virtio.c revision 1.66
1/*	$NetBSD: virtio.c,v 1.66 2023/03/23 03:27:48 yamaguchi Exp $	*/
2
3/*
4 * Copyright (c) 2020 The NetBSD Foundation, Inc.
5 * Copyright (c) 2012 Stefan Fritsch, Alexander Fiveg.
6 * Copyright (c) 2010 Minoura Makoto.
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
30#include <sys/cdefs.h>
31__KERNEL_RCSID(0, "$NetBSD: virtio.c,v 1.66 2023/03/23 03:27:48 yamaguchi Exp $");
32
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/kernel.h>
36#include <sys/atomic.h>
37#include <sys/bus.h>
38#include <sys/device.h>
39#include <sys/kmem.h>
40#include <sys/module.h>
41
42#define VIRTIO_PRIVATE
43
44#include <dev/pci/virtioreg.h> /* XXX: move to non-pci */
45#include <dev/pci/virtiovar.h> /* XXX: move to non-pci */
46
47#define MINSEG_INDIRECT		2 /* use indirect if nsegs >= this value */
48
49/* incomplete list */
50static const char *virtio_device_name[] = {
51	"unknown (0)",			/*  0 */
52	"network",			/*  1 */
53	"block",			/*  2 */
54	"console",			/*  3 */
55	"entropy",			/*  4 */
56	"memory balloon",		/*  5 */
57	"I/O memory",			/*  6 */
58	"remote processor messaging",	/*  7 */
59	"SCSI",				/*  8 */
60	"9P transport",			/*  9 */
61};
62#define NDEVNAMES	__arraycount(virtio_device_name)
63
64static void	virtio_init_vq(struct virtio_softc *,
65		    struct virtqueue *, const bool);
66
67void
68virtio_set_status(struct virtio_softc *sc, int status)
69{
70	sc->sc_ops->set_status(sc, status);
71}
72
73/*
74 * Reset the device.
75 */
76/*
77 * To reset the device to a known state, do following:
78 *	virtio_reset(sc);	     // this will stop the device activity
79 *	<dequeue finished requests>; // virtio_dequeue() still can be called
80 *	<revoke pending requests in the vqs if any>;
81 *	virtio_reinit_start(sc);     // dequeue prohibitted
82 *	newfeatures = virtio_negotiate_features(sc, requestedfeatures);
83 *	<some other initialization>;
84 *	virtio_reinit_end(sc);	     // device activated; enqueue allowed
85 * Once attached, feature negotiation can only be allowed after virtio_reset.
86 */
87void
88virtio_reset(struct virtio_softc *sc)
89{
90	virtio_device_reset(sc);
91}
92
93int
94virtio_reinit_start(struct virtio_softc *sc)
95{
96	int i, r;
97
98	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
99	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
100	for (i = 0; i < sc->sc_nvqs; i++) {
101		int n;
102		struct virtqueue *vq = &sc->sc_vqs[i];
103		n = sc->sc_ops->read_queue_size(sc, vq->vq_index);
104		if (n == 0)	/* vq disappeared */
105			continue;
106		if (n != vq->vq_num) {
107			panic("%s: virtqueue size changed, vq index %d\n",
108			    device_xname(sc->sc_dev),
109			    vq->vq_index);
110		}
111		virtio_init_vq(sc, vq, true);
112		sc->sc_ops->setup_queue(sc, vq->vq_index,
113		    vq->vq_dmamap->dm_segs[0].ds_addr);
114	}
115
116	r = sc->sc_ops->setup_interrupts(sc, 1);
117	if (r != 0)
118		goto fail;
119
120	return 0;
121
122fail:
123	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
124
125	return 1;
126}
127
128void
129virtio_reinit_end(struct virtio_softc *sc)
130{
131	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
132}
133
134/*
135 * Feature negotiation.
136 */
137void
138virtio_negotiate_features(struct virtio_softc *sc, uint64_t guest_features)
139{
140	if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) &&
141	    !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */
142		guest_features |= VIRTIO_F_RING_INDIRECT_DESC;
143	sc->sc_ops->neg_features(sc, guest_features);
144	if (sc->sc_active_features & VIRTIO_F_RING_INDIRECT_DESC)
145		sc->sc_indirect = true;
146	else
147		sc->sc_indirect = false;
148}
149
150
151/*
152 * Device configuration registers readers/writers
153 */
154#if 0
155#define DPRINTFR(n, fmt, val, index, num) \
156	printf("\n%s (", n); \
157	for (int i = 0; i < num; i++) \
158		printf("%02x ", bus_space_read_1(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index+i)); \
159	printf(") -> "); printf(fmt, val); printf("\n");
160#define DPRINTFR2(n, fmt, val_s, val_n) \
161	printf("%s ", n); \
162	printf("\n        stream "); printf(fmt, val_s); printf(" norm "); printf(fmt, val_n); printf("\n");
163#else
164#define DPRINTFR(n, fmt, val, index, num)
165#define DPRINTFR2(n, fmt, val_s, val_n)
166#endif
167
168
169uint8_t
170virtio_read_device_config_1(struct virtio_softc *sc, int index)
171{
172	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
173	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
174	uint8_t val;
175
176	val = bus_space_read_1(iot, ioh, index);
177
178	DPRINTFR("read_1", "%02x", val, index, 1);
179	return val;
180}
181
182uint16_t
183virtio_read_device_config_2(struct virtio_softc *sc, int index)
184{
185	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
186	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
187	uint16_t val;
188
189	val = bus_space_read_2(iot, ioh, index);
190	if (BYTE_ORDER != sc->sc_bus_endian)
191		val = bswap16(val);
192
193	DPRINTFR("read_2", "%04x", val, index, 2);
194	DPRINTFR2("read_2", "%04x",
195	    bus_space_read_stream_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh,
196		index),
197	    bus_space_read_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index));
198	return val;
199}
200
201uint32_t
202virtio_read_device_config_4(struct virtio_softc *sc, int index)
203{
204	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
205	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
206	uint32_t val;
207
208	val = bus_space_read_4(iot, ioh, index);
209	if (BYTE_ORDER != sc->sc_bus_endian)
210		val = bswap32(val);
211
212	DPRINTFR("read_4", "%08x", val, index, 4);
213	DPRINTFR2("read_4", "%08x",
214	    bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh,
215		index),
216	    bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index));
217	return val;
218}
219
220/*
221 * The Virtio spec explicitly tells that reading and writing 8 bytes are not
222 * considered atomic and no triggers may be connected to reading or writing
223 * it. We access it using two 32 reads. See virtio spec 4.1.3.1.
