xref: /netbsd-src/sys/dev/pci/virtio.c (revision 200d779b75dbeafa7bc01fd0f60bc61185f6967b)
1 /*	$NetBSD: virtio.c,v 1.9 2015/05/05 10:58:35 ozaki-r 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 1.9 2015/05/05 10:58:35 ozaki-r 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 
39 #include <dev/pci/pcidevs.h>
40 #include <dev/pci/pcireg.h>
41 #include <dev/pci/pcivar.h>
42 
43 #include <dev/pci/virtioreg.h>
44 #include <dev/pci/virtiovar.h>
45 
46 #define MINSEG_INDIRECT		2 /* use indirect if nsegs >= this value */
47 
48 static int	virtio_match(device_t, cfdata_t, void *);
49 static void	virtio_attach(device_t, device_t, void *);
50 static int	virtio_detach(device_t, int);
51 static int	virtio_intr(void *arg);
52 static void	virtio_soft_intr(void *arg);
53 static void	virtio_init_vq(struct virtio_softc *,
54 		    struct virtqueue *, const bool);
55 
56 CFATTACH_DECL3_NEW(virtio, sizeof(struct virtio_softc),
57     virtio_match, virtio_attach, virtio_detach, NULL, NULL, NULL,
58     DVF_DETACH_SHUTDOWN);
59 
60 static void
61 virtio_set_status(struct virtio_softc *sc, int status)
62 {
63 	int old = 0;
64 
65 	if (status != 0)
66 		old = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
67 				       VIRTIO_CONFIG_DEVICE_STATUS);
68 	bus_space_write_1(sc->sc_iot, sc->sc_ioh, VIRTIO_CONFIG_DEVICE_STATUS,
69 			  status|old);
70 }
71 
72 #define virtio_device_reset(sc)	virtio_set_status((sc), 0)
73 
74 static int
75 virtio_match(device_t parent, cfdata_t match, void *aux)
76 {
77 	struct pci_attach_args *pa;
78 
79 	pa = (struct pci_attach_args *)aux;
80 	switch (PCI_VENDOR(pa->pa_id)) {
81 	case PCI_VENDOR_QUMRANET:
82 		if ((PCI_PRODUCT_QUMRANET_VIRTIO_1000 <=
83 		     PCI_PRODUCT(pa->pa_id)) &&
84 		    (PCI_PRODUCT(pa->pa_id) <=
85 		     PCI_PRODUCT_QUMRANET_VIRTIO_103F))
86 			return 1;
87 		break;
88 	}
89 
90 	return 0;
91 }
92 
93 static const char *virtio_device_name[] = {
94 	"Unknown (0)",		/* 0 */
95 	"Network",		/* 1 */
96 	"Block",		/* 2 */
97 	"Console",		/* 3 */
98 	"Entropy",		/* 4 */
99 	"Memory Balloon",	/* 5 */
100 	"Unknown (6)",		/* 6 */
101 	"Unknown (7)",		/* 7 */
102 	"Unknown (8)",		/* 8 */
103 	"9P Transport"		/* 9 */
104 };
105 #define NDEVNAMES	(sizeof(virtio_device_name)/sizeof(char*))
106 
107 static void
108 virtio_attach(device_t parent, device_t self, void *aux)
109 {
110 	struct virtio_softc *sc = device_private(self);
111 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
112 	pci_chipset_tag_t pc = pa->pa_pc;
113 	pcitag_t tag = pa->pa_tag;
114 	int revision;
115 	pcireg_t id;
116 	char const *intrstr;
117 	pci_intr_handle_t ih;
118 	char intrbuf[PCI_INTRSTR_LEN];
119 
120 	revision = PCI_REVISION(pa->pa_class);
121 	if (revision != 0) {
122 		aprint_normal(": unknown revision 0x%02x; giving up\n",
123 			      revision);
124 		return;
125 	}
126 	aprint_normal("\n");
127 	aprint_naive("\n");
128 
129 	/* subsystem ID shows what I am */
130 	id = pci_conf_read(pc, tag, PCI_SUBSYS_ID_REG);
131 	aprint_normal_dev(self, "Virtio %s Device (rev. 0x%02x)\n",
132 			  (PCI_SUBSYS_ID(id) < NDEVNAMES?
