xref: /linux/drivers/net/ethernet/sun/sunvnet_common.c (revision ae22a94997b8a03dcb3c922857c203246711f9d4)
1 // SPDX-License-Identifier: GPL-2.0
2 /* sunvnet.c: Sun LDOM Virtual Network Driver.
3  *
4  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
5  * Copyright (C) 2016-2017 Oracle. All rights reserved.
6  */
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 #define CREATE_TRACE_POINTS
21 #include <trace/events/sunvnet.h>
22 
23 #if IS_ENABLED(CONFIG_IPV6)
24 #include <linux/icmpv6.h>
25 #endif
26 
27 #include <net/ip.h>
28 #include <net/gso.h>
29 #include <net/icmp.h>
30 #include <net/route.h>
31 
32 #include <asm/vio.h>
33 #include <asm/ldc.h>
34 
35 #include "sunvnet_common.h"
36 
37 /* Heuristic for the number of times to exponentially backoff and
38  * retry sending an LDC trigger when EAGAIN is encountered
39  */
40 #define	VNET_MAX_RETRIES	10
41 
42 MODULE_AUTHOR("David S. Miller <davem@davemloft.net>");
43 MODULE_DESCRIPTION("Sun LDOM virtual network support library");
44 MODULE_LICENSE("GPL");
45 MODULE_VERSION("1.1");
46 
47 static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
48 
49 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
50 {
51 	return vio_dring_avail(dr, VNET_TX_RING_SIZE);
52 }
53 
54 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
55 {
56 	struct vio_msg_tag *pkt = arg;
57 
58 	pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
59 	       pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
60 	pr_err("Resetting connection\n");
61 
62 	ldc_disconnect(port->vio.lp);
63 
64 	return -ECONNRESET;
65 }
66 
67 static int vnet_port_alloc_tx_ring(struct vnet_port *port);
68 
69 int sunvnet_send_attr_common(struct vio_driver_state *vio)
70 {
71 	struct vnet_port *port = to_vnet_port(vio);
72 	struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
73 	struct vio_net_attr_info pkt;
74 	int framelen = ETH_FRAME_LEN;
75 	int i, err;
76 
77 	err = vnet_port_alloc_tx_ring(to_vnet_port(vio));
78 	if (err)
79 		return err;
80 
81 	memset(&pkt, 0, sizeof(pkt));
82 	pkt.tag.type = VIO_TYPE_CTRL;
83 	pkt.tag.stype = VIO_SUBTYPE_INFO;
84 	pkt.tag.stype_env = VIO_ATTR_INFO;
85 	pkt.tag.sid = vio_send_sid(vio);
86 	if (vio_version_before(vio, 1, 2))
87 		pkt.xfer_mode = VIO_DRING_MODE;
88 	else
89 		pkt.xfer_mode = VIO_NEW_DRING_MODE;
90 	pkt.addr_type = VNET_ADDR_ETHERMAC;
91 	pkt.ack_freq = 0;
92 	for (i = 0; i < 6; i++)
93 		pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
94 	if (vio_version_after(vio, 1, 3)) {
95 		if (port->rmtu) {
96 			port->rmtu = min(VNET_MAXPACKET, port->rmtu);
97 			pkt.mtu = port->rmtu;
98 		} else {
99 			port->rmtu = VNET_MAXPACKET;
100 			pkt.mtu = port->rmtu;
101 		}
102 		if (vio_version_after_eq(vio, 1, 6))
103 			pkt.options = VIO_TX_DRING;
104 	} else if (vio_version_before(vio, 1, 3)) {
105 		pkt.mtu = framelen;
106 	} else { /* v1.3 */
107 		pkt.mtu = framelen + VLAN_HLEN;
108 	}
109 
110 	pkt.cflags = 0;
111 	if (vio_version_after_eq(vio, 1, 7) && port->tso) {
112 		pkt.cflags |= VNET_LSO_IPV4_CAPAB;
113 		if (!port->tsolen)
114 			port->tsolen = VNET_MAXTSO;
115 		pkt.ipv4_lso_maxlen = port->tsolen;
116 	}
117 
118 	pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
119 
120 	viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
121 	       "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
122 	       "cflags[0x%04x] lso_max[%u]\n",
123 	       pkt.xfer_mode, pkt.addr_type,
124 	       (unsigned long long)pkt.addr,
125 	       pkt.ack_freq, pkt.plnk_updt, pkt.options,
126 	       (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
127 
128 	return vio_ldc_send(vio, &pkt, sizeof(pkt));
129 }
130 EXPORT_SYMBOL_GPL(sunvnet_send_attr_common);
131 
132 static int handle_attr_info(struct vio_driver_state *vio,
133 			    struct vio_net_attr_info *pkt)
134 {
135 	struct vnet_port *port = to_vnet_port(vio);
136 	u64	localmtu;
137 	u8	xfer_mode;
138 
139 	viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
140 	       "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
141 	       " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
142 	       pkt->xfer_mode, pkt->addr_type,
143 	       (unsigned long long)pkt->addr,
144 	       pkt->ack_freq, pkt->plnk_updt, pkt->options,
145 	       (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
146 	       pkt->ipv4_lso_maxlen);
147 
148 	pkt->tag.sid = vio_send_sid(vio);
149 
150 	xfer_mode = pkt->xfer_mode;
151 	/* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
152 	if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
153 		xfer_mode = VIO_NEW_DRING_MODE;
154 
155 	/* MTU negotiation:
156 	 *	< v1.3 - ETH_FRAME_LEN exactly
157 	 *	> v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
158 	 *			pkt->mtu for ACK
159 	 *	= v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
160 	 */
161 	if (vio_version_before(vio, 1, 3)) {
162 		localmtu = ETH_FRAME_LEN;
163 	} else if (vio_version_after(vio, 1, 3)) {
164 		localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
165 		localmtu = min(pkt->mtu, localmtu);
166 		pkt->mtu = localmtu;
167 	} else { /* v1.3 */
168 		localmtu = ETH_FRAME_LEN + VLAN_HLEN;
169 	}
170 	port->rmtu = localmtu;
171 
172 	/* LSO negotiation */
173 	if (vio_version_after_eq(vio, 1, 7))
174 		port->tso &= !!(pkt->cflags & VNET_LSO_IPV4_CAPAB);
175 	else
176 		port->tso = false;
177 	if (port->tso) {
178 		if (!port->tsolen)
179 			port->tsolen = VNET_MAXTSO;
180 		port->tsolen = min(port->tsolen, pkt->ipv4_lso_maxlen);
181 		if (port->tsolen < VNET_MINTSO) {
182 			port->tso = false;
183 			port->tsolen = 0;
184 			pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
185 		}
186 		pkt->ipv4_lso_maxlen = port->tsolen;
187 	} else {
188 		pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
189 		pkt->ipv4_lso_maxlen = 0;
190 		port->tsolen = 0;
191 	}
192 
193 	/* for version >= 1.6, ACK packet mode we support */
194 	if (vio_version_after_eq(vio, 1, 6)) {
195 		pkt->xfer_mode = VIO_NEW_DRING_MODE;
196 		pkt->options = VIO_TX_DRING;
197 	}
198 
199 	if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
200 	    pkt->addr_type != VNET_ADDR_ETHERMAC ||
201 	    pkt->mtu != localmtu) {
202 		viodbg(HS, "SEND NET ATTR NACK\n");
203 
204 		pkt->tag.stype = VIO_SUBTYPE_NACK;
205 
206 		(void)vio_ldc_send(vio, pkt, sizeof(*pkt));
207 
208 		return -ECONNRESET;
209 	}
210 
211 	viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
212 	       "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
213 	       "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
214 	       pkt->xfer_mode, pkt->addr_type,
215 	       (unsigned long long)pkt->addr,
216 	       pkt->ack_freq, pkt->plnk_updt, pkt->options,
217 	       (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
218 	       pkt->ipv4_lso_maxlen);
219 
220 	pkt->tag.stype = VIO_SUBTYPE_ACK;
221 
222 	return vio_ldc_send(vio, pkt, sizeof(*pkt));
223 }
224 
225 static int handle_attr_ack(struct vio_driver_state *vio,
226 			   struct vio_net_attr_info *pkt)
227 {
228 	viodbg(HS, "GOT NET ATTR ACK\n");
229 
230 	return 0;
231 }
232 
233 static int handle_attr_nack(struct vio_driver_state *vio,
234 			    struct vio_net_attr_info *pkt)
235 {
236 	viodbg(HS, "GOT NET ATTR NACK\n");
237 
238 	return -ECONNRESET;
239 }
240 
241 int sunvnet_handle_attr_common(struct vio_driver_state *vio, void *arg)
242 {
243 	struct vio_net_attr_info *pkt = arg;
244 
245 	switch (pkt->tag.stype) {
246 	case VIO_SUBTYPE_INFO:
247 		return handle_attr_info(vio, pkt);
248 
249 	case VIO_SUBTYPE_ACK:
250 		return handle_attr_ack(vio, pkt);
251 
252 	case VIO_SUBTYPE_NACK:
253 		return handle_attr_nack(vio, pkt);
254 
255 	default:
256 		return -ECONNRESET;
257 	}
258 }
259 EXPORT_SYMBOL_GPL(sunvnet_handle_attr_common);
260 
261 void sunvnet_handshake_complete_common(struct vio_driver_state *vio)
262 {
263 	struct vio_dring_state *dr;
264 
265 	dr = &vio->drings[VIO_DRIVER_RX_RING];
266 	dr->rcv_nxt = 1;
267 	dr->snd_nxt = 1;
268 
269 	dr = &vio->drings[VIO_DRIVER_TX_RING];
270 	dr->rcv_nxt = 1;
271 	dr->snd_nxt = 1;
272 }
273 EXPORT_SYMBOL_GPL(sunvnet_handshake_complete_common);
274 
275 /* The hypervisor interface that implements copying to/from imported
276  * memory from another domain requires that copies are done to 8-byte
277  * aligned buffers, and that the lengths of such copies are also 8-byte
278  * multiples.
