virtio.c revision 1.23
1/*	$NetBSD: virtio.c,v 1.23 2017/03/26 10:36:10 martin 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 1.23 2017/03/26 10:36:10 martin 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#include <dev/pci/pcidevs.h>
41#include <dev/pci/pcireg.h>
42#include <dev/pci/pcivar.h>
43
44#define VIRTIO_PRIVATE
45
46#include <dev/pci/virtioreg.h>
47#include <dev/pci/virtiovar.h>
48
49#define MINSEG_INDIRECT		2 /* use indirect if nsegs >= this value */
50
51static int	virtio_match(device_t, cfdata_t, void *);
52static void	virtio_attach(device_t, device_t, void *);
53static int	virtio_rescan(device_t, const char *, const int *);
54static int	virtio_detach(device_t, int);
55static int	virtio_intr(void *arg);
56static int	virtio_msix_queue_intr(void *);
57static int	virtio_msix_config_intr(void *);
58static int	virtio_setup_msix_vectors(struct virtio_softc *);
59static int	virtio_setup_msix_interrupts(struct virtio_softc *,
60		    struct pci_attach_args *);
61static int	virtio_setup_intx_interrupt(struct virtio_softc *,
62		    struct pci_attach_args *);
63static int	virtio_setup_interrupts(struct virtio_softc *);
64static void	virtio_free_interrupts(struct virtio_softc *);
65static void	virtio_soft_intr(void *arg);
66static void	virtio_init_vq(struct virtio_softc *,
67		    struct virtqueue *, const bool);
68
69CFATTACH_DECL3_NEW(virtio, sizeof(struct virtio_softc),
70    virtio_match, virtio_attach, virtio_detach, NULL, virtio_rescan, NULL,
71    DVF_DETACH_SHUTDOWN);
72
73/* we use the legacy virtio spec, so the pci registers are host native
74 * byte order, not pci (i.e. LE) byte order */
75static inline uint16_t
76nbo_bus_space_read_2(bus_space_tag_t space, bus_space_handle_t handle,
77         bus_size_t offset)
78{
79	return le16toh(bus_space_read_2(space, handle, offset));
80}
81
82static inline uint32_t
83nbo_bus_space_read_4(bus_space_tag_t space, bus_space_handle_t handle,
84	bus_size_t offset)
85{
86	return le32toh(bus_space_read_4(space, handle, offset));
87}
88
89static void
90nbo_bus_space_write_2(bus_space_tag_t space, bus_space_handle_t handle,
91	bus_size_t offset, uint16_t value)
92{
93	bus_space_write_2(space, handle, offset, htole16(value));
94}
95
96static void
97nbo_bus_space_write_4(bus_space_tag_t space, bus_space_handle_t handle,
98	bus_size_t offset, uint16_t value)
99{
100	bus_space_write_4(space, handle, offset, htole32(value));
101}
102
103/* some functions access registers at 4 byte offset for little/high halves */
104#if BYTE_ORDER == BIG_ENDIAN
105#define REG_HI_OFF	0
106#define	REG_LO_OFF	4
107#else
108#define REG_HI_OFF	4
109#define	REG_LO_OFF	0
110#endif
111
112static void
113virtio_set_status(struct virtio_softc *sc, int status)
114{
115	int old = 0;
116
117	if (status != 0)
118		old = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
119				       VIRTIO_CONFIG_DEVICE_STATUS);
120	bus_space_write_1(sc->sc_iot, sc->sc_ioh, VIRTIO_CONFIG_DEVICE_STATUS,
121			  status|old);
122}
123
124#define virtio_device_reset(sc)	virtio_set_status((sc), 0)
125
126static int
127virtio_match(device_t parent, cfdata_t match, void *aux)
128{
129	struct pci_attach_args *pa;
130
131	pa = (struct pci_attach_args *)aux;
132	switch (PCI_VENDOR(pa->pa_id)) {
133	case PCI_VENDOR_QUMRANET:
134		if ((PCI_PRODUCT_QUMRANET_VIRTIO_1000 <=
135		     PCI_PRODUCT(pa->pa_id)) &&
136		    (PCI_PRODUCT(pa->pa_id) <=
137		     PCI_PRODUCT_QUMRANET_VIRTIO_103F))
138			return 1;
139		break;
140	}
141
142	return 0;
143}
144
145static const char *virtio_device_name[] = {
146	"Unknown (0)",			/* 0 */
147	"Network",			/* 1 */
148	"Block",			/* 2 */
149	"Console",			/* 3 */
150	"Entropy",			/* 4 */
151	"Memory Balloon",		/* 5 */
152	"I/O Memory",			/* 6 */
153	"Remote Processor Messaging",	/* 7 */
154	"SCSI",				/* 8 */
155	"9P Transport",			/* 9 */
156	"mac80211 wlan",		/* 10 */
157};
158#define NDEVNAMES	__arraycount(virtio_device_name)
159
160#define VIRTIO_MSIX_CONFIG_VECTOR_INDEX	0
161#define VIRTIO_MSIX_QUEUE_VECTOR_INDEX	1
162
163static int
164virtio_setup_msix_vectors(struct virtio_softc *sc)
165{
166	int offset, vector, ret, qid;
167
168	offset = VIRTIO_CONFIG_MSI_CONFIG_VECTOR;
169	vector = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
170
171	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh, offset, vector);
172	ret = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh, offset);
173	aprint_debug_dev(sc->sc_dev, "expected=%d, actual=%d\n",
174	    vector, ret);
175	if (ret != vector)
176		return -1;
177
178	for (qid = 0; qid < sc->sc_nvqs; qid++) {