133 			   virtio_device_name[PCI_SUBSYS_ID(id)] : "Unknown"),
134 			  revision);
135 
136 	sc->sc_dev = self;
137 	sc->sc_pc = pc;
138 	sc->sc_tag = tag;
139 	sc->sc_iot = pa->pa_iot;
140 	sc->sc_dmat = pa->pa_dmat;
141 	sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
142 
143 	if (pci_mapreg_map(pa, PCI_MAPREG_START, PCI_MAPREG_TYPE_IO, 0,
144 			   &sc->sc_iot, &sc->sc_ioh, NULL, &sc->sc_iosize)) {
145 		aprint_error_dev(self, "can't map i/o space\n");
146 		return;
147 	}
148 
149 	virtio_device_reset(sc);
150 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
151 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
152 
153 	/* XXX: use softc as aux... */
154 	sc->sc_childdevid = PCI_SUBSYS_ID(id);
155 	sc->sc_child = NULL;
156 	config_found(self, sc, NULL);
157 	if (sc->sc_child == NULL) {
158 		aprint_error_dev(self,
159 				 "no matching child driver; not configured\n");
160 		return;
161 	}
162 	if (sc->sc_child == (void*)1) { /* this shows error */
163 		aprint_error_dev(self,
164 				 "virtio configuration failed\n");
165 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
166 		return;
167 	}
168 
169 	if (pci_intr_map(pa, &ih)) {
170 		aprint_error_dev(self, "couldn't map interrupt\n");
171 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
172 		return;
173 	}
174 
175 	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
176 
177 	if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
178 		pci_intr_setattr(pc, &ih, PCI_INTR_MPSAFE, true);
179 
180 	sc->sc_ih = pci_intr_establish(pc, ih, sc->sc_ipl, virtio_intr, sc);
181 
182 	if (sc->sc_ih == NULL) {
183 		aprint_error_dev(self, "couldn't establish interrupt");
184 		if (intrstr != NULL)
185 			aprint_error(" at %s", intrstr);
186 		aprint_error("\n");
187 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
188 		return;
189 	}
190 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
191 
192 	sc->sc_soft_ih = NULL;
193 	if (sc->sc_flags & VIRTIO_F_PCI_INTR_SOFTINT) {
194 		u_int flags = SOFTINT_NET;
195 		if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
196 			flags |= SOFTINT_MPSAFE;
197 
198 		sc->sc_soft_ih = softint_establish(flags, virtio_soft_intr, sc);
199 		if (sc->sc_soft_ih == NULL)
200 			aprint_error(": failed to establish soft interrupt\n");
201 	}
202 
203 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
204 
205 	return;
206 }
207 
208 static int
209 virtio_detach(device_t self, int flags)
210 {
211 	struct virtio_softc *sc = device_private(self);
212 	int r;
213 
214 	if (sc->sc_child != 0 && sc->sc_child != (void*)1) {
215 		r = config_detach(sc->sc_child, flags);
216 		if (r)
217 			return r;
218 	}
219 	KASSERT(sc->sc_child == 0 || sc->sc_child == (void*)1);
220 	KASSERT(sc->sc_vqs == 0);
221 	if (sc->sc_ih != NULL) {
222 		pci_intr_disestablish(sc->sc_pc, sc->sc_ih);
223 		sc->sc_ih = NULL;
224 	}
225 	if (sc->sc_iosize)
226 		bus_space_unmap(sc->sc_iot, sc->sc_ioh, sc->sc_iosize);
227 	sc->sc_iosize = 0;
228 
229 	return 0;
230 }
231 
232 /*
233  * Reset the device.
234  */
235 /*
236  * To reset the device to a known state, do following:
237  *	virtio_reset(sc);	     // this will stop the device activity
238  *	<dequeue finished requests>; // virtio_dequeue() still can be called
239  *	<revoke pending requests in the vqs if any>;
240  *	virtio_reinit_begin(sc);     // dequeue prohibitted
241  *	newfeatures = virtio_negotiate_features(sc, requestedfeatures);
242  *	<some other initialization>;
243  *	virtio_reinit_end(sc);	     // device activated; enqueue allowed
244  * Once attached, feature negotiation can only be allowed after virtio_reset.
