xref: /netbsd-src/sys/dev/pci/if_vioif.c (revision e6c7e151de239c49d2e38720a061ed9d1fa99309)
1 /*	$NetBSD: if_vioif.c,v 1.52 2020/01/30 14:02:14 thorpej 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: if_vioif.c,v 1.52 2020/01/30 14:02:14 thorpej Exp $");
30 
31 #ifdef _KERNEL_OPT
32 #include "opt_net_mpsafe.h"
33 #endif
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/bus.h>
39 #include <sys/condvar.h>
40 #include <sys/device.h>
41 #include <sys/intr.h>
42 #include <sys/kmem.h>
43 #include <sys/mbuf.h>
44 #include <sys/mutex.h>
45 #include <sys/sockio.h>
46 #include <sys/cpu.h>
47 #include <sys/module.h>
48 #include <sys/pcq.h>
49 
50 #include <dev/pci/virtioreg.h>
51 #include <dev/pci/virtiovar.h>
52 
53 #include <net/if.h>
54 #include <net/if_media.h>
55 #include <net/if_ether.h>
56 
57 #include <net/bpf.h>
58 
59 #include "ioconf.h"
60 
61 #ifdef NET_MPSAFE
62 #define VIOIF_MPSAFE	1
63 #define VIOIF_MULTIQ	1
64 #endif
65 
66 #ifdef SOFTINT_INTR
67 #define VIOIF_SOFTINT_INTR	1
68 #endif
69 
70 /*
71  * if_vioifreg.h:
72  */
73 /* Configuration registers */
74 #define VIRTIO_NET_CONFIG_MAC		0 /* 8bit x 6byte */
75 #define VIRTIO_NET_CONFIG_STATUS	6 /* 16bit */
76 #define VIRTIO_NET_CONFIG_MAX_VQ_PAIRS	8 /* 16bit */
77 
78 /* Feature bits */
79 #define VIRTIO_NET_F_CSUM		__BIT(0)
80 #define VIRTIO_NET_F_GUEST_CSUM		__BIT(1)
81 #define VIRTIO_NET_F_MAC		__BIT(5)
82 #define VIRTIO_NET_F_GSO		__BIT(6)
83 #define VIRTIO_NET_F_GUEST_TSO4		__BIT(7)
84 #define VIRTIO_NET_F_GUEST_TSO6		__BIT(8)
85 #define VIRTIO_NET_F_GUEST_ECN		__BIT(9)
86 #define VIRTIO_NET_F_GUEST_UFO		__BIT(10)
87 #define VIRTIO_NET_F_HOST_TSO4		__BIT(11)
88 #define VIRTIO_NET_F_HOST_TSO6		__BIT(12)
89 #define VIRTIO_NET_F_HOST_ECN		__BIT(13)
90 #define VIRTIO_NET_F_HOST_UFO		__BIT(14)
91 #define VIRTIO_NET_F_MRG_RXBUF		__BIT(15)
92 #define VIRTIO_NET_F_STATUS		__BIT(16)
93 #define VIRTIO_NET_F_CTRL_VQ		__BIT(17)
94 #define VIRTIO_NET_F_CTRL_RX		__BIT(18)
95 #define VIRTIO_NET_F_CTRL_VLAN		__BIT(19)
96 #define VIRTIO_NET_F_CTRL_RX_EXTRA	__BIT(20)
97 #define VIRTIO_NET_F_GUEST_ANNOUNCE	__BIT(21)
98 #define VIRTIO_NET_F_MQ			__BIT(22)
99 
100 #define VIRTIO_NET_FLAG_BITS \
101 	VIRTIO_COMMON_FLAG_BITS \
102 	"\x17""MQ" \
103 	"\x16""GUEST_ANNOUNCE" \
104 	"\x15""CTRL_RX_EXTRA" \
105 	"\x14""CTRL_VLAN" \
106 	"\x13""CTRL_RX" \
107 	"\x12""CTRL_VQ" \
108 	"\x11""STATUS" \
109 	"\x10""MRG_RXBUF" \
110 	"\x0f""HOST_UFO" \
111 	"\x0e""HOST_ECN" \
112 	"\x0d""HOST_TSO6" \
113 	"\x0c""HOST_TSO4" \
114 	"\x0b""GUEST_UFO" \
115 	"\x0a""GUEST_ECN" \
116 	"\x09""GUEST_TSO6" \
117 	"\x08""GUEST_TSO4" \
118 	"\x07""GSO" \
119 	"\x06""MAC" \
120 	"\x02""GUEST_CSUM" \
121 	"\x01""CSUM"
122 
123 /* Status */
124 #define VIRTIO_NET_S_LINK_UP	1
125 
126 /* Packet header structure */
127 struct virtio_net_hdr {
128 	uint8_t		flags;
129 	uint8_t		gso_type;
130 	uint16_t	hdr_len;
131 	uint16_t	gso_size;
132 	uint16_t	csum_start;
133 	uint16_t	csum_offset;
134 #if 0
135 	uint16_t	num_buffers; /* if VIRTIO_NET_F_MRG_RXBUF enabled */
136 #endif
137 } __packed;
138 
139 #define VIRTIO_NET_HDR_F_NEEDS_CSUM	1 /* flags */
140 #define VIRTIO_NET_HDR_GSO_NONE		0 /* gso_type */
141 #define VIRTIO_NET_HDR_GSO_TCPV4	1 /* gso_type */
142 #define VIRTIO_NET_HDR_GSO_UDP		3 /* gso_type */
143 #define VIRTIO_NET_HDR_GSO_TCPV6	4 /* gso_type */
144 #define VIRTIO_NET_HDR_GSO_ECN		0x80 /* gso_type, |'ed */
145 
146 #define VIRTIO_NET_MAX_GSO_LEN		(65536+ETHER_HDR_LEN)
147 
148 /* Control virtqueue */
149 struct virtio_net_ctrl_cmd {
150 	uint8_t	class;
151 	uint8_t	command;
152 } __packed;
153 #define VIRTIO_NET_CTRL_RX		0
154 # define VIRTIO_NET_CTRL_RX_PROMISC	0
155 # define VIRTIO_NET_CTRL_RX_ALLMULTI	1
156 
157 #define VIRTIO_NET_CTRL_MAC		1
158 # define VIRTIO_NET_CTRL_MAC_TABLE_SET	0
159 
160 #define VIRTIO_NET_CTRL_VLAN		2
161 # define VIRTIO_NET_CTRL_VLAN_ADD	0
162 # define VIRTIO_NET_CTRL_VLAN_DEL	1
163 
164 #define VIRTIO_NET_CTRL_MQ			4
165 # define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET	0
166 # define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN	1
167 # define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX	0x8000
168 
169 struct virtio_net_ctrl_status {
170 	uint8_t	ack;
171 } __packed;
172 #define VIRTIO_NET_OK			0
173 #define VIRTIO_NET_ERR			1
174 
175 struct virtio_net_ctrl_rx {
176 	uint8_t	onoff;
177 } __packed;
178 
179 struct virtio_net_ctrl_mac_tbl {
180 	uint32_t nentries;
181 	uint8_t macs[][ETHER_ADDR_LEN];
182 } __packed;
183 
184 struct virtio_net_ctrl_vlan {
185 	uint16_t id;
186 } __packed;
187 
188 struct virtio_net_ctrl_mq {
189 	uint16_t virtqueue_pairs;
190 } __packed;
191 
192 struct vioif_ctrl_cmdspec {
193 	bus_dmamap_t	dmamap;
194 	void		*buf;
195 	bus_size_t	bufsize;
196 };
197 
198 /*
199  * if_vioifvar.h:
200  */
201 
202 /*
203  * Locking notes:
204  * + a field in vioif_txqueue is protected by txq_lock (a spin mutex), and
205  *   a filds in vioif_rxqueue is protected by rxq_lock (a spin mutex).
206  *      - more than one lock cannot be held at onece
207  * + ctrlq_inuse is protected by ctrlq_wait_lock.