179		offset = VIRTIO_CONFIG_QUEUE_SELECT;
180		nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh, offset, qid);
181
182		offset = VIRTIO_CONFIG_MSI_QUEUE_VECTOR;
183		vector = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
184
185		nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh, offset, vector);
186		ret = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh, offset);
187		aprint_debug_dev(sc->sc_dev, "expected=%d, actual=%d\n",
188		    vector, ret);
189		if (ret != vector)
190			return -1;
191	}
192
193	return 0;
194}
195
196static int
197virtio_setup_msix_interrupts(struct virtio_softc *sc,
198    struct pci_attach_args *pa)
199{
200	device_t self = sc->sc_dev;
201	pci_chipset_tag_t pc = pa->pa_pc;
202	char intrbuf[PCI_INTRSTR_LEN];
203	char const *intrstr;
204	int idx;
205
206	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
207	if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
208		pci_intr_setattr(pc, &sc->sc_ihp[idx], PCI_INTR_MPSAFE, true);
209
210	sc->sc_ihs[idx] = pci_intr_establish_xname(pc, sc->sc_ihp[idx],
211	    sc->sc_ipl, virtio_msix_config_intr, sc, device_xname(sc->sc_dev));
212	if (sc->sc_ihs[idx] == NULL) {
213		aprint_error_dev(self, "couldn't establish MSI-X for config\n");
214		goto error;
215	}
216
217	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
218	if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
219		pci_intr_setattr(pc, &sc->sc_ihp[idx], PCI_INTR_MPSAFE, true);
220
221	sc->sc_ihs[idx] = pci_intr_establish_xname(pc, sc->sc_ihp[idx],
222	    sc->sc_ipl, virtio_msix_queue_intr, sc, device_xname(sc->sc_dev));
223	if (sc->sc_ihs[idx] == NULL) {
224		aprint_error_dev(self, "couldn't establish MSI-X for queues\n");
225		goto error;
226	}
227
228	if (virtio_setup_msix_vectors(sc) != 0) {
229		aprint_error_dev(self, "couldn't setup MSI-X vectors\n");
230		goto error;
231	}
232
233	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
234	intrstr = pci_intr_string(pc, sc->sc_ihp[idx], intrbuf, sizeof(intrbuf));
235	aprint_normal_dev(self, "config interrupting at %s\n", intrstr);
236	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
237	intrstr = pci_intr_string(pc, sc->sc_ihp[idx], intrbuf, sizeof(intrbuf));
238	aprint_normal_dev(self, "queues interrupting at %s\n", intrstr);
239
240	return 0;
241
242error:
243	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
244	if (sc->sc_ihs[idx] != NULL)
245		pci_intr_disestablish(sc->sc_pc, sc->sc_ihs[idx]);
246	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
247	if (sc->sc_ihs[idx] != NULL)
248		pci_intr_disestablish(sc->sc_pc, sc->sc_ihs[idx]);
249
250	return -1;
251}
252
253static int
254virtio_setup_intx_interrupt(struct virtio_softc *sc,
255    struct pci_attach_args *pa)
256{
257	device_t self = sc->sc_dev;
258	pci_chipset_tag_t pc = pa->pa_pc;
259	char intrbuf[PCI_INTRSTR_LEN];
260	char const *intrstr;
261
262	if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
263		pci_intr_setattr(pc, &sc->sc_ihp[0], PCI_INTR_MPSAFE, true);
264
265	sc->sc_ihs[0] = pci_intr_establish_xname(pc, sc->sc_ihp[0],
266	    sc->sc_ipl, virtio_intr, sc, device_xname(sc->sc_dev));
267	if (sc->sc_ihs[0] == NULL) {
268		aprint_error_dev(self, "couldn't establish INTx\n");
269		return -1;
270	}
271
272	intrstr = pci_intr_string(pc, sc->sc_ihp[0], intrbuf, sizeof(intrbuf));
273	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
274
275	return 0;
276}
277
278static int
279virtio_setup_interrupts(struct virtio_softc *sc)
280{
281	device_t self = sc->sc_dev;
282	pci_chipset_tag_t pc = sc->sc_pa.pa_pc;
283	int error;
284	int nmsix;
285	int counts[PCI_INTR_TYPE_SIZE];
286	pci_intr_type_t max_type;
287
288	nmsix = pci_msix_count(sc->sc_pa.pa_pc, sc->sc_pa.pa_tag);
289	aprint_debug_dev(self, "pci_msix_count=%d\n", nmsix);
290
291	/* We need at least two: one for config and the other for queues */
292	if ((sc->sc_flags & VIRTIO_F_PCI_INTR_MSIX) == 0 || nmsix < 2) {
293		/* Try INTx only */
294		max_type = PCI_INTR_TYPE_INTX;
295		counts[PCI_INTR_TYPE_INTX] = 1;
296	} else {
297		/* Try MSI-X first and INTx second */
298		max_type = PCI_INTR_TYPE_MSIX;
299		counts[PCI_INTR_TYPE_MSIX] = 2;
300		counts[PCI_INTR_TYPE_MSI] = 0;
301		counts[PCI_INTR_TYPE_INTX] = 1;
302	}
303
304 retry:
305	error = pci_intr_alloc(&sc->sc_pa, &sc->sc_ihp, counts, max_type);
306	if (error != 0) {
307		aprint_error_dev(self, "couldn't map interrupt\n");
308		return -1;
309	}
310
311	if (pci_intr_type(pc, sc->sc_ihp[0]) == PCI_INTR_TYPE_MSIX) {
312		sc->sc_ihs = kmem_alloc(sizeof(*sc->sc_ihs) * 2,
313		    KM_SLEEP);
314		if (sc->sc_ihs == NULL) {