245  */
246 void
247 virtio_reset(struct virtio_softc *sc)
248 {
249 	virtio_device_reset(sc);
250 }
251 
252 void
253 virtio_reinit_start(struct virtio_softc *sc)
254 {
255 	int i;
256 
257 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
258 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
259 	for (i = 0; i < sc->sc_nvqs; i++) {
260 		int n;
261 		struct virtqueue *vq = &sc->sc_vqs[i];
262 		bus_space_write_2(sc->sc_iot, sc->sc_ioh,
263 				  VIRTIO_CONFIG_QUEUE_SELECT,
264 				  vq->vq_index);
265 		n = bus_space_read_2(sc->sc_iot, sc->sc_ioh,
266 				     VIRTIO_CONFIG_QUEUE_SIZE);
267 		if (n == 0)	/* vq disappeared */
268 			continue;
269 		if (n != vq->vq_num) {
270 			panic("%s: virtqueue size changed, vq index %d\n",
271 			      device_xname(sc->sc_dev),
272 			      vq->vq_index);
273 		}
274 		virtio_init_vq(sc, vq, true);
275 		bus_space_write_4(sc->sc_iot, sc->sc_ioh,
276 				  VIRTIO_CONFIG_QUEUE_ADDRESS,
277 				  (vq->vq_dmamap->dm_segs[0].ds_addr
278 				   / VIRTIO_PAGE_SIZE));
279 	}
280 }
281 
282 void
283 virtio_reinit_end(struct virtio_softc *sc)
284 {
285 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
286 }
287 
288 /*
289  * Feature negotiation.
290  */
291 uint32_t
292 virtio_negotiate_features(struct virtio_softc *sc, uint32_t guest_features)
293 {
294 	uint32_t r;
295 
296 	if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) &&
297 	    !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */
298 		guest_features |= VIRTIO_F_RING_INDIRECT_DESC;
299 	r = bus_space_read_4(sc->sc_iot, sc->sc_ioh,
300 			     VIRTIO_CONFIG_DEVICE_FEATURES);
301 	r &= guest_features;
302 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
303 			  VIRTIO_CONFIG_GUEST_FEATURES, r);
304 	sc->sc_features = r;
305 	if (r & VIRTIO_F_RING_INDIRECT_DESC)
306 		sc->sc_indirect = true;
307 	else
308 		sc->sc_indirect = false;
309 
310 	return r;
311 }
312 
313 /*
314  * Device configuration registers.
315  */
316 uint8_t
317 virtio_read_device_config_1(struct virtio_softc *sc, int index)
318 {
319 	return bus_space_read_1(sc->sc_iot, sc->sc_ioh,
320 				sc->sc_config_offset + index);
321 }
322 
323 uint16_t
324 virtio_read_device_config_2(struct virtio_softc *sc, int index)
325 {
326 	return bus_space_read_2(sc->sc_iot, sc->sc_ioh,
327 				sc->sc_config_offset + index);
328 }
329 
330 uint32_t
331 virtio_read_device_config_4(struct virtio_softc *sc, int index)
332 {
333 	return bus_space_read_4(sc->sc_iot, sc->sc_ioh,
334 				sc->sc_config_offset + index);
335 }
336 
337 uint64_t
338 virtio_read_device_config_8(struct virtio_softc *sc, int index)
339 {
340 	uint64_t r;
341 
342 	r = bus_space_read_4(sc->sc_iot, sc->sc_ioh,
343 			     sc->sc_config_offset + index + sizeof(uint32_t));
344 	r <<= 32;
345 	r += bus_space_read_4(sc->sc_iot, sc->sc_ioh,
346 			      sc->sc_config_offset + index);
347 	return r;
348 }
349 
350 void
351 virtio_write_device_config_1(struct virtio_softc *sc,
352 			     int index, uint8_t value)
353 {
354 	bus_space_write_1(sc->sc_iot, sc->sc_ioh,
355 			  sc->sc_config_offset + index, value);
356 }
357 
358 void
359 virtio_write_device_config_2(struct virtio_softc *sc,
360 			     int index, uint16_t value)
361 {
362 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
363 			  sc->sc_config_offset + index, value);
364 }
365 
366 void
367 virtio_write_device_config_4(struct virtio_softc *sc,
368 			     int index, uint32_t value)
369 {
370 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
371 			  sc->sc_config_offset + index, value);
372 }
373 
374 void
375 virtio_write_device_config_8(struct virtio_softc *sc,
376 			     int index, uint64_t value)
377 {
378 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
379 			  sc->sc_config_offset + index,
380 			  value & 0xffffffff);
381 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
382 			  sc->sc_config_offset + index + sizeof(uint32_t),
383 			  value >> 32);
384 }
385 
386 /*
387  * Interrupt handler.