208  *      - other fields in vioif_ctrlqueue are protected by ctrlq_inuse
209  *      - txq_lock or rxq_lock cannot be held along with ctrlq_wait_lock
210  */
211 
212 struct vioif_txqueue {
213 	kmutex_t		*txq_lock;	/* lock for tx operations */
214 
215 	struct virtqueue	*txq_vq;
216 	bool			txq_stopping;
217 	bool			txq_link_active;
218 	pcq_t			*txq_intrq;
219 
220 	struct virtio_net_hdr	*txq_hdrs;
221 	bus_dmamap_t		*txq_hdr_dmamaps;
222 
223 	struct mbuf		**txq_mbufs;
224 	bus_dmamap_t		*txq_dmamaps;
225 
226 	void			*txq_deferred_transmit;
227 };
228 
229 struct vioif_rxqueue {
230 	kmutex_t		*rxq_lock;	/* lock for rx operations */
231 
232 	struct virtqueue	*rxq_vq;
233 	bool			rxq_stopping;
234 
235 	struct virtio_net_hdr	*rxq_hdrs;
236 	bus_dmamap_t		*rxq_hdr_dmamaps;
237 
238 	struct mbuf		**rxq_mbufs;
239 	bus_dmamap_t		*rxq_dmamaps;
240 
241 	void			*rxq_softint;
242 };
243 
244 struct vioif_ctrlqueue {
245 	struct virtqueue		*ctrlq_vq;
246 	enum {
247 		FREE, INUSE, DONE
248 	}				ctrlq_inuse;
249 	kcondvar_t			ctrlq_wait;
250 	kmutex_t			ctrlq_wait_lock;
251 	struct lwp			*ctrlq_owner;
252 
253 	struct virtio_net_ctrl_cmd	*ctrlq_cmd;
254 	struct virtio_net_ctrl_status	*ctrlq_status;
255 	struct virtio_net_ctrl_rx	*ctrlq_rx;
256 	struct virtio_net_ctrl_mac_tbl	*ctrlq_mac_tbl_uc;
257 	struct virtio_net_ctrl_mac_tbl	*ctrlq_mac_tbl_mc;
258 	struct virtio_net_ctrl_mq	*ctrlq_mq;
259 
260 	bus_dmamap_t			ctrlq_cmd_dmamap;
261 	bus_dmamap_t			ctrlq_status_dmamap;
262 	bus_dmamap_t			ctrlq_rx_dmamap;
263 	bus_dmamap_t			ctrlq_tbl_uc_dmamap;
264 	bus_dmamap_t			ctrlq_tbl_mc_dmamap;
265 	bus_dmamap_t			ctrlq_mq_dmamap;
266 };
267 
268 struct vioif_softc {
269 	device_t		sc_dev;
270 
271 	struct virtio_softc	*sc_virtio;
272 	struct virtqueue	*sc_vqs;
273 
274 	int			sc_max_nvq_pairs;
275 	int			sc_req_nvq_pairs;
276 	int			sc_act_nvq_pairs;
277 
278 	uint8_t			sc_mac[ETHER_ADDR_LEN];
279 	struct ethercom		sc_ethercom;
280 	short			sc_deferred_init_done;
281 	bool			sc_link_active;
282 
283 	struct vioif_txqueue	*sc_txq;
284 	struct vioif_rxqueue	*sc_rxq;
285 
286 	bool			sc_has_ctrl;
287 	struct vioif_ctrlqueue	sc_ctrlq;
288 
289 	bus_dma_segment_t	sc_hdr_segs[1];
290 	void			*sc_dmamem;
291 	void			*sc_kmem;
292 
293 	void			*sc_ctl_softint;
294 };
295 #define VIRTIO_NET_TX_MAXNSEGS		(16) /* XXX */
296 #define VIRTIO_NET_CTRL_MAC_MAXENTRIES	(64) /* XXX */
297 
298 /* cfattach interface functions */
299 static int	vioif_match(device_t, cfdata_t, void *);
300 static void	vioif_attach(device_t, device_t, void *);
301 static void	vioif_deferred_init(device_t);
302 
303 /* ifnet interface functions */
304 static int	vioif_init(struct ifnet *);
305 static void	vioif_stop(struct ifnet *, int);
306 static void	vioif_start(struct ifnet *);
307 static void	vioif_start_locked(struct ifnet *, struct vioif_txqueue *);
308 static int	vioif_transmit(struct ifnet *, struct mbuf *);
309 static void	vioif_transmit_locked(struct ifnet *, struct vioif_txqueue *);
310 static int	vioif_ioctl(struct ifnet *, u_long, void *);
311 static void	vioif_watchdog(struct ifnet *);
312 
313 /* rx */
314 static int	vioif_add_rx_mbuf(struct vioif_rxqueue *, int);
315 static void	vioif_free_rx_mbuf(struct vioif_rxqueue *, int);
316 static void	vioif_populate_rx_mbufs(struct vioif_rxqueue *);
317 static void	vioif_populate_rx_mbufs_locked(struct vioif_rxqueue *);
318 static int	vioif_rx_deq(struct vioif_rxqueue *);
319 static int	vioif_rx_deq_locked(struct vioif_rxqueue *);
320 static int	vioif_rx_vq_done(struct virtqueue *);
321 static void	vioif_rx_softint(void *);
322 static void	vioif_rx_drain(struct vioif_rxqueue *);
323 
324 /* tx */
325 static int	vioif_tx_vq_done(struct virtqueue *);
326 static int	vioif_tx_vq_done_locked(struct virtqueue *);
327 static void	vioif_tx_drain(struct vioif_txqueue *);
328 static void	vioif_deferred_transmit(void *);
329 
330 /* other control */
331 static bool	vioif_is_link_up(struct vioif_softc *);
332 static void	vioif_update_link_status(struct vioif_softc *);
333 static int	vioif_ctrl_rx(struct vioif_softc *, int, bool);
334 static int	vioif_set_promisc(struct vioif_softc *, bool);
335 static int	vioif_set_allmulti(struct vioif_softc *, bool);
336 static int	vioif_set_rx_filter(struct vioif_softc *);
337 static int	vioif_rx_filter(struct vioif_softc *);
338 static int	vioif_ctrl_vq_done(struct virtqueue *);
339 static int	vioif_config_change(struct virtio_softc *);
340 static void	vioif_ctl_softint(void *);
341 static int	vioif_ctrl_mq_vq_pairs_set(struct vioif_softc *, int);
342 static void	vioif_enable_interrupt_vqpairs(struct vioif_softc *);
343 static void	vioif_disable_interrupt_vqpairs(struct vioif_softc *);
344 
345 CFATTACH_DECL_NEW(vioif, sizeof(struct vioif_softc),
346 		  vioif_match, vioif_attach, NULL, NULL);
347 
348 static int
349 vioif_match(device_t parent, cfdata_t match, void *aux)
350 {
351 	struct virtio_attach_args *va = aux;
352 
353 	if (va->sc_childdevid == PCI_PRODUCT_VIRTIO_NETWORK)
354 		return 1;
355 
356 	return 0;
357 }
358 
359 static void
360 vioif_alloc_queues(struct vioif_softc *sc)
361 {
362 	int nvq_pairs = sc->sc_max_nvq_pairs;
363 	int nvqs = nvq_pairs * 2;
364 	int i;
365 
366 	KASSERT(nvq_pairs <= VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX);
367 
368 	sc->sc_rxq = kmem_zalloc(sizeof(sc->sc_rxq[0]) * nvq_pairs,
369 	    KM_SLEEP);
370 	sc->sc_txq = kmem_zalloc(sizeof(sc->sc_txq[0]) * nvq_pairs,
371 	    KM_SLEEP);
372 
373 	if (sc->sc_has_ctrl)
374 		nvqs++;
375 
376 	sc->sc_vqs = kmem_zalloc(sizeof(sc->sc_vqs[0]) * nvqs, KM_SLEEP);
377 	nvqs = 0;
378 	for (i = 0; i < nvq_pairs; i++) {
379 		sc->sc_rxq[i].rxq_vq = &sc->sc_vqs[nvqs++];
380 		sc->sc_txq[i].txq_vq = &sc->sc_vqs[nvqs++];
381 	}
382 
383 	if (sc->sc_has_ctrl)
384 		sc->sc_ctrlq.ctrlq_vq = &sc->sc_vqs[nvqs++];
385 }
386 
387 static void
388 vioif_free_queues(struct vioif_softc *sc)
389 {
390 	int nvq_pairs = sc->sc_max_nvq_pairs;
391 	int nvqs = nvq_pairs * 2;
392 
393 	if (sc->sc_ctrlq.ctrlq_vq)
394 		nvqs++;
395 
396 	if (sc->sc_txq) {
397 		kmem_free(sc->sc_txq, sizeof(sc->sc_txq[0]) * nvq_pairs);
398 		sc->sc_txq = NULL;
399 	}
400 
401 	if (sc->sc_rxq) {
402 		kmem_free(sc->sc_rxq, sizeof(sc->sc_rxq[0]) * nvq_pairs);
403 		sc->sc_rxq = NULL;
404 	}
405 
406 	if (sc->sc_vqs) {
407 		kmem_free(sc->sc_vqs, sizeof(sc->sc_vqs[0]) * nvqs);
408 		sc->sc_vqs = NULL;
409 	}
410 }
411 
412 /* allocate memory */
413 /*
414  * dma memory is used for:
415  *   rxq_hdrs[slot]:	 metadata array for received frames (READ)
416  *   txq_hdrs[slot]:	 metadata array for frames to be sent (WRITE)
417  *   ctrlq_cmd:		 command to be sent via ctrl vq (WRITE)
418  *   ctrlq_status:	 return value for a command via ctrl vq (READ)
419  *   ctrlq_rx:		 parameter for a VIRTIO_NET_CTRL_RX class command
420  *			 (WRITE)
421  *   ctrlq_mac_tbl_uc:	 unicast MAC address filter for a VIRTIO_NET_CTRL_MAC
422  *			 class command (WRITE)
423  *   ctrlq_mac_tbl_mc:	 multicast MAC address filter for a VIRTIO_NET_CTRL_MAC
424  *			 class command (WRITE)
425  * ctrlq_* structures are allocated only one each; they are protected by
426  * ctrlq_inuse variable and ctrlq_wait condvar.
427  */
428 /*
429  * dynamically allocated memory is used for:
430  *   rxq_hdr_dmamaps[slot]:	bus_dmamap_t array for sc_rx_hdrs[slot]
431  *   txq_hdr_dmamaps[slot]:	bus_dmamap_t array for sc_tx_hdrs[slot]
432  *   rxq_dmamaps[slot]:		bus_dmamap_t array for received payload
433  *   txq_dmamaps[slot]:		bus_dmamap_t array for sent payload
434  *   rxq_mbufs[slot]:		mbuf pointer array for received frames
435  *   txq_mbufs[slot]:		mbuf pointer array for sent frames
436  */
437 static int
