if_vioif.c revision 1.115 1 /* $NetBSD: if_vioif.c,v 1.115 2025/07/15 05:09:28 ozaki-r Exp $ */
2
3 /*
4 * Copyright (c) 2020 The NetBSD Foundation, Inc.
5 * Copyright (c) 2010 Minoura Makoto.
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: if_vioif.c,v 1.115 2025/07/15 05:09:28 ozaki-r Exp $");
31
32 #ifdef _KERNEL_OPT
33 #include "opt_net_mpsafe.h"
34 #endif
35
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/atomic.h>
40 #include <sys/bus.h>
41 #include <sys/condvar.h>
42 #include <sys/device.h>
43 #include <sys/evcnt.h>
44 #include <sys/intr.h>
45 #include <sys/kmem.h>
46 #include <sys/mbuf.h>
47 #include <sys/mutex.h>
48 #include <sys/sockio.h>
49 #include <sys/syslog.h>
50 #include <sys/cpu.h>
51 #include <sys/module.h>
52 #include <sys/pcq.h>
53 #include <sys/workqueue.h>
54 #include <sys/xcall.h>
55
56 #include <dev/pci/virtioreg.h>
57 #include <dev/pci/virtiovar.h>
58
59 #include <net/if.h>
60 #include <net/if_dl.h>
61 #include <net/if_media.h>
62 #include <net/if_ether.h>
63
64 #include <net/bpf.h>
65
66 #include "ioconf.h"
67
68 #ifdef NET_MPSAFE
69 #define VIOIF_MPSAFE 1
70 #define VIOIF_MULTIQ 1
71 #endif
72
73 /*
74 * if_vioifreg.h:
75 */
76 /* Configuration registers */
77 #define VIRTIO_NET_CONFIG_MAC 0 /* 8bit x 6byte */
78 #define VIRTIO_NET_CONFIG_STATUS 6 /* 16bit */
79 #define VIRTIO_NET_CONFIG_MAX_VQ_PAIRS 8 /* 16bit */
80 #define VIRTIO_NET_CONFIG_MTU 10 /* 16bit */
81
82 /* Feature bits */
83 #define VIRTIO_NET_F_CSUM __BIT(0)
84 #define VIRTIO_NET_F_GUEST_CSUM __BIT(1)
85 #define VIRTIO_NET_F_MAC __BIT(5)
86 #define VIRTIO_NET_F_GSO __BIT(6)
87 #define VIRTIO_NET_F_GUEST_TSO4 __BIT(7)
88 #define VIRTIO_NET_F_GUEST_TSO6 __BIT(8)
89 #define VIRTIO_NET_F_GUEST_ECN __BIT(9)
90 #define VIRTIO_NET_F_GUEST_UFO __BIT(10)
91 #define VIRTIO_NET_F_HOST_TSO4 __BIT(11)
92 #define VIRTIO_NET_F_HOST_TSO6 __BIT(12)
93 #define VIRTIO_NET_F_HOST_ECN __BIT(13)
94 #define VIRTIO_NET_F_HOST_UFO __BIT(14)
95 #define VIRTIO_NET_F_MRG_RXBUF __BIT(15)
96 #define VIRTIO_NET_F_STATUS __BIT(16)
97 #define VIRTIO_NET_F_CTRL_VQ __BIT(17)
98 #define VIRTIO_NET_F_CTRL_RX __BIT(18)
99 #define VIRTIO_NET_F_CTRL_VLAN __BIT(19)
100 #define VIRTIO_NET_F_CTRL_RX_EXTRA __BIT(20)
101 #define VIRTIO_NET_F_GUEST_ANNOUNCE __BIT(21)
102 #define VIRTIO_NET_F_MQ __BIT(22)
103 #define VIRTIO_NET_F_CTRL_MAC_ADDR __BIT(23)
104
105 #define VIRTIO_NET_FLAG_BITS \
106 VIRTIO_COMMON_FLAG_BITS \
107 "b\x17" "CTRL_MAC\0" \
108 "b\x16" "MQ\0" \
109 "b\x15" "GUEST_ANNOUNCE\0" \
110 "b\x14" "CTRL_RX_EXTRA\0" \
111 "b\x13" "CTRL_VLAN\0" \
112 "b\x12" "CTRL_RX\0" \
113 "b\x11" "CTRL_VQ\0" \
114 "b\x10" "STATUS\0" \
115 "b\x0f" "MRG_RXBUF\0" \
116 "b\x0e" "HOST_UFO\0" \
117 "b\x0d" "HOST_ECN\0" \
118 "b\x0c" "HOST_TSO6\0" \
119 "b\x0b" "HOST_TSO4\0" \
120 "b\x0a" "GUEST_UFO\0" \
121 "b\x09" "GUEST_ECN\0" \
122 "b\x08" "GUEST_TSO6\0" \
123 "b\x07" "GUEST_TSO4\0" \
124 "b\x06" "GSO\0" \
125 "b\x05" "MAC\0" \
126 "b\x01" "GUEST_CSUM\0" \
127 "b\x00" "CSUM\0"
128
129 /* Status */
130 #define VIRTIO_NET_S_LINK_UP 1
131
132 /* Packet header structure */
133 struct virtio_net_hdr {
134 uint8_t flags;
135 uint8_t gso_type;
136 uint16_t hdr_len;
137 uint16_t gso_size;
138 uint16_t csum_start;
139 uint16_t csum_offset;
140
141 uint16_t num_buffers; /* VIRTIO_NET_F_MRG_RXBUF enabled or v1 */
142 } __packed;
143
144 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* flags */
145 #define VIRTIO_NET_HDR_GSO_NONE 0 /* gso_type */
146 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /* gso_type */
147 #define VIRTIO_NET_HDR_GSO_UDP 3 /* gso_type */
148 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /* gso_type */
149 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /* gso_type, |'ed */
150
151 #define VIRTIO_NET_MAX_GSO_LEN (65536+ETHER_HDR_LEN)
152
153 /* Control virtqueue */
154 struct virtio_net_ctrl_cmd {
155 uint8_t class;
156 uint8_t command;
157 } __packed;
158 #define VIRTIO_NET_CTRL_RX 0
159 # define VIRTIO_NET_CTRL_RX_PROMISC 0
160 # define VIRTIO_NET_CTRL_RX_ALLMULTI 1
161
162 #define VIRTIO_NET_CTRL_MAC 1
163 # define VIRTIO_NET_CTRL_MAC_TABLE_SET 0
164 # define VIRTIO_NET_CTRL_MAC_ADDR_SET 1
165
166 #define VIRTIO_NET_CTRL_VLAN 2
167 # define VIRTIO_NET_CTRL_VLAN_ADD 0
168 # define VIRTIO_NET_CTRL_VLAN_DEL 1
169
170 #define VIRTIO_NET_CTRL_MQ 4
171 # define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET 0
172 # define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN 1
173 # define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX 0x8000
174
175 struct virtio_net_ctrl_status {
176 uint8_t ack;
177 } __packed;
178 #define VIRTIO_NET_OK 0
179 #define VIRTIO_NET_ERR 1
180
181 struct virtio_net_ctrl_rx {
182 uint8_t onoff;
183 } __packed;
184
185 struct virtio_net_ctrl_mac_tbl {
186 uint32_t nentries;
187 uint8_t macs[][ETHER_ADDR_LEN];
188 } __packed;
189
190 struct virtio_net_ctrl_mac_addr {
191 uint8_t mac[ETHER_ADDR_LEN];
192 } __packed;
193
194 struct virtio_net_ctrl_vlan {
195 uint16_t id;
196 } __packed;
197
198 struct virtio_net_ctrl_mq {
199 uint16_t virtqueue_pairs;
200 } __packed;
201
202 /*
203 * if_vioifvar.h:
204 */
205
206 /*
207 * Locking notes:
208 * + a field in vioif_netqueue is protected by netq_lock (a spin mutex)
209 * - more than one lock cannot be held at onece
210 * + a field in vioif_tx_context and vioif_rx_context is also protected
211 * by netq_lock.
212 * + ctrlq_inuse is protected by ctrlq_wait_lock.
213 * - other fields in vioif_ctrlqueue are protected by ctrlq_inuse
214 * - netq_lock cannot be held along with ctrlq_wait_lock
215 * + fields in vioif_softc except queues are protected by
216 * sc->sc_lock(an adaptive mutex)
217 * - the lock is held before acquisition of other locks
218 */
219
220 struct vioif_ctrl_cmdspec {
221 bus_dmamap_t dmamap;
222 void *buf;
223 bus_size_t bufsize;
224 };
225
226 struct vioif_work {
227 struct work cookie;
228 void (*func)(void *);
229 void *arg;
230 unsigned int added;
231 };
232
233 struct vioif_net_map {
234 struct virtio_net_hdr *vnm_hdr;
235 bus_dmamap_t vnm_hdr_map;
236 struct mbuf *vnm_mbuf;
237 bus_dmamap_t vnm_mbuf_map;
238 };
239
240 #define VIOIF_NETQ_RX 0
241 #define VIOIF_NETQ_TX 1
242 #define VIOIF_NETQ_IDX 2
243 #define VIOIF_NETQ_DIR(n) ((n) % VIOIF_NETQ_IDX)
244 #define VIOIF_NETQ_PAIRIDX(n) ((n) / VIOIF_NETQ_IDX)
245 #define VIOIF_NETQ_RXQID(n) ((n) * VIOIF_NETQ_IDX + VIOIF_NETQ_RX)
246 #define VIOIF_NETQ_TXQID(n) ((n) * VIOIF_NETQ_IDX + VIOIF_NETQ_TX)
247
248 struct vioif_netqueue {
249 kmutex_t netq_lock;
250 struct virtqueue *netq_vq;
251 bool netq_stopping;
252 bool netq_running_handle;
253 void *netq_maps_kva;
254 struct vioif_net_map *netq_maps;
255
256 void *netq_softint;
257 struct vioif_work netq_work;
258 bool netq_workqueue;
259
260 char netq_evgroup[32];
261 struct evcnt netq_mbuf_load_failed;
262 struct evcnt netq_enqueue_failed;
263
264 void *netq_ctx;
265 };
266
267 struct vioif_tx_context {
268 bool txc_link_active;
269 bool txc_no_free_slots;
270 pcq_t *txc_intrq;
271 void *txc_deferred_transmit;
272
273 struct evcnt txc_defrag_failed;
274 struct evcnt txc_pcq_full;
275 };
276
277 struct vioif_rx_context {
278 struct evcnt rxc_mbuf_enobufs;
279 };
280 struct vioif_ctrlqueue {
281 struct virtqueue *ctrlq_vq;
282 enum {
283 FREE, INUSE, DONE
284 } ctrlq_inuse;
285 kcondvar_t ctrlq_wait;
286 kmutex_t ctrlq_wait_lock;
287 struct lwp *ctrlq_owner;
288
289 struct virtio_net_ctrl_cmd *ctrlq_cmd;
290 struct virtio_net_ctrl_status *ctrlq_status;
291 struct virtio_net_ctrl_rx *ctrlq_rx;
292 struct virtio_net_ctrl_mac_tbl *ctrlq_mac_tbl_uc;
293 struct virtio_net_ctrl_mac_tbl *ctrlq_mac_tbl_mc;
294 struct virtio_net_ctrl_mac_addr *ctrlq_mac_addr;
295 struct virtio_net_ctrl_mq *ctrlq_mq;
296
297 bus_dmamap_t ctrlq_cmd_dmamap;
298 bus_dmamap_t ctrlq_status_dmamap;
299 bus_dmamap_t ctrlq_rx_dmamap;
300 bus_dmamap_t ctrlq_tbl_uc_dmamap;
301 bus_dmamap_t ctrlq_tbl_mc_dmamap;
302 bus_dmamap_t ctrlq_mac_addr_dmamap;
303 bus_dmamap_t ctrlq_mq_dmamap;
304
305 struct evcnt ctrlq_cmd_load_failed;
306 struct evcnt ctrlq_cmd_failed;
307 };
308
309 struct vioif_softc {
310 device_t sc_dev;
311 kmutex_t sc_lock;
312 struct sysctllog *sc_sysctllog;
313
314 struct virtio_softc *sc_virtio;
315 struct virtqueue *sc_vqs;
316 u_int sc_hdr_size;
317
318 int sc_max_nvq_pairs;
319 int sc_req_nvq_pairs;
320 int sc_act_nvq_pairs;
321
322 uint8_t sc_mac[ETHER_ADDR_LEN];
323 struct ethercom sc_ethercom;
324 int sc_link_state;
325
326 struct vioif_netqueue *sc_netqs;
327
328 bool sc_has_ctrl;
329 struct vioif_ctrlqueue sc_ctrlq;
330
331 bus_dma_segment_t sc_segs[1];
332 void *sc_dmamem;
333 void *sc_kmem;
334
335 void *sc_cfg_softint;
336
337 struct workqueue *sc_txrx_workqueue;
338 bool sc_txrx_workqueue_sysctl;
339 u_int sc_tx_intr_process_limit;
340 u_int sc_tx_process_limit;
341 u_int sc_rx_intr_process_limit;
342 u_int sc_rx_process_limit;
343 };
344 #define VIRTIO_NET_TX_MAXNSEGS (16) /* XXX */
345 #define VIRTIO_NET_CTRL_MAC_MAXENTRIES (64) /* XXX */
346
347 #define VIOIF_TX_INTR_PROCESS_LIMIT 256
348 #define VIOIF_TX_PROCESS_LIMIT 256
349 #define VIOIF_RX_INTR_PROCESS_LIMIT 0U
350 #define VIOIF_RX_PROCESS_LIMIT 256
351
352 #define VIOIF_WORKQUEUE_PRI PRI_SOFTNET
353 #define VIOIF_IS_LINK_ACTIVE(_sc) ((_sc)->sc_link_state == LINK_STATE_UP ? \
354 true : false)
355
356 /* cfattach interface functions */
357 static int vioif_match(device_t, cfdata_t, void *);
358 static void vioif_attach(device_t, device_t, void *);
359 static int vioif_finalize_teardown(device_t);
360
361 /* ifnet interface functions */
362 static int vioif_init(struct ifnet *);
363 static void vioif_stop(struct ifnet *, int);
364 static void vioif_start(struct ifnet *);
365 static int vioif_transmit(struct ifnet *, struct mbuf *);
366 static int vioif_ioctl(struct ifnet *, u_long, void *);
367 static void vioif_watchdog(struct ifnet *);
368 static int vioif_ifflags(struct vioif_softc *);
369 static int vioif_ifflags_cb(struct ethercom *);
370
371 /* tx & rx */
372 static int vioif_netqueue_init(struct vioif_softc *,
373 struct virtio_softc *, size_t, u_int);
374 static void vioif_netqueue_teardown(struct vioif_softc *,
375 struct virtio_softc *, size_t);
376 static void vioif_net_intr_enable(struct vioif_softc *,
377 struct virtio_softc *);
378 static void vioif_net_intr_disable(struct vioif_softc *,
379 struct virtio_softc *);
380 static void vioif_net_sched_handle(struct vioif_softc *,
381 struct vioif_netqueue *);
382
383 /* rx */
384 static void vioif_populate_rx_mbufs_locked(struct vioif_softc *,
385 struct vioif_netqueue *);
386 static int vioif_rx_intr(void *);
387 static void vioif_rx_handle(void *);
388 static void vioif_rx_queue_clear(struct vioif_softc *,
389 struct virtio_softc *, struct vioif_netqueue *);
390
391 /* tx */
392 static void vioif_start_locked(struct ifnet *, struct vioif_netqueue *);
393 static void vioif_transmit_locked(struct ifnet *, struct vioif_netqueue *);
394 static void vioif_deferred_transmit(void *);