388  */
389 static int
390 virtio_intr(void *arg)
391 {
392 	struct virtio_softc *sc = arg;
393 	int isr, r = 0;
394 
395 	/* check and ack the interrupt */
396 	isr = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
397 			       VIRTIO_CONFIG_ISR_STATUS);
398 	if (isr == 0)
399 		return 0;
400 	if ((isr & VIRTIO_CONFIG_ISR_CONFIG_CHANGE) &&
401 	    (sc->sc_config_change != NULL))
402 		r = (sc->sc_config_change)(sc);
403 	if (sc->sc_intrhand != NULL) {
404 		if (sc->sc_soft_ih != NULL)
405 			softint_schedule(sc->sc_soft_ih);
406 		else
407 			r |= (sc->sc_intrhand)(sc);
408 	}
409 
410 	return r;
411 }
412 
413 static void
414 virtio_soft_intr(void *arg)
415 {
416 	struct virtio_softc *sc = arg;
417 
418 	KASSERT(sc->sc_intrhand != NULL);
419 
420 	(sc->sc_intrhand)(sc);
421 }
422 
423 /*
424  * dmamap sync operations for a virtqueue.
425  */
426 static inline void
427 vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops)
428 {
429 	/* availoffset == sizeof(vring_desc)*vq_num */
430 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset,
431 			ops);
432 }
433 
434 static inline void
435 vq_sync_aring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
436 {
437 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
438 			vq->vq_availoffset,
439 			offsetof(struct vring_avail, ring)
440 			 + vq->vq_num * sizeof(uint16_t),
441 			ops);
442 }
443 
444 static inline void
445 vq_sync_uring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
446 {
447 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
448 			vq->vq_usedoffset,
449 			offsetof(struct vring_used, ring)
450 			 + vq->vq_num * sizeof(struct vring_used_elem),
451 			ops);
452 }
453 
454 static inline void
455 vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot,
456 		     int ops)
457 {
458 	int offset = vq->vq_indirectoffset
459 		      + sizeof(struct vring_desc) * vq->vq_maxnsegs * slot;
460 
461 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
462 			offset, sizeof(struct vring_desc) * vq->vq_maxnsegs,
463 			ops);
464 }
465 
466 /*
467  * Can be used as sc_intrhand.
468  */
469 /*
470  * Scan vq, bus_dmamap_sync for the vqs (not for the payload),
471  * and calls (*vq_done)() if some entries are consumed.
472  */
473 int
474 virtio_vq_intr(struct virtio_softc *sc)
475 {
476 	struct virtqueue *vq;
477 	int i, r = 0;
478 
479 	for (i = 0; i < sc->sc_nvqs; i++) {
480 		vq = &sc->sc_vqs[i];
481 		if (vq->vq_queued) {
482 			vq->vq_queued = 0;
483 			vq_sync_aring(sc, vq, BUS_DMASYNC_POSTWRITE);
484 		}
485 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
486 		membar_consumer();
487 		if (vq->vq_used_idx != vq->vq_used->idx) {
488 			if (vq->vq_done)
489 				r |= (vq->vq_done)(vq);
490 		}
491 	}
492 
493 	return r;
494 }
495 
496 /*
497  * Start/stop vq interrupt.  No guarantee.
498  */
499 void
500 virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
501 {
502 	vq->vq_avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
503 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
504 	vq->vq_queued++;
505 }
506 
507 void
508 virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
509 {
510 	vq->vq_avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
511 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
512 	vq->vq_queued++;
513 }
514 
515 /*
516  * Initialize vq structure.
