xref: /netbsd-src/sys/dev/pci/virtio.c (revision 5bbd2a12505d72a8177929a37b5cee489d0a1cfd)
1 /*	$NetBSD: virtio.c,v 1.3 2011/11/02 23:05:52 njoly 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 1.3 2011/11/02 23:05:52 njoly 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_init_vq(struct virtio_softc *,
53 		    struct virtqueue *, const bool);
54 
55 CFATTACH_DECL3_NEW(virtio, sizeof(struct virtio_softc),
56     virtio_match, virtio_attach, virtio_detach, NULL, NULL, NULL,
57     DVF_DETACH_SHUTDOWN);
58 
59 static void
60 virtio_set_status(struct virtio_softc *sc, int status)
61 {
62 	int old = 0;
63 
64 	if (status != 0)
65 		old = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
66 				       VIRTIO_CONFIG_DEVICE_STATUS);
67 	bus_space_write_1(sc->sc_iot, sc->sc_ioh, VIRTIO_CONFIG_DEVICE_STATUS,
68 			  status|old);
69 }
70 
71 #define virtio_device_reset(sc)	virtio_set_status((sc), 0)
72 
73 static int
74 virtio_match(device_t parent, cfdata_t match, void *aux)
75 {
76 	struct pci_attach_args *pa;
77 
78 	pa = (struct pci_attach_args *)aux;
79 	switch (PCI_VENDOR(pa->pa_id)) {
80 	case PCI_VENDOR_QUMRANET:
81 		if ((PCI_PRODUCT_QUMRANET_VIRTIO_1000 <=
82 		     PCI_PRODUCT(pa->pa_id)) &&
83 		    (PCI_PRODUCT(pa->pa_id) <=
84 		     PCI_PRODUCT_QUMRANET_VIRTIO_103F))
85 			return 1;
86 		break;
87 	}
88 
89 	return 0;
90 }
91 
92 static const char *virtio_device_name[] = {
93 	"Unknown (0)",		/* 0 */
94 	"Network",		/* 1 */
95 	"Block",		/* 2 */
96 	"Console",		/* 3 */
97 	"Entropy",		/* 4 */
98 	"Memory Balloon",	/* 5 */
99 	"Unknown (6)",		/* 6 */
100 	"Unknown (7)",		/* 7 */
101 	"Unknown (8)",		/* 8 */
102 	"9P Transport"		/* 9 */
103 };
104 #define NDEVNAMES	(sizeof(virtio_device_name)/sizeof(char*))
105 
106 static void
107 virtio_attach(device_t parent, device_t self, void *aux)
108 {
109 	struct virtio_softc *sc = device_private(self);
110 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
111 	pci_chipset_tag_t pc = pa->pa_pc;
112 	pcitag_t tag = pa->pa_tag;
113 	int revision;
114 	pcireg_t id;
115 	char const *intrstr;
116 	pci_intr_handle_t ih;
117 
118 	revision = PCI_REVISION(pa->pa_class);
119 	if (revision != 0) {
120 		aprint_normal(": unknown revision 0x%02x; giving up\n",
121 			      revision);
122 		return;
123 	}
124 	aprint_normal("\n");
125 	aprint_naive("\n");
126 
127 	/* subsystem ID shows what I am */
128 	id = pci_conf_read(pc, tag, PCI_SUBSYS_ID_REG);
129 	aprint_normal_dev(self, "Virtio %s Device (rev. 0x%02x)\n",
130 			  (PCI_PRODUCT(id) < NDEVNAMES?