438 vioif_alloc_mems(struct vioif_softc *sc)
439 {
440 	struct virtio_softc *vsc = sc->sc_virtio;
441 	struct vioif_txqueue *txq;
442 	struct vioif_rxqueue *rxq;
443 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
444 	int allocsize, allocsize2, r, rsegs, i, qid;
445 	void *vaddr;
446 	intptr_t p;
447 
448 	allocsize = 0;
449 	for (qid = 0; qid < sc->sc_max_nvq_pairs; qid++) {
450 		rxq = &sc->sc_rxq[qid];
451 		txq = &sc->sc_txq[qid];
452 
453 		allocsize +=
454 		    sizeof(struct virtio_net_hdr) * rxq->rxq_vq->vq_num;
455 		allocsize +=
456 		    sizeof(struct virtio_net_hdr) * txq->txq_vq->vq_num;
457 	}
458 	if (sc->sc_has_ctrl) {
459 		allocsize += sizeof(struct virtio_net_ctrl_cmd) * 1;
460 		allocsize += sizeof(struct virtio_net_ctrl_status) * 1;
461 		allocsize += sizeof(struct virtio_net_ctrl_rx) * 1;
462 		allocsize += sizeof(struct virtio_net_ctrl_mac_tbl)
463 		    + sizeof(struct virtio_net_ctrl_mac_tbl)
464 		    + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES;
465 		allocsize += sizeof(struct virtio_net_ctrl_mq) * 1;
466 	}
467 	r = bus_dmamem_alloc(virtio_dmat(vsc), allocsize, 0, 0,
468 	    &sc->sc_hdr_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
469 	if (r != 0) {
470 		aprint_error_dev(sc->sc_dev,
471 		    "DMA memory allocation failed, size %d, "
472 		    "error code %d\n", allocsize, r);
473 		goto err_none;
474 	}
475 	r = bus_dmamem_map(virtio_dmat(vsc),
476 	    &sc->sc_hdr_segs[0], 1, allocsize, &vaddr, BUS_DMA_NOWAIT);
477 	if (r != 0) {
478 		aprint_error_dev(sc->sc_dev,
479 		    "DMA memory map failed, error code %d\n", r);
480 		goto err_dmamem_alloc;
481 	}
482 
483 #define P(p, p0, p0size)	do { p0 = (void *) p;		\
484 				     p += p0size; } while (0)
485 	memset(vaddr, 0, allocsize);
486 	sc->sc_dmamem = vaddr;
487 	p = (intptr_t) vaddr;
488 
489 	for (qid = 0; qid < sc->sc_max_nvq_pairs; qid++) {
490 		rxq = &sc->sc_rxq[qid];
491 		txq = &sc->sc_txq[qid];
492 
493 		P(p, rxq->rxq_hdrs,
494 		    sizeof(rxq->rxq_hdrs[0]) * rxq->rxq_vq->vq_num);
495 		P(p, txq->txq_hdrs,
496 		    sizeof(txq->txq_hdrs[0]) * txq->txq_vq->vq_num);
497 	}
498 	if (sc->sc_has_ctrl) {
499 		P(p, ctrlq->ctrlq_cmd, sizeof(*ctrlq->ctrlq_cmd));
500 		P(p, ctrlq->ctrlq_status, sizeof(*ctrlq->ctrlq_status));
501 		P(p, ctrlq->ctrlq_rx, sizeof(*ctrlq->ctrlq_rx));
502 		P(p, ctrlq->ctrlq_mac_tbl_uc, sizeof(*ctrlq->ctrlq_mac_tbl_uc));
503 		P(p, ctrlq->ctrlq_mac_tbl_mc, sizeof(*ctrlq->ctrlq_mac_tbl_mc)
504 		    + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES);
505 		P(p, ctrlq->ctrlq_mq, sizeof(*ctrlq->ctrlq_mq));
506 	}
507 
508 	allocsize2 = 0;
509 	for (qid = 0; qid < sc->sc_max_nvq_pairs; qid++) {
510 		int rxqsize, txqsize;
511 
512 		rxq = &sc->sc_rxq[qid];
513 		txq = &sc->sc_txq[qid];
514 		rxqsize = rxq->rxq_vq->vq_num;
515 		txqsize = txq->txq_vq->vq_num;
516 
517 		allocsize2 += sizeof(rxq->rxq_dmamaps[0]) * rxqsize;
518 		allocsize2 += sizeof(rxq->rxq_hdr_dmamaps[0]) * rxqsize;
519 		allocsize2 += sizeof(rxq->rxq_mbufs[0]) * rxqsize;
520 
521 		allocsize2 += sizeof(txq->txq_dmamaps[0]) * txqsize;
522 		allocsize2 += sizeof(txq->txq_hdr_dmamaps[0]) * txqsize;
523 		allocsize2 += sizeof(txq->txq_mbufs[0]) * txqsize;
524 	}
525 	vaddr = kmem_zalloc(allocsize2, KM_SLEEP);
526 	sc->sc_kmem = vaddr;
527 	p = (intptr_t) vaddr;
528 
529 	for (qid = 0; qid < sc->sc_max_nvq_pairs; qid++) {
530 		int rxqsize, txqsize;
531 		rxq = &sc->sc_rxq[qid];
532 		txq = &sc->sc_txq[qid];
533 		rxqsize = rxq->rxq_vq->vq_num;
534 		txqsize = txq->txq_vq->vq_num;
535 
536 		P(p, rxq->rxq_hdr_dmamaps,
537 		    sizeof(rxq->rxq_hdr_dmamaps[0]) * rxqsize);
538 		P(p, txq->txq_hdr_dmamaps,
539 		    sizeof(txq->txq_hdr_dmamaps[0]) * txqsize);
540 		P(p, rxq->rxq_dmamaps, sizeof(rxq->rxq_dmamaps[0]) * rxqsize);
541 		P(p, txq->txq_dmamaps, sizeof(txq->txq_dmamaps[0]) * txqsize);
542 		P(p, rxq->rxq_mbufs, sizeof(rxq->rxq_mbufs[0]) * rxqsize);
543 		P(p, txq->txq_mbufs, sizeof(txq->txq_mbufs[0]) * txqsize);
544 	}
545 #undef P
546 
547 #define C(map, size, nsegs, usage)					      \
548 	do {								      \
549 		r = bus_dmamap_create(virtio_dmat(vsc), size, nsegs, size, 0, \
550 		    BUS_DMA_NOWAIT | BUS_DMA_ALLOCNOW,			      \
551 		    &map);						      \
552 		if (r != 0) {						      \
553 			aprint_error_dev(sc->sc_dev,			      \
554 			    usage " dmamap creation failed, "		      \
555 			    "error code %d\n", r);			      \
556 			goto err_reqs;					      \
557 		}							      \
558 	} while (0)
559 #define C_L(map, buf, size, nsegs, rw, usage)				\
560 	C(map, size, nsegs, usage);					\
561 	do {								\
562 		r = bus_dmamap_load(virtio_dmat(vsc), map,		\
563 				    buf, size, NULL,			\
564 				    rw | BUS_DMA_NOWAIT);		\
565 		if (r != 0) {						\
566 			aprint_error_dev(sc->sc_dev,			\
567 			    usage " dmamap load failed, "		\
568 			    "error code %d\n", r);			\
569 			goto err_reqs;					\
570 		}							\
571 	} while (0)
572 
573 	for (qid = 0; qid < sc->sc_max_nvq_pairs; qid++) {
574 		rxq = &sc->sc_rxq[qid];
575 		txq = &sc->sc_txq[qid];
576 
577 		for (i = 0; i < rxq->rxq_vq->vq_num; i++) {
578 			C_L(rxq->rxq_hdr_dmamaps[i], &rxq->rxq_hdrs[i],
579 			    sizeof(rxq->rxq_hdrs[0]), 1,
580 			    BUS_DMA_READ, "rx header");
581 			C(rxq->rxq_dmamaps[i], MCLBYTES, 1, "rx payload");
582 		}
583 
584 		for (i = 0; i < txq->txq_vq->vq_num; i++) {
585 			C_L(txq->txq_hdr_dmamaps[i], &txq->txq_hdrs[i],
586 			    sizeof(txq->txq_hdrs[0]), 1,
587 			    BUS_DMA_READ, "tx header");
588 			C(txq->txq_dmamaps[i], ETHER_MAX_LEN,
589 			    VIRTIO_NET_TX_MAXNSEGS, "tx payload");
590 		}
591 	}
592 
593 	if (sc->sc_has_ctrl) {
594 		/* control vq class & command */
595 		C_L(ctrlq->ctrlq_cmd_dmamap,
596 		    ctrlq->ctrlq_cmd, sizeof(*ctrlq->ctrlq_cmd), 1,
597 		    BUS_DMA_WRITE, "control command");
598 		C_L(ctrlq->ctrlq_status_dmamap,
599 		    ctrlq->ctrlq_status, sizeof(*ctrlq->ctrlq_status), 1,
600 		    BUS_DMA_READ, "control status");
601 
602 		/* control vq rx mode command parameter */
603 		C_L(ctrlq->ctrlq_rx_dmamap,
604 		    ctrlq->ctrlq_rx, sizeof(*ctrlq->ctrlq_rx), 1,
605 		    BUS_DMA_WRITE, "rx mode control command");
606 
607 		/* multiqueue set command */
608 		C_L(ctrlq->ctrlq_mq_dmamap,
609 		    ctrlq->ctrlq_mq, sizeof(*ctrlq->ctrlq_mq), 1,
610 		    BUS_DMA_WRITE, "multiqueue set command");
611 
612 		/* control vq MAC filter table for unicast */
613 		/* do not load now since its length is variable */
614 		C(ctrlq->ctrlq_tbl_uc_dmamap,
615 		    sizeof(*ctrlq->ctrlq_mac_tbl_uc) + 0, 1,
616 		    "unicast MAC address filter command");
617 
618 		/* control vq MAC filter table for multicast */
619 		C(ctrlq->ctrlq_tbl_mc_dmamap,
620 		    sizeof(*ctrlq->ctrlq_mac_tbl_mc)
621 		    + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES, 1,
622 		    "multicast MAC address filter command");
623 	}
624 #undef C_L
625 #undef C
626 
627 	return 0;
628 
629 err_reqs:
630 #define D(map)								\
631 	do {								\
632 		if (map) {						\
633 			bus_dmamap_destroy(virtio_dmat(vsc), map);	\
634 			map = NULL;					\
635 		}							\
636 	} while (0)
637 	D(ctrlq->ctrlq_tbl_mc_dmamap);
638 	D(ctrlq->ctrlq_tbl_uc_dmamap);
639 	D(ctrlq->ctrlq_rx_dmamap);
640 	D(ctrlq->ctrlq_status_dmamap);
641 	D(ctrlq->ctrlq_cmd_dmamap);
642 	for (qid = 0; qid < sc->sc_max_nvq_pairs; qid++) {
643 		rxq = &sc->sc_rxq[qid];
644 		txq = &sc->sc_txq[qid];
645 
646 		for (i = 0; i < txq->txq_vq->vq_num; i++) {
647 			D(txq->txq_dmamaps[i]);
648 			D(txq->txq_hdr_dmamaps[i]);
649 		}
650 		for (i = 0; i < rxq->rxq_vq->vq_num; i++) {
651 			D(rxq->rxq_dmamaps[i]);
652 			D(rxq->rxq_hdr_dmamaps[i]);
653 		}
654 	}
655 #undef D
656 	if (sc->sc_kmem) {
