xref: /netbsd-src/sys/dev/pci/virtio.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: virtio.c,v 1.5 2014/03/29 19:28:25 christos 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 1.5 2014/03/29 19:28:25 christos 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 	char intrbuf[PCI_INTRSTR_LEN];
118 
119 	revision = PCI_REVISION(pa->pa_class);
120 	if (revision != 0) {
121 		aprint_normal(": unknown revision 0x%02x; giving up\n",
122 			      revision);
123 		return;
124 	}
125 	aprint_normal("\n");
126 	aprint_naive("\n");
127 
128 	/* subsystem ID shows what I am */
129 	id = pci_conf_read(pc, tag, PCI_SUBSYS_ID_REG);
130 	aprint_normal_dev(self, "Virtio %s Device (rev. 0x%02x)\n",
131 			  (PCI_PRODUCT(id) < NDEVNAMES?
132 			   virtio_device_name[PCI_PRODUCT(id)] : "Unknown"),
133 			  revision);
134 
135 	sc->sc_dev = self;
136 	sc->sc_pc = pc;
137 	sc->sc_tag = tag;
138 	sc->sc_iot = pa->pa_iot;
139 	sc->sc_dmat = pa->pa_dmat;
140 	sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
141 
142 	if (pci_mapreg_map(pa, PCI_MAPREG_START, PCI_MAPREG_TYPE_IO, 0,
143 			   &sc->sc_iot, &sc->sc_ioh, NULL, &sc->sc_iosize)) {
144 		aprint_error_dev(self, "can't map i/o space\n");
145 		return;
146 	}
147 
148 	virtio_device_reset(sc);
149 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
150 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
151 
152 	/* XXX: use softc as aux... */
153 	sc->sc_childdevid = PCI_PRODUCT(id);
154 	sc->sc_child = NULL;
155 	config_found(self, sc, NULL);
156 	if (sc->sc_child == NULL) {
157 		aprint_error_dev(self,
158 				 "no matching child driver; not configured\n");
159 		return;
160 	}
161 	if (sc->sc_child == (void*)1) { /* this shows error */
162 		aprint_error_dev(self,
163 				 "virtio configuration failed\n");
164 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
165 		return;
166 	}
167 
168 	if (pci_intr_map(pa, &ih)) {
169 		aprint_error_dev(self, "couldn't map interrupt\n");
170 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
171 		return;
172 	}
173 	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
174 	sc->sc_ih = pci_intr_establish(pc, ih, sc->sc_ipl, virtio_intr, sc);
175 	if (sc->sc_ih == NULL) {
176 		aprint_error_dev(self, "couldn't establish interrupt");
177 		if (intrstr != NULL)
178 			aprint_error(" at %s", intrstr);
179 		aprint_error("\n");
180 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
181 		return;
182 	}
183 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
184 
185 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
186 
187 	return;
188 }
189 
190 static int
191 virtio_detach(device_t self, int flags)
192 {
193 	struct virtio_softc *sc = device_private(self);
194 	int r;
195 
196 	if (sc->sc_child != 0 && sc->sc_child != (void*)1) {
197 		r = config_detach(sc->sc_child, flags);
198 		if (r)
199 			return r;
200 	}
201 	KASSERT(sc->sc_child == 0 || sc->sc_child == (void*)1);
202 	KASSERT(sc->sc_vqs == 0);
203 	if (sc->sc_ih != NULL) {
204 		pci_intr_disestablish(sc->sc_pc, sc->sc_ih);
205 		sc->sc_ih = NULL;
206 	}
207 	if (sc->sc_iosize)
208 		bus_space_unmap(sc->sc_iot, sc->sc_ioh, sc->sc_iosize);
209 	sc->sc_iosize = 0;
210 
211 	return 0;
212 }
213 
214 /*
215  * Reset the device.
216  */
217 /*
218  * To reset the device to a known state, do following:
219  *	virtio_reset(sc);	     // this will stop the device activity
220  *	<dequeue finished requests>; // virtio_dequeue() still can be called
221  *	<revoke pending requests in the vqs if any>;
222  *	virtio_reinit_begin(sc);     // dequeue prohibitted
223  *	newfeatures = virtio_negotiate_features(sc, requestedfeatures);
224  *	<some other initialization>;
225  *	virtio_reinit_end(sc);	     // device activated; enqueue allowed
226  * Once attached, feature negotiation can only be allowed after virtio_reset.