131 			   virtio_device_name[PCI_PRODUCT(id)] : "Unknown"),
132 			  revision);
133 
134 	sc->sc_dev = self;
135 	sc->sc_pc = pc;
136 	sc->sc_tag = tag;
137 	sc->sc_iot = pa->pa_iot;
138 	sc->sc_dmat = pa->pa_dmat;
139 	sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
140 
141 	if (pci_mapreg_map(pa, PCI_MAPREG_START, PCI_MAPREG_TYPE_IO, 0,
142 			   &sc->sc_iot, &sc->sc_ioh, NULL, &sc->sc_iosize)) {
143 		aprint_error_dev(self, "can't map i/o space\n");
144 		return;
145 	}
146 
147 	virtio_device_reset(sc);
148 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
149 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
150 
151 	/* XXX: use softc as aux... */
152 	sc->sc_childdevid = PCI_PRODUCT(id);
153 	sc->sc_child = NULL;
154 	config_found(self, sc, NULL);
155 	if (sc->sc_child == NULL) {
156 		aprint_error_dev(self,
157 				 "no matching child driver; not configured\n");
158 		return;
159 	}
160 	if (sc->sc_child == (void*)1) { /* this shows error */
161 		aprint_error_dev(self,
162 				 "virtio configuration failed\n");
163 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
164 		return;
165 	}
166 
167 	if (pci_intr_map(pa, &ih)) {
168 		aprint_error_dev(self, "couldn't map interrupt\n");
169 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
170 		return;
171 	}
172 	intrstr = pci_intr_string(pc, ih);
173 	sc->sc_ih = pci_intr_establish(pc, ih, sc->sc_ipl, virtio_intr, sc);
174 	if (sc->sc_ih == NULL) {
175 		aprint_error_dev(self, "couldn't establish interrupt");
176 		if (intrstr != NULL)
177 			aprint_error(" at %s", intrstr);
178 		aprint_error("\n");
179 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
180 		return;
181 	}
182 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
183 
184 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
185 
186 	return;
187 }
188 
189 static int
190 virtio_detach(device_t self, int flags)
191 {
192 	struct virtio_softc *sc = device_private(self);
193 	int r;
194 
195 	if (sc->sc_child != 0 && sc->sc_child != (void*)1) {
196 		r = config_detach(sc->sc_child, flags);
197 		if (r)
198 			return r;
199 	}
200 	KASSERT(sc->sc_child == 0 || sc->sc_child == (void*)1);
201 	KASSERT(sc->sc_vqs == 0);
202 	pci_intr_disestablish(sc->sc_pc, sc->sc_ih);
203 	sc->sc_ih = 0;
204 	if (sc->sc_iosize)
205 		bus_space_unmap(sc->sc_iot, sc->sc_ioh, sc->sc_iosize);
206 	sc->sc_iosize = 0;
207 
208 	return 0;
209 }
210 
211 /*
212  * Reset the device.
213  */
214 /*
215  * To reset the device to a known state, do following:
216  *	virtio_reset(sc);	     // this will stop the device activity
217  *	<dequeue finished requests>; // virtio_dequeue() still can be called
218  *	<revoke pending requests in the vqs if any>;
219  *	virtio_reinit_begin(sc);     // dequeue prohibitted
220  *	newfeatures = virtio_negotiate_features(sc, requestedfeatures);
221  *	<some other initialization>;
222  *	virtio_reinit_end(sc);	     // device activated; enqueue allowed
223  * Once attached, feature negotiation can only be allowed after virtio_reset.
224  */
225 void
226 virtio_reset(struct virtio_softc *sc)
227 {
228 	virtio_device_reset(sc);
229 }
230 
231 void
232 virtio_reinit_start(struct virtio_softc *sc)
233 {
234 	int i;
235 
236 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
237 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
238 	for (i = 0; i < sc->sc_nvqs; i++) {
239 		int n;
240 		struct virtqueue *vq = &sc->sc_vqs[i];
241 		bus_space_write_2(sc->sc_iot, sc->sc_ioh,
242 				  VIRTIO_CONFIG_QUEUE_SELECT,
243 				  vq->vq_index);
244 		n = bus_space_read_2(sc->sc_iot, sc->sc_ioh,
245 				     VIRTIO_CONFIG_QUEUE_SIZE);
246 		if (n == 0)	/* vq disappeared */
247 			continue;
248 		if (n != vq->vq_num) {
249 			panic("%s: virtqueue size changed, vq index %d\n",
250 			      device_xname(sc->sc_dev),
251 			      vq->vq_index);
252 		}
253 		virtio_init_vq(sc, vq, true);
254 		bus_space_write_4(sc->sc_iot, sc->sc_ioh,
255 				  VIRTIO_CONFIG_QUEUE_ADDRESS,
256 				  (vq->vq_dmamap->dm_segs[0].ds_addr
257 				   / VIRTIO_PAGE_SIZE));
258 	}
259 }
260 
261 void
262 virtio_reinit_end(struct virtio_softc *sc)
263 {
264 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
265 }
266 
267 /*
268  * Feature negotiation.