395 static int vioif_tx_intr(void *);
396 static void vioif_tx_handle(void *);
397 static void vioif_tx_queue_clear(struct vioif_softc *, struct virtio_softc *,
398 struct vioif_netqueue *);
399
400 /* controls */
401 static int vioif_ctrl_intr(void *);
402 static int vioif_ctrl_rx(struct vioif_softc *, int, bool);
403 static int vioif_set_promisc(struct vioif_softc *, bool);
404 static int vioif_set_allmulti(struct vioif_softc *, bool);
405 static int vioif_set_rx_filter(struct vioif_softc *);
406 static int vioif_rx_filter(struct vioif_softc *);
407 static int vioif_set_mac_addr(struct vioif_softc *);
408 static int vioif_ctrl_mq_vq_pairs_set(struct vioif_softc *, int);
409
410 /* config interrupt */
411 static int vioif_config_change(struct virtio_softc *);
412 static void vioif_cfg_softint(void *);
413 static void vioif_update_link_status(struct vioif_softc *);
414
415 /* others */
416 static void vioif_alloc_queues(struct vioif_softc *);
417 static void vioif_free_queues(struct vioif_softc *);
418 static int vioif_alloc_mems(struct vioif_softc *);
419 static struct workqueue*
420 vioif_workq_create(const char *, pri_t, int, int);
421 static void vioif_workq_destroy(struct workqueue *);
422 static void vioif_work_set(struct vioif_work *, void(*)(void *), void *);
423 static void vioif_work_add(struct workqueue *, struct vioif_work *);
424 static void vioif_work_wait(struct workqueue *, struct vioif_work *);
425 static int vioif_setup_sysctl(struct vioif_softc *);
426 static void vioif_setup_stats(struct vioif_softc *);
427
428 CFATTACH_DECL_NEW(vioif, sizeof(struct vioif_softc),
429 vioif_match, vioif_attach, NULL, NULL);
430
431 static void
432 vioif_intr_barrier(void)
433 {
434
435 /* wait for finish all interrupt handler */
436 xc_barrier(0);
437 }
438
439 static void
440 vioif_notify(struct virtio_softc *vsc, struct virtqueue *vq)
441 {
442
443 virtio_enqueue_commit(vsc, vq, -1, true);
444 }
445
446 static int
447 vioif_match(device_t parent, cfdata_t match, void *aux)
448 {
449 struct virtio_attach_args *va = aux;
450
451 if (va->sc_childdevid == VIRTIO_DEVICE_ID_NETWORK)
452 return 1;
453
454 return 0;
455 }
456
457 static void
458 vioif_attach(device_t parent, device_t self, void *aux)
459 {
460 struct vioif_softc *sc = device_private(self);
461 struct virtio_softc *vsc = device_private(parent);
462 struct vioif_netqueue *txq0;
463 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
464 uint64_t features, req_features;
465 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
466 u_int softint_flags;
467 int r, i, req_flags;
468 char xnamebuf[MAXCOMLEN];
469 size_t nvqs;
470
471 if (virtio_child(vsc) != NULL) {
472 aprint_normal(": child already attached for %s; "
473 "something wrong...\n", device_xname(parent));
474 return;
475 }
476
477 sc->sc_dev = self;
478 sc->sc_virtio = vsc;
479 sc->sc_link_state = LINK_STATE_UNKNOWN;
480
481 sc->sc_max_nvq_pairs = 1;
482 sc->sc_req_nvq_pairs = 1;
483 sc->sc_act_nvq_pairs = 1;
484 sc->sc_txrx_workqueue_sysctl = true;
485 sc->sc_tx_intr_process_limit = VIOIF_TX_INTR_PROCESS_LIMIT;
486 sc->sc_tx_process_limit = VIOIF_TX_PROCESS_LIMIT;
487 sc->sc_rx_intr_process_limit = VIOIF_RX_INTR_PROCESS_LIMIT;
488 sc->sc_rx_process_limit = VIOIF_RX_PROCESS_LIMIT;
489
490 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE);
491
492 snprintf(xnamebuf, sizeof(xnamebuf), "%s_txrx", device_xname(self));
493 sc->sc_txrx_workqueue = vioif_workq_create(xnamebuf, VIOIF_WORKQUEUE_PRI,
494 IPL_NET, WQ_PERCPU | WQ_MPSAFE);
495 if (sc->sc_txrx_workqueue == NULL)
496 goto err;
497
498 req_flags = 0;
499
500 #ifdef VIOIF_MPSAFE
501 req_flags |= VIRTIO_F_INTR_MPSAFE;
502 #endif
503 req_flags |= VIRTIO_F_INTR_MSIX;
504
505 req_features =
506 VIRTIO_NET_F_MAC | VIRTIO_NET_F_STATUS | VIRTIO_NET_F_CTRL_VQ |
507 VIRTIO_NET_F_CTRL_RX | VIRTIO_F_NOTIFY_ON_EMPTY;
508 req_features |= VIRTIO_F_RING_EVENT_IDX;
509 req_features |= VIRTIO_NET_F_CTRL_MAC_ADDR;
510 #ifdef VIOIF_MULTIQ
511 req_features |= VIRTIO_NET_F_MQ;
512 #endif
513
514 virtio_child_attach_start(vsc, self, IPL_NET,
515 req_features, VIRTIO_NET_FLAG_BITS);
516 features = virtio_features(vsc);
517
518 if (features == 0)
519 goto err;
520
521 if (features & VIRTIO_NET_F_MAC) {
522 for (i = 0; i < __arraycount(sc->sc_mac); i++) {
523 sc->sc_mac[i] = virtio_read_device_config_1(vsc,
524 VIRTIO_NET_CONFIG_MAC + i);
525 }
526 } else {
527 /* code stolen from sys/net/if_tap.c */
528 struct timeval tv;
529 uint32_t ui;
530 getmicrouptime(&tv);
531 ui = (tv.tv_sec ^ tv.tv_usec) & 0xffffff;
532 memcpy(sc->sc_mac+3, (uint8_t *)&ui, 3);
533 for (i = 0; i < __arraycount(sc->sc_mac); i++) {
534 virtio_write_device_config_1(vsc,
535 VIRTIO_NET_CONFIG_MAC + i, sc->sc_mac[i]);
536 }
537 }
538
539 /* 'Ethernet' with capital follows other ethernet driver attachment */
540 aprint_normal_dev(self, "Ethernet address %s\n",
541 ether_sprintf(sc->sc_mac));
542
543 if (features & (VIRTIO_NET_F_MRG_RXBUF | VIRTIO_F_VERSION_1)) {
544 sc->sc_hdr_size = sizeof(struct virtio_net_hdr);
545 } else {
546 sc->sc_hdr_size = offsetof(struct virtio_net_hdr, num_buffers);
547 }
548
549 if ((features & VIRTIO_NET_F_CTRL_VQ) &&
550 (features & VIRTIO_NET_F_CTRL_RX)) {
551 sc->sc_has_ctrl = true;
552
553 cv_init(&ctrlq->ctrlq_wait, "ctrl_vq");
554 mutex_init(&ctrlq->ctrlq_wait_lock, MUTEX_DEFAULT, IPL_NET);
555 ctrlq->ctrlq_inuse = FREE;
556 } else {
557 sc->sc_has_ctrl = false;
558 }
559
560 if (sc->sc_has_ctrl && (features & VIRTIO_NET_F_MQ)) {
561 sc->sc_max_nvq_pairs = virtio_read_device_config_2(vsc,
562 VIRTIO_NET_CONFIG_MAX_VQ_PAIRS);
563
564 if (sc->sc_max_nvq_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX)
565 goto err;
566
567 /* Limit the number of queue pairs to use */
568 sc->sc_req_nvq_pairs = MIN(sc->sc_max_nvq_pairs, ncpu);
569
570 if (sc->sc_max_nvq_pairs > 1)
571 req_flags |= VIRTIO_F_INTR_PERVQ;
572 }
573
574 vioif_alloc_queues(sc);
575
576 #ifdef VIOIF_MPSAFE
577 softint_flags = SOFTINT_NET | SOFTINT_MPSAFE;
578 #else
579 softint_flags = SOFTINT_NET;
580 #endif
581
582 /*
583 * Initialize network queues
584 */
585 nvqs = sc->sc_max_nvq_pairs * 2;
586 for (i = 0; i < nvqs; i++) {
587 r = vioif_netqueue_init(sc, vsc, i, softint_flags);
588 if (r != 0)
589 goto err;
590 }
591
592 if (sc->sc_has_ctrl) {
593 int ctrlq_idx = nvqs;
594
595 nvqs++;
596 /*
597 * Allocating a virtqueue for control channel
598 */
599 sc->sc_ctrlq.ctrlq_vq = &sc->sc_vqs[ctrlq_idx];
600 virtio_init_vq(vsc, ctrlq->ctrlq_vq, ctrlq_idx,
601 vioif_ctrl_intr, ctrlq);
602
603 r = virtio_alloc_vq(vsc, ctrlq->ctrlq_vq, NBPG, 1, "control");
604 if (r != 0) {
605 aprint_error_dev(self, "failed to allocate "
606 "a virtqueue for control channel, error code %d\n",
607 r);
608
609 sc->sc_has_ctrl = false;
610 cv_destroy(&ctrlq->ctrlq_wait);
611 mutex_destroy(&ctrlq->ctrlq_wait_lock);
612 }
613 }
614
615 sc->sc_cfg_softint = softint_establish(softint_flags,
616 vioif_cfg_softint, sc);
617 if (sc->sc_cfg_softint == NULL) {
618 aprint_error_dev(self, "cannot establish ctl softint\n");
619 goto err;
620 }
621
622 if (vioif_alloc_mems(sc) < 0)
623 goto err;
624
625 r = virtio_child_attach_finish(vsc, sc->sc_vqs, nvqs,
626 vioif_config_change, req_flags);
627 if (r != 0)
628 goto err;
629
630 if (vioif_setup_sysctl(sc) != 0) {
631 aprint_error_dev(self, "unable to create sysctl node\n");
632 /* continue */
633 }
634
635 vioif_setup_stats(sc);
636
637 strlcpy(ifp->if_xname, device_xname(self), IFNAMSIZ);
638 ifp->if_softc = sc;
639 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
640 #ifdef VIOIF_MPSAFE
641 ifp->if_extflags = IFEF_MPSAFE;
642 #endif
643 ifp->if_start = vioif_start;
644 if (sc->sc_req_nvq_pairs > 1)
645 ifp->if_transmit = vioif_transmit;
646 ifp->if_ioctl = vioif_ioctl;
647 ifp->if_init = vioif_init;
648 ifp->if_stop = vioif_stop;
649 ifp->if_capabilities = 0;
650 ifp->if_watchdog = vioif_watchdog;
651 txq0 = &sc->sc_netqs[VIOIF_NETQ_TXQID(0)];
652 IFQ_SET_MAXLEN(&ifp->if_snd, MAX(txq0->netq_vq->vq_num, IFQ_MAXLEN));
653 IFQ_SET_READY(&ifp->if_snd);
654
655 sc->sc_ethercom.ec_capabilities |= ETHERCAP_VLAN_MTU;
656
657 if_attach(ifp);
658 if_deferred_start_init(ifp, NULL);
659 ether_ifattach(ifp, sc->sc_mac);
660 ether_set_ifflags_cb(&sc->sc_ethercom, vioif_ifflags_cb);
661
662 return;
663
664 err:
665 nvqs = sc->sc_max_nvq_pairs * 2;
666 for (i = 0; i < nvqs; i++) {
667 vioif_netqueue_teardown(sc, vsc, i);
668 }
669
670 if (sc->sc_has_ctrl) {
671 cv_destroy(&ctrlq->ctrlq_wait);
672 mutex_destroy(&ctrlq->ctrlq_wait_lock);
673 virtio_free_vq(vsc, ctrlq->ctrlq_vq);
674 ctrlq->ctrlq_vq = NULL;
675 }
676
677 vioif_free_queues(sc);
678 mutex_destroy(&sc->sc_lock);
679 virtio_child_attach_failed(vsc);
680 config_finalize_register(self, vioif_finalize_teardown);
681
682 return;
683 }
684
685 static int
686 vioif_finalize_teardown(device_t self)
687 {
688 struct vioif_softc *sc = device_private(self);
689
690 if (sc->sc_txrx_workqueue != NULL) {
691 vioif_workq_destroy(sc->sc_txrx_workqueue);
692 sc->sc_txrx_workqueue = NULL;
693 }
694
695 return 0;
696 }
697
698 /*
699 * Interface functions for ifnet
700 */
701 static int
702 vioif_init(struct ifnet *ifp)
703 {
704 struct vioif_softc *sc = ifp->if_softc;
705 struct virtio_softc *vsc = sc->sc_virtio;
706 struct vioif_netqueue *netq;
707 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
708 int r, i;
709
710 vioif_stop(ifp, 0);
711
712 r = virtio_reinit_start(vsc);
713 if (r != 0) {
714 log(LOG_ERR, "%s: reset failed\n", ifp->if_xname);
715 return EIO;
716 }
717
718 virtio_negotiate_features(vsc, virtio_features(vsc));
719
720 for (i = 0; i < sc->sc_req_nvq_pairs; i++) {
721 netq = &sc->sc_netqs[VIOIF_NETQ_RXQID(i)];
722
723 mutex_enter(&netq->netq_lock);
724 vioif_populate_rx_mbufs_locked(sc, netq);
725 mutex_exit(&netq->netq_lock);
726 }
727
728 virtio_reinit_end(vsc);
729
730 if (sc->sc_has_ctrl)
731 virtio_start_vq_intr(vsc, ctrlq->ctrlq_vq);
732
733 r = vioif_ctrl_mq_vq_pairs_set(sc, sc->sc_req_nvq_pairs);
734 if (r == 0)
735 sc->sc_act_nvq_pairs = sc->sc_req_nvq_pairs;
736 else
737 sc->sc_act_nvq_pairs = 1;
738
739 SET(ifp->if_flags, IFF_RUNNING);
740
741 vioif_net_intr_enable(sc, vsc);
742
743 vioif_update_link_status(sc);
744 r = vioif_rx_filter(sc);
745
746 return r;
747 }
748
749 static void
750 vioif_stop(struct ifnet *ifp, int disable)
751 {
752 struct vioif_softc *sc = ifp->if_softc;
753 struct virtio_softc *vsc = sc->sc_virtio;
754 struct vioif_netqueue *netq;
755 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
756 size_t i, act_qnum;
757
758 act_qnum = sc->sc_act_nvq_pairs * 2;
759
760 CLR(ifp->if_flags, IFF_RUNNING);
761 for (i = 0; i < act_qnum; i++) {
762 netq = &sc->sc_netqs[i];
763
764 mutex_enter(&netq->netq_lock);
765 netq->netq_stopping = true;
766 mutex_exit(&netq->netq_lock);
767 }
768
769 /* disable interrupts */
770 vioif_net_intr_disable(sc, vsc);
771 if (sc->sc_has_ctrl)
772 virtio_stop_vq_intr(vsc, ctrlq->ctrlq_vq);
773
774 /*
775 * only way to stop interrupt, I/O and DMA is resetting...
