xref: /netbsd-src/sys/dev/pci/if_vioif.c (revision c29d51755812ace2e87aeefdb06cb2b4dac7087a)
1 /*	$NetBSD: if_vioif.c,v 1.16 2015/05/05 10:56:13 ozaki-r Exp $	*/
2 
3 /*
4  * Copyright (c) 2010 Minoura Makoto.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __KERNEL_RCSID(0, "$NetBSD: if_vioif.c,v 1.16 2015/05/05 10:56:13 ozaki-r 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 
48 #include <dev/pci/pcidevs.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/virtioreg.h>
52 #include <dev/pci/virtiovar.h>
53 
54 #include <net/if.h>
55 #include <net/if_media.h>
56 #include <net/if_ether.h>
57 
58 #include <net/bpf.h>
59 
60 
61 #ifdef NET_MPSAFE
62 #define VIOIF_MPSAFE	1
63 #endif
64 
65 #ifdef SOFTINT_INTR
66 #define VIOIF_SOFTINT_INTR	1
67 #endif
68 
69 /*
70  * if_vioifreg.h:
71  */
72 /* Configuration registers */
73 #define VIRTIO_NET_CONFIG_MAC		0 /* 8bit x 6byte */
74 #define VIRTIO_NET_CONFIG_STATUS	6 /* 16bit */
75 
76 /* Feature bits */
77 #define VIRTIO_NET_F_CSUM	(1<<0)
78 #define VIRTIO_NET_F_GUEST_CSUM	(1<<1)
79 #define VIRTIO_NET_F_MAC	(1<<5)
80 #define VIRTIO_NET_F_GSO	(1<<6)
81 #define VIRTIO_NET_F_GUEST_TSO4	(1<<7)
82 #define VIRTIO_NET_F_GUEST_TSO6	(1<<8)
83 #define VIRTIO_NET_F_GUEST_ECN	(1<<9)
84 #define VIRTIO_NET_F_GUEST_UFO	(1<<10)
85 #define VIRTIO_NET_F_HOST_TSO4	(1<<11)
86 #define VIRTIO_NET_F_HOST_TSO6	(1<<12)
87 #define VIRTIO_NET_F_HOST_ECN	(1<<13)
88 #define VIRTIO_NET_F_HOST_UFO	(1<<14)
89 #define VIRTIO_NET_F_MRG_RXBUF	(1<<15)
90 #define VIRTIO_NET_F_STATUS	(1<<16)
91 #define VIRTIO_NET_F_CTRL_VQ	(1<<17)
92 #define VIRTIO_NET_F_CTRL_RX	(1<<18)
93 #define VIRTIO_NET_F_CTRL_VLAN	(1<<19)
94 
95 /* Status */
96 #define VIRTIO_NET_S_LINK_UP	1
97 
98 /* Packet header structure */
99 struct virtio_net_hdr {
100 	uint8_t		flags;
101 	uint8_t		gso_type;
102 	uint16_t	hdr_len;
103 	uint16_t	gso_size;
104 	uint16_t	csum_start;
105 	uint16_t	csum_offset;
106 #if 0
107 	uint16_t	num_buffers; /* if VIRTIO_NET_F_MRG_RXBUF enabled */
108 #endif
109 } __packed;
110 
111 #define VIRTIO_NET_HDR_F_NEEDS_CSUM	1 /* flags */
112 #define VIRTIO_NET_HDR_GSO_NONE		0 /* gso_type */
113 #define VIRTIO_NET_HDR_GSO_TCPV4	1 /* gso_type */
114 #define VIRTIO_NET_HDR_GSO_UDP		3 /* gso_type */
115 #define VIRTIO_NET_HDR_GSO_TCPV6	4 /* gso_type */
116 #define VIRTIO_NET_HDR_GSO_ECN		0x80 /* gso_type, |'ed */
117 
118 #define VIRTIO_NET_MAX_GSO_LEN		(65536+ETHER_HDR_LEN)
119 
120 /* Control virtqueue */
121 struct virtio_net_ctrl_cmd {
122 	uint8_t	class;
123 	uint8_t	command;
124 } __packed;
125 #define VIRTIO_NET_CTRL_RX		0
126 # define VIRTIO_NET_CTRL_RX_PROMISC	0
127 # define VIRTIO_NET_CTRL_RX_ALLMULTI	1
128 
129 #define VIRTIO_NET_CTRL_MAC		1
130 # define VIRTIO_NET_CTRL_MAC_TABLE_SET	0
131 
132 #define VIRTIO_NET_CTRL_VLAN		2
133 # define VIRTIO_NET_CTRL_VLAN_ADD	0
134 # define VIRTIO_NET_CTRL_VLAN_DEL	1
135 
136 struct virtio_net_ctrl_status {
137 	uint8_t	ack;
138 } __packed;
139 #define VIRTIO_NET_OK			0
140 #define VIRTIO_NET_ERR			1
141 
142 struct virtio_net_ctrl_rx {
143 	uint8_t	onoff;
144 } __packed;
145 
146 struct virtio_net_ctrl_mac_tbl {
147 	uint32_t nentries;
148 	uint8_t macs[][ETHER_ADDR_LEN];
149 } __packed;
150 
151 struct virtio_net_ctrl_vlan {
152 	uint16_t id;
153 } __packed;
154 
155 
156 /*
157  * if_vioifvar.h:
158  */
159 struct vioif_softc {
160 	device_t		sc_dev;
161 
162 	struct virtio_softc	*sc_virtio;
163 	struct virtqueue	sc_vq[3];
164 
165 	uint8_t			sc_mac[ETHER_ADDR_LEN];
166 	struct ethercom		sc_ethercom;
167 	short			sc_deferred_init_done;
168 
169 	/* bus_dmamem */
170 	bus_dma_segment_t	sc_hdr_segs[1];
171 	struct virtio_net_hdr	*sc_hdrs;
172 #define sc_rx_hdrs	sc_hdrs
173 	struct virtio_net_hdr	*sc_tx_hdrs;
174 	struct virtio_net_ctrl_cmd *sc_ctrl_cmd;
175 	struct virtio_net_ctrl_status *sc_ctrl_status;
176 	struct virtio_net_ctrl_rx *sc_ctrl_rx;
177 	struct virtio_net_ctrl_mac_tbl *sc_ctrl_mac_tbl_uc;
178 	struct virtio_net_ctrl_mac_tbl *sc_ctrl_mac_tbl_mc;
179 
180 	/* kmem */
181 	bus_dmamap_t		*sc_arrays;
182 #define sc_rxhdr_dmamaps sc_arrays
183 	bus_dmamap_t		*sc_txhdr_dmamaps;
184 	bus_dmamap_t		*sc_rx_dmamaps;
185 	bus_dmamap_t		*sc_tx_dmamaps;
186 	struct mbuf		**sc_rx_mbufs;
187 	struct mbuf		**sc_tx_mbufs;
188 
189 	bus_dmamap_t		sc_ctrl_cmd_dmamap;
190 	bus_dmamap_t		sc_ctrl_status_dmamap;
191 	bus_dmamap_t		sc_ctrl_rx_dmamap;
192 	bus_dmamap_t		sc_ctrl_tbl_uc_dmamap;
193 	bus_dmamap_t		sc_ctrl_tbl_mc_dmamap;
194 
195 	void			*sc_rx_softint;
196 
197 	enum {
198 		FREE, INUSE, DONE
199 	}			sc_ctrl_inuse;
200 	kcondvar_t		sc_ctrl_wait;
201 	kmutex_t		sc_ctrl_wait_lock;
202 	kmutex_t		*sc_tx_lock;
203 	kmutex_t		*sc_rx_lock;
204 	bool			sc_stopping;
205 };
206 #define VIRTIO_NET_TX_MAXNSEGS		(16) /* XXX */
207 #define VIRTIO_NET_CTRL_MAC_MAXENTRIES	(64) /* XXX */
208 
209 #define VIOIF_TX_LOCK(_sc)	if ((_sc)->sc_tx_lock) mutex_enter((_sc)->sc_tx_lock)
210 #define VIOIF_TX_UNLOCK(_sc)	if ((_sc)->sc_tx_lock) mutex_exit((_sc)->sc_tx_lock)
211 #define VIOIF_TX_LOCKED(_sc)	(!(_sc)->sc_tx_lock || mutex_owned((_sc)->sc_tx_lock))
212 #define VIOIF_RX_LOCK(_sc)	if ((_sc)->sc_rx_lock) mutex_enter((_sc)->sc_rx_lock)
213 #define VIOIF_RX_UNLOCK(_sc)	if ((_sc)->sc_rx_lock) mutex_exit((_sc)->sc_rx_lock)
214 #define VIOIF_RX_LOCKED(_sc)	(!(_sc)->sc_rx_lock || mutex_owned((_sc)->sc_rx_lock))
215 
216 /* cfattach interface functions */
217 static int	vioif_match(device_t, cfdata_t, void *);
218 static void	vioif_attach(device_t, device_t, void *);
219 static void	vioif_deferred_init(device_t);
220 
221 /* ifnet interface functions */