227  */
228 void
229 virtio_reset(struct virtio_softc *sc)
230 {
231 	virtio_device_reset(sc);
232 }
233 
234 void
235 virtio_reinit_start(struct virtio_softc *sc)
236 {
237 	int i;
238 
239 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
240 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
241 	for (i = 0; i < sc->sc_nvqs; i++) {
242 		int n;
243 		struct virtqueue *vq = &sc->sc_vqs[i];
244 		bus_space_write_2(sc->sc_iot, sc->sc_ioh,
245 				  VIRTIO_CONFIG_QUEUE_SELECT,
246 				  vq->vq_index);
247 		n = bus_space_read_2(sc->sc_iot, sc->sc_ioh,
248 				     VIRTIO_CONFIG_QUEUE_SIZE);
249 		if (n == 0)	/* vq disappeared */
250 			continue;
251 		if (n != vq->vq_num) {
252 			panic("%s: virtqueue size changed, vq index %d\n",
253 			      device_xname(sc->sc_dev),
254 			      vq->vq_index);
255 		}
256 		virtio_init_vq(sc, vq, true);
257 		bus_space_write_4(sc->sc_iot, sc->sc_ioh,
258 				  VIRTIO_CONFIG_QUEUE_ADDRESS,
259 				  (vq->vq_dmamap->dm_segs[0].ds_addr
260 				   / VIRTIO_PAGE_SIZE));
261 	}
262 }
263 
264 void
265 virtio_reinit_end(struct virtio_softc *sc)
266 {
267 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
268 }
269 
270 /*
271  * Feature negotiation.
272  */
273 uint32_t
274 virtio_negotiate_features(struct virtio_softc *sc, uint32_t guest_features)
275 {
276 	uint32_t r;
277 
278 	if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) &&
279 	    !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */
280 		guest_features |= VIRTIO_F_RING_INDIRECT_DESC;
281 	r = bus_space_read_4(sc->sc_iot, sc->sc_ioh,
282 			     VIRTIO_CONFIG_DEVICE_FEATURES);
283 	r &= guest_features;
284 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
285 			  VIRTIO_CONFIG_GUEST_FEATURES, r);
286 	sc->sc_features = r;
287 	if (r & VIRTIO_F_RING_INDIRECT_DESC)
288 		sc->sc_indirect = true;
289 	else
290 		sc->sc_indirect = false;
291 
292 	return r;
293 }
294 
295 /*
296  * Device configuration registers.
297  */
298 uint8_t
299 virtio_read_device_config_1(struct virtio_softc *sc, int index)
300 {
301 	return bus_space_read_1(sc->sc_iot, sc->sc_ioh,
302 				sc->sc_config_offset + index);
303 }
304 
305 uint16_t
306 virtio_read_device_config_2(struct virtio_softc *sc, int index)
307 {
308 	return bus_space_read_2(sc->sc_iot, sc->sc_ioh,
309 				sc->sc_config_offset + index);
310 }
311 
312 uint32_t
313 virtio_read_device_config_4(struct virtio_softc *sc, int index)
314 {
315 	return bus_space_read_4(sc->sc_iot, sc->sc_ioh,
316 				sc->sc_config_offset + index);
317 }
318 
319 uint64_t
320 virtio_read_device_config_8(struct virtio_softc *sc, int index)
321 {
322 	uint64_t r;
323 
324 	r = bus_space_read_4(sc->sc_iot, sc->sc_ioh,
325 			     sc->sc_config_offset + index + sizeof(uint32_t));
326 	r <<= 32;
327 	r += bus_space_read_4(sc->sc_iot, sc->sc_ioh,
328 			      sc->sc_config_offset + index);
329 	return r;
330 }
331 
332 void
333 virtio_write_device_config_1(struct virtio_softc *sc,
334 			     int index, uint8_t value)
335 {
336 	bus_space_write_1(sc->sc_iot, sc->sc_ioh,
337 			  sc->sc_config_offset + index, value);
338 }
339 
340 void
341 virtio_write_device_config_2(struct virtio_softc *sc,
342 			     int index, uint16_t value)
343 {
344 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
345 			  sc->sc_config_offset + index, value);
346 }
347 
348 void
349 virtio_write_device_config_4(struct virtio_softc *sc,
350 			     int index, uint32_t value)
351 {
352 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
353 			  sc->sc_config_offset + index, value);
354 }
355 
356 void
357 virtio_write_device_config_8(struct virtio_softc *sc,
358 			     int index, uint64_t value)
359 {
360 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
361 			  sc->sc_config_offset + index,
362 			  value & 0xffffffff);
363 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
364 			  sc->sc_config_offset + index + sizeof(uint32_t),
365 			  value >> 32);
366 }
367 
368 /*
369  * Interrupt handler.