776 *
777 * NOTE: Devices based on VirtIO draft specification can not
778 * stop interrupt completely even if virtio_stop_vq_intr() is called.
779 */
780 virtio_reset(vsc);
781
782 vioif_intr_barrier();
783
784 for (i = 0; i < act_qnum; i++) {
785 netq = &sc->sc_netqs[i];
786 vioif_work_wait(sc->sc_txrx_workqueue, &netq->netq_work);
787 }
788
789 for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
790 netq = &sc->sc_netqs[VIOIF_NETQ_RXQID(i)];
791 vioif_rx_queue_clear(sc, vsc, netq);
792
793 netq = &sc->sc_netqs[VIOIF_NETQ_TXQID(i)];
794 vioif_tx_queue_clear(sc, vsc, netq);
795 }
796
797 /* all packet processing is stopped */
798 for (i = 0; i < act_qnum; i++) {
799 netq = &sc->sc_netqs[i];
800
801 mutex_enter(&netq->netq_lock);
802 netq->netq_stopping = false;
803 mutex_exit(&netq->netq_lock);
804 }
805 }
806
807 static void
808 vioif_start(struct ifnet *ifp)
809 {
810 struct vioif_softc *sc = ifp->if_softc;
811 struct vioif_netqueue *txq0 = &sc->sc_netqs[VIOIF_NETQ_TXQID(0)];
812
813 #ifdef VIOIF_MPSAFE
814 KASSERT(if_is_mpsafe(ifp));
815 #endif
816
817 mutex_enter(&txq0->netq_lock);
818 vioif_start_locked(ifp, txq0);
819 mutex_exit(&txq0->netq_lock);
820 }
821
822 static inline int
823 vioif_select_txqueue(struct ifnet *ifp, struct mbuf *m)
824 {
825 struct vioif_softc *sc = ifp->if_softc;
826 u_int cpuid = cpu_index(curcpu());
827
828 return VIOIF_NETQ_TXQID(cpuid % sc->sc_act_nvq_pairs);
829 }
830
831 static int
832 vioif_transmit(struct ifnet *ifp, struct mbuf *m)
833 {
834 struct vioif_softc *sc = ifp->if_softc;
835 struct vioif_netqueue *netq;
836 struct vioif_tx_context *txc;
837 int qid;
838
839 qid = vioif_select_txqueue(ifp, m);
840 netq = &sc->sc_netqs[qid];
841 txc = netq->netq_ctx;
842
843 if (__predict_false(!pcq_put(txc->txc_intrq, m))) {
844 txc->txc_pcq_full.ev_count++;
845 m_freem(m);
846 return ENOBUFS;
847 }
848
849 net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
850 if_statadd_ref(ifp, nsr, if_obytes, m->m_pkthdr.len);
851 if (m->m_flags & M_MCAST)
852 if_statinc_ref(ifp, nsr, if_omcasts);
853 IF_STAT_PUTREF(ifp);
854
855 if (mutex_tryenter(&netq->netq_lock)) {
856 vioif_transmit_locked(ifp, netq);
857 mutex_exit(&netq->netq_lock);
858 }
859
860 return 0;
861 }
862
863 void
864 vioif_watchdog(struct ifnet *ifp)
865 {
866 struct vioif_softc *sc = ifp->if_softc;
867 struct vioif_netqueue *netq;
868 int i;
869
870 if (ISSET(ifp->if_flags, IFF_RUNNING)) {
871 if (ISSET(ifp->if_flags, IFF_DEBUG)) {
872 log(LOG_DEBUG, "%s: watchdog timed out\n",
873 ifp->if_xname);
874 }
875
876 for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
877 netq = &sc->sc_netqs[VIOIF_NETQ_TXQID(i)];
878
879 mutex_enter(&netq->netq_lock);
880 if (!netq->netq_running_handle) {
881 netq->netq_running_handle = true;
882 vioif_net_sched_handle(sc, netq);
883 }
884 mutex_exit(&netq->netq_lock);
885 }
886 }
887 }
888
889 static int
890 vioif_ioctl(struct ifnet *ifp, u_long cmd, void *data)
891 {
892 int s, r;
893
894 s = splnet();
895
896 r = ether_ioctl(ifp, cmd, data);
897 if (r == ENETRESET && (cmd == SIOCADDMULTI || cmd == SIOCDELMULTI)) {
898 if (ifp->if_flags & IFF_RUNNING) {
899 r = vioif_rx_filter(ifp->if_softc);
900 } else {
901 r = 0;
902 }
903 }
904
905 splx(s);
906
907 return r;
908 }
909
910 static int
911 vioif_ifflags(struct vioif_softc *sc)
912 {
913 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
914 bool onoff;
915 int r;
916
917 if (!sc->sc_has_ctrl) {
918 /* no ctrl vq; always promisc and allmulti */
919 ifp->if_flags |= (IFF_PROMISC | IFF_ALLMULTI);
920 return 0;
921 }
922
923 onoff = ifp->if_flags & IFF_ALLMULTI ? true : false;
924 r = vioif_set_allmulti(sc, onoff);
925 if (r != 0) {
926 log(LOG_WARNING,
927 "%s: couldn't %sable ALLMULTI\n",
928 ifp->if_xname, onoff ? "en" : "dis");
929 if (onoff) {
930 CLR(ifp->if_flags, IFF_ALLMULTI);
931 } else {
932 SET(ifp->if_flags, IFF_ALLMULTI);
933 }
934 }
935
936 onoff = ifp->if_flags & IFF_PROMISC ? true : false;
937 r = vioif_set_promisc(sc, onoff);
938 if (r != 0) {
939 log(LOG_WARNING,
940 "%s: couldn't %sable PROMISC\n",
941 ifp->if_xname, onoff ? "en" : "dis");
942 if (onoff) {
943 CLR(ifp->if_flags, IFF_PROMISC);
944 } else {
945 SET(ifp->if_flags, IFF_PROMISC);
946 }
947 }
948
949 return 0;
950 }
951
952 static int
953 vioif_ifflags_cb(struct ethercom *ec)
954 {
955 struct ifnet *ifp = &ec->ec_if;
956 struct vioif_softc *sc = ifp->if_softc;
957
958 return vioif_ifflags(sc);
959 }
960
961 static int
962 vioif_setup_sysctl(struct vioif_softc *sc)
963 {
964 const char *devname;
965 struct sysctllog **log;
966 const struct sysctlnode *rnode, *rxnode, *txnode;
967 int error;
968
969 log = &sc->sc_sysctllog;
970 devname = device_xname(sc->sc_dev);
971
972 error = sysctl_createv(log, 0, NULL, &rnode,
973 0, CTLTYPE_NODE, devname,
974 SYSCTL_DESCR("virtio-net information and settings"),
975 NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL);
976 if (error)
977 goto out;
978
979 error = sysctl_createv(log, 0, &rnode, NULL,
980 CTLFLAG_READWRITE, CTLTYPE_BOOL, "txrx_workqueue",
981 SYSCTL_DESCR("Use workqueue for packet processing"),
982 NULL, 0, &sc->sc_txrx_workqueue_sysctl, 0, CTL_CREATE, CTL_EOL);
983 if (error)
984 goto out;
985
986 error = sysctl_createv(log, 0, &rnode, &rxnode,
987 0, CTLTYPE_NODE, "rx",
988 SYSCTL_DESCR("virtio-net information and settings for Rx"),
989 NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL);
990 if (error)
991 goto out;
992
993 error = sysctl_createv(log, 0, &rxnode, NULL,
994 CTLFLAG_READWRITE, CTLTYPE_INT, "intr_process_limit",
995 SYSCTL_DESCR("max number of Rx packets to process for interrupt processing"),
996 NULL, 0, &sc->sc_rx_intr_process_limit, 0, CTL_CREATE, CTL_EOL);
997 if (error)
998 goto out;
999
1000 error = sysctl_createv(log, 0, &rxnode, NULL,
1001 CTLFLAG_READWRITE, CTLTYPE_INT, "process_limit",
1002 SYSCTL_DESCR("max number of Rx packets to process for deferred processing"),
1003 NULL, 0, &sc->sc_rx_process_limit, 0, CTL_CREATE, CTL_EOL);
1004 if (error)
1005 goto out;
1006
1007 error = sysctl_createv(log, 0, &rnode, &txnode,
1008 0, CTLTYPE_NODE, "tx",
1009 SYSCTL_DESCR("virtio-net information and settings for Tx"),
1010 NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL);
1011 if (error)
1012 goto out;
1013
1014 error = sysctl_createv(log, 0, &txnode, NULL,
1015 CTLFLAG_READWRITE, CTLTYPE_INT, "intr_process_limit",
1016 SYSCTL_DESCR("max number of Tx packets to process for interrupt processing"),
1017 NULL, 0, &sc->sc_tx_intr_process_limit, 0, CTL_CREATE, CTL_EOL);
1018 if (error)
1019 goto out;
1020
1021 error = sysctl_createv(log, 0, &txnode, NULL,
1022 CTLFLAG_READWRITE, CTLTYPE_INT, "process_limit",
1023 SYSCTL_DESCR("max number of Tx packets to process for deferred processing"),
1024 NULL, 0, &sc->sc_tx_process_limit, 0, CTL_CREATE, CTL_EOL);
1025
1026 out:
1027 if (error)
1028 sysctl_teardown(log);
1029
1030 return error;
1031 }
1032
1033 static void
1034 vioif_setup_stats(struct vioif_softc *sc)
1035 {
1036 struct vioif_netqueue *netq;
1037 struct vioif_tx_context *txc;
1038 struct vioif_rx_context *rxc;
1039 size_t i, netq_num;
1040
1041 netq_num = sc->sc_max_nvq_pairs * 2;
1042 for (i = 0; i < netq_num; i++) {
1043 netq = &sc->sc_netqs[i];
1044 evcnt_attach_dynamic(&netq->netq_mbuf_load_failed, EVCNT_TYPE_MISC,
1045 NULL, netq->netq_evgroup, "failed to load mbuf to DMA");
1046 evcnt_attach_dynamic(&netq->netq_enqueue_failed,
1047 EVCNT_TYPE_MISC, NULL, netq->netq_evgroup,
1048 "virtqueue enqueue failed failed");
1049
1050 switch (VIOIF_NETQ_DIR(i)) {
1051 case VIOIF_NETQ_RX:
1052 rxc = netq->netq_ctx;
1053 evcnt_attach_dynamic(&rxc->rxc_mbuf_enobufs,
1054 EVCNT_TYPE_MISC, NULL, netq->netq_evgroup,
1055 "no receive buffer");
1056 break;
1057 case VIOIF_NETQ_TX:
1058 txc = netq->netq_ctx;
1059 evcnt_attach_dynamic(&txc->txc_defrag_failed,
1060 EVCNT_TYPE_MISC, NULL, netq->netq_evgroup,
1061 "m_defrag() failed");
1062 evcnt_attach_dynamic(&txc->txc_pcq_full,
1063 EVCNT_TYPE_MISC, NULL, netq->netq_evgroup,
1064 "pcq full");
1065 break;
1066 }
1067 }
1068
1069 evcnt_attach_dynamic(&sc->sc_ctrlq.ctrlq_cmd_load_failed, EVCNT_TYPE_MISC,
1070 NULL, device_xname(sc->sc_dev), "control command dmamap load failed");
1071 evcnt_attach_dynamic(&sc->sc_ctrlq.ctrlq_cmd_failed, EVCNT_TYPE_MISC,
1072 NULL, device_xname(sc->sc_dev), "control command failed");
1073 }
1074
1075 /*
1076 * allocate memory
1077 */
1078 static int
1079 vioif_dmamap_create(struct vioif_softc *sc, bus_dmamap_t *map,
1080 bus_size_t size, int nsegs, const char *usage)
1081 {
1082 int r;
1083
1084 r = bus_dmamap_create(virtio_dmat(sc->sc_virtio), size,
1085 nsegs, size, 0, BUS_DMA_NOWAIT | BUS_DMA_ALLOCNOW, map);
1086
1087 if (r != 0) {
1088 aprint_error_dev(sc->sc_dev, "%s dmamap creation failed, "
1089 "error code %d\n", usage, r);
1090 }
1091
1092 return r;
1093 }
1094
1095 static void
1096 vioif_dmamap_destroy(struct vioif_softc *sc, bus_dmamap_t *map)
1097 {
1098
1099 if (*map) {
1100 bus_dmamap_destroy(virtio_dmat(sc->sc_virtio), *map);
1101 *map = NULL;
1102 }
1103 }
1104
1105 static int
1106 vioif_dmamap_create_load(struct vioif_softc *sc, bus_dmamap_t *map,
1107 void *buf, bus_size_t size, int nsegs, int rw, const char *usage)
1108 {
1109 int r;
1110
1111 r = vioif_dmamap_create(sc, map, size, nsegs, usage);
1112 if (r != 0)
1113 return 1;
1114
1115 r = bus_dmamap_load(virtio_dmat(sc->sc_virtio), *map, buf,
1116 size, NULL, rw | BUS_DMA_NOWAIT);
1117 if (r != 0) {
1118 vioif_dmamap_destroy(sc, map);
1119 aprint_error_dev(sc->sc_dev, "%s dmamap load failed. "
1120 "error code %d\n", usage, r);
1121 }
1122
1123 return r;
1124 }
1125
1126 static void *
1127 vioif_assign_mem(intptr_t *p, size_t size)
1128 {
1129 intptr_t rv;
1130
1131 rv = *p;
1132 *p += size;
1133
1134 return (void *)rv;
1135 }
1136
1137 /*
1138 * dma memory is used for:
1139 * netq_maps_kva: metadata array for received frames (READ) and
1140 * sent frames (WRITE)
1141 * ctrlq_cmd: command to be sent via ctrl vq (WRITE)
1142 * ctrlq_status: return value for a command via ctrl vq (READ)
1143 * ctrlq_rx: parameter for a VIRTIO_NET_CTRL_RX class command
1144 * (WRITE)
1145 * ctrlq_mac_tbl_uc: unicast MAC address filter for a VIRTIO_NET_CTRL_MAC
1146 * class command (WRITE)
1147 * ctrlq_mac_tbl_mc: multicast MAC address filter for a VIRTIO_NET_CTRL_MAC
1148 * class command (WRITE)
1149 * ctrlq_* structures are allocated only one each; they are protected by
1150 * ctrlq_inuse variable and ctrlq_wait condvar.