315			pci_intr_release(pc, sc->sc_ihp, 2);
316
317			/* Retry INTx */
318			max_type = PCI_INTR_TYPE_INTX;
319			counts[PCI_INTR_TYPE_INTX] = 1;
320			goto retry;
321		}
322
323		error = virtio_setup_msix_interrupts(sc, &sc->sc_pa);
324		if (error != 0) {
325			kmem_free(sc->sc_ihs, sizeof(*sc->sc_ihs) * 2);
326			pci_intr_release(pc, sc->sc_ihp, 2);
327
328			/* Retry INTx */
329			max_type = PCI_INTR_TYPE_INTX;
330			counts[PCI_INTR_TYPE_INTX] = 1;
331			goto retry;
332		}
333
334		sc->sc_ihs_num = 2;
335		sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_MSI;
336	} else if (pci_intr_type(pc, sc->sc_ihp[0]) == PCI_INTR_TYPE_INTX) {
337		sc->sc_ihs = kmem_alloc(sizeof(*sc->sc_ihs) * 1,
338		    KM_SLEEP);
339		if (sc->sc_ihs == NULL) {
340			pci_intr_release(pc, sc->sc_ihp, 1);
341			return -1;
342		}
343
344		error = virtio_setup_intx_interrupt(sc, &sc->sc_pa);
345		if (error != 0) {
346			kmem_free(sc->sc_ihs, sizeof(*sc->sc_ihs) * 1);
347			pci_intr_release(pc, sc->sc_ihp, 1);
348			return -1;
349		}
350
351		sc->sc_ihs_num = 1;
352		sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
353	}
354
355	KASSERT(sc->sc_soft_ih == NULL);
356	if (sc->sc_flags & VIRTIO_F_PCI_INTR_SOFTINT) {
357		u_int flags = SOFTINT_NET;
358		if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
359			flags |= SOFTINT_MPSAFE;
360
361		sc->sc_soft_ih = softint_establish(flags, virtio_soft_intr, sc);
362		if (sc->sc_soft_ih == NULL) {
363			virtio_free_interrupts(sc);
364			aprint_error_dev(sc->sc_dev,
365			    "failed to establish soft interrupt\n");
366			return -1;
367		}
368	}
369
370	return 0;
371}
372
373static void
374virtio_free_interrupts(struct virtio_softc *sc)
375{
376	for (int i = 0; i < sc->sc_ihs_num; i++) {
377		if (sc->sc_ihs[i] == NULL)
378			continue;
379		pci_intr_disestablish(sc->sc_pc, sc->sc_ihs[i]);
380		sc->sc_ihs[i] = NULL;
381	}
382
383	if (sc->sc_ihs_num > 0)
384		pci_intr_release(sc->sc_pc, sc->sc_ihp, sc->sc_ihs_num);
385
386	if (sc->sc_soft_ih) {
387		softint_disestablish(sc->sc_soft_ih);
388		sc->sc_soft_ih = NULL;
389	}
390
391	if (sc->sc_ihs != NULL) {
392		kmem_free(sc->sc_ihs, sizeof(*sc->sc_ihs) * sc->sc_ihs_num);
393		sc->sc_ihs = NULL;
394	}
395	sc->sc_ihs_num = 0;
396}
397
398static void
399virtio_attach(device_t parent, device_t self, void *aux)
400{
401	struct virtio_softc *sc = device_private(self);
402	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
403	pci_chipset_tag_t pc = pa->pa_pc;
404	pcitag_t tag = pa->pa_tag;
405	int revision;
406	pcireg_t id;
407
408	revision = PCI_REVISION(pa->pa_class);
409	if (revision != 0) {
410		aprint_normal(": unknown revision 0x%02x; giving up\n",
411			      revision);
412		return;
413	}
414	aprint_normal("\n");
415	aprint_naive("\n");
416
417	/* subsystem ID shows what I am */
418	id = pci_conf_read(pc, tag, PCI_SUBSYS_ID_REG);
419	aprint_normal_dev(self, "Virtio %s Device (rev. 0x%02x)\n",
420			  (PCI_SUBSYS_ID(id) < NDEVNAMES?
421			   virtio_device_name[PCI_SUBSYS_ID(id)] : "Unknown"),
422			  revision);
423
424	sc->sc_dev = self;
425	sc->sc_pc = pc;
426	sc->sc_tag = tag;
427	sc->sc_iot = pa->pa_iot;
428	if (pci_dma64_available(pa))
429		sc->sc_dmat = pa->pa_dmat64;
430	else
431		sc->sc_dmat = pa->pa_dmat;
432	sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
433
434	if (pci_mapreg_map(pa, PCI_MAPREG_START, PCI_MAPREG_TYPE_IO, 0,
435			   &sc->sc_iot, &sc->sc_ioh, NULL, &sc->sc_iosize)) {
436		aprint_error_dev(self, "can't map i/o space\n");
437		return;
438	}
439
440	virtio_device_reset(sc);
441	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
442	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
443
444	sc->sc_childdevid = PCI_SUBSYS_ID(id);
445	sc->sc_child = NULL;
446	sc->sc_pa = *pa;
447	virtio_rescan(self, "virtio", 0);
448	return;
449}
450
451/* ARGSUSED */
452static int
453virtio_rescan(device_t self, const char *attr, const int *scan_flags)
454{
455	struct virtio_softc *sc;
456	struct virtio_attach_args va;
457
458	sc = device_private(self);
459	if (sc->sc_child)	/* Child already attached? */
460		return 0;
461
462	memset(&va, 0, sizeof(va));
463	va.sc_childdevid = sc->sc_childdevid;
464
465	config_found_ia(self, attr, &va, NULL);
466
467	if (sc->sc_child == NULL) {
468		aprint_error_dev(self,
469				 "no matching child driver; not configured\n");
470		return 0;
471	}
472
473	if (sc->sc_child == VIRTIO_CHILD_FAILED) {
474		aprint_error_dev(self,
475				 "virtio configuration failed\n");
476		return 0;
477	}
478
479	/*
480	 * Make sure child drivers initialize interrupts via call
481	 * to virtio_child_attach_finish().
