xref: /linux/drivers/net/ethernet/sun/sunvnet.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /* sunvnet.c: Sun LDOM Virtual Network Driver.
2  *
3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include <linux/module.h>
9 #include <linux/kernel.h>
10 #include <linux/types.h>
11 #include <linux/slab.h>
12 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/netdevice.h>
15 #include <linux/ethtool.h>
16 #include <linux/etherdevice.h>
17 #include <linux/mutex.h>
18 #include <linux/highmem.h>
19 #include <linux/if_vlan.h>
20 
21 #if IS_ENABLED(CONFIG_IPV6)
22 #include <linux/icmpv6.h>
23 #endif
24 
25 #include <net/ip.h>
26 #include <net/icmp.h>
27 #include <net/route.h>
28 
29 #include <asm/vio.h>
30 #include <asm/ldc.h>
31 
32 #include "sunvnet.h"
33 
34 #define DRV_MODULE_NAME		"sunvnet"
35 #define DRV_MODULE_VERSION	"1.0"
36 #define DRV_MODULE_RELDATE	"June 25, 2007"
37 
38 static char version[] =
39 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
40 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
41 MODULE_DESCRIPTION("Sun LDOM virtual network driver");
42 MODULE_LICENSE("GPL");
43 MODULE_VERSION(DRV_MODULE_VERSION);
44 
45 #define	VNET_MAX_TXQS		16
46 
47 /* Heuristic for the number of times to exponentially backoff and
48  * retry sending an LDC trigger when EAGAIN is encountered
49  */
50 #define	VNET_MAX_RETRIES	10
51 
52 static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
53 static void vnet_port_reset(struct vnet_port *port);
54 
55 /* Ordered from largest major to lowest */
56 static struct vio_version vnet_versions[] = {
57 	{ .major = 1, .minor = 8 },
58 	{ .major = 1, .minor = 7 },
59 	{ .major = 1, .minor = 6 },
60 	{ .major = 1, .minor = 0 },
61 };
62 
63 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
64 {
65 	return vio_dring_avail(dr, VNET_TX_RING_SIZE);
66 }
67 
68 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
69 {
70 	struct vio_msg_tag *pkt = arg;
71 
72 	pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
73 	       pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
74 	pr_err("Resetting connection\n");
75 
76 	ldc_disconnect(port->vio.lp);
77 
78 	return -ECONNRESET;
79 }
80 
81 static int vnet_port_alloc_tx_ring(struct vnet_port *port);
82 
83 static int vnet_send_attr(struct vio_driver_state *vio)
84 {
85 	struct vnet_port *port = to_vnet_port(vio);
86 	struct net_device *dev = port->vp->dev;
87 	struct vio_net_attr_info pkt;
88 	int framelen = ETH_FRAME_LEN;
89 	int i, err;
90 
91 	err = vnet_port_alloc_tx_ring(to_vnet_port(vio));
92 	if (err)
93 		return err;
94 
95 	memset(&pkt, 0, sizeof(pkt));
96 	pkt.tag.type = VIO_TYPE_CTRL;
97 	pkt.tag.stype = VIO_SUBTYPE_INFO;
98 	pkt.tag.stype_env = VIO_ATTR_INFO;
99 	pkt.tag.sid = vio_send_sid(vio);
100 	if (vio_version_before(vio, 1, 2))
101 		pkt.xfer_mode = VIO_DRING_MODE;
102 	else
103 		pkt.xfer_mode = VIO_NEW_DRING_MODE;
104 	pkt.addr_type = VNET_ADDR_ETHERMAC;
105 	pkt.ack_freq = 0;
106 	for (i = 0; i < 6; i++)
107 		pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
108 	if (vio_version_after(vio, 1, 3)) {
109 		if (port->rmtu) {
110 			port->rmtu = min(VNET_MAXPACKET, port->rmtu);
111 			pkt.mtu = port->rmtu;
112 		} else {
113 			port->rmtu = VNET_MAXPACKET;
114 			pkt.mtu = port->rmtu;
115 		}
116 		if (vio_version_after_eq(vio, 1, 6))
117 			pkt.options = VIO_TX_DRING;
118 	} else if (vio_version_before(vio, 1, 3)) {
119 		pkt.mtu = framelen;
120 	} else { /* v1.3 */
121 		pkt.mtu = framelen + VLAN_HLEN;
122 	}
123 
124 	pkt.cflags = 0;
125 	if (vio_version_after_eq(vio, 1, 7) && port->tso) {
126 		pkt.cflags |= VNET_LSO_IPV4_CAPAB;
127 		if (!port->tsolen)
128 			port->tsolen = VNET_MAXTSO;
129 		pkt.ipv4_lso_maxlen = port->tsolen;
130 	}
131 
132 	pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
133 
134 	viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
135 	       "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
136 	       "cflags[0x%04x] lso_max[%u]\n",
137 	       pkt.xfer_mode, pkt.addr_type,
138 	       (unsigned long long)pkt.addr,
139 	       pkt.ack_freq, pkt.plnk_updt, pkt.options,
140 	       (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
141 
142 
143 	return vio_ldc_send(vio, &pkt, sizeof(pkt));
144 }
145 
146 static int handle_attr_info(struct vio_driver_state *vio,
147 			    struct vio_net_attr_info *pkt)
148 {
149 	struct vnet_port *port = to_vnet_port(vio);
150 	u64	localmtu;
151 	u8	xfer_mode;
152 
153 	viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
154 	       "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
155 	       " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
156 	       pkt->xfer_mode, pkt->addr_type,
157 	       (unsigned long long)pkt->addr,
158 	       pkt->ack_freq, pkt->plnk_updt, pkt->options,
159 	       (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
160 	       pkt->ipv4_lso_maxlen);
161 
162 	pkt->tag.sid = vio_send_sid(vio);
163 
164 	xfer_mode = pkt->xfer_mode;
165 	/* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
166 	if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
167 		xfer_mode = VIO_NEW_DRING_MODE;
168 
169 	/* MTU negotiation:
170 	 *	< v1.3 - ETH_FRAME_LEN exactly
171 	 *	> v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
172 	 *			pkt->mtu for ACK
173 	 *	= v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
174 	 */
175 	if (vio_version_before(vio, 1, 3)) {
176 		localmtu = ETH_FRAME_LEN;
177 	} else if (vio_version_after(vio, 1, 3)) {
178 		localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
179 		localmtu = min(pkt->mtu, localmtu);
180 		pkt->mtu = localmtu;
181 	} else { /* v1.3 */
182 		localmtu = ETH_FRAME_LEN + VLAN_HLEN;
183 	}
184 	port->rmtu = localmtu;
185 
186 	/* LSO negotiation */
187 	if (vio_version_after_eq(vio, 1, 7))
188 		port->tso &= !!(pkt->cflags & VNET_LSO_IPV4_CAPAB);
189 	else
190 		port->tso = false;
191 	if (port->tso) {
192 		if (!port->tsolen)
193 			port->tsolen = VNET_MAXTSO;
194 		port->tsolen = min(port->tsolen, pkt->ipv4_lso_maxlen);
195 		if (port->tsolen < VNET_MINTSO) {
196 			port->tso = false;
197 			port->tsolen = 0;
198 			pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
199 		}
200 		pkt->ipv4_lso_maxlen = port->tsolen;
201 	} else {
202 		pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
203 		pkt->ipv4_lso_maxlen = 0;
204 	}
205 
206 	/* for version >= 1.6, ACK packet mode we support */
207 	if (vio_version_after_eq(vio, 1, 6)) {
208 		pkt->xfer_mode = VIO_NEW_DRING_MODE;
209 		pkt->options = VIO_TX_DRING;
210 	}
211 
212 	if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
213 	    pkt->addr_type != VNET_ADDR_ETHERMAC ||
214 	    pkt->mtu != localmtu) {
215 		viodbg(HS, "SEND NET ATTR NACK\n");
216 
217 		pkt->tag.stype = VIO_SUBTYPE_NACK;
218 
219 		(void) vio_ldc_send(vio, pkt, sizeof(*pkt));
220 
221 		return -ECONNRESET;
222 	} else {
223 		viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
224 		       "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
225 		       "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
226 		       pkt->xfer_mode, pkt->addr_type,
227 		       (unsigned long long)pkt->addr,
228 		       pkt->ack_freq, pkt->plnk_updt, pkt->options,
229 		       (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
230 		       pkt->ipv4_lso_maxlen);
231 
232 		pkt->tag.stype = VIO_SUBTYPE_ACK;
233 
234 		return vio_ldc_send(vio, pkt, sizeof(*pkt));
235 	}
236 
237 }
238 
239 static int handle_attr_ack(struct vio_driver_state *vio,
240 			   struct vio_net_attr_info *pkt)
241 {
242 	viodbg(HS, "GOT NET ATTR ACK\n");
243 
244 	return 0;
245 }
246 
247 static int handle_attr_nack(struct vio_driver_state *vio,
248 			    struct vio_net_attr_info *pkt)
249 {
250 	viodbg(HS, "GOT NET ATTR NACK\n");
251 
252 	return -ECONNRESET;
253 }
254 
255 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
256 {
257 	struct vio_net_attr_info *pkt = arg;
258 
259 	switch (pkt->tag.stype) {
260 	case VIO_SUBTYPE_INFO:
261 		return handle_attr_info(vio, pkt);
262 
263 	case VIO_SUBTYPE_ACK:
264 		return handle_attr_ack(vio, pkt);
265 
266 	case VIO_SUBTYPE_NACK:
267 		return handle_attr_nack(vio, pkt);
268 
269 	default:
270 		return -ECONNRESET;
271 	}
272 }
273 
274 static void vnet_handshake_complete(struct vio_driver_state *vio)
275 {
276 	struct vio_dring_state *dr;
277 
278 	dr = &vio->drings[VIO_DRIVER_RX_RING];
279 	dr->snd_nxt = dr->rcv_nxt = 1;
280 
281 	dr = &vio->drings[VIO_DRIVER_TX_RING];
282 	dr->snd_nxt = dr->rcv_nxt = 1;
283 }
284 
285 /* The hypervisor interface that implements copying to/from imported
286  * memory from another domain requires that copies are done to 8-byte
287  * aligned buffers, and that the lengths of such copies are also 8-byte
288  * multiples.