370  */
371 static int
372 virtio_intr(void *arg)
373 {
374 	struct virtio_softc *sc = arg;
375 	int isr, r = 0;
376 
377 	/* check and ack the interrupt */
378 	isr = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
379 			       VIRTIO_CONFIG_ISR_STATUS);
380 	if (isr == 0)
381 		return 0;
382 	if ((isr & VIRTIO_CONFIG_ISR_CONFIG_CHANGE) &&
383 	    (sc->sc_config_change != NULL))
384 		r = (sc->sc_config_change)(sc);
385 	if (sc->sc_intrhand != NULL)
386 		r |= (sc->sc_intrhand)(sc);
387 
388 	return r;
389 }
390 
391 /*
392  * dmamap sync operations for a virtqueue.
393  */
394 static inline void
395 vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops)
396 {
397 	/* availoffset == sizeof(vring_desc)*vq_num */
398 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset,
399 			ops);
400 }
401 
402 static inline void
403 vq_sync_aring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
404 {
405 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
406 			vq->vq_availoffset,
407 			offsetof(struct vring_avail, ring)
408 			 + vq->vq_num * sizeof(uint16_t),
409 			ops);
410 }
411 
412 static inline void
413 vq_sync_uring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
414 {
415 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
416 			vq->vq_usedoffset,
417 			offsetof(struct vring_used, ring)
418 			 + vq->vq_num * sizeof(struct vring_used_elem),
419 			ops);
420 }
421 
422 static inline void
423 vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot,
424 		     int ops)
425 {
426 	int offset = vq->vq_indirectoffset
427 		      + sizeof(struct vring_desc) * vq->vq_maxnsegs * slot;
428 
429 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
430 			offset, sizeof(struct vring_desc) * vq->vq_maxnsegs,
431 			ops);
432 }
433 
434 /*
435  * Can be used as sc_intrhand.
436  */
437 /*
438  * Scan vq, bus_dmamap_sync for the vqs (not for the payload),
439  * and calls (*vq_done)() if some entries are consumed.
440  */
441 int
442 virtio_vq_intr(struct virtio_softc *sc)
443 {
444 	struct virtqueue *vq;
445 	int i, r = 0;
446 
447 	for (i = 0; i < sc->sc_nvqs; i++) {
448 		vq = &sc->sc_vqs[i];
449 		if (vq->vq_queued) {
450 			vq->vq_queued = 0;
451 			vq_sync_aring(sc, vq, BUS_DMASYNC_POSTWRITE);
452 		}
453 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
454 		membar_consumer();
455 		if (vq->vq_used_idx != vq->vq_used->idx) {
456 			if (vq->vq_done)
457 				r |= (vq->vq_done)(vq);
458 		}
459 	}
460 
461 
462 	return r;
463 }
464 
465 /*
466  * Start/stop vq interrupt.  No guarantee.
467  */
468 void
469 virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
470 {
471 	vq->vq_avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
472 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
473 	vq->vq_queued++;
474 }
475 
476 void
477 virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
478 {
479 	vq->vq_avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
480 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
481 	vq->vq_queued++;
482 }
483 
484 /*
485  * Initialize vq structure.
