virtio.c revision 1.22
1/*	$NetBSD: virtio.c,v 1.22 2017/03/25 18:02:06 jdolecek 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 1.22 2017/03/25 18:02:06 jdolecek 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
847		for (i = 0; i < vq_size; i++) {
848			vd = vq->vq_indirect;
849			vd += vq->vq_maxnsegs * i;
850			for (j = 0; j < vq->vq_maxnsegs-1; j++)
851				vd[j].next = j + 1;
852		}
853	}
854
855	/* free slot management */
856	SIMPLEQ_INIT(&vq->vq_freelist);
857	for (i = 0; i < vq_size; i++) {
858		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
859				    &vq->vq_entries[i], qe_list);
860		vq->vq_entries[i].qe_index = i;
861	}
862	if (!reinit)
863		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
864
865	/* enqueue/dequeue status */
866	vq->vq_avail_idx = 0;
867	vq->vq_used_idx = 0;
868	vq->vq_queued = 0;
869	if (!reinit) {
870		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
871		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
872	}
873	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
874	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
875	vq->vq_queued++;
876}
877
878/*
879 * Allocate/free a vq.
880 */
881int
882virtio_alloc_vq(struct virtio_softc *sc, struct virtqueue *vq, int index,
883    int maxsegsize, int maxnsegs, const char *name)
884{
885	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
886	int rsegs, r;
887#define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
888				 ~(VIRTIO_PAGE_SIZE-1))
889
890	/* Make sure callers allocate vqs in order */
891	KASSERT(sc->sc_nvqs == index);
892
893	memset(vq, 0, sizeof(*vq));
894
895	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
896			  VIRTIO_CONFIG_QUEUE_SELECT, index);
897	vq_size = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh,
898				   VIRTIO_CONFIG_QUEUE_SIZE);
899	if (vq_size == 0) {
900		aprint_error_dev(sc->sc_dev,
901				 "virtqueue not exist, index %d for %s\n",
902				 index, name);
903		goto err;
904	}
905	/* allocsize1: descriptor table + avail ring + pad */
906	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
907				     + sizeof(uint16_t)*(2+vq_size));
908	/* allocsize2: used ring + pad */
909	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
910				     + sizeof(struct vring_used_elem)*vq_size);
911	/* allocsize3: indirect table */
912	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
913		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
914	else
915		allocsize3 = 0;
916	allocsize = allocsize1 + allocsize2 + allocsize3;
917
918	/* alloc and map the memory */
919	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
920			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
921	if (r != 0) {
922		aprint_error_dev(sc->sc_dev,
923				 "virtqueue %d for %s allocation failed, "
924				 "error code %d\n", index, name, r);
925		goto err;
926	}
927	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
928			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
929	if (r != 0) {
930		aprint_error_dev(sc->sc_dev,
931				 "virtqueue %d for %s map failed, "
932				 "error code %d\n", index, name, r);
933		goto err;
934	}
935	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
936			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
937	if (r != 0) {
938		aprint_error_dev(sc->sc_dev,
939				 "virtqueue %d for %s dmamap creation failed, "
940				 "error code %d\n", index, name, r);
941		goto err;
942	}
943	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
944			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
945	if (r != 0) {
946		aprint_error_dev(sc->sc_dev,
947				 "virtqueue %d for %s dmamap load failed, "
948				 "error code %d\n", index, name, r);
949		goto err;
950	}
951
952	/* set the vq address */
953	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
954			  VIRTIO_CONFIG_QUEUE_ADDRESS,
