_dev,
377 				 "virtqueue not exist, index %d for %s\n",
378 				 index, name);
379 		goto err;
380 	}
381 	/* allocsize1: descriptor table + avail ring + pad */
382 	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
383 				     + sizeof(uint16_t)*(2+vq_size));
384 	/* allocsize2: used ring + pad */
385 	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
386 				     + sizeof(struct vring_used_elem)*vq_size);
387 	/* allocsize3: indirect table */
388 	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
389 		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
390 	else
391 		allocsize3 = 0;
392 	allocsize = allocsize1 + allocsize2 + allocsize3;
393 
394 	/* alloc and map the memory */
395 	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
396 			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
397 	if (r != 0) {
398 		aprint_error_dev(sc->sc_dev,
399 				 "virtqueue %d for %s allocation failed, "
400 				 "error code %d\n", index, name, r);
401 		goto err;
402 	}
403 	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
404 			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
405 	if (r != 0) {
406 		aprint_error_dev(sc->sc_dev,
407 				 "virtqueue %d for %s map failed, "
408 				 "error code %d\n", index, name, r);
409 		goto err;
410 	}
411 	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
412 			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
413 	if (r != 0) {
414 		aprint_error_dev(sc->sc_dev,
415 				 "virtqueue %d for %s dmamap creation failed, "
416 				 "error code %d\n", index, name, r);
417 		goto err;
418 	}
419 	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
420 			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
421 	if (r != 0) {
422 		aprint_error_dev(sc->sc_dev,
423 				 "virtqueue %d for %s dmamap load failed, "
424 				 "error code %d\n", index, name, r);
425 		goto err;
426 	}
427 
428 	/* set the vq address */
429 	sc->sc_ops->setup_queue(sc, index,
430 	    vq->vq_dmamap->dm_segs[0].ds_addr / VIRTIO_PAGE_SIZE);
431 
432 	/* remember addresses and offsets for later use */
433 	vq->vq_owner = sc;
434 	vq->vq_num = vq_size;
435 	vq->vq_index = index;
436 	vq->vq_desc = vq->vq_vaddr;
437 	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
438 	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
439 	vq->vq_usedoffset = allocsize1;
440 	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
441 	if (allocsize3 > 0) {
442 		vq->vq_indirectoffset = allocsize1 + allocsize2;
443 		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
444 					  + vq->vq_indirectoffset);
445 	}
446 	vq->vq_bytesize = allocsize;
447 	vq->vq_maxsegsize = maxsegsize;
448 	vq->vq_maxnsegs = maxnsegs;
449 
450 	/* free slot management */
451 	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
452 				     KM_SLEEP);
453 	virtio_init_vq(sc, vq, false);
454 
455 	aprint_verbose_dev(sc->sc_dev,
456 			   "allocated %u byte for virtqueue %d for %s, "
457 			   "size %d\n", allocsize, index, name, vq_size);
458 	if (allocsize3 > 0)
459 		aprint_verbose_dev(sc->sc_dev,
460 				   "using %d byte (%d entries) "
461 				   "indirect descriptors\n",
462 				   allocsize3, maxnsegs * vq_size);
463 
464 	sc->sc_nvqs++;
465 
466 	return 0;
467 
468 err:
469 	sc->sc_ops->setup_queue(sc, index, 0);
470 	if (vq->vq_dmamap)
471 		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
472 	if (vq->vq_vaddr)
473 		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
474 	if (vq->vq_segs[0].ds_addr)
475 		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
476 	memset(vq, 0, sizeof(*vq));
477 
478 	return -1;
479 }
480 
481 int
482 virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
483 {
484 	struct vq_entry *qe;
485 	int i = 0;
486 
487 	/* device must be already deactivated */
488 	/* confirm the vq is empty */
489 	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
490 		i++;
491 	}
492 	if (i != vq->vq_num) {
493 		printf("%s: freeing non-empty vq, index %d\n",
494 		       device_xname(sc->sc_dev), vq->vq_index);
495 		return EBUSY;
496 	}
497 
498 	/* tell device that there's no virtqueue any longer */
499 	sc->sc_ops->setup_queue(sc, vq->vq_index, 0);
500 
501 	kmem_free(vq->vq_entries, sizeof(*vq->vq_entries) * vq->vq_num);
502 	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
503 	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
504 	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
505 	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
506 	mutex_destroy(&vq->vq_freelist_lock);
507 	mutex_destroy(&vq->vq_uring_lock);
508 	mutex_destroy(&vq->vq_aring_lock);
509 	memset(vq, 0, sizeof(*vq));
510 
511 	sc->sc_nvqs--;
512 
513 	return 0;
514 }
515 
516 /*
517  * Free descriptor management.