486  */
487 static void
488 virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq, const bool reinit)
489 {
490 	int i, j;
491 	int vq_size = vq->vq_num;
492 
493 	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
494 
495 	/* build the indirect descriptor chain */
496 	if (vq->vq_indirect != NULL) {
497 		struct vring_desc *vd;
498 
499 		for (i = 0; i < vq_size; i++) {
500 			vd = vq->vq_indirect;
501 			vd += vq->vq_maxnsegs * i;
502 			for (j = 0; j < vq->vq_maxnsegs-1; j++)
503 				vd[j].next = j + 1;
504 		}
505 	}
506 
507 	/* free slot management */
508 	SIMPLEQ_INIT(&vq->vq_freelist);
509 	for (i = 0; i < vq_size; i++) {
510 		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
511 				    &vq->vq_entries[i], qe_list);
512 		vq->vq_entries[i].qe_index = i;
513 	}
514 	if (!reinit)
515 		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
516 
517 	/* enqueue/dequeue status */
518 	vq->vq_avail_idx = 0;
519 	vq->vq_used_idx = 0;
520 	vq->vq_queued = 0;
521 	if (!reinit) {
522 		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
523 		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
524 	}
525 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
526 	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
527 	vq->vq_queued++;
528 }
529 
530 /*
531  * Allocate/free a vq.
532  */
533 int
534 virtio_alloc_vq(struct virtio_softc *sc,
535 		struct virtqueue *vq, int index, int maxsegsize, int maxnsegs,
536 		const char *name)
537 {
538 	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
539 	int rsegs, r;
540 #define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
541 				 ~(VIRTIO_PAGE_SIZE-1))
542 
543 	memset(vq, 0, sizeof(*vq));
544 
545 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
546 			  VIRTIO_CONFIG_QUEUE_SELECT, index);
547 	vq_size = bus_space_read_2(sc->sc_iot, sc->sc_ioh,
548 				   VIRTIO_CONFIG_QUEUE_SIZE);
549 	if (vq_size == 0) {
550 		aprint_error_dev(sc->sc_dev,
551 				 "virtqueue not exist, index %d for %s\n",
552 				 index, name);
553 		goto err;
554 	}
555 	/* allocsize1: descriptor table + avail ring + pad */
556 	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
557 				     + sizeof(uint16_t)*(2+vq_size));
558 	/* allocsize2: used ring + pad */
559 	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
560 				     + sizeof(struct vring_used_elem)*vq_size);
561 	/* allocsize3: indirect table */
562 	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
563 		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
564 	else
565 		allocsize3 = 0;
566 	allocsize = allocsize1 + allocsize2 + allocsize3;
567 
568 	/* alloc and map the memory */
569 	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
570 			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
571 	if (r != 0) {
572 		aprint_error_dev(sc->sc_dev,
573 				 "virtqueue %d for %s allocation failed, "
574 				 "error code %d\n", index, name, r);
575 		goto err;
576 	}
577 	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
578 			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
579 	if (r != 0) {
580 		aprint_error_dev(sc->sc_dev,
581 				 "virtqueue %d for %s map failed, "
582 				 "error code %d\n", index, name, r);
583 		goto err;
584 	}
585 	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
586 			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
587 	if (r != 0) {
588 		aprint_error_dev(sc->sc_dev,
589 				 "virtqueue %d for %s dmamap creation failed, "
590 				 "error code %d\n", index, name, r);
591 		goto err;
592 	}
593 	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
594 			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
595 	if (r != 0) {
596 		aprint_error_dev(sc->sc_dev,
597 				 "virtqueue %d for %s dmamap load failed, "
598 				 "error code %d\n", index, name, r);
599 		goto err;
600 	}
601 
602 	/* set the vq address */