222 static int	vioif_init(struct ifnet *);
223 static void	vioif_stop(struct ifnet *, int);
224 static void	vioif_start(struct ifnet *);
225 static int	vioif_ioctl(struct ifnet *, u_long, void *);
226 static void	vioif_watchdog(struct ifnet *);
227 
228 /* rx */
229 static int	vioif_add_rx_mbuf(struct vioif_softc *, int);
230 static void	vioif_free_rx_mbuf(struct vioif_softc *, int);
231 static void	vioif_populate_rx_mbufs(struct vioif_softc *);
232 static void	vioif_populate_rx_mbufs_locked(struct vioif_softc *);
233 static int	vioif_rx_deq(struct vioif_softc *);
234 static int	vioif_rx_deq_locked(struct vioif_softc *);
235 static int	vioif_rx_vq_done(struct virtqueue *);
236 static void	vioif_rx_softint(void *);
237 static void	vioif_rx_drain(struct vioif_softc *);
238 
239 /* tx */
240 static int	vioif_tx_vq_done(struct virtqueue *);
241 static int	vioif_tx_vq_done_locked(struct virtqueue *);
242 static void	vioif_tx_drain(struct vioif_softc *);
243 
244 /* other control */
245 static int	vioif_updown(struct vioif_softc *, bool);
246 static int	vioif_ctrl_rx(struct vioif_softc *, int, bool);
247 static int	vioif_set_promisc(struct vioif_softc *, bool);
248 static int	vioif_set_allmulti(struct vioif_softc *, bool);
249 static int	vioif_set_rx_filter(struct vioif_softc *);
250 static int	vioif_rx_filter(struct vioif_softc *);
251 static int	vioif_ctrl_vq_done(struct virtqueue *);
252 
253 CFATTACH_DECL_NEW(vioif, sizeof(struct vioif_softc),
254 		  vioif_match, vioif_attach, NULL, NULL);
255 
256 static int
257 vioif_match(device_t parent, cfdata_t match, void *aux)
258 {
259 	struct virtio_softc *va = aux;
260 
261 	if (va->sc_childdevid == PCI_PRODUCT_VIRTIO_NETWORK)
262 		return 1;
263 
264 	return 0;
265 }
266 
267 /* allocate memory */
268 /*
269  * dma memory is used for:
270  *   sc_rx_hdrs[slot]:	 metadata array for recieved frames (READ)
271  *   sc_tx_hdrs[slot]:	 metadata array for frames to be sent (WRITE)
272  *   sc_ctrl_cmd:	 command to be sent via ctrl vq (WRITE)
273  *   sc_ctrl_status:	 return value for a command via ctrl vq (READ)
274  *   sc_ctrl_rx:	 parameter for a VIRTIO_NET_CTRL_RX class command
275  *			 (WRITE)
276  *   sc_ctrl_mac_tbl_uc: unicast MAC address filter for a VIRTIO_NET_CTRL_MAC
277  *			 class command (WRITE)
278  *   sc_ctrl_mac_tbl_mc: multicast MAC address filter for a VIRTIO_NET_CTRL_MAC
279  *			 class command (WRITE)
280  * sc_ctrl_* structures are allocated only one each; they are protected by
281  * sc_ctrl_inuse variable and sc_ctrl_wait condvar.
282  */
283 /*
284  * dynamically allocated memory is used for:
285  *   sc_rxhdr_dmamaps[slot]:	bus_dmamap_t array for sc_rx_hdrs[slot]
286  *   sc_txhdr_dmamaps[slot]:	bus_dmamap_t array for sc_tx_hdrs[slot]
287  *   sc_rx_dmamaps[slot]:	bus_dmamap_t array for recieved payload
288  *   sc_tx_dmamaps[slot]:	bus_dmamap_t array for sent payload
289  *   sc_rx_mbufs[slot]:		mbuf pointer array for recieved frames
290  *   sc_tx_mbufs[slot]:		mbuf pointer array for sent frames
291  */
292 static int
293 vioif_alloc_mems(struct vioif_softc *sc)
294 {
295 	struct virtio_softc *vsc = sc->sc_virtio;
296 	int allocsize, allocsize2, r, rsegs, i;
297 	void *vaddr;
298 	intptr_t p;
299 	int rxqsize, txqsize;
300 
301 	rxqsize = vsc->sc_vqs[0].vq_num;
302 	txqsize = vsc->sc_vqs[1].vq_num;
303 
304 	allocsize = sizeof(struct virtio_net_hdr) * rxqsize;
305 	allocsize += sizeof(struct virtio_net_hdr) * txqsize;
306 	if (vsc->sc_nvqs == 3) {
307 		allocsize += sizeof(struct virtio_net_ctrl_cmd) * 1;
308 		allocsize += sizeof(struct virtio_net_ctrl_status) * 1;
309 		allocsize += sizeof(struct virtio_net_ctrl_rx) * 1;
310 		allocsize += sizeof(struct virtio_net_ctrl_mac_tbl)
311 			+ sizeof(struct virtio_net_ctrl_mac_tbl)
312 			+ ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES;
313 	}
314 	r = bus_dmamem_alloc(vsc->sc_dmat, allocsize, 0, 0,
315 			     &sc->sc_hdr_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
316 	if (r != 0) {
317 		aprint_error_dev(sc->sc_dev,
318 				 "DMA memory allocation failed, size %d, "
319 				 "error code %d\n", allocsize, r);
320 		goto err_none;
321 	}
322 	r = bus_dmamem_map(vsc->sc_dmat,
323 			   &sc->sc_hdr_segs[0], 1, allocsize,
324 			   &vaddr, BUS_DMA_NOWAIT);
325 	if (r != 0) {
326 		aprint_error_dev(sc->sc_dev,
327 				 "DMA memory map failed, "
328 				 "error code %d\n", r);
329 		goto err_dmamem_alloc;
330 	}
331 	sc->sc_hdrs = vaddr;
332 	memset(vaddr, 0, allocsize);
333 	p = (intptr_t) vaddr;
334 	p += sizeof(struct virtio_net_hdr) * rxqsize;
335 #define P(name,size)	do { sc->sc_ ##name = (void*) p;	\
336 			     p += size; } while (0)
337 	P(tx_hdrs, sizeof(struct virtio_net_hdr) * txqsize);
338 	if (vsc->sc_nvqs == 3) {
339 		P(ctrl_cmd, sizeof(struct virtio_net_ctrl_cmd));
340 		P(ctrl_status, sizeof(struct virtio_net_ctrl_status));
341 		P(ctrl_rx, sizeof(struct virtio_net_ctrl_rx));
342 		P(ctrl_mac_tbl_uc, sizeof(struct virtio_net_ctrl_mac_tbl));
343 		P(ctrl_mac_tbl_mc,
344 		  (sizeof(struct virtio_net_ctrl_mac_tbl)
345 		   + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES));
346 	}
347 #undef P
348 
349 	allocsize2 = sizeof(bus_dmamap_t) * (rxqsize + txqsize);
350 	allocsize2 += sizeof(bus_dmamap_t) * (rxqsize + txqsize);
351 	allocsize2 += sizeof(struct mbuf*) * (rxqsize + txqsize);
352 	sc->sc_arrays = kmem_zalloc(allocsize2, KM_SLEEP);
353 	if (sc->sc_arrays == NULL)
354 		goto err_dmamem_map;
355 	sc->sc_txhdr_dmamaps = sc->sc_arrays + rxqsize;
356 	sc->sc_rx_dmamaps = sc->sc_txhdr_dmamaps + txqsize;
357 	sc->sc_tx_dmamaps = sc->sc_rx_dmamaps + rxqsize;