518  */
519 static struct vq_entry *
520 vq_alloc_entry(struct virtqueue *vq)
521 {
522 	struct vq_entry *qe;
523 
524 	mutex_enter(&vq->vq_freelist_lock);
525 	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
526 		mutex_exit(&vq->vq_freelist_lock);
527 		return NULL;
528 	}
529 	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
530 	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
531 	mutex_exit(&vq->vq_freelist_lock);
532 
533 	return qe;
534 }
535 
536 static void
537 vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
538 {
539 	mutex_enter(&vq->vq_freelist_lock);
540 	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
541 	mutex_exit(&vq->vq_freelist_lock);
542 
543 	return;
544 }
545 
546 /*
547  * Enqueue several dmamaps as a single request.
548  */
549 /*
550  * Typical usage:
551  *  <queue size> number of followings are stored in arrays
552  *  - command blocks (in dmamem) should be pre-allocated and mapped
553  *  - dmamaps for command blocks should be pre-allocated and loaded
554  *  - dmamaps for payload should be pre-allocated
555  *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
556  *	if (r)		// currently 0 or EAGAIN
557  *	  return r;
558  *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
559  *	if (r) {
560  *	  virtio_enqueue_abort(sc, vq, slot);
561  *	  return r;
562  *	}
563  *	r = virtio_enqueue_reserve(sc, vq, slot,
564  *				   dmamap_payload[slot]->dm_nsegs+1);
565  *							// ^ +1 for command
566  *	if (r) {	// currently 0 or EAGAIN
567  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
568  *	  return r;					// do not call abort()
569  *	}
570  *	<setup and prepare commands>
571  *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
572  *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
573  *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
574  *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
575  *	virtio_enqueue_commit(sc, vq, slot, true);
576  */
577 
578 /*
579  * enqueue_prep: allocate a slot number
580  */
581 int
582 virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
583 {
584 	struct vq_entry *qe1;
585 
586 	KASSERT(slotp != NULL);
587 
588 	qe1 = vq_alloc_entry(vq);
589 	if (qe1 == NULL)
590 		return EAGAIN;
591 	/* next slot is not allocated yet */
592 	qe1->qe_next = -1;
593 	*slotp = qe1->qe_index;
594 
595 	return 0;
596 }
597 
598 /*
599  * enqueue_reserve: allocate remaining slots and build the descriptor chain.