517  */
518 static void
519 virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq, const bool reinit)
520 {
521 	int i, j;
522 	int vq_size = vq->vq_num;
523 
524 	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
525 
526 	/* build the indirect descriptor chain */
527 	if (vq->vq_indirect != NULL) {
528 		struct vring_desc *vd;
529 
530 		for (i = 0; i < vq_size; i++) {
531 			vd = vq->vq_indirect;
532 			vd += vq->vq_maxnsegs * i;
533 			for (j = 0; j < vq->vq_maxnsegs-1; j++)
534 				vd[j].next = j + 1;
535 		}
536 	}
537 
538 	/* free slot management */
539 	SIMPLEQ_INIT(&vq->vq_freelist);
540 	for (i = 0; i < vq_size; i++) {
541 		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
542 				    &vq->vq_entries[i], qe_list);
543 		vq->vq_entries[i].qe_index = i;
544 	}
545 	if (!reinit)
546 		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
547 
548 	/* enqueue/dequeue status */
549 	vq->vq_avail_idx = 0;
550 	vq->vq_used_idx = 0;
551 	vq->vq_queued = 0;
552 	if (!reinit) {
553 		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
554 		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
555 	}
556 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
557 	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
558 	vq->vq_queued++;
559 }
560 
561 /*
562  * Allocate/free a vq.
563  */
564 int
565 virtio_alloc_vq(struct virtio_softc *sc,
566 		struct virtqueue *vq, int index, int maxsegsize, int maxnsegs,
567 		const char *name)
568 {
569 	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
570 	int rsegs, r;
571 #define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
572 				 ~(VIRTIO_PAGE_SIZE-1))
573 
574 	memset(vq, 0, sizeof(*vq));
575 
576 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
577 			  VIRTIO_CONFIG_QUEUE_SELECT, index);
578 	vq_size = bus_space_read_2(sc->sc_iot, sc->sc_ioh,
579 				   VIRTIO_CONFIG_QUEUE_SIZE);
580 	if (vq_size == 0) {
581 		aprint_error_dev(sc->sc_dev,
582 				 "virtqueue not exist, index %d for %s\n",
583 				 index, name);
584 		goto err;
585 	}
586 	/* allocsize1: descriptor table + avail ring + pad */
587 	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
588 				     + sizeof(uint16_t)*(2+vq_size));
589 	/* allocsize2: used ring + pad */
590 	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
591 				     + sizeof(struct vring_used_elem)*vq_size);
592 	/* allocsize3: indirect table */
593 	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
594 		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
595 	else
596 		allocsize3 = 0;
597 	allocsize = allocsize1 + allocsize2 + allocsize3;
598 
599 	/* alloc and map the memory */
600 	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
601 			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
602 	if (r != 0) {
603 		aprint_error_dev(sc->sc_dev,
604 				 "virtqueue %d for %s allocation failed, "
605 				 "error code %d\n", index, name, r);
606 		goto err;
607 	}
608 	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
609 			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
610 	if (r != 0) {
611 		aprint_error_dev(sc->sc_dev,
612 				 "virtqueue %d for %s map failed, "
613 				 "error code %d\n", index, name, r);
614 		goto err;
615 	}
616 	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
617 			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
618 	if (r != 0) {
619 		aprint_error_dev(sc->sc_dev,
620 				 "virtqueue %d for %s dmamap creation failed, "
621 				 "error code %d\n", index, name, r);
622 		goto err;
623 	}
624 	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
625 			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
626 	if (r != 0) {
627 		aprint_error_dev(sc->sc_dev,
628 				 "virtqueue %d for %s dmamap load failed, "
629 				 "error code %d\n", index, name, r);
630 		goto err;
631 	}
632 
633 	/* set the vq address */
634 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
635 			  VIRTIO_CONFIG_QUEUE_ADDRESS,
636 			  (vq->vq_dmamap->dm_segs[0].ds_addr
637 			   / VIRTIO_PAGE_SIZE));
638 
639 	/* remember addresses and offsets for later use */
640 	vq->vq_owner = sc;
641 	vq->vq_num = vq_size;
642 	vq->vq_index = index;
643 	vq->vq_desc = vq->vq_vaddr;
644 	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
645 	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
646 	vq->vq_usedoffset = allocsize1;
647 	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
648 	if (allocsize3 > 0) {
649 		vq->vq_indirectoffset = allocsize1 + allocsize2;
650 		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
651 					  + vq->vq_indirectoffset);
652 	}
653 	vq->vq_bytesize = allocsize;
654 	vq->vq_maxsegsize = maxsegsize;
655 	vq->vq_maxnsegs = maxnsegs;
656 
657 	/* free slot management */
