xref: /linux/drivers/infiniband/hw/cxgb4/cm.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
1 /*
2  * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *	  copyright notice, this list of conditions and the following
16  *	  disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *	  copyright notice, this list of conditions and the following
20  *	  disclaimer in the documentation and/or other materials
21  *	  provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
42 
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
49 
50 #include <rdma/ib_addr.h>
51 
52 #include "iw_cxgb4.h"
53 #include "clip_tbl.h"
54 
55 static char *states[] = {
56 	"idle",
57 	"listen",
58 	"connecting",
59 	"mpa_wait_req",
60 	"mpa_req_sent",
61 	"mpa_req_rcvd",
62 	"mpa_rep_sent",
63 	"fpdu_mode",
64 	"aborting",
65 	"closing",
66 	"moribund",
67 	"dead",
68 	NULL,
69 };
70 
71 static int nocong;
72 module_param(nocong, int, 0644);
73 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
74 
75 static int enable_ecn;
76 module_param(enable_ecn, int, 0644);
77 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
78 
79 static int dack_mode = 1;
80 module_param(dack_mode, int, 0644);
81 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
82 
83 uint c4iw_max_read_depth = 32;
84 module_param(c4iw_max_read_depth, int, 0644);
85 MODULE_PARM_DESC(c4iw_max_read_depth,
86 		 "Per-connection max ORD/IRD (default=32)");
87 
88 static int enable_tcp_timestamps;
89 module_param(enable_tcp_timestamps, int, 0644);
90 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
91 
92 static int enable_tcp_sack;
93 module_param(enable_tcp_sack, int, 0644);
94 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
95 
96 static int enable_tcp_window_scaling = 1;
97 module_param(enable_tcp_window_scaling, int, 0644);
98 MODULE_PARM_DESC(enable_tcp_window_scaling,
99 		 "Enable tcp window scaling (default=1)");
100 
101 int c4iw_debug;
102 module_param(c4iw_debug, int, 0644);
103 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
104 
105 static int peer2peer = 1;
106 module_param(peer2peer, int, 0644);
107 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
108 
109 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
110 module_param(p2p_type, int, 0644);
111 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
112 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
113 
114 static int ep_timeout_secs = 60;
115 module_param(ep_timeout_secs, int, 0644);
116 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
117 				   "in seconds (default=60)");
118 
119 static int mpa_rev = 2;
120 module_param(mpa_rev, int, 0644);
121 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
122 		"1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
123 		" compliant (default=2)");
124 
125 static int markers_enabled;
126 module_param(markers_enabled, int, 0644);
127 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
128 
129 static int crc_enabled = 1;
130 module_param(crc_enabled, int, 0644);
131 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
132 
133 static int rcv_win = 256 * 1024;
134 module_param(rcv_win, int, 0644);
135 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
136 
137 static int snd_win = 128 * 1024;
138 module_param(snd_win, int, 0644);
139 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
140 
141 static struct workqueue_struct *workq;
142 
143 static struct sk_buff_head rxq;
144 
145 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
146 static void ep_timeout(unsigned long arg);
147 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
148 
149 static LIST_HEAD(timeout_list);
150 static spinlock_t timeout_lock;
151 
152 static void deref_qp(struct c4iw_ep *ep)
153 {
154 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
155 	clear_bit(QP_REFERENCED, &ep->com.flags);
156 }
157 
158 static void ref_qp(struct c4iw_ep *ep)
159 {
160 	set_bit(QP_REFERENCED, &ep->com.flags);
161 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
162 }
163 
164 static void start_ep_timer(struct c4iw_ep *ep)
165 {
166 	PDBG("%s ep %p\n", __func__, ep);
167 	if (timer_pending(&ep->timer)) {
168 		pr_err("%s timer already started! ep %p\n",
169 		       __func__, ep);
170 		return;
171 	}
172 	clear_bit(TIMEOUT, &ep->com.flags);
173 	c4iw_get_ep(&ep->com);
174 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
175 	ep->timer.data = (unsigned long)ep;
176 	ep->timer.function = ep_timeout;
177 	add_timer(&ep->timer);
178 }
179 
180 static int stop_ep_timer(struct c4iw_ep *ep)
181 {
182 	PDBG("%s ep %p stopping\n", __func__, ep);
183 	del_timer_sync(&ep->timer);
184 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
185 		c4iw_put_ep(&ep->com);
186 		return 0;
187 	}
188 	return 1;
189 }
190 
191 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
192 		  struct l2t_entry *l2e)
193 {
194 	int	error = 0;
195 
196 	if (c4iw_fatal_error(rdev)) {
197 		kfree_skb(skb);
198 		PDBG("%s - device in error state - dropping\n", __func__);
199 		return -EIO;
200 	}
201 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
202 	if (error < 0)
203 		kfree_skb(skb);
204 	return error < 0 ? error : 0;
205 }
206 
207 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
208 {
209 	int	error = 0;
210 
211 	if (c4iw_fatal_error(rdev)) {
212 		kfree_skb(skb);
213 		PDBG("%s - device in error state - dropping\n", __func__);
214 		return -EIO;
215 	}
216 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
217 	if (error < 0)
218 		kfree_skb(skb);
219 	return error < 0 ? error : 0;
220 }
221 
222 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
223 {
224 	struct cpl_tid_release *req;
225 
226 	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
227 	if (!skb)
228 		return;
229 	req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
230 	INIT_TP_WR(req, hwtid);
231 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
232 	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
233 	c4iw_ofld_send(rdev, skb);
234 	return;
235 }
236 
237 static void set_emss(struct c4iw_ep *ep, u16 opt)
238 {
239 	ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
240 		   ((AF_INET == ep->com.remote_addr.ss_family) ?
241 		    sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
242 		   sizeof(struct tcphdr);
243 	ep->mss = ep->emss;
244 	if (TCPOPT_TSTAMP_G(opt))
245 		ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
246 	if (ep->emss < 128)
247 		ep->emss = 128;
248 	if (ep->emss & 7)
249 		PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
250 		     TCPOPT_MSS_G(opt), ep->mss, ep->emss);
251 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
252 	     ep->mss, ep->emss);
253 }
254 
255 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
256 {
257 	enum c4iw_ep_state state;
258 
259 	mutex_lock(&epc->mutex);
260 	state = epc->state;
261 	mutex_unlock(&epc->mutex);
262 	return state;
263 }
264 
265 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
266 {
267 	epc->state = new;
268 }
269 
270 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
271 {
272 	mutex_lock(&epc->mutex);
273 	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
274 	__state_set(epc, new);
275 	mutex_unlock(&epc->mutex);
276 	return;
277 }
278 
279 static void *alloc_ep(int size, gfp_t gfp)
280 {
281 	struct c4iw_ep_common *epc;
282 
283 	epc = kzalloc(size, gfp);
284 	if (epc) {
285 		kref_init(&epc->kref);
286 		mutex_init(&epc->mutex);
287 		c4iw_init_wr_wait(&epc->wr_wait);
288 	}
289 	PDBG("%s alloc ep %p\n", __func__, epc);
290 	return epc;
291 }
292 
293 void _c4iw_free_ep(struct kref *kref)
294 {
295 	struct c4iw_ep *ep;
296 
297 	ep = container_of(kref, struct c4iw_ep, com.kref);
298 	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
299 	if (test_bit(QP_REFERENCED, &ep->com.flags))
300 		deref_qp(ep);
301 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
302 		if (ep->com.remote_addr.ss_family == AF_INET6) {
303 			struct sockaddr_in6 *sin6 =
304 					(struct sockaddr_in6 *)
305 					&ep->com.mapped_local_addr;
306 
307 			cxgb4_clip_release(
308 					ep->com.dev->rdev.lldi.ports[0],
309 					(const u32 *)&sin6->sin6_addr.s6_addr,
310 					1);
311 		}
312 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
313 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
314 		dst_release(ep->dst);
315 		cxgb4_l2t_release(ep->l2t);
316 	}
317 	if (test_bit(RELEASE_MAPINFO, &ep->com.flags)) {
318 		print_addr(&ep->com, __func__, "remove_mapinfo/mapping");
319 		iwpm_remove_mapinfo(&ep->com.local_addr,
320 				    &ep->com.mapped_local_addr);
321 		iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
322 	}
323 	kfree(ep);
324 }
325 
326 static void release_ep_resources(struct c4iw_ep *ep)
327 {
328 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
329 	c4iw_put_ep(&ep->com);
330 }
331 
332 static int status2errno(int status)
333 {
334 	switch (status) {
335 	case CPL_ERR_NONE:
336 		return 0;
337 	case CPL_ERR_CONN_RESET:
338 		return -ECONNRESET;
339 	case CPL_ERR_ARP_MISS:
340 		return -EHOSTUNREACH;
341 	case CPL_ERR_CONN_TIMEDOUT:
342 		return -ETIMEDOUT;
343 	case CPL_ERR_TCAM_FULL:
344 		return -ENOMEM;
345 	case CPL_ERR_CONN_EXIST:
346 		return -EADDRINUSE;
347 	default:
348 		return -EIO;
349 	}
350 }
351 
352 /*
353  * Try and reuse skbs already allocated...