269  */
270 uint32_t
271 virtio_negotiate_features(struct virtio_softc *sc, uint32_t guest_features)
272 {
273 	uint32_t r;
274 
275 	if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) &&
276 	    !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */
277 		guest_features |= VIRTIO_F_RING_INDIRECT_DESC;
278 	r = bus_space_read_4(sc->sc_iot, sc->sc_ioh,
279 			     VIRTIO_CONFIG_DEVICE_FEATURES);
280 	r &= guest_features;
281 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
282 			  VIRTIO_CONFIG_GUEST_FEATURES, r);
283 	sc->sc_features = r;
284 	if (r & VIRTIO_F_RING_INDIRECT_DESC)
285 		sc->sc_indirect = true;
286 	else
287 		sc->sc_indirect = false;
288 
289 	return r;
290 }
291 
292 /*
293  * Device configuration registers.
294  */
295 uint8_t
296 virtio_read_device_config_1(struct virtio_softc *sc, int index)
297 {
298 	return bus_space_read_1(sc->sc_iot, sc->sc_ioh,
299 				sc->sc_config_offset + index);
300 }
301 
302 uint16_t
303 virtio_read_device_config_2(struct virtio_softc *sc, int index)
304 {
305 	return bus_space_read_2(sc->sc_iot, sc->sc_ioh,
306 				sc->sc_config_offset + index);
307 }
308 
309 uint32_t
310 virtio_read_device_config_4(struct virtio_softc *sc, int index)
311 {
312 	return bus_space_read_4(sc->sc_iot, sc->sc_ioh,
313 				sc->sc_config_offset + index);
314 }
315 
316 uint64_t
317 virtio_read_device_config_8(struct virtio_softc *sc, int index)
318 {
319 	uint64_t r;
320 
321 	r = bus_space_read_4(sc->sc_iot, sc->sc_ioh,
322 			     sc->sc_config_offset + index + sizeof(uint32_t));
323 	r <<= 32;
324 	r += bus_space_read_4(sc->sc_iot, sc->sc_ioh,
325 			      sc->sc_config_offset + index);
326 	return r;
327 }
328 
329 void
330 virtio_write_device_config_1(struct virtio_softc *sc,
331 			     int index, uint8_t value)
332 {
333 	bus_space_write_1(sc->sc_iot, sc->sc_ioh,
334 			  sc->sc_config_offset + index, value);
335 }
336 
337 void
338 virtio_write_device_config_2(struct virtio_softc *sc,
339 			     int index, uint16_t value)
340 {
341 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
342 			  sc->sc_config_offset + index, value);
343 }
344 
345 void
346 virtio_write_device_config_4(struct virtio_softc *sc,
347 			     int index, uint32_t value)
348 {
349 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
350 			  sc->sc_config_offset + index, value);
351 }
352 
353 void
354 virtio_write_device_config_8(struct virtio_softc *sc,
355 			     int index, uint64_t value)
356 {
357 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
358 			  sc->sc_config_offset + index,
359 			  value & 0xffffffff);
360 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
361 			  sc->sc_config_offset + index + sizeof(uint32_t),
362 			  value >> 32);
363 }
364 
365 /*
366  * Interrupt handler.