289  *
290  * So we align skb->data to an 8-byte multiple and pad-out the data
291  * area so we can round the copy length up to the next multiple of
292  * 8 for the copy.
293  *
294  * The transmitter puts the actual start of the packet 6 bytes into
295  * the buffer it sends over, so that the IP headers after the ethernet
296  * header are aligned properly.  These 6 bytes are not in the descriptor
297  * length, they are simply implied.  This offset is represented using
298  * the VNET_PACKET_SKIP macro.
299  */
300 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
301 					   unsigned int len)
302 {
303 	struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
304 	unsigned long addr, off;
305 
306 	if (unlikely(!skb))
307 		return NULL;
308 
309 	addr = (unsigned long) skb->data;
310 	off = ((addr + 7UL) & ~7UL) - addr;
311 	if (off)
312 		skb_reserve(skb, off);
313 
314 	return skb;
315 }
316 
317 static inline void vnet_fullcsum(struct sk_buff *skb)
318 {
319 	struct iphdr *iph = ip_hdr(skb);
320 	int offset = skb_transport_offset(skb);
321 
322 	if (skb->protocol != htons(ETH_P_IP))
323 		return;
324 	if (iph->protocol != IPPROTO_TCP &&
325 	    iph->protocol != IPPROTO_UDP)
326 		return;
327 	skb->ip_summed = CHECKSUM_NONE;
328 	skb->csum_level = 1;
329 	skb->csum = 0;
330 	if (iph->protocol == IPPROTO_TCP) {
331 		struct tcphdr *ptcp = tcp_hdr(skb);
332 
333 		ptcp->check = 0;
334 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
335 		ptcp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
336 						skb->len - offset, IPPROTO_TCP,
337 						skb->csum);
338 	} else if (iph->protocol == IPPROTO_UDP) {
339 		struct udphdr *pudp = udp_hdr(skb);
340 
341 		pudp->check = 0;
342 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
343 		pudp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
344 						skb->len - offset, IPPROTO_UDP,
345 						skb->csum);
346 	}
347 }
348 
349 static int vnet_rx_one(struct vnet_port *port, struct vio_net_desc *desc)
350 {
351 	struct net_device *dev = port->vp->dev;
352 	unsigned int len = desc->size;
353 	unsigned int copy_len;
354 	struct sk_buff *skb;
355 	int maxlen;
356 	int err;
357 
358 	err = -EMSGSIZE;
359 	if (port->tso && port->tsolen > port->rmtu)
360 		maxlen = port->tsolen;
361 	else
362 		maxlen = port->rmtu;
363 	if (unlikely(len < ETH_ZLEN || len > maxlen)) {
364 		dev->stats.rx_length_errors++;
365 		goto out_dropped;
366 	}
367 
368 	skb = alloc_and_align_skb(dev, len);
369 	err = -ENOMEM;
370 	if (unlikely(!skb)) {
371 		dev->stats.rx_missed_errors++;
372 		goto out_dropped;
373 	}
374 
375 	copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
376 	skb_put(skb, copy_len);
377 	err = ldc_copy(port->vio.lp, LDC_COPY_IN,
378 		       skb->data, copy_len, 0,
379 		       desc->cookies, desc->ncookies);
380 	if (unlikely(err < 0)) {
381 		dev->stats.rx_frame_errors++;
382 		goto out_free_skb;
383 	}
384 
385 	skb_pull(skb, VNET_PACKET_SKIP);
386 	skb_trim(skb, len);
387 	skb->protocol = eth_type_trans(skb, dev);
388 
389 	if (vio_version_after_eq(&port->vio, 1, 8)) {
390 		struct vio_net_dext *dext = vio_net_ext(desc);
391 
392 		if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM) {
393 			if (skb->protocol == ETH_P_IP) {
394 				struct iphdr *iph = (struct iphdr *)skb->data;
395 
396 				iph->check = 0;
397 				ip_send_check(iph);
398 			}
399 		}
400 		if ((dext->flags & VNET_PKT_HCK_FULLCKSUM) &&
401 		    skb->ip_summed == CHECKSUM_NONE)
402 			vnet_fullcsum(skb);
403 		if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM_OK) {
404 			skb->ip_summed = CHECKSUM_PARTIAL;
405 			skb->csum_level = 0;
406 			if (dext->flags & VNET_PKT_HCK_FULLCKSUM_OK)
407 				skb->csum_level = 1;
408 		}
409 	}
410 
411 	skb->ip_summed = port->switch_port ? CHECKSUM_NONE : CHECKSUM_PARTIAL;
412 
413 	dev->stats.rx_packets++;
414 	dev->stats.rx_bytes += len;
415 	napi_gro_receive(&port->napi, skb);
416 	return 0;
417 
418 out_free_skb:
419 	kfree_skb(skb);
420 
421 out_dropped:
422 	dev->stats.rx_dropped++;
423 	return err;
424 }
425 
426 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
427 			 u32 start, u32 end, u8 vio_dring_state)
428 {
429 	struct vio_dring_data hdr = {
430 		.tag = {
431 			.type		= VIO_TYPE_DATA,
432 			.stype		= VIO_SUBTYPE_ACK,
433 			.stype_env	= VIO_DRING_DATA,
434 			.sid		= vio_send_sid(&port->vio),
435 		},
436 		.dring_ident		= dr->ident,
437 		.start_idx		= start,
438 		.end_idx		= end,
439 		.state			= vio_dring_state,
440 	};
441 	int err, delay;
442 	int retries = 0;
443 
444 	hdr.seq = dr->snd_nxt;
445 	delay = 1;
446 	do {
447 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
448 		if (err > 0) {
449 			dr->snd_nxt++;
450 			break;
451 		}
452 		udelay(delay);
453 		if ((delay <<= 1) > 128)
454 			delay = 128;
455 		if (retries++ > VNET_MAX_RETRIES) {
456 			pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
457 				port->raddr[0], port->raddr[1],
458 				port->raddr[2], port->raddr[3],
459 				port->raddr[4], port->raddr[5]);
460 			break;
461 		}
462 	} while (err == -EAGAIN);
463 
464 	if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
465 		port->stop_rx_idx = end;
466 		port->stop_rx = true;
467 	} else {
468 		port->stop_rx_idx = 0;
469 		port->stop_rx = false;
470 	}
471 
472 	return err;
473 }
474 
475 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
476 					struct vio_dring_state *dr,
477 					u32 index)
478 {
479 	struct vio_net_desc *desc = port->vio.desc_buf;
480 	int err;
481 
482 	err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
483 				  (index * dr->entry_size),
484 				  dr->cookies, dr->ncookies);
485 	if (err < 0)
486 		return ERR_PTR(err);
487 
488 	return desc;
489 }
490 
491 static int put_rx_desc(struct vnet_port *port,
492 		       struct vio_dring_state *dr,
493 		       struct vio_net_desc *desc,
494 		       u32 index)
495 {
496 	int err;
497 
498 	err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
499 				  (index * dr->entry_size),
500 				  dr->cookies, dr->ncookies);
501 	if (err < 0)
502 		return err;
503 
504 	return 0;
505 }
506 
507 static int vnet_walk_rx_one(struct vnet_port *port,
508 			    struct vio_dring_state *dr,
509 			    u32 index, int *needs_ack)
510 {
511 	struct vio_net_desc *desc = get_rx_desc(port, dr, index);
512 	struct vio_driver_state *vio = &port->vio;
513 	int err;
514 
515 	BUG_ON(desc == NULL);
516 	if (IS_ERR(desc))
517 		return PTR_ERR(desc);
518 
519 	if (desc->hdr.state != VIO_DESC_READY)
520 		return 1;
521 
522 	dma_rmb();
523 
524 	viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
525 	       desc->hdr.state, desc->hdr.ack,
526 	       desc->size, desc->ncookies,
527 	       desc->cookies[0].cookie_addr,
528 	       desc->cookies[0].cookie_size);
529 
530 	err = vnet_rx_one(port, desc);
531 	if (err == -ECONNRESET)
532 		return err;
533 	desc->hdr.state = VIO_DESC_DONE;
534 	err = put_rx_desc(port, dr, desc, index);
535 	if (err < 0)
536 		return err;
537 	*needs_ack = desc->hdr.ack;
538 	return 0;
539 }
540 
541 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
542 			u32 start, u32 end, int *npkts, int budget)
543 {
544 	struct vio_driver_state *vio = &port->vio;
545 	int ack_start = -1, ack_end = -1;
546 	bool send_ack = true;
547 
548 	end = (end == (u32) -1) ? vio_dring_prev(dr, start)