358 	sc->sc_rx_mbufs = (void*) (sc->sc_tx_dmamaps + txqsize);
359 	sc->sc_tx_mbufs = sc->sc_rx_mbufs + rxqsize;
360 
361 #define C(map, buf, size, nsegs, rw, usage)				\
362 	do {								\
363 		r = bus_dmamap_create(vsc->sc_dmat, size, nsegs, size, 0, \
364 				      BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW,	\
365 				      &sc->sc_ ##map);			\
366 		if (r != 0) {						\
367 			aprint_error_dev(sc->sc_dev,			\
368 					 usage " dmamap creation failed, " \
369 					 "error code %d\n", r);		\
370 					 goto err_reqs;			\
371 		}							\
372 	} while (0)
373 #define C_L1(map, buf, size, nsegs, rw, usage)				\
374 	C(map, buf, size, nsegs, rw, usage);				\
375 	do {								\
376 		r = bus_dmamap_load(vsc->sc_dmat, sc->sc_ ##map,	\
377 				    &sc->sc_ ##buf, size, NULL,		\
378 				    BUS_DMA_ ##rw | BUS_DMA_NOWAIT);	\
379 		if (r != 0) {						\
380 			aprint_error_dev(sc->sc_dev,			\
381 					 usage " dmamap load failed, "	\
382 					 "error code %d\n", r);		\
383 			goto err_reqs;					\
384 		}							\
385 	} while (0)
386 #define C_L2(map, buf, size, nsegs, rw, usage)				\
387 	C(map, buf, size, nsegs, rw, usage);				\
388 	do {								\
389 		r = bus_dmamap_load(vsc->sc_dmat, sc->sc_ ##map,	\
390 				    sc->sc_ ##buf, size, NULL,		\
391 				    BUS_DMA_ ##rw | BUS_DMA_NOWAIT);	\
392 		if (r != 0) {						\
393 			aprint_error_dev(sc->sc_dev,			\
394 					 usage " dmamap load failed, "	\
395 					 "error code %d\n", r);		\
396 			goto err_reqs;					\
397 		}							\
398 	} while (0)
399 	for (i = 0; i < rxqsize; i++) {
400 		C_L1(rxhdr_dmamaps[i], rx_hdrs[i],
401 		    sizeof(struct virtio_net_hdr), 1,
402 		    READ, "rx header");
403 		C(rx_dmamaps[i], NULL, MCLBYTES, 1, 0, "rx payload");
404 	}
405 
406 	for (i = 0; i < txqsize; i++) {
407 		C_L1(txhdr_dmamaps[i], rx_hdrs[i],
408 		    sizeof(struct virtio_net_hdr), 1,
409 		    WRITE, "tx header");
410 		C(tx_dmamaps[i], NULL, ETHER_MAX_LEN, 256 /* XXX */, 0,
411 		  "tx payload");
412 	}
413 
414 	if (vsc->sc_nvqs == 3) {
415 		/* control vq class & command */
416 		C_L2(ctrl_cmd_dmamap, ctrl_cmd,
417 		    sizeof(struct virtio_net_ctrl_cmd), 1, WRITE,
418 		    "control command");
419 
420 		/* control vq status */
421 		C_L2(ctrl_status_dmamap, ctrl_status,
422 		    sizeof(struct virtio_net_ctrl_status), 1, READ,
423 		    "control status");
424 
425 		/* control vq rx mode command parameter */
426 		C_L2(ctrl_rx_dmamap, ctrl_rx,
427 		    sizeof(struct virtio_net_ctrl_rx), 1, WRITE,
428 		    "rx mode control command");
429 
430 		/* control vq MAC filter table for unicast */
431 		/* do not load now since its length is variable */
432 		C(ctrl_tbl_uc_dmamap, NULL,
433 		  sizeof(struct virtio_net_ctrl_mac_tbl) + 0, 1, WRITE,
434 		  "unicast MAC address filter command");
435 
436 		/* control vq MAC filter table for multicast */
437 		C(ctrl_tbl_mc_dmamap, NULL,
438 		  (sizeof(struct virtio_net_ctrl_mac_tbl)
439 		   + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES),
440 		  1, WRITE, "multicast MAC address filter command");
441 	}
442 #undef C_L2
443 #undef C_L1
444 #undef C
445 
446 	return 0;
447 
448 err_reqs:
449 #define D(map)								\
450 	do {								\
451 		if (sc->sc_ ##map) {					\
452 			bus_dmamap_destroy(vsc->sc_dmat, sc->sc_ ##map); \
453 			sc->sc_ ##map = NULL;				\
454 		}							\
455 	} while (0)
456 	D(ctrl_tbl_mc_dmamap);
457 	D(ctrl_tbl_uc_dmamap);
458 	D(ctrl_rx_dmamap);
459 	D(ctrl_status_dmamap);
460 	D(ctrl_cmd_dmamap);
461 	for (i = 0; i < txqsize; i++) {
462 		D(tx_dmamaps[i]);
463 		D(txhdr_dmamaps[i]);
464 	}
465 	for (i = 0; i < rxqsize; i++) {
466 		D(rx_dmamaps[i]);
467 		D(rxhdr_dmamaps[i]);
468 	}
469 #undef D
470 	if (sc->sc_arrays) {
471 		kmem_free(sc->sc_arrays, allocsize2);
472 		sc->sc_arrays = 0;
473 	}
474 err_dmamem_map:
475 	bus_dmamem_unmap(vsc->sc_dmat, sc->sc_hdrs, allocsize);
476 err_dmamem_alloc:
477 	bus_dmamem_free(vsc->sc_dmat, &sc->sc_hdr_segs[0], 1);
478 err_none:
479 	return -1;
480 }
481 
482 static void
483 vioif_attach(device_t parent, device_t self, void *aux)
484 {
485 	struct vioif_softc *sc = device_private(self);
486 	struct virtio_softc *vsc = device_private(parent);
487 	uint32_t features;
488 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
489 	u_int flags;
490 
491 	if (vsc->sc_child != NULL) {
492 		aprint_normal(": child already attached for %s; "
493 			      "something wrong...\n",
494 			      device_xname(parent));
495 		return;
496 	}
497 
498 	sc->sc_dev = self;
499 	sc->sc_virtio = vsc;
500 
501 	vsc->sc_child = self;
502 	vsc->sc_ipl = IPL_NET;
503 	vsc->sc_vqs = &sc->sc_vq[0];
504 	vsc->sc_config_change = NULL;
505 	vsc->sc_intrhand = virtio_vq_intr;
506 	vsc->sc_flags = 0;
507 
508 #ifdef VIOIF_MPSAFE
509 	vsc->sc_flags |= VIRTIO_F_PCI_INTR_MPSAFE;
510 #endif
511 #ifdef VIOIF_SOFTINT_INTR
512 	vsc->sc_flags |= VIRTIO_F_PCI_INTR_SOFTINT;
513 #endif
514 
515 	features = virtio_negotiate_features(vsc,
516 					     (VIRTIO_NET_F_MAC |
517 					      VIRTIO_NET_F_STATUS |
518 					      VIRTIO_NET_F_CTRL_VQ |
519 					      VIRTIO_NET_F_CTRL_RX |
520 					      VIRTIO_F_NOTIFY_ON_EMPTY));
521 	if (features & VIRTIO_NET_F_MAC) {
522 		sc->sc_mac[0] = virtio_read_device_config_1(vsc,
523 						    VIRTIO_NET_CONFIG_MAC+0);
524 		sc->sc_mac[1] = virtio_read_device_config_1(vsc,
525 						    VIRTIO_NET_CONFIG_MAC+1);
526 		sc->sc_mac[2] = virtio_read_device_config_1(vsc,
527 						    VIRTIO_NET_CONFIG_MAC+2);
528 		sc->sc_mac[3] = virtio_read_device_config_1(vsc,
529 						    VIRTIO_NET_CONFIG_MAC+3);