367  */
368 static int
369 virtio_intr(void *arg)
370 {
371 	struct virtio_softc *sc = arg;
372 	int isr, r = 0;
373 
374 	/* check and ack the interrupt */
375 	isr = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
376 			       VIRTIO_CONFIG_ISR_STATUS);
377 	if (isr == 0)
378 		return 0;
379 	if ((isr & VIRTIO_CONFIG_ISR_CONFIG_CHANGE) &&
380 	    (sc->sc_config_change != NULL))
381 		r = (sc->sc_config_change)(sc);
382 	if (sc->sc_intrhand != NULL)
383 		r |= (sc->sc_intrhand)(sc);
384 
385 	return r;
386 }
387 
388 /*
389  * dmamap sync operations for a virtqueue.
390  */
391 static inline void
392 vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops)
393 {
394 	/* availoffset == sizeof(vring_desc)*vq_num */
395 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset,
396 			ops);
397 }
398 
399 static inline void
400 vq_sync_aring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
401 {
402 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
403 			vq->vq_availoffset,
404 			offsetof(struct vring_avail, ring)
405 			 + vq->vq_num * sizeof(uint16_t),
406 			ops);
407 }
408 
409 static inline void
410 vq_sync_uring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
411 {
412 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
413 			vq->vq_usedoffset,
414 			offsetof(struct vring_used, ring)
415 			 + vq->vq_num * sizeof(struct vring_used_elem),
416 			ops);
417 }
418 
419 static inline void
420 vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot,
421 		     int ops)
422 {
423 	int offset = vq->vq_indirectoffset
424 		      + sizeof(struct vring_desc) * vq->vq_maxnsegs * slot;
425 
426 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
427 			offset, sizeof(struct vring_desc) * vq->vq_maxnsegs,
428 			ops);
429 }
430 
431 /*
432  * Can be used as sc_intrhand.
433  */
434 /*
435  * Scan vq, bus_dmamap_sync for the vqs (not for the payload),
436  * and calls (*vq_done)() if some entries are consumed.
437  */
438 int
439 virtio_vq_intr(struct virtio_softc *sc)
440 {
441 	struct virtqueue *vq;
442 	int i, r = 0;
443 
444 	for (i = 0; i < sc->sc_nvqs; i++) {
445 		vq = &sc->sc_vqs[i];
446 		if (vq->vq_queued) {
447 			vq->vq_queued = 0;
448 			vq_sync_aring(sc, vq, BUS_DMASYNC_POSTWRITE);
449 		}
450 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
451 		membar_consumer();
452 		if (vq->vq_used_idx != vq->vq_used->idx) {
453 			if (vq->vq_done)
454 				r |= (vq->vq_done)(vq);
455 		}
456 	}
457 
458 
459 	return r;
460 }
461 
462 /*
463  * Start/stop vq interrupt.  No guarantee.
464  */
465 void
466 virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
467 {
468 	vq->vq_avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
469 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
470 	vq->vq_queued++;
471 }
472 
473 void
474 virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
475 {
476 	vq->vq_avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
477 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
478 	vq->vq_queued++;
479 }
480 
481 /*
482  * Initialize vq structure.
483  */
484 static void
485 virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq, const bool reinit)
486 {
487 	int i, j;
488 	int vq_size = vq->vq_num;
489 
490 	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
491 
492 	/* build the indirect descriptor chain */
493 	if (vq->vq_indirect != NULL) {
494 		struct vring_desc *vd;
495 
496 		for (i = 0; i < vq_size; i++) {
497 			vd = vq->vq_indirect;
498 			vd += vq->vq_maxnsegs * i;
499 			for (j = 0; j < vq->vq_maxnsegs-1; j++)
500 				vd[j].next = j + 1;
501 		}
502 	}
503 
504 	/* free slot management */
505 	SIMPLEQ_INIT(&vq->vq_freelist);
506 	for (i = 0; i < vq_size; i++) {
507 		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
508 				    &vq->vq_entries[i], qe_list);
509 		vq->vq_entries[i].qe_index = i;
510 	}
511 	if (!reinit)
512 		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
513 
514 	/* enqueue/dequeue status */
515 	vq->vq_avail_idx = 0;
516 	vq->vq_used_idx = 0;
517 	vq->vq_queued = 0;
518 	if (!reinit) {
519 		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
520 		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
521 	}
522 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
523 	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
524 	vq->vq_queued++;
525 }
526 
527 /*
528  * Allocate/free a vq.