549 				: vio_dring_next(dr, end);
550 
551 	viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
552 
553 	while (start != end) {
554 		int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
555 		if (err == -ECONNRESET)
556 			return err;
557 		if (err != 0)
558 			break;
559 		(*npkts)++;
560 		if (ack_start == -1)
561 			ack_start = start;
562 		ack_end = start;
563 		start = vio_dring_next(dr, start);
564 		if (ack && start != end) {
565 			err = vnet_send_ack(port, dr, ack_start, ack_end,
566 					    VIO_DRING_ACTIVE);
567 			if (err == -ECONNRESET)
568 				return err;
569 			ack_start = -1;
570 		}
571 		if ((*npkts) >= budget) {
572 			send_ack = false;
573 			break;
574 		}
575 	}
576 	if (unlikely(ack_start == -1))
577 		ack_start = ack_end = vio_dring_prev(dr, start);
578 	if (send_ack) {
579 		port->napi_resume = false;
580 		return vnet_send_ack(port, dr, ack_start, ack_end,
581 				     VIO_DRING_STOPPED);
582 	} else  {
583 		port->napi_resume = true;
584 		port->napi_stop_idx = ack_end;
585 		return 1;
586 	}
587 }
588 
589 static int vnet_rx(struct vnet_port *port, void *msgbuf, int *npkts,
590 		   int budget)
591 {
592 	struct vio_dring_data *pkt = msgbuf;
593 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
594 	struct vio_driver_state *vio = &port->vio;
595 
596 	viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
597 	       pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
598 
599 	if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
600 		return 0;
601 	if (unlikely(pkt->seq != dr->rcv_nxt)) {
602 		pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
603 		       pkt->seq, dr->rcv_nxt);
604 		return 0;
605 	}
606 
607 	if (!port->napi_resume)
608 		dr->rcv_nxt++;
609 
610 	/* XXX Validate pkt->start_idx and pkt->end_idx XXX */
611 
612 	return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx,
613 			    npkts, budget);
614 }
615 
616 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
617 {
618 	u32 idx = dr->cons;
619 	int found = 0;
620 
621 	while (idx != dr->prod) {
622 		if (idx == end) {
623 			found = 1;
624 			break;
625 		}
626 		idx = vio_dring_next(dr, idx);
627 	}
628 	return found;
629 }
630 
631 static int vnet_ack(struct vnet_port *port, void *msgbuf)
632 {
633 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
634 	struct vio_dring_data *pkt = msgbuf;
635 	struct net_device *dev;
636 	struct vnet *vp;
637 	u32 end;
638 	struct vio_net_desc *desc;
639 	struct netdev_queue *txq;
640 
641 	if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
642 		return 0;
643 
644 	end = pkt->end_idx;
645 	vp = port->vp;
646 	dev = vp->dev;
647 	netif_tx_lock(dev);
648 	if (unlikely(!idx_is_pending(dr, end))) {
649 		netif_tx_unlock(dev);
650 		return 0;
651 	}
652 
653 	/* sync for race conditions with vnet_start_xmit() and tell xmit it
654 	 * is time to send a trigger.
655 	 */
656 	dr->cons = vio_dring_next(dr, end);
657 	desc = vio_dring_entry(dr, dr->cons);
658 	if (desc->hdr.state == VIO_DESC_READY && !port->start_cons) {
659 		/* vnet_start_xmit() just populated this dring but missed
660 		 * sending the "start" LDC message to the consumer.
661 		 * Send a "start" trigger on its behalf.
662 		 */
663 		if (__vnet_tx_trigger(port, dr->cons) > 0)
664 			port->start_cons = false;
665 		else
666 			port->start_cons = true;
667 	} else {
668 		port->start_cons = true;
669 	}
670 	netif_tx_unlock(dev);
671 
672 	txq = netdev_get_tx_queue(dev, port->q_index);
673 	if (unlikely(netif_tx_queue_stopped(txq) &&
674 		     vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
675 		return 1;
676 
677 	return 0;
678 }
679 
680 static int vnet_nack(struct vnet_port *port, void *msgbuf)
681 {
682 	/* XXX just reset or similar XXX */
683 	return 0;
684 }
685 
686 static int handle_mcast(struct vnet_port *port, void *msgbuf)
687 {
688 	struct vio_net_mcast_info *pkt = msgbuf;
689 
690 	if (pkt->tag.stype != VIO_SUBTYPE_ACK)
691 		pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
692 		       port->vp->dev->name,
693 		       pkt->tag.type,
694 		       pkt->tag.stype,
695 		       pkt->tag.stype_env,
696 		       pkt->tag.sid);
697 
698 	return 0;
699 }
700 
701 /* Got back a STOPPED LDC message on port. If the queue is stopped,
702  * wake it up so that we'll send out another START message at the
703  * next TX.
704  */
705 static void maybe_tx_wakeup(struct vnet_port *port)
706 {
707 	struct netdev_queue *txq;
708 
709 	txq = netdev_get_tx_queue(port->vp->dev, port->q_index);
710 	__netif_tx_lock(txq, smp_processor_id());
711 	if (likely(netif_tx_queue_stopped(txq))) {
712 		struct vio_dring_state *dr;
713 
714 		dr = &port->vio.drings[VIO_DRIVER_TX_RING];
715 		netif_tx_wake_queue(txq);
716 	}
717 	__netif_tx_unlock(txq);
718 }
719 
720 static inline bool port_is_up(struct vnet_port *vnet)
721 {
722 	struct vio_driver_state *vio = &vnet->vio;
723 
724 	return !!(vio->hs_state & VIO_HS_COMPLETE);
725 }
726 
727 static int vnet_event_napi(struct vnet_port *port, int budget)
728 {
729 	struct vio_driver_state *vio = &port->vio;
730 	int tx_wakeup, err;
731 	int npkts = 0;
732 	int event = (port->rx_event & LDC_EVENT_RESET);
733 
734 ldc_ctrl:
735 	if (unlikely(event == LDC_EVENT_RESET ||
736 		     event == LDC_EVENT_UP)) {
737 		vio_link_state_change(vio, event);
738 
739 		if (event == LDC_EVENT_RESET) {
740 			vnet_port_reset(port);
741 			vio_port_up(vio);
742 		}
743 		port->rx_event = 0;
744 		return 0;
745 	}
746 	/* We may have multiple LDC events in rx_event. Unroll send_events() */
747 	event = (port->rx_event & LDC_EVENT_UP);
748 	port->rx_event &= ~(LDC_EVENT_RESET|LDC_EVENT_UP);
749 	if (event == LDC_EVENT_UP)
750 		goto ldc_ctrl;
751 	event = port->rx_event;
752 	if (!(event & LDC_EVENT_DATA_READY))
753 		return 0;
754 
755 	/* we dont expect any other bits than RESET, UP, DATA_READY */
756 	BUG_ON(event != LDC_EVENT_DATA_READY);
757 
758 	tx_wakeup = err = 0;
759 	while (1) {
760 		union {
761 			struct vio_msg_tag tag;
762 			u64 raw[8];
763 		} msgbuf;
764 
765 		if (port->napi_resume) {
766 			struct vio_dring_data *pkt =
767 				(struct vio_dring_data *)&msgbuf;
768 			struct vio_dring_state *dr =
769 				&port->vio.drings[VIO_DRIVER_RX_RING];
770 
771 			pkt->tag.type = VIO_TYPE_DATA;
772 			pkt->tag.stype = VIO_SUBTYPE_INFO;
773 			pkt->tag.stype_env = VIO_DRING_DATA;
774 			pkt->seq = dr->rcv_nxt;
775 			pkt->start_idx = vio_dring_next(dr, port->napi_stop_idx);
776 			pkt->end_idx = -1;
777 			goto napi_resume;
778 		}
779 		err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
780 		if (unlikely(err < 0)) {
781 			if (err == -ECONNRESET)
782 				vio_conn_reset(vio);
783 			break;
784 		}
785 		if (err == 0)
786 			break;
787 		viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
788 		       msgbuf.tag.type,
789 		       msgbuf.tag.stype,
790 		       msgbuf.tag.stype_env,
791 		       msgbuf.tag.sid);
792 		err = vio_validate_sid(vio, &msgbuf.tag);
793 		if (err < 0)
794 			break;
795 napi_resume:
796 		if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
797 			if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
798 				if (!port_is_up(port)) {
799 					/* failures like handshake_failure()
800 					 * may have cleaned up dring, but
801 					 * NAPI polling may bring us here.