955			  (vq->vq_dmamap->dm_segs[0].ds_addr
956			   / VIRTIO_PAGE_SIZE));
957
958	/* remember addresses and offsets for later use */
959	vq->vq_owner = sc;
960	vq->vq_num = vq_size;
961	vq->vq_index = index;
962	vq->vq_desc = vq->vq_vaddr;
963	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
964	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
965	vq->vq_usedoffset = allocsize1;
966	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
967	if (allocsize3 > 0) {
968		vq->vq_indirectoffset = allocsize1 + allocsize2;
969		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
970					  + vq->vq_indirectoffset);
971	}
972	vq->vq_bytesize = allocsize;
973	vq->vq_maxsegsize = maxsegsize;
974	vq->vq_maxnsegs = maxnsegs;
975
976	/* free slot management */
977	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
978				     KM_NOSLEEP);
979	if (vq->vq_entries == NULL) {
980		r = ENOMEM;
981		goto err;
982	}
983
984	virtio_init_vq(sc, vq, false);
985
986	aprint_verbose_dev(sc->sc_dev,
987			   "allocated %u byte for virtqueue %d for %s, "
988			   "size %d\n", allocsize, index, name, vq_size);
989	if (allocsize3 > 0)
990		aprint_verbose_dev(sc->sc_dev,
991				   "using %d byte (%d entries) "
992				   "indirect descriptors\n",
993				   allocsize3, maxnsegs * vq_size);
994
995	sc->sc_nvqs++;
996
997	return 0;
998
999err:
1000	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
1001			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
1002	if (vq->vq_dmamap)
1003		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
1004	if (vq->vq_vaddr)
1005		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
1006	if (vq->vq_segs[0].ds_addr)
1007		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
1008	memset(vq, 0, sizeof(*vq));
1009
1010	return -1;
1011}
1012
1013int
1014virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
1015{
1016	struct vq_entry *qe;
1017	int i = 0;
1018
1019	/* device must be already deactivated */
1020	/* confirm the vq is empty */
1021	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
1022		i++;
1023	}
1024	if (i != vq->vq_num) {
1025		printf("%s: freeing non-empty vq, index %d\n",
1026		       device_xname(sc->sc_dev), vq->vq_index);
1027		return EBUSY;
1028	}
1029
1030	/* tell device that there's no virtqueue any longer */
1031	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
1032			  VIRTIO_CONFIG_QUEUE_SELECT, vq->vq_index);
1033	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
1034			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
1035
1036	kmem_free(vq->vq_entries, sizeof(*vq->vq_entries) * vq->vq_num);
1037	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
1038	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
1039	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
1040	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
1041	mutex_destroy(&vq->vq_freelist_lock);
1042	mutex_destroy(&vq->vq_uring_lock);
1043	mutex_destroy(&vq->vq_aring_lock);
1044	memset(vq, 0, sizeof(*vq));
1045
1046	sc->sc_nvqs--;
1047
1048	return 0;
1049}
1050
1051/*
1052 * Free descriptor management.
1053 */
1054static struct vq_entry *
1055vq_alloc_entry(struct virtqueue *vq)
1056{
1057	struct vq_entry *qe;
1058
1059	mutex_enter(&vq->vq_freelist_lock);
1060	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
1061		mutex_exit(&vq->vq_freelist_lock);
1062		return NULL;
1063	}
1064	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
1065	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
1066	mutex_exit(&vq->vq_freelist_lock);
1067
1068	return qe;
1069}
1070
1071static void
1072vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
1073{
1074	mutex_enter(&vq->vq_freelist_lock);
1075	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
1076	mutex_exit(&vq->vq_freelist_lock);
1077
1078	return;
1079}
1080
1081/*
1082 * Enqueue several dmamaps as a single request.