530 		sc->sc_mac[4] = virtio_read_device_config_1(vsc,
531 						    VIRTIO_NET_CONFIG_MAC+4);
532 		sc->sc_mac[5] = virtio_read_device_config_1(vsc,
533 						    VIRTIO_NET_CONFIG_MAC+5);
534 	} else {
535 		/* code stolen from sys/net/if_tap.c */
536 		struct timeval tv;
537 		uint32_t ui;
538 		getmicrouptime(&tv);
539 		ui = (tv.tv_sec ^ tv.tv_usec) & 0xffffff;
540 		memcpy(sc->sc_mac+3, (uint8_t *)&ui, 3);
541 		virtio_write_device_config_1(vsc,
542 					     VIRTIO_NET_CONFIG_MAC+0,
543 					     sc->sc_mac[0]);
544 		virtio_write_device_config_1(vsc,
545 					     VIRTIO_NET_CONFIG_MAC+1,
546 					     sc->sc_mac[1]);
547 		virtio_write_device_config_1(vsc,
548 					     VIRTIO_NET_CONFIG_MAC+2,
549 					     sc->sc_mac[2]);
550 		virtio_write_device_config_1(vsc,
551 					     VIRTIO_NET_CONFIG_MAC+3,
552 					     sc->sc_mac[3]);
553 		virtio_write_device_config_1(vsc,
554 					     VIRTIO_NET_CONFIG_MAC+4,
555 					     sc->sc_mac[4]);
556 		virtio_write_device_config_1(vsc,
557 					     VIRTIO_NET_CONFIG_MAC+5,
558 					     sc->sc_mac[5]);
559 	}
560 	aprint_normal(": Ethernet address %s\n", ether_sprintf(sc->sc_mac));
561 	aprint_naive("\n");
562 
563 	if (virtio_alloc_vq(vsc, &sc->sc_vq[0], 0,
564 			    MCLBYTES+sizeof(struct virtio_net_hdr), 2,
565 			    "rx") != 0) {
566 		goto err;
567 	}
568 	vsc->sc_nvqs = 1;
569 	sc->sc_vq[0].vq_done = vioif_rx_vq_done;
570 	if (virtio_alloc_vq(vsc, &sc->sc_vq[1], 1,
571 			    (sizeof(struct virtio_net_hdr)
572 			     + (ETHER_MAX_LEN - ETHER_HDR_LEN)),
573 			    VIRTIO_NET_TX_MAXNSEGS + 1,
574 			    "tx") != 0) {
575 		goto err;
576 	}
577 
578 #ifdef VIOIF_MPSAFE
579 	sc->sc_tx_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
580 	sc->sc_rx_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
581 #else
582 	sc->sc_tx_lock = NULL;
583 	sc->sc_rx_lock = NULL;
584 #endif
585 	sc->sc_stopping = false;
586 
587 	vsc->sc_nvqs = 2;
588 	sc->sc_vq[1].vq_done = vioif_tx_vq_done;
589 	virtio_start_vq_intr(vsc, &sc->sc_vq[0]);
590 	virtio_stop_vq_intr(vsc, &sc->sc_vq[1]); /* not urgent; do it later */
591 	if ((features & VIRTIO_NET_F_CTRL_VQ)
592 	    && (features & VIRTIO_NET_F_CTRL_RX)) {
593 		if (virtio_alloc_vq(vsc, &sc->sc_vq[2], 2,
594 				    NBPG, 1, "control") == 0) {
595 			sc->sc_vq[2].vq_done = vioif_ctrl_vq_done;
596 			cv_init(&sc->sc_ctrl_wait, "ctrl_vq");
597 			mutex_init(&sc->sc_ctrl_wait_lock,
598 				   MUTEX_DEFAULT, IPL_NET);
599 			sc->sc_ctrl_inuse = FREE;
600 			virtio_start_vq_intr(vsc, &sc->sc_vq[2]);
601 			vsc->sc_nvqs = 3;
602 		}
603 	}
604 
605 #ifdef VIOIF_MPSAFE
606 	flags = SOFTINT_NET | SOFTINT_MPSAFE;
607 #else
608 	flags = SOFTINT_NET;
609 #endif
610 	sc->sc_rx_softint = softint_establish(flags, vioif_rx_softint, sc);
611 	if (sc->sc_rx_softint == NULL) {
612 		aprint_error_dev(self, "cannot establish softint\n");
613 		goto err;
614 	}
615 
616 	if (vioif_alloc_mems(sc) < 0)
617 		goto err;
618 
619 	strlcpy(ifp->if_xname, device_xname(self), IFNAMSIZ);
620 	ifp->if_softc = sc;
621 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
622 	ifp->if_start = vioif_start;
623 	ifp->if_ioctl = vioif_ioctl;
624 	ifp->if_init = vioif_init;
625 	ifp->if_stop = vioif_stop;
626 	ifp->if_capabilities = 0;
627 	ifp->if_watchdog = vioif_watchdog;
628 
629 	sc->sc_ethercom.ec_capabilities |= ETHERCAP_VLAN_MTU;
630 
631 	if_attach(ifp);
632 	ether_ifattach(ifp, sc->sc_mac);
633 
634 	return;
635 
636 err:
637 	if (sc->sc_tx_lock)
638 		mutex_obj_free(sc->sc_tx_lock);
639 	if (sc->sc_rx_lock)
640 		mutex_obj_free(sc->sc_rx_lock);
641 
642 	if (vsc->sc_nvqs == 3) {
643 		virtio_free_vq(vsc, &sc->sc_vq[2]);
644 		cv_destroy(&sc->sc_ctrl_wait);
645 		mutex_destroy(&sc->sc_ctrl_wait_lock);
646 		vsc->sc_nvqs = 2;
647 	}
648 	if (vsc->sc_nvqs == 2) {
649 		virtio_free_vq(vsc, &sc->sc_vq[1]);
650 		vsc->sc_nvqs = 1;
651 	}
652 	if (vsc->sc_nvqs == 1) {
653 		virtio_free_vq(vsc, &sc->sc_vq[0]);
654 		vsc->sc_nvqs = 0;
655 	}
656 	vsc->sc_child = (void*)1;
657 	return;
658 }
659 
660 /* we need interrupts to make promiscuous mode off */
661 static void
662 vioif_deferred_init(device_t self)
663 {
664 	struct vioif_softc *sc = device_private(self);
665 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
666 	int r;
667 
668 	if (ifp->if_flags & IFF_PROMISC)
669 		return;
670 
671 	r =  vioif_set_promisc(sc, false);
672 	if (r != 0)
673 		aprint_error_dev(self, "resetting promisc mode failed, "
674 				 "errror code %d\n", r);
675 }
676 
677 /*
678  * Interface functions for ifnet
679  */
680 static int
681 vioif_init(struct ifnet *ifp)
682 {
683 	struct vioif_softc *sc = ifp->if_softc;
684 
685 	vioif_stop(ifp, 0);
686 
687 	if (!sc->sc_deferred_init_done) {
688 		struct virtio_softc *vsc = sc->sc_virtio;
689 
690 		sc->sc_deferred_init_done = 1;
691 		if (vsc->sc_nvqs == 3)
692 			vioif_deferred_init(sc->sc_dev);
693 	}
694 
695 	/* Have to set false before vioif_populate_rx_mbufs */
696 	sc->sc_stopping = false;
697 
698 	vioif_populate_rx_mbufs(sc);
699 
700 	vioif_updown(sc, true);
701 	ifp->if_flags |= IFF_RUNNING;
702 	ifp->if_flags &= ~IFF_OACTIVE;
703 	vioif_rx_filter(sc);
704 
705 	return 0;
706 }
707 
708 static void
709 vioif_stop(struct ifnet *ifp, int disable)
710 {
711 	struct vioif_softc *sc = ifp->if_softc;
712 	struct virtio_softc *vsc = sc->sc_virtio;
713 
714 	/* Take the locks to ensure that ongoing TX/RX finish */
715 	VIOIF_TX_LOCK(sc);
716 	VIOIF_RX_LOCK(sc);
717 	sc->sc_stopping = true;
718 	VIOIF_RX_UNLOCK(sc);
719 	VIOIF_TX_UNLOCK(sc);
720 
721 	/* only way to stop I/O and DMA is resetting... */