482	 */
483	KASSERT(sc->sc_ihs_num != 0);
484
485	return 0;
486}
487
488static int
489virtio_detach(device_t self, int flags)
490{
491	struct virtio_softc *sc = device_private(self);
492	int r;
493
494	if (sc->sc_child != NULL) {
495		r = config_detach(sc->sc_child, flags);
496		if (r)
497			return r;
498	}
499
500	/* Check that child detached properly */
501	KASSERT(sc->sc_child == NULL);
502	KASSERT(sc->sc_vqs == NULL);
503	KASSERT(sc->sc_ihs_num == 0);
504
505	if (sc->sc_iosize)
506		bus_space_unmap(sc->sc_iot, sc->sc_ioh, sc->sc_iosize);
507	sc->sc_iosize = 0;
508
509	return 0;
510}
511
512/*
513 * Reset the device.
514 */
515/*
516 * To reset the device to a known state, do following:
517 *	virtio_reset(sc);	     // this will stop the device activity
518 *	<dequeue finished requests>; // virtio_dequeue() still can be called
519 *	<revoke pending requests in the vqs if any>;
520 *	virtio_reinit_begin(sc);     // dequeue prohibitted
521 *	newfeatures = virtio_negotiate_features(sc, requestedfeatures);
522 *	<some other initialization>;
523 *	virtio_reinit_end(sc);	     // device activated; enqueue allowed
524 * Once attached, feature negotiation can only be allowed after virtio_reset.
525 */
526void
527virtio_reset(struct virtio_softc *sc)
528{
529	virtio_device_reset(sc);
530}
531
532void
533virtio_reinit_start(struct virtio_softc *sc)
534{
535	int i;
536
537	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
538	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
539	for (i = 0; i < sc->sc_nvqs; i++) {
540		int n;
541		struct virtqueue *vq = &sc->sc_vqs[i];
542		nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
543				  VIRTIO_CONFIG_QUEUE_SELECT,
544				  vq->vq_index);
545		n = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh,
546				     VIRTIO_CONFIG_QUEUE_SIZE);
547		if (n == 0)	/* vq disappeared */
548			continue;
549		if (n != vq->vq_num) {
550			panic("%s: virtqueue size changed, vq index %d\n",
551			      device_xname(sc->sc_dev),
552			      vq->vq_index);
553		}
554		virtio_init_vq(sc, vq, true);
555		nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
556				  VIRTIO_CONFIG_QUEUE_ADDRESS,
557				  (vq->vq_dmamap->dm_segs[0].ds_addr
558				   / VIRTIO_PAGE_SIZE));
559	}
560
561	/* MSI-X should have more than one handles where INTx has just one */
562	if (sc->sc_ihs_num > 1) {
563		if (virtio_setup_msix_vectors(sc) != 0) {
564			aprint_error_dev(sc->sc_dev,
565			    "couldn't setup MSI-X vectors\n");
566			return;
567		}
568	}
569}
570
571void
572virtio_reinit_end(struct virtio_softc *sc)
573{
574	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
575}
576
577/*
578 * Feature negotiation.
579 */
580uint32_t
581virtio_negotiate_features(struct virtio_softc *sc, uint32_t guest_features)
582{
583	uint32_t r;
584
585	if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) &&
586	    !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */
587		guest_features |= VIRTIO_F_RING_INDIRECT_DESC;
588	r = nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
589			     VIRTIO_CONFIG_DEVICE_FEATURES);
590	r &= guest_features;
591	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
592			  VIRTIO_CONFIG_GUEST_FEATURES, r);
593	sc->sc_features = r;
594	if (r & VIRTIO_F_RING_INDIRECT_DESC)
595		sc->sc_indirect = true;
596	else
597		sc->sc_indirect = false;
598
599	return r;
600}
601
602/*
603 * Device configuration registers.
604 */
605uint8_t
606virtio_read_device_config_1(struct virtio_softc *sc, int index)
607{
608	return bus_space_read_1(sc->sc_iot, sc->sc_ioh,
609				sc->sc_config_offset + index);
610}
611
612uint16_t
613virtio_read_device_config_2(struct virtio_softc *sc, int index)
614{
615	return nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh,
616				sc->sc_config_offset + index);
617}
618
619uint32_t
620virtio_read_device_config_4(struct virtio_softc *sc, int index)
621{
622	return nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
623				sc->sc_config_offset + index);
624}
625
626uint64_t
627virtio_read_device_config_8(struct virtio_softc *sc, int index)
628{
629	uint64_t r;
630
631	r = nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
632			     sc->sc_config_offset + index + REG_HI_OFF);
633	r <<= 32;
634	r |= nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
635			      sc->sc_config_offset + index + REG_LO_OFF);
636
637	return r;
638}
639
640void
641virtio_write_device_config_1(struct virtio_softc *sc,
642			     int index, uint8_t value)
643{
644	bus_space_write_1(sc->sc_iot, sc->sc_ioh,
645			  sc->sc_config_offset + index, value);
646}
647
648void
649virtio_write_device_config_2(struct virtio_softc *sc,
650			     int index, uint16_t value)
651{
652	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
653			  sc->sc_config_offset + index, value);
654}
655
656void
657virtio_write_device_config_4(struct virtio_softc *sc,
658			     int index, uint32_t value)
659{
660	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
661			  sc->sc_config_offset + index, value);
662}
663
664void
665virtio_write_device_config_8(struct virtio_softc *sc,
666			     int index, uint64_t value)
667{
668	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
669			  sc->sc_config_offset + index + REG_LO_OFF,
670			  value & 0xffffffff);
671	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
672			  sc->sc_config_offset + index + REG_HI_OFF,
673			  value >> 32);
674}
675
676/*
677 * Interrupt handler.