1151 */
1152 static int
1153 vioif_alloc_mems(struct vioif_softc *sc)
1154 {
1155 struct virtio_softc *vsc = sc->sc_virtio;
1156 struct vioif_netqueue *netq;
1157 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
1158 struct vioif_net_map *maps;
1159 unsigned int vq_num;
1160 int r, rsegs;
1161 bus_size_t dmamemsize;
1162 size_t qid, i, netq_num, kmemsize;
1163 void *vaddr;
1164 intptr_t p;
1165
1166 netq_num = sc->sc_max_nvq_pairs * 2;
1167
1168 /* allocate DMA memory */
1169 dmamemsize = 0;
1170
1171 for (qid = 0; qid < netq_num; qid++) {
1172 maps = sc->sc_netqs[qid].netq_maps;
1173 vq_num = sc->sc_netqs[qid].netq_vq->vq_num;
1174 dmamemsize += sizeof(*maps[0].vnm_hdr) * vq_num;
1175 }
1176
1177 if (sc->sc_has_ctrl) {
1178 dmamemsize += sizeof(struct virtio_net_ctrl_cmd);
1179 dmamemsize += sizeof(struct virtio_net_ctrl_status);
1180 dmamemsize += sizeof(struct virtio_net_ctrl_rx);
1181 dmamemsize += sizeof(struct virtio_net_ctrl_mac_tbl)
1182 + ETHER_ADDR_LEN;
1183 dmamemsize += sizeof(struct virtio_net_ctrl_mac_tbl)
1184 + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES;
1185 dmamemsize += sizeof(struct virtio_net_ctrl_mac_addr);
1186 dmamemsize += sizeof(struct virtio_net_ctrl_mq);
1187 }
1188
1189 r = bus_dmamem_alloc(virtio_dmat(vsc), dmamemsize, 0, 0,
1190 &sc->sc_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
1191 if (r != 0) {
1192 aprint_error_dev(sc->sc_dev,
1193 "DMA memory allocation failed, size %" PRIuBUSSIZE ", "
1194 "error code %d\n", dmamemsize, r);
1195 goto err_none;
1196 }
1197 r = bus_dmamem_map(virtio_dmat(vsc), &sc->sc_segs[0], 1,
1198 dmamemsize, &vaddr, BUS_DMA_NOWAIT);
1199 if (r != 0) {
1200 aprint_error_dev(sc->sc_dev,
1201 "DMA memory map failed, error code %d\n", r);
1202 goto err_dmamem_alloc;
1203 }
1204
1205 /* assign DMA memory */
1206 memset(vaddr, 0, dmamemsize);
1207 sc->sc_dmamem = vaddr;
1208 p = (intptr_t) vaddr;
1209
1210 for (qid = 0; qid < netq_num; qid++) {
1211 netq = &sc->sc_netqs[qid];
1212 maps = netq->netq_maps;
1213 vq_num = netq->netq_vq->vq_num;
1214
1215 netq->netq_maps_kva = vioif_assign_mem(&p,
1216 sizeof(*maps[0].vnm_hdr) * vq_num);
1217 }
1218
1219 if (sc->sc_has_ctrl) {
1220 ctrlq->ctrlq_cmd = vioif_assign_mem(&p,
1221 sizeof(*ctrlq->ctrlq_cmd));
1222 ctrlq->ctrlq_status = vioif_assign_mem(&p,
1223 sizeof(*ctrlq->ctrlq_status));
1224 ctrlq->ctrlq_rx = vioif_assign_mem(&p,
1225 sizeof(*ctrlq->ctrlq_rx));
1226 ctrlq->ctrlq_mac_tbl_uc = vioif_assign_mem(&p,
1227 sizeof(*ctrlq->ctrlq_mac_tbl_uc)
1228 + ETHER_ADDR_LEN);
1229 ctrlq->ctrlq_mac_tbl_mc = vioif_assign_mem(&p,
1230 sizeof(*ctrlq->ctrlq_mac_tbl_mc)
1231 + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES);
1232 ctrlq->ctrlq_mac_addr = vioif_assign_mem(&p,
1233 sizeof(*ctrlq->ctrlq_mac_addr));
1234 ctrlq->ctrlq_mq = vioif_assign_mem(&p, sizeof(*ctrlq->ctrlq_mq));
1235 }
1236
1237 /* allocate kmem */
1238 kmemsize = 0;
1239
1240 for (qid = 0; qid < netq_num; qid++) {
1241 netq = &sc->sc_netqs[qid];
1242 vq_num = netq->netq_vq->vq_num;
1243
1244 kmemsize += sizeof(netq->netq_maps[0]) * vq_num;
1245 }
1246
1247 vaddr = kmem_zalloc(kmemsize, KM_SLEEP);
1248 sc->sc_kmem = vaddr;
1249
1250 /* assign allocated kmem */
1251 p = (intptr_t) vaddr;
1252
1253 for (qid = 0; qid < netq_num; qid++) {
1254 netq = &sc->sc_netqs[qid];
1255 vq_num = netq->netq_vq->vq_num;
1256
1257 netq->netq_maps = vioif_assign_mem(&p,
1258 sizeof(netq->netq_maps[0]) * vq_num);
1259 }
1260
1261 /* prepare dmamaps */
1262 for (qid = 0; qid < netq_num; qid++) {
1263 static const struct {
1264 const char *msg_hdr;
1265 const char *msg_payload;
1266 int dma_flag;
1267 bus_size_t dma_size;
1268 int dma_nsegs;
1269 } dmaparams[VIOIF_NETQ_IDX] = {
1270 [VIOIF_NETQ_RX] = {
1271 .msg_hdr = "rx header",
1272 .msg_payload = "rx payload",
1273 .dma_flag = BUS_DMA_READ,
1274 .dma_size = MCLBYTES - ETHER_ALIGN,
1275 .dma_nsegs = 1,
1276 },
1277 [VIOIF_NETQ_TX] = {
1278 .msg_hdr = "tx header",
1279 .msg_payload = "tx payload",
1280 .dma_flag = BUS_DMA_WRITE,
1281 .dma_size = ETHER_MAX_LEN,
1282 .dma_nsegs = VIRTIO_NET_TX_MAXNSEGS,
1283 }
1284 };
1285
1286 struct virtio_net_hdr *hdrs;
1287 int dir;
1288 int nsegs;
1289
1290 dir = VIOIF_NETQ_DIR(qid);
1291 netq = &sc->sc_netqs[qid];
1292 vq_num = netq->netq_vq->vq_num;
1293 maps = netq->netq_maps;
1294 hdrs = netq->netq_maps_kva;
1295 nsegs = uimin(dmaparams[dir].dma_nsegs, vq_num - 1/*hdr*/);
1296
1297 for (i = 0; i < vq_num; i++) {
1298 maps[i].vnm_hdr = &hdrs[i];
1299
1300 r = vioif_dmamap_create_load(sc, &maps[i].vnm_hdr_map,
1301 maps[i].vnm_hdr, sc->sc_hdr_size, 1,
1302 dmaparams[dir].dma_flag, dmaparams[dir].msg_hdr);
1303 if (r != 0)
1304 goto err_reqs;
1305
1306 r = vioif_dmamap_create(sc, &maps[i].vnm_mbuf_map,
1307 dmaparams[dir].dma_size, nsegs,
1308 dmaparams[dir].msg_payload);
1309 if (r != 0)
1310 goto err_reqs;
1311 }
1312 }
1313
1314 if (sc->sc_has_ctrl) {
1315 /* control vq class & command */
1316 r = vioif_dmamap_create_load(sc, &ctrlq->ctrlq_cmd_dmamap,
1317 ctrlq->ctrlq_cmd, sizeof(*ctrlq->ctrlq_cmd), 1,
1318 BUS_DMA_WRITE, "control command");
1319 if (r != 0)
1320 goto err_reqs;
1321
1322 r = vioif_dmamap_create_load(sc, &ctrlq->ctrlq_status_dmamap,
1323 ctrlq->ctrlq_status, sizeof(*ctrlq->ctrlq_status), 1,
1324 BUS_DMA_READ, "control status");
1325 if (r != 0)
1326 goto err_reqs;
1327
1328 /* control vq rx mode command parameter */
1329 r = vioif_dmamap_create_load(sc, &ctrlq->ctrlq_rx_dmamap,
1330 ctrlq->ctrlq_rx, sizeof(*ctrlq->ctrlq_rx), 1,
1331 BUS_DMA_WRITE, "rx mode control command");
1332 if (r != 0)
1333 goto err_reqs;
1334
1335 /* multiqueue set command */
1336 r = vioif_dmamap_create_load(sc, &ctrlq->ctrlq_mq_dmamap,
1337 ctrlq->ctrlq_mq, sizeof(*ctrlq->ctrlq_mq), 1,
1338 BUS_DMA_WRITE, "multiqueue set command");
1339 if (r != 0)
1340 goto err_reqs;
1341
1342 /* control vq MAC filter table for unicast */
1343 /* do not load now since its length is variable */
1344 r = vioif_dmamap_create(sc, &ctrlq->ctrlq_tbl_uc_dmamap,
1345 sizeof(*ctrlq->ctrlq_mac_tbl_uc)
1346 + ETHER_ADDR_LEN, 1,
1347 "unicast MAC address filter command");
1348 if (r != 0)
1349 goto err_reqs;
1350
1351 /* control vq MAC filter table for multicast */
1352 r = vioif_dmamap_create(sc, &ctrlq->ctrlq_tbl_mc_dmamap,
1353 sizeof(*ctrlq->ctrlq_mac_tbl_mc)
1354 + ETHER_ADDR_LEN * VIRTIO_NET_CTRL_MAC_MAXENTRIES, 1,
1355 "multicast MAC address filter command");
1356 if (r != 0)
1357 goto err_reqs;
1358
1359 /* control vq MAC address set command */
1360 r = vioif_dmamap_create_load(sc,
1361 &ctrlq->ctrlq_mac_addr_dmamap,
1362 ctrlq->ctrlq_mac_addr,
1363 sizeof(*ctrlq->ctrlq_mac_addr), 1,
1364 BUS_DMA_WRITE, "mac addr set command");
1365 if (r != 0)
1366 goto err_reqs;
1367 }
1368
1369 return 0;
1370
1371 err_reqs:
1372 vioif_dmamap_destroy(sc, &ctrlq->ctrlq_tbl_mc_dmamap);
1373 vioif_dmamap_destroy(sc, &ctrlq->ctrlq_tbl_uc_dmamap);
1374 vioif_dmamap_destroy(sc, &ctrlq->ctrlq_rx_dmamap);
1375 vioif_dmamap_destroy(sc, &ctrlq->ctrlq_status_dmamap);
1376 vioif_dmamap_destroy(sc, &ctrlq->ctrlq_cmd_dmamap);
1377 vioif_dmamap_destroy(sc, &ctrlq->ctrlq_mac_addr_dmamap);
1378 for (qid = 0; qid < netq_num; qid++) {
1379 vq_num = sc->sc_netqs[qid].netq_vq->vq_num;
1380 maps = sc->sc_netqs[qid].netq_maps;
1381
1382 for (i = 0; i < vq_num; i++) {
1383 vioif_dmamap_destroy(sc, &maps[i].vnm_mbuf_map);
1384 vioif_dmamap_destroy(sc, &maps[i].vnm_hdr_map);
1385 }
1386 }
1387 if (sc->sc_kmem) {
1388 kmem_free(sc->sc_kmem, kmemsize);
1389 sc->sc_kmem = NULL;
1390 }
1391 bus_dmamem_unmap(virtio_dmat(vsc), sc->sc_dmamem, dmamemsize);
1392 err_dmamem_alloc:
1393 bus_dmamem_free(virtio_dmat(vsc), &sc->sc_segs[0], 1);
1394 err_none:
1395 return -1;
1396 }
1397
1398 static void
1399 vioif_alloc_queues(struct vioif_softc *sc)
1400 {
1401 int nvq_pairs = sc->sc_max_nvq_pairs;
1402 size_t nvqs, netq_num;
1403
1404 KASSERT(nvq_pairs <= VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX);
1405
1406 nvqs = netq_num = sc->sc_max_nvq_pairs * 2;
1407 if (sc->sc_has_ctrl)
1408 nvqs++;
1409
1410 sc->sc_vqs = kmem_zalloc(sizeof(sc->sc_vqs[0]) * nvqs, KM_SLEEP);
1411 sc->sc_netqs = kmem_zalloc(sizeof(sc->sc_netqs[0]) * netq_num,
1412 KM_SLEEP);
1413 }
1414
1415 static void
1416 vioif_free_queues(struct vioif_softc *sc)
1417 {
1418 size_t nvqs, netq_num;
1419
1420 nvqs = netq_num = sc->sc_max_nvq_pairs * 2;
1421 if (sc->sc_ctrlq.ctrlq_vq)
1422 nvqs++;
1423
1424 kmem_free(sc->sc_netqs, sizeof(sc->sc_netqs[0]) * netq_num);
1425 kmem_free(sc->sc_vqs, sizeof(sc->sc_vqs[0]) * nvqs);
1426 sc->sc_netqs = NULL;
1427 sc->sc_vqs = NULL;
1428 }
1429
1430 /*
1431 * Network queues
1432 */
1433 static int
1434 vioif_netqueue_init(struct vioif_softc *sc, struct virtio_softc *vsc,
1435 size_t qid, u_int softint_flags)
1436 {
1437 static const struct {
1438 const char *dirname;
1439 int segsize;
1440 int nsegs;
1441 int (*intrhand)(void *);
1442 void (*sihand)(void *);
1443 } params[VIOIF_NETQ_IDX] = {
1444 [VIOIF_NETQ_RX] = {
1445 .dirname = "rx",