722 	virtio_reset(vsc);
723 	vioif_rx_deq(sc);
724 	vioif_tx_drain(sc);
725 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
726 
727 	if (disable)
728 		vioif_rx_drain(sc);
729 
730 	virtio_reinit_start(vsc);
731 	virtio_negotiate_features(vsc, vsc->sc_features);
732 	virtio_start_vq_intr(vsc, &sc->sc_vq[0]);
733 	virtio_stop_vq_intr(vsc, &sc->sc_vq[1]);
734 	if (vsc->sc_nvqs >= 3)
735 		virtio_start_vq_intr(vsc, &sc->sc_vq[2]);
736 	virtio_reinit_end(vsc);
737 	vioif_updown(sc, false);
738 }
739 
740 static void
741 vioif_start(struct ifnet *ifp)
742 {
743 	struct vioif_softc *sc = ifp->if_softc;
744 	struct virtio_softc *vsc = sc->sc_virtio;
745 	struct virtqueue *vq = &sc->sc_vq[1]; /* tx vq */
746 	struct mbuf *m;
747 	int queued = 0, retry = 0;
748 
749 	VIOIF_TX_LOCK(sc);
750 
751 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
752 		goto out;
753 
754 	if (sc->sc_stopping)
755 		goto out;
756 
757 	for (;;) {
758 		int slot, r;
759 
760 		IFQ_DEQUEUE(&ifp->if_snd, m);
761 
762 		if (m == NULL)
763 			break;
764 
765 retry:
766 		r = virtio_enqueue_prep(vsc, vq, &slot);
767 		if (r == EAGAIN) {
768 			ifp->if_flags |= IFF_OACTIVE;
769 			vioif_tx_vq_done_locked(vq);
770 			if (retry++ == 0)
771 				goto retry;
772 			else
773 				break;
774 		}
775 		if (r != 0)
776 			panic("enqueue_prep for a tx buffer");
777 		r = bus_dmamap_load_mbuf(vsc->sc_dmat,
778 					 sc->sc_tx_dmamaps[slot],
779 					 m, BUS_DMA_WRITE|BUS_DMA_NOWAIT);
780 		if (r != 0) {
781 			virtio_enqueue_abort(vsc, vq, slot);
782 			printf("%s: tx dmamap load failed, error code %d\n",
783 			       device_xname(sc->sc_dev), r);
784 			break;
785 		}
786 		r = virtio_enqueue_reserve(vsc, vq, slot,
787 					sc->sc_tx_dmamaps[slot]->dm_nsegs + 1);
788 		if (r != 0) {
789 			bus_dmamap_unload(vsc->sc_dmat,
790 					  sc->sc_tx_dmamaps[slot]);
791 			ifp->if_flags |= IFF_OACTIVE;
792 			vioif_tx_vq_done_locked(vq);
793 			if (retry++ == 0)
794 				goto retry;
795 			else
796 				break;
797 		}
798 
799 		sc->sc_tx_mbufs[slot] = m;
800 
801 		memset(&sc->sc_tx_hdrs[slot], 0, sizeof(struct virtio_net_hdr));
802 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_tx_dmamaps[slot],
803 				0, sc->sc_tx_dmamaps[slot]->dm_mapsize,
804 				BUS_DMASYNC_PREWRITE);
805 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_txhdr_dmamaps[slot],
806 				0, sc->sc_txhdr_dmamaps[slot]->dm_mapsize,
807 				BUS_DMASYNC_PREWRITE);
808 		virtio_enqueue(vsc, vq, slot, sc->sc_txhdr_dmamaps[slot], true);
809 		virtio_enqueue(vsc, vq, slot, sc->sc_tx_dmamaps[slot], true);
810 		virtio_enqueue_commit(vsc, vq, slot, false);
811 		queued++;
812 		bpf_mtap(ifp, m);
813 	}
814 
815 	if (m != NULL) {
816 		ifp->if_flags |= IFF_OACTIVE;
817 		m_freem(m);
818 	}
819 
820 	if (queued > 0) {
821 		virtio_enqueue_commit(vsc, vq, -1, true);
822 		ifp->if_timer = 5;
823 	}
824 
825 out:
826 	VIOIF_TX_UNLOCK(sc);
827 }
828 
829 static int
830 vioif_ioctl(struct ifnet *ifp, u_long cmd, void *data)
831 {
832 	int s, r;
833 
834 	s = splnet();
835 
836 	r = ether_ioctl(ifp, cmd, data);
837 	if ((r == 0 && cmd == SIOCSIFFLAGS) ||
838 	    (r == ENETRESET && (cmd == SIOCADDMULTI || cmd == SIOCDELMULTI))) {
839 		if (ifp->if_flags & IFF_RUNNING)
840 			r = vioif_rx_filter(ifp->if_softc);
841 		else
842 			r = 0;
843 	}
844 
845 	splx(s);
846 
847 	return r;
848 }
849 
850 void
851 vioif_watchdog(struct ifnet *ifp)
852 {
853 	struct vioif_softc *sc = ifp->if_softc;
854 
855 	if (ifp->if_flags & IFF_RUNNING)
856 		vioif_tx_vq_done(&sc->sc_vq[1]);
857 }
858 
859 
860 /*
861  * Recieve implementation
862  */
863 /* allocate and initialize a mbuf for recieve */
864 static int
865 vioif_add_rx_mbuf(struct vioif_softc *sc, int i)
866 {
867 	struct mbuf *m;
868 	int r;
869 
870 	MGETHDR(m, M_DONTWAIT, MT_DATA);
871 	if (m == NULL)
872 		return ENOBUFS;
873 	MCLGET(m, M_DONTWAIT);
874 	if ((m->m_flags & M_EXT) == 0) {
875 		m_freem(m);
876 		return ENOBUFS;
877 	}
878 	sc->sc_rx_mbufs[i] = m;
879 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
880 	r = bus_dmamap_load_mbuf(sc->sc_virtio->sc_dmat,
881 				 sc->sc_rx_dmamaps[i],
882 				 m, BUS_DMA_READ|BUS_DMA_NOWAIT);
883 	if (r) {
884 		m_freem(m);
885 		sc->sc_rx_mbufs[i] = 0;
886 		return r;
887 	}
888 
889 	return 0;
890 }
891 
892 /* free a mbuf for recieve */
893 static void
894 vioif_free_rx_mbuf(struct vioif_softc *sc, int i)
895 {
896 	bus_dmamap_unload(sc->sc_virtio->sc_dmat, sc->sc_rx_dmamaps[i]);
897 	m_freem(sc->sc_rx_mbufs[i]);
898 	sc->sc_rx_mbufs[i] = NULL;
899 }
900 
901 /* add mbufs for all the empty recieve slots */
902 static void
903 vioif_populate_rx_mbufs(struct vioif_softc *sc)
904 {
905 	VIOIF_RX_LOCK(sc);
906 	vioif_populate_rx_mbufs_locked(sc);
907 	VIOIF_RX_UNLOCK(sc);
908 }
909 
910 static void
911 vioif_populate_rx_mbufs_locked(struct vioif_softc *sc)
912 {
913 	struct virtio_softc *vsc = sc->sc_virtio;
914 	int i, r, ndone = 0;
915 	struct virtqueue *vq = &sc->sc_vq[0]; /* rx vq */
916 
917 	KASSERT(VIOIF_RX_LOCKED(sc));
918 
919 	if (sc->sc_stopping)
920 		return;
921 
922 	for (i = 0; i < vq->vq_num; i++) {
923 		int slot;
924 		r = virtio_enqueue_prep(vsc, vq, &slot);
925 		if (r == EAGAIN)
926 			break;
927 		if (r != 0)
928 			panic("enqueue_prep for rx buffers");
929 		if (sc->sc_rx_mbufs[slot] == NULL) {
930 			r = vioif_add_rx_mbuf(sc, slot);
931 			if (r != 0) {
932 				printf("%s: rx mbuf allocation failed, "
933 				       "error code %d\n",
934 				       device_xname(sc->sc_dev), r);
935 				break;
936 			}
937 		}