224 */
225uint64_t
226virtio_read_device_config_8(struct virtio_softc *sc, int index)
227{
228	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
229	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
230	union {
231		uint64_t u64;
232		uint32_t l[2];
233	} v;
234	uint64_t val;
235
236	v.l[0] = bus_space_read_4(iot, ioh, index);
237	v.l[1] = bus_space_read_4(iot, ioh, index + 4);
238	if (sc->sc_bus_endian != sc->sc_struct_endian) {
239		v.l[0] = bswap32(v.l[0]);
240		v.l[1] = bswap32(v.l[1]);
241	}
242	val = v.u64;
243
244	if (BYTE_ORDER != sc->sc_struct_endian)
245		val = bswap64(val);
246
247	DPRINTFR("read_8", "%08"PRIx64, val, index, 8);
248	DPRINTFR2("read_8 low ", "%08x",
249	    bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh,
250		index),
251	    bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index));
252	DPRINTFR2("read_8 high ", "%08x",
253	    bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh,
254		index + 4),
255	    bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index + 4));
256	return val;
257}
258
259/*
260 * In the older virtio spec, device config registers are host endian. On newer
261 * they are little endian. Some newer devices however explicitly specify their
262 * register to always be little endian. These functions cater for these.
263 */
264uint16_t
265virtio_read_device_config_le_2(struct virtio_softc *sc, int index)
266{
267	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
268	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
269	uint16_t val;
270
271	val = bus_space_read_2(iot, ioh, index);
272	if (sc->sc_bus_endian != LITTLE_ENDIAN)
273		val = bswap16(val);
274
275	DPRINTFR("read_le_2", "%04x", val, index, 2);
276	DPRINTFR2("read_le_2", "%04x",
277	    bus_space_read_stream_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0),
278	    bus_space_read_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0));
279	return val;
280}
281
282uint32_t
283virtio_read_device_config_le_4(struct virtio_softc *sc, int index)
284{
285	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
286	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
287	uint32_t val;
288
289	val = bus_space_read_4(iot, ioh, index);
290	if (sc->sc_bus_endian != LITTLE_ENDIAN)
291		val = bswap32(val);
292
293	DPRINTFR("read_le_4", "%08x", val, index, 4);
294	DPRINTFR2("read_le_4", "%08x",
295	    bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0),
296	    bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0));
297	return val;
298}
299
300void
301virtio_write_device_config_1(struct virtio_softc *sc, int index, uint8_t value)
302{
303	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
304	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
305
306	bus_space_write_1(iot, ioh, index, value);
307}
308
309void
310virtio_write_device_config_2(struct virtio_softc *sc, int index,
311    uint16_t value)
312{
313	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
314	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
315
316	if (BYTE_ORDER != sc->sc_bus_endian)
317		value = bswap16(value);
318	bus_space_write_2(iot, ioh, index, value);
319}
320
321void
322virtio_write_device_config_4(struct virtio_softc *sc, int index,
323    uint32_t value)
324{
325	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
326	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
327
328	if (BYTE_ORDER != sc->sc_bus_endian)
329		value = bswap32(value);
330	bus_space_write_4(iot, ioh, index, value);
331}
332
333/*
334 * The Virtio spec explicitly tells that reading and writing 8 bytes are not
335 * considered atomic and no triggers may be connected to reading or writing
336 * it. We access it using two 32 bit writes. For good measure it is stated to
337 * always write lsb first just in case of a hypervisor bug. See See virtio
338 * spec 4.1.3.1.
339 */
340void
341virtio_write_device_config_8(struct virtio_softc *sc, int index,
342    uint64_t value)
343{
344	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
345	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
346	union {
347		uint64_t u64;
348		uint32_t l[2];
349	} v;
350
351	if (BYTE_ORDER != sc->sc_struct_endian)
352		value = bswap64(value);
353
354	v.u64 = value;
355	if (sc->sc_bus_endian != sc->sc_struct_endian) {
356		v.l[0] = bswap32(v.l[0]);
357		v.l[1] = bswap32(v.l[1]);
358	}
359
360	if (sc->sc_struct_endian == LITTLE_ENDIAN) {
361		bus_space_write_4(iot, ioh, index,     v.l[0]);
362		bus_space_write_4(iot, ioh, index + 4, v.l[1]);
363	} else {
364		bus_space_write_4(iot, ioh, index + 4, v.l[1]);
365		bus_space_write_4(iot, ioh, index,     v.l[0]);
366	}
367}
368
369/*
370 * In the older virtio spec, device config registers are host endian. On newer
371 * they are little endian. Some newer devices however explicitly specify their
372 * register to always be little endian. These functions cater for these.
373 */
374void
375virtio_write_device_config_le_2(struct virtio_softc *sc, int index,
376    uint16_t value)
377{
378	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
379	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
380
381	if (sc->sc_bus_endian != LITTLE_ENDIAN)
382		value = bswap16(value);
383	bus_space_write_2(iot, ioh, index, value);
384}
385
386void
387virtio_write_device_config_le_4(struct virtio_softc *sc, int index,
388    uint32_t value)
389{
390	bus_space_tag_t	   iot = sc->sc_devcfg_iot;
391	bus_space_handle_t ioh = sc->sc_devcfg_ioh;
392
393	if (sc->sc_bus_endian != LITTLE_ENDIAN)
394		value = bswap32(value);
395	bus_space_write_4(iot, ioh, index, value);
396}
397
398
399/*
400 * data structures endian helpers
401 */
402uint16_t
403virtio_rw16(struct virtio_softc *sc, uint16_t val)
404{
405	KASSERT(sc);
406	return BYTE_ORDER != sc->sc_struct_endian ? bswap16(val) : val;
407}
408
409uint32_t
410virtio_rw32(struct virtio_softc *sc, uint32_t val)
411{
412	KASSERT(sc);
413	return BYTE_ORDER != sc->sc_struct_endian ? bswap32(val) : val;
414}
415
416uint64_t
417virtio_rw64(struct virtio_softc *sc, uint64_t val)
418{
419	KASSERT(sc);
420	return BYTE_ORDER != sc->sc_struct_endian ? bswap64(val) : val;
421}
422
423
424/*
425 * Interrupt handler.