657 		kmem_free(sc->sc_kmem, allocsize2);
658 		sc->sc_kmem = NULL;
659 	}
660 	bus_dmamem_unmap(virtio_dmat(vsc), sc->sc_dmamem, allocsize);
661 err_dmamem_alloc:
662 	bus_dmamem_free(virtio_dmat(vsc), &sc->sc_hdr_segs[0], 1);
663 err_none:
664 	return -1;
665 }
666 
667 static void
668 vioif_attach(device_t parent, device_t self, void *aux)
669 {
670 	struct vioif_softc *sc = device_private(self);
671 	struct virtio_softc *vsc = device_private(parent);
672 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
673 	struct vioif_txqueue *txq;
674 	struct vioif_rxqueue *rxq;
675 	uint32_t features, req_features;
676 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
677 	u_int softint_flags;
678 	int r, i, nvqs=0, req_flags;
679 
680 	if (virtio_child(vsc) != NULL) {
681 		aprint_normal(": child already attached for %s; "
682 		    "something wrong...\n", device_xname(parent));
683 		return;
684 	}
685 
686 	sc->sc_dev = self;
687 	sc->sc_virtio = vsc;
688 	sc->sc_link_active = false;
689 
690 	sc->sc_max_nvq_pairs = 1;
691 	sc->sc_req_nvq_pairs = 1;
692 	sc->sc_act_nvq_pairs = 1;
693 
694 	req_flags = 0;
695 
696 #ifdef VIOIF_MPSAFE
697 	req_flags |= VIRTIO_F_PCI_INTR_MPSAFE;
698 #endif
699 #ifdef VIOIF_SOFTINT_INTR
700 	req_flags |= VIRTIO_F_PCI_INTR_SOFTINT;
701 #endif
702 	req_flags |= VIRTIO_F_PCI_INTR_MSIX;
703 
704 	req_features =
705 	    VIRTIO_NET_F_MAC | VIRTIO_NET_F_STATUS | VIRTIO_NET_F_CTRL_VQ |
706 	    VIRTIO_NET_F_CTRL_RX | VIRTIO_F_NOTIFY_ON_EMPTY;
707 #ifdef VIOIF_MULTIQ
708 	req_features |= VIRTIO_NET_F_MQ;
709 #endif
710 	virtio_child_attach_start(vsc, self, IPL_NET, NULL,
711 	    vioif_config_change, virtio_vq_intr, req_flags,
712 	    req_features, VIRTIO_NET_FLAG_BITS);
713 
714 	features = virtio_features(vsc);
715 
716 	if (features & VIRTIO_NET_F_MAC) {
717 		for (i = 0; i < __arraycount(sc->sc_mac); i++) {
718 			sc->sc_mac[i] = virtio_read_device_config_1(vsc,
719 			    VIRTIO_NET_CONFIG_MAC + i);
720 		}
721 	} else {
722 		/* code stolen from sys/net/if_tap.c */
723 		struct timeval tv;
724 		uint32_t ui;
725 		getmicrouptime(&tv);
726 		ui = (tv.tv_sec ^ tv.tv_usec) & 0xffffff;
727 		memcpy(sc->sc_mac+3, (uint8_t *)&ui, 3);
728 		for (i = 0; i < __arraycount(sc->sc_mac); i++) {
729 			virtio_write_device_config_1(vsc,
730 			    VIRTIO_NET_CONFIG_MAC + i, sc->sc_mac[i]);
731 		}
732 	}
733 
734 	aprint_normal_dev(self, "Ethernet address %s\n",
735 	    ether_sprintf(sc->sc_mac));
736 
737 	if ((features & VIRTIO_NET_F_CTRL_VQ) &&
738 	    (features & VIRTIO_NET_F_CTRL_RX)) {
739 		sc->sc_has_ctrl = true;
740 
741 		cv_init(&ctrlq->ctrlq_wait, "ctrl_vq");
742 		mutex_init(&ctrlq->ctrlq_wait_lock, MUTEX_DEFAULT, IPL_NET);
743 		ctrlq->ctrlq_inuse = FREE;
744 	} else {
745 		sc->sc_has_ctrl = false;
746 	}
747 
748 	if (sc->sc_has_ctrl && (features & VIRTIO_NET_F_MQ)) {
749 		sc->sc_max_nvq_pairs = virtio_read_device_config_2(vsc,
750 		    VIRTIO_NET_CONFIG_MAX_VQ_PAIRS);
751 
752 		if (sc->sc_max_nvq_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
753 			goto err;
754 
755 		/* Limit the number of queue pairs to use */
756 		sc->sc_req_nvq_pairs = MIN(sc->sc_max_nvq_pairs, ncpu);
757 	}
758 
759 	vioif_alloc_queues(sc);
760 	virtio_child_attach_set_vqs(vsc, sc->sc_vqs, sc->sc_req_nvq_pairs);
761 
762 #ifdef VIOIF_MPSAFE
763 	softint_flags = SOFTINT_NET | SOFTINT_MPSAFE;
764 #else
765 	softint_flags = SOFTINT_NET;
766 #endif
767 
768 	/*
769 	 * Allocating a virtqueues
770 	 */
771 	for (i = 0; i < sc->sc_max_nvq_pairs; i++) {
772 		rxq = &sc->sc_rxq[i];
773 		txq = &sc->sc_txq[i];
774 		char qname[32];
775 
776 		rxq->rxq_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
777 
778 		rxq->rxq_softint = softint_establish(softint_flags,
779 		    vioif_rx_softint, rxq);
780 		if (rxq->rxq_softint == NULL) {
781 			aprint_error_dev(self, "cannot establish rx softint\n");
782 			goto err;
783 		}
784 		snprintf(qname, sizeof(qname), "rx%d", i);
785 		r = virtio_alloc_vq(vsc, rxq->rxq_vq, nvqs,
786 		    MCLBYTES+sizeof(struct virtio_net_hdr), nvqs, qname);
787 		if (r != 0)
788 			goto err;
789 		nvqs++;
790 		rxq->rxq_vq->vq_done = vioif_rx_vq_done;
791 		rxq->rxq_vq->vq_done_ctx = (void *)rxq;
792 		rxq->rxq_stopping = true;
793 
794 		txq->txq_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
795 		txq->txq_deferred_transmit = softint_establish(softint_flags,
796 		    vioif_deferred_transmit, txq);
797 		if (txq->txq_deferred_transmit == NULL) {
798 			aprint_error_dev(self, "cannot establish tx softint\n");
799 			goto err;
800 		}
801 		snprintf(qname, sizeof(qname), "tx%d", i);
802 		r = virtio_alloc_vq(vsc, txq->txq_vq, nvqs,
803 		    sizeof(struct virtio_net_hdr)
804 		    + (ETHER_MAX_LEN - ETHER_HDR_LEN),
805 		    VIRTIO_NET_TX_MAXNSEGS + 1, qname);
806 		if (r != 0)
807 			goto err;
808 		nvqs++;
809 		txq->txq_vq->vq_done = vioif_tx_vq_done;
810 		txq->txq_vq->vq_done_ctx = (void *)txq;
811 		txq->txq_link_active = sc->sc_link_active;
812 		txq->txq_stopping = false;
813 		txq->txq_intrq = pcq_create(txq->txq_vq->vq_num, KM_SLEEP);
814 	}
815 
816 	if (sc->sc_has_ctrl) {
817 		/*
818 		 * Allocating a virtqueue for control channel
819 		 */
820 		r = virtio_alloc_vq(vsc, ctrlq->ctrlq_vq, nvqs,
821 		    NBPG, 1, "control");
822 		if (r != 0) {
823 			aprint_error_dev(self, "failed to allocate "
824 			    "a virtqueue for control channel, error code %d\n",
825 			    r);
826 
827 			sc->sc_has_ctrl = false;
828 			cv_destroy(&ctrlq->ctrlq_wait);
829 			mutex_destroy(&ctrlq->ctrlq_wait_lock);
830 		} else {
831 			nvqs++;
832 			ctrlq->ctrlq_vq->vq_done = vioif_ctrl_vq_done;
833 			ctrlq->ctrlq_vq->vq_done_ctx = (void *) ctrlq;
834 		}
835 	}
836 
837 	sc->sc_ctl_softint = softint_establish(softint_flags,
838 	    vioif_ctl_softint, sc);
839 	if (sc->sc_ctl_softint == NULL) {
840 		aprint_error_dev(self, "cannot establish ctl softint\n");
841 		goto err;
842 	}
843 
844 	if (vioif_alloc_mems(sc) < 0)
845 		goto err;
846 
847 	if (virtio_child_attach_finish(vsc) != 0)
848 		goto err;
849 
850 	strlcpy(ifp->if_xname, device_xname(self), IFNAMSIZ);
851 	ifp->if_softc = sc;
852 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
853 #ifdef VIOIF_MPSAFE
854 	ifp->if_extflags = IFEF_MPSAFE;
855 #endif
856 	ifp->if_start = vioif_start;
857 	if (sc->sc_req_nvq_pairs > 1)
858 		ifp->if_transmit = vioif_transmit;
859 	ifp->if_ioctl = vioif_ioctl;
860 	ifp->if_init = vioif_init;
861 	ifp->if_stop = vioif_stop;
862 	ifp->if_capabilities = 0;
863 	ifp->if_watchdog = vioif_watchdog;
864 	txq = &sc->sc_txq[0];
865 	IFQ_SET_MAXLEN(&ifp->if_snd, MAX(txq->txq_vq->vq_num, IFQ_MAXLEN));
866 	IFQ_SET_READY(&ifp->if_snd);
867 
868 	sc->sc_ethercom.ec_capabilities |= ETHERCAP_VLAN_MTU;
869 
870 	if_attach(ifp);
871 	if_deferred_start_init(ifp, NULL);
872 	ether_ifattach(ifp, sc->sc_mac);
873 
874 	return;
875 
876 err:
877 	for (i = 0; i < sc->sc_max_nvq_pairs; i++) {
878 		rxq = &sc->sc_rxq[i];
879 		txq = &sc->sc_txq[i];
880 
881 		if (rxq->rxq_lock) {
882 			mutex_obj_free(rxq->rxq_lock);
883 			rxq->rxq_lock = NULL;
884 		}
885 
886 		if (rxq->rxq_softint) {
887 			softint_disestablish(rxq->rxq_softint);
888 			rxq->rxq_softint = NULL;
889 		}
890 
891 		if (txq->txq_lock) {
892 			mutex_obj_free(txq->txq_lock);
893 			txq->txq_lock = NULL;
894 		}
895 
896 		if (txq->txq_deferred_transmit) {
897 			softint_disestablish(txq->txq_deferred_transmit);
898 			txq->txq_deferred_transmit = NULL;
899 		}
900 
901 		if (txq->txq_intrq) {
902 			pcq_destroy(txq->txq_intrq);
903 			txq->txq_intrq = NULL;
904 		}
905 	}
906 
907 	if (sc->sc_has_ctrl) {
908 		cv_destroy(&ctrlq->ctrlq_wait);
909 		mutex_destroy(&ctrlq->ctrlq_wait_lock);