354  */
355 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
356 {
357 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
358 		skb_trim(skb, 0);
359 		skb_get(skb);
360 		skb_reset_transport_header(skb);
361 	} else {
362 		skb = alloc_skb(len, gfp);
363 	}
364 	t4_set_arp_err_handler(skb, NULL, NULL);
365 	return skb;
366 }
367 
368 static struct net_device *get_real_dev(struct net_device *egress_dev)
369 {
370 	return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
371 }
372 
373 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
374 {
375 	int i;
376 
377 	egress_dev = get_real_dev(egress_dev);
378 	for (i = 0; i < dev->rdev.lldi.nports; i++)
379 		if (dev->rdev.lldi.ports[i] == egress_dev)
380 			return 1;
381 	return 0;
382 }
383 
384 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
385 				     __u8 *peer_ip, __be16 local_port,
386 				     __be16 peer_port, u8 tos,
387 				     __u32 sin6_scope_id)
388 {
389 	struct dst_entry *dst = NULL;
390 
391 	if (IS_ENABLED(CONFIG_IPV6)) {
392 		struct flowi6 fl6;
393 
394 		memset(&fl6, 0, sizeof(fl6));
395 		memcpy(&fl6.daddr, peer_ip, 16);
396 		memcpy(&fl6.saddr, local_ip, 16);
397 		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
398 			fl6.flowi6_oif = sin6_scope_id;
399 		dst = ip6_route_output(&init_net, NULL, &fl6);
400 		if (!dst)
401 			goto out;
402 		if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
403 		    !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
404 			dst_release(dst);
405 			dst = NULL;
406 		}
407 	}
408 
409 out:
410 	return dst;
411 }
412 
413 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
414 				 __be32 peer_ip, __be16 local_port,
415 				 __be16 peer_port, u8 tos)
416 {
417 	struct rtable *rt;
418 	struct flowi4 fl4;
419 	struct neighbour *n;
420 
421 	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
422 				   peer_port, local_port, IPPROTO_TCP,
423 				   tos, 0);
424 	if (IS_ERR(rt))
425 		return NULL;
426 	n = dst_neigh_lookup(&rt->dst, &peer_ip);
427 	if (!n)
428 		return NULL;
429 	if (!our_interface(dev, n->dev) &&
430 	    !(n->dev->flags & IFF_LOOPBACK)) {
431 		neigh_release(n);
432 		dst_release(&rt->dst);
433 		return NULL;
434 	}
435 	neigh_release(n);
436 	return &rt->dst;
437 }
438 
439 static void arp_failure_discard(void *handle, struct sk_buff *skb)
440 {
441 	PDBG("%s c4iw_dev %p\n", __func__, handle);
442 	kfree_skb(skb);
443 }
444 
445 /*
446  * Handle an ARP failure for an active open.
447  */
448 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
449 {
450 	struct c4iw_ep *ep = handle;
451 
452 	printk(KERN_ERR MOD "ARP failure during connect\n");
453 	kfree_skb(skb);
454 	connect_reply_upcall(ep, -EHOSTUNREACH);
455 	state_set(&ep->com, DEAD);
456 	if (ep->com.remote_addr.ss_family == AF_INET6) {
457 		struct sockaddr_in6 *sin6 =
458 			(struct sockaddr_in6 *)&ep->com.mapped_local_addr;
459 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
460 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
461 	}
462 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
463 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
464 	dst_release(ep->dst);
465 	cxgb4_l2t_release(ep->l2t);
466 	c4iw_put_ep(&ep->com);
467 }
468 
469 /*
470  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
471  * and send it along.
472  */
473 static void abort_arp_failure(void *handle, struct sk_buff *skb)
474 {
475 	struct c4iw_rdev *rdev = handle;
476 	struct cpl_abort_req *req = cplhdr(skb);
477 
478 	PDBG("%s rdev %p\n", __func__, rdev);
479 	req->cmd = CPL_ABORT_NO_RST;
480 	c4iw_ofld_send(rdev, skb);
481 }
482 
483 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
484 {
485 	unsigned int flowclen = 80;
486 	struct fw_flowc_wr *flowc;
487 	int i;
488 
489 	skb = get_skb(skb, flowclen, GFP_KERNEL);
490 	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
491 
492 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
493 					   FW_FLOWC_WR_NPARAMS_V(8));
494 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(flowclen,
495 					  16)) | FW_WR_FLOWID_V(ep->hwtid));
496 
497 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
498 	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
499 					    (ep->com.dev->rdev.lldi.pf));
500 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
501 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
502 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
503 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
504 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
505 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
506 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
507 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
508 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
509 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
510 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
511 	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
512 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
513 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
514 	/* Pad WR to 16 byte boundary */
515 	flowc->mnemval[8].mnemonic = 0;
516 	flowc->mnemval[8].val = 0;
517 	for (i = 0; i < 9; i++) {
518 		flowc->mnemval[i].r4[0] = 0;
519 		flowc->mnemval[i].r4[1] = 0;
520 		flowc->mnemval[i].r4[2] = 0;
521 	}
522 
523 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
524 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
525 }
526 
527 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
528 {
529 	struct cpl_close_con_req *req;
530 	struct sk_buff *skb;
531 	int wrlen = roundup(sizeof *req, 16);
532 
533 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
534 	skb = get_skb(NULL, wrlen, gfp);
535 	if (!skb) {
536 		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
537 		return -ENOMEM;
538 	}
539 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
540 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
541 	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
542 	memset(req, 0, wrlen);
543 	INIT_TP_WR(req, ep->hwtid);
544 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
545 						    ep->hwtid));
546 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
547 }
548 
549 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
550 {
551 	struct cpl_abort_req *req;
552 	int wrlen = roundup(sizeof *req, 16);
553 
554 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
555 	skb = get_skb(skb, wrlen, gfp);
556 	if (!skb) {
557 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
558 		       __func__);
559 		return -ENOMEM;
560 	}
561 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
562 	t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
563 	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
564 	memset(req, 0, wrlen);
565 	INIT_TP_WR(req, ep->hwtid);
566 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
567 	req->cmd = CPL_ABORT_SEND_RST;
568 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
569 }
570 
571 /*
572  * c4iw_form_pm_msg - Form a port mapper message with mapping info
573  */
574 static void c4iw_form_pm_msg(struct c4iw_ep *ep,
575 				struct iwpm_sa_data *pm_msg)
576 {
577 	memcpy(&pm_msg->loc_addr, &ep->com.local_addr,
578 		sizeof(ep->com.local_addr));
579 	memcpy(&pm_msg->rem_addr, &ep->com.remote_addr,
580 		sizeof(ep->com.remote_addr));
581 }
582 
583 /*
584  * c4iw_form_reg_msg - Form a port mapper message with dev info
585  */
586 static void c4iw_form_reg_msg(struct c4iw_dev *dev,
587 				struct iwpm_dev_data *pm_msg)
588 {
589 	memcpy(pm_msg->dev_name, dev->ibdev.name, IWPM_DEVNAME_SIZE);
590 	memcpy(pm_msg->if_name, dev->rdev.lldi.ports[0]->name,
591 				IWPM_IFNAME_SIZE);
592 }
593 
594 static void c4iw_record_pm_msg(struct c4iw_ep *ep,
595 			struct iwpm_sa_data *pm_msg)
596 {
597 	memcpy(&ep->com.mapped_local_addr, &pm_msg->mapped_loc_addr,
598 		sizeof(ep->com.mapped_local_addr));
599 	memcpy(&ep->com.mapped_remote_addr, &pm_msg->mapped_rem_addr,
600 		sizeof(ep->com.mapped_remote_addr));
601 }
602 
603 static int get_remote_addr(struct c4iw_ep *parent_ep, struct c4iw_ep *child_ep)
604 {
605 	int ret;
606 
607 	print_addr(&parent_ep->com, __func__, "get_remote_addr parent_ep ");
608 	print_addr(&child_ep->com, __func__, "get_remote_addr child_ep ");
609 
610 	ret = iwpm_get_remote_info(&parent_ep->com.mapped_local_addr,
611 				   &child_ep->com.mapped_remote_addr,
612 				   &child_ep->com.remote_addr, RDMA_NL_C4IW);
613 	if (ret)
614 		PDBG("Unable to find remote peer addr info - err %d\n", ret);
615 
616 	return ret;
617 }
618 
619 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
620 		     unsigned int *idx, int use_ts, int ipv6)
621 {
622 	unsigned short hdr_size = (ipv6 ?