1446 .segsize = MCLBYTES,
1447 .nsegs = 2,
1448 .intrhand = vioif_rx_intr,
1449 .sihand = vioif_rx_handle,
1450 },
1451 [VIOIF_NETQ_TX] = {
1452 .dirname = "tx",
1453 .segsize = ETHER_MAX_LEN - ETHER_HDR_LEN,
1454 .nsegs = 2,
1455 .intrhand = vioif_tx_intr,
1456 .sihand = vioif_tx_handle,
1457 }
1458 };
1459
1460 struct virtqueue *vq;
1461 struct vioif_netqueue *netq;
1462 struct vioif_tx_context *txc;
1463 struct vioif_rx_context *rxc;
1464 char qname[32];
1465 int r, dir;
1466
1467 txc = NULL;
1468 rxc = NULL;
1469 netq = &sc->sc_netqs[qid];
1470 vq = &sc->sc_vqs[qid];
1471 dir = VIOIF_NETQ_DIR(qid);
1472
1473 netq->netq_vq = &sc->sc_vqs[qid];
1474 netq->netq_stopping = false;
1475 netq->netq_running_handle = false;
1476
1477 snprintf(qname, sizeof(qname), "%s%zu",
1478 params[dir].dirname, VIOIF_NETQ_PAIRIDX(qid));
1479 snprintf(netq->netq_evgroup, sizeof(netq->netq_evgroup),
1480 "%s-%s", device_xname(sc->sc_dev), qname);
1481
1482 mutex_init(&netq->netq_lock, MUTEX_DEFAULT, IPL_NET);
1483 virtio_init_vq(vsc, vq, qid, params[dir].intrhand, netq);
1484
1485 r = virtio_alloc_vq(vsc, vq,
1486 params[dir].segsize + sc->sc_hdr_size,
1487 params[dir].nsegs, qname);
1488 if (r != 0)
1489 goto err;
1490 netq->netq_vq = vq;
1491
1492 netq->netq_softint = softint_establish(softint_flags,
1493 params[dir].sihand, netq);
1494 if (netq->netq_softint == NULL) {
1495 aprint_error_dev(sc->sc_dev,
1496 "couldn't establish %s softint\n",
1497 params[dir].dirname);
1498 goto err;
1499 }
1500 vioif_work_set(&netq->netq_work, params[dir].sihand, netq);
1501
1502 switch (dir) {
1503 case VIOIF_NETQ_RX:
1504 rxc = kmem_zalloc(sizeof(*rxc), KM_SLEEP);
1505 netq->netq_ctx = rxc;
1506 /* nothing to do */
1507 break;
1508 case VIOIF_NETQ_TX:
1509 txc = kmem_zalloc(sizeof(*txc), KM_SLEEP);
1510 netq->netq_ctx = (void *)txc;
1511 txc->txc_deferred_transmit = softint_establish(softint_flags,
1512 vioif_deferred_transmit, netq);
1513 if (txc->txc_deferred_transmit == NULL) {
1514 aprint_error_dev(sc->sc_dev,
1515 "couldn't establish softint for "
1516 "tx deferred transmit\n");
1517 goto err;
1518 }
1519 txc->txc_link_active = VIOIF_IS_LINK_ACTIVE(sc);
1520 txc->txc_no_free_slots = false;
1521 txc->txc_intrq = pcq_create(vq->vq_num, KM_SLEEP);
1522 break;
1523 }
1524
1525 return 0;
1526
1527 err:
1528 netq->netq_ctx = NULL;
1529
1530 if (rxc != NULL) {
1531 kmem_free(rxc, sizeof(*rxc));
1532 }
1533
1534 if (txc != NULL) {
1535 if (txc->txc_deferred_transmit != NULL)
1536 softint_disestablish(txc->txc_deferred_transmit);
1537 if (txc->txc_intrq != NULL)
1538 pcq_destroy(txc->txc_intrq);
1539 kmem_free(txc, sizeof(*txc));
1540 }
1541
1542 vioif_work_set(&netq->netq_work, NULL, NULL);
1543 if (netq->netq_softint != NULL) {
1544 softint_disestablish(netq->netq_softint);
1545 netq->netq_softint = NULL;
1546 }
1547
1548 virtio_free_vq(vsc, vq);
1549 mutex_destroy(&netq->netq_lock);
1550 netq->netq_vq = NULL;
1551
1552 return -1;
1553 }
1554
1555 static void
1556 vioif_netqueue_teardown(struct vioif_softc *sc, struct virtio_softc *vsc,
1557 size_t qid)
1558 {
1559 struct vioif_netqueue *netq;
1560 struct vioif_rx_context *rxc;
1561 struct vioif_tx_context *txc;
1562 int dir;
1563
1564 netq = &sc->sc_netqs[qid];
1565
1566 if (netq->netq_vq == NULL)
1567 return;
1568
1569 netq = &sc->sc_netqs[qid];
1570 dir = VIOIF_NETQ_DIR(qid);
1571 switch (dir) {
1572 case VIOIF_NETQ_RX:
1573 rxc = netq->netq_ctx;
1574 netq->netq_ctx = NULL;
1575 kmem_free(rxc, sizeof(*rxc));
1576 break;
1577 case VIOIF_NETQ_TX:
1578 txc = netq->netq_ctx;
1579 netq->netq_ctx = NULL;
1580 softint_disestablish(txc->txc_deferred_transmit);
1581 pcq_destroy(txc->txc_intrq);
1582 kmem_free(txc, sizeof(*txc));
1583 break;
1584 }
1585
1586 softint_disestablish(netq->netq_softint);
1587 virtio_free_vq(vsc, netq->netq_vq);
1588 mutex_destroy(&netq->netq_lock);
1589 netq->netq_vq = NULL;
1590 }
1591
1592 static void
1593 vioif_net_sched_handle(struct vioif_softc *sc, struct vioif_netqueue *netq)
1594 {
1595
1596 KASSERT(mutex_owned(&netq->netq_lock));
1597 KASSERT(!netq->netq_stopping);
1598
1599 if (netq->netq_workqueue) {
1600 vioif_work_add(sc->sc_txrx_workqueue, &netq->netq_work);
1601 } else {
1602 softint_schedule(netq->netq_softint);
1603 }
1604 }
1605
1606 static int
1607 vioif_net_load_mbuf(struct virtio_softc *vsc, struct vioif_net_map *map,
1608 struct mbuf *m, int dma_flags)
1609 {
1610 int r;
1611
1612 KASSERT(map->vnm_mbuf == NULL);
1613
1614 r = bus_dmamap_load_mbuf(virtio_dmat(vsc),
1615 map->vnm_mbuf_map, m, dma_flags | BUS_DMA_NOWAIT);
1616 if (r == 0) {
1617 map->vnm_mbuf = m;
1618 }
1619
1620 return r;
1621 }
1622
1623 static void
1624 vioif_net_unload_mbuf(struct virtio_softc *vsc, struct vioif_net_map *map)
1625 {
1626
1627 KASSERT(map->vnm_mbuf != NULL);
1628 bus_dmamap_unload(virtio_dmat(vsc), map->vnm_mbuf_map);
1629 map->vnm_mbuf = NULL;
1630 }
1631
1632 static int
1633 vioif_net_enqueue(struct virtio_softc *vsc, struct virtqueue *vq,
1634 int slot, struct vioif_net_map *map, int dma_ops, bool is_write)
1635 {
1636 int r;
1637
1638 KASSERT(map->vnm_mbuf != NULL);
1639
1640 /* This should actually never fail */
1641 r = virtio_enqueue_reserve(vsc, vq, slot,
1642 map->vnm_mbuf_map->dm_nsegs + 1);
1643 if (r != 0) {
1644 /* slot already freed by virtio_enqueue_reserve */
1645 return r;
1646 }
1647
1648 bus_dmamap_sync(virtio_dmat(vsc), map->vnm_mbuf_map,
1649 0, map->vnm_mbuf_map->dm_mapsize, dma_ops);
1650 bus_dmamap_sync(virtio_dmat(vsc), map->vnm_hdr_map,
1651 0, map->vnm_hdr_map->dm_mapsize, dma_ops);
1652
1653 virtio_enqueue(vsc, vq, slot, map->vnm_hdr_map, is_write);
1654 virtio_enqueue(vsc, vq, slot, map->vnm_mbuf_map, is_write);
1655 virtio_enqueue_commit(vsc, vq, slot, false);
1656
1657 return 0;
1658 }
1659
1660 static int
1661 vioif_net_enqueue_tx(struct virtio_softc *vsc, struct virtqueue *vq,
1662 int slot, struct vioif_net_map *map)
1663 {
1664
1665 return vioif_net_enqueue(vsc, vq, slot, map,
1666 BUS_DMASYNC_PREWRITE, true);
1667 }
1668
1669 static int
1670 vioif_net_enqueue_rx(struct virtio_softc *vsc, struct virtqueue *vq,
1671 int slot, struct vioif_net_map *map)
1672 {
1673
1674 return vioif_net_enqueue(vsc, vq, slot, map,
1675 BUS_DMASYNC_PREREAD, false);
1676 }
1677
1678 static struct mbuf *
1679 vioif_net_dequeue_commit(struct virtio_softc *vsc, struct virtqueue *vq,
1680 int slot, struct vioif_net_map *map, int dma_flags)
1681 {
1682 struct mbuf *m;
1683
1684 m = map->vnm_mbuf;
1685 KASSERT(m != NULL);
1686 map->vnm_mbuf = NULL;
1687
1688 bus_dmamap_sync(virtio_dmat(vsc), map->vnm_hdr_map,
1689 0, map->vnm_hdr_map->dm_mapsize, dma_flags);
1690 bus_dmamap_sync(virtio_dmat(vsc), map->vnm_mbuf_map,
1691 0, map->vnm_mbuf_map->dm_mapsize, dma_flags);
1692
1693 bus_dmamap_unload(virtio_dmat(vsc), map->vnm_mbuf_map);
1694 virtio_dequeue_commit(vsc, vq, slot);
1695
1696 return m;
1697 }
1698
1699 static void
1700 vioif_net_intr_enable(struct vioif_softc *sc, struct virtio_softc *vsc)
1701 {
1702 struct vioif_netqueue *netq;
1703 size_t i, act_qnum;
1704 int enqueued;
1705
1706 act_qnum = sc->sc_act_nvq_pairs * 2;
1707 for (i = 0; i < act_qnum; i++) {
1708 netq = &sc->sc_netqs[i];
1709
1710 KASSERT(!netq->netq_stopping);
1711 KASSERT(!netq->netq_running_handle);
1712
1713 enqueued = virtio_start_vq_intr(vsc, netq->netq_vq);
1714 if (enqueued != 0) {
1715 virtio_stop_vq_intr(vsc, netq->netq_vq);
1716
1717 mutex_enter(&netq->netq_lock);
1718 netq->netq_running_handle = true;
1719 vioif_net_sched_handle(sc, netq);
1720 mutex_exit(&netq->netq_lock);
1721 }
1722 }
1723 }
1724
1725 static void
1726 vioif_net_intr_disable(struct vioif_softc *sc, struct virtio_softc *vsc)
1727 {
1728 struct vioif_netqueue *netq;
1729 size_t i, act_qnum;
1730
1731 act_qnum = sc->sc_act_nvq_pairs * 2;
1732 for (i = 0; i < act_qnum; i++) {
1733 netq = &sc->sc_netqs[i];
1734
1735 virtio_stop_vq_intr(vsc, netq->netq_vq);
1736 }
1737 }
1738
1739 /*
1740 * Receive implementation
1741 */
1742 /* enqueue mbufs to receive slots */
1743 static void
1744 vioif_populate_rx_mbufs_locked(struct vioif_softc *sc, struct vioif_netqueue *netq)
1745 {
1746 struct virtqueue *vq = netq->netq_vq;
1747 struct virtio_softc *vsc = vq->vq_owner;
1748 struct vioif_rx_context *rxc;
1749 struct vioif_net_map *map;
1750 struct mbuf *m;
1751 int i, r, ndone = 0;
1752
1753 KASSERT(mutex_owned(&netq->netq_lock));
1754
1755 rxc = netq->netq_ctx;
1756
1757 for (i = 0; i < vq->vq_num; i++) {
1758 int slot;
1759 r = virtio_enqueue_prep(vsc, vq, &slot);
1760 if (r == EAGAIN)
1761 break;
1762 if (__predict_false(r != 0))
1763 panic("enqueue_prep for rx buffers");
1764
1765 MGETHDR(m, M_DONTWAIT, MT_DATA);
1766 if (m == NULL) {
1767 virtio_enqueue_abort(vsc, vq, slot);
1768 rxc->rxc_mbuf_enobufs.ev_count++;
1769 break;
1770 }
1771 MCLAIM(m, &sc->sc_ethercom.ec_rx_mowner);
1772 MCLGET(m, M_DONTWAIT);
1773 if ((m->m_flags & M_EXT) == 0) {
1774 virtio_enqueue_abort(vsc, vq, slot);
1775 m_freem(m);
1776 rxc->rxc_mbuf_enobufs.ev_count++;
1777 break;