802 					 */
803 					err = -ECONNRESET;
804 					break;
805 				}
806 				err = vnet_rx(port, &msgbuf, &npkts, budget);
807 				if (npkts >= budget)
808 					break;
809 				if (npkts == 0)
810 					break;
811 			} else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
812 				err = vnet_ack(port, &msgbuf);
813 				if (err > 0)
814 					tx_wakeup |= err;
815 			} else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
816 				err = vnet_nack(port, &msgbuf);
817 			}
818 		} else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
819 			if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
820 				err = handle_mcast(port, &msgbuf);
821 			else
822 				err = vio_control_pkt_engine(vio, &msgbuf);
823 			if (err)
824 				break;
825 		} else {
826 			err = vnet_handle_unknown(port, &msgbuf);
827 		}
828 		if (err == -ECONNRESET)
829 			break;
830 	}
831 	if (unlikely(tx_wakeup && err != -ECONNRESET))
832 		maybe_tx_wakeup(port);
833 	return npkts;
834 }
835 
836 static int vnet_poll(struct napi_struct *napi, int budget)
837 {
838 	struct vnet_port *port = container_of(napi, struct vnet_port, napi);
839 	struct vio_driver_state *vio = &port->vio;
840 	int processed = vnet_event_napi(port, budget);
841 
842 	if (processed < budget) {
843 		napi_complete(napi);
844 		port->rx_event &= ~LDC_EVENT_DATA_READY;
845 		vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
846 	}
847 	return processed;
848 }
849 
850 static void vnet_event(void *arg, int event)
851 {
852 	struct vnet_port *port = arg;
853 	struct vio_driver_state *vio = &port->vio;
854 
855 	port->rx_event |= event;
856 	vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
857 	napi_schedule(&port->napi);
858 
859 }
860 
861 static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
862 {
863 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
864 	struct vio_dring_data hdr = {
865 		.tag = {
866 			.type		= VIO_TYPE_DATA,
867 			.stype		= VIO_SUBTYPE_INFO,
868 			.stype_env	= VIO_DRING_DATA,
869 			.sid		= vio_send_sid(&port->vio),
870 		},
871 		.dring_ident		= dr->ident,
872 		.start_idx		= start,
873 		.end_idx		= (u32) -1,
874 	};
875 	int err, delay;
876 	int retries = 0;
877 
878 	if (port->stop_rx) {
879 		err = vnet_send_ack(port,
880 				    &port->vio.drings[VIO_DRIVER_RX_RING],
881 				    port->stop_rx_idx, -1,
882 				    VIO_DRING_STOPPED);
883 		if (err <= 0)
884 			return err;
885 	}
886 
887 	hdr.seq = dr->snd_nxt;
888 	delay = 1;
889 	do {
890 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
891 		if (err > 0) {
892 			dr->snd_nxt++;
893 			break;
894 		}
895 		udelay(delay);
896 		if ((delay <<= 1) > 128)
897 			delay = 128;
898 		if (retries++ > VNET_MAX_RETRIES)
899 			break;
900 	} while (err == -EAGAIN);
901 
902 	return err;
903 }
904 
905 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
906 {
907 	unsigned int hash = vnet_hashfn(skb->data);
908 	struct hlist_head *hp = &vp->port_hash[hash];
909 	struct vnet_port *port;
910 
911 	hlist_for_each_entry_rcu(port, hp, hash) {
912 		if (!port_is_up(port))
913 			continue;
914 		if (ether_addr_equal(port->raddr, skb->data))
915 			return port;
916 	}
917 	list_for_each_entry_rcu(port, &vp->port_list, list) {
918 		if (!port->switch_port)
919 			continue;
920 		if (!port_is_up(port))
921 			continue;
922 		return port;
923 	}
924 	return NULL;
925 }
926 
927 static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
928 					  unsigned *pending)
929 {
930 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
931 	struct sk_buff *skb = NULL;
932 	int i, txi;
933 
934 	*pending = 0;
935 
936 	txi = dr->prod;
937 	for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
938 		struct vio_net_desc *d;
939 
940 		--txi;
941 		if (txi < 0)
942 			txi = VNET_TX_RING_SIZE-1;
943 
944 		d = vio_dring_entry(dr, txi);
945 
946 		if (d->hdr.state == VIO_DESC_READY) {
947 			(*pending)++;
948 			continue;
949 		}
950 		if (port->tx_bufs[txi].skb) {
951 			if (d->hdr.state != VIO_DESC_DONE)
952 				pr_notice("invalid ring buffer state %d\n",
953 					  d->hdr.state);
954 			BUG_ON(port->tx_bufs[txi].skb->next);
955 
956 			port->tx_bufs[txi].skb->next = skb;
957 			skb = port->tx_bufs[txi].skb;
958 			port->tx_bufs[txi].skb = NULL;
959 
960 			ldc_unmap(port->vio.lp,
961 				  port->tx_bufs[txi].cookies,
962 				  port->tx_bufs[txi].ncookies);
963 		} else if (d->hdr.state == VIO_DESC_FREE)
964 			break;
965 		d->hdr.state = VIO_DESC_FREE;
966 	}
967 	return skb;
968 }
969 
970 static inline void vnet_free_skbs(struct sk_buff *skb)
971 {
972 	struct sk_buff *next;
973 
974 	while (skb) {
975 		next = skb->next;
976 		skb->next = NULL;
977 		dev_kfree_skb(skb);
978 		skb = next;
979 	}
980 }
981 
982 static void vnet_clean_timer_expire(unsigned long port0)
983 {
984 	struct vnet_port *port = (struct vnet_port *)port0;
985 	struct sk_buff *freeskbs;
986 	unsigned pending;
987 
988 	netif_tx_lock(port->vp->dev);
989 	freeskbs = vnet_clean_tx_ring(port, &pending);
990 	netif_tx_unlock(port->vp->dev);
991 
992 	vnet_free_skbs(freeskbs);
993 
994 	if (pending)
995 		(void)mod_timer(&port->clean_timer,
996 				jiffies + VNET_CLEAN_TIMEOUT);
997 	 else
998 		del_timer(&port->clean_timer);
999 }
1000 
1001 static inline int vnet_skb_map(struct ldc_channel *lp, struct sk_buff *skb,
1002 			       struct ldc_trans_cookie *cookies, int ncookies,
1003 			       unsigned int map_perm)
1004 {
1005 	int i, nc, err, blen;
1006 
1007 	/* header */
1008 	blen = skb_headlen(skb);
1009 	if (blen < ETH_ZLEN)
1010 		blen = ETH_ZLEN;
1011 	blen += VNET_PACKET_SKIP;
1012 	blen += 8 - (blen & 7);
1013 
1014 	err = ldc_map_single(lp, skb->data-VNET_PACKET_SKIP, blen, cookies,
1015 			     ncookies, map_perm);
1016 	if (err < 0)
1017 		return err;
1018 	nc = err;
1019 
1020 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1021 		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1022 		u8 *vaddr;
1023 
1024 		if (nc < ncookies) {
1025 			vaddr = kmap_atomic(skb_frag_page(f));
1026 			blen = skb_frag_size(f);
1027 			blen += 8 - (blen & 7);
1028 			err = ldc_map_single(lp, vaddr + f->page_offset,
1029 					     blen, cookies + nc, ncookies - nc,
1030 					     map_perm);
1031 			kunmap_atomic(vaddr);
1032 		} else {
1033 			err = -EMSGSIZE;
1034 		}
1035 
1036 		if (err < 0) {
1037 			ldc_unmap(lp, cookies, nc);
1038 			return err;
1039 		}
1040 		nc += err;
1041 	}
1042 	return nc;
1043 }
1044 
1045 static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, int ncookies)
1046 {
1047 	struct sk_buff *nskb;
1048 	int i, len, pad, docopy;
1049 
1050 	len = skb->len;
1051 	pad = 0;
1052 	if (len < ETH_ZLEN) {
1053 		pad += ETH_ZLEN - skb->len;
1054 		len += pad;
1055 	}
1056 	len += VNET_PACKET_SKIP;
1057 	pad += 8 - (len & 7);
1058 
1059 	/* make sure we have enough cookies and alignment in every frag */
1060 	docopy = skb_shinfo(skb)->nr_frags >= ncookies;
1061 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1062 		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1063 
1064 		docopy |= f->page_offset & 7;
1065 	}
1066 	if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
1067 	    skb_tailroom(skb) < pad ||
1068 	    skb_headroom(skb) < VNET_PACKET_SKIP || docopy) {
1069 		int start = 0, offset;
1070 		__wsum csum;
1071 
1072 		len = skb->len > ETH_ZLEN ? skb->len : ETH_ZLEN;
1073 		nskb = alloc_and_align_skb(skb->dev, len);
1074 		if (nskb == NULL) {
1075 			dev_kfree_skb(skb);
1076 			return NULL;
1077 		}
1078 		skb_reserve(nskb, VNET_PACKET_SKIP);
1079 
1080 		nskb->protocol = skb->protocol;
1081 		offset = skb_mac_header(skb) - skb->data;
1082 		skb_set_mac_header(nskb, offset);
1083 		offset = skb_network_header(skb) - skb->data;
1084 		skb_set_network_header(nskb, offset);
1085 		offset = skb_transport_header(skb) - skb->data;
1086 		skb_set_transport_header(nskb, offset);
1087 
1088 		offset = 0;
1089 		nskb->csum_offset = skb->csum_offset;
1090 		nskb->ip_summed = skb->ip_summed;
1091 
1092 		if (skb->ip_summed == CHECKSUM_PARTIAL)
1093 			start = skb_checksum_start_offset(skb);
1094 		if (start) {
1095 			struct iphdr *iph = ip_hdr(nskb);
1096 			int offset = start + nskb->csum_offset;
1097 