910 	}
911 
912 	while (nvqs > 0)
913 		virtio_free_vq(vsc, &sc->sc_vqs[--nvqs]);
914 
915 	vioif_free_queues(sc);
916 
917 	virtio_child_attach_failed(vsc);
918 	return;
919 }
920 
921 /* we need interrupts to make promiscuous mode off */
922 static void
923 vioif_deferred_init(device_t self)
924 {
925 	struct vioif_softc *sc = device_private(self);
926 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
927 	int r;
928 
929 	if (ifp->if_flags & IFF_PROMISC)
930 		return;
931 
932 	r =  vioif_set_promisc(sc, false);
933 	if (r != 0)
934 		aprint_error_dev(self, "resetting promisc mode failed, "
935 		    "error code %d\n", r);
936 }
937 
938 static void
939 vioif_enable_interrupt_vqpairs(struct vioif_softc *sc)
940 {
941 	struct virtio_softc *vsc = sc->sc_virtio;
942 	struct vioif_txqueue *txq;
943 	struct vioif_rxqueue *rxq;
944 	int i;
945 
946 	for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
947 		txq = &sc->sc_txq[i];
948 		rxq = &sc->sc_rxq[i];
949 
950 		virtio_start_vq_intr(vsc, txq->txq_vq);
951 		virtio_start_vq_intr(vsc, rxq->rxq_vq);
952 	}
953 }
954 
955 static void
956 vioif_disable_interrupt_vqpairs(struct vioif_softc *sc)
957 {
958 	struct virtio_softc *vsc = sc->sc_virtio;
959 	struct vioif_txqueue *txq;
960 	struct vioif_rxqueue *rxq;
961 	int i;
962 
963 	for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
964 		txq = &sc->sc_txq[i];
965 		rxq = &sc->sc_rxq[i];
966 
967 		virtio_stop_vq_intr(vsc, txq->txq_vq);
968 		virtio_stop_vq_intr(vsc, rxq->rxq_vq);
969 	}
970 }
971 
972 /*
973  * Interface functions for ifnet
974  */
975 static int
976 vioif_init(struct ifnet *ifp)
977 {
978 	struct vioif_softc *sc = ifp->if_softc;
979 	struct virtio_softc *vsc = sc->sc_virtio;
980 	struct vioif_rxqueue *rxq;
981 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
982 	int r, i;
983 
984 	vioif_stop(ifp, 0);
985 
986 	virtio_reinit_start(vsc);
987 	virtio_negotiate_features(vsc, virtio_features(vsc));
988 
989 	for (i = 0; i < sc->sc_req_nvq_pairs; i++) {
990 		rxq = &sc->sc_rxq[i];
991 
992 		/* Have to set false before vioif_populate_rx_mbufs */
993 		rxq->rxq_stopping = false;
994 		vioif_populate_rx_mbufs(rxq);
995 	}
996 
997 	virtio_reinit_end(vsc);
998 
999 	if (sc->sc_has_ctrl)
1000 		virtio_start_vq_intr(vsc, ctrlq->ctrlq_vq);
1001 
1002 	r = vioif_ctrl_mq_vq_pairs_set(sc, sc->sc_req_nvq_pairs);
1003 	if (r == 0)
1004 		sc->sc_act_nvq_pairs = sc->sc_req_nvq_pairs;
1005 	else
1006 		sc->sc_act_nvq_pairs = 1;
1007 
1008 	for (i = 0; i < sc->sc_act_nvq_pairs; i++)
1009 		sc->sc_txq[i].txq_stopping = false;
1010 
1011 	vioif_enable_interrupt_vqpairs(sc);
1012 
1013 	if (!sc->sc_deferred_init_done) {
1014 		sc->sc_deferred_init_done = 1;
1015 		if (sc->sc_has_ctrl)
1016 			vioif_deferred_init(sc->sc_dev);
1017 	}
1018 
1019 	vioif_update_link_status(sc);
1020 	ifp->if_flags |= IFF_RUNNING;
1021 	ifp->if_flags &= ~IFF_OACTIVE;
1022 	vioif_rx_filter(sc);
1023 
1024 	return 0;
1025 }
1026 
1027 static void
1028 vioif_stop(struct ifnet *ifp, int disable)
1029 {
1030 	struct vioif_softc *sc = ifp->if_softc;
1031 	struct virtio_softc *vsc = sc->sc_virtio;
1032 	struct vioif_txqueue *txq;
1033 	struct vioif_rxqueue *rxq;
1034 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
1035 	int i;
1036 
1037 	/* Take the locks to ensure that ongoing TX/RX finish */
1038 	for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
1039 		txq = &sc->sc_txq[i];
1040 		rxq = &sc->sc_rxq[i];
1041 
1042 		mutex_enter(txq->txq_lock);
1043 		txq->txq_stopping = true;
1044 		mutex_exit(txq->txq_lock);
1045 
1046 		mutex_enter(rxq->rxq_lock);
1047 		rxq->rxq_stopping = true;
1048 		mutex_exit(rxq->rxq_lock);
1049 	}
1050 
1051 	/* disable interrupts */
1052 	vioif_disable_interrupt_vqpairs(sc);
1053 
1054 	if (sc->sc_has_ctrl)
1055 		virtio_stop_vq_intr(vsc, ctrlq->ctrlq_vq);
1056 
1057 	/* only way to stop I/O and DMA is resetting... */
1058 	virtio_reset(vsc);
1059 	for (i = 0; i < sc->sc_act_nvq_pairs; i++)
1060 		vioif_rx_deq(&sc->sc_rxq[i]);
1061 
1062 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1063 	sc->sc_link_active = false;
1064 
1065 	for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
1066 		txq = &sc->sc_txq[i];
1067 		rxq = &sc->sc_rxq[i];
1068 
1069 		txq->txq_link_active = false;
1070 
1071 		if (disable)
1072 			vioif_rx_drain(rxq);
1073 
1074 		vioif_tx_drain(txq);
1075 	}
1076 }
1077 
1078 static void
1079 vioif_send_common_locked(struct ifnet *ifp, struct vioif_txqueue *txq,
1080     bool is_transmit)
1081 {
1082 	struct vioif_softc *sc = ifp->if_softc;
1083 	struct virtio_softc *vsc = sc->sc_virtio;
1084 	struct virtqueue *vq = txq->txq_vq;
1085 	struct mbuf *m;
1086 	int queued = 0;
1087 
1088 	KASSERT(mutex_owned(txq->txq_lock));
1089 
1090 	if ((ifp->if_flags & IFF_RUNNING) == 0)
1091 		return;
1092 
1093 	if (!txq->txq_link_active || txq->txq_stopping)
1094 		return;
1095 
1096 	if ((ifp->if_flags & IFF_OACTIVE) != 0 && !is_transmit)
1097 		return;
1098 
1099 	for (;;) {
1100 		int slot, r;
1101 
1102 		if (is_transmit)
1103 			m = pcq_get(txq->txq_intrq);
1104 		else
1105 			IFQ_DEQUEUE(&ifp->if_snd, m);
1106 
1107 		if (m == NULL)
1108 			break;
1109 
1110 		r = virtio_enqueue_prep(vsc, vq, &slot);
1111 		if (r == EAGAIN) {
1112 			ifp->if_flags |= IFF_OACTIVE;
1113 			m_freem(m);
1114 			break;
1115 		}
1116 		if (r != 0)
1117 			panic("enqueue_prep for a tx buffer");
1118 
1119 		r = bus_dmamap_load_mbuf(virtio_dmat(vsc),
1120 		    txq->txq_dmamaps[slot], m, BUS_DMA_WRITE | BUS_DMA_NOWAIT);
1121 		if (r != 0) {
1122 			/* maybe just too fragmented */
1123 			struct mbuf *newm;
1124 
1125 			newm = m_defrag(m, M_NOWAIT);
1126 			if (newm == NULL) {
1127 				aprint_error_dev(sc->sc_dev,
1128 				    "m_defrag() failed\n");
1129 				goto skip;
1130 			}
1131 
1132 			m = newm;
1133 			r = bus_dmamap_load_mbuf(virtio_dmat(vsc),
1134 			    txq->txq_dmamaps[slot], m,
1135 			    BUS_DMA_WRITE | BUS_DMA_NOWAIT);
1136 			if (r != 0) {
1137 				aprint_error_dev(sc->sc_dev,
1138 				    "tx dmamap load failed, error code %d\n",
1139 				    r);
1140 skip:
1141 				m_freem(m);
1142 				virtio_enqueue_abort(vsc, vq, slot);
1143 				continue;
1144 			}
1145 		}
1146 
1147 		/* This should actually never fail */
1148 		r = virtio_enqueue_reserve(vsc, vq, slot,
1149 					txq->txq_dmamaps[slot]->dm_nsegs + 1);
1150 		if (r != 0) {
1151 			aprint_error_dev(sc->sc_dev,
1152 			    "virtio_enqueue_reserve failed, error code %d\n",
1153 			    r);
1154 			bus_dmamap_unload(virtio_dmat(vsc),
1155 					  txq->txq_dmamaps[slot]);
1156 			/* slot already freed by virtio_enqueue_reserve */
1157 			m_freem(m);
1158 			continue;
1159 		}
1160 
1161 		txq->txq_mbufs[slot] = m;
1162 
1163 		memset(&txq->txq_hdrs[slot], 0, sizeof(struct virtio_net_hdr));
1164 		bus_dmamap_sync(virtio_dmat(vsc), txq->txq_dmamaps[slot],
1165 		    0, txq->txq_dmamaps[slot]->dm_mapsize,
1166 		    BUS_DMASYNC_PREWRITE);
1167 		bus_dmamap_sync(virtio_dmat(vsc), txq->txq_hdr_dmamaps[slot],
1168 		    0, txq->txq_hdr_dmamaps[slot]->dm_mapsize,
1169 		    BUS_DMASYNC_PREWRITE);
1170 		virtio_enqueue(vsc, vq, slot, txq->txq_hdr_dmamaps[slot], true);
1171 		virtio_enqueue(vsc, vq, slot, txq->txq_dmamaps[slot], true);
1172 		virtio_enqueue_commit(vsc, vq, slot, false);
1173 
1174 		queued++;
1175 		bpf_mtap(ifp, m, BPF_D_OUT);
1176 	}
1177 
1178 	if (queued > 0) {
1179 		virtio_enqueue_commit(vsc, vq, -1, true);
1180 		ifp->if_timer = 5;
1181 	}
1182 }
1183 
1184 static void
1185 vioif_start_locked(struct ifnet *ifp, struct vioif_txqueue *txq)
1186 {
1187 
1188 	/*
1189 	 * ifp->if_obytes and ifp->if_omcasts are added in if_transmit()@if.c.