623 				   sizeof(struct ipv6hdr) :
624 				   sizeof(struct iphdr)) +
625 				  sizeof(struct tcphdr) +
626 				  (use_ts ?
627 				   round_up(TCPOLEN_TIMESTAMP, 4) : 0);
628 	unsigned short data_size = mtu - hdr_size;
629 
630 	cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
631 }
632 
633 static int send_connect(struct c4iw_ep *ep)
634 {
635 	struct cpl_act_open_req *req = NULL;
636 	struct cpl_t5_act_open_req *t5req = NULL;
637 	struct cpl_t6_act_open_req *t6req = NULL;
638 	struct cpl_act_open_req6 *req6 = NULL;
639 	struct cpl_t5_act_open_req6 *t5req6 = NULL;
640 	struct cpl_t6_act_open_req6 *t6req6 = NULL;
641 	struct sk_buff *skb;
642 	u64 opt0;
643 	u32 opt2;
644 	unsigned int mtu_idx;
645 	int wscale;
646 	int win, sizev4, sizev6, wrlen;
647 	struct sockaddr_in *la = (struct sockaddr_in *)
648 				 &ep->com.mapped_local_addr;
649 	struct sockaddr_in *ra = (struct sockaddr_in *)
650 				 &ep->com.mapped_remote_addr;
651 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
652 				   &ep->com.mapped_local_addr;
653 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
654 				   &ep->com.mapped_remote_addr;
655 	int ret;
656 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
657 	u32 isn = (prandom_u32() & ~7UL) - 1;
658 
659 	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
660 	case CHELSIO_T4:
661 		sizev4 = sizeof(struct cpl_act_open_req);
662 		sizev6 = sizeof(struct cpl_act_open_req6);
663 		break;
664 	case CHELSIO_T5:
665 		sizev4 = sizeof(struct cpl_t5_act_open_req);
666 		sizev6 = sizeof(struct cpl_t5_act_open_req6);
667 		break;
668 	case CHELSIO_T6:
669 		sizev4 = sizeof(struct cpl_t6_act_open_req);
670 		sizev6 = sizeof(struct cpl_t6_act_open_req6);
671 		break;
672 	default:
673 		pr_err("T%d Chip is not supported\n",
674 		       CHELSIO_CHIP_VERSION(adapter_type));
675 		return -EINVAL;
676 	}
677 
678 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
679 			roundup(sizev4, 16) :
680 			roundup(sizev6, 16);
681 
682 	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
683 
684 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
685 	if (!skb) {
686 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
687 		       __func__);
688 		return -ENOMEM;
689 	}
690 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
691 
692 	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
693 		 enable_tcp_timestamps,
694 		 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
695 	wscale = compute_wscale(rcv_win);
696 
697 	/*
698 	 * Specify the largest window that will fit in opt0. The
699 	 * remainder will be specified in the rx_data_ack.
700 	 */
701 	win = ep->rcv_win >> 10;
702 	if (win > RCV_BUFSIZ_M)
703 		win = RCV_BUFSIZ_M;
704 
705 	opt0 = (nocong ? NO_CONG_F : 0) |
706 	       KEEP_ALIVE_F |
707 	       DELACK_F |
708 	       WND_SCALE_V(wscale) |
709 	       MSS_IDX_V(mtu_idx) |
710 	       L2T_IDX_V(ep->l2t->idx) |
711 	       TX_CHAN_V(ep->tx_chan) |
712 	       SMAC_SEL_V(ep->smac_idx) |
713 	       DSCP_V(ep->tos) |
714 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
715 	       RCV_BUFSIZ_V(win);
716 	opt2 = RX_CHANNEL_V(0) |
717 	       CCTRL_ECN_V(enable_ecn) |
718 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
719 	if (enable_tcp_timestamps)
720 		opt2 |= TSTAMPS_EN_F;
721 	if (enable_tcp_sack)
722 		opt2 |= SACK_EN_F;
723 	if (wscale && enable_tcp_window_scaling)
724 		opt2 |= WND_SCALE_EN_F;
725 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
726 		if (peer2peer)
727 			isn += 4;
728 
729 		opt2 |= T5_OPT_2_VALID_F;
730 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
731 		opt2 |= T5_ISS_F;
732 	}
733 
734 	if (ep->com.remote_addr.ss_family == AF_INET6)
735 		cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
736 			       (const u32 *)&la6->sin6_addr.s6_addr, 1);
737 
738 	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
739 
740 	if (ep->com.remote_addr.ss_family == AF_INET) {
741 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
742 		case CHELSIO_T4:
743 			req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
744 			INIT_TP_WR(req, 0);
745 			break;
746 		case CHELSIO_T5:
747 			t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
748 					wrlen);
749 			INIT_TP_WR(t5req, 0);
750 			req = (struct cpl_act_open_req *)t5req;
751 			break;
752 		case CHELSIO_T6:
753 			t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
754 					wrlen);
755 			INIT_TP_WR(t6req, 0);
756 			req = (struct cpl_act_open_req *)t6req;
757 			t5req = (struct cpl_t5_act_open_req *)t6req;
758 			break;
759 		default:
760 			pr_err("T%d Chip is not supported\n",
761 			       CHELSIO_CHIP_VERSION(adapter_type));
762 			ret = -EINVAL;
763 			goto clip_release;
764 		}
765 
766 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
767 					((ep->rss_qid<<14) | ep->atid)));
768 		req->local_port = la->sin_port;
769 		req->peer_port = ra->sin_port;
770 		req->local_ip = la->sin_addr.s_addr;
771 		req->peer_ip = ra->sin_addr.s_addr;
772 		req->opt0 = cpu_to_be64(opt0);
773 
774 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
775 			req->params = cpu_to_be32(cxgb4_select_ntuple(
776 						ep->com.dev->rdev.lldi.ports[0],
777 						ep->l2t));
778 			req->opt2 = cpu_to_be32(opt2);
779 		} else {
780 			t5req->params = cpu_to_be64(FILTER_TUPLE_V(
781 						cxgb4_select_ntuple(
782 						ep->com.dev->rdev.lldi.ports[0],
783 						ep->l2t)));
784 			t5req->rsvd = cpu_to_be32(isn);
785 			PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
786 			t5req->opt2 = cpu_to_be32(opt2);
787 		}
788 	} else {
789 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
790 		case CHELSIO_T4:
791 			req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
792 			INIT_TP_WR(req6, 0);
793 			break;
794 		case CHELSIO_T5:
795 			t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
796 					wrlen);
797 			INIT_TP_WR(t5req6, 0);
798 			req6 = (struct cpl_act_open_req6 *)t5req6;
799 			break;
800 		case CHELSIO_T6:
801 			t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
802 					wrlen);
803 			INIT_TP_WR(t6req6, 0);
804 			req6 = (struct cpl_act_open_req6 *)t6req6;
805 			t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
806 			break;
807 		default:
808 			pr_err("T%d Chip is not supported\n",
809 			       CHELSIO_CHIP_VERSION(adapter_type));
810 			ret = -EINVAL;
811 			goto clip_release;
812 		}
813 
814 		OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
815 					((ep->rss_qid<<14)|ep->atid)));
816 		req6->local_port = la6->sin6_port;
817 		req6->peer_port = ra6->sin6_port;
818 		req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
819 		req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
820 		req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