678 */
679static int
680virtio_intr(void *arg)
681{
682	struct virtio_softc *sc = arg;
683	int isr, r = 0;
684
685	/* check and ack the interrupt */
686	isr = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
687			       VIRTIO_CONFIG_ISR_STATUS);
688	if (isr == 0)
689		return 0;
690	if ((isr & VIRTIO_CONFIG_ISR_CONFIG_CHANGE) &&
691	    (sc->sc_config_change != NULL))
692		r = (sc->sc_config_change)(sc);
693	if (sc->sc_intrhand != NULL) {
694		if (sc->sc_soft_ih != NULL)
695			softint_schedule(sc->sc_soft_ih);
696		else
697			r |= (sc->sc_intrhand)(sc);
698	}
699
700	return r;
701}
702
703static int
704virtio_msix_queue_intr(void *arg)
705{
706	struct virtio_softc *sc = arg;
707	int r = 0;
708
709	if (sc->sc_intrhand != NULL) {
710		if (sc->sc_soft_ih != NULL)
711			softint_schedule(sc->sc_soft_ih);
712		else
713			r |= (sc->sc_intrhand)(sc);
714	}
715
716	return r;
717}
718
719static int
720virtio_msix_config_intr(void *arg)
721{
722	struct virtio_softc *sc = arg;
723
724	/* TODO: handle events */
725	aprint_debug_dev(sc->sc_dev, "%s\n", __func__);
726	return 1;
727}
728
729static void
730virtio_soft_intr(void *arg)
731{
732	struct virtio_softc *sc = arg;
733
734	KASSERT(sc->sc_intrhand != NULL);
735
736	(sc->sc_intrhand)(sc);
737}
738
739/*
740 * dmamap sync operations for a virtqueue.
741 */
742static inline void
743vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops)
744{
745	/* availoffset == sizeof(vring_desc)*vq_num */
746	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset,
747			ops);
748}
749
750static inline void
751vq_sync_aring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
752{
753	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
754			vq->vq_availoffset,
755			offsetof(struct vring_avail, ring)
756			 + vq->vq_num * sizeof(uint16_t),
757			ops);
758}
759
760static inline void
761vq_sync_uring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
762{
763	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
764			vq->vq_usedoffset,
765			offsetof(struct vring_used, ring)
766			 + vq->vq_num * sizeof(struct vring_used_elem),
767			ops);
768}
769
770static inline void
771vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot,
772		     int ops)
773{
774	int offset = vq->vq_indirectoffset
775		      + sizeof(struct vring_desc) * vq->vq_maxnsegs * slot;
776
777	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
778			offset, sizeof(struct vring_desc) * vq->vq_maxnsegs,
779			ops);
780}
781
782/*
783 * Can be used as sc_intrhand.
784 */
785/*
786 * Scan vq, bus_dmamap_sync for the vqs (not for the payload),
787 * and calls (*vq_done)() if some entries are consumed.
788 */
789int
790virtio_vq_intr(struct virtio_softc *sc)
791{
792	struct virtqueue *vq;
793	int i, r = 0;
794
795	for (i = 0; i < sc->sc_nvqs; i++) {
796		vq = &sc->sc_vqs[i];
797		if (vq->vq_queued) {
798			vq->vq_queued = 0;
799			vq_sync_aring(sc, vq, BUS_DMASYNC_POSTWRITE);
800		}
801		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
802		membar_consumer();
803		if (vq->vq_used_idx != vq->vq_used->idx) {
804			if (vq->vq_done)
805				r |= (vq->vq_done)(vq);
806		}
807	}
808
809	return r;
810}
811
812/*
813 * Start/stop vq interrupt.  No guarantee.
814 */
815void
816virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
817{
818	vq->vq_avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
819	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
820	vq->vq_queued++;
821}
822
823void
824virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
825{
826	vq->vq_avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
827	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
828	vq->vq_queued++;
829}
830
831/*
832 * Initialize vq structure.
833 */
834static void
835virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq,
836    const bool reinit)
837{
838	int i, j;
839	int vq_size = vq->vq_num;
840
841	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
842
843	/* build the indirect descriptor chain */
844	if (vq->vq_indirect != NULL) {
845		struct vring_desc *vd;
846
847printf("vq_size: %d, vq_maxnsegs: %d\n", vq_size, vq->vq_maxnsegs);
848
849		for (i = 0; i < vq_size; i++) {
850			vd = vq->vq_indirect;
851			vd += vq->vq_maxnsegs * i;
852			for (j = 0; j < vq->vq_maxnsegs-1; j++) {
853				vd[j].next = j + 1;
854			}
855		}
856	}
857
858	/* free slot management */
859	SIMPLEQ_INIT(&vq->vq_freelist);
860	for (i = 0; i < vq_size; i++) {
861		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
862				    &vq->vq_entries[i], qe_list);
863		vq->vq_entries[i].qe_index = i;
864	}
865	if (!reinit)
866		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
867
868	/* enqueue/dequeue status */
869	vq->vq_avail_idx = 0;
870	vq->vq_used_idx = 0;
871	vq->vq_queued = 0;
872	if (!reinit) {
873		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
874		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
875	}
876	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
877	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
878	vq->vq_queued++;
879}
880
881/*
882 * Allocate/free a vq.
