xref: /netbsd-src/sys/dev/pci/virtio.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: virtio.c,v 1.8 2014/12/19 06:54:40 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 1.8 2014/12/19 06:54:40 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 
494 	return r;
495 }
496 
497 /*
498  * Start/stop vq interrupt.  No guarantee.
499  */
500 void
501 virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
502 {
503 	vq->vq_avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
504 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
505 	vq->vq_queued++;
506 }
507 
508 void
509 virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
510 {
511 	vq->vq_avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
512 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
513 	vq->vq_queued++;
514 }
515 
516 /*
517  * Initialize vq structure.
518  */
519 static void
520 virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq, const bool reinit)
521 {
522 	int i, j;
523 	int vq_size = vq->vq_num;
524 
525 	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
526 
527 	/* build the indirect descriptor chain */
528 	if (vq->vq_indirect != NULL) {
529 		struct vring_desc *vd;
530 
531 		for (i = 0; i < vq_size; i++) {
532 			vd = vq->vq_indirect;
533 			vd += vq->vq_maxnsegs * i;
534 			for (j = 0; j < vq->vq_maxnsegs-1; j++)
535 				vd[j].next = j + 1;
536 		}
537 	}
538 
539 	/* free slot management */
540 	SIMPLEQ_INIT(&vq->vq_freelist);
541 	for (i = 0; i < vq_size; i++) {
542 		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
543 				    &vq->vq_entries[i], qe_list);
544 		vq->vq_entries[i].qe_index = i;
545 	}
546 	if (!reinit)
547 		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
548 
549 	/* enqueue/dequeue status */
550 	vq->vq_avail_idx = 0;
551 	vq->vq_used_idx = 0;
552 	vq->vq_queued = 0;
553 	if (!reinit) {
554 		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
555 		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
556 	}
557 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
558 	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
559 	vq->vq_queued++;
560 }
561 
562 /*
563  * Allocate/free a vq.
564  */
565 int
566 virtio_alloc_vq(struct virtio_softc *sc,
567 		struct virtqueue *vq, int index, int maxsegsize, int maxnsegs,
568 		const char *name)
569 {
570 	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
571 	int rsegs, r;
572 #define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
573 				 ~(VIRTIO_PAGE_SIZE-1))
574 
575 	memset(vq, 0, sizeof(*vq));
576 
577 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
578 			  VIRTIO_CONFIG_QUEUE_SELECT, index);
579 	vq_size = bus_space_read_2(sc->sc_iot, sc->sc_ioh,
580 				   VIRTIO_CONFIG_QUEUE_SIZE);
581 	if (vq_size == 0) {
582 		aprint_error_dev(sc->sc_dev,
583 				 "virtqueue not exist, index %d for %s\n",
584 				 index, name);
585 		goto err;
586 	}
587 	/* allocsize1: descriptor table + avail ring + pad */
588 	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
589 				     + sizeof(uint16_t)*(2+vq_size));
590 	/* allocsize2: used ring + pad */
591 	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
592 				     + sizeof(struct vring_used_elem)*vq_size);
593 	/* allocsize3: indirect table */
594 	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
595 		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
596 	else
597 		allocsize3 = 0;
598 	allocsize = allocsize1 + allocsize2 + allocsize3;
599 
600 	/* alloc and map the memory */
601 	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
602 			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
603 	if (r != 0) {
604 		aprint_error_dev(sc->sc_dev,
605 				 "virtqueue %d for %s allocation failed, "
606 				 "error code %d\n", index, name, r);
607 		goto err;
608 	}
609 	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
610 			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
611 	if (r != 0) {
612 		aprint_error_dev(sc->sc_dev,
613 				 "virtqueue %d for %s map failed, "
614 				 "error code %d\n", index, name, r);
615 		goto err;
616 	}
617 	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
618 			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
619 	if (r != 0) {
620 		aprint_error_dev(sc->sc_dev,
621 				 "virtqueue %d for %s dmamap creation failed, "