426 */
427static void
428virtio_soft_intr(void *arg)
429{
430	struct virtio_softc *sc = arg;
431
432	KASSERT(sc->sc_intrhand != NULL);
433
434	(*sc->sc_intrhand)(sc);
435}
436
437/*
438 * dmamap sync operations for a virtqueue.
439 */
440static inline void
441vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops)
442{
443
444	/* availoffset == sizeof(vring_desc) * vq_num */
445	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset,
446	    ops);
447}
448
449static inline void
450vq_sync_aring_all(struct virtio_softc *sc, struct virtqueue *vq, int ops)
451{
452	uint16_t hdrlen = offsetof(struct vring_avail, ring);
453	size_t payloadlen = vq->vq_num * sizeof(uint16_t);
454	size_t usedlen = 0;
455
456	if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX)
457		usedlen = sizeof(uint16_t);
458	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
459	    vq->vq_availoffset, hdrlen + payloadlen + usedlen, ops);
460}
461
462static inline void
463vq_sync_aring_header(struct virtio_softc *sc, struct virtqueue *vq, int ops)
464{
465	uint16_t hdrlen = offsetof(struct vring_avail, ring);
466
467	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
468	    vq->vq_availoffset, hdrlen, ops);
469}
470
471static inline void
472vq_sync_aring_payload(struct virtio_softc *sc, struct virtqueue *vq, int ops)
473{
474	uint16_t hdrlen = offsetof(struct vring_avail, ring);
475	size_t payloadlen = vq->vq_num * sizeof(uint16_t);
476
477	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
478	    vq->vq_availoffset + hdrlen, payloadlen, ops);
479}
480
481static inline void
482vq_sync_aring_used(struct virtio_softc *sc, struct virtqueue *vq, int ops)
483{
484	uint16_t hdrlen = offsetof(struct vring_avail, ring);
485	size_t payloadlen = vq->vq_num * sizeof(uint16_t);
486	size_t usedlen = sizeof(uint16_t);
487
488	if ((sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) == 0)
489		return;
490	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
491	    vq->vq_availoffset + hdrlen + payloadlen, usedlen, ops);
492}
493
494static inline void
495vq_sync_uring_all(struct virtio_softc *sc, struct virtqueue *vq, int ops)
496{
497	uint16_t hdrlen = offsetof(struct vring_used, ring);
498	size_t payloadlen = vq->vq_num * sizeof(struct vring_used_elem);
499	size_t availlen = 0;
500
501	if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX)
502		availlen = sizeof(uint16_t);
503	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
504	    vq->vq_usedoffset, hdrlen + payloadlen + availlen, ops);
505}
506
507static inline void
508vq_sync_uring_header(struct virtio_softc *sc, struct virtqueue *vq, int ops)
509{
510	uint16_t hdrlen = offsetof(struct vring_used, ring);
511
512	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
513	    vq->vq_usedoffset, hdrlen, ops);
514}
515
516static inline void
517vq_sync_uring_payload(struct virtio_softc *sc, struct virtqueue *vq, int ops)
518{
519	uint16_t hdrlen = offsetof(struct vring_used, ring);
520	size_t payloadlen = vq->vq_num * sizeof(struct vring_used_elem);
521
522	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
523	    vq->vq_usedoffset + hdrlen, payloadlen, ops);
524}
525
526static inline void
527vq_sync_uring_avail(struct virtio_softc *sc, struct virtqueue *vq, int ops)
528{
529	uint16_t hdrlen = offsetof(struct vring_used, ring);
530	size_t payloadlen = vq->vq_num * sizeof(struct vring_used_elem);
531	size_t availlen = sizeof(uint16_t);
532
533	if ((sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) == 0)
534		return;
535	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
536	    vq->vq_usedoffset + hdrlen + payloadlen, availlen, ops);
537}
538
539static inline void
540vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot,
541    int ops)
542{
543	int offset = vq->vq_indirectoffset +
544	    sizeof(struct vring_desc) * vq->vq_maxnsegs * slot;
545
546	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
547	    offset, sizeof(struct vring_desc) * vq->vq_maxnsegs, ops);
548}
549
550bool
551virtio_vq_is_enqueued(struct virtio_softc *sc, struct virtqueue *vq)
552{
553
554	if (vq->vq_queued) {
555		vq->vq_queued = 0;
556		vq_sync_aring_all(sc, vq, BUS_DMASYNC_POSTWRITE);
557	}
558
559	vq_sync_uring_header(sc, vq, BUS_DMASYNC_POSTREAD);
560	if (vq->vq_used_idx == virtio_rw16(sc, vq->vq_used->idx))
561		return 0;
562	vq_sync_uring_payload(sc, vq, BUS_DMASYNC_POSTREAD);
563	return 1;
564}
565
566/*
567 * Scan vq, bus_dmamap_sync for the vqs (not for the payload),
568 * and calls (*vq_done)() if some entries are consumed.
569 *
570 * Can be used as sc_intrhand.