821 		req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
822 		req6->opt0 = cpu_to_be64(opt0);
823 
824 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
825 			req6->params = cpu_to_be32(cxgb4_select_ntuple(
826 						ep->com.dev->rdev.lldi.ports[0],
827 						ep->l2t));
828 			req6->opt2 = cpu_to_be32(opt2);
829 		} else {
830 			t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
831 						cxgb4_select_ntuple(
832 						ep->com.dev->rdev.lldi.ports[0],
833 						ep->l2t)));
834 			t5req6->rsvd = cpu_to_be32(isn);
835 			PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
836 			t5req6->opt2 = cpu_to_be32(opt2);
837 		}
838 	}
839 
840 	set_bit(ACT_OPEN_REQ, &ep->com.history);
841 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
842 clip_release:
843 	if (ret && ep->com.remote_addr.ss_family == AF_INET6)
844 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
845 				   (const u32 *)&la6->sin6_addr.s6_addr, 1);
846 	return ret;
847 }
848 
849 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
850 		u8 mpa_rev_to_use)
851 {
852 	int mpalen, wrlen;
853 	struct fw_ofld_tx_data_wr *req;
854 	struct mpa_message *mpa;
855 	struct mpa_v2_conn_params mpa_v2_params;
856 
857 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
858 
859 	BUG_ON(skb_cloned(skb));
860 
861 	mpalen = sizeof(*mpa) + ep->plen;
862 	if (mpa_rev_to_use == 2)
863 		mpalen += sizeof(struct mpa_v2_conn_params);
864 	wrlen = roundup(mpalen + sizeof *req, 16);
865 	skb = get_skb(skb, wrlen, GFP_KERNEL);
866 	if (!skb) {
867 		connect_reply_upcall(ep, -ENOMEM);
868 		return;
869 	}
870 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
871 
872 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
873 	memset(req, 0, wrlen);
874 	req->op_to_immdlen = cpu_to_be32(
875 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
876 		FW_WR_COMPL_F |
877 		FW_WR_IMMDLEN_V(mpalen));
878 	req->flowid_len16 = cpu_to_be32(
879 		FW_WR_FLOWID_V(ep->hwtid) |
880 		FW_WR_LEN16_V(wrlen >> 4));
881 	req->plen = cpu_to_be32(mpalen);
882 	req->tunnel_to_proxy = cpu_to_be32(
883 		FW_OFLD_TX_DATA_WR_FLUSH_F |
884 		FW_OFLD_TX_DATA_WR_SHOVE_F);
885 
886 	mpa = (struct mpa_message *)(req + 1);
887 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
888 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
889 		     (markers_enabled ? MPA_MARKERS : 0) |
890 		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
891 	mpa->private_data_size = htons(ep->plen);
892 	mpa->revision = mpa_rev_to_use;
893 	if (mpa_rev_to_use == 1) {
894 		ep->tried_with_mpa_v1 = 1;
895 		ep->retry_with_mpa_v1 = 0;
896 	}
897 
898 	if (mpa_rev_to_use == 2) {
899 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
900 					       sizeof (struct mpa_v2_conn_params));
901 		PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
902 		     ep->ord);
903 		mpa_v2_params.ird = htons((u16)ep->ird);
904 		mpa_v2_params.ord = htons((u16)ep->ord);
905 
906 		if (peer2peer) {
907 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
908 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
909 				mpa_v2_params.ord |=
910 					htons(MPA_V2_RDMA_WRITE_RTR);
911 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
912 				mpa_v2_params.ord |=
913 					htons(MPA_V2_RDMA_READ_RTR);
914 		}
915 		memcpy(mpa->private_data, &mpa_v2_params,
916 		       sizeof(struct mpa_v2_conn_params));
917 
918 		if (ep->plen)
919 			memcpy(mpa->private_data +
920 			       sizeof(struct mpa_v2_conn_params),
921 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
922 	} else
923 		if (ep->plen)
924 			memcpy(mpa->private_data,
925 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
926 
927 	/*
928 	 * Reference the mpa skb.  This ensures the data area
929 	 * will remain in memory until the hw acks the tx.
930 	 * Function fw4_ack() will deref it.
931 	 */
932 	skb_get(skb);
933 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
934 	BUG_ON(ep->mpa_skb);
935 	ep->mpa_skb = skb;
936 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
937 	start_ep_timer(ep);
938 	__state_set(&ep->com, MPA_REQ_SENT);
939 	ep->mpa_attr.initiator = 1;
940 	ep->snd_seq += mpalen;
941 	return;
942 }
943 
944 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
945 {
946 	int mpalen, wrlen;
947 	struct fw_ofld_tx_data_wr *req;
948 	struct mpa_message *mpa;
949 	struct sk_buff *skb;
950 	struct mpa_v2_conn_params mpa_v2_params;
951 
952 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
953 
954 	mpalen = sizeof(*mpa) + plen;
955 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
956 		mpalen += sizeof(struct mpa_v2_conn_params);
957 	wrlen = roundup(mpalen + sizeof *req, 16);
958 
959 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
960 	if (!skb) {
961 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
962 		return -ENOMEM;
963 	}
964 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
965 
966 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
967 	memset(req, 0, wrlen);
968 	req->op_to_immdlen = cpu_to_be32(
969 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
970 		FW_WR_COMPL_F |
971 		FW_WR_IMMDLEN_V(mpalen));
972 	req->flowid_len16 = cpu_to_be32(
973 		FW_WR_FLOWID_V(ep->hwtid) |
974 		FW_WR_LEN16_V(wrlen >> 4));
975 	req->plen = cpu_to_be32(mpalen);
976 	req->tunnel_to_proxy = cpu_to_be32(
977 		FW_OFLD_TX_DATA_WR_FLUSH_F |
978 		FW_OFLD_TX_DATA_WR_SHOVE_F);
979 
980 	mpa = (struct mpa_message *)(req + 1);
981 	memset(mpa, 0, sizeof(*mpa));
982 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
983 	mpa->flags = MPA_REJECT;
984 	mpa->revision = ep->mpa_attr.version;
985 	mpa->private_data_size = htons(plen);
986 
987 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
988 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
989 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
990 					       sizeof (struct mpa_v2_conn_params));
991 		mpa_v2_params.ird = htons(((u16)ep->ird) |
992 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
993 					   0));
994 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
995 					  (p2p_type ==
996 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
997 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
998 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
999 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
1000 		memcpy(mpa->private_data, &mpa_v2_params,
1001 		       sizeof(struct mpa_v2_conn_params));
1002 
1003 		if (ep->plen)
1004 			memcpy(mpa->private_data +
1005 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1006 	} else
1007 		if (plen)
1008 			memcpy(mpa->private_data, pdata, plen);
1009 
1010 	/*
1011 	 * Reference the mpa skb again.  This ensures the data area
1012 	 * will remain in memory until the hw acks the tx.