622 				 "error code %d\n", index, name, r);
623 		goto err;
624 	}
625 	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
626 			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
627 	if (r != 0) {
628 		aprint_error_dev(sc->sc_dev,
629 				 "virtqueue %d for %s dmamap load failed, "
630 				 "error code %d\n", index, name, r);
631 		goto err;
632 	}
633 
634 	/* set the vq address */
635 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
636 			  VIRTIO_CONFIG_QUEUE_ADDRESS,
637 			  (vq->vq_dmamap->dm_segs[0].ds_addr
638 			   / VIRTIO_PAGE_SIZE));
639 
640 	/* remember addresses and offsets for later use */
641 	vq->vq_owner = sc;
642 	vq->vq_num = vq_size;
643 	vq->vq_index = index;
644 	vq->vq_desc = vq->vq_vaddr;
645 	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
646 	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
647 	vq->vq_usedoffset = allocsize1;
648 	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
649 	if (allocsize3 > 0) {
650 		vq->vq_indirectoffset = allocsize1 + allocsize2;
651 		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
652 					  + vq->vq_indirectoffset);
653 	}
654 	vq->vq_bytesize = allocsize;
655 	vq->vq_maxsegsize = maxsegsize;
656 	vq->vq_maxnsegs = maxnsegs;
657 
658 	/* free slot management */
659 	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
660 				     KM_NOSLEEP);
661 	if (vq->vq_entries == NULL) {
662 		r = ENOMEM;
663 		goto err;
664 	}
665 
666 	virtio_init_vq(sc, vq, false);
667 
668 	aprint_verbose_dev(sc->sc_dev,
669 			   "allocated %u byte for virtqueue %d for %s, "
670 			   "size %d\n", allocsize, index, name, vq_size);
671 	if (allocsize3 > 0)
672 		aprint_verbose_dev(sc->sc_dev,
673 				   "using %d byte (%d entries) "
674 				   "indirect descriptors\n",
675 				   allocsize3, maxnsegs * vq_size);
676 	return 0;
677 
678 err:
679 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
680 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
681 	if (vq->vq_dmamap)
682 		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
683 	if (vq->vq_vaddr)
684 		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
685 	if (vq->vq_segs[0].ds_addr)
686 		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
687 	memset(vq, 0, sizeof(*vq));
688 
689 	return -1;
690 }
691 
692 int
693 virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
694 {
695 	struct vq_entry *qe;
696 	int i = 0;
697 
698 	/* device must be already deactivated */
699 	/* confirm the vq is empty */
700 	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
701 		i++;
702 	}
703 	if (i != vq->vq_num) {
704 		printf("%s: freeing non-empty vq, index %d\n",
705 		       device_xname(sc->sc_dev), vq->vq_index);
706 		return EBUSY;
707 	}
708 
709 	/* tell device that there's no virtqueue any longer */
710 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
711 			  VIRTIO_CONFIG_QUEUE_SELECT, vq->vq_index);
712 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
713 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
714 
715 	kmem_free(vq->vq_entries, vq->vq_bytesize);
716 	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
717 	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
718 	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
719 	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
720 	mutex_destroy(&vq->vq_freelist_lock);
721 	mutex_destroy(&vq->vq_uring_lock);
722 	mutex_destroy(&vq->vq_aring_lock);
723 	memset(vq, 0, sizeof(*vq));
724 
725 	return 0;
726 }
727 
728 /*
729  * Free descriptor management.
730  */
731 static struct vq_entry *
732 vq_alloc_entry(struct virtqueue *vq)
733 {
734 	struct vq_entry *qe;
735 
736 	mutex_enter(&vq->vq_freelist_lock);
737 	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
738 		mutex_exit(&vq->vq_freelist_lock);
739 		return NULL;
740 	}
741 	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
742 	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
743 	mutex_exit(&vq->vq_freelist_lock);
744 
745 	return qe;
746 }
747 
748 static void
749 vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
750 {
751 	mutex_enter(&vq->vq_freelist_lock);
752 	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
753 	mutex_exit(&vq->vq_freelist_lock);
754 
755 	return;
756 }
757 
758 /*
759  * Enqueue several dmamaps as a single request.