1190 	 */
1191 	vioif_send_common_locked(ifp, txq, false);
1192 
1193 }
1194 
1195 static void
1196 vioif_start(struct ifnet *ifp)
1197 {
1198 	struct vioif_softc *sc = ifp->if_softc;
1199 	struct vioif_txqueue *txq = &sc->sc_txq[0];
1200 
1201 #ifdef VIOIF_MPSAFE
1202 	KASSERT(if_is_mpsafe(ifp));
1203 #endif
1204 
1205 	mutex_enter(txq->txq_lock);
1206 	if (!txq->txq_stopping)
1207 		vioif_start_locked(ifp, txq);
1208 	mutex_exit(txq->txq_lock);
1209 }
1210 
1211 static inline int
1212 vioif_select_txqueue(struct ifnet *ifp, struct mbuf *m)
1213 {
1214 	struct vioif_softc *sc = ifp->if_softc;
1215 	u_int cpuid = cpu_index(curcpu());
1216 
1217 	return cpuid % sc->sc_act_nvq_pairs;
1218 }
1219 
1220 static void
1221 vioif_transmit_locked(struct ifnet *ifp, struct vioif_txqueue *txq)
1222 {
1223 
1224 	vioif_send_common_locked(ifp, txq, true);
1225 }
1226 
1227 static int
1228 vioif_transmit(struct ifnet *ifp, struct mbuf *m)
1229 {
1230 	struct vioif_softc *sc = ifp->if_softc;
1231 	struct vioif_txqueue *txq;
1232 	int qid;
1233 
1234 	qid = vioif_select_txqueue(ifp, m);
1235 	txq = &sc->sc_txq[qid];
1236 
1237 	if (__predict_false(!pcq_put(txq->txq_intrq, m))) {
1238 		m_freem(m);
1239 		return ENOBUFS;
1240 	}
1241 
1242 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
1243 	if_statadd_ref(nsr, if_obytes, m->m_pkthdr.len);
1244 	if (m->m_flags & M_MCAST)
1245 		if_statinc_ref(nsr, if_omcasts);
1246 	IF_STAT_PUTREF(ifp);
1247 
1248 	if (mutex_tryenter(txq->txq_lock)) {
1249 		if (!txq->txq_stopping)
1250 			vioif_transmit_locked(ifp, txq);
1251 		mutex_exit(txq->txq_lock);
1252 	}
1253 
1254 	return 0;
1255 }
1256 
1257 static void
1258 vioif_deferred_transmit(void *arg)
1259 {
1260 	struct vioif_txqueue *txq = arg;
1261 	struct virtio_softc *vsc = txq->txq_vq->vq_owner;
1262 	struct vioif_softc *sc = device_private(virtio_child(vsc));
1263 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1264 
1265 	if (mutex_tryenter(txq->txq_lock)) {
1266 		vioif_send_common_locked(ifp, txq, true);
1267 		mutex_exit(txq->txq_lock);
1268 	}
1269 }
1270 
1271 static int
1272 vioif_ioctl(struct ifnet *ifp, u_long cmd, void *data)
1273 {
1274 	int s, r;
1275 
1276 	s = splnet();
1277 
1278 	r = ether_ioctl(ifp, cmd, data);
1279 	if ((r == 0 && cmd == SIOCSIFFLAGS) ||
1280 	    (r == ENETRESET && (cmd == SIOCADDMULTI || cmd == SIOCDELMULTI))) {
1281 		if (ifp->if_flags & IFF_RUNNING)
1282 			r = vioif_rx_filter(ifp->if_softc);
1283 		else
1284 			r = 0;
1285 	}
1286 
1287 	splx(s);
1288 
1289 	return r;
1290 }
1291 
1292 void
1293 vioif_watchdog(struct ifnet *ifp)
1294 {
1295 	struct vioif_softc *sc = ifp->if_softc;
1296 	int i;
1297 
1298 	if (ifp->if_flags & IFF_RUNNING) {
1299 		for (i = 0; i < sc->sc_act_nvq_pairs; i++)
1300 			vioif_tx_vq_done(sc->sc_txq[i].txq_vq);
1301 	}
1302 }
1303 
1304 
1305 /*
1306  * Receive implementation
1307  */
1308 /* allocate and initialize a mbuf for receive */
1309 static int
1310 vioif_add_rx_mbuf(struct vioif_rxqueue *rxq, int i)
1311 {
1312 	struct virtio_softc *vsc = rxq->rxq_vq->vq_owner;
1313 	struct mbuf *m;
1314 	int r;
1315 
1316 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1317 	if (m == NULL)
1318 		return ENOBUFS;
1319 	MCLGET(m, M_DONTWAIT);
1320 	if ((m->m_flags & M_EXT) == 0) {
1321 		m_freem(m);
1322 		return ENOBUFS;
1323 	}
1324 	rxq->rxq_mbufs[i] = m;
1325 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
1326 	r = bus_dmamap_load_mbuf(virtio_dmat(vsc),
1327 	    rxq->rxq_dmamaps[i], m, BUS_DMA_READ | BUS_DMA_NOWAIT);
1328 	if (r) {
1329 		m_freem(m);
1330 		rxq->rxq_mbufs[i] = NULL;
1331 		return r;
1332 	}
1333 
1334 	return 0;
1335 }
1336 
1337 /* free a mbuf for receive */
1338 static void
1339 vioif_free_rx_mbuf(struct vioif_rxqueue *rxq, int i)
1340 {
1341 	struct virtio_softc *vsc = rxq->rxq_vq->vq_owner;
1342 
1343 	bus_dmamap_unload(virtio_dmat(vsc), rxq->rxq_dmamaps[i]);
1344 	m_freem(rxq->rxq_mbufs[i]);
1345 	rxq->rxq_mbufs[i] = NULL;
1346 }
1347 
1348 /* add mbufs for all the empty receive slots */
1349 static void
1350 vioif_populate_rx_mbufs(struct vioif_rxqueue *rxq)
1351 {
1352 
1353 	mutex_enter(rxq->rxq_lock);
1354 	vioif_populate_rx_mbufs_locked(rxq);
1355 	mutex_exit(rxq->rxq_lock);
1356 }
1357 
1358 static void
1359 vioif_populate_rx_mbufs_locked(struct vioif_rxqueue *rxq)
1360 {
1361 	struct virtqueue *vq = rxq->rxq_vq;
1362 	struct virtio_softc *vsc = vq->vq_owner;
1363 	struct vioif_softc *sc = device_private(virtio_child(vsc));
1364 	int i, r, ndone = 0;
1365 
1366 	KASSERT(mutex_owned(rxq->rxq_lock));
1367 
1368 	if (rxq->rxq_stopping)
1369 		return;
1370 
1371 	for (i = 0; i < vq->vq_num; i++) {
1372 		int slot;
1373 		r = virtio_enqueue_prep(vsc, vq, &slot);
1374 		if (r == EAGAIN)
1375 			break;
1376 		if (r != 0)
1377 			panic("enqueue_prep for rx buffers");
1378 		if (rxq->rxq_mbufs[slot] == NULL) {
1379 			r = vioif_add_rx_mbuf(rxq, slot);
1380 			if (r != 0) {
1381 				aprint_error_dev(sc->sc_dev,
1382 				    "rx mbuf allocation failed, "
1383 				    "error code %d\n", r);
1384 				break;
1385 			}
1386 		}
1387 		r = virtio_enqueue_reserve(vsc, vq, slot,
1388 		    rxq->rxq_dmamaps[slot]->dm_nsegs + 1);
1389 		if (r != 0) {
1390 			vioif_free_rx_mbuf(rxq, slot);
1391 			break;
1392 		}
1393 		bus_dmamap_sync(virtio_dmat(vsc), rxq->rxq_hdr_dmamaps[slot],
1394 		    0, sizeof(struct virtio_net_hdr), BUS_DMASYNC_PREREAD);
1395 		bus_dmamap_sync(virtio_dmat(vsc), rxq->rxq_dmamaps[slot],
1396 		    0, MCLBYTES, BUS_DMASYNC_PREREAD);
1397 		virtio_enqueue(vsc, vq, slot, rxq->rxq_hdr_dmamaps[slot],
1398 		    false);
1399 		virtio_enqueue(vsc, vq, slot, rxq->rxq_dmamaps[slot], false);
1400 		virtio_enqueue_commit(vsc, vq, slot, false);
1401 		ndone++;
1402 	}
1403 	if (ndone > 0)
1404 		virtio_enqueue_commit(vsc, vq, -1, true);
1405 }
1406 
1407 /* dequeue received packets */
1408 static int
1409 vioif_rx_deq(struct vioif_rxqueue *rxq)
1410 {
1411 	int r;
1412 
1413 	KASSERT(rxq->rxq_stopping);
1414 
1415 	mutex_enter(rxq->rxq_lock);
1416 	r = vioif_rx_deq_locked(rxq);
1417 	mutex_exit(rxq->rxq_lock);
1418 
1419 	return r;
1420 }
1421 
1422 /* dequeue received packets */
1423 static int
1424 vioif_rx_deq_locked(struct vioif_rxqueue *rxq)
1425 {
1426 	struct virtqueue *vq = rxq->rxq_vq;
1427 	struct virtio_softc *vsc = vq->vq_owner;
1428 	struct vioif_softc *sc = device_private(virtio_child(vsc));
1429 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1430 	struct mbuf *m;
1431 	int r = 0;
1432 	int slot, len;
1433 
1434 	KASSERT(mutex_owned(rxq->rxq_lock));
1435 
1436 	while (virtio_dequeue(vsc, vq, &slot, &len) == 0) {
1437 		len -= sizeof(struct virtio_net_hdr);
1438 		r = 1;
1439 		bus_dmamap_sync(virtio_dmat(vsc), rxq->rxq_hdr_dmamaps[slot],
1440 		    0, sizeof(struct virtio_net_hdr), BUS_DMASYNC_POSTREAD);
1441 		bus_dmamap_sync(virtio_dmat(vsc), rxq->rxq_dmamaps[slot],
1442 		    0, MCLBYTES, BUS_DMASYNC_POSTREAD);
1443 		m = rxq->rxq_mbufs[slot];
1444 		KASSERT(m != NULL);
1445 		bus_dmamap_unload(virtio_dmat(vsc), rxq->rxq_dmamaps[slot]);
1446 		rxq->rxq_mbufs[slot] = NULL;
1447 		virtio_dequeue_commit(vsc, vq, slot);
1448 		m_set_rcvif(m, ifp);
1449 		m->m_len = m->m_pkthdr.len = len;
1450 
1451 		mutex_exit(rxq->rxq_lock);
1452 		if_percpuq_enqueue(ifp->if_percpuq, m);
1453 		mutex_enter(rxq->rxq_lock);
1454 
1455 		if (rxq->rxq_stopping)
1456 			break;
1457 	}
1458 
1459 	return r;
1460 }
1461 
1462 /* rx interrupt; call _dequeue above and schedule a softint */
1463 static int
1464 vioif_rx_vq_done(struct virtqueue *vq)
1465 {
1466 	struct vioif_rxqueue *rxq = vq->vq_done_ctx;
1467 	int r = 0;
1468 
1469 #ifdef VIOIF_SOFTINT_INTR
1470 	KASSERT(!cpu_intr_p());
1471 #endif
1472 
1473 	mutex_enter(rxq->rxq_lock);
1474 
1475 	if (rxq->rxq_stopping)
1476 		goto out;
1477 
1478 	r = vioif_rx_deq_locked(rxq);
1479 	if (r)
1480 #ifdef VIOIF_SOFTINT_INTR
1481 		vioif_populate_rx_mbufs_locked(rxq);
1482 #else
1483 		softint_schedule(rxq->rxq_softint);
1484 #endif
1485 
1486 out:
1487 	mutex_exit(rxq->rxq_lock);
1488 	return r;
1489 }
1490 
1491 /* softint: enqueue receive requests for new incoming packets */