571 */
572int
573virtio_vq_intr(struct virtio_softc *sc)
574{
575	struct virtqueue *vq;
576	int i, r = 0;
577
578	for (i = 0; i < sc->sc_nvqs; i++) {
579		vq = &sc->sc_vqs[i];
580		if (virtio_vq_is_enqueued(sc, vq) == 1) {
581			if (vq->vq_done)
582				r |= (*vq->vq_done)(vq);
583		}
584	}
585
586	return r;
587}
588
589int
590virtio_vq_intrhand(struct virtio_softc *sc)
591{
592	struct virtqueue *vq;
593	int i, r = 0;
594
595	for (i = 0; i < sc->sc_nvqs; i++) {
596		vq = &sc->sc_vqs[i];
597		r |= (*vq->vq_intrhand)(vq->vq_intrhand_arg);
598	}
599
600	return r;
601}
602
603
604/*
605 * Increase the event index in order to delay interrupts.
606 */
607int
608virtio_postpone_intr(struct virtio_softc *sc, struct virtqueue *vq,
609    uint16_t nslots)
610{
611	uint16_t	idx, nused;
612
613	idx = vq->vq_used_idx + nslots;
614
615	/* set the new event index: avail_ring->used_event = idx */
616	*vq->vq_used_event = virtio_rw16(sc, idx);
617	vq_sync_aring_used(vq->vq_owner, vq, BUS_DMASYNC_PREWRITE);
618	vq->vq_queued++;
619
620	nused = (uint16_t)
621	    (virtio_rw16(sc, vq->vq_used->idx) - vq->vq_used_idx);
622	KASSERT(nused <= vq->vq_num);
623
624	return nslots < nused;
625}
626
627/*
628 * Postpone interrupt until 3/4 of the available descriptors have been
629 * consumed.
630 */
631int
632virtio_postpone_intr_smart(struct virtio_softc *sc, struct virtqueue *vq)
633{
634	uint16_t	nslots;
635
636	nslots = (uint16_t)
637	    (virtio_rw16(sc, vq->vq_avail->idx) - vq->vq_used_idx) * 3 / 4;
638
639	return virtio_postpone_intr(sc, vq, nslots);
640}
641
642/*
643 * Postpone interrupt until all of the available descriptors have been
644 * consumed.
645 */
646int
647virtio_postpone_intr_far(struct virtio_softc *sc, struct virtqueue *vq)
648{
649	uint16_t	nslots;
650
651	nslots = (uint16_t)
652	    (virtio_rw16(sc, vq->vq_avail->idx) - vq->vq_used_idx);
653
654	return virtio_postpone_intr(sc, vq, nslots);
655}
656
657/*
658 * Start/stop vq interrupt.  No guarantee.
659 */
660void
661virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
662{
663
664	if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) {
665		/*
666		 * No way to disable the interrupt completely with
667		 * RingEventIdx. Instead advance used_event by half the
668		 * possible value. This won't happen soon and is far enough in
669		 * the past to not trigger a spurios interrupt.
670		 */
671		*vq->vq_used_event = virtio_rw16(sc, vq->vq_used_idx + 0x8000);
672		vq_sync_aring_used(sc, vq, BUS_DMASYNC_PREWRITE);
673	} else {
674		vq->vq_avail->flags |=
675		    virtio_rw16(sc, VRING_AVAIL_F_NO_INTERRUPT);
676		vq_sync_aring_header(sc, vq, BUS_DMASYNC_PREWRITE);
677	}
678	vq->vq_queued++;
679}
680
681int
682virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
683{
684
685	if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) {
686		/*
687		 * If event index feature is negotiated, enabling interrupts
688		 * is done through setting the latest consumed index in the
689		 * used_event field
690		 */
691		*vq->vq_used_event = virtio_rw16(sc, vq->vq_used_idx);
692		vq_sync_aring_used(sc, vq, BUS_DMASYNC_PREWRITE);
693	} else {
694		vq->vq_avail->flags &=
695		    ~virtio_rw16(sc, VRING_AVAIL_F_NO_INTERRUPT);
696		vq_sync_aring_header(sc, vq, BUS_DMASYNC_PREWRITE);
697	}
698	vq->vq_queued++;
699
700	vq_sync_uring_header(sc, vq, BUS_DMASYNC_POSTREAD);
701	if (vq->vq_used_idx == virtio_rw16(sc, vq->vq_used->idx))
702		return 0;
703	vq_sync_uring_payload(sc, vq, BUS_DMASYNC_POSTREAD);
704	return 1;
705}
706
707/*
708 * Initialize vq structure.
709 */
710static void
711virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq,
712    const bool reinit)
713{
714	int i, j;
715	int vq_size = vq->vq_num;
716
717	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
718
719	/* build the indirect descriptor chain */
720	if (vq->vq_indirect != NULL) {
721		struct vring_desc *vd;
722
723		for (i = 0; i < vq_size; i++) {
724			vd = vq->vq_indirect;
725			vd += vq->vq_maxnsegs * i;
726			for (j = 0; j < vq->vq_maxnsegs - 1; j++) {
727				vd[j].next = virtio_rw16(sc, j + 1);
728			}
729		}
730	}
731
732	/* free slot management */
733	SIMPLEQ_INIT(&vq->vq_freelist);
734	for (i = 0; i < vq_size; i++) {
735		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, &vq->vq_entries[i],
736		    qe_list);
737		vq->vq_entries[i].qe_index = i;
738	}
739	if (!reinit)
740		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
741
742	/* enqueue/dequeue status */
743	vq->vq_avail_idx = 0;
744	vq->vq_used_idx = 0;
745	vq->vq_queued = 0;
746	if (!reinit) {
747		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
748		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
749	}
750	vq_sync_uring_all(sc, vq, BUS_DMASYNC_PREREAD);
751	vq->vq_queued++;
752}
753
754/*
755 * Allocate/free a vq.