760  */
761 /*
762  * Typical usage:
763  *  <queue size> number of followings are stored in arrays
764  *  - command blocks (in dmamem) should be pre-allocated and mapped
765  *  - dmamaps for command blocks should be pre-allocated and loaded
766  *  - dmamaps for payload should be pre-allocated
767  *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
768  *	if (r)		// currently 0 or EAGAIN
769  *	  return r;
770  *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
771  *	if (r) {
772  *	  virtio_enqueue_abort(sc, vq, slot);
773  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
774  *	  return r;
775  *	}
776  *	r = virtio_enqueue_reserve(sc, vq, slot,
777  *				   dmamap_payload[slot]->dm_nsegs+1);
778  *							// ^ +1 for command
779  *	if (r) {	// currently 0 or EAGAIN
780  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
781  *	  return r;					// do not call abort()
782  *	}
783  *	<setup and prepare commands>
784  *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
785  *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
786  *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
787  *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
788  *	virtio_enqueue_commit(sc, vq, slot, true);
789  */
790 
791 /*
792  * enqueue_prep: allocate a slot number
793  */
794 int
795 virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
796 {
797 	struct vq_entry *qe1;
798 
799 	KASSERT(slotp != NULL);
800 
801 	qe1 = vq_alloc_entry(vq);
802 	if (qe1 == NULL)
803 		return EAGAIN;
804 	/* next slot is not allocated yet */
805 	qe1->qe_next = -1;
806 	*slotp = qe1->qe_index;
807 
808 	return 0;
809 }
810 
811 /*
812  * enqueue_reserve: allocate remaining slots and build the descriptor chain.
813  */
814 int
815 virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
816 		       int slot, int nsegs)
817 {
818 	int indirect;
819 	struct vq_entry *qe1 = &vq->vq_entries[slot];
820 
821 	KASSERT(qe1->qe_next == -1);
822 	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
823 
824 	if ((vq->vq_indirect != NULL) &&
825 	    (nsegs >= MINSEG_INDIRECT) &&
826 	    (nsegs <= vq->vq_maxnsegs))
827 		indirect = 1;
828 	else
829 		indirect = 0;
830 	qe1->qe_indirect = indirect;
831 
832 	if (indirect) {
833 		struct vring_desc *vd;
834 		int i;
835 
836 		vd = &vq->vq_desc[qe1->qe_index];
837 		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
838 			+ vq->vq_indirectoffset;
839 		vd->addr += sizeof(struct vring_desc)
840 			* vq->vq_maxnsegs * qe1->qe_index;
841 		vd->len = sizeof(struct vring_desc) * nsegs;
842 		vd->flags = VRING_DESC_F_INDIRECT;
843 
844 		vd = vq->vq_indirect;
845 		vd += vq->vq_maxnsegs * qe1->qe_index;
846 		qe1->qe_desc_base = vd;
847 
848 		for (i = 0; i < nsegs-1; i++) {
849 			vd[i].flags = VRING_DESC_F_NEXT;
850 		}
851 		vd[i].flags = 0;
852 		qe1->qe_next = 0;
853 
854 		return 0;
855 	} else {
856 		struct vring_desc *vd;
857 		struct vq_entry *qe;
858 		int i, s;
859 
860 		vd = &vq->vq_desc[0];
861 		qe1->qe_desc_base = vd;
862 		qe1->qe_next = qe1->qe_index;
863 		s = slot;
864 		for (i = 0; i < nsegs - 1; i++) {
865 			qe = vq_alloc_entry(vq);
866 			if (qe == NULL) {
867 				vd[s].flags = 0;
868 				virtio_enqueue_abort(sc, vq, slot);
869 				return EAGAIN;
870 			}
871 			vd[s].flags = VRING_DESC_F_NEXT;
872 			vd[s].next = qe->qe_index;
873 			s = qe->qe_index;
874 		}
875 		vd[s].flags = 0;
876 
877 		return 0;
878 	}
879 }
880 
881 /*
882  * enqueue: enqueue a single dmamap.
883  */
884 int
885 virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
886 	       bus_dmamap_t dmamap, bool write)
887 {
888 	struct vq_entry *qe1 = &vq->vq_entries[slot];
889 	struct vring_desc *vd = qe1->qe_desc_base;
890 	int i;
891 	int s = qe1->qe_next;
892 
893 	KASSERT(s >= 0);
894 	KASSERT(dmamap->dm_nsegs > 0);
895 
896 	for (i = 0; i < dmamap->dm_nsegs; i++) {
897 		vd[s].addr = dmamap->dm_segs[i].ds_addr;
898 		vd[s].len = dmamap->dm_segs[i].ds_len;
899 		if (!write)
900 			vd[s].flags |= VRING_DESC_F_WRITE;
901 		s = vd[s].next;
902 	}
903 	qe1->qe_next = s;
904 
905 	return 0;
906 }
907 
908 int
909 virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
910 		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
911 		 bool write)
912 {
913 	struct vq_entry *qe1 = &vq->vq_entries[slot];
914 	struct vring_desc *vd = qe1->qe_desc_base;
915 	int s = qe1->qe_next;
916 
917 	KASSERT(s >= 0);
918 	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
919 	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
920 		(dmamap->dm_segs[0].ds_len >= start + len));
921 
922 	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
923 	vd[s].len = len;
924 	if (!write)
925 		vd[s].flags |= VRING_DESC_F_WRITE;
926 	qe1->qe_next = vd[s].next;
927 
928 	return 0;
929 }
930 
931 /*
932  * enqueue_commit: add it to the aring.