279  *
280  * So we align skb->data to an 8-byte multiple and pad-out the data
281  * area so we can round the copy length up to the next multiple of
282  * 8 for the copy.
283  *
284  * The transmitter puts the actual start of the packet 6 bytes into
285  * the buffer it sends over, so that the IP headers after the ethernet
286  * header are aligned properly.  These 6 bytes are not in the descriptor
287  * length, they are simply implied.  This offset is represented using
288  * the VNET_PACKET_SKIP macro.
289  */
290 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
291 					   unsigned int len)
292 {
293 	struct sk_buff *skb;
294 	unsigned long addr, off;
295 
296 	skb = netdev_alloc_skb(dev, len + VNET_PACKET_SKIP + 8 + 8);
297 	if (unlikely(!skb))
298 		return NULL;
299 
300 	addr = (unsigned long)skb->data;
301 	off = ((addr + 7UL) & ~7UL) - addr;
302 	if (off)
303 		skb_reserve(skb, off);
304 
305 	return skb;
306 }
307 
308 static inline void vnet_fullcsum_ipv4(struct sk_buff *skb)
309 {
310 	struct iphdr *iph = ip_hdr(skb);
311 	int offset = skb_transport_offset(skb);
312 
313 	if (skb->protocol != htons(ETH_P_IP))
314 		return;
315 	if (iph->protocol != IPPROTO_TCP &&
316 	    iph->protocol != IPPROTO_UDP)
317 		return;
318 	skb->ip_summed = CHECKSUM_NONE;
319 	skb->csum_level = 1;
320 	skb->csum = 0;
321 	if (iph->protocol == IPPROTO_TCP) {
322 		struct tcphdr *ptcp = tcp_hdr(skb);
323 
324 		ptcp->check = 0;
325 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
326 		ptcp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
327 						skb->len - offset, IPPROTO_TCP,
328 						skb->csum);
329 	} else if (iph->protocol == IPPROTO_UDP) {
330 		struct udphdr *pudp = udp_hdr(skb);
331 
332 		pudp->check = 0;
333 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
334 		pudp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
335 						skb->len - offset, IPPROTO_UDP,
336 						skb->csum);
337 	}
338 }
339 
340 #if IS_ENABLED(CONFIG_IPV6)
341 static inline void vnet_fullcsum_ipv6(struct sk_buff *skb)
342 {
343 	struct ipv6hdr *ip6h = ipv6_hdr(skb);
344 	int offset = skb_transport_offset(skb);
345 
346 	if (skb->protocol != htons(ETH_P_IPV6))
347 		return;
348 	if (ip6h->nexthdr != IPPROTO_TCP &&
349 	    ip6h->nexthdr != IPPROTO_UDP)
350 		return;
351 	skb->ip_summed = CHECKSUM_NONE;
352 	skb->csum_level = 1;
353 	skb->csum = 0;
354 	if (ip6h->nexthdr == IPPROTO_TCP) {
355 		struct tcphdr *ptcp = tcp_hdr(skb);
356 
357 		ptcp->check = 0;
358 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
359 		ptcp->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
360 					      skb->len - offset, IPPROTO_TCP,
361 					      skb->csum);
362 	} else if (ip6h->nexthdr == IPPROTO_UDP) {
363 		struct udphdr *pudp = udp_hdr(skb);
364 
365 		pudp->check = 0;
366 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
367 		pudp->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
368 					      skb->len - offset, IPPROTO_UDP,
369 					      skb->csum);
370 	}
371 }
372 #endif
373 
374 static int vnet_rx_one(struct vnet_port *port, struct vio_net_desc *desc)
375 {
376 	struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
377 	unsigned int len = desc->size;
378 	unsigned int copy_len;
379 	struct sk_buff *skb;
380 	int maxlen;
381 	int err;
382 
383 	err = -EMSGSIZE;
384 	if (port->tso && port->tsolen > port->rmtu)
385 		maxlen = port->tsolen;
386 	else
387 		maxlen = port->rmtu;
388 	if (unlikely(len < ETH_ZLEN || len > maxlen)) {
389 		dev->stats.rx_length_errors++;
390 		goto out_dropped;
391 	}
392 
393 	skb = alloc_and_align_skb(dev, len);
394 	err = -ENOMEM;
395 	if (unlikely(!skb)) {
396 		dev->stats.rx_missed_errors++;
397 		goto out_dropped;
398 	}
399 
400 	copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
401 	skb_put(skb, copy_len);
402 	err = ldc_copy(port->vio.lp, LDC_COPY_IN,
403 		       skb->data, copy_len, 0,
404 		       desc->cookies, desc->ncookies);
405 	if (unlikely(err < 0)) {
406 		dev->stats.rx_frame_errors++;
407 		goto out_free_skb;
408 	}
409 
410 	skb_pull(skb, VNET_PACKET_SKIP);
411 	skb_trim(skb, len);
412 	skb->protocol = eth_type_trans(skb, dev);
413 
414 	if (vio_version_after_eq(&port->vio, 1, 8)) {
415 		struct vio_net_dext *dext = vio_net_ext(desc);
416 
417 		skb_reset_network_header(skb);
418 
419 		if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM) {
420 			if (skb->protocol == ETH_P_IP) {
421 				struct iphdr *iph = ip_hdr(skb);
422 
423 				iph->check = 0;
424 				ip_send_check(iph);
425 			}
426 		}
427 		if ((dext->flags & VNET_PKT_HCK_FULLCKSUM) &&
428 		    skb->ip_summed == CHECKSUM_NONE) {
429 			if (skb->protocol == htons(ETH_P_IP)) {
430 				struct iphdr *iph = ip_hdr(skb);
431 				int ihl = iph->ihl * 4;
432 
433 				skb_set_transport_header(skb, ihl);
434 				vnet_fullcsum_ipv4(skb);
435 #if IS_ENABLED(CONFIG_IPV6)
436 			} else if (skb->protocol == htons(ETH_P_IPV6)) {
437 				skb_set_transport_header(skb,
438 							 sizeof(struct ipv6hdr));
439 				vnet_fullcsum_ipv6(skb);
440 #endif
441 			}
442 		}
443 		if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM_OK) {
444 			skb->ip_summed = CHECKSUM_PARTIAL;
445 			skb->csum_level = 0;
446 			if (dext->flags & VNET_PKT_HCK_FULLCKSUM_OK)
447 				skb->csum_level = 1;
448 		}
449 	}
450 
451 	skb->ip_summed = port->switch_port ? CHECKSUM_NONE : CHECKSUM_PARTIAL;
452 
453 	if (unlikely(is_multicast_ether_addr(eth_hdr(skb)->h_dest)))
454 		dev->stats.multicast++;
455 	dev->stats.rx_packets++;
456 	dev->stats.rx_bytes += len;
457 	port->stats.rx_packets++;
458 	port->stats.rx_bytes += len;
459 	napi_gro_receive(&port->napi, skb);
460 	return 0;
461 
462 out_free_skb:
463 	kfree_skb(skb);
464 
465 out_dropped:
466 	dev->stats.rx_dropped++;
467 	return err;
468 }
469 
470 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
471 			 u32 start, u32 end, u8 vio_dring_state)
472 {
473 	struct vio_dring_data hdr = {
474 		.tag = {
475 			.type		= VIO_TYPE_DATA,
476 			.stype		= VIO_SUBTYPE_ACK,
477 			.stype_env	= VIO_DRING_DATA,
478 			.sid		= vio_send_sid(&port->vio),
479 		},
480 		.dring_ident		= dr->ident,
481 		.start_idx		= start,
482 		.end_idx		= end,
483 		.state			= vio_dring_state,
484 	};
485 	int err, delay;
486 	int retries = 0;
487 
488 	hdr.seq = dr->snd_nxt;
489 	delay = 1;
490 	do {