756 */
757int
758virtio_alloc_vq(struct virtio_softc *sc, struct virtqueue *vq, int index,
759    int maxsegsize, int maxnsegs, const char *name)
760{
761	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
762	int rsegs, r, hdrlen;
763#define VIRTQUEUE_ALIGN(n)	roundup(n, VIRTIO_PAGE_SIZE)
764
765	memset(vq, 0, sizeof(*vq));
766
767	vq_size = sc->sc_ops->read_queue_size(sc, index);
768	if (vq_size == 0) {
769		aprint_error_dev(sc->sc_dev,
770		    "virtqueue not exist, index %d for %s\n",
771		    index, name);
772		goto err;
773	}
774
775	hdrlen = sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX ? 3 : 2;
776
777	/* allocsize1: descriptor table + avail ring + pad */
778	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc) * vq_size
779	    + sizeof(uint16_t) * (hdrlen + vq_size));
780	/* allocsize2: used ring + pad */
781	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t) * hdrlen
782	    + sizeof(struct vring_used_elem) * vq_size);
783	/* allocsize3: indirect table */
784	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
785		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
786	else
787		allocsize3 = 0;
788	allocsize = allocsize1 + allocsize2 + allocsize3;
789
790	/* alloc and map the memory */
791	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
792	    &vq->vq_segs[0], 1, &rsegs, BUS_DMA_WAITOK);
793	if (r != 0) {
794		aprint_error_dev(sc->sc_dev,
795		    "virtqueue %d for %s allocation failed, "
796		    "error code %d\n", index, name, r);
797		goto err;
798	}
799	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], rsegs, allocsize,
800	    &vq->vq_vaddr, BUS_DMA_WAITOK);
801	if (r != 0) {
802		aprint_error_dev(sc->sc_dev,
803		    "virtqueue %d for %s map failed, "
804		    "error code %d\n", index, name, r);
805		goto err;
806	}
807	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
808	    BUS_DMA_WAITOK, &vq->vq_dmamap);
809	if (r != 0) {
810		aprint_error_dev(sc->sc_dev,
811		    "virtqueue %d for %s dmamap creation failed, "
812		    "error code %d\n", index, name, r);
813		goto err;
814	}
815	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
816	    vq->vq_vaddr, allocsize, NULL, BUS_DMA_WAITOK);
817	if (r != 0) {
818		aprint_error_dev(sc->sc_dev,
819		    "virtqueue %d for %s dmamap load failed, "
820		    "error code %d\n", index, name, r);
821		goto err;
822	}
823
824	/* remember addresses and offsets for later use */
825	vq->vq_owner = sc;
826	vq->vq_num = vq_size;
827	vq->vq_index = index;
828	vq->vq_desc = vq->vq_vaddr;
829	vq->vq_availoffset = sizeof(struct vring_desc) * vq_size;
830	vq->vq_avail = (void *)(((char *)vq->vq_desc) + vq->vq_availoffset);
831	vq->vq_used_event = (uint16_t *)((char *)vq->vq_avail +
832	    offsetof(struct vring_avail, ring[vq->vq_num]));
833	vq->vq_usedoffset = allocsize1;
834	vq->vq_used = (void *)(((char *)vq->vq_desc) + vq->vq_usedoffset);
835	vq->vq_avail_event = (uint16_t *)((char *)vq->vq_used +
836	    offsetof(struct vring_used, ring[vq->vq_num]));
837
838	if (allocsize3 > 0) {
839		vq->vq_indirectoffset = allocsize1 + allocsize2;
840		vq->vq_indirect = (void *)(((char *)vq->vq_desc)
841		    + vq->vq_indirectoffset);
842	}
843	vq->vq_bytesize = allocsize;
844	vq->vq_maxsegsize = maxsegsize;
845	vq->vq_maxnsegs = maxnsegs;
846
847	/* free slot management */
848	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry) * vq_size,
849	    KM_SLEEP);
850	virtio_init_vq(sc, vq, false);
851
852	/* set the vq address */
853	sc->sc_ops->setup_queue(sc, index,
854	    vq->vq_dmamap->dm_segs[0].ds_addr);
855
856	aprint_verbose_dev(sc->sc_dev,
857	    "allocated %u byte for virtqueue %d for %s, size %d\n",
858	    allocsize, index, name, vq_size);
859	if (allocsize3 > 0)
860		aprint_verbose_dev(sc->sc_dev,
861		    "using %d byte (%d entries) indirect descriptors\n",
862		    allocsize3, maxnsegs * vq_size);
863
864	return 0;
865
866err:
867	sc->sc_ops->setup_queue(sc, index, 0);
868	if (vq->vq_dmamap)
869		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
870	if (vq->vq_vaddr)
871		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
872	if (vq->vq_segs[0].ds_addr)
873		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
874	memset(vq, 0, sizeof(*vq));
875
876	return -1;
877}
878
879int
880virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
881{
882	struct vq_entry *qe;
883	int i = 0;
884
885	/* device must be already deactivated */
886	/* confirm the vq is empty */
887	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
888		i++;
889	}
890	if (i != vq->vq_num) {
891		printf("%s: freeing non-empty vq, index %d\n",
892		    device_xname(sc->sc_dev), vq->vq_index);
893		return EBUSY;
894	}
895
896	/* tell device that there's no virtqueue any longer */
897	sc->sc_ops->setup_queue(sc, vq->vq_index, 0);
898
899	vq_sync_aring_all(sc, vq, BUS_DMASYNC_POSTWRITE);
900
901	kmem_free(vq->vq_entries, sizeof(*vq->vq_entries) * vq->vq_num);
902	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
903	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
904	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
905	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
906	mutex_destroy(&vq->vq_freelist_lock);
907	mutex_destroy(&vq->vq_uring_lock);
908	mutex_destroy(&vq->vq_aring_lock);
909	memset(vq, 0, sizeof(*vq));
910
911	return 0;
912}
913
914/*
915 * Free descriptor management.
916 */
917static struct vq_entry *
918vq_alloc_entry(struct virtqueue *vq)
919{
920	struct vq_entry *qe;
921
922	mutex_enter(&vq->vq_freelist_lock);
923	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
924		mutex_exit(&vq->vq_freelist_lock);
925		return NULL;
926	}
927	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
928	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
929	mutex_exit(&vq->vq_freelist_lock);
930
931	return qe;
932}
933
934static void
935vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
936{
937	mutex_enter(&vq->vq_freelist_lock);
938	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
939	mutex_exit(&vq->vq_freelist_lock);
940
941	return;
942}
943
944/*
945 * Enqueue several dmamaps as a single request.