603 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
604 			  VIRTIO_CONFIG_QUEUE_ADDRESS,
605 			  (vq->vq_dmamap->dm_segs[0].ds_addr
606 			   / VIRTIO_PAGE_SIZE));
607 
608 	/* remember addresses and offsets for later use */
609 	vq->vq_owner = sc;
610 	vq->vq_num = vq_size;
611 	vq->vq_index = index;
612 	vq->vq_desc = vq->vq_vaddr;
613 	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
614 	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
615 	vq->vq_usedoffset = allocsize1;
616 	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
617 	if (allocsize3 > 0) {
618 		vq->vq_indirectoffset = allocsize1 + allocsize2;
619 		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
620 					  + vq->vq_indirectoffset);
621 	}
622 	vq->vq_bytesize = allocsize;
623 	vq->vq_maxsegsize = maxsegsize;
624 	vq->vq_maxnsegs = maxnsegs;
625 
626 	/* free slot management */
627 	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
628 				     KM_NOSLEEP);
629 	if (vq->vq_entries == NULL) {
630 		r = ENOMEM;
631 		goto err;
632 	}
633 
634 	virtio_init_vq(sc, vq, false);
635 
636 	aprint_verbose_dev(sc->sc_dev,
637 			   "allocated %u byte for virtqueue %d for %s, "
638 			   "size %d\n", allocsize, index, name, vq_size);
639 	if (allocsize3 > 0)
640 		aprint_verbose_dev(sc->sc_dev,
641 				   "using %d byte (%d entries) "
642 				   "indirect descriptors\n",
643 				   allocsize3, maxnsegs * vq_size);
644 	return 0;
645 
646 err:
647 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
648 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
649 	if (vq->vq_dmamap)
650 		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
651 	if (vq->vq_vaddr)
652 		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
653 	if (vq->vq_segs[0].ds_addr)
654 		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
655 	memset(vq, 0, sizeof(*vq));
656 
657 	return -1;
658 }
659 
660 int
661 virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
662 {
663 	struct vq_entry *qe;
664 	int i = 0;
665 
666 	/* device must be already deactivated */
667 	/* confirm the vq is empty */
668 	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
669 		i++;
670 	}
671 	if (i != vq->vq_num) {
672 		printf("%s: freeing non-empty vq, index %d\n",
673 		       device_xname(sc->sc_dev), vq->vq_index);
674 		return EBUSY;
675 	}
676 
677 	/* tell device that there's no virtqueue any longer */
678 	bus_space_write_2(sc->sc_iot, sc->sc_ioh,
679 			  VIRTIO_CONFIG_QUEUE_SELECT, vq->vq_index);
680 	bus_space_write_4(sc->sc_iot, sc->sc_ioh,
681 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
682 
683 	kmem_free(vq->vq_entries, vq->vq_bytesize);
684 	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
685 	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
686 	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
687 	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
688 	mutex_destroy(&vq->vq_freelist_lock);
689 	mutex_destroy(&vq->vq_uring_lock);
690 	mutex_destroy(&vq->vq_aring_lock);
691 	memset(vq, 0, sizeof(*vq));
692 
693 	return 0;
694 }
695 
696 /*
697  * Free descriptor management.
698  */
699 static struct vq_entry *
700 vq_alloc_entry(struct virtqueue *vq)
701 {
702 	struct vq_entry *qe;
703 
704 	mutex_enter(&vq->vq_freelist_lock);
705 	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
706 		mutex_exit(&vq->vq_freelist_lock);
707 		return NULL;
708 	}
709 	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
710 	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
711 	mutex_exit(&vq->vq_freelist_lock);
712 
713 	return qe;
714 }
715 
716 static void
717 vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
718 {
719 	mutex_enter(&vq->vq_freelist_lock);
720 	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
721 	mutex_exit(&vq->vq_freelist_lock);
722 
723 	return;
724 }
725 
726 /*
727  * Enqueue several dmamaps as a single request.