883 */
884int
885virtio_alloc_vq(struct virtio_softc *sc, struct virtqueue *vq, int index,
886    int maxsegsize, int maxnsegs, const char *name)
887{
888	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
889	int rsegs, r;
890#define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
891				 ~(VIRTIO_PAGE_SIZE-1))
892
893	/* Make sure callers allocate vqs in order */
894	KASSERT(sc->sc_nvqs == index);
895
896	memset(vq, 0, sizeof(*vq));
897
898	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
899			  VIRTIO_CONFIG_QUEUE_SELECT, index);
900	vq_size = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh,
901				   VIRTIO_CONFIG_QUEUE_SIZE);
902	if (vq_size == 0) {
903		aprint_error_dev(sc->sc_dev,
904				 "virtqueue not exist, index %d for %s\n",
905				 index, name);
906		goto err;
907	}
908	/* allocsize1: descriptor table + avail ring + pad */
909	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
910				     + sizeof(uint16_t)*(2+vq_size));
911	/* allocsize2: used ring + pad */
912	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
913				     + sizeof(struct vring_used_elem)*vq_size);
914	/* allocsize3: indirect table */
915	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
916		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
917	else
918		allocsize3 = 0;
919	allocsize = allocsize1 + allocsize2 + allocsize3;
920printf("virtio: allocsize1: %d, allocsize2: %d, allocsize3: %d\n",
921   allocsize1, allocsize2, allocsize3);
922
923	/* alloc and map the memory */
924	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
925			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
926	if (r != 0) {
927		aprint_error_dev(sc->sc_dev,
928				 "virtqueue %d for %s allocation failed, "
929				 "error code %d\n", index, name, r);
930		goto err;
931	}
932	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
933			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
934	if (r != 0) {
935		aprint_error_dev(sc->sc_dev,
936				 "virtqueue %d for %s map failed, "
937				 "error code %d\n", index, name, r);
938		goto err;
939	}
940	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
941			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
942	if (r != 0) {
943		aprint_error_dev(sc->sc_dev,
944				 "virtqueue %d for %s dmamap creation failed, "
945				 "error code %d\n", index, name, r);
946		goto err;
947	}
948	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
949			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
950	if (r != 0) {
951		aprint_error_dev(sc->sc_dev,
952				 "virtqueue %d for %s dmamap load failed, "
953				 "error code %d\n", index, name, r);
954		goto err;
955	}
956
957	/* set the vq address */
958	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
959			  VIRTIO_CONFIG_QUEUE_ADDRESS,
960			  (vq->vq_dmamap->dm_segs[0].ds_addr
961			   / VIRTIO_PAGE_SIZE));
962
963	/* remember addresses and offsets for later use */
964	vq->vq_owner = sc;
965	vq->vq_num = vq_size;
966	vq->vq_index = index;
967	vq->vq_desc = vq->vq_vaddr;
968	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
969	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
970	vq->vq_usedoffset = allocsize1;
971	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
972	if (allocsize3 > 0) {
973		vq->vq_indirectoffset = allocsize1 + allocsize2;
974		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
975					  + vq->vq_indirectoffset);
976printf("virtio: vq_indirectoffset: %d, vq_indirect: %p\n",
977    vq->vq_indirectoffset, vq->vq_indirect);
978	}
979	vq->vq_bytesize = allocsize;
980	vq->vq_maxsegsize = maxsegsize;
981	vq->vq_maxnsegs = maxnsegs;
982
983	/* free slot management */
984	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
985				     KM_NOSLEEP);
986	if (vq->vq_entries == NULL) {
987		r = ENOMEM;
988		goto err;
989	}
990
991	virtio_init_vq(sc, vq, false);
992
993	aprint_verbose_dev(sc->sc_dev,
994			   "allocated %u byte for virtqueue %d for %s, "
995			   "size %d\n", allocsize, index, name, vq_size);
996	if (allocsize3 > 0)
997		aprint_verbose_dev(sc->sc_dev,
998				   "using %d byte (%d entries) "
999				   "indirect descriptors\n",
1000				   allocsize3, maxnsegs * vq_size);
1001
1002	sc->sc_nvqs++;
1003
1004	return 0;
1005
1006err:
1007	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
1008			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
1009	if (vq->vq_dmamap)
1010		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
1011	if (vq->vq_vaddr)
1012		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
1013	if (vq->vq_segs[0].ds_addr)
1014		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
1015	memset(vq, 0, sizeof(*vq));
1016
1017	return -1;
1018}
1019
1020int
1021virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
1022{
1023	struct vq_entry *qe;
1024	int i = 0;
1025
1026	/* device must be already deactivated */
1027	/* confirm the vq is empty */
1028	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
1029		i++;
1030	}
1031	if (i != vq->vq_num) {
1032		printf("%s: freeing non-empty vq, index %d\n",
1033		       device_xname(sc->sc_dev), vq->vq_index);
1034		return EBUSY;
1035	}
1036
1037	/* tell device that there's no virtqueue any longer */
1038	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
1039			  VIRTIO_CONFIG_QUEUE_SELECT, vq->vq_index);
1040	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
1041			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
1042
1043	kmem_free(vq->vq_entries, sizeof(*vq->vq_entries) * vq->vq_num);
1044	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
1045	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
1046	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
1047	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
1048	mutex_destroy(&vq->vq_freelist_lock);
1049	mutex_destroy(&vq->vq_uring_lock);
1050	mutex_destroy(&vq->vq_aring_lock);
1051	memset(vq, 0, sizeof(*vq));
1052
1053	sc->sc_nvqs--;
1054
1055	return 0;
1056}
1057
1058/*
1059 * Free descriptor management.
1060 */
1061static struct vq_entry *
1062vq_alloc_entry(struct virtqueue *vq)
1063{
1064	struct vq_entry *qe;
1065
1066	mutex_enter(&vq->vq_freelist_lock);
1067	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
1068		mutex_exit(&vq->vq_freelist_lock);
1069		return NULL;
1070	}
1071	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
1072	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
1073	mutex_exit(&vq->vq_freelist_lock);
1074
1075	return qe;
1076}
1077
1078static void
1079vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
1080{
1081	mutex_enter(&vq->vq_freelist_lock);
1082	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
1083	mutex_exit(&vq->vq_freelist_lock);
1084
1085	return;
1086}
1087
1088/*
1089 * Enqueue several dmamaps as a single request.