600  */
601 int
602 virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
603 		       int slot, int nsegs)
604 {
605 	int indirect;
606 	struct vq_entry *qe1 = &vq->vq_entries[slot];
607 
608 	KASSERT(qe1->qe_next == -1);
609 	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
610 
611 	if ((vq->vq_indirect != NULL) &&
612 	    (nsegs >= MINSEG_INDIRECT) &&
613 	    (nsegs <= vq->vq_maxnsegs))
614 		indirect = 1;
615 	else
616 		indirect = 0;
617 	qe1->qe_indirect = indirect;
618 
619 	if (indirect) {
620 		struct vring_desc *vd;
621 		int i;
622 
623 		vd = &vq->vq_desc[qe1->qe_index];
624 		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
625 			+ vq->vq_indirectoffset;
626 		vd->addr += sizeof(struct vring_desc)
627 			* vq->vq_maxnsegs * qe1->qe_index;
628 		vd->len = sizeof(struct vring_desc) * nsegs;
629 		vd->flags = VRING_DESC_F_INDIRECT;
630 
631 		vd = vq->vq_indirect;
632 		vd += vq->vq_maxnsegs * qe1->qe_index;
633 		qe1->qe_desc_base = vd;
634 
635 		for (i = 0; i < nsegs-1; i++) {
636 			vd[i].flags = VRING_DESC_F_NEXT;
637 		}
638 		vd[i].flags = 0;
639 		qe1->qe_next = 0;
640 
641 		return 0;
642 	} else {
643 		struct vring_desc *vd;
644 		struct vq_entry *qe;
645 		int i, s;
646 
647 		vd = &vq->vq_desc[0];
648 		qe1->qe_desc_base = vd;
649 		qe1->qe_next = qe1->qe_index;
650 		s = slot;
651 		for (i = 0; i < nsegs - 1; i++) {
652 			qe = vq_alloc_entry(vq);
653 			if (qe == NULL) {
654 				vd[s].flags = 0;
655 				virtio_enqueue_abort(sc, vq, slot);
656 				return EAGAIN;
657 			}
658 			vd[s].flags = VRING_DESC_F_NEXT;
659 			vd[s].next = qe->qe_index;
660 			s = qe->qe_index;
661 		}
662 		vd[s].flags = 0;
663 
664 		return 0;
665 	}
666 }
667 
668 /*
669  * enqueue: enqueue a single dmamap.
670  */
671 int
672 virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
673 	       bus_dmamap_t dmamap, bool write)
674 {
675 	struct vq_entry *qe1 = &vq->vq_entries[slot];
676 	struct vring_desc *vd = qe1->qe_desc_base;
677 	int i;
678 	int s = qe1->qe_next;
679 
680 	KASSERT(s >= 0);
681 	KASSERT(dmamap->dm_nsegs > 0);
682 
683 	for (i = 0; i < dmamap->dm_nsegs; i++) {
684 		vd[s].addr = dmamap->dm_segs[i].ds_addr;
685 		vd[s].len = dmamap->dm_segs[i].ds_len;
686 		if (!write)
687 			vd[s].flags |= VRING_DESC_F_WRITE;
688 		s = vd[s].next;
689 	}
690 	qe1->qe_next = s;
691 
692 	return 0;
693 }
694 
695 int
696 virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
697 		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
698 		 bool write)
699 {
700 	struct vq_entry *qe1 = &vq->vq_entries[slot];
701 	struct vring_desc *vd = qe1->qe_desc_base;
702 	int s = qe1->qe_next;
703 
704 	KASSERT(s >= 0);
705 	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
706 	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
707 		(dmamap->dm_segs[0].ds_len >= start + len));
708 
709 	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
710 	vd[s].len = len;
711 	if (!write)
712 		vd[s].flags |= VRING_DESC_F_WRITE;
713 	qe1->qe_next = vd[s].next;
714 
715 	return 0;
716 }
717 
718 /*
719  * enqueue_commit: add it to the aring.
720  */
721 int
722 virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
723 		      bool notifynow)
724 {
725 	struct vq_entry *qe1;
726 
727 	if (slot < 0) {
728 		mutex_enter(&vq->vq_aring_lock);
729 		goto notify;
730 	}
731 	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
732 	qe1 = &vq->vq_entries[slot];
733 	if (qe1->qe_indirect)
734 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
735 	mutex_enter(&vq->vq_aring_lock);
736 	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
737 
738 notify:
739 	if (notifynow) {
740 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
741 		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
742 		membar_producer();
743 		vq->vq_avail->idx = vq->vq_avail_idx;
744 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
745 		membar_producer();
746 		vq->vq_queued++;
747 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
748 		membar_consumer();
749 		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
750 			sc->sc_ops->kick(sc, vq->vq_index);
751 	}
752 	mutex_exit(&vq->vq_aring_lock);
753 
754 	return 0;
755 }
756 
757 /*
758  * enqueue_abort: rollback.