1098 			if (skb_copy_bits(skb, 0, nskb->data, start)) {
1099 				dev_kfree_skb(nskb);
1100 				dev_kfree_skb(skb);
1101 				return NULL;
1102 			}
1103 			*(__sum16 *)(skb->data + offset) = 0;
1104 			csum = skb_copy_and_csum_bits(skb, start,
1105 						      nskb->data + start,
1106 						      skb->len - start, 0);
1107 			if (iph->protocol == IPPROTO_TCP ||
1108 			    iph->protocol == IPPROTO_UDP) {
1109 				csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
1110 							 skb->len - start,
1111 							 iph->protocol, csum);
1112 			}
1113 			*(__sum16 *)(nskb->data + offset) = csum;
1114 
1115 			nskb->ip_summed = CHECKSUM_NONE;
1116 		} else if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
1117 			dev_kfree_skb(nskb);
1118 			dev_kfree_skb(skb);
1119 			return NULL;
1120 		}
1121 		(void)skb_put(nskb, skb->len);
1122 		if (skb_is_gso(skb)) {
1123 			skb_shinfo(nskb)->gso_size = skb_shinfo(skb)->gso_size;
1124 			skb_shinfo(nskb)->gso_type = skb_shinfo(skb)->gso_type;
1125 		}
1126 		nskb->queue_mapping = skb->queue_mapping;
1127 		dev_kfree_skb(skb);
1128 		skb = nskb;
1129 	}
1130 	return skb;
1131 }
1132 
1133 static u16
1134 vnet_select_queue(struct net_device *dev, struct sk_buff *skb,
1135 		  void *accel_priv, select_queue_fallback_t fallback)
1136 {
1137 	struct vnet *vp = netdev_priv(dev);
1138 	struct vnet_port *port = __tx_port_find(vp, skb);
1139 
1140 	if (port == NULL)
1141 		return 0;
1142 	return port->q_index;
1143 }
1144 
1145 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev);
1146 
1147 static int vnet_handle_offloads(struct vnet_port *port, struct sk_buff *skb)
1148 {
1149 	struct net_device *dev = port->vp->dev;
1150 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1151 	struct sk_buff *segs;
1152 	int maclen, datalen;
1153 	int status;
1154 	int gso_size, gso_type, gso_segs;
1155 	int hlen = skb_transport_header(skb) - skb_mac_header(skb);
1156 	int proto = IPPROTO_IP;
1157 
1158 	if (skb->protocol == htons(ETH_P_IP))
1159 		proto = ip_hdr(skb)->protocol;
1160 	else if (skb->protocol == htons(ETH_P_IPV6))
1161 		proto = ipv6_hdr(skb)->nexthdr;
1162 
1163 	if (proto == IPPROTO_TCP)
1164 		hlen += tcp_hdr(skb)->doff * 4;
1165 	else if (proto == IPPROTO_UDP)
1166 		hlen += sizeof(struct udphdr);
1167 	else {
1168 		pr_err("vnet_handle_offloads GSO with unknown transport "
1169 		       "protocol %d tproto %d\n", skb->protocol, proto);
1170 		hlen = 128; /* XXX */
1171 	}
1172 	datalen = port->tsolen - hlen;
1173 
1174 	gso_size = skb_shinfo(skb)->gso_size;
1175 	gso_type = skb_shinfo(skb)->gso_type;
1176 	gso_segs = skb_shinfo(skb)->gso_segs;
1177 
1178 	if (port->tso && gso_size < datalen)
1179 		gso_segs = DIV_ROUND_UP(skb->len - hlen, datalen);
1180 
1181 	if (unlikely(vnet_tx_dring_avail(dr) < gso_segs)) {
1182 		struct netdev_queue *txq;
1183 
1184 		txq  = netdev_get_tx_queue(dev, port->q_index);
1185 		netif_tx_stop_queue(txq);
1186 		if (vnet_tx_dring_avail(dr) < skb_shinfo(skb)->gso_segs)
1187 			return NETDEV_TX_BUSY;
1188 		netif_tx_wake_queue(txq);
1189 	}
1190 
1191 	maclen = skb_network_header(skb) - skb_mac_header(skb);
1192 	skb_pull(skb, maclen);
1193 
1194 	if (port->tso && gso_size < datalen) {
1195 		if (skb_unclone(skb, GFP_ATOMIC))
1196 			goto out_dropped;
1197 
1198 		/* segment to TSO size */
1199 		skb_shinfo(skb)->gso_size = datalen;
1200 		skb_shinfo(skb)->gso_segs = gso_segs;
1201 	}
1202 	segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO);
1203 	if (IS_ERR(segs))
1204 		goto out_dropped;
1205 
1206 	skb_push(skb, maclen);
1207 	skb_reset_mac_header(skb);
1208 
1209 	status = 0;
1210 	while (segs) {
1211 		struct sk_buff *curr = segs;
1212 
1213 		segs = segs->next;
1214 		curr->next = NULL;
1215 		if (port->tso && curr->len > dev->mtu) {
1216 			skb_shinfo(curr)->gso_size = gso_size;
1217 			skb_shinfo(curr)->gso_type = gso_type;
1218 			skb_shinfo(curr)->gso_segs =
1219 				DIV_ROUND_UP(curr->len - hlen, gso_size);
1220 		} else
1221 			skb_shinfo(curr)->gso_size = 0;
1222 
1223 		skb_push(curr, maclen);
1224 		skb_reset_mac_header(curr);
1225 		memcpy(skb_mac_header(curr), skb_mac_header(skb),
1226 		       maclen);
1227 		curr->csum_start = skb_transport_header(curr) - curr->head;
1228 		if (ip_hdr(curr)->protocol == IPPROTO_TCP)
1229 			curr->csum_offset = offsetof(struct tcphdr, check);
1230 		else if (ip_hdr(curr)->protocol == IPPROTO_UDP)
1231 			curr->csum_offset = offsetof(struct udphdr, check);
1232 
1233 		if (!(status & NETDEV_TX_MASK))
1234 			status = vnet_start_xmit(curr, dev);
1235 		if (status & NETDEV_TX_MASK)
1236 			dev_kfree_skb_any(curr);
1237 	}
1238 
1239 	if (!(status & NETDEV_TX_MASK))
1240 		dev_kfree_skb_any(skb);
1241 	return status;
1242 out_dropped:
1243 	dev->stats.tx_dropped++;
1244 	dev_kfree_skb_any(skb);
1245 	return NETDEV_TX_OK;
1246 }
1247 
1248 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
1249 {
1250 	struct vnet *vp = netdev_priv(dev);
1251 	struct vnet_port *port = NULL;
1252 	struct vio_dring_state *dr;
1253 	struct vio_net_desc *d;
1254 	unsigned int len;
1255 	struct sk_buff *freeskbs = NULL;
1256 	int i, err, txi;
1257 	unsigned pending = 0;
1258 	struct netdev_queue *txq;
1259 
1260 	rcu_read_lock();
1261 	port = __tx_port_find(vp, skb);
1262 	if (unlikely(!port)) {
1263 		rcu_read_unlock();
1264 		goto out_dropped;
1265 	}
1266 
1267 	if (skb_is_gso(skb) && skb->len > port->tsolen) {
1268 		err = vnet_handle_offloads(port, skb);
1269 		rcu_read_unlock();
1270 		return err;
1271 	}
1272 
1273 	if (!skb_is_gso(skb) && skb->len > port->rmtu) {
1274 		unsigned long localmtu = port->rmtu - ETH_HLEN;
1275 
1276 		if (vio_version_after_eq(&port->vio, 1, 3))
1277 			localmtu -= VLAN_HLEN;
1278 
1279 		if (skb->protocol == htons(ETH_P_IP)) {
1280 			struct flowi4 fl4;
1281 			struct rtable *rt = NULL;
1282 
1283 			memset(&fl4, 0, sizeof(fl4));
1284 			fl4.flowi4_oif = dev->ifindex;
1285 			fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
1286 			fl4.daddr = ip_hdr(skb)->daddr;
1287 			fl4.saddr = ip_hdr(skb)->saddr;
1288 
1289 			rt = ip_route_output_key(dev_net(dev), &fl4);
1290 			rcu_read_unlock();
1291 			if (!IS_ERR(rt)) {
1292 				skb_dst_set(skb, &rt->dst);
1293 				icmp_send(skb, ICMP_DEST_UNREACH,
1294 					  ICMP_FRAG_NEEDED,
1295 					  htonl(localmtu));
1296 			}
1297 		}
1298 #if IS_ENABLED(CONFIG_IPV6)
1299 		else if (skb->protocol == htons(ETH_P_IPV6))
1300 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1301 #endif
1302 		goto out_dropped;
1303 	}
1304 
1305 	skb = vnet_skb_shape(skb, 2);
1306 
1307 	if (unlikely(!skb))
1308 		goto out_dropped;
1309 
1310 	if (skb->ip_summed == CHECKSUM_PARTIAL)
1311 		vnet_fullcsum(skb);
1312 
1313 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1314 	i = skb_get_queue_mapping(skb);
1315 	txq = netdev_get_tx_queue(dev, i);
1316 	if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1317 		if (!netif_tx_queue_stopped(txq)) {
1318 			netif_tx_stop_queue(txq);
1319 
1320 			/* This is a hard error, log it. */
1321 			netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
1322 			dev->stats.tx_errors++;
1323 		}
1324 		rcu_read_unlock();
1325 		return NETDEV_TX_BUSY;
1326 	}
1327 
1328 	d = vio_dring_cur(dr);
1329 
1330 	txi = dr->prod;
1331 
1332 	freeskbs = vnet_clean_tx_ring(port, &pending);
1333 
1334 	BUG_ON(port->tx_bufs[txi].skb);
1335 
1336 	len = skb->len;
1337 	if (len < ETH_ZLEN)
1338 		len = ETH_ZLEN;
1339 
1340 	err = vnet_skb_map(port->vio.lp, skb, port->tx_bufs[txi].cookies, 2,
1341 			   (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1342 	if (err < 0) {
1343 		netdev_info(dev, "tx buffer map error %d\n", err);
1344 		goto out_dropped;
1345 	}
1346 
1347 	port->tx_bufs[txi].skb = skb;
1348 	skb = NULL;
1349 	port->tx_bufs[txi].ncookies = err;
1350 
1351 	/* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1352 	 * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1353 	 * the protocol itself does not require it as long as the peer
1354 	 * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1355 	 *
1356 	 * An ACK for every packet in the ring is expensive as the
1357 	 * sending of LDC messages is slow and affects performance.