938 		r = virtio_enqueue_reserve(vsc, vq, slot,
939 					sc->sc_rx_dmamaps[slot]->dm_nsegs + 1);
940 		if (r != 0) {
941 			vioif_free_rx_mbuf(sc, slot);
942 			break;
943 		}
944 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_rxhdr_dmamaps[slot],
945 			0, sizeof(struct virtio_net_hdr), BUS_DMASYNC_PREREAD);
946 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_rx_dmamaps[slot],
947 			0, MCLBYTES, BUS_DMASYNC_PREREAD);
948 		virtio_enqueue(vsc, vq, slot, sc->sc_rxhdr_dmamaps[slot], false);
949 		virtio_enqueue(vsc, vq, slot, sc->sc_rx_dmamaps[slot], false);
950 		virtio_enqueue_commit(vsc, vq, slot, false);
951 		ndone++;
952 	}
953 	if (ndone > 0)
954 		virtio_enqueue_commit(vsc, vq, -1, true);
955 }
956 
957 /* dequeue recieved packets */
958 static int
959 vioif_rx_deq(struct vioif_softc *sc)
960 {
961 	int r;
962 
963 	KASSERT(sc->sc_stopping);
964 
965 	VIOIF_RX_LOCK(sc);
966 	r = vioif_rx_deq_locked(sc);
967 	VIOIF_RX_UNLOCK(sc);
968 
969 	return r;
970 }
971 
972 /* dequeue recieved packets */
973 static int
974 vioif_rx_deq_locked(struct vioif_softc *sc)
975 {
976 	struct virtio_softc *vsc = sc->sc_virtio;
977 	struct virtqueue *vq = &sc->sc_vq[0];
978 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
979 	struct mbuf *m;
980 	int r = 0;
981 	int slot, len;
982 
983 	KASSERT(VIOIF_RX_LOCKED(sc));
984 
985 	while (virtio_dequeue(vsc, vq, &slot, &len) == 0) {
986 		len -= sizeof(struct virtio_net_hdr);
987 		r = 1;
988 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_rxhdr_dmamaps[slot],
989 				0, sizeof(struct virtio_net_hdr),
990 				BUS_DMASYNC_POSTREAD);
991 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_rx_dmamaps[slot],
992 				0, MCLBYTES,
993 				BUS_DMASYNC_POSTREAD);
994 		m = sc->sc_rx_mbufs[slot];
995 		KASSERT(m != NULL);
996 		bus_dmamap_unload(vsc->sc_dmat, sc->sc_rx_dmamaps[slot]);
997 		sc->sc_rx_mbufs[slot] = 0;
998 		virtio_dequeue_commit(vsc, vq, slot);
999 		m->m_pkthdr.rcvif = ifp;
1000 		m->m_len = m->m_pkthdr.len = len;
1001 		ifp->if_ipackets++;
1002 		bpf_mtap(ifp, m);
1003 
1004 		VIOIF_RX_UNLOCK(sc);
1005 		(*ifp->if_input)(ifp, m);
1006 		VIOIF_RX_LOCK(sc);
1007 
1008 		if (sc->sc_stopping)
1009 			break;
1010 	}
1011 
1012 	return r;
1013 }
1014 
1015 /* rx interrupt; call _dequeue above and schedule a softint */
1016 static int
1017 vioif_rx_vq_done(struct virtqueue *vq)
1018 {
1019 	struct virtio_softc *vsc = vq->vq_owner;
1020 	struct vioif_softc *sc = device_private(vsc->sc_child);
1021 	int r = 0;
1022 
1023 #ifdef VIOIF_SOFTINT_INTR
1024 	KASSERT(!cpu_intr_p());
1025 #endif
1026 
1027 	VIOIF_RX_LOCK(sc);
1028 
1029 	if (sc->sc_stopping)
1030 		goto out;
1031 
1032 	r = vioif_rx_deq_locked(sc);
1033 	if (r)
1034 #ifdef VIOIF_SOFTINT_INTR
1035 		vioif_populate_rx_mbufs_locked(sc);
1036 #else
1037 		softint_schedule(sc->sc_rx_softint);
1038 #endif
1039 
1040 out:
1041 	VIOIF_RX_UNLOCK(sc);
1042 	return r;
1043 }
1044 
1045 /* softint: enqueue recieve requests for new incoming packets */
1046 static void
1047 vioif_rx_softint(void *arg)
1048 {
1049 	struct vioif_softc *sc = arg;
1050 
1051 	vioif_populate_rx_mbufs(sc);
1052 }
1053 
1054 /* free all the mbufs; called from if_stop(disable) */
1055 static void
1056 vioif_rx_drain(struct vioif_softc *sc)
1057 {
1058 	struct virtqueue *vq = &sc->sc_vq[0];
1059 	int i;
1060 
1061 	for (i = 0; i < vq->vq_num; i++) {
1062 		if (sc->sc_rx_mbufs[i] == NULL)
1063 			continue;
1064 		vioif_free_rx_mbuf(sc, i);
1065 	}
1066 }
1067 
1068 
1069 /*
1070  * Transmition implementation
1071  */
1072 /* actual transmission is done in if_start */
1073 /* tx interrupt; dequeue and free mbufs */
1074 /*
1075  * tx interrupt is actually disabled; this should be called upon
1076  * tx vq full and watchdog
1077  */
1078 static int
1079 vioif_tx_vq_done(struct virtqueue *vq)
1080 {
1081 	struct virtio_softc *vsc = vq->vq_owner;
1082 	struct vioif_softc *sc = device_private(vsc->sc_child);
1083 	int r = 0;
1084 
1085 	VIOIF_TX_LOCK(sc);
1086 
1087 	if (sc->sc_stopping)
1088 		goto out;
1089 
1090 	r = vioif_tx_vq_done_locked(vq);
1091 
1092 out:
1093 	VIOIF_TX_UNLOCK(sc);
1094 	return r;
1095 }
1096 
1097 static int
1098 vioif_tx_vq_done_locked(struct virtqueue *vq)
1099 {
1100 	struct virtio_softc *vsc = vq->vq_owner;
1101 	struct vioif_softc *sc = device_private(vsc->sc_child);
1102 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1103 	struct mbuf *m;
1104 	int r = 0;
1105 	int slot, len;
1106 
1107 	KASSERT(VIOIF_TX_LOCKED(sc));
1108 
1109 	while (virtio_dequeue(vsc, vq, &slot, &len) == 0) {
1110 		r++;
1111 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_txhdr_dmamaps[slot],
1112 				0, sizeof(struct virtio_net_hdr),
1113 				BUS_DMASYNC_POSTWRITE);
1114 		bus_dmamap_sync(vsc->sc_dmat, sc->sc_tx_dmamaps[slot],
1115 				0, sc->sc_tx_dmamaps[slot]->dm_mapsize,
1116 				BUS_DMASYNC_POSTWRITE);
1117 		m = sc->sc_tx_mbufs[slot];
1118 		bus_dmamap_unload(vsc->sc_dmat, sc->sc_tx_dmamaps[slot]);
1119 		sc->sc_tx_mbufs[slot] = 0;
1120 		virtio_dequeue_commit(vsc, vq, slot);
1121 		ifp->if_opackets++;
1122 		m_freem(m);
1123 	}
1124 
1125 	if (r)
1126 		ifp->if_flags &= ~IFF_OACTIVE;
1127 	return r;
1128 }
1129 
1130 /* free all the mbufs already put on vq; called from if_stop(disable) */
1131 static void
1132 vioif_tx_drain(struct vioif_softc *sc)
1133 {
1134 	struct virtio_softc *vsc = sc->sc_virtio;
1135 	struct virtqueue *vq = &sc->sc_vq[1];
1136 	int i;
1137 
1138 	KASSERT(sc->sc_stopping);
1139 
1140 	for (i = 0; i < vq->vq_num; i++) {
1141 		if (sc->sc_tx_mbufs[i] == NULL)
1142 			continue;