1013 	 * Function fw4_ack() will deref it.
1014 	 */
1015 	skb_get(skb);
1016 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1017 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1018 	BUG_ON(ep->mpa_skb);
1019 	ep->mpa_skb = skb;
1020 	ep->snd_seq += mpalen;
1021 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1022 }
1023 
1024 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1025 {
1026 	int mpalen, wrlen;
1027 	struct fw_ofld_tx_data_wr *req;
1028 	struct mpa_message *mpa;
1029 	struct sk_buff *skb;
1030 	struct mpa_v2_conn_params mpa_v2_params;
1031 
1032 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1033 
1034 	mpalen = sizeof(*mpa) + plen;
1035 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1036 		mpalen += sizeof(struct mpa_v2_conn_params);
1037 	wrlen = roundup(mpalen + sizeof *req, 16);
1038 
1039 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1040 	if (!skb) {
1041 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1042 		return -ENOMEM;
1043 	}
1044 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1045 
1046 	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1047 	memset(req, 0, wrlen);
1048 	req->op_to_immdlen = cpu_to_be32(
1049 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1050 		FW_WR_COMPL_F |
1051 		FW_WR_IMMDLEN_V(mpalen));
1052 	req->flowid_len16 = cpu_to_be32(
1053 		FW_WR_FLOWID_V(ep->hwtid) |
1054 		FW_WR_LEN16_V(wrlen >> 4));
1055 	req->plen = cpu_to_be32(mpalen);
1056 	req->tunnel_to_proxy = cpu_to_be32(
1057 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1058 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1059 
1060 	mpa = (struct mpa_message *)(req + 1);
1061 	memset(mpa, 0, sizeof(*mpa));
1062 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1063 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1064 		     (markers_enabled ? MPA_MARKERS : 0);
1065 	mpa->revision = ep->mpa_attr.version;
1066 	mpa->private_data_size = htons(plen);
1067 
1068 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1069 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1070 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1071 					       sizeof (struct mpa_v2_conn_params));
1072 		mpa_v2_params.ird = htons((u16)ep->ird);
1073 		mpa_v2_params.ord = htons((u16)ep->ord);
1074 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1075 					FW_RI_INIT_P2PTYPE_DISABLED)) {
1076 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1077 
1078 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1079 				mpa_v2_params.ord |=
1080 					htons(MPA_V2_RDMA_WRITE_RTR);
1081 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1082 				mpa_v2_params.ord |=
1083 					htons(MPA_V2_RDMA_READ_RTR);
1084 		}
1085 
1086 		memcpy(mpa->private_data, &mpa_v2_params,
1087 		       sizeof(struct mpa_v2_conn_params));
1088 
1089 		if (ep->plen)
1090 			memcpy(mpa->private_data +
1091 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1092 	} else
1093 		if (plen)
1094 			memcpy(mpa->private_data, pdata, plen);
1095 
1096 	/*
1097 	 * Reference the mpa skb.  This ensures the data area
1098 	 * will remain in memory until the hw acks the tx.
1099 	 * Function fw4_ack() will deref it.
1100 	 */
1101 	skb_get(skb);
1102 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1103 	ep->mpa_skb = skb;
1104 	__state_set(&ep->com, MPA_REP_SENT);
1105 	ep->snd_seq += mpalen;
1106 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1107 }
1108 
1109 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1110 {
1111 	struct c4iw_ep *ep;
1112 	struct cpl_act_establish *req = cplhdr(skb);
1113 	unsigned int tid = GET_TID(req);
1114 	unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1115 	struct tid_info *t = dev->rdev.lldi.tids;
1116 
1117 	ep = lookup_atid(t, atid);
1118 
1119 	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1120 	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1121 
1122 	mutex_lock(&ep->com.mutex);
1123 	dst_confirm(ep->dst);
1124 
1125 	/* setup the hwtid for this connection */
1126 	ep->hwtid = tid;
1127 	cxgb4_insert_tid(t, ep, tid);
1128 	insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
1129 
1130 	ep->snd_seq = be32_to_cpu(req->snd_isn);
1131 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1132 
1133 	set_emss(ep, ntohs(req->tcp_opt));
1134 
1135 	/* dealloc the atid */
1136 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1137 	cxgb4_free_atid(t, atid);
1138 	set_bit(ACT_ESTAB, &ep->com.history);
1139 
1140 	/* start MPA negotiation */
1141 	send_flowc(ep, NULL);
1142 	if (ep->retry_with_mpa_v1)
1143 		send_mpa_req(ep, skb, 1);
1144 	else
1145 		send_mpa_req(ep, skb, mpa_rev);
1146 	mutex_unlock(&ep->com.mutex);
1147 	return 0;
1148 }
1149 
1150 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1151 {
1152 	struct iw_cm_event event;
1153 
1154 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1155 	memset(&event, 0, sizeof(event));
1156 	event.event = IW_CM_EVENT_CLOSE;
1157 	event.status = status;
1158 	if (ep->com.cm_id) {
1159 		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1160 		     ep, ep->com.cm_id, ep->hwtid);
1161 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1162 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1163 		ep->com.cm_id = NULL;
1164 		set_bit(CLOSE_UPCALL, &ep->com.history);
1165 	}
1166 }
1167 
1168 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1169 {
1170 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1171 	__state_set(&ep->com, ABORTING);
1172 	set_bit(ABORT_CONN, &ep->com.history);
1173 	return send_abort(ep, skb, gfp);
1174 }
1175 
1176 static void peer_close_upcall(struct c4iw_ep *ep)
1177 {
1178 	struct iw_cm_event event;
1179 
1180 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1181 	memset(&event, 0, sizeof(event));
1182 	event.event = IW_CM_EVENT_DISCONNECT;
1183 	if (ep->com.cm_id) {
1184 		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1185 		     ep, ep->com.cm_id, ep->hwtid);
1186 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1187 		set_bit(DISCONN_UPCALL, &ep->com.history);
1188 	}
1189 }
1190 
1191 static void peer_abort_upcall(struct c4iw_ep *ep)
1192 {
1193 	struct iw_cm_event event;
1194 
1195 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1196 	memset(&event, 0, sizeof(event));
1197 	event.event = IW_CM_EVENT_CLOSE;
1198 	event.status = -ECONNRESET;
1199 	if (ep->com.cm_id) {
1200 		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1201 		     ep->com.cm_id, ep->hwtid);
1202 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1203 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1204 		ep->com.cm_id = NULL;
1205 		set_bit(ABORT_UPCALL, &ep->com.history);
1206 	}
1207 }
1208 
1209 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1210 {
1211 	struct iw_cm_event event;
1212 
1213 	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1214 	memset(&event, 0, sizeof(event));
1215 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1216 	event.status = status;