491 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
492 		if (err > 0) {
493 			dr->snd_nxt++;
494 			break;
495 		}
496 		udelay(delay);
497 		if ((delay <<= 1) > 128)
498 			delay = 128;
499 		if (retries++ > VNET_MAX_RETRIES) {
500 			pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
501 				port->raddr[0], port->raddr[1],
502 				port->raddr[2], port->raddr[3],
503 				port->raddr[4], port->raddr[5]);
504 			break;
505 		}
506 	} while (err == -EAGAIN);
507 
508 	if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
509 		port->stop_rx_idx = end;
510 		port->stop_rx = true;
511 	} else {
512 		port->stop_rx_idx = 0;
513 		port->stop_rx = false;
514 	}
515 
516 	return err;
517 }
518 
519 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
520 					struct vio_dring_state *dr,
521 					u32 index)
522 {
523 	struct vio_net_desc *desc = port->vio.desc_buf;
524 	int err;
525 
526 	err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
527 				  (index * dr->entry_size),
528 				  dr->cookies, dr->ncookies);
529 	if (err < 0)
530 		return ERR_PTR(err);
531 
532 	return desc;
533 }
534 
535 static int put_rx_desc(struct vnet_port *port,
536 		       struct vio_dring_state *dr,
537 		       struct vio_net_desc *desc,
538 		       u32 index)
539 {
540 	int err;
541 
542 	err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
543 				  (index * dr->entry_size),
544 				  dr->cookies, dr->ncookies);
545 	if (err < 0)
546 		return err;
547 
548 	return 0;
549 }
550 
551 static int vnet_walk_rx_one(struct vnet_port *port,
552 			    struct vio_dring_state *dr,
553 			    u32 index, int *needs_ack)
554 {
555 	struct vio_net_desc *desc = get_rx_desc(port, dr, index);
556 	struct vio_driver_state *vio = &port->vio;
557 	int err;
558 
559 	BUG_ON(!desc);
560 	if (IS_ERR(desc))
561 		return PTR_ERR(desc);
562 
563 	if (desc->hdr.state != VIO_DESC_READY)
564 		return 1;
565 
566 	dma_rmb();
567 
568 	viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
569 	       desc->hdr.state, desc->hdr.ack,
570 	       desc->size, desc->ncookies,
571 	       desc->cookies[0].cookie_addr,
572 	       desc->cookies[0].cookie_size);
573 
574 	err = vnet_rx_one(port, desc);
575 	if (err == -ECONNRESET)
576 		return err;
577 	trace_vnet_rx_one(port->vio._local_sid, port->vio._peer_sid,
578 			  index, desc->hdr.ack);
579 	desc->hdr.state = VIO_DESC_DONE;
580 	err = put_rx_desc(port, dr, desc, index);
581 	if (err < 0)
582 		return err;
583 	*needs_ack = desc->hdr.ack;
584 	return 0;
585 }
586 
587 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
588 			u32 start, u32 end, int *npkts, int budget)
589 {
590 	struct vio_driver_state *vio = &port->vio;
591 	int ack_start = -1, ack_end = -1;
592 	bool send_ack = true;
593 
594 	end = (end == (u32)-1) ? vio_dring_prev(dr, start)
595 			       : vio_dring_next(dr, end);
596 
597 	viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
598 
599 	while (start != end) {
600 		int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
601 
602 		if (err == -ECONNRESET)
603 			return err;
604 		if (err != 0)
605 			break;
606 		(*npkts)++;
607 		if (ack_start == -1)
608 			ack_start = start;
609 		ack_end = start;
610 		start = vio_dring_next(dr, start);
611 		if (ack && start != end) {
612 			err = vnet_send_ack(port, dr, ack_start, ack_end,
613 					    VIO_DRING_ACTIVE);
614 			if (err == -ECONNRESET)
615 				return err;
616 			ack_start = -1;
617 		}
618 		if ((*npkts) >= budget) {
619 			send_ack = false;
620 			break;
621 		}
622 	}
623 	if (unlikely(ack_start == -1)) {
624 		ack_end = vio_dring_prev(dr, start);
625 		ack_start = ack_end;
626 	}
627 	if (send_ack) {
628 		port->napi_resume = false;
629 		trace_vnet_tx_send_stopped_ack(port->vio._local_sid,
630 					       port->vio._peer_sid,
631 					       ack_end, *npkts);
632 		return vnet_send_ack(port, dr, ack_start, ack_end,
633 				     VIO_DRING_STOPPED);
634 	} else  {
635 		trace_vnet_tx_defer_stopped_ack(port->vio._local_sid,
636 						port->vio._peer_sid,
637 						ack_end, *npkts);
638 		port->napi_resume = true;
639 		port->napi_stop_idx = ack_end;
640 		return 1;
641 	}
642 }
643 
644 static int vnet_rx(struct vnet_port *port, void *msgbuf, int *npkts,
645 		   int budget)
646 {
647 	struct vio_dring_data *pkt = msgbuf;
648 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
649 	struct vio_driver_state *vio = &port->vio;
650 
651 	viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
652 	       pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
653 
654 	if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
655 		return 0;
656 	if (unlikely(pkt->seq != dr->rcv_nxt)) {
657 		pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
658 		       pkt->seq, dr->rcv_nxt);
659 		return 0;
660 	}
661 
662 	if (!port->napi_resume)
663 		dr->rcv_nxt++;
664 
665 	/* XXX Validate pkt->start_idx and pkt->end_idx XXX */
666 
667 	return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx,
668 			    npkts, budget);
669 }
670 
671 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
672 {
673 	u32 idx = dr->cons;
674 	int found = 0;
675 
676 	while (idx != dr->prod) {
677 		if (idx == end) {
678 			found = 1;
679 			break;
680 		}
681 		idx = vio_dring_next(dr, idx);
682 	}
683 	return found;
684 }
685 
686 static int vnet_ack(struct vnet_port *port, void *msgbuf)
687 {
688 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
689 	struct vio_dring_data *pkt = msgbuf;
690 	struct net_device *dev;
691 	u32 end;
692 	struct vio_net_desc *desc;
693 	struct netdev_queue *txq;
694 
695 	if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
696 		return 0;
697 
698 	end = pkt->end_idx;
699 	dev = VNET_PORT_TO_NET_DEVICE(port);
700 	netif_tx_lock(dev);
701 	if (unlikely(!idx_is_pending(dr, end))) {
702 		netif_tx_unlock(dev);
703 		return 0;
704 	}
705 
706 	/* sync for race conditions with vnet_start_xmit() and tell xmit it
707 	 * is time to send a trigger.
708 	 */
709 	trace_vnet_rx_stopped_ack(port->vio._local_sid,
710 				  port->vio._peer_sid, end);
711 	dr->cons = vio_dring_next(dr, end);
712 	desc = vio_dring_entry(dr, dr->cons);
713 	if (desc->hdr.state == VIO_DESC_READY && !port->start_cons) {
714 		/* vnet_start_xmit() just populated this dring but missed
715 		 * sending the "start" LDC message to the consumer.