728  */
729 /*
730  * Typical usage:
731  *  <queue size> number of followings are stored in arrays
732  *  - command blocks (in dmamem) should be pre-allocated and mapped
733  *  - dmamaps for command blocks should be pre-allocated and loaded
734  *  - dmamaps for payload should be pre-allocated
735  *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
736  *	if (r)		// currently 0 or EAGAIN
737  *	  return r;
738  *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
739  *	if (r) {
740  *	  virtio_enqueue_abort(sc, vq, slot);
741  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
742  *	  return r;
743  *	}
744  *	r = virtio_enqueue_reserve(sc, vq, slot,
745  *				   dmamap_payload[slot]->dm_nsegs+1);
746  *							// ^ +1 for command
747  *	if (r) {	// currently 0 or EAGAIN
748  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
749  *	  return r;					// do not call abort()
750  *	}
751  *	<setup and prepare commands>
752  *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
753  *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
754  *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
755  *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
756  *	virtio_enqueue_commit(sc, vq, slot, true);
757  */
758 
759 /*
760  * enqueue_prep: allocate a slot number
761  */
762 int
763 virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
764 {
765 	struct vq_entry *qe1;
766 
767 	KASSERT(slotp != NULL);
768 
769 	qe1 = vq_alloc_entry(vq);
770 	if (qe1 == NULL)
771 		return EAGAIN;
772 	/* next slot is not allocated yet */
773 	qe1->qe_next = -1;
774 	*slotp = qe1->qe_index;
775 
776 	return 0;
777 }
778 
779 /*
780  * enqueue_reserve: allocate remaining slots and build the descriptor chain.
781  */
782 int
783 virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
784 		       int slot, int nsegs)
785 {
786 	int indirect;
787 	struct vq_entry *qe1 = &vq->vq_entries[slot];
788 
789 	KASSERT(qe1->qe_next == -1);
790 	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
791 
792 	if ((vq->vq_indirect != NULL) &&
793 	    (nsegs >= MINSEG_INDIRECT) &&
794 	    (nsegs <= vq->vq_maxnsegs))
795 		indirect = 1;
796 	else
797 		indirect = 0;
798 	qe1->qe_indirect = indirect;
799 
800 	if (indirect) {
801 		struct vring_desc *vd;
802 		int i;
803 
804 		vd = &vq->vq_desc[qe1->qe_index];
805 		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
806 			+ vq->vq_indirectoffset;
807 		vd->addr += sizeof(struct vring_desc)
808 			* vq->vq_maxnsegs * qe1->qe_index;
809 		vd->len = sizeof(struct vring_desc) * nsegs;
810 		vd->flags = VRING_DESC_F_INDIRECT;
811 
812 		vd = vq->vq_indirect;
813 		vd += vq->vq_maxnsegs * qe1->qe_index;
814 		qe1->qe_desc_base = vd;
815 
816 		for (i = 0; i < nsegs-1; i++) {
817 			vd[i].flags = VRING_DESC_F_NEXT;
818 		}
819 		vd[i].flags = 0;
820 		qe1->qe_next = 0;
821 
822 		return 0;
823 	} else {
824 		struct vring_desc *vd;
825 		struct vq_entry *qe;
826 		int i, s;
827 
828 		vd = &vq->vq_desc[0];
829 		qe1->qe_desc_base = vd;
830 		qe1->qe_next = qe1->qe_index;
831 		s = slot;
832 		for (i = 0; i < nsegs - 1; i++) {
833 			qe = vq_alloc_entry(vq);
834 			if (qe == NULL) {
835 				vd[s].flags = 0;
836 				virtio_enqueue_abort(sc, vq, slot);
837 				return EAGAIN;
838 			}
839 			vd[s].flags = VRING_DESC_F_NEXT;
840 			vd[s].next = qe->qe_index;
841 			s = qe->qe_index;
842 		}
843 		vd[s].flags = 0;
844 
845 		return 0;
846 	}
847 }
848 
849 /*
850  * enqueue: enqueue a single dmamap.
851  */
852 int
853 virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
854 	       bus_dmamap_t dmamap, bool write)
855 {
856 	struct vq_entry *qe1 = &vq->vq_entries[slot];
857 	struct vring_desc *vd = qe1->qe_desc_base;
858 	int i;
859 	int s = qe1->qe_next;
860 
861 	KASSERT(s >= 0);
862 	KASSERT(dmamap->dm_nsegs > 0);
863 
864 	for (i = 0; i < dmamap->dm_nsegs; i++) {
865 		vd[s].addr = dmamap->dm_segs[i].ds_addr;
866 		vd[s].len = dmamap->dm_segs[i].ds_len;
867 		if (!write)
868 			vd[s].flags |= VRING_DESC_F_WRITE;
869 		s = vd[s].next;
870 	}
871 	qe1->qe_next = s;
872 
873 	return 0;
874 }
875 
876 int
877 virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
878 		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
879 		 bool write)
880 {
881 	struct vq_entry *qe1 = &vq->vq_entries[slot];
882 	struct vring_desc *vd = qe1->qe_desc_base;
883 	int s = qe1->qe_next;
884 
885 	KASSERT(s >= 0);
886 	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
887 	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
888 		(dmamap->dm_segs[0].ds_len >= start + len));
889 
890 	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
891 	vd[s].len = len;
892 	if (!write)
893 		vd[s].flags |= VRING_DESC_F_WRITE;
894 	qe1->qe_next = vd[s].next;
895 
896 	return 0;
897 }
898 
899 /*
900  * enqueue_commit: add it to the aring.