1217 	memcpy(&event.local_addr, &ep->com.local_addr,
1218 	       sizeof(ep->com.local_addr));
1219 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1220 	       sizeof(ep->com.remote_addr));
1221 
1222 	if ((status == 0) || (status == -ECONNREFUSED)) {
1223 		if (!ep->tried_with_mpa_v1) {
1224 			/* this means MPA_v2 is used */
1225 			event.ord = ep->ird;
1226 			event.ird = ep->ord;
1227 			event.private_data_len = ep->plen -
1228 				sizeof(struct mpa_v2_conn_params);
1229 			event.private_data = ep->mpa_pkt +
1230 				sizeof(struct mpa_message) +
1231 				sizeof(struct mpa_v2_conn_params);
1232 		} else {
1233 			/* this means MPA_v1 is used */
1234 			event.ord = cur_max_read_depth(ep->com.dev);
1235 			event.ird = cur_max_read_depth(ep->com.dev);
1236 			event.private_data_len = ep->plen;
1237 			event.private_data = ep->mpa_pkt +
1238 				sizeof(struct mpa_message);
1239 		}
1240 	}
1241 
1242 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1243 	     ep->hwtid, status);
1244 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1245 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1246 
1247 	if (status < 0) {
1248 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1249 		ep->com.cm_id = NULL;
1250 	}
1251 }
1252 
1253 static int connect_request_upcall(struct c4iw_ep *ep)
1254 {
1255 	struct iw_cm_event event;
1256 	int ret;
1257 
1258 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1259 	memset(&event, 0, sizeof(event));
1260 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1261 	memcpy(&event.local_addr, &ep->com.local_addr,
1262 	       sizeof(ep->com.local_addr));
1263 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1264 	       sizeof(ep->com.remote_addr));
1265 	event.provider_data = ep;
1266 	if (!ep->tried_with_mpa_v1) {
1267 		/* this means MPA_v2 is used */
1268 		event.ord = ep->ord;
1269 		event.ird = ep->ird;
1270 		event.private_data_len = ep->plen -
1271 			sizeof(struct mpa_v2_conn_params);
1272 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1273 			sizeof(struct mpa_v2_conn_params);
1274 	} else {
1275 		/* this means MPA_v1 is used. Send max supported */
1276 		event.ord = cur_max_read_depth(ep->com.dev);
1277 		event.ird = cur_max_read_depth(ep->com.dev);
1278 		event.private_data_len = ep->plen;
1279 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1280 	}
1281 	c4iw_get_ep(&ep->com);
1282 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1283 						      &event);
1284 	if (ret)
1285 		c4iw_put_ep(&ep->com);
1286 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1287 	c4iw_put_ep(&ep->parent_ep->com);
1288 	return ret;
1289 }
1290 
1291 static void established_upcall(struct c4iw_ep *ep)
1292 {
1293 	struct iw_cm_event event;
1294 
1295 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1296 	memset(&event, 0, sizeof(event));
1297 	event.event = IW_CM_EVENT_ESTABLISHED;
1298 	event.ird = ep->ord;
1299 	event.ord = ep->ird;
1300 	if (ep->com.cm_id) {
1301 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1302 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1303 		set_bit(ESTAB_UPCALL, &ep->com.history);
1304 	}
1305 }
1306 
1307 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1308 {
1309 	struct cpl_rx_data_ack *req;
1310 	struct sk_buff *skb;
1311 	int wrlen = roundup(sizeof *req, 16);
1312 
1313 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1314 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1315 	if (!skb) {
1316 		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1317 		return 0;
1318 	}
1319 
1320 	/*
1321 	 * If we couldn't specify the entire rcv window at connection setup
1322 	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1323 	 * then add the overage in to the credits returned.
1324 	 */
1325 	if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1326 		credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1327 
1328 	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1329 	memset(req, 0, wrlen);
1330 	INIT_TP_WR(req, ep->hwtid);
1331 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1332 						    ep->hwtid));
1333 	req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1334 				       RX_DACK_CHANGE_F |
1335 				       RX_DACK_MODE_V(dack_mode));
1336 	set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1337 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1338 	return credits;
1339 }
1340 
1341 #define RELAXED_IRD_NEGOTIATION 1
1342 
1343 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1344 {
1345 	struct mpa_message *mpa;
1346 	struct mpa_v2_conn_params *mpa_v2_params;
1347 	u16 plen;
1348 	u16 resp_ird, resp_ord;
1349 	u8 rtr_mismatch = 0, insuff_ird = 0;
1350 	struct c4iw_qp_attributes attrs;
1351 	enum c4iw_qp_attr_mask mask;
1352 	int err;
1353 	int disconnect = 0;
1354 
1355 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1356 
1357 	/*
1358 	 * Stop mpa timer.  If it expired, then
1359 	 * we ignore the MPA reply.  process_timeout()
1360 	 * will abort the connection.
1361 	 */
1362 	if (stop_ep_timer(ep))
1363 		return 0;
1364 
1365 	/*
1366 	 * If we get more than the supported amount of private data
1367 	 * then we must fail this connection.
1368 	 */
1369 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1370 		err = -EINVAL;
1371 		goto err;
1372 	}
1373 
1374 	/*
1375 	 * copy the new data into our accumulation buffer.
1376 	 */
1377 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1378 				  skb->len);
1379 	ep->mpa_pkt_len += skb->len;
1380 
1381 	/*
1382 	 * if we don't even have the mpa message, then bail.
1383 	 */
1384 	if (ep->mpa_pkt_len < sizeof(*mpa))
1385 		return 0;
1386 	mpa = (struct mpa_message *) ep->mpa_pkt;
1387 
1388 	/* Validate MPA header. */
1389 	if (mpa->revision > mpa_rev) {
1390 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1391 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1392 		err = -EPROTO;
1393 		goto err;
1394 	}
1395 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1396 		err = -EPROTO;
1397 		goto err;
1398 	}
1399 
1400 	plen = ntohs(mpa->private_data_size);
1401 
1402 	/*
1403 	 * Fail if there's too much private data.
1404 	 */
1405 	if (plen > MPA_MAX_PRIVATE_DATA) {
1406 		err = -EPROTO;
1407 		goto err;
1408 	}
1409 
1410 	/*
1411 	 * If plen does not account for pkt size
1412 	 */
1413 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1414 		err = -EPROTO;
1415 		goto err;
1416 	}
1417 
1418 	ep->plen = (u8) plen;
1419 
1420 	/*
1421 	 * If we don't have all the pdata yet, then bail.
1422 	 * We'll continue process when more data arrives.
1423 	 */
1424 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1425 		return 0;
1426 
1427 	if (mpa->flags & MPA_REJECT) {
1428 		err = -ECONNREFUSED;
1429 		goto err;
1430 	}
1431 
1432 	/*
1433 	 * If we get here we have accumulated the entire mpa
1434 	 * start reply message including private data. And
1435 	 * the MPA header is valid.