658 	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
659 				     KM_NOSLEEP);
660 	if (vq->vq_entries == NULL) {
661 		r = ENOMEM;
662 		goto err;
663 	}
664 
665 	virtio_init_vq(sc, vq, false);
666 
667 	aprint_verbose_dev(sc->sc_dev,
668 			   "allocated %u byte for virtqueue %d for %s, "
669 			   "size %d\n", allocsize, index, name, vq_size);
670 	if (allocsize3 > 0)
671 		aprint_verbose_dev(sc->sc_dev,
672 				   "using %d byte (%d entries) "
673 				   "indirect descriptors\n",
674 				   allocsize3, maxnsegs * vq_size);
675 	return 0;
676 
677 err:
678 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
679 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
680 	if (vq->vq_dmamap)
681 		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
682 	if (vq->vq_vaddr)
683 		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
684 	if (vq->vq_segs[0].ds_addr)
685 		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
686 	memset(vq, 0, sizeof(*vq));
687 
688 	return -1;
689 }
690 
691 int
692 virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
693 {
694 	struct vq_entry *qe;
695 	int i = 0;
696 
697 	/* device must be already deactivated */
698 	/* confirm the vq is empty */
699 	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
700 		i++;
701 	}
702 	if (i != vq->vq_num) {
703 		printf("%s: freeing non-empty vq, index %d\n",
704 		       device_xname(sc->sc_dev), vq->vq_index);
705 		return EBUSY;
706 	}
707 
708 	/* tell device that there's no virtqueue any longer */
709 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
710 			  VIRTIO_CONFIG_QUEUE_SELECT, vq->vq_index);
711 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
712 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
713 
714 	kmem_free(vq->vq_entries, vq->vq_bytesize);
715 	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
716 	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
717 	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
718 	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
719 	mutex_destroy(&vq->vq_freelist_lock);
720 	mutex_destroy(&vq->vq_uring_lock);
721 	mutex_destroy(&vq->vq_aring_lock);
722 	memset(vq, 0, sizeof(*vq));
723 
724 	return 0;
725 }
726 
727 /*
728  * Free descriptor management.
729  */
730 static struct vq_entry *
731 vq_alloc_entry(struct virtqueue *vq)
732 {
733 	struct vq_entry *qe;
734 
735 	mutex_enter(&vq->vq_freelist_lock);
736 	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
737 		mutex_exit(&vq->vq_freelist_lock);
738 		return NULL;
739 	}
740 	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
741 	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
742 	mutex_exit(&vq->vq_freelist_lock);
743 
744 	return qe;
745 }
746 
747 static void
748 vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
749 {
750 	mutex_enter(&vq->vq_freelist_lock);
751 	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
752 	mutex_exit(&vq->vq_freelist_lock);
753 
754 	return;
755 }
756 
757 /*
758  * Enqueue several dmamaps as a single request.
759  */
760 /*
761  * Typical usage:
762  *  <queue size> number of followings are stored in arrays
763  *  - command blocks (in dmamem) should be pre-allocated and mapped
764  *  - dmamaps for command blocks should be pre-allocated and loaded
765  *  - dmamaps for payload should be pre-allocated
766  *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
767  *	if (r)		// currently 0 or EAGAIN
768  *	  return r;
769  *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
770  *	if (r) {
771  *	  virtio_enqueue_abort(sc, vq, slot);
772  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
773  *	  return r;
774  *	}
775  *	r = virtio_enqueue_reserve(sc, vq, slot,
776  *				   dmamap_payload[slot]->dm_nsegs+1);
777  *							// ^ +1 for command
778  *	if (r) {	// currently 0 or EAGAIN
779  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
780  *	  return r;					// do not call abort()
781  *	}
782  *	<setup and prepare commands>
783  *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
784  *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
785  *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
786  *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
787  *	virtio_enqueue_commit(sc, vq, slot, true);
788  */
789 
790 /*
791  * enqueue_prep: allocate a slot number
792  */
793 int
794 virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
795 {
796 	struct vq_entry *qe1;
797 
798 	KASSERT(slotp != NULL);
799 
800 	qe1 = vq_alloc_entry(vq);
801 	if (qe1 == NULL)
802 		return EAGAIN;
803 	/* next slot is not allocated yet */
804 	qe1->qe_next = -1;
805 	*slotp = qe1->qe_index;
806 
807 	return 0;
808 }
809 
810 /*
811  * enqueue_reserve: allocate remaining slots and build the descriptor chain.