946 */
947/*
948 * Typical usage:
949 *  <queue size> number of followings are stored in arrays
950 *  - command blocks (in dmamem) should be pre-allocated and mapped
951 *  - dmamaps for command blocks should be pre-allocated and loaded
952 *  - dmamaps for payload should be pre-allocated
953 *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
954 *	if (r)		// currently 0 or EAGAIN
955 *		return r;
956 *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
957 *	if (r) {
958 *		virtio_enqueue_abort(sc, vq, slot);
959 *		return r;
960 *	}
961 *	r = virtio_enqueue_reserve(sc, vq, slot,
962 *	    dmamap_payload[slot]->dm_nsegs + 1);
963 *							// ^ +1 for command
964 *	if (r) {	// currently 0 or EAGAIN
965 *		bus_dmamap_unload(dmat, dmamap_payload[slot]);
966 *		return r;				// do not call abort()
967 *	}
968 *	<setup and prepare commands>
969 *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
970 *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
971 *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
972 *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
973 *	virtio_enqueue_commit(sc, vq, slot, true);
974 */
975
976/*
977 * enqueue_prep: allocate a slot number
978 */
979int
980virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
981{
982	struct vq_entry *qe1;
983
984	KASSERT(slotp != NULL);
985
986	qe1 = vq_alloc_entry(vq);
987	if (qe1 == NULL)
988		return EAGAIN;
989	/* next slot is not allocated yet */
990	qe1->qe_next = -1;
991	*slotp = qe1->qe_index;
992
993	return 0;
994}
995
996/*
997 * enqueue_reserve: allocate remaining slots and build the descriptor chain.
998 */
999int
1000virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
1001    int slot, int nsegs)
1002{
1003	int indirect;
1004	struct vq_entry *qe1 = &vq->vq_entries[slot];
1005
1006	KASSERT(qe1->qe_next == -1);
1007	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
1008
1009	if ((vq->vq_indirect != NULL) &&
1010	    (nsegs >= MINSEG_INDIRECT) &&
1011	    (nsegs <= vq->vq_maxnsegs))
1012		indirect = 1;
1013	else
1014		indirect = 0;
1015	qe1->qe_indirect = indirect;
1016
1017	if (indirect) {
1018		struct vring_desc *vd;
1019		uint64_t addr;
1020		int i;
1021
1022		vd = &vq->vq_desc[qe1->qe_index];
1023		addr = vq->vq_dmamap->dm_segs[0].ds_addr
1024		    + vq->vq_indirectoffset;
1025		addr += sizeof(struct vring_desc)
1026		    * vq->vq_maxnsegs * qe1->qe_index;
1027		vd->addr  = virtio_rw64(sc, addr);
1028		vd->len   = virtio_rw32(sc, sizeof(struct vring_desc) * nsegs);
1029		vd->flags = virtio_rw16(sc, VRING_DESC_F_INDIRECT);
1030
1031		vd = vq->vq_indirect;
1032		vd += vq->vq_maxnsegs * qe1->qe_index;
1033		qe1->qe_desc_base = vd;
1034
1035		for (i = 0; i < nsegs - 1; i++) {
1036			vd[i].flags = virtio_rw16(sc, VRING_DESC_F_NEXT);
1037		}
1038		vd[i].flags  = virtio_rw16(sc, 0);
1039		qe1->qe_next = 0;
1040
1041		return 0;
1042	} else {
1043		struct vring_desc *vd;
1044		struct vq_entry *qe;
1045		int i, s;
1046
1047		vd = &vq->vq_desc[0];
1048		qe1->qe_desc_base = vd;
1049		qe1->qe_next = qe1->qe_index;
1050		s = slot;
1051		for (i = 0; i < nsegs - 1; i++) {
1052			qe = vq_alloc_entry(vq);
1053			if (qe == NULL) {
1054				vd[s].flags = virtio_rw16(sc, 0);
1055				virtio_enqueue_abort(sc, vq, slot);
1056				return EAGAIN;
1057			}
1058			vd[s].flags = virtio_rw16(sc, VRING_DESC_F_NEXT);
1059			vd[s].next  = virtio_rw16(sc, qe->qe_index);
1060			s = qe->qe_index;
1061		}
1062		vd[s].flags = virtio_rw16(sc, 0);
1063
1064		return 0;
1065	}
1066}
1067
1068/*
1069 * enqueue: enqueue a single dmamap.
1070 */
1071int
1072virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1073    bus_dmamap_t dmamap, bool write)
1074{
1075	struct vq_entry *qe1 = &vq->vq_entries[slot];
1076	struct vring_desc *vd = qe1->qe_desc_base;
1077	int i;
1078	int s = qe1->qe_next;
1079
1080	KASSERT(s >= 0);
1081	KASSERT(dmamap->dm_nsegs > 0);
1082
1083	for (i = 0; i < dmamap->dm_nsegs; i++) {
1084		vd[s].addr = virtio_rw64(sc, dmamap->dm_segs[i].ds_addr);
1085		vd[s].len  = virtio_rw32(sc, dmamap->dm_segs[i].ds_len);
1086		if (!write)
1087			vd[s].flags |= virtio_rw16(sc, VRING_DESC_F_WRITE);
1088		s = virtio_rw16(sc, vd[s].next);
1089	}
1090	qe1->qe_next = s;
1091
1092	return 0;
1093}
1094
1095int
1096virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1097    bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
1098    bool write)
1099{
1100	struct vq_entry *qe1 = &vq->vq_entries[slot];
1101	struct vring_desc *vd = qe1->qe_desc_base;
1102	int s = qe1->qe_next;
1103
1104	KASSERT(s >= 0);
1105	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
1106	KASSERT(dmamap->dm_segs[0].ds_len > start);
1107	KASSERT(dmamap->dm_segs[0].ds_len >= start + len);
1108
1109	vd[s].addr = virtio_rw64(sc, dmamap->dm_segs[0].ds_addr + start);
1110	vd[s].len  = virtio_rw32(sc, len);
1111	if (!write)
1112		vd[s].flags |= virtio_rw16(sc, VRING_DESC_F_WRITE);
1113	qe1->qe_next = virtio_rw16(sc, vd[s].next);
1114
1115	return 0;
1116}
1117
1118/*
1119 * enqueue_commit: add it to the aring.