901  */
902 int
903 virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
904 		      bool notifynow)
905 {
906 	struct vq_entry *qe1;
907 
908 	if (slot < 0) {
909 		mutex_enter(&vq->vq_aring_lock);
910 		goto notify;
911 	}
912 	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
913 	qe1 = &vq->vq_entries[slot];
914 	if (qe1->qe_indirect)
915 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
916 	mutex_enter(&vq->vq_aring_lock);
917 	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
918 
919 notify:
920 	if (notifynow) {
921 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
922 		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
923 		membar_producer();
924 		vq->vq_avail->idx = vq->vq_avail_idx;
925 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
926 		membar_producer();
927 		vq->vq_queued++;
928 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
929 		membar_consumer();
930 		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
931 			bus_space_write_2(sc->sc_iot, sc->sc_ioh,
932 					  VIRTIO_CONFIG_QUEUE_NOTIFY,
933 					  vq->vq_index);
934 	}
935 	mutex_exit(&vq->vq_aring_lock);
936 
937 	return 0;
938 }
939 
940 /*
941  * enqueue_abort: rollback.
942  */
943 int
944 virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
945 {
946 	struct vq_entry *qe = &vq->vq_entries[slot];
947 	struct vring_desc *vd;
948 	int s;
949 
950 	if (qe->qe_next < 0) {
951 		vq_free_entry(vq, qe);
952 		return 0;
953 	}
954 
955 	s = slot;
956 	vd = &vq->vq_desc[0];
957 	while (vd[s].flags & VRING_DESC_F_NEXT) {
958 		s = vd[s].next;
959 		vq_free_entry(vq, qe);
960 		qe = &vq->vq_entries[s];
961 	}
962 	vq_free_entry(vq, qe);
963 	return 0;
964 }
965 
966 /*
967  * Dequeue a request.
968  */
969 /*
970  * dequeue: dequeue a request from uring; dmamap_sync for uring is
971  *	    already done in the interrupt handler.
972  */
973 int
974 virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
975 	       int *slotp, int *lenp)
976 {
977 	uint16_t slot, usedidx;
978 	struct vq_entry *qe;
979 
980 	if (vq->vq_used_idx == vq->vq_used->idx)
981 		return ENOENT;
982 	mutex_enter(&vq->vq_uring_lock);
983 	usedidx = vq->vq_used_idx++;
984 	mutex_exit(&vq->vq_uring_lock);
985 	usedidx %= vq->vq_num;
986 	slot = vq->vq_used->ring[usedidx].id;
987 	qe = &vq->vq_entries[slot];
988 
989 	if (qe->qe_indirect)
990 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
991 
992 	if (slotp)
993 		*slotp = slot;
994 	if (lenp)
995 		*lenp = vq->vq_used->ring[usedidx].len;
996 
997 	return 0;
998 }
999 
1000 /*
1001  * dequeue_commit: complete dequeue; the slot is recycled for future use.
1002  *                 if you forget to call this the slot will be leaked.
1003  */
1004 int
1005 virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
1006 {
1007 	struct vq_entry *qe = &vq->vq_entries[slot];
1008 	struct vring_desc *vd = &vq->vq_desc[0];
1009 	int s = slot;
1010 
1011 	while (vd[s].flags & VRING_DESC_F_NEXT) {
1012 		s = vd[s].next;
1013 		vq_free_entry(vq, qe);
1014 		qe = &vq->vq_entries[s];
1015 	}
1016 	vq_free_entry(vq, qe);
1017 
1018 	return 0;
1019 }
1020