1083 */
1084/*
1085 * Typical usage:
1086 *  <queue size> number of followings are stored in arrays
1087 *  - command blocks (in dmamem) should be pre-allocated and mapped
1088 *  - dmamaps for command blocks should be pre-allocated and loaded
1089 *  - dmamaps for payload should be pre-allocated
1090 *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
1091 *	if (r)		// currently 0 or EAGAIN
1092 *	  return r;
1093 *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
1094 *	if (r) {
1095 *	  virtio_enqueue_abort(sc, vq, slot);
1096 *	  return r;
1097 *	}
1098 *	r = virtio_enqueue_reserve(sc, vq, slot,
1099 *				   dmamap_payload[slot]->dm_nsegs+1);
1100 *							// ^ +1 for command
1101 *	if (r) {	// currently 0 or EAGAIN
1102 *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
1103 *	  return r;					// do not call abort()
1104 *	}
1105 *	<setup and prepare commands>
1106 *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
1107 *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
1108 *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
1109 *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
1110 *	virtio_enqueue_commit(sc, vq, slot, true);
1111 */
1112
1113/*
1114 * enqueue_prep: allocate a slot number
1115 */
1116int
1117virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
1118{
1119	struct vq_entry *qe1;
1120
1121	KASSERT(slotp != NULL);
1122
1123	qe1 = vq_alloc_entry(vq);
1124	if (qe1 == NULL)
1125		return EAGAIN;
1126	/* next slot is not allocated yet */
1127	qe1->qe_next = -1;
1128	*slotp = qe1->qe_index;
1129
1130	return 0;
1131}
1132
1133/*
1134 * enqueue_reserve: allocate remaining slots and build the descriptor chain.
1135 */
1136int
1137virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
1138		       int slot, int nsegs)
1139{
1140	int indirect;
1141	struct vq_entry *qe1 = &vq->vq_entries[slot];
1142
1143	KASSERT(qe1->qe_next == -1);
1144	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
1145
1146	if ((vq->vq_indirect != NULL) &&
1147	    (nsegs >= MINSEG_INDIRECT) &&
1148	    (nsegs <= vq->vq_maxnsegs))
1149		indirect = 1;
1150	else
1151		indirect = 0;
1152	qe1->qe_indirect = indirect;
1153
1154	if (indirect) {
1155		struct vring_desc *vd;
1156		int i;
1157
1158		vd = &vq->vq_desc[qe1->qe_index];
1159		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
1160			+ vq->vq_indirectoffset;
1161		vd->addr += sizeof(struct vring_desc)
1162			* vq->vq_maxnsegs * qe1->qe_index;
1163		vd->len = sizeof(struct vring_desc) * nsegs;
1164		vd->flags = VRING_DESC_F_INDIRECT;
1165
1166		vd = vq->vq_indirect;
1167		vd += vq->vq_maxnsegs * qe1->qe_index;
1168		qe1->qe_desc_base = vd;
1169
1170		for (i = 0; i < nsegs-1; i++) {
1171			vd[i].flags = VRING_DESC_F_NEXT;
1172		}
1173		vd[i].flags = 0;
1174		qe1->qe_next = 0;
1175
1176		return 0;
1177	} else {
1178		struct vring_desc *vd;
1179		struct vq_entry *qe;
1180		int i, s;
1181
1182		vd = &vq->vq_desc[0];
1183		qe1->qe_desc_base = vd;
1184		qe1->qe_next = qe1->qe_index;
1185		s = slot;
1186		for (i = 0; i < nsegs - 1; i++) {
1187			qe = vq_alloc_entry(vq);
1188			if (qe == NULL) {
1189				vd[s].flags = 0;
1190				virtio_enqueue_abort(sc, vq, slot);
1191				return EAGAIN;
1192			}
1193			vd[s].flags = VRING_DESC_F_NEXT;
1194			vd[s].next = qe->qe_index;
1195			s = qe->qe_index;
1196		}
1197		vd[s].flags = 0;
1198
1199		return 0;
1200	}
1201}
1202
1203/*
1204 * enqueue: enqueue a single dmamap.