1143 		bus_dmamap_unload(vsc->sc_dmat, sc->sc_tx_dmamaps[i]);
1144 		m_freem(sc->sc_tx_mbufs[i]);
1145 		sc->sc_tx_mbufs[i] = NULL;
1146 	}
1147 }
1148 
1149 /*
1150  * Control vq
1151  */
1152 /* issue a VIRTIO_NET_CTRL_RX class command and wait for completion */
1153 static int
1154 vioif_ctrl_rx(struct vioif_softc *sc, int cmd, bool onoff)
1155 {
1156 	struct virtio_softc *vsc = sc->sc_virtio;
1157 	struct virtqueue *vq = &sc->sc_vq[2];
1158 	int r, slot;
1159 
1160 	if (vsc->sc_nvqs < 3)
1161 		return ENOTSUP;
1162 
1163 	mutex_enter(&sc->sc_ctrl_wait_lock);
1164 	while (sc->sc_ctrl_inuse != FREE)
1165 		cv_wait(&sc->sc_ctrl_wait, &sc->sc_ctrl_wait_lock);
1166 	sc->sc_ctrl_inuse = INUSE;
1167 	mutex_exit(&sc->sc_ctrl_wait_lock);
1168 
1169 	sc->sc_ctrl_cmd->class = VIRTIO_NET_CTRL_RX;
1170 	sc->sc_ctrl_cmd->command = cmd;
1171 	sc->sc_ctrl_rx->onoff = onoff;
1172 
1173 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_cmd_dmamap,
1174 			0, sizeof(struct virtio_net_ctrl_cmd),
1175 			BUS_DMASYNC_PREWRITE);
1176 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_rx_dmamap,
1177 			0, sizeof(struct virtio_net_ctrl_rx),
1178 			BUS_DMASYNC_PREWRITE);
1179 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_status_dmamap,
1180 			0, sizeof(struct virtio_net_ctrl_status),
1181 			BUS_DMASYNC_PREREAD);
1182 
1183 	r = virtio_enqueue_prep(vsc, vq, &slot);
1184 	if (r != 0)
1185 		panic("%s: control vq busy!?", device_xname(sc->sc_dev));
1186 	r = virtio_enqueue_reserve(vsc, vq, slot, 3);
1187 	if (r != 0)
1188 		panic("%s: control vq busy!?", device_xname(sc->sc_dev));
1189 	virtio_enqueue(vsc, vq, slot, sc->sc_ctrl_cmd_dmamap, true);
1190 	virtio_enqueue(vsc, vq, slot, sc->sc_ctrl_rx_dmamap, true);
1191 	virtio_enqueue(vsc, vq, slot, sc->sc_ctrl_status_dmamap, false);
1192 	virtio_enqueue_commit(vsc, vq, slot, true);
1193 
1194 	/* wait for done */
1195 	mutex_enter(&sc->sc_ctrl_wait_lock);
1196 	while (sc->sc_ctrl_inuse != DONE)
1197 		cv_wait(&sc->sc_ctrl_wait, &sc->sc_ctrl_wait_lock);
1198 	mutex_exit(&sc->sc_ctrl_wait_lock);
1199 	/* already dequeueued */
1200 
1201 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_cmd_dmamap, 0,
1202 			sizeof(struct virtio_net_ctrl_cmd),
1203 			BUS_DMASYNC_POSTWRITE);
1204 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_rx_dmamap, 0,
1205 			sizeof(struct virtio_net_ctrl_rx),
1206 			BUS_DMASYNC_POSTWRITE);
1207 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_status_dmamap, 0,
1208 			sizeof(struct virtio_net_ctrl_status),
1209 			BUS_DMASYNC_POSTREAD);
1210 
1211 	if (sc->sc_ctrl_status->ack == VIRTIO_NET_OK)
1212 		r = 0;
1213 	else {
1214 		printf("%s: failed setting rx mode\n",
1215 		       device_xname(sc->sc_dev));
1216 		r = EIO;
1217 	}
1218 
1219 	mutex_enter(&sc->sc_ctrl_wait_lock);
1220 	sc->sc_ctrl_inuse = FREE;
1221 	cv_signal(&sc->sc_ctrl_wait);
1222 	mutex_exit(&sc->sc_ctrl_wait_lock);
1223 
1224 	return r;
1225 }
1226 
1227 static int
1228 vioif_set_promisc(struct vioif_softc *sc, bool onoff)
1229 {
1230 	int r;
1231 
1232 	r = vioif_ctrl_rx(sc, VIRTIO_NET_CTRL_RX_PROMISC, onoff);
1233 
1234 	return r;
1235 }
1236 
1237 static int
1238 vioif_set_allmulti(struct vioif_softc *sc, bool onoff)
1239 {
1240 	int r;
1241 
1242 	r = vioif_ctrl_rx(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, onoff);
1243 
1244 	return r;
1245 }
1246 
1247 /* issue VIRTIO_NET_CTRL_MAC_TABLE_SET command and wait for completion */
1248 static int
1249 vioif_set_rx_filter(struct vioif_softc *sc)
1250 {
1251 	/* filter already set in sc_ctrl_mac_tbl */
1252 	struct virtio_softc *vsc = sc->sc_virtio;
1253 	struct virtqueue *vq = &sc->sc_vq[2];
1254 	int r, slot;
1255 
1256 	if (vsc->sc_nvqs < 3)
1257 		return ENOTSUP;
1258 
1259 	mutex_enter(&sc->sc_ctrl_wait_lock);
1260 	while (sc->sc_ctrl_inuse != FREE)
1261 		cv_wait(&sc->sc_ctrl_wait, &sc->sc_ctrl_wait_lock);
1262 	sc->sc_ctrl_inuse = INUSE;
1263 	mutex_exit(&sc->sc_ctrl_wait_lock);
1264 
1265 	sc->sc_ctrl_cmd->class = VIRTIO_NET_CTRL_MAC;
1266 	sc->sc_ctrl_cmd->command = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1267 
1268 	r = bus_dmamap_load(vsc->sc_dmat, sc->sc_ctrl_tbl_uc_dmamap,
1269 			    sc->sc_ctrl_mac_tbl_uc,
1270 			    (sizeof(struct virtio_net_ctrl_mac_tbl)
1271 			  + ETHER_ADDR_LEN * sc->sc_ctrl_mac_tbl_uc->nentries),
1272 			    NULL, BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1273 	if (r) {
1274 		printf("%s: control command dmamap load failed, "
1275 		       "error code %d\n", device_xname(sc->sc_dev), r);
1276 		goto out;
1277 	}
1278 	r = bus_dmamap_load(vsc->sc_dmat, sc->sc_ctrl_tbl_mc_dmamap,
1279 			    sc->sc_ctrl_mac_tbl_mc,
1280 			    (sizeof(struct virtio_net_ctrl_mac_tbl)
1281 			  + ETHER_ADDR_LEN * sc->sc_ctrl_mac_tbl_mc->nentries),
1282 			    NULL, BUS_DMA_WRITE|BUS_DMA_NOWAIT);
1283 	if (r) {
1284 		printf("%s: control command dmamap load failed, "
1285 		       "error code %d\n", device_xname(sc->sc_dev), r);
1286 		bus_dmamap_unload(vsc->sc_dmat, sc->sc_ctrl_tbl_uc_dmamap);
1287 		goto out;
1288 	}
1289 
1290 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_cmd_dmamap,
1291 			0, sizeof(struct virtio_net_ctrl_cmd),
1292 			BUS_DMASYNC_PREWRITE);
1293 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_tbl_uc_dmamap, 0,
1294 			(sizeof(struct virtio_net_ctrl_mac_tbl)
1295 			 + ETHER_ADDR_LEN * sc->sc_ctrl_mac_tbl_uc->nentries),
1296 			BUS_DMASYNC_PREWRITE);
1297 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_tbl_mc_dmamap, 0,
1298 			(sizeof(struct virtio_net_ctrl_mac_tbl)
1299 			 + ETHER_ADDR_LEN * sc->sc_ctrl_mac_tbl_mc->nentries),