1358 	 */
1359 	d->hdr.ack = VIO_ACK_DISABLE;
1360 	d->size = len;
1361 	d->ncookies = port->tx_bufs[txi].ncookies;
1362 	for (i = 0; i < d->ncookies; i++)
1363 		d->cookies[i] = port->tx_bufs[txi].cookies[i];
1364 	if (vio_version_after_eq(&port->vio, 1, 7)) {
1365 		struct vio_net_dext *dext = vio_net_ext(d);
1366 
1367 		memset(dext, 0, sizeof(*dext));
1368 		if (skb_is_gso(port->tx_bufs[txi].skb)) {
1369 			dext->ipv4_lso_mss = skb_shinfo(port->tx_bufs[txi].skb)
1370 					     ->gso_size;
1371 			dext->flags |= VNET_PKT_IPV4_LSO;
1372 		}
1373 		if (vio_version_after_eq(&port->vio, 1, 8) &&
1374 		    !port->switch_port) {
1375 			dext->flags |= VNET_PKT_HCK_IPV4_HDRCKSUM_OK;
1376 			dext->flags |= VNET_PKT_HCK_FULLCKSUM_OK;
1377 		}
1378 	}
1379 
1380 	/* This has to be a non-SMP write barrier because we are writing
1381 	 * to memory which is shared with the peer LDOM.
1382 	 */
1383 	dma_wmb();
1384 
1385 	d->hdr.state = VIO_DESC_READY;
1386 
1387 	/* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1388 	 * to notify the consumer that some descriptors are READY.
1389 	 * After that "start" trigger, no additional triggers are needed until
1390 	 * a DRING_STOPPED is received from the consumer. The dr->cons field
1391 	 * (set up by vnet_ack()) has the value of the next dring index
1392 	 * that has not yet been ack-ed. We send a "start" trigger here
1393 	 * if, and only if, start_cons is true (reset it afterward). Conversely,
1394 	 * vnet_ack() should check if the dring corresponding to cons
1395 	 * is marked READY, but start_cons was false.
1396 	 * If so, vnet_ack() should send out the missed "start" trigger.
1397 	 *
1398 	 * Note that the dma_wmb() above makes sure the cookies et al. are
1399 	 * not globally visible before the VIO_DESC_READY, and that the
1400 	 * stores are ordered correctly by the compiler. The consumer will
1401 	 * not proceed until the VIO_DESC_READY is visible assuring that
1402 	 * the consumer does not observe anything related to descriptors
1403 	 * out of order. The HV trap from the LDC start trigger is the
1404 	 * producer to consumer announcement that work is available to the
1405 	 * consumer
1406 	 */
1407 	if (!port->start_cons)
1408 		goto ldc_start_done; /* previous trigger suffices */
1409 
1410 	err = __vnet_tx_trigger(port, dr->cons);
1411 	if (unlikely(err < 0)) {
1412 		netdev_info(dev, "TX trigger error %d\n", err);
1413 		d->hdr.state = VIO_DESC_FREE;
1414 		skb = port->tx_bufs[txi].skb;
1415 		port->tx_bufs[txi].skb = NULL;
1416 		dev->stats.tx_carrier_errors++;
1417 		goto out_dropped;
1418 	}
1419 
1420 ldc_start_done:
1421 	port->start_cons = false;
1422 
1423 	dev->stats.tx_packets++;
1424 	dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1425 
1426 	dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1427 	if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1428 		netif_tx_stop_queue(txq);
1429 		if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1430 			netif_tx_wake_queue(txq);
1431 	}
1432 
1433 	(void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1434 	rcu_read_unlock();
1435 
1436 	vnet_free_skbs(freeskbs);
1437 
1438 	return NETDEV_TX_OK;
1439 
1440 out_dropped:
1441 	if (pending)
1442 		(void)mod_timer(&port->clean_timer,
1443 				jiffies + VNET_CLEAN_TIMEOUT);
1444 	else if (port)
1445 		del_timer(&port->clean_timer);
1446 	if (port)
1447 		rcu_read_unlock();
1448 	if (skb)
1449 		dev_kfree_skb(skb);
1450 	vnet_free_skbs(freeskbs);
1451 	dev->stats.tx_dropped++;
1452 	return NETDEV_TX_OK;
1453 }
1454 
1455 static void vnet_tx_timeout(struct net_device *dev)
1456 {
1457 	/* XXX Implement me XXX */
1458 }
1459 
1460 static int vnet_open(struct net_device *dev)
1461 {
1462 	netif_carrier_on(dev);
1463 	netif_tx_start_all_queues(dev);
1464 
1465 	return 0;
1466 }
1467 
1468 static int vnet_close(struct net_device *dev)
1469 {
1470 	netif_tx_stop_all_queues(dev);
1471 	netif_carrier_off(dev);
1472 
1473 	return 0;
1474 }
1475 
1476 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1477 {
1478 	struct vnet_mcast_entry *m;
1479 
1480 	for (m = vp->mcast_list; m; m = m->next) {
1481 		if (ether_addr_equal(m->addr, addr))
1482 			return m;
1483 	}
1484 	return NULL;
1485 }
1486 
1487 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1488 {
1489 	struct netdev_hw_addr *ha;
1490 
1491 	netdev_for_each_mc_addr(ha, dev) {
1492 		struct vnet_mcast_entry *m;
1493 
1494 		m = __vnet_mc_find(vp, ha->addr);
1495 		if (m) {
1496 			m->hit = 1;
1497 			continue;
1498 		}
1499 
1500 		if (!m) {
1501 			m = kzalloc(sizeof(*m), GFP_ATOMIC);
1502 			if (!m)
1503 				continue;
1504 			memcpy(m->addr, ha->addr, ETH_ALEN);
1505 			m->hit = 1;
1506 
1507 			m->next = vp->mcast_list;
1508 			vp->mcast_list = m;
1509 		}
1510 	}
1511 }
1512 
1513 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1514 {
1515 	struct vio_net_mcast_info info;
1516 	struct vnet_mcast_entry *m, **pp;
1517 	int n_addrs;
1518 
1519 	memset(&info, 0, sizeof(info));
1520 
1521 	info.tag.type = VIO_TYPE_CTRL;
1522 	info.tag.stype = VIO_SUBTYPE_INFO;
1523 	info.tag.stype_env = VNET_MCAST_INFO;
1524 	info.tag.sid = vio_send_sid(&port->vio);
1525 	info.set = 1;
1526 
1527 	n_addrs = 0;
1528 	for (m = vp->mcast_list; m; m = m->next) {
1529 		if (m->sent)
1530 			continue;
1531 		m->sent = 1;
1532 		memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1533 		       m->addr, ETH_ALEN);
1534 		if (++n_addrs == VNET_NUM_MCAST) {
1535 			info.count = n_addrs;
1536 
1537 			(void) vio_ldc_send(&port->vio, &info,
1538 					    sizeof(info));
1539 			n_addrs = 0;
1540 		}
1541 	}
1542 	if (n_addrs) {
1543 		info.count = n_addrs;
1544 		(void) vio_ldc_send(&port->vio, &info, sizeof(info));
1545 	}
1546 
1547 	info.set = 0;
1548 
1549 	n_addrs = 0;
1550 	pp = &vp->mcast_list;
1551 	while ((m = *pp) != NULL) {
1552 		if (m->hit) {
1553 			m->hit = 0;
1554 			pp = &m->next;
1555 			continue;
1556 		}
1557 
1558 		memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1559 		       m->addr, ETH_ALEN);
1560 		if (++n_addrs == VNET_NUM_MCAST) {
1561 			info.count = n_addrs;
1562 			(void) vio_ldc_send(&port->vio, &info,
1563 					    sizeof(info));
1564 			n_addrs = 0;
1565 		}
1566 
1567 		*pp = m->next;
1568 		kfree(m);
1569 	}
1570 	if (n_addrs) {
1571 		info.count = n_addrs;
1572 		(void) vio_ldc_send(&port->vio, &info, sizeof(info));
1573 	}
1574 }
1575 
1576 static void vnet_set_rx_mode(struct net_device *dev)
1577 {
1578 	struct vnet *vp = netdev_priv(dev);
1579 	struct vnet_port *port;