1492 static void
1493 vioif_rx_softint(void *arg)
1494 {
1495 	struct vioif_rxqueue *rxq = arg;
1496 
1497 	vioif_populate_rx_mbufs(rxq);
1498 }
1499 
1500 /* free all the mbufs; called from if_stop(disable) */
1501 static void
1502 vioif_rx_drain(struct vioif_rxqueue *rxq)
1503 {
1504 	struct virtqueue *vq = rxq->rxq_vq;
1505 	int i;
1506 
1507 	for (i = 0; i < vq->vq_num; i++) {
1508 		if (rxq->rxq_mbufs[i] == NULL)
1509 			continue;
1510 		vioif_free_rx_mbuf(rxq, i);
1511 	}
1512 }
1513 
1514 
1515 /*
1516  * Transmition implementation
1517  */
1518 /* actual transmission is done in if_start */
1519 /* tx interrupt; dequeue and free mbufs */
1520 /*
1521  * tx interrupt is actually disabled; this should be called upon
1522  * tx vq full and watchdog
1523  */
1524 static int
1525 vioif_tx_vq_done(struct virtqueue *vq)
1526 {
1527 	struct virtio_softc *vsc = vq->vq_owner;
1528 	struct vioif_softc *sc = device_private(virtio_child(vsc));
1529 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1530 	struct vioif_txqueue *txq = vq->vq_done_ctx;
1531 	int r = 0;
1532 
1533 	mutex_enter(txq->txq_lock);
1534 
1535 	if (txq->txq_stopping)
1536 		goto out;
1537 
1538 	r = vioif_tx_vq_done_locked(vq);
1539 
1540 out:
1541 	mutex_exit(txq->txq_lock);
1542 	if (r) {
1543 		if_schedule_deferred_start(ifp);
1544 
1545 		KASSERT(txq->txq_deferred_transmit != NULL);
1546 		softint_schedule(txq->txq_deferred_transmit);
1547 	}
1548 	return r;
1549 }
1550 
1551 static int
1552 vioif_tx_vq_done_locked(struct virtqueue *vq)
1553 {
1554 	struct virtio_softc *vsc = vq->vq_owner;
1555 	struct vioif_softc *sc = device_private(virtio_child(vsc));
1556 	struct vioif_txqueue *txq = vq->vq_done_ctx;
1557 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1558 	struct mbuf *m;
1559 	int r = 0;
1560 	int slot, len;
1561 
1562 	KASSERT(mutex_owned(txq->txq_lock));
1563 
1564 	while (virtio_dequeue(vsc, vq, &slot, &len) == 0) {
1565 		r++;
1566 		bus_dmamap_sync(virtio_dmat(vsc), txq->txq_hdr_dmamaps[slot],
1567 		    0, sizeof(struct virtio_net_hdr), BUS_DMASYNC_POSTWRITE);
1568 		bus_dmamap_sync(virtio_dmat(vsc), txq->txq_dmamaps[slot],
1569 		    0, txq->txq_dmamaps[slot]->dm_mapsize,
1570 		    BUS_DMASYNC_POSTWRITE);
1571 		m = txq->txq_mbufs[slot];
1572 		bus_dmamap_unload(virtio_dmat(vsc), txq->txq_dmamaps[slot]);
1573 		txq->txq_mbufs[slot] = NULL;
1574 		virtio_dequeue_commit(vsc, vq, slot);
1575 		if_statinc(ifp, if_opackets);
1576 		m_freem(m);
1577 	}
1578 
1579 	if (r)
1580 		ifp->if_flags &= ~IFF_OACTIVE;
1581 	return r;
1582 }
1583 
1584 /* free all the mbufs already put on vq; called from if_stop(disable) */
1585 static void
1586 vioif_tx_drain(struct vioif_txqueue *txq)
1587 {
1588 	struct virtqueue *vq = txq->txq_vq;
1589 	struct virtio_softc *vsc = vq->vq_owner;
1590 	int i;
1591 
1592 	KASSERT(txq->txq_stopping);
1593 
1594 	for (i = 0; i < vq->vq_num; i++) {
1595 		if (txq->txq_mbufs[i] == NULL)
1596 			continue;
1597 		bus_dmamap_unload(virtio_dmat(vsc), txq->txq_dmamaps[i]);
1598 		m_freem(txq->txq_mbufs[i]);
1599 		txq->txq_mbufs[i] = NULL;
1600 	}
1601 }
1602 
1603 /*
1604  * Control vq
1605  */
1606 /* issue a VIRTIO_NET_CTRL_RX class command and wait for completion */
1607 static void
1608 vioif_ctrl_acquire(struct vioif_softc *sc)
1609 {
1610 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
1611 
1612 	mutex_enter(&ctrlq->ctrlq_wait_lock);
1613 	while (ctrlq->ctrlq_inuse != FREE)
1614 		cv_wait(&ctrlq->ctrlq_wait, &ctrlq->ctrlq_wait_lock);
1615 	ctrlq->ctrlq_inuse = INUSE;
1616 	ctrlq->ctrlq_owner = curlwp;
1617 	mutex_exit(&ctrlq->ctrlq_wait_lock);
1618 }
1619 
1620 static void
1621 vioif_ctrl_release(struct vioif_softc *sc)
1622 {
1623 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
1624 
1625 	KASSERT(ctrlq->ctrlq_inuse != FREE);
1626 	KASSERT(ctrlq->ctrlq_owner == curlwp);
1627 
1628 	mutex_enter(&ctrlq->ctrlq_wait_lock);
1629 	ctrlq->ctrlq_inuse = FREE;
1630 	ctrlq->ctrlq_owner = NULL;
1631 	cv_signal(&ctrlq->ctrlq_wait);
1632 	mutex_exit(&ctrlq->ctrlq_wait_lock);
1633 }
1634 
1635 static int
1636 vioif_ctrl_load_cmdspec(struct vioif_softc *sc,
1637     struct vioif_ctrl_cmdspec *specs, int nspecs)
1638 {
1639 	struct virtio_softc *vsc = sc->sc_virtio;
1640 	int i, r, loaded;
1641 
1642 	loaded = 0;
1643 	for (i = 0; i < nspecs; i++) {
1644 		r = bus_dmamap_load(virtio_dmat(vsc),
1645 		    specs[i].dmamap, specs[i].buf, specs[i].bufsize,
1646 		    NULL, BUS_DMA_WRITE | BUS_DMA_NOWAIT);
1647 		if (r) {
1648 			aprint_error_dev(sc->sc_dev, "control command dmamap"
1649 			    " load failed, error code %d\n", r);
1650 			goto err;
1651 		}
1652 		loaded++;
1653 
1654 	}
1655 
1656 	return r;
1657 
1658 err:
1659 	for (i = 0; i < loaded; i++) {
1660 		bus_dmamap_unload(virtio_dmat(vsc), specs[i].dmamap);
1661 	}
1662 
1663 	return r;
1664 }
1665 
1666 static void
1667 vioif_ctrl_unload_cmdspec(struct vioif_softc *sc,
1668     struct vioif_ctrl_cmdspec *specs, int nspecs)
1669 {
1670 	struct virtio_softc *vsc = sc->sc_virtio;
1671 	int i;
1672 
1673 	for (i = 0; i < nspecs; i++) {
1674 		bus_dmamap_unload(virtio_dmat(vsc), specs[i].dmamap);
1675 	}
1676 }
1677 
1678 static int
1679 vioif_ctrl_send_command(struct vioif_softc *sc, uint8_t class, uint8_t cmd,
1680     struct vioif_ctrl_cmdspec *specs, int nspecs)
1681 {
1682 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
1683 	struct virtqueue *vq = ctrlq->ctrlq_vq;
1684 	struct virtio_softc *vsc = sc->sc_virtio;
1685 	int i, r, slot;
1686 
1687 	ctrlq->ctrlq_cmd->class = class;
1688 	ctrlq->ctrlq_cmd->command = cmd;
1689 
1690 	bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_cmd_dmamap,
1691 	    0, sizeof(struct virtio_net_ctrl_cmd), BUS_DMASYNC_PREWRITE);
1692 	for (i = 0; i < nspecs; i++) {
1693 		bus_dmamap_sync(virtio_dmat(vsc), specs[i].dmamap,
1694 		    0, specs[i].bufsize, BUS_DMASYNC_PREWRITE);
1695 	}
1696 	bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_status_dmamap,
1697 	    0, sizeof(struct virtio_net_ctrl_status), BUS_DMASYNC_PREREAD);
1698 
1699 	r = virtio_enqueue_prep(vsc, vq, &slot);
1700 	if (r != 0)
1701 		panic("%s: control vq busy!?", device_xname(sc->sc_dev));
1702 	r = virtio_enqueue_reserve(vsc, vq, slot, nspecs + 2);
1703 	if (r != 0)
1704 		panic("%s: control vq busy!?", device_xname(sc->sc_dev));
1705 	virtio_enqueue(vsc, vq, slot, ctrlq->ctrlq_cmd_dmamap, true);
1706 	for (i = 0; i < nspecs; i++) {
1707 		virtio_enqueue(vsc, vq, slot, specs[i].dmamap, true);
1708 	}
1709 	virtio_enqueue(vsc, vq, slot, ctrlq->ctrlq_status_dmamap, false);
1710 	virtio_enqueue_commit(vsc, vq, slot, true);
1711 
1712 	/* wait for done */
1713 	mutex_enter(&ctrlq->ctrlq_wait_lock);
1714 	while (ctrlq->ctrlq_inuse != DONE)
1715 		cv_wait(&ctrlq->ctrlq_wait, &ctrlq->ctrlq_wait_lock);
1716 	mutex_exit(&ctrlq->ctrlq_wait_lock);
1717 	/* already dequeueued */
1718 
1719 	bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_cmd_dmamap, 0,
1720 	    sizeof(struct virtio_net_ctrl_cmd), BUS_DMASYNC_POSTWRITE);
1721 	for (i = 0; i < nspecs; i++) {
1722 		bus_dmamap_sync(virtio_dmat(vsc), specs[i].dmamap, 0,
1723 		    specs[i].bufsize, BUS_DMASYNC_POSTWRITE);
1724 	}
1725 	bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_status_dmamap, 0,
1726 	    sizeof(struct virtio_net_ctrl_status), BUS_DMASYNC_POSTREAD);
1727 
1728 	if (ctrlq->ctrlq_status->ack == VIRTIO_NET_OK)
1729 		r = 0;
1730 	else {
1731 		aprint_error_dev(sc->sc_dev, "failed setting rx mode\n");
1732 		r = EIO;
1733 	}
1734 
1735 	return r;
1736 }
1737 
1738 static int
1739 vioif_ctrl_rx(struct vioif_softc *sc, int cmd, bool onoff)
1740 {
1741 	struct virtio_net_ctrl_rx *rx = sc->sc_ctrlq.ctrlq_rx;
1742 	struct vioif_ctrl_cmdspec specs[1];
1743 	int r;
1744 
1745 	if (!sc->sc_has_ctrl)
1746 		return ENOTSUP;
1747 
1748 	vioif_ctrl_acquire(sc);
1749 
1750 	rx->onoff = onoff;
1751 	specs[0].dmamap = sc->sc_ctrlq.ctrlq_rx_dmamap;
1752 	specs[0].buf = rx;
1753 	specs[0].bufsize = sizeof(*rx);
1754 
1755 	r = vioif_ctrl_send_command(sc, VIRTIO_NET_CTRL_RX, cmd,
1756 	    specs, __arraycount(specs));
1757 
1758 	vioif_ctrl_release(sc);