1205 */
1206int
1207virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1208	       bus_dmamap_t dmamap, bool write)
1209{
1210	struct vq_entry *qe1 = &vq->vq_entries[slot];
1211	struct vring_desc *vd = qe1->qe_desc_base;
1212	int i;
1213	int s = qe1->qe_next;
1214
1215	KASSERT(s >= 0);
1216	KASSERT(dmamap->dm_nsegs > 0);
1217
1218	for (i = 0; i < dmamap->dm_nsegs; i++) {
1219		vd[s].addr = dmamap->dm_segs[i].ds_addr;
1220		vd[s].len = dmamap->dm_segs[i].ds_len;
1221		if (!write)
1222			vd[s].flags |= VRING_DESC_F_WRITE;
1223		s = vd[s].next;
1224	}
1225	qe1->qe_next = s;
1226
1227	return 0;
1228}
1229
1230int
1231virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1232		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
1233		 bool write)
1234{
1235	struct vq_entry *qe1 = &vq->vq_entries[slot];
1236	struct vring_desc *vd = qe1->qe_desc_base;
1237	int s = qe1->qe_next;
1238
1239	KASSERT(s >= 0);
1240	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
1241	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
1242		(dmamap->dm_segs[0].ds_len >= start + len));
1243
1244	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
1245	vd[s].len = len;
1246	if (!write)
1247		vd[s].flags |= VRING_DESC_F_WRITE;
1248	qe1->qe_next = vd[s].next;
1249
1250	return 0;
1251}
1252
1253/*
1254 * enqueue_commit: add it to the aring.
1255 */
1256int
1257virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
1258		      bool notifynow)
1259{
1260	struct vq_entry *qe1;
1261
1262	if (slot < 0) {
1263		mutex_enter(&vq->vq_aring_lock);
1264		goto notify;
1265	}
1266	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
1267	qe1 = &vq->vq_entries[slot];
1268	if (qe1->qe_indirect)
1269		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
1270	mutex_enter(&vq->vq_aring_lock);
1271	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
1272
1273notify:
1274	if (notifynow) {
1275		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
1276		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
1277		membar_producer();
1278		vq->vq_avail->idx = vq->vq_avail_idx;
1279		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
1280		membar_producer();
1281		vq->vq_queued++;
1282		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
1283		membar_consumer();
1284		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
1285			nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
1286					  VIRTIO_CONFIG_QUEUE_NOTIFY,
1287					  vq->vq_index);
1288	}
1289	mutex_exit(&vq->vq_aring_lock);
1290
1291	return 0;
1292}
1293
1294/*
1295 * enqueue_abort: rollback.
1296 */
1297int
1298virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1299{
1300	struct vq_entry *qe = &vq->vq_entries[slot];
1301	struct vring_desc *vd;
1302	int s;
1303
1304	if (qe->qe_next < 0) {
1305		vq_free_entry(vq, qe);
1306		return 0;
1307	}
1308
1309	s = slot;
1310	vd = &vq->vq_desc[0];
1311	while (vd[s].flags & VRING_DESC_F_NEXT) {
1312		s = vd[s].next;
1313		vq_free_entry(vq, qe);
1314		qe = &vq->vq_entries[s];
1315	}
1316	vq_free_entry(vq, qe);
1317	return 0;
1318}
1319
1320/*
1321 * Dequeue a request.
1322 */
1323/*
1324 * dequeue: dequeue a request from uring; dmamap_sync for uring is
1325 *	    already done in the interrupt handler.
1326 */
1327int
1328virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
1329	       int *slotp, int *lenp)
1330{
1331	uint16_t slot, usedidx;
1332	struct vq_entry *qe;
1333
1334	if (vq->vq_used_idx == vq->vq_used->idx)
1335		return ENOENT;
1336	mutex_enter(&vq->vq_uring_lock);
1337	usedidx = vq->vq_used_idx++;
1338	mutex_exit(&vq->vq_uring_lock);
1339	usedidx %= vq->vq_num;
1340	slot = vq->vq_used->ring[usedidx].id;
1341	qe = &vq->vq_entries[slot];
1342
1343	if (qe->qe_indirect)
1344		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
1345
1346	if (slotp)
1347		*slotp = slot;
1348	if (lenp)
1349		*lenp = vq->vq_used->ring[usedidx].len;
1350
1351	return 0;
1352}
1353
1354/*
1355 * dequeue_commit: complete dequeue; the slot is recycled for future use.