529  */
530 int
531 virtio_alloc_vq(struct virtio_softc *sc,
532 		struct virtqueue *vq, int index, int maxsegsize, int maxnsegs,
533 		const char *name)
534 {
535 	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
536 	int rsegs, r;
537 #define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
538 				 ~(VIRTIO_PAGE_SIZE-1))
539 
540 	memset(vq, 0, sizeof(*vq));
541 
542 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
543 			  VIRTIO_CONFIG_QUEUE_SELECT, index);
544 	vq_size = bus_space_read_2(sc->sc_iot, sc->sc_ioh,
545 				   VIRTIO_CONFIG_QUEUE_SIZE);
546 	if (vq_size == 0) {
547 		aprint_error_dev(sc->sc_dev,
548 				 "virtqueue not exist, index %d for %s\n",
549 				 index, name);
550 		goto err;
551 	}
552 	/* allocsize1: descriptor table + avail ring + pad */
553 	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
554 				     + sizeof(uint16_t)*(2+vq_size));
555 	/* allocsize2: used ring + pad */
556 	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
557 				     + sizeof(struct vring_used_elem)*vq_size);
558 	/* allocsize3: indirect table */
559 	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
560 		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
561 	else
562 		allocsize3 = 0;
563 	allocsize = allocsize1 + allocsize2 + allocsize3;
564 
565 	/* alloc and map the memory */
566 	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
567 			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
568 	if (r != 0) {
569 		aprint_error_dev(sc->sc_dev,
570 				 "virtqueue %d for %s allocation failed, "
571 				 "error code %d\n", index, name, r);
572 		goto err;
573 	}
574 	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
575 			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
576 	if (r != 0) {
577 		aprint_error_dev(sc->sc_dev,
578 				 "virtqueue %d for %s map failed, "
579 				 "error code %d\n", index, name, r);
580 		goto err;
581 	}
582 	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
583 			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
584 	if (r != 0) {
585 		aprint_error_dev(sc->sc_dev,
586 				 "virtqueue %d for %s dmamap creation failed, "
587 				 "error code %d\n", index, name, r);
588 		goto err;
589 	}
590 	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
591 			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
592 	if (r != 0) {
593 		aprint_error_dev(sc->sc_dev,
594 				 "virtqueue %d for %s dmamap load failed, "
595 				 "error code %d\n", index, name, r);
596 		goto err;
597 	}
598 
599 	/* set the vq address */
600 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
601 			  VIRTIO_CONFIG_QUEUE_ADDRESS,
602 			  (vq->vq_dmamap->dm_segs[0].ds_addr
603 			   / VIRTIO_PAGE_SIZE));
604 
605 	/* remember addresses and offsets for later use */
606 	vq->vq_owner = sc;
607 	vq->vq_num = vq_size;
608 	vq->vq_index = index;
609 	vq->vq_desc = vq->vq_vaddr;
610 	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
611 	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
612 	vq->vq_usedoffset = allocsize1;
613 	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
614 	if (allocsize3 > 0) {
615 		vq->vq_indirectoffset = allocsize1 + allocsize2;
616 		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
617 					  + vq->vq_indirectoffset);
618 	}
619 	vq->vq_bytesize = allocsize;
620 	vq->vq_maxsegsize = maxsegsize;
621 	vq->vq_maxnsegs = maxnsegs;
622 
623 	/* free slot management */
624 	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
625 				     KM_NOSLEEP);
626 	if (vq->vq_entries == NULL) {
627 		r = ENOMEM;
628 		goto err;
629 	}
630 
631 	virtio_init_vq(sc, vq, false);
632 
633 	aprint_verbose_dev(sc->sc_dev,
634 			   "allocated %u byte for virtqueue %d for %s, "
635 			   "size %d\n", allocsize, index, name, vq_size);
636 	if (allocsize3 > 0)
637 		aprint_verbose_dev(sc->sc_dev,
638 				   "using %d byte (%d entries) "
639 				   "indirect descriptors\n",
640 				   allocsize3, maxnsegs * vq_size);
641 	return 0;
642 
643 err:
644 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
645 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