933  */
934 int
935 virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
936 		      bool notifynow)
937 {
938 	struct vq_entry *qe1;
939 
940 	if (slot < 0) {
941 		mutex_enter(&vq->vq_aring_lock);
942 		goto notify;
943 	}
944 	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
945 	qe1 = &vq->vq_entries[slot];
946 	if (qe1->qe_indirect)
947 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
948 	mutex_enter(&vq->vq_aring_lock);
949 	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
950 
951 notify:
952 	if (notifynow) {
953 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
954 		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
955 		membar_producer();
956 		vq->vq_avail->idx = vq->vq_avail_idx;
957 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
958 		membar_producer();
959 		vq->vq_queued++;
960 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
961 		membar_consumer();
962 		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
963 			bus_space_write_2(sc->sc_iot, sc->sc_ioh,
964 					  VIRTIO_CONFIG_QUEUE_NOTIFY,
965 					  vq->vq_index);
966 	}
967 	mutex_exit(&vq->vq_aring_lock);
968 
969 	return 0;
970 }
971 
972 /*
973  * enqueue_abort: rollback.
974  */
975 int
976 virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
977 {
978 	struct vq_entry *qe = &vq->vq_entries[slot];
979 	struct vring_desc *vd;
980 	int s;
981 
982 	if (qe->qe_next < 0) {
983 		vq_free_entry(vq, qe);
984 		return 0;
985 	}
986 
987 	s = slot;
988 	vd = &vq->vq_desc[0];
989 	while (vd[s].flags & VRING_DESC_F_NEXT) {
990 		s = vd[s].next;
991 		vq_free_entry(vq, qe);
992 		qe = &vq->vq_entries[s];
993 	}
994 	vq_free_entry(vq, qe);
995 	return 0;
996 }
997 
998 /*
999  * Dequeue a request.
1000  */
1001 /*
1002  * dequeue: dequeue a request from uring; dmamap_sync for uring is
1003  *	    already done in the interrupt handler.
1004  */
1005 int
1006 virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
1007 	       int *slotp, int *lenp)
1008 {
1009 	uint16_t slot, usedidx;
1010 	struct vq_entry *qe;
1011 
1012 	if (vq->vq_used_idx == vq->vq_used->idx)
1013 		return ENOENT;
1014 	mutex_enter(&vq->vq_uring_lock);
1015 	usedidx = vq->vq_used_idx++;
1016 	mutex_exit(&vq->vq_uring_lock);
1017 	usedidx %= vq->vq_num;
1018 	slot = vq->vq_used->ring[usedidx].id;
1019 	qe = &vq->vq_entries[slot];
1020 
1021 	if (qe->qe_indirect)
1022 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
1023 
1024 	if (slotp)
1025 		*slotp = slot;
1026 	if (lenp)
1027 		*lenp = vq->vq_used->ring[usedidx].len;
1028 
1029 	return 0;
1030 }
1031 
1032 /*
1033  * dequeue_commit: complete dequeue; the slot is recycled for future use.
1034  *                 if you forget to call this the slot will be leaked.
1035  */
1036 int
1037 virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1038 {
1039 	struct vq_entry *qe = &vq->vq_entries[slot];
1040 	struct vring_desc *vd = &vq->vq_desc[0];
1041 	int s = slot;
1042 
1043 	while (vd[s].flags & VRING_DESC_F_NEXT) {
1044 		s = vd[s].next;
1045 		vq_free_entry(vq, qe);
1046 		qe = &vq->vq_entries[s];
1047 	}
1048 	vq_free_entry(vq, qe);
1049 
1050 	return 0;
1051 }
1052