716 		 * Send a "start" trigger on its behalf.
717 		 */
718 		if (__vnet_tx_trigger(port, dr->cons) > 0)
719 			port->start_cons = false;
720 		else
721 			port->start_cons = true;
722 	} else {
723 		port->start_cons = true;
724 	}
725 	netif_tx_unlock(dev);
726 
727 	txq = netdev_get_tx_queue(dev, port->q_index);
728 	if (unlikely(netif_tx_queue_stopped(txq) &&
729 		     vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
730 		return 1;
731 
732 	return 0;
733 }
734 
735 static int vnet_nack(struct vnet_port *port, void *msgbuf)
736 {
737 	/* XXX just reset or similar XXX */
738 	return 0;
739 }
740 
741 static int handle_mcast(struct vnet_port *port, void *msgbuf)
742 {
743 	struct vio_net_mcast_info *pkt = msgbuf;
744 	struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
745 
746 	if (pkt->tag.stype != VIO_SUBTYPE_ACK)
747 		pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
748 		       dev->name,
749 		       pkt->tag.type,
750 		       pkt->tag.stype,
751 		       pkt->tag.stype_env,
752 		       pkt->tag.sid);
753 
754 	return 0;
755 }
756 
757 /* If the queue is stopped, wake it up so that we'll
758  * send out another START message at the next TX.
759  */
760 static void maybe_tx_wakeup(struct vnet_port *port)
761 {
762 	struct netdev_queue *txq;
763 
764 	txq = netdev_get_tx_queue(VNET_PORT_TO_NET_DEVICE(port),
765 				  port->q_index);
766 	__netif_tx_lock(txq, smp_processor_id());
767 	if (likely(netif_tx_queue_stopped(txq)))
768 		netif_tx_wake_queue(txq);
769 	__netif_tx_unlock(txq);
770 }
771 
772 bool sunvnet_port_is_up_common(struct vnet_port *vnet)
773 {
774 	struct vio_driver_state *vio = &vnet->vio;
775 
776 	return !!(vio->hs_state & VIO_HS_COMPLETE);
777 }
778 EXPORT_SYMBOL_GPL(sunvnet_port_is_up_common);
779 
780 static int vnet_event_napi(struct vnet_port *port, int budget)
781 {
782 	struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
783 	struct vio_driver_state *vio = &port->vio;
784 	int tx_wakeup, err;
785 	int npkts = 0;
786 
787 	/* we don't expect any other bits */
788 	BUG_ON(port->rx_event & ~(LDC_EVENT_DATA_READY |
789 				  LDC_EVENT_RESET |
790 				  LDC_EVENT_UP));
791 
792 	/* RESET takes precedent over any other event */
793 	if (port->rx_event & LDC_EVENT_RESET) {
794 		/* a link went down */
795 
796 		if (port->vsw == 1) {
797 			netif_tx_stop_all_queues(dev);
798 			netif_carrier_off(dev);
799 		}
800 
801 		vio_link_state_change(vio, LDC_EVENT_RESET);
802 		vnet_port_reset(port);
803 		vio_port_up(vio);
804 
805 		/* If the device is running but its tx queue was
806 		 * stopped (due to flow control), restart it.
807 		 * This is necessary since vnet_port_reset()
808 		 * clears the tx drings and thus we may never get
809 		 * back a VIO_TYPE_DATA ACK packet - which is
810 		 * the normal mechanism to restart the tx queue.
811 		 */
812 		if (netif_running(dev))
813 			maybe_tx_wakeup(port);
814 
815 		port->rx_event = 0;
816 		port->stats.event_reset++;
817 		return 0;
818 	}
819 
820 	if (port->rx_event & LDC_EVENT_UP) {
821 		/* a link came up */
822 
823 		if (port->vsw == 1) {
824 			netif_carrier_on(port->dev);
825 			netif_tx_start_all_queues(port->dev);
826 		}
827 
828 		vio_link_state_change(vio, LDC_EVENT_UP);
829 		port->rx_event = 0;
830 		port->stats.event_up++;
831 		return 0;
832 	}
833 
834 	err = 0;
835 	tx_wakeup = 0;
836 	while (1) {
837 		union {
838 			struct vio_msg_tag tag;
839 			u64 raw[8];
840 		} msgbuf;
841 
842 		if (port->napi_resume) {
843 			struct vio_dring_data *pkt =
844 				(struct vio_dring_data *)&msgbuf;
845 			struct vio_dring_state *dr =
846 				&port->vio.drings[VIO_DRIVER_RX_RING];
847 
848 			pkt->tag.type = VIO_TYPE_DATA;
849 			pkt->tag.stype = VIO_SUBTYPE_INFO;
850 			pkt->tag.stype_env = VIO_DRING_DATA;
851 			pkt->seq = dr->rcv_nxt;
852 			pkt->start_idx = vio_dring_next(dr,
853 							port->napi_stop_idx);
854 			pkt->end_idx = -1;
855 		} else {
856 			err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
857 			if (unlikely(err < 0)) {
858 				if (err == -ECONNRESET)
859 					vio_conn_reset(vio);
860 				break;
861 			}
862 			if (err == 0)
863 				break;
864 			viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
865 			       msgbuf.tag.type,
866 			       msgbuf.tag.stype,
867 			       msgbuf.tag.stype_env,
868 			       msgbuf.tag.sid);
869 			err = vio_validate_sid(vio, &msgbuf.tag);
870 			if (err < 0)
871 				break;
872 		}
873 
874 		if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
875 			if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
876 				if (!sunvnet_port_is_up_common(port)) {
877 					/* failures like handshake_failure()
878 					 * may have cleaned up dring, but
879 					 * NAPI polling may bring us here.
880 					 */
881 					err = -ECONNRESET;
882 					break;
883 				}
884 				err = vnet_rx(port, &msgbuf, &npkts, budget);
885 				if (npkts >= budget)
886 					break;
887 				if (npkts == 0)
888 					break;
889 			} else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
890 				err = vnet_ack(port, &msgbuf);
891 				if (err > 0)
892 					tx_wakeup |= err;
893 			} else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
894 				err = vnet_nack(port, &msgbuf);
895 			}
896 		} else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
897 			if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
898 				err = handle_mcast(port, &msgbuf);
899 			else
900 				err = vio_control_pkt_engine(vio, &msgbuf);
901 			if (err)
902 				break;
903 		} else {
904 			err = vnet_handle_unknown(port, &msgbuf);
905 		}
906 		if (err == -ECONNRESET)
907 			break;
908 	}
909 	if (unlikely(tx_wakeup && err != -ECONNRESET))
910 		maybe_tx_wakeup(port);
911 	return npkts;
912 }
913 
914 int sunvnet_poll_common(struct napi_struct *napi, int budget)
915 {
916 	struct vnet_port *port = container_of(napi, struct vnet_port, napi);
917 	struct vio_driver_state *vio = &port->vio;
918 	int processed = vnet_event_napi(port, budget);
919 
920 	if (processed < budget) {
921 		napi_complete_done(napi, processed);
922 		port->rx_event &= ~LDC_EVENT_DATA_READY;
923 		vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
924 	}
925 	return processed;
926 }
927 EXPORT_SYMBOL_GPL(sunvnet_poll_common);
928 
929 void sunvnet_event_common(void *arg, int event)
930 {
931 	struct vnet_port *port = arg;
932 	struct vio_driver_state *vio = &port->vio;
933 
934 	port->rx_event |= event;
935 	vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
936 	napi_schedule(&port->napi);
937 }
938 EXPORT_SYMBOL_GPL(sunvnet_event_common);
939 
940 static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
941 {
942 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
943 	struct vio_dring_data hdr = {
944 		.tag = {
945 			.type		= VIO_TYPE_DATA,
946 			.stype		= VIO_SUBTYPE_INFO,
947 			.stype_env	= VIO_DRING_DATA,
948 			.sid		= vio_send_sid(&port->vio),
949 		},
950 		.dring_ident		= dr->ident,
951 		.start_idx		= start,
952 		.end_idx		= (u32)-1,
953 	};
954 	int err, delay;
955 	int retries = 0;
956 
957 	if (port->stop_rx) {