646 	if (vq->vq_dmamap)
647 		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
648 	if (vq->vq_vaddr)
649 		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
650 	if (vq->vq_segs[0].ds_addr)
651 		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
652 	memset(vq, 0, sizeof(*vq));
653 
654 	return -1;
655 }
656 
657 int
658 virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
659 {
660 	struct vq_entry *qe;
661 	int i = 0;
662 
663 	/* device must be already deactivated */
664 	/* confirm the vq is empty */
665 	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
666 		i++;
667 	}
668 	if (i != vq->vq_num) {
669 		printf("%s: freeing non-empty vq, index %d\n",
670 		       device_xname(sc->sc_dev), vq->vq_index);
671 		return EBUSY;
672 	}
673 
674 	/* tell device that there's no virtqueue any longer */
675 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
676 			  VIRTIO_CONFIG_QUEUE_SELECT, vq->vq_index);
677 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
678 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
679 
680 	kmem_free(vq->vq_entries, vq->vq_bytesize);
681 	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
682 	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
683 	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
684 	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
685 	mutex_destroy(&vq->vq_freelist_lock);
686 	mutex_destroy(&vq->vq_uring_lock);
687 	mutex_destroy(&vq->vq_aring_lock);
688 	memset(vq, 0, sizeof(*vq));
689 
690 	return 0;
691 }
692 
693 /*
694  * Free descriptor management.
695  */
696 static struct vq_entry *
697 vq_alloc_entry(struct virtqueue *vq)
698 {
699 	struct vq_entry *qe;
700 
701 	mutex_enter(&vq->vq_freelist_lock);
702 	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
703 		mutex_exit(&vq->vq_freelist_lock);
704 		return NULL;
705 	}
706 	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
707 	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
708 	mutex_exit(&vq->vq_freelist_lock);
709 
710 	return qe;
711 }
712 
713 static void
714 vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
715 {
716 	mutex_enter(&vq->vq_freelist_lock);
717 	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
718 	mutex_exit(&vq->vq_freelist_lock);
719 
720 	return;
721 }
722 
723 /*
724  * Enqueue several dmamaps as a single request.
725  */
726 /*
727  * Typical usage:
728  *  <queue size> number of followings are stored in arrays
729  *  - command blocks (in dmamem) should be pre-allocated and mapped
730  *  - dmamaps for command blocks should be pre-allocated and loaded
731  *  - dmamaps for payload should be pre-allocated
732  *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
733  *	if (r)		// currently 0 or EAGAIN
734  *	  return r;
735  *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
736  *	if (r) {
737  *	  virtio_enqueue_abort(sc, vq, slot);
738  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
739  *	  return r;
740  *	}
741  *	r = virtio_enqueue_reserve(sc, vq, slot,
742  *				   dmamap_payload[slot]->dm_nsegs+1);
743  *							// ^ +1 for command
744  *	if (r) {	// currently 0 or EAGAIN
745  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
746  *	  return r;					// do not call abort()
747  *	}
748  *	<setup and prepare commands>
749  *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
750  *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
751  *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
752  *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
753  *	virtio_enqueue_commit(sc, vq, slot, true);
754  */
755 
756 /*
757  * enqueue_prep: allocate a slot number
758  */
759 int
760 virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
761 {
762 	struct vq_entry *qe1;
763 
764 	KASSERT(slotp != NULL);
765 
766 	qe1 = vq_alloc_entry(vq);
767 	if (qe1 == NULL)
768 		return EAGAIN;
769 	/* next slot is not allocated yet */
770 	qe1->qe_next = -1;
771 	*slotp = qe1->qe_index;
772 
773 	return 0;
774 }
775 
776 /*
777  * enqueue_reserve: allocate remaining slots and build the descriptor chain.