1778 }
1779
1780 m->m_len = m->m_pkthdr.len = MCLBYTES;
1781 m_adj(m, ETHER_ALIGN);
1782
1783 map = &netq->netq_maps[slot];
1784 r = vioif_net_load_mbuf(vsc, map, m, BUS_DMA_READ);
1785 if (r != 0) {
1786 virtio_enqueue_abort(vsc, vq, slot);
1787 m_freem(m);
1788 netq->netq_mbuf_load_failed.ev_count++;
1789 break;
1790 }
1791
1792 r = vioif_net_enqueue_rx(vsc, vq, slot, map);
1793 if (r != 0) {
1794 vioif_net_unload_mbuf(vsc, map);
1795 netq->netq_enqueue_failed.ev_count++;
1796 m_freem(m);
1797 /* slot already freed by vioif_net_enqueue_rx */
1798 break;
1799 }
1800
1801 ndone++;
1802 }
1803
1804 if (ndone > 0)
1805 vioif_notify(vsc, vq);
1806 }
1807
1808 /* dequeue received packets */
1809 static bool
1810 vioif_rx_deq_locked(struct vioif_softc *sc, struct virtio_softc *vsc,
1811 struct vioif_netqueue *netq, u_int limit, size_t *ndeqp)
1812 {
1813 struct virtqueue *vq = netq->netq_vq;
1814 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1815 struct vioif_net_map *map;
1816 struct mbuf *m;
1817 int slot, len;
1818 bool more;
1819 size_t ndeq;
1820
1821 KASSERT(mutex_owned(&netq->netq_lock));
1822
1823 more = false;
1824 ndeq = 0;
1825
1826 if (virtio_vq_is_enqueued(vsc, vq) == false)
1827 goto done;
1828
1829 for (;;ndeq++) {
1830 if (ndeq >= limit) {
1831 more = true;
1832 break;
1833 }
1834
1835 if (virtio_dequeue(vsc, vq, &slot, &len) != 0)
1836 break;
1837
1838 map = &netq->netq_maps[slot];
1839 KASSERT(map->vnm_mbuf != NULL);
1840 m = vioif_net_dequeue_commit(vsc, vq, slot,
1841 map, BUS_DMASYNC_POSTREAD);
1842 KASSERT(m != NULL);
1843
1844 m->m_len = m->m_pkthdr.len = len - sc->sc_hdr_size;
1845 m_set_rcvif(m, ifp);
1846 if_percpuq_enqueue(ifp->if_percpuq, m);
1847 }
1848
1849 done:
1850 if (ndeqp != NULL)
1851 *ndeqp = ndeq;
1852
1853 return more;
1854 }
1855
1856 static void
1857 vioif_rx_queue_clear(struct vioif_softc *sc, struct virtio_softc *vsc,
1858 struct vioif_netqueue *netq)
1859 {
1860 struct vioif_net_map *map;
1861 struct mbuf *m;
1862 unsigned int i, vq_num;
1863 bool more;
1864
1865 mutex_enter(&netq->netq_lock);
1866
1867 vq_num = netq->netq_vq->vq_num;
1868 for (;;) {
1869 more = vioif_rx_deq_locked(sc, vsc, netq, vq_num, NULL);
1870 if (more == false)
1871 break;
1872 }
1873
1874 for (i = 0; i < vq_num; i++) {
1875 map = &netq->netq_maps[i];
1876
1877 m = map->vnm_mbuf;
1878 if (m == NULL)
1879 continue;
1880
1881 vioif_net_unload_mbuf(vsc, map);
1882 m_freem(m);
1883 }
1884 mutex_exit(&netq->netq_lock);
1885 }
1886
1887 static void
1888 vioif_rx_handle_locked(void *xnetq, u_int limit)
1889 {
1890 struct vioif_netqueue *netq = xnetq;
1891 struct virtqueue *vq = netq->netq_vq;
1892 struct virtio_softc *vsc = vq->vq_owner;
1893 struct vioif_softc *sc = device_private(virtio_child(vsc));
1894 bool more;
1895 int enqueued;
1896 size_t ndeq;
1897
1898 KASSERT(mutex_owned(&netq->netq_lock));
1899 KASSERT(!netq->netq_stopping);
1900
1901 more = vioif_rx_deq_locked(sc, vsc, netq, limit, &ndeq);
1902 if (ndeq > 0)
1903 vioif_populate_rx_mbufs_locked(sc, netq);
1904
1905 if (more) {
1906 vioif_net_sched_handle(sc, netq);
1907 return;
1908 }
1909
1910 enqueued = virtio_start_vq_intr(vsc, netq->netq_vq);
1911 if (enqueued != 0) {
1912 virtio_stop_vq_intr(vsc, netq->netq_vq);
1913 vioif_net_sched_handle(sc, netq);
1914 return;
1915 }
1916
1917 netq->netq_running_handle = false;
1918 }
1919
1920 static int
1921 vioif_rx_intr(void *arg)
1922 {
1923 struct vioif_netqueue *netq = arg;
1924 struct virtqueue *vq = netq->netq_vq;
1925 struct virtio_softc *vsc = vq->vq_owner;
1926 struct vioif_softc *sc = device_private(virtio_child(vsc));
1927 u_int limit;
1928
1929 mutex_enter(&netq->netq_lock);
1930
1931 /* handler is already running in softint/workqueue */
1932 if (netq->netq_running_handle)
1933 goto done;
1934
1935 if (netq->netq_stopping)
1936 goto done;
1937
1938 netq->netq_running_handle = true;
1939
1940 limit = sc->sc_rx_intr_process_limit;
1941 virtio_stop_vq_intr(vsc, vq);
1942 vioif_rx_handle_locked(netq, limit);
1943
1944 done:
1945 mutex_exit(&netq->netq_lock);
1946 return 1;
1947 }
1948
1949 static void
1950 vioif_rx_handle(void *xnetq)
1951 {
1952 struct vioif_netqueue *netq = xnetq;
1953 struct virtqueue *vq = netq->netq_vq;
1954 struct virtio_softc *vsc = vq->vq_owner;
1955 struct vioif_softc *sc = device_private(virtio_child(vsc));
1956 u_int limit;
1957
1958 mutex_enter(&netq->netq_lock);
1959
1960 KASSERT(netq->netq_running_handle);
1961
1962 if (netq->netq_stopping) {
1963 netq->netq_running_handle = false;
1964 goto done;
1965 }
1966
1967 limit = sc->sc_rx_process_limit;
1968 vioif_rx_handle_locked(netq, limit);
1969
1970 done:
1971 mutex_exit(&netq->netq_lock);
1972 }
1973
1974 /*
1975 * Transmission implementation
1976 */
1977 /* enqueue mbufs to send */
1978 static void
1979 vioif_send_common_locked(struct ifnet *ifp, struct vioif_netqueue *netq,
1980 bool is_transmit)
1981 {
1982 struct vioif_softc *sc = ifp->if_softc;
1983 struct virtio_softc *vsc = sc->sc_virtio;
1984 struct virtqueue *vq = netq->netq_vq;
1985 struct vioif_tx_context *txc;
1986 struct vioif_net_map *map;
1987 struct mbuf *m;
1988 int queued = 0;
1989
1990 KASSERT(mutex_owned(&netq->netq_lock));
1991
1992 if (netq->netq_stopping ||
1993 !ISSET(ifp->if_flags, IFF_RUNNING))
1994 return;
1995
1996 txc = netq->netq_ctx;
1997
1998 if (!txc->txc_link_active ||
1999 txc->txc_no_free_slots)
2000 return;
2001
2002 for (;;) {
2003 int slot, r;
2004 r = virtio_enqueue_prep(vsc, vq, &slot);
2005 if (r == EAGAIN) {
2006 txc->txc_no_free_slots = true;
2007 break;
2008 }
2009 if (__predict_false(r != 0))
2010 panic("enqueue_prep for tx buffers");
2011
2012 if (is_transmit)
2013 m = pcq_get(txc->txc_intrq);
2014 else
2015 IFQ_DEQUEUE(&ifp->if_snd, m);
2016
2017 if (m == NULL) {
2018 virtio_enqueue_abort(vsc, vq, slot);
2019 break;
2020 }
2021
2022 map = &netq->netq_maps[slot];
2023 KASSERT(map->vnm_mbuf == NULL);
2024
2025 r = vioif_net_load_mbuf(vsc, map, m, BUS_DMA_WRITE);
2026 if (r != 0) {
2027 /* maybe just too fragmented */
2028 struct mbuf *newm;
2029
2030 newm = m_defrag(m, M_NOWAIT);
2031 if (newm != NULL) {
2032 m = newm;
2033 r = vioif_net_load_mbuf(vsc, map, m,
2034 BUS_DMA_WRITE);
2035 } else {
2036 txc->txc_defrag_failed.ev_count++;
2037 r = -1;
2038 }
2039
2040 if (r != 0) {
2041 netq->netq_mbuf_load_failed.ev_count++;
2042 m_freem(m);
2043 if_statinc(ifp, if_oerrors);
2044 virtio_enqueue_abort(vsc, vq, slot);
2045 continue;
2046 }
2047 }
2048
2049 memset(map->vnm_hdr, 0, sc->sc_hdr_size);
2050
2051 r = vioif_net_enqueue_tx(vsc, vq, slot, map);
2052 if (r != 0) {
2053 netq->netq_enqueue_failed.ev_count++;
2054 vioif_net_unload_mbuf(vsc, map);
2055 m_freem(m);
2056 /* slot already freed by vioif_net_enqueue_tx */
2057
2058 if_statinc(ifp, if_oerrors);
2059 continue;
2060 }
2061
2062 queued++;
2063 bpf_mtap(ifp, m, BPF_D_OUT);
2064 }
2065
2066 if (queued > 0) {
2067 vioif_notify(vsc, vq);
2068 ifp->if_timer = 5;
2069 }
2070 }
2071
2072 /* dequeue sent mbufs */
2073 static bool
2074 vioif_tx_deq_locked(struct vioif_softc *sc, struct virtio_softc *vsc,
2075 struct vioif_netqueue *netq, u_int limit, size_t *ndeqp)
2076 {
2077 struct virtqueue *vq = netq->netq_vq;
2078 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2079 struct vioif_net_map *map;
2080 struct mbuf *m;
2081 int slot, len;
2082 bool more;
2083 size_t ndeq;
2084
2085 KASSERT(mutex_owned(&netq->netq_lock));
2086
2087 more = false;
2088 ndeq = 0;
2089
2090 if (virtio_vq_is_enqueued(vsc, vq) == false)
2091 goto done;
2092
2093 for (;;ndeq++) {
2094 if (limit-- == 0) {
2095 more = true;
2096 break;
2097 }
2098
2099 if (virtio_dequeue(vsc, vq, &slot, &len) != 0)
2100 break;
2101
2102 map = &netq->netq_maps[slot];
2103 KASSERT(map->vnm_mbuf != NULL);
2104 m = vioif_net_dequeue_commit(vsc, vq, slot,
2105 map, BUS_DMASYNC_POSTWRITE);
2106 KASSERT(m != NULL);
2107
2108 if_statinc(ifp, if_opackets);
2109 m_freem(m);
2110 }
2111
2112 done:
2113 if (ndeqp != NULL)
2114 *ndeqp = ndeq;
2115 return more;
2116 }
2117
2118 static void
2119 vioif_tx_queue_clear(struct vioif_softc *sc, struct virtio_softc *vsc,
2120 struct vioif_netqueue *netq)
2121 {
2122 struct vioif_tx_context *txc;
2123 struct vioif_net_map *map;
2124 struct mbuf *m;
2125 unsigned int i, vq_num;
2126 bool more;
2127
2128 mutex_enter(&netq->netq_lock);
2129
2130 txc = netq->netq_ctx;
2131 vq_num = netq->netq_vq->vq_num;
2132
2133 for (;;) {
2134 more = vioif_tx_deq_locked(sc, vsc, netq, vq_num, NULL);
2135 if (more == false)
2136 break;
2137 }
2138
2139 for (i = 0; i < vq_num; i++) {
2140 map = &netq->netq_maps[i];
2141
2142 m = map->vnm_mbuf;
2143 if (m == NULL)
2144 continue;
2145
2146 vioif_net_unload_mbuf(vsc, map);
2147 m_freem(m);
2148 }
2149
2150 txc->txc_no_free_slots = false;
2151
2152 mutex_exit(&netq->netq_lock);
2153 }
2154
2155 static void
2156 vioif_start_locked(struct ifnet *ifp, struct vioif_netqueue *netq)
2157 {
2158
2159 /*
2160 * ifp->if_obytes and ifp->if_omcasts are added in if_transmit()@if.c.