1436 	 */
1437 	__state_set(&ep->com, FPDU_MODE);
1438 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1439 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1440 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1441 	ep->mpa_attr.version = mpa->revision;
1442 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1443 
1444 	if (mpa->revision == 2) {
1445 		ep->mpa_attr.enhanced_rdma_conn =
1446 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1447 		if (ep->mpa_attr.enhanced_rdma_conn) {
1448 			mpa_v2_params = (struct mpa_v2_conn_params *)
1449 				(ep->mpa_pkt + sizeof(*mpa));
1450 			resp_ird = ntohs(mpa_v2_params->ird) &
1451 				MPA_V2_IRD_ORD_MASK;
1452 			resp_ord = ntohs(mpa_v2_params->ord) &
1453 				MPA_V2_IRD_ORD_MASK;
1454 			PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1455 			     __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1456 
1457 			/*
1458 			 * This is a double-check. Ideally, below checks are
1459 			 * not required since ird/ord stuff has been taken
1460 			 * care of in c4iw_accept_cr
1461 			 */
1462 			if (ep->ird < resp_ord) {
1463 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1464 				    ep->com.dev->rdev.lldi.max_ordird_qp)
1465 					ep->ird = resp_ord;
1466 				else
1467 					insuff_ird = 1;
1468 			} else if (ep->ird > resp_ord) {
1469 				ep->ird = resp_ord;
1470 			}
1471 			if (ep->ord > resp_ird) {
1472 				if (RELAXED_IRD_NEGOTIATION)
1473 					ep->ord = resp_ird;
1474 				else
1475 					insuff_ird = 1;
1476 			}
1477 			if (insuff_ird) {
1478 				err = -ENOMEM;
1479 				ep->ird = resp_ord;
1480 				ep->ord = resp_ird;
1481 			}
1482 
1483 			if (ntohs(mpa_v2_params->ird) &
1484 					MPA_V2_PEER2PEER_MODEL) {
1485 				if (ntohs(mpa_v2_params->ord) &
1486 						MPA_V2_RDMA_WRITE_RTR)
1487 					ep->mpa_attr.p2p_type =
1488 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1489 				else if (ntohs(mpa_v2_params->ord) &
1490 						MPA_V2_RDMA_READ_RTR)
1491 					ep->mpa_attr.p2p_type =
1492 						FW_RI_INIT_P2PTYPE_READ_REQ;
1493 			}
1494 		}
1495 	} else if (mpa->revision == 1)
1496 		if (peer2peer)
1497 			ep->mpa_attr.p2p_type = p2p_type;
1498 
1499 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1500 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1501 	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
1502 	     ep->mpa_attr.recv_marker_enabled,
1503 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1504 	     ep->mpa_attr.p2p_type, p2p_type);
1505 
1506 	/*
1507 	 * If responder's RTR does not match with that of initiator, assign
1508 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1509 	 * generated when moving QP to RTS state.
1510 	 * A TERM message will be sent after QP has moved to RTS state
1511 	 */
1512 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1513 			(ep->mpa_attr.p2p_type != p2p_type)) {
1514 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1515 		rtr_mismatch = 1;
1516 	}
1517 
1518 	attrs.mpa_attr = ep->mpa_attr;
1519 	attrs.max_ird = ep->ird;
1520 	attrs.max_ord = ep->ord;
1521 	attrs.llp_stream_handle = ep;
1522 	attrs.next_state = C4IW_QP_STATE_RTS;
1523 
1524 	mask = C4IW_QP_ATTR_NEXT_STATE |
1525 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1526 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1527 
1528 	/* bind QP and TID with INIT_WR */
1529 	err = c4iw_modify_qp(ep->com.qp->rhp,
1530 			     ep->com.qp, mask, &attrs, 1);
1531 	if (err)
1532 		goto err;
1533 
1534 	/*
1535 	 * If responder's RTR requirement did not match with what initiator
1536 	 * supports, generate TERM message
1537 	 */
1538 	if (rtr_mismatch) {
1539 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1540 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1541 		attrs.ecode = MPA_NOMATCH_RTR;
1542 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1543 		attrs.send_term = 1;
1544 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1545 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1546 		err = -ENOMEM;
1547 		disconnect = 1;
1548 		goto out;
1549 	}
1550 
1551 	/*
1552 	 * Generate TERM if initiator IRD is not sufficient for responder
1553 	 * provided ORD. Currently, we do the same behaviour even when
1554 	 * responder provided IRD is also not sufficient as regards to
1555 	 * initiator ORD.
1556 	 */
1557 	if (insuff_ird) {
1558 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1559 				__func__);
1560 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1561 		attrs.ecode = MPA_INSUFF_IRD;
1562 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1563 		attrs.send_term = 1;
1564 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1565 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1566 		err = -ENOMEM;
1567 		disconnect = 1;
1568 		goto out;
1569 	}
1570 	goto out;
1571 err:
1572 	__state_set(&ep->com, ABORTING);
1573 	send_abort(ep, skb, GFP_KERNEL);
1574 out:
1575 	connect_reply_upcall(ep, err);
1576 	return disconnect;
1577 }
1578 
1579 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1580 {
1581 	struct mpa_message *mpa;
1582 	struct mpa_v2_conn_params *mpa_v2_params;
1583 	u16 plen;
1584 
1585 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1586 
1587 	/*
1588 	 * If we get more than the supported amount of private data
1589 	 * then we must fail this connection.
1590 	 */
1591 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1592 		(void)stop_ep_timer(ep);
1593 		abort_connection(ep, skb, GFP_KERNEL);
1594 		return;
1595 	}
1596 
1597 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1598 
1599 	/*
1600 	 * Copy the new data into our accumulation buffer.
1601 	 */
1602 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1603 				  skb->len);
1604 	ep->mpa_pkt_len += skb->len;
1605 
1606 	/*
1607 	 * If we don't even have the mpa message, then bail.
1608 	 * We'll continue process when more data arrives.
1609 	 */
1610 	if (ep->mpa_pkt_len < sizeof(*mpa))
1611 		return;
1612 
1613 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1614 	mpa = (struct mpa_message *) ep->mpa_pkt;
1615 
1616 	/*
1617 	 * Validate MPA Header.
1618 	 */
1619 	if (mpa->revision > mpa_rev) {
1620 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1621 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1622 		(void)stop_ep_timer(ep);
1623 		abort_connection(ep, skb, GFP_KERNEL);
1624 		return;
1625 	}
1626 
1627 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1628 		(void)stop_ep_timer(ep);
1629 		abort_connection(ep, skb, GFP_KERNEL);
1630 		return;
1631 	}
1632 
1633 	plen = ntohs(mpa->private_data_size);
1634 
1635 	/*
1636 	 * Fail if there's too much private data.
1637 	 */
1638 	if (plen > MPA_MAX_PRIVATE_DATA) {
1639 		(void)stop_ep_timer(ep);
1640 		abort_connection(ep, skb, GFP_KERNEL);
1641 		return;
1642 	}
1643 
1644 	/*
1645 	 * If plen does not account for pkt size
1646 	 */
1647 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1648 		(void)stop_ep_timer(ep);
1649 		abort_connection(ep, skb, GFP_KERNEL);
1650 		return;
1651 	}
1652 	ep->plen = (u8) plen;
1653 
1654 	/*
1655 	 * If we don't have all the pdata yet, then bail.
1656 	 */
1657 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1658 		return;
1659 
1660 	/*
1661 	 * If we get here we have accumulated the entire mpa
1662 	 * start reply message including private data.
1663 	 */
1664 	ep->mpa_attr.initiator = 0;
1665 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1666 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1667 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1668 	ep->mpa_attr.version = mpa->revision;
1669 	if (mpa->revision == 1)
1670 		ep->tried_with_mpa_v1 = 1;
1671 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1672 
1673 	if (mpa->revision == 2) {
1674 		ep->mpa_attr.enhanced_rdma_conn =
1675 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1676 		if (ep->mpa_attr.enhanced_rdma_conn) {
1677 			mpa_v2_params = (struct mpa_v2_conn_params *)
1678 				(ep->mpa_pkt + sizeof(*mpa));
1679 			ep->ird = ntohs(mpa_v2_params->ird) &
1680 				MPA_V2_IRD_ORD_MASK;
1681 			ep->ord = ntohs(mpa_v2_params->ord) &
1682 				MPA_V2_IRD_ORD_MASK;
1683 			PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1684 			     ep->ord);
1685 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1686 				if (peer2peer) {
1687 					if (ntohs(mpa_v2_params->ord) &
1688 							MPA_V2_RDMA_WRITE_RTR)
1689 						ep->mpa_attr.p2p_type =
1690 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1691 					else if (ntohs(mpa_v2_params->ord) &
1692 							MPA_V2_RDMA_READ_RTR)
1693 						ep->mpa_attr.p2p_type =
1694 						FW_RI_INIT_P2PTYPE_READ_REQ;
1695 				}
1696 		}
1697 	} else if (mpa->revision == 1)
1698 		if (peer2peer)
1699 			ep->mpa_attr.p2p_type = p2p_type;
1700 
1701 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1702 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1703 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1704 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1705 	     ep->mpa_attr.p2p_type);
1706 
1707 	/*
1708 	 * If the endpoint timer already expired, then we ignore
1709 	 * the start request.  process_timeout() will abort
1710 	 * the connection.