778  */
779 int
780 virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
781 		       int slot, int nsegs)
782 {
783 	int indirect;
784 	struct vq_entry *qe1 = &vq->vq_entries[slot];
785 
786 	KASSERT(qe1->qe_next == -1);
787 	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
788 
789 	if ((vq->vq_indirect != NULL) &&
790 	    (nsegs >= MINSEG_INDIRECT) &&
791 	    (nsegs <= vq->vq_maxnsegs))
792 		indirect = 1;
793 	else
794 		indirect = 0;
795 	qe1->qe_indirect = indirect;
796 
797 	if (indirect) {
798 		struct vring_desc *vd;
799 		int i;
800 
801 		vd = &vq->vq_desc[qe1->qe_index];
802 		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
803 			+ vq->vq_indirectoffset;
804 		vd->addr += sizeof(struct vring_desc)
805 			* vq->vq_maxnsegs * qe1->qe_index;
806 		vd->len = sizeof(struct vring_desc) * nsegs;
807 		vd->flags = VRING_DESC_F_INDIRECT;
808 
809 		vd = vq->vq_indirect;
810 		vd += vq->vq_maxnsegs * qe1->qe_index;
811 		qe1->qe_desc_base = vd;
812 
813 		for (i = 0; i < nsegs-1; i++) {
814 			vd[i].flags = VRING_DESC_F_NEXT;
815 		}
816 		vd[i].flags = 0;
817 		qe1->qe_next = 0;
818 
819 		return 0;
820 	} else {
821 		struct vring_desc *vd;
822 		struct vq_entry *qe;
823 		int i, s;
824 
825 		vd = &vq->vq_desc[0];
826 		qe1->qe_desc_base = vd;
827 		qe1->qe_next = qe1->qe_index;
828 		s = slot;
829 		for (i = 0; i < nsegs - 1; i++) {
830 			qe = vq_alloc_entry(vq);
831 			if (qe == NULL) {
832 				vd[s].flags = 0;
833 				virtio_enqueue_abort(sc, vq, slot);
834 				return EAGAIN;
835 			}
836 			vd[s].flags = VRING_DESC_F_NEXT;
837 			vd[s].next = qe->qe_index;
838 			s = qe->qe_index;
839 		}
840 		vd[s].flags = 0;
841 
842 		return 0;
843 	}
844 }
845 
846 /*
847  * enqueue: enqueue a single dmamap.
848  */
849 int
850 virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
851 	       bus_dmamap_t dmamap, bool write)
852 {
853 	struct vq_entry *qe1 = &vq->vq_entries[slot];
854 	struct vring_desc *vd = qe1->qe_desc_base;
855 	int i;
856 	int s = qe1->qe_next;
857 
858 	KASSERT(s >= 0);
859 	KASSERT(dmamap->dm_nsegs > 0);
860 
861 	for (i = 0; i < dmamap->dm_nsegs; i++) {
862 		vd[s].addr = dmamap->dm_segs[i].ds_addr;
863 		vd[s].len = dmamap->dm_segs[i].ds_len;
864 		if (!write)
865 			vd[s].flags |= VRING_DESC_F_WRITE;
866 		s = vd[s].next;
867 	}
868 	qe1->qe_next = s;
869 
870 	return 0;
871 }
872 
873 int
874 virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
875 		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
876 		 bool write)
877 {
878 	struct vq_entry *qe1 = &vq->vq_entries[slot];
879 	struct vring_desc *vd = qe1->qe_desc_base;
880 	int s = qe1->qe_next;
881 
882 	KASSERT(s >= 0);
883 	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
884 	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
885 		(dmamap->dm_segs[0].ds_len >= start + len));
886 
887 	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
888 	vd[s].len = len;
889 	if (!write)
890 		vd[s].flags |= VRING_DESC_F_WRITE;
891 	qe1->qe_next = vd[s].next;
892 
893 	return 0;
894 }
895 
896 /*
897  * enqueue_commit: add it to the aring.