1356 *                 if you forget to call this the slot will be leaked.
1357 */
1358int
1359virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1360{
1361	struct vq_entry *qe = &vq->vq_entries[slot];
1362	struct vring_desc *vd = &vq->vq_desc[0];
1363	int s = slot;
1364
1365	while (vd[s].flags & VRING_DESC_F_NEXT) {
1366		s = vd[s].next;
1367		vq_free_entry(vq, qe);
1368		qe = &vq->vq_entries[s];
1369	}
1370	vq_free_entry(vq, qe);
1371
1372	return 0;
1373}
1374
1375/*
1376 * Attach a child, fill all the members.
1377 */
1378void
1379virtio_child_attach_start(struct virtio_softc *sc, device_t child, int ipl,
1380		    struct virtqueue *vqs,
1381		    virtio_callback config_change,
1382		    virtio_callback intr_hand,
1383		    int req_flags, int req_features, const char *feat_bits)
1384{
1385	char buf[256];
1386	int features;
1387
1388	sc->sc_child = child;
1389	sc->sc_ipl = ipl;
1390	sc->sc_vqs = vqs;
1391	sc->sc_config_change = config_change;
1392	sc->sc_intrhand = intr_hand;
1393	sc->sc_flags = req_flags;
1394
1395	features = virtio_negotiate_features(sc, req_features);
1396	snprintb(buf, sizeof(buf), feat_bits, features);
1397	aprint_normal(": Features: %s\n", buf);
1398	aprint_naive("\n");
1399}
1400
1401int
1402virtio_child_attach_finish(struct virtio_softc *sc)
1403{
1404	int r;
1405
1406	r = virtio_setup_interrupts(sc);
1407	if (r != 0) {
1408		aprint_error_dev(sc->sc_dev, "failed to setup interrupts\n");
1409		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
1410		return 1;
1411	}
1412
1413	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
1414
1415	return 0;
1416}
1417
1418void
1419virtio_child_detach(struct virtio_softc *sc)
1420{
1421	sc->sc_child = NULL;
1422	sc->sc_vqs = NULL;
1423
1424	virtio_device_reset(sc);
1425
1426	virtio_free_interrupts(sc);
1427}
1428
1429void
1430virtio_child_attach_failed(struct virtio_softc *sc)
1431{
1432	virtio_child_detach(sc);
1433
1434	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
1435
1436	sc->sc_child = VIRTIO_CHILD_FAILED;
1437}
1438
1439bus_dma_tag_t
1440virtio_dmat(struct virtio_softc *sc)
1441{
1442	return sc->sc_dmat;
1443}
1444
1445device_t
1446virtio_child(struct virtio_softc *sc)
1447{
1448	return sc->sc_child;
1449}
1450
1451int
1452virtio_intrhand(struct virtio_softc *sc)
1453{
1454	return (sc->sc_intrhand)(sc);
1455}
1456
1457uint32_t
1458virtio_features(struct virtio_softc *sc)
1459{
1460	return sc->sc_features;
1461}
1462
1463MODULE(MODULE_CLASS_DRIVER, virtio, "pci");
1464
1465#ifdef _MODULE
1466#include "ioconf.c"
1467#endif
1468
1469static int
1470virtio_modcmd(modcmd_t cmd, void *opaque)
1471{
1472	int error = 0;
1473
1474#ifdef _MODULE
1475	switch (cmd) {
1476	case MODULE_CMD_INIT:
1477		error = config_init_component(cfdriver_ioconf_virtio,
1478		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
1479		break;
1480	case MODULE_CMD_FINI:
1481		error = config_fini_component(cfdriver_ioconf_virtio,
1482		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
1483		break;
1484	default:
1485		error = ENOTTY;
1486		break;
1487	}
1488#endif
1489
1490	return error;
1491}
1492