1300 			BUS_DMASYNC_PREWRITE);
1301 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_status_dmamap,
1302 			0, sizeof(struct virtio_net_ctrl_status),
1303 			BUS_DMASYNC_PREREAD);
1304 
1305 	r = virtio_enqueue_prep(vsc, vq, &slot);
1306 	if (r != 0)
1307 		panic("%s: control vq busy!?", device_xname(sc->sc_dev));
1308 	r = virtio_enqueue_reserve(vsc, vq, slot, 4);
1309 	if (r != 0)
1310 		panic("%s: control vq busy!?", device_xname(sc->sc_dev));
1311 	virtio_enqueue(vsc, vq, slot, sc->sc_ctrl_cmd_dmamap, true);
1312 	virtio_enqueue(vsc, vq, slot, sc->sc_ctrl_tbl_uc_dmamap, true);
1313 	virtio_enqueue(vsc, vq, slot, sc->sc_ctrl_tbl_mc_dmamap, true);
1314 	virtio_enqueue(vsc, vq, slot, sc->sc_ctrl_status_dmamap, false);
1315 	virtio_enqueue_commit(vsc, vq, slot, true);
1316 
1317 	/* wait for done */
1318 	mutex_enter(&sc->sc_ctrl_wait_lock);
1319 	while (sc->sc_ctrl_inuse != DONE)
1320 		cv_wait(&sc->sc_ctrl_wait, &sc->sc_ctrl_wait_lock);
1321 	mutex_exit(&sc->sc_ctrl_wait_lock);
1322 	/* already dequeueued */
1323 
1324 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_cmd_dmamap, 0,
1325 			sizeof(struct virtio_net_ctrl_cmd),
1326 			BUS_DMASYNC_POSTWRITE);
1327 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_tbl_uc_dmamap, 0,
1328 			(sizeof(struct virtio_net_ctrl_mac_tbl)
1329 			 + ETHER_ADDR_LEN * sc->sc_ctrl_mac_tbl_uc->nentries),
1330 			BUS_DMASYNC_POSTWRITE);
1331 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_tbl_mc_dmamap, 0,
1332 			(sizeof(struct virtio_net_ctrl_mac_tbl)
1333 			 + ETHER_ADDR_LEN * sc->sc_ctrl_mac_tbl_mc->nentries),
1334 			BUS_DMASYNC_POSTWRITE);
1335 	bus_dmamap_sync(vsc->sc_dmat, sc->sc_ctrl_status_dmamap, 0,
1336 			sizeof(struct virtio_net_ctrl_status),
1337 			BUS_DMASYNC_POSTREAD);
1338 	bus_dmamap_unload(vsc->sc_dmat, sc->sc_ctrl_tbl_uc_dmamap);
1339 	bus_dmamap_unload(vsc->sc_dmat, sc->sc_ctrl_tbl_mc_dmamap);
1340 
1341 	if (sc->sc_ctrl_status->ack == VIRTIO_NET_OK)
1342 		r = 0;
1343 	else {
1344 		printf("%s: failed setting rx filter\n",
1345 		       device_xname(sc->sc_dev));
1346 		r = EIO;
1347 	}
1348 
1349 out:
1350 	mutex_enter(&sc->sc_ctrl_wait_lock);
1351 	sc->sc_ctrl_inuse = FREE;
1352 	cv_signal(&sc->sc_ctrl_wait);
1353 	mutex_exit(&sc->sc_ctrl_wait_lock);
1354 
1355 	return r;
1356 }
1357 
1358 /* ctrl vq interrupt; wake up the command issuer */
1359 static int
1360 vioif_ctrl_vq_done(struct virtqueue *vq)
1361 {
1362 	struct virtio_softc *vsc = vq->vq_owner;
1363 	struct vioif_softc *sc = device_private(vsc->sc_child);
1364 	int r, slot;
1365 
1366 	r = virtio_dequeue(vsc, vq, &slot, NULL);
1367 	if (r == ENOENT)
1368 		return 0;
1369 	virtio_dequeue_commit(vsc, vq, slot);
1370 
1371 	mutex_enter(&sc->sc_ctrl_wait_lock);
1372 	sc->sc_ctrl_inuse = DONE;
1373 	cv_signal(&sc->sc_ctrl_wait);
1374 	mutex_exit(&sc->sc_ctrl_wait_lock);
1375 
1376 	return 1;
1377 }
1378 
1379 /*
1380  * If IFF_PROMISC requested,  set promiscuous
1381  * If multicast filter small enough (<=MAXENTRIES) set rx filter
1382  * If large multicast filter exist use ALLMULTI
1383  */
1384 /*
1385  * If setting rx filter fails fall back to ALLMULTI
1386  * If ALLMULTI fails fall back to PROMISC
1387  */
1388 static int
1389 vioif_rx_filter(struct vioif_softc *sc)
1390 {
1391 	struct virtio_softc *vsc = sc->sc_virtio;
1392 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1393 	struct ether_multi *enm;
1394 	struct ether_multistep step;
1395 	int nentries;
1396 	int promisc = 0, allmulti = 0, rxfilter = 0;
1397 	int r;
1398 
1399 	if (vsc->sc_nvqs < 3) {	/* no ctrl vq; always promisc */
1400 		ifp->if_flags |= IFF_PROMISC;
1401 		return 0;
1402 	}
1403 
1404 	if (ifp->if_flags & IFF_PROMISC) {
1405 		promisc = 1;
1406 		goto set;
1407 	}
1408 
1409 	nentries = -1;
1410 	ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
1411 	while (nentries++, enm != NULL) {
1412 		if (nentries >= VIRTIO_NET_CTRL_MAC_MAXENTRIES) {
1413 			allmulti = 1;
1414 			goto set;
1415 		}
1416 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
1417 			   ETHER_ADDR_LEN)) {
1418 			allmulti = 1;
1419 			goto set;
1420 		}
1421 		memcpy(sc->sc_ctrl_mac_tbl_mc->macs[nentries],
1422 		       enm->enm_addrlo, ETHER_ADDR_LEN);
1423 		ETHER_NEXT_MULTI(step, enm);
1424 	}
1425 	rxfilter = 1;
1426 
1427 set:
1428 	if (rxfilter) {
1429 		sc->sc_ctrl_mac_tbl_uc->nentries = 0;
1430 		sc->sc_ctrl_mac_tbl_mc->nentries = nentries;
1431 		r = vioif_set_rx_filter(sc);
1432 		if (r != 0) {
1433 			rxfilter = 0;
1434 			allmulti = 1; /* fallback */
1435 		}
1436 	} else {
1437 		/* remove rx filter */
1438 		sc->sc_ctrl_mac_tbl_uc->nentries = 0;
1439 		sc->sc_ctrl_mac_tbl_mc->nentries = 0;
1440 		r = vioif_set_rx_filter(sc);
1441 		/* what to do on failure? */
1442 	}
1443 	if (allmulti) {
1444 		r = vioif_set_allmulti(sc, true);
1445 		if (r != 0) {
1446 			allmulti = 0;
1447 			promisc = 1; /* fallback */
1448 		}
1449 	} else {
1450 		r = vioif_set_allmulti(sc, false);
1451 		/* what to do on failure? */
1452 	}
1453 	if (promisc) {
1454 		r = vioif_set_promisc(sc, true);
1455 	} else {
1456 		r = vioif_set_promisc(sc, false);
1457 	}
1458 
1459 	return r;
1460 }
1461 
1462 /* change link status */
1463 static int
1464 vioif_updown(struct vioif_softc *sc, bool isup)
1465 {
1466 	struct virtio_softc *vsc = sc->sc_virtio;
1467 
1468 	if (!(vsc->sc_features & VIRTIO_NET_F_STATUS))
1469 		return ENODEV;
1470 	virtio_write_device_config_1(vsc,
1471 				     VIRTIO_NET_CONFIG_STATUS,
1472 				     isup?VIRTIO_NET_S_LINK_UP:0);
1473 	return 0;
1474 }
1475