958 		trace_vnet_tx_pending_stopped_ack(port->vio._local_sid,
959 						  port->vio._peer_sid,
960 						  port->stop_rx_idx, -1);
961 		err = vnet_send_ack(port,
962 				    &port->vio.drings[VIO_DRIVER_RX_RING],
963 				    port->stop_rx_idx, -1,
964 				    VIO_DRING_STOPPED);
965 		if (err <= 0)
966 			return err;
967 	}
968 
969 	hdr.seq = dr->snd_nxt;
970 	delay = 1;
971 	do {
972 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
973 		if (err > 0) {
974 			dr->snd_nxt++;
975 			break;
976 		}
977 		udelay(delay);
978 		if ((delay <<= 1) > 128)
979 			delay = 128;
980 		if (retries++ > VNET_MAX_RETRIES)
981 			break;
982 	} while (err == -EAGAIN);
983 	trace_vnet_tx_trigger(port->vio._local_sid,
984 			      port->vio._peer_sid, start, err);
985 
986 	return err;
987 }
988 
989 static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
990 					  unsigned *pending)
991 {
992 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
993 	struct sk_buff *skb = NULL;
994 	int i, txi;
995 
996 	*pending = 0;
997 
998 	txi = dr->prod;
999 	for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1000 		struct vio_net_desc *d;
1001 
1002 		--txi;
1003 		if (txi < 0)
1004 			txi = VNET_TX_RING_SIZE - 1;
1005 
1006 		d = vio_dring_entry(dr, txi);
1007 
1008 		if (d->hdr.state == VIO_DESC_READY) {
1009 			(*pending)++;
1010 			continue;
1011 		}
1012 		if (port->tx_bufs[txi].skb) {
1013 			if (d->hdr.state != VIO_DESC_DONE)
1014 				pr_notice("invalid ring buffer state %d\n",
1015 					  d->hdr.state);
1016 			BUG_ON(port->tx_bufs[txi].skb->next);
1017 
1018 			port->tx_bufs[txi].skb->next = skb;
1019 			skb = port->tx_bufs[txi].skb;
1020 			port->tx_bufs[txi].skb = NULL;
1021 
1022 			ldc_unmap(port->vio.lp,
1023 				  port->tx_bufs[txi].cookies,
1024 				  port->tx_bufs[txi].ncookies);
1025 		} else if (d->hdr.state == VIO_DESC_FREE) {
1026 			break;
1027 		}
1028 		d->hdr.state = VIO_DESC_FREE;
1029 	}
1030 	return skb;
1031 }
1032 
1033 static inline void vnet_free_skbs(struct sk_buff *skb)
1034 {
1035 	struct sk_buff *next;
1036 
1037 	while (skb) {
1038 		next = skb->next;
1039 		skb->next = NULL;
1040 		dev_kfree_skb(skb);
1041 		skb = next;
1042 	}
1043 }
1044 
1045 void sunvnet_clean_timer_expire_common(struct timer_list *t)
1046 {
1047 	struct vnet_port *port = from_timer(port, t, clean_timer);
1048 	struct sk_buff *freeskbs;
1049 	unsigned pending;
1050 
1051 	netif_tx_lock(VNET_PORT_TO_NET_DEVICE(port));
1052 	freeskbs = vnet_clean_tx_ring(port, &pending);
1053 	netif_tx_unlock(VNET_PORT_TO_NET_DEVICE(port));
1054 
1055 	vnet_free_skbs(freeskbs);
1056 
1057 	if (pending)
1058 		(void)mod_timer(&port->clean_timer,
1059 				jiffies + VNET_CLEAN_TIMEOUT);
1060 	 else
1061 		del_timer(&port->clean_timer);
1062 }
1063 EXPORT_SYMBOL_GPL(sunvnet_clean_timer_expire_common);
1064 
1065 static inline int vnet_skb_map(struct ldc_channel *lp, struct sk_buff *skb,
1066 			       struct ldc_trans_cookie *cookies, int ncookies,
1067 			       unsigned int map_perm)
1068 {
1069 	int i, nc, err, blen;
1070 
1071 	/* header */
1072 	blen = skb_headlen(skb);
1073 	if (blen < ETH_ZLEN)
1074 		blen = ETH_ZLEN;
1075 	blen += VNET_PACKET_SKIP;
1076 	blen += 8 - (blen & 7);
1077 
1078 	err = ldc_map_single(lp, skb->data - VNET_PACKET_SKIP, blen, cookies,
1079 			     ncookies, map_perm);
1080 	if (err < 0)
1081 		return err;
1082 	nc = err;
1083 
1084 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1085 		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1086 		u8 *vaddr;
1087 
1088 		if (nc < ncookies) {
1089 			vaddr = kmap_local_page(skb_frag_page(f));
1090 			blen = skb_frag_size(f);
1091 			blen += 8 - (blen & 7);
1092 			err = ldc_map_single(lp, vaddr + skb_frag_off(f),
1093 					     blen, cookies + nc, ncookies - nc,
1094 					     map_perm);
1095 			kunmap_local(vaddr);
1096 		} else {
1097 			err = -EMSGSIZE;
1098 		}
1099 
1100 		if (err < 0) {
1101 			ldc_unmap(lp, cookies, nc);
1102 			return err;
1103 		}
1104 		nc += err;
1105 	}
1106 	return nc;
1107 }
1108 
1109 static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, int ncookies)
1110 {
1111 	struct sk_buff *nskb;
1112 	int i, len, pad, docopy;
1113 
1114 	len = skb->len;
1115 	pad = 0;
1116 	if (len < ETH_ZLEN) {
1117 		pad += ETH_ZLEN - skb->len;
1118 		len += pad;
1119 	}
1120 	len += VNET_PACKET_SKIP;
1121 	pad += 8 - (len & 7);
1122 
1123 	/* make sure we have enough cookies and alignment in every frag */
1124 	docopy = skb_shinfo(skb)->nr_frags >= ncookies;
1125 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1126 		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1127 
1128 		docopy |= skb_frag_off(f) & 7;
1129 	}
1130 	if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
1131 	    skb_tailroom(skb) < pad ||
1132 	    skb_headroom(skb) < VNET_PACKET_SKIP || docopy) {
1133 		int start = 0, offset;
1134 		__wsum csum;
1135 
1136 		len = skb->len > ETH_ZLEN ? skb->len : ETH_ZLEN;
1137 		nskb = alloc_and_align_skb(skb->dev, len);
1138 		if (!nskb) {
1139 			dev_kfree_skb(skb);
1140 			return NULL;
1141 		}
1142 		skb_reserve(nskb, VNET_PACKET_SKIP);
1143 
1144 		nskb->protocol = skb->protocol;
1145 		offset = skb_mac_header(skb) - skb->data;
1146 		skb_set_mac_header(nskb, offset);
1147 		offset = skb_network_offset(skb);
1148 		skb_set_network_header(nskb, offset);
1149 		offset = skb_transport_offset(skb);
1150 		skb_set_transport_header(nskb, offset);
1151 
1152 		offset = 0;
1153 		nskb->csum_offset = skb->csum_offset;
1154 		nskb->ip_summed = skb->ip_summed;
1155 
1156 		if (skb->ip_summed == CHECKSUM_PARTIAL)
1157 			start = skb_checksum_start_offset(skb);
1158 		if (start) {
1159 			int offset = start + nskb->csum_offset;
1160 
1161 			/* copy the headers, no csum here */
1162 			if (skb_copy_bits(skb, 0, nskb->data, start)) {
1163 				dev_kfree_skb(nskb);
1164 				dev_kfree_skb(skb);
1165 				return NULL;
1166 			}
1167 
1168 			/* copy the rest, with csum calculation */
1169 			*(__sum16 *)(skb->data + offset) = 0;
1170 			csum = skb_copy_and_csum_bits(skb, start,
1171 						      nskb->data + start,
1172 						      skb->len - start);
1173 
1174 			/* add in the header checksums */
1175 			if (skb->protocol == htons(ETH_P_IP)) {
1176 				struct iphdr *iph = ip_hdr(nskb);
1177 
1178 				if (iph->protocol == IPPROTO_TCP ||
1179 				    iph->protocol == IPPROTO_UDP) {
1180 					csum = csum_tcpudp_magic(iph->saddr,
1181 								 iph->daddr,
1182 								 skb->len - start,
1183 								 iph->protocol,
1184 								 csum);
1185 				}
1186 			} else if (skb->protocol == htons(ETH_P_IPV6)) {
1187 				struct ipv6hdr *ip6h = ipv6_hdr(nskb);
1188 
1189 				if (ip6h->nexthdr == IPPROTO_TCP ||
1190 				    ip6h->nexthdr == IPPROTO_UDP) {
1191 					csum = csum_ipv6_magic(&ip6h->saddr,
1192 							       &ip6h->daddr,
1193 							       skb->len - start,
1194 							       ip6h->nexthdr,
1195 							       csum);
1196 				}
1197 			}
1198 
1199 			/* save the final result */
1200 			*(__sum16 *)(nskb->data + offset) = csum;
1201 
1202 			nskb->ip_summed = CHECKSUM_NONE;