759  */
760 int
761 virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
762 {
763 	struct vq_entry *qe = &vq->vq_entries[slot];
764 	struct vring_desc *vd;
765 	int s;
766 
767 	if (qe->qe_next < 0) {
768 		vq_free_entry(vq, qe);
769 		return 0;
770 	}
771 
772 	s = slot;
773 	vd = &vq->vq_desc[0];
774 	while (vd[s].flags & VRING_DESC_F_NEXT) {
775 		s = vd[s].next;
776 		vq_free_entry(vq, qe);
777 		qe = &vq->vq_entries[s];
778 	}
779 	vq_free_entry(vq, qe);
780 	return 0;
781 }
782 
783 /*
784  * Dequeue a request.
785  */
786 /*
787  * dequeue: dequeue a request from uring; dmamap_sync for uring is
788  *	    already done in the interrupt handler.
789  */
790 int
791 virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
792 	       int *slotp, int *lenp)
793 {
794 	uint16_t slot, usedidx;
795 	struct vq_entry *qe;
796 
797 	if (vq->vq_used_idx == vq->vq_used->idx)
798 		return ENOENT;
799 	mutex_enter(&vq->vq_uring_lock);
800 	usedidx = vq->vq_used_idx++;
801 	mutex_exit(&vq->vq_uring_lock);
802 	usedidx %= vq->vq_num;
803 	slot = vq->vq_used->ring[usedidx].id;
804 	qe = &vq->vq_entries[slot];
805 
806 	if (qe->qe_indirect)
807 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
808 
809 	if (slotp)
810 		*slotp = slot;
811 	if (lenp)
812 		*lenp = vq->vq_used->ring[usedidx].len;
813 
814 	return 0;
815 }
816 
817 /*
818  * dequeue_commit: complete dequeue; the slot is recycled for future use.
819  *                 if you forget to call this the slot will be leaked.
820  */
821 int
822 virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
823 {
824 	struct vq_entry *qe = &vq->vq_entries[slot];
825 	struct vring_desc *vd = &vq->vq_desc[0];
826 	int s = slot;
827 
828 	while (vd[s].flags & VRING_DESC_F_NEXT) {
829 		s = vd[s].next;
830 		vq_free_entry(vq, qe);
831 		qe = &vq->vq_entries[s];
832 	}
833 	vq_free_entry(vq, qe);
834 
835 	return 0;
836 }
837 
838 /*
839  * Attach a child, fill all the members.
840  */
841 void
842 virtio_child_attach_start(struct virtio_softc *sc, device_t child, int ipl,
843 		    struct virtqueue *vqs,
844 		    virtio_callback config_change,
845 		    virtio_callback intr_hand,
846 		    int req_flags, int req_features, const char *feat_bits)
847 {
848 	char buf[256];
849 	int features;
850 
851 	sc->sc_child = child;
852 	sc->sc_ipl = ipl;
853 	sc->sc_vqs = vqs;
854 	sc->sc_config_change = config_change;
855 	sc->sc_intrhand = intr_hand;
856 	sc->sc_flags = req_flags;
857 
858 	features = virtio_negotiate_features(sc, req_features);
859 	snprintb(buf, sizeof(buf), feat_bits, features);
860 	aprint_normal(": Features: %s\n", buf);
861 	aprint_naive("\n");
862 }
863 
864 void
865 virtio_child_attach_set_vqs(struct virtio_softc *sc,
866     struct virtqueue *vqs, int nvq_pairs)
867 {
868 
869 	KASSERT(nvq_pairs == 1 ||
870 	    (sc->sc_flags & VIRTIO_F_PCI_INTR_SOFTINT) == 0);
871 	if (nvq_pairs > 1)
872 		sc->sc_child_mq = true;
873 
874 	sc->sc_vqs = vqs;
875 }