1580 
1581 	rcu_read_lock();
1582 	list_for_each_entry_rcu(port, &vp->port_list, list) {
1583 
1584 		if (port->switch_port) {
1585 			__update_mc_list(vp, dev);
1586 			__send_mc_list(vp, port);
1587 			break;
1588 		}
1589 	}
1590 	rcu_read_unlock();
1591 }
1592 
1593 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
1594 {
1595 	if (new_mtu < 68 || new_mtu > 65535)
1596 		return -EINVAL;
1597 
1598 	dev->mtu = new_mtu;
1599 	return 0;
1600 }
1601 
1602 static int vnet_set_mac_addr(struct net_device *dev, void *p)
1603 {
1604 	return -EINVAL;
1605 }
1606 
1607 static void vnet_get_drvinfo(struct net_device *dev,
1608 			     struct ethtool_drvinfo *info)
1609 {
1610 	strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
1611 	strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
1612 }
1613 
1614 static u32 vnet_get_msglevel(struct net_device *dev)
1615 {
1616 	struct vnet *vp = netdev_priv(dev);
1617 	return vp->msg_enable;
1618 }
1619 
1620 static void vnet_set_msglevel(struct net_device *dev, u32 value)
1621 {
1622 	struct vnet *vp = netdev_priv(dev);
1623 	vp->msg_enable = value;
1624 }
1625 
1626 static const struct ethtool_ops vnet_ethtool_ops = {
1627 	.get_drvinfo		= vnet_get_drvinfo,
1628 	.get_msglevel		= vnet_get_msglevel,
1629 	.set_msglevel		= vnet_set_msglevel,
1630 	.get_link		= ethtool_op_get_link,
1631 };
1632 
1633 static void vnet_port_free_tx_bufs(struct vnet_port *port)
1634 {
1635 	struct vio_dring_state *dr;
1636 	int i;
1637 
1638 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1639 
1640 	if (dr->base == NULL)
1641 		return;
1642 
1643 	for (i = 0; i < VNET_TX_RING_SIZE; i++) {
1644 		struct vio_net_desc *d;
1645 		void *skb = port->tx_bufs[i].skb;
1646 
1647 		if (!skb)
1648 			continue;
1649 
1650 		d = vio_dring_entry(dr, i);
1651 
1652 		ldc_unmap(port->vio.lp,
1653 			  port->tx_bufs[i].cookies,
1654 			  port->tx_bufs[i].ncookies);
1655 		dev_kfree_skb(skb);
1656 		port->tx_bufs[i].skb = NULL;
1657 		d->hdr.state = VIO_DESC_FREE;
1658 	}
1659 	ldc_free_exp_dring(port->vio.lp, dr->base,
1660 			   (dr->entry_size * dr->num_entries),
1661 			   dr->cookies, dr->ncookies);
1662 	dr->base = NULL;
1663 	dr->entry_size = 0;
1664 	dr->num_entries = 0;
1665 	dr->pending = 0;
1666 	dr->ncookies = 0;
1667 }
1668 
1669 static void vnet_port_reset(struct vnet_port *port)
1670 {
1671 	del_timer(&port->clean_timer);
1672 	vnet_port_free_tx_bufs(port);
1673 	port->rmtu = 0;
1674 	port->tso = true;
1675 	port->tsolen = 0;
1676 }
1677 
1678 static int vnet_port_alloc_tx_ring(struct vnet_port *port)
1679 {
1680 	struct vio_dring_state *dr;
1681 	unsigned long len, elen;
1682 	int i, err, ncookies;
1683 	void *dring;
1684 
1685 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1686 
1687 	elen = sizeof(struct vio_net_desc) +
1688 	       sizeof(struct ldc_trans_cookie) * 2;
1689 	if (vio_version_after_eq(&port->vio, 1, 7))
1690 		elen += sizeof(struct vio_net_dext);
1691 	len = VNET_TX_RING_SIZE * elen;
1692 
1693 	ncookies = VIO_MAX_RING_COOKIES;
1694 	dring = ldc_alloc_exp_dring(port->vio.lp, len,
1695 				    dr->cookies, &ncookies,
1696 				    (LDC_MAP_SHADOW |
1697 				     LDC_MAP_DIRECT |
1698 				     LDC_MAP_RW));
1699 	if (IS_ERR(dring)) {
1700 		err = PTR_ERR(dring);
1701 		goto err_out;
1702 	}
1703 
1704 	dr->base = dring;
1705 	dr->entry_size = elen;
1706 	dr->num_entries = VNET_TX_RING_SIZE;
1707 	dr->prod = dr->cons = 0;
1708 	port->start_cons  = true; /* need an initial trigger */
1709 	dr->pending = VNET_TX_RING_SIZE;
1710 	dr->ncookies = ncookies;
1711 
1712 	for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1713 		struct vio_net_desc *d;
1714 
1715 		d = vio_dring_entry(dr, i);
1716 		d->hdr.state = VIO_DESC_FREE;
1717 	}
1718 	return 0;
1719 
1720 err_out:
1721 	vnet_port_free_tx_bufs(port);
1722 
1723 	return err;
1724 }
1725 
1726 #ifdef CONFIG_NET_POLL_CONTROLLER
1727 static void vnet_poll_controller(struct net_device *dev)
1728 {
1729 	struct vnet *vp = netdev_priv(dev);
1730 	struct vnet_port *port;
1731 	unsigned long flags;
1732 
1733 	spin_lock_irqsave(&vp->lock, flags);
1734 	if (!list_empty(&vp->port_list)) {
1735 		port = list_entry(vp->port_list.next, struct vnet_port, list);
1736 		napi_schedule(&port->napi);
1737 	}
1738 	spin_unlock_irqrestore(&vp->lock, flags);
1739 }
1740 #endif
1741 static LIST_HEAD(vnet_list);
1742 static DEFINE_MUTEX(vnet_list_mutex);
1743 
1744 static const struct net_device_ops vnet_ops = {
1745 	.ndo_open		= vnet_open,
1746 	.ndo_stop		= vnet_close,
1747 	.ndo_set_rx_mode	= vnet_set_rx_mode,
1748 	.ndo_set_mac_address	= vnet_set_mac_addr,
1749 	.ndo_validate_addr	= eth_validate_addr,
1750 	.ndo_tx_timeout		= vnet_tx_timeout,
1751 	.ndo_change_mtu		= vnet_change_mtu,
1752 	.ndo_start_xmit		= vnet_start_xmit,
1753 	.ndo_select_queue	= vnet_select_queue,
1754 #ifdef CONFIG_NET_POLL_CONTROLLER
1755 	.ndo_poll_controller	= vnet_poll_controller,
1756 #endif
1757 };
1758 
1759 static struct vnet *vnet_new(const u64 *local_mac,
1760 			     struct vio_dev *vdev)
1761 {
1762 	struct net_device *dev;
1763 	struct vnet *vp;
1764 	int err, i;
1765 
1766 	dev = alloc_etherdev_mqs(sizeof(*vp), VNET_MAX_TXQS, 1);
1767 	if (!dev)
1768 		return ERR_PTR(-ENOMEM);
1769 	dev->needed_headroom = VNET_PACKET_SKIP + 8;
1770 	dev->needed_tailroom = 8;
1771 
1772 	for (i = 0; i < ETH_ALEN; i++)
1773 		dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1774 
1775 	vp = netdev_priv(dev);
1776 
1777 	spin_lock_init(&vp->lock);
1778 	vp->dev = dev;
1779 
1780 	INIT_LIST_HEAD(&vp->port_list);
1781 	for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1782 		INIT_HLIST_HEAD(&vp->port_hash[i]);
1783 	INIT_LIST_HEAD(&vp->list);
1784 	vp->local_mac = *local_mac;
1785 
1786 	dev->netdev_ops = &vnet_ops;
1787 	dev->ethtool_ops = &vnet_ethtool_ops;
1788 	dev->watchdog_timeo = VNET_TX_TIMEOUT;
1789 
1790 	dev->hw_features = NETIF_F_TSO | NETIF_F_GSO | NETIF_F_GSO_SOFTWARE |
1791 			   NETIF_F_HW_CSUM | NETIF_F_SG;
1792 	dev->features = dev->hw_features;
1793 
1794 	SET_NETDEV_DEV(dev, &vdev->dev);
1795 
1796 	err = register_netdev(dev);
1797 	if (err) {
1798 		pr_err("Cannot register net device, aborting\n");
1799 		goto err_out_free_dev;
1800 	}
1801 
1802 	netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1803 
1804 	list_add(&vp->list, &vnet_list);
1805 
1806 	return vp;
1807 
1808 err_out_free_dev:
1809 	free_netdev(dev);
1810 
1811 	return ERR_PTR(err);
1812 }
1813 
1814 static struct vnet *vnet_find_or_create(const u64 *local_mac,