1203 		} else if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
1204 			dev_kfree_skb(nskb);
1205 			dev_kfree_skb(skb);
1206 			return NULL;
1207 		}
1208 		(void)skb_put(nskb, skb->len);
1209 		if (skb_is_gso(skb)) {
1210 			skb_shinfo(nskb)->gso_size = skb_shinfo(skb)->gso_size;
1211 			skb_shinfo(nskb)->gso_type = skb_shinfo(skb)->gso_type;
1212 		}
1213 		nskb->queue_mapping = skb->queue_mapping;
1214 		dev_kfree_skb(skb);
1215 		skb = nskb;
1216 	}
1217 	return skb;
1218 }
1219 
1220 static netdev_tx_t
1221 vnet_handle_offloads(struct vnet_port *port, struct sk_buff *skb,
1222 		     struct vnet_port *(*vnet_tx_port)
1223 		     (struct sk_buff *, struct net_device *))
1224 {
1225 	struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
1226 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1227 	struct sk_buff *segs, *curr, *next;
1228 	int maclen, datalen;
1229 	int status;
1230 	int gso_size, gso_type, gso_segs;
1231 	int hlen = skb_transport_header(skb) - skb_mac_header(skb);
1232 	int proto = IPPROTO_IP;
1233 
1234 	if (skb->protocol == htons(ETH_P_IP))
1235 		proto = ip_hdr(skb)->protocol;
1236 	else if (skb->protocol == htons(ETH_P_IPV6))
1237 		proto = ipv6_hdr(skb)->nexthdr;
1238 
1239 	if (proto == IPPROTO_TCP) {
1240 		hlen += tcp_hdr(skb)->doff * 4;
1241 	} else if (proto == IPPROTO_UDP) {
1242 		hlen += sizeof(struct udphdr);
1243 	} else {
1244 		pr_err("vnet_handle_offloads GSO with unknown transport "
1245 		       "protocol %d tproto %d\n", skb->protocol, proto);
1246 		hlen = 128; /* XXX */
1247 	}
1248 	datalen = port->tsolen - hlen;
1249 
1250 	gso_size = skb_shinfo(skb)->gso_size;
1251 	gso_type = skb_shinfo(skb)->gso_type;
1252 	gso_segs = skb_shinfo(skb)->gso_segs;
1253 
1254 	if (port->tso && gso_size < datalen)
1255 		gso_segs = DIV_ROUND_UP(skb->len - hlen, datalen);
1256 
1257 	if (unlikely(vnet_tx_dring_avail(dr) < gso_segs)) {
1258 		struct netdev_queue *txq;
1259 
1260 		txq  = netdev_get_tx_queue(dev, port->q_index);
1261 		netif_tx_stop_queue(txq);
1262 		if (vnet_tx_dring_avail(dr) < skb_shinfo(skb)->gso_segs)
1263 			return NETDEV_TX_BUSY;
1264 		netif_tx_wake_queue(txq);
1265 	}
1266 
1267 	maclen = skb_network_header(skb) - skb_mac_header(skb);
1268 	skb_pull(skb, maclen);
1269 
1270 	if (port->tso && gso_size < datalen) {
1271 		if (skb_unclone(skb, GFP_ATOMIC))
1272 			goto out_dropped;
1273 
1274 		/* segment to TSO size */
1275 		skb_shinfo(skb)->gso_size = datalen;
1276 		skb_shinfo(skb)->gso_segs = gso_segs;
1277 	}
1278 	segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO);
1279 	if (IS_ERR(segs))
1280 		goto out_dropped;
1281 
1282 	skb_push(skb, maclen);
1283 	skb_reset_mac_header(skb);
1284 
1285 	status = 0;
1286 	skb_list_walk_safe(segs, curr, next) {
1287 		skb_mark_not_on_list(curr);
1288 		if (port->tso && curr->len > dev->mtu) {
1289 			skb_shinfo(curr)->gso_size = gso_size;
1290 			skb_shinfo(curr)->gso_type = gso_type;
1291 			skb_shinfo(curr)->gso_segs =
1292 				DIV_ROUND_UP(curr->len - hlen, gso_size);
1293 		} else {
1294 			skb_shinfo(curr)->gso_size = 0;
1295 		}
1296 
1297 		skb_push(curr, maclen);
1298 		skb_reset_mac_header(curr);
1299 		memcpy(skb_mac_header(curr), skb_mac_header(skb),
1300 		       maclen);
1301 		curr->csum_start = skb_transport_header(curr) - curr->head;
1302 		if (ip_hdr(curr)->protocol == IPPROTO_TCP)
1303 			curr->csum_offset = offsetof(struct tcphdr, check);
1304 		else if (ip_hdr(curr)->protocol == IPPROTO_UDP)
1305 			curr->csum_offset = offsetof(struct udphdr, check);
1306 
1307 		if (!(status & NETDEV_TX_MASK))
1308 			status = sunvnet_start_xmit_common(curr, dev,
1309 							   vnet_tx_port);
1310 		if (status & NETDEV_TX_MASK)
1311 			dev_kfree_skb_any(curr);
1312 	}
1313 
1314 	if (!(status & NETDEV_TX_MASK))
1315 		dev_kfree_skb_any(skb);
1316 	return status;
1317 out_dropped:
1318 	dev->stats.tx_dropped++;
1319 	dev_kfree_skb_any(skb);
1320 	return NETDEV_TX_OK;
1321 }
1322 
1323 netdev_tx_t
1324 sunvnet_start_xmit_common(struct sk_buff *skb, struct net_device *dev,
1325 			  struct vnet_port *(*vnet_tx_port)
1326 			  (struct sk_buff *, struct net_device *))
1327 {
1328 	struct vnet_port *port = NULL;
1329 	struct vio_dring_state *dr;
1330 	struct vio_net_desc *d;
1331 	unsigned int len;
1332 	struct sk_buff *freeskbs = NULL;
1333 	int i, err, txi;
1334 	unsigned pending = 0;
1335 	struct netdev_queue *txq;
1336 
1337 	rcu_read_lock();
1338 	port = vnet_tx_port(skb, dev);
1339 	if (unlikely(!port))
1340 		goto out_dropped;
1341 
1342 	if (skb_is_gso(skb) && skb->len > port->tsolen) {
1343 		err = vnet_handle_offloads(port, skb, vnet_tx_port);
1344 		rcu_read_unlock();
1345 		return err;
1346 	}
1347 
1348 	if (!skb_is_gso(skb) && skb->len > port->rmtu) {
1349 		unsigned long localmtu = port->rmtu - ETH_HLEN;
1350 
1351 		if (vio_version_after_eq(&port->vio, 1, 3))
1352 			localmtu -= VLAN_HLEN;
1353 
1354 		if (skb->protocol == htons(ETH_P_IP))
1355 			icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
1356 				      htonl(localmtu));
1357 #if IS_ENABLED(CONFIG_IPV6)
1358 		else if (skb->protocol == htons(ETH_P_IPV6))
1359 			icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1360 #endif
1361 		goto out_dropped;
1362 	}
1363 
1364 	skb = vnet_skb_shape(skb, 2);
1365 
1366 	if (unlikely(!skb))
1367 		goto out_dropped;
1368 
1369 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1370 		if (skb->protocol == htons(ETH_P_IP))
1371 			vnet_fullcsum_ipv4(skb);
1372 #if IS_ENABLED(CONFIG_IPV6)
1373 		else if (skb->protocol == htons(ETH_P_IPV6))
1374 			vnet_fullcsum_ipv6(skb);
1375 #endif
1376 	}
1377 
1378 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1379 	i = skb_get_queue_mapping(skb);
1380 	txq = netdev_get_tx_queue(dev, i);
1381 	if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1382 		if (!netif_tx_queue_stopped(txq)) {
1383 			netif_tx_stop_queue(txq);
1384 
1385 			/* This is a hard error, log it. */
1386 			netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
1387 			dev->stats.tx_errors++;
1388 		}
1389 		rcu_read_unlock();
1390 		return NETDEV_TX_BUSY;
1391 	}
1392 
1393 	d = vio_dring_cur(dr);
1394 
1395 	txi = dr->prod;
1396 
1397 	freeskbs = vnet_clean_tx_ring(port, &pending);
1398 
1399 	BUG_ON(port->tx_bufs[txi].skb);
1400 
1401 	len = skb->len;
1402 	if (len < ETH_ZLEN)
1403 		len = ETH_ZLEN;
1404 
1405 	err = vnet_skb_map(port->vio.lp, skb, port->tx_bufs[txi].cookies, 2,
1406 			   (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1407 	if (err < 0) {
1408 		netdev_info(dev, "tx buffer map error %d\n", err);
1409 		goto out_dropped;
1410 	}
1411 
1412 	port->tx_bufs[txi].skb = skb;
1413 	skb = NULL;
1414 	port->tx_bufs[txi].ncookies = err;
1415 
1416 	/* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1417 	 * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1418 	 * the protocol itself does not require it as long as the peer
1419 	 * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1420 	 *
1421 	 * An ACK for every packet in the ring is expensive as the
1422 	 * sending of LDC messages is slow and affects performance.