1120 */
1121int
1122virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1123    bool notifynow)
1124{
1125	struct vq_entry *qe1;
1126
1127	if (slot < 0) {
1128		mutex_enter(&vq->vq_aring_lock);
1129		goto notify;
1130	}
1131	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
1132	qe1 = &vq->vq_entries[slot];
1133	if (qe1->qe_indirect)
1134		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
1135	mutex_enter(&vq->vq_aring_lock);
1136	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] =
1137	    virtio_rw16(sc, slot);
1138
1139notify:
1140	if (notifynow) {
1141		uint16_t o, n, t;
1142		uint16_t flags;
1143
1144		o = virtio_rw16(sc, vq->vq_avail->idx) - 1;
1145		n = vq->vq_avail_idx;
1146
1147		/*
1148		 * Prepare for `device->CPU' (host->guest) transfer
1149		 * into the buffer.  This must happen before we commit
1150		 * the vq->vq_avail->idx update to ensure we're not
1151		 * still using the buffer in case program-prior loads
1152		 * or stores in it get delayed past the store to
1153		 * vq->vq_avail->idx.
1154		 */
1155		vq_sync_uring_all(sc, vq, BUS_DMASYNC_PREREAD);
1156
1157		/* ensure payload is published, then avail idx */
1158		vq_sync_aring_payload(sc, vq, BUS_DMASYNC_PREWRITE);
1159		vq->vq_avail->idx = virtio_rw16(sc, vq->vq_avail_idx);
1160		vq_sync_aring_header(sc, vq, BUS_DMASYNC_PREWRITE);
1161		vq->vq_queued++;
1162
1163		if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) {
1164			vq_sync_uring_avail(sc, vq, BUS_DMASYNC_POSTREAD);
1165			t = virtio_rw16(sc, *vq->vq_avail_event) + 1;
1166			if ((uint16_t) (n - t) < (uint16_t) (n - o))
1167				sc->sc_ops->kick(sc, vq->vq_index);
1168		} else {
1169			vq_sync_uring_header(sc, vq, BUS_DMASYNC_POSTREAD);
1170			flags = virtio_rw16(sc, vq->vq_used->flags);
1171			if (!(flags & VRING_USED_F_NO_NOTIFY))
1172				sc->sc_ops->kick(sc, vq->vq_index);
1173		}
1174	}
1175	mutex_exit(&vq->vq_aring_lock);
1176
1177	return 0;
1178}
1179
1180/*
1181 * enqueue_abort: rollback.
1182 */
1183int
1184virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1185{
1186	struct vq_entry *qe = &vq->vq_entries[slot];
1187	struct vring_desc *vd;
1188	int s;
1189
1190	if (qe->qe_next < 0) {
1191		vq_free_entry(vq, qe);
1192		return 0;
1193	}
1194
1195	s = slot;
1196	vd = &vq->vq_desc[0];
1197	while (virtio_rw16(sc, vd[s].flags) & VRING_DESC_F_NEXT) {
1198		s = virtio_rw16(sc, vd[s].next);
1199		vq_free_entry(vq, qe);
1200		qe = &vq->vq_entries[s];
1201	}
1202	vq_free_entry(vq, qe);
1203	return 0;
1204}
1205
1206/*
1207 * Dequeue a request.
1208 */
1209/*
1210 * dequeue: dequeue a request from uring; dmamap_sync for uring is
1211 *	    already done in the interrupt handler.
1212 */
1213int
1214virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
1215    int *slotp, int *lenp)
1216{
1217	uint16_t slot, usedidx;
1218	struct vq_entry *qe;
1219
1220	if (vq->vq_used_idx == virtio_rw16(sc, vq->vq_used->idx))
1221		return ENOENT;
1222	mutex_enter(&vq->vq_uring_lock);
1223	usedidx = vq->vq_used_idx++;
1224	mutex_exit(&vq->vq_uring_lock);
1225	usedidx %= vq->vq_num;
1226	slot = virtio_rw32(sc, vq->vq_used->ring[usedidx].id);
1227	qe = &vq->vq_entries[slot];
1228
1229	if (qe->qe_indirect)
1230		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
1231
1232	if (slotp)
1233		*slotp = slot;
1234	if (lenp)
1235		*lenp = virtio_rw32(sc, vq->vq_used->ring[usedidx].len);
1236
1237	return 0;
1238}
1239
1240/*
1241 * dequeue_commit: complete dequeue; the slot is recycled for future use.
1242 *                 if you forget to call this the slot will be leaked.
1243 */
1244int
1245virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1246{
1247	struct vq_entry *qe = &vq->vq_entries[slot];
1248	struct vring_desc *vd = &vq->vq_desc[0];
1249	int s = slot;
1250
1251	while (virtio_rw16(sc, vd[s].flags) & VRING_DESC_F_NEXT) {
1252		s = virtio_rw16(sc, vd[s].next);
1253		vq_free_entry(vq, qe);
1254		qe = &vq->vq_entries[s];
1255	}
1256	vq_free_entry(vq, qe);
1257
1258	return 0;
1259}
1260
1261/*
1262 * Attach a child, fill all the members.