1815 					struct vio_dev *vdev)
1816 {
1817 	struct vnet *iter, *vp;
1818 
1819 	mutex_lock(&vnet_list_mutex);
1820 	vp = NULL;
1821 	list_for_each_entry(iter, &vnet_list, list) {
1822 		if (iter->local_mac == *local_mac) {
1823 			vp = iter;
1824 			break;
1825 		}
1826 	}
1827 	if (!vp)
1828 		vp = vnet_new(local_mac, vdev);
1829 	mutex_unlock(&vnet_list_mutex);
1830 
1831 	return vp;
1832 }
1833 
1834 static void vnet_cleanup(void)
1835 {
1836 	struct vnet *vp;
1837 	struct net_device *dev;
1838 
1839 	mutex_lock(&vnet_list_mutex);
1840 	while (!list_empty(&vnet_list)) {
1841 		vp = list_first_entry(&vnet_list, struct vnet, list);
1842 		list_del(&vp->list);
1843 		dev = vp->dev;
1844 		/* vio_unregister_driver() should have cleaned up port_list */
1845 		BUG_ON(!list_empty(&vp->port_list));
1846 		unregister_netdev(dev);
1847 		free_netdev(dev);
1848 	}
1849 	mutex_unlock(&vnet_list_mutex);
1850 }
1851 
1852 static const char *local_mac_prop = "local-mac-address";
1853 
1854 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1855 				     u64 port_node,
1856 				     struct vio_dev *vdev)
1857 {
1858 	const u64 *local_mac = NULL;
1859 	u64 a;
1860 
1861 	mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1862 		u64 target = mdesc_arc_target(hp, a);
1863 		const char *name;
1864 
1865 		name = mdesc_get_property(hp, target, "name", NULL);
1866 		if (!name || strcmp(name, "network"))
1867 			continue;
1868 
1869 		local_mac = mdesc_get_property(hp, target,
1870 					       local_mac_prop, NULL);
1871 		if (local_mac)
1872 			break;
1873 	}
1874 	if (!local_mac)
1875 		return ERR_PTR(-ENODEV);
1876 
1877 	return vnet_find_or_create(local_mac, vdev);
1878 }
1879 
1880 static struct ldc_channel_config vnet_ldc_cfg = {
1881 	.event		= vnet_event,
1882 	.mtu		= 64,
1883 	.mode		= LDC_MODE_UNRELIABLE,
1884 };
1885 
1886 static struct vio_driver_ops vnet_vio_ops = {
1887 	.send_attr		= vnet_send_attr,
1888 	.handle_attr		= vnet_handle_attr,
1889 	.handshake_complete	= vnet_handshake_complete,
1890 };
1891 
1892 static void print_version(void)
1893 {
1894 	printk_once(KERN_INFO "%s", version);
1895 }
1896 
1897 const char *remote_macaddr_prop = "remote-mac-address";
1898 
1899 static void
1900 vnet_port_add_txq(struct vnet_port *port)
1901 {
1902 	struct vnet *vp = port->vp;
1903 	int n;
1904 
1905 	n = vp->nports++;
1906 	n = n & (VNET_MAX_TXQS - 1);
1907 	port->q_index = n;
1908 	netif_tx_wake_queue(netdev_get_tx_queue(vp->dev, port->q_index));
1909 }
1910 
1911 static void
1912 vnet_port_rm_txq(struct vnet_port *port)
1913 {
1914 	port->vp->nports--;
1915 	netif_tx_stop_queue(netdev_get_tx_queue(port->vp->dev, port->q_index));
1916 }
1917 
1918 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1919 {
1920 	struct mdesc_handle *hp;
1921 	struct vnet_port *port;
1922 	unsigned long flags;
1923 	struct vnet *vp;
1924 	const u64 *rmac;
1925 	int len, i, err, switch_port;
1926 
1927 	print_version();
1928 
1929 	hp = mdesc_grab();
1930 
1931 	vp = vnet_find_parent(hp, vdev->mp, vdev);
1932 	if (IS_ERR(vp)) {
1933 		pr_err("Cannot find port parent vnet\n");
1934 		err = PTR_ERR(vp);
1935 		goto err_out_put_mdesc;
1936 	}
1937 
1938 	rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1939 	err = -ENODEV;
1940 	if (!rmac) {
1941 		pr_err("Port lacks %s property\n", remote_macaddr_prop);
1942 		goto err_out_put_mdesc;
1943 	}
1944 
1945 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1946 	err = -ENOMEM;
1947 	if (!port)
1948 		goto err_out_put_mdesc;
1949 
1950 	for (i = 0; i < ETH_ALEN; i++)
1951 		port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1952 
1953 	port->vp = vp;
1954 
1955 	err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1956 			      vnet_versions, ARRAY_SIZE(vnet_versions),
1957 			      &vnet_vio_ops, vp->dev->name);
1958 	if (err)
1959 		goto err_out_free_port;
1960 
1961 	err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1962 	if (err)
1963 		goto err_out_free_port;
1964 
1965 	netif_napi_add(port->vp->dev, &port->napi, vnet_poll, NAPI_POLL_WEIGHT);
1966 
1967 	INIT_HLIST_NODE(&port->hash);
1968 	INIT_LIST_HEAD(&port->list);
1969 
1970 	switch_port = 0;
1971 	if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1972 		switch_port = 1;
1973 	port->switch_port = switch_port;
1974 	port->tso = true;
1975 	port->tsolen = 0;
1976 
1977 	spin_lock_irqsave(&vp->lock, flags);
1978 	if (switch_port)
1979 		list_add_rcu(&port->list, &vp->port_list);
1980 	else
1981 		list_add_tail_rcu(&port->list, &vp->port_list);
1982 	hlist_add_head_rcu(&port->hash,
1983 			   &vp->port_hash[vnet_hashfn(port->raddr)]);
1984 	vnet_port_add_txq(port);
1985 	spin_unlock_irqrestore(&vp->lock, flags);
1986 
1987 	dev_set_drvdata(&vdev->dev, port);
1988 
1989 	pr_info("%s: PORT ( remote-mac %pM%s )\n",
1990 		vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1991 
1992 	setup_timer(&port->clean_timer, vnet_clean_timer_expire,
1993 		    (unsigned long)port);
1994 
1995 	napi_enable(&port->napi);
1996 	vio_port_up(&port->vio);
1997 
1998 	mdesc_release(hp);
1999 
2000 	return 0;
2001 
2002 err_out_free_port:
2003 	kfree(port);
2004 
2005 err_out_put_mdesc:
2006 	mdesc_release(hp);
2007 	return err;
2008 }
2009 
2010 static int vnet_port_remove(struct vio_dev *vdev)
2011 {
2012 	struct vnet_port *port = dev_get_drvdata(&vdev->dev);
2013 
2014 	if (port) {
2015 
2016 		del_timer_sync(&port->vio.timer);
2017 
2018 		napi_disable(&port->napi);
2019 
2020 		list_del_rcu(&port->list);
2021 		hlist_del_rcu(&port->hash);
2022 
2023 		synchronize_rcu();
2024 		del_timer_sync(&port->clean_timer);
2025 		vnet_port_rm_txq(port);
2026 		netif_napi_del(&port->napi);
2027 		vnet_port_free_tx_bufs(port);
2028 		vio_ldc_free(&port->vio);
2029 
2030 		dev_set_drvdata(&vdev->dev, NULL);
2031 
2032 		kfree(port);
2033 
2034 	}
2035 	return 0;
2036 }
2037 
2038 static const struct vio_device_id vnet_port_match[] = {
2039 	{
2040 		.type = "vnet-port",
2041 	},
2042 	{},
2043 };
2044 MODULE_DEVICE_TABLE(vio, vnet_port_match);
2045 
2046 static struct vio_driver vnet_port_driver = {
2047 	.id_table	= vnet_port_match,
2048 	.probe		= vnet_port_probe,
2049 	.remove		= vnet_port_remove,
2050 	.name		= "vnet_port",
2051 };
2052 
2053 static int __init vnet_init(void)
2054 {
2055 	return vio_register_driver(&vnet_port_driver);
2056 }
2057 
2058 static void __exit vnet_exit(void)
2059 {
2060 	vio_unregister_driver(&vnet_port_driver);
2061 	vnet_cleanup();
2062 }
2063 
2064 module_init(vnet_init);
2065 module_exit(vnet_exit);
2066