812  */
813 int
814 virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
815 		       int slot, int nsegs)
816 {
817 	int indirect;
818 	struct vq_entry *qe1 = &vq->vq_entries[slot];
819 
820 	KASSERT(qe1->qe_next == -1);
821 	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
822 
823 	if ((vq->vq_indirect != NULL) &&
824 	    (nsegs >= MINSEG_INDIRECT) &&
825 	    (nsegs <= vq->vq_maxnsegs))
826 		indirect = 1;
827 	else
828 		indirect = 0;
829 	qe1->qe_indirect = indirect;
830 
831 	if (indirect) {
832 		struct vring_desc *vd;
833 		int i;
834 
835 		vd = &vq->vq_desc[qe1->qe_index];
836 		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
837 			+ vq->vq_indirectoffset;
838 		vd->addr += sizeof(struct vring_desc)
839 			* vq->vq_maxnsegs * qe1->qe_index;
840 		vd->len = sizeof(struct vring_desc) * nsegs;
841 		vd->flags = VRING_DESC_F_INDIRECT;
842 
843 		vd = vq->vq_indirect;
844 		vd += vq->vq_maxnsegs * qe1->qe_index;
845 		qe1->qe_desc_base = vd;
846 
847 		for (i = 0; i < nsegs-1; i++) {
848 			vd[i].flags = VRING_DESC_F_NEXT;
849 		}
850 		vd[i].flags = 0;
851 		qe1->qe_next = 0;
852 
853 		return 0;
854 	} else {
855 		struct vring_desc *vd;
856 		struct vq_entry *qe;
857 		int i, s;
858 
859 		vd = &vq->vq_desc[0];
860 		qe1->qe_desc_base = vd;
861 		qe1->qe_next = qe1->qe_index;
862 		s = slot;
863 		for (i = 0; i < nsegs - 1; i++) {
864 			qe = vq_alloc_entry(vq);
865 			if (qe == NULL) {
866 				vd[s].flags = 0;
867 				virtio_enqueue_abort(sc, vq, slot);
868 				return EAGAIN;
869 			}
870 			vd[s].flags = VRING_DESC_F_NEXT;
871 			vd[s].next = qe->qe_index;
872 			s = qe->qe_index;
873 		}
874 		vd[s].flags = 0;
875 
876 		return 0;
877 	}
878 }
879 
880 /*
881  * enqueue: enqueue a single dmamap.
882  */
883 int
884 virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
885 	       bus_dmamap_t dmamap, bool write)
886 {
887 	struct vq_entry *qe1 = &vq->vq_entries[slot];
888 	struct vring_desc *vd = qe1->qe_desc_base;
889 	int i;
890 	int s = qe1->qe_next;
891 
892 	KASSERT(s >= 0);
893 	KASSERT(dmamap->dm_nsegs > 0);
894 
895 	for (i = 0; i < dmamap->dm_nsegs; i++) {
896 		vd[s].addr = dmamap->dm_segs[i].ds_addr;
897 		vd[s].len = dmamap->dm_segs[i].ds_len;
898 		if (!write)
899 			vd[s].flags |= VRING_DESC_F_WRITE;
900 		s = vd[s].next;
901 	}
902 	qe1->qe_next = s;
903 
904 	return 0;
905 }
906 
907 int
908 virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
909 		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
910 		 bool write)
911 {
912 	struct vq_entry *qe1 = &vq->vq_entries[slot];
913 	struct vring_desc *vd = qe1->qe_desc_base;
914 	int s = qe1->qe_next;
915 
916 	KASSERT(s >= 0);
917 	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
918 	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
919 		(dmamap->dm_segs[0].ds_len >= start + len));
920 
921 	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
922 	vd[s].len = len;
923 	if (!write)
924 		vd[s].flags |= VRING_DESC_F_WRITE;
925 	qe1->qe_next = vd[s].next;
926 
927 	return 0;
928 }
929 
930 /*
931  * enqueue_commit: add it to the aring.