2161 */
2162 vioif_send_common_locked(ifp, netq, false);
2163
2164 }
2165
2166 static void
2167 vioif_transmit_locked(struct ifnet *ifp, struct vioif_netqueue *netq)
2168 {
2169
2170 vioif_send_common_locked(ifp, netq, true);
2171 }
2172
2173 static void
2174 vioif_deferred_transmit(void *arg)
2175 {
2176 struct vioif_netqueue *netq = arg;
2177 struct virtio_softc *vsc = netq->netq_vq->vq_owner;
2178 struct vioif_softc *sc = device_private(virtio_child(vsc));
2179 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2180
2181 mutex_enter(&netq->netq_lock);
2182 vioif_send_common_locked(ifp, netq, true);
2183 mutex_exit(&netq->netq_lock);
2184 }
2185
2186 static void
2187 vioif_tx_handle_locked(struct vioif_netqueue *netq, u_int limit)
2188 {
2189 struct virtqueue *vq = netq->netq_vq;
2190 struct vioif_tx_context *txc = netq->netq_ctx;
2191 struct virtio_softc *vsc = vq->vq_owner;
2192 struct vioif_softc *sc = device_private(virtio_child(vsc));
2193 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2194 bool more;
2195 int enqueued;
2196 size_t ndeq;
2197
2198 KASSERT(mutex_owned(&netq->netq_lock));
2199 KASSERT(!netq->netq_stopping);
2200
2201 more = vioif_tx_deq_locked(sc, vsc, netq, limit, &ndeq);
2202 if (txc->txc_no_free_slots && ndeq > 0) {
2203 txc->txc_no_free_slots = false;
2204 softint_schedule(txc->txc_deferred_transmit);
2205 }
2206
2207 if (more) {
2208 vioif_net_sched_handle(sc, netq);
2209 return;
2210 }
2211
2212 enqueued = (virtio_features(vsc) & VIRTIO_F_RING_EVENT_IDX) ?
2213 virtio_postpone_intr_smart(vsc, vq):
2214 virtio_start_vq_intr(vsc, vq);
2215 if (enqueued != 0) {
2216 virtio_stop_vq_intr(vsc, vq);
2217 vioif_net_sched_handle(sc, netq);
2218 return;
2219 }
2220
2221 netq->netq_running_handle = false;
2222
2223 /* for ALTQ */
2224 if (netq == &sc->sc_netqs[VIOIF_NETQ_TXQID(0)])
2225 if_schedule_deferred_start(ifp);
2226
2227 softint_schedule(txc->txc_deferred_transmit);
2228 }
2229
2230 static int
2231 vioif_tx_intr(void *arg)
2232 {
2233 struct vioif_netqueue *netq = arg;
2234 struct virtqueue *vq = netq->netq_vq;
2235 struct virtio_softc *vsc = vq->vq_owner;
2236 struct vioif_softc *sc = device_private(virtio_child(vsc));
2237 u_int limit;
2238
2239 mutex_enter(&netq->netq_lock);
2240
2241 /* tx handler is already running in softint/workqueue */
2242 if (netq->netq_running_handle)
2243 goto done;
2244
2245 if (netq->netq_stopping)
2246 goto done;
2247
2248 netq->netq_running_handle = true;
2249
2250 virtio_stop_vq_intr(vsc, vq);
2251 netq->netq_workqueue = sc->sc_txrx_workqueue_sysctl;
2252 limit = sc->sc_tx_intr_process_limit;
2253 vioif_tx_handle_locked(netq, limit);
2254
2255 done:
2256 mutex_exit(&netq->netq_lock);
2257 return 1;
2258 }
2259
2260 static void
2261 vioif_tx_handle(void *xnetq)
2262 {
2263 struct vioif_netqueue *netq = xnetq;
2264 struct virtqueue *vq = netq->netq_vq;
2265 struct virtio_softc *vsc = vq->vq_owner;
2266 struct vioif_softc *sc = device_private(virtio_child(vsc));
2267 u_int limit;
2268
2269 mutex_enter(&netq->netq_lock);
2270
2271 KASSERT(netq->netq_running_handle);
2272
2273 if (netq->netq_stopping) {
2274 netq->netq_running_handle = false;
2275 goto done;
2276 }
2277
2278 limit = sc->sc_tx_process_limit;
2279 vioif_tx_handle_locked(netq, limit);
2280
2281 done:
2282 mutex_exit(&netq->netq_lock);
2283 }
2284
2285 /*
2286 * Control vq
2287 */
2288 /* issue a VIRTIO_NET_CTRL_RX class command and wait for completion */
2289 static void
2290 vioif_ctrl_acquire(struct vioif_softc *sc)
2291 {
2292 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
2293
2294 mutex_enter(&ctrlq->ctrlq_wait_lock);
2295 while (ctrlq->ctrlq_inuse != FREE)
2296 cv_wait(&ctrlq->ctrlq_wait, &ctrlq->ctrlq_wait_lock);
2297 ctrlq->ctrlq_inuse = INUSE;
2298 ctrlq->ctrlq_owner = curlwp;
2299 mutex_exit(&ctrlq->ctrlq_wait_lock);
2300 }
2301
2302 static void
2303 vioif_ctrl_release(struct vioif_softc *sc)
2304 {
2305 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
2306
2307 KASSERT(ctrlq->ctrlq_inuse != FREE);
2308 KASSERT(ctrlq->ctrlq_owner == curlwp);
2309
2310 mutex_enter(&ctrlq->ctrlq_wait_lock);
2311 ctrlq->ctrlq_inuse = FREE;
2312 ctrlq->ctrlq_owner = NULL;
2313 cv_signal(&ctrlq->ctrlq_wait);
2314 mutex_exit(&ctrlq->ctrlq_wait_lock);
2315 }
2316
2317 static int
2318 vioif_ctrl_load_cmdspec(struct vioif_softc *sc,
2319 struct vioif_ctrl_cmdspec *specs, int nspecs)
2320 {
2321 struct virtio_softc *vsc = sc->sc_virtio;
2322 int i, r, loaded;
2323
2324 loaded = 0;
2325 for (i = 0; i < nspecs; i++) {
2326 r = bus_dmamap_load(virtio_dmat(vsc),
2327 specs[i].dmamap, specs[i].buf, specs[i].bufsize,
2328 NULL, BUS_DMA_WRITE | BUS_DMA_NOWAIT);
2329 if (r) {
2330 sc->sc_ctrlq.ctrlq_cmd_load_failed.ev_count++;
2331 goto err;
2332 }
2333 loaded++;
2334
2335 }
2336
2337 return r;
2338
2339 err:
2340 for (i = 0; i < loaded; i++) {
2341 bus_dmamap_unload(virtio_dmat(vsc), specs[i].dmamap);
2342 }
2343
2344 return r;
2345 }
2346
2347 static void
2348 vioif_ctrl_unload_cmdspec(struct vioif_softc *sc,
2349 struct vioif_ctrl_cmdspec *specs, int nspecs)
2350 {
2351 struct virtio_softc *vsc = sc->sc_virtio;
2352 int i;
2353
2354 for (i = 0; i < nspecs; i++) {
2355 bus_dmamap_unload(virtio_dmat(vsc), specs[i].dmamap);
2356 }
2357 }
2358
2359 static int
2360 vioif_ctrl_send_command(struct vioif_softc *sc, uint8_t class, uint8_t cmd,
2361 struct vioif_ctrl_cmdspec *specs, int nspecs)
2362 {
2363 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
2364 struct virtqueue *vq = ctrlq->ctrlq_vq;
2365 struct virtio_softc *vsc = sc->sc_virtio;
2366 int i, r, slot;
2367
2368 ctrlq->ctrlq_cmd->class = class;
2369 ctrlq->ctrlq_cmd->command = cmd;
2370
2371 bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_cmd_dmamap,
2372 0, sizeof(struct virtio_net_ctrl_cmd), BUS_DMASYNC_PREWRITE);
2373 for (i = 0; i < nspecs; i++) {
2374 bus_dmamap_sync(virtio_dmat(vsc), specs[i].dmamap,
2375 0, specs[i].bufsize, BUS_DMASYNC_PREWRITE);
2376 }
2377 bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_status_dmamap,
2378 0, sizeof(struct virtio_net_ctrl_status), BUS_DMASYNC_PREREAD);
2379
2380 /* we need to explicitly (re)start vq intr when using RING EVENT IDX */
2381 if (virtio_features(vsc) & VIRTIO_F_RING_EVENT_IDX)
2382 virtio_start_vq_intr(vsc, ctrlq->ctrlq_vq);
2383
2384 r = virtio_enqueue_prep(vsc, vq, &slot);
2385 if (r != 0)
2386 panic("%s: control vq busy!?", device_xname(sc->sc_dev));
2387 r = virtio_enqueue_reserve(vsc, vq, slot, nspecs + 2);
2388 if (r != 0)
2389 panic("%s: control vq busy!?", device_xname(sc->sc_dev));
2390 virtio_enqueue(vsc, vq, slot, ctrlq->ctrlq_cmd_dmamap, true);
2391 for (i = 0; i < nspecs; i++) {
2392 virtio_enqueue(vsc, vq, slot, specs[i].dmamap, true);
2393 }
2394 virtio_enqueue(vsc, vq, slot, ctrlq->ctrlq_status_dmamap, false);
2395 virtio_enqueue_commit(vsc, vq, slot, true);
2396
2397 /* wait for done */
2398 mutex_enter(&ctrlq->ctrlq_wait_lock);
2399 while (ctrlq->ctrlq_inuse != DONE)
2400 cv_wait(&ctrlq->ctrlq_wait, &ctrlq->ctrlq_wait_lock);
2401 mutex_exit(&ctrlq->ctrlq_wait_lock);
2402 /* already dequeued */
2403
2404 bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_cmd_dmamap, 0,
2405 sizeof(struct virtio_net_ctrl_cmd), BUS_DMASYNC_POSTWRITE);
2406 for (i = 0; i < nspecs; i++) {
2407 bus_dmamap_sync(virtio_dmat(vsc), specs[i].dmamap, 0,
2408 specs[i].bufsize, BUS_DMASYNC_POSTWRITE);
2409 }
2410 bus_dmamap_sync(virtio_dmat(vsc), ctrlq->ctrlq_status_dmamap, 0,
2411 sizeof(struct virtio_net_ctrl_status), BUS_DMASYNC_POSTREAD);
2412
2413 if (ctrlq->ctrlq_status->ack == VIRTIO_NET_OK)
2414 r = 0;
2415 else {
2416 device_printf(sc->sc_dev, "failed setting rx mode\n");
2417 sc->sc_ctrlq.ctrlq_cmd_failed.ev_count++;
2418 r = EIO;
2419 }
2420
2421 return r;
2422 }
2423
2424 /* ctrl vq interrupt; wake up the command issuer */
2425 static int
2426 vioif_ctrl_intr(void *arg)
2427 {
2428 struct vioif_ctrlqueue *ctrlq = arg;
2429 struct virtqueue *vq = ctrlq->ctrlq_vq;
2430 struct virtio_softc *vsc = vq->vq_owner;
2431 int r, slot;
2432
2433 if (virtio_vq_is_enqueued(vsc, vq) == false)
2434 return 0;
2435
2436 r = virtio_dequeue(vsc, vq, &slot, NULL);
2437 if (r == ENOENT)
2438 return 0;
2439 virtio_dequeue_commit(vsc, vq, slot);
2440
2441 mutex_enter(&ctrlq->ctrlq_wait_lock);
2442 ctrlq->ctrlq_inuse = DONE;
2443 cv_signal(&ctrlq->ctrlq_wait);
2444 mutex_exit(&ctrlq->ctrlq_wait_lock);
2445
2446 return 1;
2447 }
2448
2449 static int
2450 vioif_ctrl_rx(struct vioif_softc *sc, int cmd, bool onoff)
2451 {
2452 struct virtio_net_ctrl_rx *rx = sc->sc_ctrlq.ctrlq_rx;
2453 struct vioif_ctrl_cmdspec specs[1];
2454 int r;
2455
2456 if (!sc->sc_has_ctrl)
2457 return ENOTSUP;
2458
2459 vioif_ctrl_acquire(sc);
2460
2461 rx->onoff = onoff;
2462 specs[0].dmamap = sc->sc_ctrlq.ctrlq_rx_dmamap;
2463 specs[0].buf = rx;
2464 specs[0].bufsize = sizeof(*rx);
2465
2466 r = vioif_ctrl_send_command(sc, VIRTIO_NET_CTRL_RX, cmd,
2467 specs, __arraycount(specs));
2468
2469 vioif_ctrl_release(sc);
2470 return r;
2471 }
2472
2473 static int
2474 vioif_set_promisc(struct vioif_softc *sc, bool onoff)
2475 {
2476 return vioif_ctrl_rx(sc, VIRTIO_NET_CTRL_RX_PROMISC, onoff);
2477 }
2478
2479 static int
2480 vioif_set_allmulti(struct vioif_softc *sc, bool onoff)