876 
877 int
878 virtio_child_attach_finish(struct virtio_softc *sc)
879 {
880 	int r;
881 
882 	r = sc->sc_ops->setup_interrupts(sc);
883 	if (r != 0) {
884 		aprint_error_dev(sc->sc_dev, "failed to setup interrupts\n");
885 		goto fail;
886 	}
887 
888 	KASSERT(sc->sc_soft_ih == NULL);
889 	if (sc->sc_flags & VIRTIO_F_PCI_INTR_SOFTINT) {
890 		u_int flags = SOFTINT_NET;
891 		if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
892 			flags |= SOFTINT_MPSAFE;
893 
894 		sc->sc_soft_ih = softint_establish(flags, virtio_soft_intr, sc);
895 		if (sc->sc_soft_ih == NULL) {
896 			sc->sc_ops->free_interrupts(sc);
897 			aprint_error_dev(sc->sc_dev,
898 			    "failed to establish soft interrupt\n");
899 			goto fail;
900 		}
901 	}
902 
903 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
904 	return 0;
905 
906 fail:
907 	if (sc->sc_soft_ih) {
908 		softint_disestablish(sc->sc_soft_ih);
909 		sc->sc_soft_ih = NULL;
910 	}
911 
912 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
913 	return 1;
914 }
915 
916 void
917 virtio_child_detach(struct virtio_softc *sc)
918 {
919 	sc->sc_child = NULL;
920 	sc->sc_vqs = NULL;
921 
922 	virtio_device_reset(sc);
923 
924 	sc->sc_ops->free_interrupts(sc);
925 
926 	if (sc->sc_soft_ih) {
927 		softint_disestablish(sc->sc_soft_ih);
928 		sc->sc_soft_ih = NULL;
929 	}
930 }
931 
932 void
933 virtio_child_attach_failed(struct virtio_softc *sc)
934 {
935 	virtio_child_detach(sc);
936 
937 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
938 
939 	sc->sc_child = VIRTIO_CHILD_FAILED;
940 }
941 
942 bus_dma_tag_t
943 virtio_dmat(struct virtio_softc *sc)
944 {
945 	return sc->sc_dmat;
946 }
947 
948 device_t
949 virtio_child(struct virtio_softc *sc)
950 {
951 	return sc->sc_child;
952 }
953 
954 int
955 virtio_intrhand(struct virtio_softc *sc)
956 {
957 	return (sc->sc_intrhand)(sc);
958 }
959 
960 uint32_t
961 virtio_features(struct virtio_softc *sc)
962 {
963 	return sc->sc_features;
964 }
965 
966 int
967 virtiobusprint(void *aux, const char *pnp)
968 {
969 	struct virtio_attach_args * const va = aux;
970 
971 	if (va->sc_childdevid == 0)
972 		return QUIET;	/* No device present */
973 
974 	if (pnp)
975 		aprint_normal("Device ID %d at %s", va->sc_childdevid, pnp);
976 
977 	return UNCONF;
978 }
979 
980 MODULE(MODULE_CLASS_DRIVER, virtio, NULL);
981 
982 #ifdef _MODULE
983 #include "ioconf.c"
984 #endif
985 
986 static int
987 virtio_modcmd(modcmd_t cmd, void *opaque)
988 {
989 	int error = 0;
990 
991 #ifdef _MODULE
992 	switch (cmd) {
993 	case MODULE_CMD_INIT:
994 		error = config_init_component(cfdriver_ioconf_virtio,
995 		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
996 		break;
997 	case MODULE_CMD_FINI:
998 		error = config_fini_component(cfdriver_ioconf_virtio,
999 		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
1000 		break;
1001 	default:
1002 		error = ENOTTY;
1003 		break;
1004 	}
1005 #endif
1006 
1007 	return error;
1008 }
1009