1090 */
1091/*
1092 * Typical usage:
1093 *  <queue size> number of followings are stored in arrays
1094 *  - command blocks (in dmamem) should be pre-allocated and mapped
1095 *  - dmamaps for command blocks should be pre-allocated and loaded
1096 *  - dmamaps for payload should be pre-allocated
1097 *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
1098 *	if (r)		// currently 0 or EAGAIN
1099 *	  return r;
1100 *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
1101 *	if (r) {
1102 *	  virtio_enqueue_abort(sc, vq, slot);
1103 *	  return r;
1104 *	}
1105 *	r = virtio_enqueue_reserve(sc, vq, slot,
1106 *				   dmamap_payload[slot]->dm_nsegs+1);
1107 *							// ^ +1 for command
1108 *	if (r) {	// currently 0 or EAGAIN
1109 *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
1110 *	  return r;					// do not call abort()
1111 *	}
1112 *	<setup and prepare commands>
1113 *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
1114 *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
1115 *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
1116 *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
1117 *	virtio_enqueue_commit(sc, vq, slot, true);
1118 */
1119
1120/*
1121 * enqueue_prep: allocate a slot number
1122 */
1123int
1124virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
1125{
1126	struct vq_entry *qe1;
1127
1128	KASSERT(slotp != NULL);
1129
1130	qe1 = vq_alloc_entry(vq);
1131	if (qe1 == NULL)
1132		return EAGAIN;
1133	/* next slot is not allocated yet */
1134	qe1->qe_next = -1;
1135	*slotp = qe1->qe_index;
1136
1137	return 0;
1138}
1139
1140/*
1141 * enqueue_reserve: allocate remaining slots and build the descriptor chain.
1142 */
1143int
1144virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
1145		       int slot, int nsegs)
1146{
1147	int indirect;
1148	struct vq_entry *qe1 = &vq->vq_entries[slot];
1149
1150	KASSERT(qe1->qe_next == -1);
1151	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
1152
1153	if ((vq->vq_indirect != NULL) &&
1154	    (nsegs >= MINSEG_INDIRECT) &&
1155	    (nsegs <= vq->vq_maxnsegs))
1156		indirect = 1;
1157	else
1158		indirect = 0;
1159	qe1->qe_indirect = indirect;
1160
1161	if (indirect) {
1162		struct vring_desc *vd;
1163		int i;
1164
1165		vd = &vq->vq_desc[qe1->qe_index];
1166		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
1167			+ vq->vq_indirectoffset;
1168		vd->addr += sizeof(struct vring_desc)
1169			* vq->vq_maxnsegs * qe1->qe_index;
1170		vd->len = sizeof(struct vring_desc) * nsegs;
1171		vd->flags = VRING_DESC_F_INDIRECT;
1172
1173		vd = vq->vq_indirect;
1174		vd += vq->vq_maxnsegs * qe1->qe_index;
1175		qe1->qe_desc_base = vd;
1176
1177		for (i = 0; i < nsegs-1; i++) {
1178			vd[i].flags = VRING_DESC_F_NEXT;
1179		}
1180		vd[i].flags = 0;
1181		qe1->qe_next = 0;
1182
1183		return 0;
1184	} else {
1185		struct vring_desc *vd;
1186		struct vq_entry *qe;
1187		int i, s;
1188
1189		vd = &vq->vq_desc[0];
1190		qe1->qe_desc_base = vd;
1191		qe1->qe_next = qe1->qe_index;
1192		s = slot;
1193		for (i = 0; i < nsegs - 1; i++) {
1194			qe = vq_alloc_entry(vq);
1195			if (qe == NULL) {
1196				vd[s].flags = 0;
1197				virtio_enqueue_abort(sc, vq, slot);
1198				return EAGAIN;
1199			}
1200			vd[s].flags = VRING_DESC_F_NEXT;
1201			vd[s].next = qe->qe_index;
1202			s = qe->qe_index;
1203		}
1204		vd[s].flags = 0;
1205
1206		return 0;
1207	}
1208}
1209
1210/*
1211 * enqueue: enqueue a single dmamap.
1212 */
1213int
1214virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1215	       bus_dmamap_t dmamap, bool write)
1216{
1217	struct vq_entry *qe1 = &vq->vq_entries[slot];
1218	struct vring_desc *vd = qe1->qe_desc_base;
1219	int i;
1220	int s = qe1->qe_next;
1221
1222	KASSERT(s >= 0);
1223	KASSERT(dmamap->dm_nsegs > 0);
1224
1225	for (i = 0; i < dmamap->dm_nsegs; i++) {
1226		vd[s].addr = dmamap->dm_segs[i].ds_addr;
1227		vd[s].len = dmamap->dm_segs[i].ds_len;
1228		if (!write)
1229			vd[s].flags |= VRING_DESC_F_WRITE;
1230		s = vd[s].next;
1231	}
1232	qe1->qe_next = s;
1233
1234	return 0;
1235}
1236
1237int
1238virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1239		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
1240		 bool write)
1241{
1242	struct vq_entry *qe1 = &vq->vq_entries[slot];
1243	struct vring_desc *vd = qe1->qe_desc_base;
1244	int s = qe1->qe_next;
1245
1246	KASSERT(s >= 0);
1247	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
1248	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
1249		(dmamap->dm_segs[0].ds_len >= start + len));
1250
1251	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
1252	vd[s].len = len;
1253	if (!write)
1254		vd[s].flags |= VRING_DESC_F_WRITE;
1255	qe1->qe_next = vd[s].next;
1256
1257	return 0;
1258}
1259
1260/*
1261 * enqueue_commit: add it to the aring.
1262 */
1263int
1264virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1265		      bool notifynow)
1266{
1267	struct vq_entry *qe1;
1268
1269	if (slot < 0) {
1270		mutex_enter(&vq->vq_aring_lock);
1271		goto notify;
1272	}
1273	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
1274	qe1 = &vq->vq_entries[slot];
1275	if (qe1->qe_indirect)
1276		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
1277	mutex_enter(&vq->vq_aring_lock);
1278	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
1279
1280notify:
1281	if (notifynow) {
1282		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
1283		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
1284		membar_producer();
1285		vq->vq_avail->idx = vq->vq_avail_idx;
1286		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
1287		membar_producer();
1288		vq->vq_queued++;
1289		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
1290		membar_consumer();
1291		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
1292			nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
1293					  VIRTIO_CONFIG_QUEUE_NOTIFY,
1294					  vq->vq_index);
1295	}
1296	mutex_exit(&vq->vq_aring_lock);
1297
1298	return 0;
1299}
1300
1301/*
1302 * enqueue_abort: rollback.