1423 	 */
1424 	d->hdr.ack = VIO_ACK_DISABLE;
1425 	d->size = len;
1426 	d->ncookies = port->tx_bufs[txi].ncookies;
1427 	for (i = 0; i < d->ncookies; i++)
1428 		d->cookies[i] = port->tx_bufs[txi].cookies[i];
1429 	if (vio_version_after_eq(&port->vio, 1, 7)) {
1430 		struct vio_net_dext *dext = vio_net_ext(d);
1431 
1432 		memset(dext, 0, sizeof(*dext));
1433 		if (skb_is_gso(port->tx_bufs[txi].skb)) {
1434 			dext->ipv4_lso_mss = skb_shinfo(port->tx_bufs[txi].skb)
1435 					     ->gso_size;
1436 			dext->flags |= VNET_PKT_IPV4_LSO;
1437 		}
1438 		if (vio_version_after_eq(&port->vio, 1, 8) &&
1439 		    !port->switch_port) {
1440 			dext->flags |= VNET_PKT_HCK_IPV4_HDRCKSUM_OK;
1441 			dext->flags |= VNET_PKT_HCK_FULLCKSUM_OK;
1442 		}
1443 	}
1444 
1445 	/* This has to be a non-SMP write barrier because we are writing
1446 	 * to memory which is shared with the peer LDOM.
1447 	 */
1448 	dma_wmb();
1449 
1450 	d->hdr.state = VIO_DESC_READY;
1451 
1452 	/* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1453 	 * to notify the consumer that some descriptors are READY.
1454 	 * After that "start" trigger, no additional triggers are needed until
1455 	 * a DRING_STOPPED is received from the consumer. The dr->cons field
1456 	 * (set up by vnet_ack()) has the value of the next dring index
1457 	 * that has not yet been ack-ed. We send a "start" trigger here
1458 	 * if, and only if, start_cons is true (reset it afterward). Conversely,
1459 	 * vnet_ack() should check if the dring corresponding to cons
1460 	 * is marked READY, but start_cons was false.
1461 	 * If so, vnet_ack() should send out the missed "start" trigger.
1462 	 *
1463 	 * Note that the dma_wmb() above makes sure the cookies et al. are
1464 	 * not globally visible before the VIO_DESC_READY, and that the
1465 	 * stores are ordered correctly by the compiler. The consumer will
1466 	 * not proceed until the VIO_DESC_READY is visible assuring that
1467 	 * the consumer does not observe anything related to descriptors
1468 	 * out of order. The HV trap from the LDC start trigger is the
1469 	 * producer to consumer announcement that work is available to the
1470 	 * consumer
1471 	 */
1472 	if (!port->start_cons) { /* previous trigger suffices */
1473 		trace_vnet_skip_tx_trigger(port->vio._local_sid,
1474 					   port->vio._peer_sid, dr->cons);
1475 		goto ldc_start_done;
1476 	}
1477 
1478 	err = __vnet_tx_trigger(port, dr->cons);
1479 	if (unlikely(err < 0)) {
1480 		netdev_info(dev, "TX trigger error %d\n", err);
1481 		d->hdr.state = VIO_DESC_FREE;
1482 		skb = port->tx_bufs[txi].skb;
1483 		port->tx_bufs[txi].skb = NULL;
1484 		dev->stats.tx_carrier_errors++;
1485 		goto out_dropped;
1486 	}
1487 
1488 ldc_start_done:
1489 	port->start_cons = false;
1490 
1491 	dev->stats.tx_packets++;
1492 	dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1493 	port->stats.tx_packets++;
1494 	port->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1495 
1496 	dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1497 	if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1498 		netif_tx_stop_queue(txq);
1499 		smp_rmb();
1500 		if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1501 			netif_tx_wake_queue(txq);
1502 	}
1503 
1504 	(void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1505 	rcu_read_unlock();
1506 
1507 	vnet_free_skbs(freeskbs);
1508 
1509 	return NETDEV_TX_OK;
1510 
1511 out_dropped:
1512 	if (pending)
1513 		(void)mod_timer(&port->clean_timer,
1514 				jiffies + VNET_CLEAN_TIMEOUT);
1515 	else if (port)
1516 		del_timer(&port->clean_timer);
1517 	rcu_read_unlock();
1518 	dev_kfree_skb(skb);
1519 	vnet_free_skbs(freeskbs);
1520 	dev->stats.tx_dropped++;
1521 	return NETDEV_TX_OK;
1522 }
1523 EXPORT_SYMBOL_GPL(sunvnet_start_xmit_common);
1524 
1525 void sunvnet_tx_timeout_common(struct net_device *dev, unsigned int txqueue)
1526 {
1527 	/* XXX Implement me XXX */
1528 }
1529 EXPORT_SYMBOL_GPL(sunvnet_tx_timeout_common);
1530 
1531 int sunvnet_open_common(struct net_device *dev)
1532 {
1533 	netif_carrier_on(dev);
1534 	netif_tx_start_all_queues(dev);
1535 
1536 	return 0;
1537 }
1538 EXPORT_SYMBOL_GPL(sunvnet_open_common);
1539 
1540 int sunvnet_close_common(struct net_device *dev)
1541 {
1542 	netif_tx_stop_all_queues(dev);
1543 	netif_carrier_off(dev);
1544 
1545 	return 0;
1546 }
1547 EXPORT_SYMBOL_GPL(sunvnet_close_common);
1548 
1549 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1550 {
1551 	struct vnet_mcast_entry *m;
1552 
1553 	for (m = vp->mcast_list; m; m = m->next) {
1554 		if (ether_addr_equal(m->addr, addr))
1555 			return m;
1556 	}
1557 	return NULL;
1558 }
1559 
1560 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1561 {
1562 	struct netdev_hw_addr *ha;
1563 
1564 	netdev_for_each_mc_addr(ha, dev) {
1565 		struct vnet_mcast_entry *m;
1566 
1567 		m = __vnet_mc_find(vp, ha->addr);
1568 		if (m) {
1569 			m->hit = 1;
1570 			continue;
1571 		}
1572 
1573 		if (!m) {
1574 			m = kzalloc(sizeof(*m), GFP_ATOMIC);
1575 			if (!m)
1576 				continue;
1577 			memcpy(m->addr, ha->addr, ETH_ALEN);
1578 			m->hit = 1;
1579 
1580 			m->next = vp->mcast_list;
1581 			vp->mcast_list = m;
1582 		}
1583 	}
1584 }
1585 
1586 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1587 {
1588 	struct vio_net_mcast_info info;
1589 	struct vnet_mcast_entry *m, **pp;
1590 	int n_addrs;
1591 
1592 	memset(&info, 0, sizeof(info));
1593 
1594 	info.tag.type = VIO_TYPE_CTRL;
1595 	info.tag.stype = VIO_SUBTYPE_INFO;
1596 	info.tag.stype_env = VNET_MCAST_INFO;
1597 	info.tag.sid = vio_send_sid(&port->vio);
1598 	info.set = 1;
1599 
1600 	n_addrs = 0;
1601 	for (m = vp->mcast_list; m; m = m->next) {
1602 		if (m->sent)
1603 			continue;