1263 */
1264void
1265virtio_child_attach_start(struct virtio_softc *sc, device_t child, int ipl,
1266    uint64_t req_features, const char *feat_bits)
1267{
1268	char buf[1024];
1269
1270	sc->sc_child = child;
1271	sc->sc_ipl = ipl;
1272
1273	virtio_negotiate_features(sc, req_features);
1274	snprintb(buf, sizeof(buf), feat_bits, sc->sc_active_features);
1275	aprint_normal(": features: %s\n", buf);
1276	aprint_naive("\n");
1277}
1278
1279int
1280virtio_child_attach_finish(struct virtio_softc *sc,
1281    struct virtqueue *vqs, size_t nvqs,
1282    virtio_callback config_change, virtio_callback intr_hand,
1283    int req_flags)
1284{
1285	int r;
1286
1287#ifdef DIAGNOSTIC
1288	KASSERT(nvqs > 0);
1289#define VIRTIO_ASSERT_FLAGS	(VIRTIO_F_INTR_SOFTINT | VIRTIO_F_INTR_PERVQ)
1290	KASSERT((req_flags & VIRTIO_ASSERT_FLAGS) != VIRTIO_ASSERT_FLAGS);
1291#undef VIRTIO_ASSERT_FLAGS
1292
1293	for (size_t _i = 0; _i < nvqs; _i++){
1294		KASSERT(vqs[_i].vq_index == _i);
1295		KASSERT((req_flags & VIRTIO_F_INTR_PERVQ) == 0 ||
1296		    vqs[_i].vq_intrhand != NULL);
1297	}
1298#endif
1299
1300	sc->sc_finished_called = true;
1301
1302	sc->sc_vqs = vqs;
1303	sc->sc_nvqs = nvqs;
1304	sc->sc_config_change = config_change;
1305	sc->sc_intrhand = intr_hand;
1306	sc->sc_flags = req_flags;
1307
1308	r = sc->sc_ops->alloc_interrupts(sc);
1309	if (r != 0) {
1310		aprint_error_dev(sc->sc_dev,
1311		    "failed to allocate interrupts\n");
1312		goto fail;
1313	}
1314
1315	r = sc->sc_ops->setup_interrupts(sc, 0);
1316	if (r != 0) {
1317		aprint_error_dev(sc->sc_dev, "failed to setup interrupts\n");
1318		goto fail;
1319	}
1320
1321	KASSERT(sc->sc_soft_ih == NULL);
1322	if (sc->sc_flags & VIRTIO_F_INTR_SOFTINT) {
1323		u_int flags = SOFTINT_NET;
1324		if (sc->sc_flags & VIRTIO_F_INTR_MPSAFE)
1325			flags |= SOFTINT_MPSAFE;
1326
1327		sc->sc_soft_ih = softint_establish(flags, virtio_soft_intr,
1328		    sc);
1329		if (sc->sc_soft_ih == NULL) {
1330			sc->sc_ops->free_interrupts(sc);
1331			aprint_error_dev(sc->sc_dev,
1332			    "failed to establish soft interrupt\n");
1333			goto fail;
1334		}
1335	}
1336
1337	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
1338	return 0;
1339
1340fail:
1341	if (sc->sc_soft_ih) {
1342		softint_disestablish(sc->sc_soft_ih);
1343		sc->sc_soft_ih = NULL;
1344	}
1345
1346	sc->sc_ops->free_interrupts(sc);
1347
1348	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
1349	return 1;
1350}
1351
1352void
1353virtio_child_detach(struct virtio_softc *sc)
1354{
1355	sc->sc_child = NULL;
1356	sc->sc_vqs = NULL;
1357
1358	virtio_device_reset(sc);
1359
1360	sc->sc_ops->free_interrupts(sc);
1361
1362	if (sc->sc_soft_ih) {
1363		softint_disestablish(sc->sc_soft_ih);
1364		sc->sc_soft_ih = NULL;
1365	}
1366}
1367
1368void
1369virtio_child_attach_failed(struct virtio_softc *sc)
1370{
1371	virtio_child_detach(sc);
1372
1373	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
1374
1375	sc->sc_child = VIRTIO_CHILD_FAILED;
1376}
1377
1378bus_dma_tag_t
1379virtio_dmat(struct virtio_softc *sc)
1380{
1381	return sc->sc_dmat;
1382}
1383
1384device_t
1385virtio_child(struct virtio_softc *sc)
1386{
1387	return sc->sc_child;
1388}
1389
1390int
1391virtio_intrhand(struct virtio_softc *sc)
1392{
1393	return (*sc->sc_intrhand)(sc);
1394}
1395
1396uint64_t
1397virtio_features(struct virtio_softc *sc)
1398{
1399	return sc->sc_active_features;
1400}
1401
1402int
1403virtio_attach_failed(struct virtio_softc *sc)
1404{
1405	device_t self = sc->sc_dev;
1406
1407	/* no error if its not connected, but its failed */
1408	if (sc->sc_childdevid == 0)
1409		return 1;
1410
1411	if (sc->sc_child == NULL) {
1412		aprint_error_dev(self,
1413		    "no matching child driver; not configured\n");
1414		return 1;
1415	}
1416
1417	if (sc->sc_child == VIRTIO_CHILD_FAILED) {
1418		aprint_error_dev(self, "virtio configuration failed\n");
1419		return 1;
1420	}
1421
1422	/* sanity check */
1423	if (!sc->sc_finished_called) {
1424		aprint_error_dev(self, "virtio internal error, child driver "
1425		    "signaled OK but didn't initialize interrupts\n");
1426		return 1;
1427	}
1428
1429	return 0;
1430}
1431
1432void
1433virtio_print_device_type(device_t self, int id, int revision)
1434{
1435	aprint_normal_dev(self, "%s device (id %d, rev. 0x%02x)\n",
1436	    (id < NDEVNAMES ? virtio_device_name[id] : "Unknown"),
1437	    id,
1438	    revision);
1439}
1440
1441
1442MODULE(MODULE_CLASS_DRIVER, virtio, NULL);
1443
1444#ifdef _MODULE
1445#include "ioconf.c"
1446#endif
1447
1448static int
1449virtio_modcmd(modcmd_t cmd, void *opaque)
1450{
1451	int error = 0;
1452
1453#ifdef _MODULE
1454	switch (cmd) {
1455	case MODULE_CMD_INIT:
1456		error = config_init_component(cfdriver_ioconf_virtio,
1457		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
1458		break;
1459	case MODULE_CMD_FINI:
1460		error = config_fini_component(cfdriver_ioconf_virtio,
1461		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
1462		break;
1463	default:
1464		error = ENOTTY;
1465		break;
1466	}
1467#endif
1468
1469	return error;
1470}
1471