932  */
933 int
934 virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
935 		      bool notifynow)
936 {
937 	struct vq_entry *qe1;
938 
939 	if (slot < 0) {
940 		mutex_enter(&vq->vq_aring_lock);
941 		goto notify;
942 	}
943 	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
944 	qe1 = &vq->vq_entries[slot];
945 	if (qe1->qe_indirect)
946 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
947 	mutex_enter(&vq->vq_aring_lock);
948 	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
949 
950 notify:
951 	if (notifynow) {
952 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
953 		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
954 		membar_producer();
955 		vq->vq_avail->idx = vq->vq_avail_idx;
956 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
957 		membar_producer();
958 		vq->vq_queued++;
959 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
960 		membar_consumer();
961 		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
962 			bus_space_write_2(sc->sc_iot, sc->sc_ioh,
963 					  VIRTIO_CONFIG_QUEUE_NOTIFY,
964 					  vq->vq_index);
965 	}
966 	mutex_exit(&vq->vq_aring_lock);
967 
968 	return 0;
969 }
970 
971 /*
972  * enqueue_abort: rollback.
973  */
974 int
975 virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
976 {
977 	struct vq_entry *qe = &vq->vq_entries[slot];
978 	struct vring_desc *vd;
979 	int s;
980 
981 	if (qe->qe_next < 0) {
982 		vq_free_entry(vq, qe);
983 		return 0;
984 	}
985 
986 	s = slot;
987 	vd = &vq->vq_desc[0];
988 	while (vd[s].flags & VRING_DESC_F_NEXT) {
989 		s = vd[s].next;
990 		vq_free_entry(vq, qe);
991 		qe = &vq->vq_entries[s];
992 	}
993 	vq_free_entry(vq, qe);
994 	return 0;
995 }
996 
997 /*
998  * Dequeue a request.
999  */
1000 /*
1001  * dequeue: dequeue a request from uring; dmamap_sync for uring is
1002  *	    already done in the interrupt handler.
1003  */
1004 int
1005 virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
1006 	       int *slotp, int *lenp)
1007 {
1008 	uint16_t slot, usedidx;
1009 	struct vq_entry *qe;
1010 
1011 	if (vq->vq_used_idx == vq->vq_used->idx)
1012 		return ENOENT;
1013 	mutex_enter(&vq->vq_uring_lock);
1014 	usedidx = vq->vq_used_idx++;
1015 	mutex_exit(&vq->vq_uring_lock);
1016 	usedidx %= vq->vq_num;
1017 	slot = vq->vq_used->ring[usedidx].id;
1018 	qe = &vq->vq_entries[slot];
1019 
1020 	if (qe->qe_indirect)
1021 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
1022 
1023 	if (slotp)
1024 		*slotp = slot;
1025 	if (lenp)
1026 		*lenp = vq->vq_used->ring[usedidx].len;
1027 
1028 	return 0;
1029 }
1030 
1031 /*
1032  * dequeue_commit: complete dequeue; the slot is recycled for future use.
1033  *                 if you forget to call this the slot will be leaked.
1034  */
1035 int
1036 virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1037 {
1038 	struct vq_entry *qe = &vq->vq_entries[slot];
1039 	struct vring_desc *vd = &vq->vq_desc[0];
1040 	int s = slot;
1041 
1042 	while (vd[s].flags & VRING_DESC_F_NEXT) {
1043 		s = vd[s].next;
1044 		vq_free_entry(vq, qe);
1045 		qe = &vq->vq_entries[s];
1046 	}
1047 	vq_free_entry(vq, qe);
1048 
1049 	return 0;
1050 }
1051