898  */
899 int
900 virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
901 		      bool notifynow)
902 {
903 	struct vq_entry *qe1;
904 
905 	if (slot < 0) {
906 		mutex_enter(&vq->vq_aring_lock);
907 		goto notify;
908 	}
909 	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
910 	qe1 = &vq->vq_entries[slot];
911 	if (qe1->qe_indirect)
912 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
913 	mutex_enter(&vq->vq_aring_lock);
914 	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
915 
916 notify:
917 	if (notifynow) {
918 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
919 		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
920 		membar_producer();
921 		vq->vq_avail->idx = vq->vq_avail_idx;
922 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
923 		membar_producer();
924 		vq->vq_queued++;
925 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
926 		membar_consumer();
927 		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
928 			bus_space_write_2(sc->sc_iot, sc->sc_ioh,
929 					  VIRTIO_CONFIG_QUEUE_NOTIFY,
930 					  vq->vq_index);
931 	}
932 	mutex_exit(&vq->vq_aring_lock);
933 
934 	return 0;
935 }
936 
937 /*
938  * enqueue_abort: rollback.
939  */
940 int
941 virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
942 {
943 	struct vq_entry *qe = &vq->vq_entries[slot];
944 	struct vring_desc *vd;
945 	int s;
946 
947 	if (qe->qe_next < 0) {
948 		vq_free_entry(vq, qe);
949 		return 0;
950 	}
951 
952 	s = slot;
953 	vd = &vq->vq_desc[0];
954 	while (vd[s].flags & VRING_DESC_F_NEXT) {
955 		s = vd[s].next;
956 		vq_free_entry(vq, qe);
957 		qe = &vq->vq_entries[s];
958 	}
959 	vq_free_entry(vq, qe);
960 	return 0;
961 }
962 
963 /*
964  * Dequeue a request.
965  */
966 /*
967  * dequeue: dequeue a request from uring; dmamap_sync for uring is
968  *	    already done in the interrupt handler.
969  */
970 int
971 virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
972 	       int *slotp, int *lenp)
973 {
974 	uint16_t slot, usedidx;
975 	struct vq_entry *qe;
976 
977 	if (vq->vq_used_idx == vq->vq_used->idx)
978 		return ENOENT;
979 	mutex_enter(&vq->vq_uring_lock);
980 	usedidx = vq->vq_used_idx++;
981 	mutex_exit(&vq->vq_uring_lock);
982 	usedidx %= vq->vq_num;
983 	slot = vq->vq_used->ring[usedidx].id;
984 	qe = &vq->vq_entries[slot];
985 
986 	if (qe->qe_indirect)
987 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
988 
989 	if (slotp)
990 		*slotp = slot;
991 	if (lenp)
992 		*lenp = vq->vq_used->ring[usedidx].len;
993 
994 	return 0;
995 }
996 
997 /*
998  * dequeue_commit: complete dequeue; the slot is recycled for future use.
999  *                 if you forget to call this the slot will be leaked.
1000  */
1001 int
1002 virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1003 {
1004 	struct vq_entry *qe = &vq->vq_entries[slot];
1005 	struct vring_desc *vd = &vq->vq_desc[0];
1006 	int s = slot;
1007 
1008 	while (vd[s].flags & VRING_DESC_F_NEXT) {
1009 		s = vd[s].next;
1010 		vq_free_entry(vq, qe);
1011 		qe = &vq->vq_entries[s];
1012 	}
1013 	vq_free_entry(vq, qe);
1014 
1015 	return 0;
1016 }
1017