1759 	return r;
1760 }
1761 
1762 static int
1763 vioif_set_promisc(struct vioif_softc *sc, bool onoff)
1764 {
1765 	return vioif_ctrl_rx(sc, VIRTIO_NET_CTRL_RX_PROMISC, onoff);
1766 }
1767 
1768 static int
1769 vioif_set_allmulti(struct vioif_softc *sc, bool onoff)
1770 {
1771 	return vioif_ctrl_rx(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, onoff);
1772 }
1773 
1774 /* issue VIRTIO_NET_CTRL_MAC_TABLE_SET command and wait for completion */
1775 static int
1776 vioif_set_rx_filter(struct vioif_softc *sc)
1777 {
1778 	/* filter already set in ctrlq->ctrlq_mac_tbl */
1779 	struct virtio_net_ctrl_mac_tbl *mac_tbl_uc, *mac_tbl_mc;
1780 	struct vioif_ctrl_cmdspec specs[2];
1781 	int nspecs = __arraycount(specs);
1782 	int r;
1783 
1784 	mac_tbl_uc = sc->sc_ctrlq.ctrlq_mac_tbl_uc;
1785 	mac_tbl_mc = sc->sc_ctrlq.ctrlq_mac_tbl_mc;
1786 
1787 	if (!sc->sc_has_ctrl)
1788 		return ENOTSUP;
1789 
1790 	vioif_ctrl_acquire(sc);
1791 
1792 	specs[0].dmamap = sc->sc_ctrlq.ctrlq_tbl_uc_dmamap;
1793 	specs[0].buf = mac_tbl_uc;
1794 	specs[0].bufsize = sizeof(*mac_tbl_uc)
1795 	    + (ETHER_ADDR_LEN * mac_tbl_uc->nentries);
1796 
1797 	specs[1].dmamap = sc->sc_ctrlq.ctrlq_tbl_mc_dmamap;
1798 	specs[1].buf = mac_tbl_mc;
1799 	specs[1].bufsize = sizeof(*mac_tbl_mc)
1800 	    + (ETHER_ADDR_LEN * mac_tbl_mc->nentries);
1801 
1802 	r = vioif_ctrl_load_cmdspec(sc, specs, nspecs);
1803 	if (r != 0)
1804 		goto out;
1805 
1806 	r = vioif_ctrl_send_command(sc,
1807 	    VIRTIO_NET_CTRL_MAC, VIRTIO_NET_CTRL_MAC_TABLE_SET,
1808 	    specs, nspecs);
1809 
1810 	vioif_ctrl_unload_cmdspec(sc, specs, nspecs);
1811 
1812 out:
1813 	vioif_ctrl_release(sc);
1814 
1815 	return r;
1816 }
1817 
1818 static int
1819 vioif_ctrl_mq_vq_pairs_set(struct vioif_softc *sc, int nvq_pairs)
1820 {
1821 	struct virtio_net_ctrl_mq *mq = sc->sc_ctrlq.ctrlq_mq;
1822 	struct vioif_ctrl_cmdspec specs[1];
1823 	int r;
1824 
1825 	if (!sc->sc_has_ctrl)
1826 		return ENOTSUP;
1827 
1828 	if (nvq_pairs <= 1)
1829 		return EINVAL;
1830 
1831 	vioif_ctrl_acquire(sc);
1832 
1833 	mq->virtqueue_pairs = nvq_pairs;
1834 	specs[0].dmamap = sc->sc_ctrlq.ctrlq_mq_dmamap;
1835 	specs[0].buf = mq;
1836 	specs[0].bufsize = sizeof(*mq);
1837 
1838 	r = vioif_ctrl_send_command(sc,
1839 	    VIRTIO_NET_CTRL_MQ, VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET,
1840 	    specs, __arraycount(specs));
1841 
1842 	vioif_ctrl_release(sc);
1843 
1844 	return r;
1845 }
1846 
1847 /* ctrl vq interrupt; wake up the command issuer */
1848 static int
1849 vioif_ctrl_vq_done(struct virtqueue *vq)
1850 {
1851 	struct virtio_softc *vsc = vq->vq_owner;
1852 	struct vioif_ctrlqueue *ctrlq = vq->vq_done_ctx;
1853 	int r, slot;
1854 
1855 	r = virtio_dequeue(vsc, vq, &slot, NULL);
1856 	if (r == ENOENT)
1857 		return 0;
1858 	virtio_dequeue_commit(vsc, vq, slot);
1859 
1860 	mutex_enter(&ctrlq->ctrlq_wait_lock);
1861 	ctrlq->ctrlq_inuse = DONE;
1862 	cv_signal(&ctrlq->ctrlq_wait);
1863 	mutex_exit(&ctrlq->ctrlq_wait_lock);
1864 
1865 	return 1;
1866 }
1867 
1868 /*
1869  * If IFF_PROMISC requested,  set promiscuous
1870  * If multicast filter small enough (<=MAXENTRIES) set rx filter
1871  * If large multicast filter exist use ALLMULTI
1872  */
1873 /*
1874  * If setting rx filter fails fall back to ALLMULTI
1875  * If ALLMULTI fails fall back to PROMISC
1876  */
1877 static int
1878 vioif_rx_filter(struct vioif_softc *sc)
1879 {
1880 	struct ethercom *ec = &sc->sc_ethercom;
1881 	struct ifnet *ifp = &ec->ec_if;
1882 	struct ether_multi *enm;
1883 	struct ether_multistep step;
1884 	struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
1885 	int nentries;
1886 	int promisc = 0, allmulti = 0, rxfilter = 0;
1887 	int r;
1888 
1889 	if (!sc->sc_has_ctrl) {	/* no ctrl vq; always promisc */
1890 		ifp->if_flags |= IFF_PROMISC;
1891 		return 0;
1892 	}
1893 
1894 	if (ifp->if_flags & IFF_PROMISC) {
1895 		promisc = 1;
1896 		goto set;
1897 	}
1898 
1899 	nentries = -1;
1900 	ETHER_LOCK(ec);
1901 	ETHER_FIRST_MULTI(step, ec, enm);
1902 	while (nentries++, enm != NULL) {
1903 		if (nentries >= VIRTIO_NET_CTRL_MAC_MAXENTRIES) {
1904 			allmulti = 1;
1905 			goto set_unlock;
1906 		}
1907 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1908 			allmulti = 1;
1909 			goto set_unlock;
1910 		}
1911 		memcpy(ctrlq->ctrlq_mac_tbl_mc->macs[nentries],
1912 		    enm->enm_addrlo, ETHER_ADDR_LEN);
1913 		ETHER_NEXT_MULTI(step, enm);
1914 	}
1915 	rxfilter = 1;
1916 
1917 set_unlock:
1918 	ETHER_UNLOCK(ec);
1919 
1920 set:
1921 	if (rxfilter) {
1922 		ctrlq->ctrlq_mac_tbl_uc->nentries = 0;
1923 		ctrlq->ctrlq_mac_tbl_mc->nentries = nentries;
1924 		r = vioif_set_rx_filter(sc);
1925 		if (r != 0) {
1926 			rxfilter = 0;
1927 			allmulti = 1; /* fallback */
1928 		}
1929 	} else {
1930 		/* remove rx filter */
1931 		ctrlq->ctrlq_mac_tbl_uc->nentries = 0;
1932 		ctrlq->ctrlq_mac_tbl_mc->nentries = 0;
1933 		r = vioif_set_rx_filter(sc);
1934 		/* what to do on failure? */
1935 	}
1936 	if (allmulti) {
1937 		r = vioif_set_allmulti(sc, true);
1938 		if (r != 0) {
1939 			allmulti = 0;
1940 			promisc = 1; /* fallback */
1941 		}
1942 	} else {
1943 		r = vioif_set_allmulti(sc, false);
1944 		/* what to do on failure? */
1945 	}
1946 	if (promisc) {
1947 		r = vioif_set_promisc(sc, true);
1948 	} else {
1949 		r = vioif_set_promisc(sc, false);
1950 	}
1951 
1952 	return r;
1953 }
1954 
1955 static bool
1956 vioif_is_link_up(struct vioif_softc *sc)
1957 {
1958 	struct virtio_softc *vsc = sc->sc_virtio;
1959 	uint16_t status;
1960 
1961 	if (virtio_features(vsc) & VIRTIO_NET_F_STATUS)
1962 		status = virtio_read_device_config_2(vsc,
1963 		    VIRTIO_NET_CONFIG_STATUS);
1964 	else
1965 		status = VIRTIO_NET_S_LINK_UP;
1966 
1967 	return ((status & VIRTIO_NET_S_LINK_UP) != 0);
1968 }
1969 
1970 /* change link status */
1971 static void
1972 vioif_update_link_status(struct vioif_softc *sc)
1973 {
1974 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1975 	struct vioif_txqueue *txq;
1976 	bool active, changed;
1977 	int link, i;
1978 
1979 	active = vioif_is_link_up(sc);
1980 	changed = false;
1981 
1982 	if (active) {
1983 		if (!sc->sc_link_active)
1984 			changed = true;
1985 
1986 		link = LINK_STATE_UP;
1987 		sc->sc_link_active = true;
1988 	} else {
1989 		if (sc->sc_link_active)
1990 			changed = true;
1991 
1992 		link = LINK_STATE_DOWN;
1993 		sc->sc_link_active = false;
1994 	}
1995 
1996 	if (changed) {
1997 		for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
1998 			txq = &sc->sc_txq[i];
1999 
2000 			mutex_enter(txq->txq_lock);
2001 			txq->txq_link_active = sc->sc_link_active;
2002 			mutex_exit(txq->txq_lock);
2003 		}
2004 
2005 		if_link_state_change(ifp, link);
2006 	}
2007 }
2008 
2009 static int
2010 vioif_config_change(struct virtio_softc *vsc)
2011 {
2012 	struct vioif_softc *sc = device_private(virtio_child(vsc));
2013 
2014 #ifdef VIOIF_SOFTINT_INTR
2015 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2016 #endif
2017 
2018 #ifdef VIOIF_SOFTINT_INTR
2019 	KASSERT(!cpu_intr_p());
2020 	vioif_update_link_status(sc);
2021 	vioif_start(ifp);
2022 #else
2023 	softint_schedule(sc->sc_ctl_softint);
2024 #endif
2025 
2026 	return 0;
2027 }
2028 
2029 static void
2030 vioif_ctl_softint(void *arg)
2031 {
2032 	struct vioif_softc *sc = arg;
2033 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2034 
2035 	vioif_update_link_status(sc);
2036 	vioif_start(ifp);
2037 }
2038 
2039 MODULE(MODULE_CLASS_DRIVER, if_vioif, "virtio");
2040 
2041 #ifdef _MODULE
2042 #include "ioconf.c"
2043 #endif
2044 
2045 static int
2046 if_vioif_modcmd(modcmd_t cmd, void *opaque)
2047 {
2048 	int error = 0;
2049 
2050 #ifdef _MODULE
2051 	switch (cmd) {
2052 	case MODULE_CMD_INIT:
2053 		error = config_init_component(cfdriver_ioconf_if_vioif,
2054 		    cfattach_ioconf_if_vioif, cfdata_ioconf_if_vioif);
2055 		break;
2056 	case MODULE_CMD_FINI:
2057 		error = config_fini_component(cfdriver_ioconf_if_vioif,
2058 		    cfattach_ioconf_if_vioif, cfdata_ioconf_if_vioif);
2059 		break;
2060 	default:
2061 		error = ENOTTY;
2062 		break;
2063 	}
2064 #endif
2065 
2066 	return error;
2067 }
2068