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