1303 */
1304int
1305virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1306{
1307	struct vq_entry *qe = &vq->vq_entries[slot];
1308	struct vring_desc *vd;
1309	int s;
1310
1311	if (qe->qe_next < 0) {
1312		vq_free_entry(vq, qe);
1313		return 0;
1314	}
1315
1316	s = slot;
1317	vd = &vq->vq_desc[0];
1318	while (vd[s].flags & VRING_DESC_F_NEXT) {
1319		s = vd[s].next;
1320		vq_free_entry(vq, qe);
1321		qe = &vq->vq_entries[s];
1322	}
1323	vq_free_entry(vq, qe);
1324	return 0;
1325}
1326
1327/*
1328 * Dequeue a request.
1329 */
1330/*
1331 * dequeue: dequeue a request from uring; dmamap_sync for uring is
1332 *	    already done in the interrupt handler.
1333 */
1334int
1335virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
1336	       int *slotp, int *lenp)
1337{
1338	uint16_t slot, usedidx;
1339	struct vq_entry *qe;
1340
1341	if (vq->vq_used_idx == vq->vq_used->idx)
1342		return ENOENT;
1343	mutex_enter(&vq->vq_uring_lock);
1344	usedidx = vq->vq_used_idx++;
1345	mutex_exit(&vq->vq_uring_lock);
1346	usedidx %= vq->vq_num;
1347	slot = vq->vq_used->ring[usedidx].id;
1348	qe = &vq->vq_entries[slot];
1349
1350	if (qe->qe_indirect)
1351		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
1352
1353	if (slotp)
1354		*slotp = slot;
1355	if (lenp)
1356		*lenp = vq->vq_used->ring[usedidx].len;
1357
1358	return 0;
1359}
1360
1361/*
1362 * dequeue_commit: complete dequeue; the slot is recycled for future use.
1363 *                 if you forget to call this the slot will be leaked.
1364 */
1365int
1366virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1367{
1368	struct vq_entry *qe = &vq->vq_entries[slot];
1369	struct vring_desc *vd = &vq->vq_desc[0];
1370	int s = slot;
1371
1372	while (vd[s].flags & VRING_DESC_F_NEXT) {
1373		s = vd[s].next;
1374		vq_free_entry(vq, qe);
1375		qe = &vq->vq_entries[s];
1376	}
1377	vq_free_entry(vq, qe);
1378
1379	return 0;
1380}
1381
1382/*
1383 * Attach a child, fill all the members.
1384 */
1385void
1386virtio_child_attach_start(struct virtio_softc *sc, device_t child, int ipl,
1387		    struct virtqueue *vqs,
1388		    virtio_callback config_change,
1389		    virtio_callback intr_hand,
1390		    int req_flags, int req_features, const char *feat_bits)
1391{
1392	char buf[256];
1393	int features;
1394
1395	sc->sc_child = child;
1396	sc->sc_ipl = ipl;
1397	sc->sc_vqs = vqs;
1398	sc->sc_config_change = config_change;
1399	sc->sc_intrhand = intr_hand;
1400	sc->sc_flags = req_flags;
1401
1402	features = virtio_negotiate_features(sc, req_features);
1403	snprintb(buf, sizeof(buf), feat_bits, features);
1404	aprint_normal(": Features: %s\n", buf);
1405	aprint_naive("\n");
1406}
1407
1408int
1409virtio_child_attach_finish(struct virtio_softc *sc)
1410{
1411	int r;
1412
1413	r = virtio_setup_interrupts(sc);
1414	if (r != 0) {
1415		aprint_error_dev(sc->sc_dev, "failed to setup interrupts\n");
1416		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
1417		return 1;
1418	}
1419
1420	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
1421
1422	return 0;
1423}
1424
1425void
1426virtio_child_detach(struct virtio_softc *sc)
1427{
1428	sc->sc_child = NULL;
1429	sc->sc_vqs = NULL;
1430
1431	virtio_device_reset(sc);
1432
1433	virtio_free_interrupts(sc);
1434}
1435
1436void
1437virtio_child_attach_failed(struct virtio_softc *sc)
1438{
1439	virtio_child_detach(sc);
1440
1441	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
1442
1443	sc->sc_child = VIRTIO_CHILD_FAILED;
1444}
1445
1446bus_dma_tag_t
1447virtio_dmat(struct virtio_softc *sc)
1448{
1449	return sc->sc_dmat;
1450}
1451
1452device_t
1453virtio_child(struct virtio_softc *sc)
1454{
1455	return sc->sc_child;
1456}
1457
1458int
1459virtio_intrhand(struct virtio_softc *sc)
1460{
1461	return (sc->sc_intrhand)(sc);
1462}
1463
1464uint32_t
1465virtio_features(struct virtio_softc *sc)
1466{
1467	return sc->sc_features;
1468}
1469
1470struct pci_attach_args *
1471virtio_pci_attach_args(struct virtio_softc *sc)
1472{
1473	return &sc->sc_pa;
1474}
1475
1476
1477MODULE(MODULE_CLASS_DRIVER, virtio, "pci");
1478
1479#ifdef _MODULE
1480#include "ioconf.c"
1481#endif
1482
1483static int
1484virtio_modcmd(modcmd_t cmd, void *opaque)
1485{
1486	int error = 0;
1487
1488#ifdef _MODULE
1489	switch (cmd) {
1490	case MODULE_CMD_INIT:
1491		error = config_init_component(cfdriver_ioconf_virtio,
1492		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
1493		break;
1494	case MODULE_CMD_FINI:
1495		error = config_fini_component(cfdriver_ioconf_virtio,
1496		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
1497		break;
1498	default:
1499		error = ENOTTY;
1500		break;
1501	}
1502#endif
1503
1504	return error;
1505}
1506