1604 		m->sent = 1;
1605 		memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1606 		       m->addr, ETH_ALEN);
1607 		if (++n_addrs == VNET_NUM_MCAST) {
1608 			info.count = n_addrs;
1609 
1610 			(void)vio_ldc_send(&port->vio, &info,
1611 					   sizeof(info));
1612 			n_addrs = 0;
1613 		}
1614 	}
1615 	if (n_addrs) {
1616 		info.count = n_addrs;
1617 		(void)vio_ldc_send(&port->vio, &info, sizeof(info));
1618 	}
1619 
1620 	info.set = 0;
1621 
1622 	n_addrs = 0;
1623 	pp = &vp->mcast_list;
1624 	while ((m = *pp) != NULL) {
1625 		if (m->hit) {
1626 			m->hit = 0;
1627 			pp = &m->next;
1628 			continue;
1629 		}
1630 
1631 		memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1632 		       m->addr, ETH_ALEN);
1633 		if (++n_addrs == VNET_NUM_MCAST) {
1634 			info.count = n_addrs;
1635 			(void)vio_ldc_send(&port->vio, &info,
1636 					   sizeof(info));
1637 			n_addrs = 0;
1638 		}
1639 
1640 		*pp = m->next;
1641 		kfree(m);
1642 	}
1643 	if (n_addrs) {
1644 		info.count = n_addrs;
1645 		(void)vio_ldc_send(&port->vio, &info, sizeof(info));
1646 	}
1647 }
1648 
1649 void sunvnet_set_rx_mode_common(struct net_device *dev, struct vnet *vp)
1650 {
1651 	struct vnet_port *port;
1652 
1653 	rcu_read_lock();
1654 	list_for_each_entry_rcu(port, &vp->port_list, list) {
1655 		if (port->switch_port) {
1656 			__update_mc_list(vp, dev);
1657 			__send_mc_list(vp, port);
1658 			break;
1659 		}
1660 	}
1661 	rcu_read_unlock();
1662 }
1663 EXPORT_SYMBOL_GPL(sunvnet_set_rx_mode_common);
1664 
1665 int sunvnet_set_mac_addr_common(struct net_device *dev, void *p)
1666 {
1667 	return -EINVAL;
1668 }
1669 EXPORT_SYMBOL_GPL(sunvnet_set_mac_addr_common);
1670 
1671 void sunvnet_port_free_tx_bufs_common(struct vnet_port *port)
1672 {
1673 	struct vio_dring_state *dr;
1674 	int i;
1675 
1676 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1677 
1678 	if (!dr->base)
1679 		return;
1680 
1681 	for (i = 0; i < VNET_TX_RING_SIZE; i++) {
1682 		struct vio_net_desc *d;
1683 		void *skb = port->tx_bufs[i].skb;
1684 
1685 		if (!skb)
1686 			continue;
1687 
1688 		d = vio_dring_entry(dr, i);
1689 
1690 		ldc_unmap(port->vio.lp,
1691 			  port->tx_bufs[i].cookies,
1692 			  port->tx_bufs[i].ncookies);
1693 		dev_kfree_skb(skb);
1694 		port->tx_bufs[i].skb = NULL;
1695 		d->hdr.state = VIO_DESC_FREE;
1696 	}
1697 	ldc_free_exp_dring(port->vio.lp, dr->base,
1698 			   (dr->entry_size * dr->num_entries),
1699 			   dr->cookies, dr->ncookies);
1700 	dr->base = NULL;
1701 	dr->entry_size = 0;
1702 	dr->num_entries = 0;
1703 	dr->pending = 0;
1704 	dr->ncookies = 0;
1705 }
1706 EXPORT_SYMBOL_GPL(sunvnet_port_free_tx_bufs_common);
1707 
1708 void vnet_port_reset(struct vnet_port *port)
1709 {
1710 	del_timer(&port->clean_timer);
1711 	sunvnet_port_free_tx_bufs_common(port);
1712 	port->rmtu = 0;
1713 	port->tso = (port->vsw == 0);  /* no tso in vsw, misbehaves in bridge */
1714 	port->tsolen = 0;
1715 }
1716 EXPORT_SYMBOL_GPL(vnet_port_reset);
1717 
1718 static int vnet_port_alloc_tx_ring(struct vnet_port *port)
1719 {
1720 	struct vio_dring_state *dr;
1721 	unsigned long len, elen;
1722 	int i, err, ncookies;
1723 	void *dring;
1724 
1725 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1726 
1727 	elen = sizeof(struct vio_net_desc) +
1728 	       sizeof(struct ldc_trans_cookie) * 2;
1729 	if (vio_version_after_eq(&port->vio, 1, 7))
1730 		elen += sizeof(struct vio_net_dext);
1731 	len = VNET_TX_RING_SIZE * elen;
1732 
1733 	ncookies = VIO_MAX_RING_COOKIES;
1734 	dring = ldc_alloc_exp_dring(port->vio.lp, len,
1735 				    dr->cookies, &ncookies,
1736 				    (LDC_MAP_SHADOW |
1737 				     LDC_MAP_DIRECT |
1738 				     LDC_MAP_RW));
1739 	if (IS_ERR(dring)) {
1740 		err = PTR_ERR(dring);
1741 		goto err_out;
1742 	}
1743 
1744 	dr->base = dring;
1745 	dr->entry_size = elen;
1746 	dr->num_entries = VNET_TX_RING_SIZE;
1747 	dr->prod = 0;
1748 	dr->cons = 0;
1749 	port->start_cons  = true; /* need an initial trigger */
1750 	dr->pending = VNET_TX_RING_SIZE;
1751 	dr->ncookies = ncookies;
1752 
1753 	for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1754 		struct vio_net_desc *d;
1755 
1756 		d = vio_dring_entry(dr, i);
1757 		d->hdr.state = VIO_DESC_FREE;
1758 	}
1759 	return 0;
1760 
1761 err_out:
1762 	sunvnet_port_free_tx_bufs_common(port);
1763 
1764 	return err;
1765 }
1766 
1767 #ifdef CONFIG_NET_POLL_CONTROLLER
1768 void sunvnet_poll_controller_common(struct net_device *dev, struct vnet *vp)
1769 {
1770 	struct vnet_port *port;
1771 	unsigned long flags;
1772 
1773 	spin_lock_irqsave(&vp->lock, flags);
1774 	if (!list_empty(&vp->port_list)) {
1775 		port = list_entry(vp->port_list.next, struct vnet_port, list);
1776 		napi_schedule(&port->napi);
1777 	}
1778 	spin_unlock_irqrestore(&vp->lock, flags);
1779 }
1780 EXPORT_SYMBOL_GPL(sunvnet_poll_controller_common);
1781 #endif
1782 
1783 void sunvnet_port_add_txq_common(struct vnet_port *port)
1784 {
1785 	struct vnet *vp = port->vp;
1786 	int smallest = 0;
1787 	int i;
1788 
1789 	/* find the first least-used q
1790 	 * When there are more ldoms than q's, we start to
1791 	 * double up on ports per queue.
1792 	 */
1793 	for (i = 0; i < VNET_MAX_TXQS; i++) {
1794 		if (vp->q_used[i] == 0) {
1795 			smallest = i;
1796 			break;
1797 		}
1798 		if (vp->q_used[i] < vp->q_used[smallest])
1799 			smallest = i;
1800 	}
1801 
1802 	vp->nports++;
1803 	vp->q_used[smallest]++;
1804 	port->q_index = smallest;
1805 }
1806 EXPORT_SYMBOL_GPL(sunvnet_port_add_txq_common);
1807 
1808 void sunvnet_port_rm_txq_common(struct vnet_port *port)
1809 {
1810 	port->vp->nports--;
1811 	port->vp->q_used[port->q_index]--;
1812 	port->q_index = 0;
1813 }
1814 EXPORT_SYMBOL_GPL(sunvnet_port_rm_txq_common);
1815