2481 {
2482 return vioif_ctrl_rx(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, onoff);
2483 }
2484
2485 static int
2486 vioif_ctrl_mq_vq_pairs_set(struct vioif_softc *sc, int nvq_pairs)
2487 {
2488 struct virtio_net_ctrl_mq *mq = sc->sc_ctrlq.ctrlq_mq;
2489 struct vioif_ctrl_cmdspec specs[1];
2490 int r;
2491
2492 if (!sc->sc_has_ctrl)
2493 return ENOTSUP;
2494
2495 if (nvq_pairs <= 1)
2496 return EINVAL;
2497
2498 vioif_ctrl_acquire(sc);
2499
2500 mq->virtqueue_pairs = virtio_rw16(sc->sc_virtio, nvq_pairs);
2501 specs[0].dmamap = sc->sc_ctrlq.ctrlq_mq_dmamap;
2502 specs[0].buf = mq;
2503 specs[0].bufsize = sizeof(*mq);
2504
2505 r = vioif_ctrl_send_command(sc,
2506 VIRTIO_NET_CTRL_MQ, VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET,
2507 specs, __arraycount(specs));
2508
2509 vioif_ctrl_release(sc);
2510
2511 return r;
2512 }
2513
2514 static int
2515 vioif_set_mac_addr(struct vioif_softc *sc)
2516 {
2517 struct virtio_net_ctrl_mac_addr *ma =
2518 sc->sc_ctrlq.ctrlq_mac_addr;
2519 struct vioif_ctrl_cmdspec specs[1];
2520 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2521 int nspecs = __arraycount(specs);
2522 uint64_t features;
2523 int r;
2524 size_t i;
2525
2526 if (!sc->sc_has_ctrl)
2527 return ENOTSUP;
2528
2529 if (memcmp(CLLADDR(ifp->if_sadl), sc->sc_mac,
2530 ETHER_ADDR_LEN) == 0) {
2531 return 0;
2532 }
2533
2534 memcpy(sc->sc_mac, CLLADDR(ifp->if_sadl), ETHER_ADDR_LEN);
2535
2536 features = virtio_features(sc->sc_virtio);
2537 if (features & VIRTIO_NET_F_CTRL_MAC_ADDR) {
2538 vioif_ctrl_acquire(sc);
2539
2540 memcpy(ma->mac, sc->sc_mac, ETHER_ADDR_LEN);
2541 specs[0].dmamap = sc->sc_ctrlq.ctrlq_mac_addr_dmamap;
2542 specs[0].buf = ma;
2543 specs[0].bufsize = sizeof(*ma);
2544
2545 r = vioif_ctrl_send_command(sc,
2546 VIRTIO_NET_CTRL_MAC, VIRTIO_NET_CTRL_MAC_ADDR_SET,
2547 specs, nspecs);
2548
2549 vioif_ctrl_release(sc);
2550 } else {
2551 for (i = 0; i < __arraycount(sc->sc_mac); i++) {
2552 virtio_write_device_config_1(sc->sc_virtio,
2553 VIRTIO_NET_CONFIG_MAC + i, sc->sc_mac[i]);
2554 }
2555 r = 0;
2556 }
2557
2558 return r;
2559 }
2560
2561 static int
2562 vioif_set_rx_filter(struct vioif_softc *sc)
2563 {
2564 /* filter already set in ctrlq->ctrlq_mac_tbl */
2565 struct virtio_softc *vsc = sc->sc_virtio;
2566 struct virtio_net_ctrl_mac_tbl *mac_tbl_uc, *mac_tbl_mc;
2567 struct vioif_ctrl_cmdspec specs[2];
2568 int nspecs = __arraycount(specs);
2569 int r;
2570
2571 mac_tbl_uc = sc->sc_ctrlq.ctrlq_mac_tbl_uc;
2572 mac_tbl_mc = sc->sc_ctrlq.ctrlq_mac_tbl_mc;
2573
2574 if (!sc->sc_has_ctrl)
2575 return ENOTSUP;
2576
2577 vioif_ctrl_acquire(sc);
2578
2579 specs[0].dmamap = sc->sc_ctrlq.ctrlq_tbl_uc_dmamap;
2580 specs[0].buf = mac_tbl_uc;
2581 specs[0].bufsize = sizeof(*mac_tbl_uc)
2582 + (ETHER_ADDR_LEN * virtio_rw32(vsc, mac_tbl_uc->nentries));
2583
2584 specs[1].dmamap = sc->sc_ctrlq.ctrlq_tbl_mc_dmamap;
2585 specs[1].buf = mac_tbl_mc;
2586 specs[1].bufsize = sizeof(*mac_tbl_mc)
2587 + (ETHER_ADDR_LEN * virtio_rw32(vsc, mac_tbl_mc->nentries));
2588
2589 r = vioif_ctrl_load_cmdspec(sc, specs, nspecs);
2590 if (r != 0)
2591 goto out;
2592
2593 r = vioif_ctrl_send_command(sc,
2594 VIRTIO_NET_CTRL_MAC, VIRTIO_NET_CTRL_MAC_TABLE_SET,
2595 specs, nspecs);
2596
2597 vioif_ctrl_unload_cmdspec(sc, specs, nspecs);
2598
2599 out:
2600 vioif_ctrl_release(sc);
2601
2602 return r;
2603 }
2604
2605 /*
2606 * If multicast filter small enough (<=MAXENTRIES) set rx filter
2607 * If large multicast filter exist use ALLMULTI
2608 * If setting rx filter fails fall back to ALLMULTI
2609 */
2610 static int
2611 vioif_rx_filter(struct vioif_softc *sc)
2612 {
2613 struct virtio_softc *vsc = sc->sc_virtio;
2614 struct ethercom *ec = &sc->sc_ethercom;
2615 struct ifnet *ifp = &ec->ec_if;
2616 struct ether_multi *enm;
2617 struct ether_multistep step;
2618 struct vioif_ctrlqueue *ctrlq = &sc->sc_ctrlq;
2619 int nentries;
2620 bool allmulti = 0;
2621 int r;
2622
2623 if (!sc->sc_has_ctrl) {
2624 goto set_ifflags;
2625 }
2626
2627 memcpy(ctrlq->ctrlq_mac_tbl_uc->macs[0],
2628 CLLADDR(ifp->if_sadl), ETHER_ADDR_LEN);
2629
2630 nentries = 0;
2631 allmulti = false;
2632
2633 ETHER_LOCK(ec);
2634 for (ETHER_FIRST_MULTI(step, ec, enm); enm != NULL;
2635 ETHER_NEXT_MULTI(step, enm)) {
2636 if (nentries >= VIRTIO_NET_CTRL_MAC_MAXENTRIES) {
2637 allmulti = true;
2638 break;
2639 }
2640 if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2641 allmulti = true;
2642 break;
2643 }
2644
2645 memcpy(ctrlq->ctrlq_mac_tbl_mc->macs[nentries],
2646 enm->enm_addrlo, ETHER_ADDR_LEN);
2647 nentries++;
2648 }
2649 ETHER_UNLOCK(ec);
2650
2651 r = vioif_set_mac_addr(sc);
2652 if (r != 0) {
2653 log(LOG_WARNING, "%s: couldn't set MAC address\n",
2654 ifp->if_xname);
2655 }
2656
2657 if (!allmulti) {
2658 ctrlq->ctrlq_mac_tbl_uc->nentries = virtio_rw32(vsc, 1);
2659 ctrlq->ctrlq_mac_tbl_mc->nentries = virtio_rw32(vsc, nentries);
2660 r = vioif_set_rx_filter(sc);
2661 if (r != 0) {
2662 allmulti = true; /* fallback */
2663 }
2664 }
2665
2666 if (allmulti) {
2667 ctrlq->ctrlq_mac_tbl_uc->nentries = virtio_rw32(vsc, 0);
2668 ctrlq->ctrlq_mac_tbl_mc->nentries = virtio_rw32(vsc, 0);
2669 r = vioif_set_rx_filter(sc);
2670 if (r != 0) {
2671 log(LOG_DEBUG, "%s: couldn't clear RX filter\n",
2672 ifp->if_xname);
2673 /* what to do on failure? */
2674 }
2675
2676 ifp->if_flags |= IFF_ALLMULTI;
2677 }
2678
2679 set_ifflags:
2680 r = vioif_ifflags(sc);
2681
2682 return r;
2683 }
2684
2685 /*
2686 * VM configuration changes
2687 */
2688 static int
2689 vioif_config_change(struct virtio_softc *vsc)
2690 {
2691 struct vioif_softc *sc = device_private(virtio_child(vsc));
2692
2693 softint_schedule(sc->sc_cfg_softint);
2694 return 0;
2695 }
2696
2697 static void
2698 vioif_cfg_softint(void *arg)
2699 {
2700 struct vioif_softc *sc = arg;
2701 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2702
2703 vioif_update_link_status(sc);
2704 vioif_start(ifp);
2705 }
2706
2707 static int
2708 vioif_get_link_status(struct vioif_softc *sc)
2709 {
2710 struct virtio_softc *vsc = sc->sc_virtio;
2711 uint16_t status;
2712
2713 if (virtio_features(vsc) & VIRTIO_NET_F_STATUS)
2714 status = virtio_read_device_config_2(vsc,
2715 VIRTIO_NET_CONFIG_STATUS);
2716 else
2717 status = VIRTIO_NET_S_LINK_UP;
2718
2719 if ((status & VIRTIO_NET_S_LINK_UP) != 0)
2720 return LINK_STATE_UP;
2721
2722 return LINK_STATE_DOWN;
2723 }
2724
2725 static void
2726 vioif_update_link_status(struct vioif_softc *sc)
2727 {
2728 struct ifnet *ifp = &sc->sc_ethercom.ec_if;
2729 struct vioif_netqueue *netq;
2730 struct vioif_tx_context *txc;
2731 bool active;
2732 int link, i;
2733
2734 mutex_enter(&sc->sc_lock);
2735
2736 link = vioif_get_link_status(sc);
2737
2738 if (link == sc->sc_link_state)
2739 goto done;
2740
2741 sc->sc_link_state = link;
2742
2743 active = VIOIF_IS_LINK_ACTIVE(sc);
2744 for (i = 0; i < sc->sc_act_nvq_pairs; i++) {
2745 netq = &sc->sc_netqs[VIOIF_NETQ_TXQID(i)];
2746
2747 mutex_enter(&netq->netq_lock);
2748 txc = netq->netq_ctx;
2749 txc->txc_link_active = active;
2750 mutex_exit(&netq->netq_lock);
2751 }
2752
2753 if_link_state_change(ifp, sc->sc_link_state);
2754
2755 done:
2756 mutex_exit(&sc->sc_lock);
2757 }
2758
2759 static void
2760 vioif_workq_work(struct work *wk, void *context)
2761 {
2762 struct vioif_work *work;
2763
2764 work = container_of(wk, struct vioif_work, cookie);
2765
2766 atomic_store_relaxed(&work->added, 0);
2767 work->func(work->arg);
2768 }
2769
2770 static struct workqueue *
2771 vioif_workq_create(const char *name, pri_t prio, int ipl, int flags)
2772 {
2773 struct workqueue *wq;
2774 int error;
2775
2776 error = workqueue_create(&wq, name, vioif_workq_work, NULL,
2777 prio, ipl, flags);
2778
2779 if (error)
2780 return NULL;
2781
2782 return wq;
2783 }
2784
2785 static void
2786 vioif_workq_destroy(struct workqueue *wq)
2787 {
2788
2789 workqueue_destroy(wq);
2790 }
2791
2792 static void
2793 vioif_work_set(struct vioif_work *work, void (*func)(void *), void *arg)
2794 {
2795
2796 memset(work, 0, sizeof(*work));
2797 work->func = func;
2798 work->arg = arg;
2799 }
2800
2801 static void
2802 vioif_work_add(struct workqueue *wq, struct vioif_work *work)
2803 {
2804
2805 if (atomic_load_relaxed(&work->added) != 0)
2806 return;
2807
2808 atomic_store_relaxed(&work->added, 1);
2809 kpreempt_disable();
2810 workqueue_enqueue(wq, &work->cookie, NULL);
2811 kpreempt_enable();
2812 }
2813
2814 static void
2815 vioif_work_wait(struct workqueue *wq, struct vioif_work *work)
2816 {
2817
2818 workqueue_wait(wq, &work->cookie);
2819 }
2820
2821 MODULE(MODULE_CLASS_DRIVER, if_vioif, "virtio");
2822
2823 #ifdef _MODULE
2824 #include "ioconf.c"
2825 #endif
2826
2827 static int
2828 if_vioif_modcmd(modcmd_t cmd, void *opaque)
2829 {
2830 int error = 0;
2831
2832 #ifdef _MODULE
2833 switch (cmd) {
2834 case MODULE_CMD_INIT:
2835 error = config_init_component(cfdriver_ioconf_if_vioif,
2836 cfattach_ioconf_if_vioif, cfdata_ioconf_if_vioif);
2837 break;
2838 case MODULE_CMD_FINI:
2839 error = config_fini_component(cfdriver_ioconf_if_vioif,
2840 cfattach_ioconf_if_vioif, cfdata_ioconf_if_vioif);
2841 break;
2842 default:
2843 error = ENOTTY;
2844 break;
2845 }
2846 #endif
2847
2848 return error;
2849 }
2850