1711 	 */
1712 	if (!stop_ep_timer(ep)) {
1713 		__state_set(&ep->com, MPA_REQ_RCVD);
1714 
1715 		/* drive upcall */
1716 		mutex_lock_nested(&ep->parent_ep->com.mutex,
1717 				  SINGLE_DEPTH_NESTING);
1718 		if (ep->parent_ep->com.state != DEAD) {
1719 			if (connect_request_upcall(ep))
1720 				abort_connection(ep, skb, GFP_KERNEL);
1721 		} else {
1722 			abort_connection(ep, skb, GFP_KERNEL);
1723 		}
1724 		mutex_unlock(&ep->parent_ep->com.mutex);
1725 	}
1726 	return;
1727 }
1728 
1729 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1730 {
1731 	struct c4iw_ep *ep;
1732 	struct cpl_rx_data *hdr = cplhdr(skb);
1733 	unsigned int dlen = ntohs(hdr->len);
1734 	unsigned int tid = GET_TID(hdr);
1735 	struct tid_info *t = dev->rdev.lldi.tids;
1736 	__u8 status = hdr->status;
1737 	int disconnect = 0;
1738 
1739 	ep = lookup_tid(t, tid);
1740 	if (!ep)
1741 		return 0;
1742 	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1743 	skb_pull(skb, sizeof(*hdr));
1744 	skb_trim(skb, dlen);
1745 	mutex_lock(&ep->com.mutex);
1746 
1747 	/* update RX credits */
1748 	update_rx_credits(ep, dlen);
1749 
1750 	switch (ep->com.state) {
1751 	case MPA_REQ_SENT:
1752 		ep->rcv_seq += dlen;
1753 		disconnect = process_mpa_reply(ep, skb);
1754 		break;
1755 	case MPA_REQ_WAIT:
1756 		ep->rcv_seq += dlen;
1757 		process_mpa_request(ep, skb);
1758 		break;
1759 	case FPDU_MODE: {
1760 		struct c4iw_qp_attributes attrs;
1761 		BUG_ON(!ep->com.qp);
1762 		if (status)
1763 			pr_err("%s Unexpected streaming data." \
1764 			       " qpid %u ep %p state %d tid %u status %d\n",
1765 			       __func__, ep->com.qp->wq.sq.qid, ep,
1766 			       ep->com.state, ep->hwtid, status);
1767 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1768 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1769 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1770 		disconnect = 1;
1771 		break;
1772 	}
1773 	default:
1774 		break;
1775 	}
1776 	mutex_unlock(&ep->com.mutex);
1777 	if (disconnect)
1778 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1779 	return 0;
1780 }
1781 
1782 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1783 {
1784 	struct c4iw_ep *ep;
1785 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1786 	int release = 0;
1787 	unsigned int tid = GET_TID(rpl);
1788 	struct tid_info *t = dev->rdev.lldi.tids;
1789 
1790 	ep = lookup_tid(t, tid);
1791 	if (!ep) {
1792 		printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1793 		return 0;
1794 	}
1795 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1796 	mutex_lock(&ep->com.mutex);
1797 	switch (ep->com.state) {
1798 	case ABORTING:
1799 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1800 		__state_set(&ep->com, DEAD);
1801 		release = 1;
1802 		break;
1803 	default:
1804 		printk(KERN_ERR "%s ep %p state %d\n",
1805 		     __func__, ep, ep->com.state);
1806 		break;
1807 	}
1808 	mutex_unlock(&ep->com.mutex);
1809 
1810 	if (release)
1811 		release_ep_resources(ep);
1812 	return 0;
1813 }
1814 
1815 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1816 {
1817 	struct sk_buff *skb;
1818 	struct fw_ofld_connection_wr *req;
1819 	unsigned int mtu_idx;
1820 	int wscale;
1821 	struct sockaddr_in *sin;
1822 	int win;
1823 
1824 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1825 	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1826 	memset(req, 0, sizeof(*req));
1827 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1828 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1829 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1830 				     ep->com.dev->rdev.lldi.ports[0],
1831 				     ep->l2t));
1832 	sin = (struct sockaddr_in *)&ep->com.mapped_local_addr;
1833 	req->le.lport = sin->sin_port;
1834 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1835 	sin = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
1836 	req->le.pport = sin->sin_port;
1837 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1838 	req->tcb.t_state_to_astid =
1839 			htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1840 			FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1841 	req->tcb.cplrxdataack_cplpassacceptrpl =
1842 			htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1843 	req->tcb.tx_max = (__force __be32) jiffies;
1844 	req->tcb.rcv_adv = htons(1);
1845 	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1846 		 enable_tcp_timestamps,
1847 		 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
1848 	wscale = compute_wscale(rcv_win);
1849 
1850 	/*
1851 	 * Specify the largest window that will fit in opt0. The
1852 	 * remainder will be specified in the rx_data_ack.
1853 	 */
1854 	win = ep->rcv_win >> 10;
1855 	if (win > RCV_BUFSIZ_M)
1856 		win = RCV_BUFSIZ_M;
1857 
1858 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1859 		(nocong ? NO_CONG_F : 0) |
1860 		KEEP_ALIVE_F |
1861 		DELACK_F |
1862 		WND_SCALE_V(wscale) |
1863 		MSS_IDX_V(mtu_idx) |
1864 		L2T_IDX_V(ep->l2t->idx) |
1865 		TX_CHAN_V(ep->tx_chan) |
1866 		SMAC_SEL_V(ep->smac_idx) |
1867 		DSCP_V(ep->tos) |
1868 		ULP_MODE_V(ULP_MODE_TCPDDP) |
1869 		RCV_BUFSIZ_V(win));
1870 	req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1871 		TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1872 		RX_CHANNEL_V(0) |
1873 		CCTRL_ECN_V(enable_ecn) |
1874 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1875 	if (enable_tcp_timestamps)
1876 		req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1877 	if (enable_tcp_sack)
1878 		req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1879 	if (wscale && enable_tcp_window_scaling)
1880 		req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1881 	req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1882 	req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1883 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1884 	set_bit(ACT_OFLD_CONN, &ep->com.history);
1885 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1886 }
1887 
1888 /*
1889  * Return whether a failed active open has allocated a TID
1890  */
1891 static inline int act_open_has_tid(int status)
1892 {
1893 	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1894 	       status != CPL_ERR_ARP_MISS;
1895 }
1896 
1897 /* Returns whether a CPL status conveys negative advice.
1898  */
1899 static int is_neg_adv(unsigned int status)
1900 {
1901 	return status == CPL_ERR_RTX_NEG_ADVICE ||
1902 	       status == CPL_ERR_PERSIST_NEG_ADVICE ||
1903 	       status == CPL_ERR_KEEPALV_NEG_ADVICE;
1904 }
1905 
1906 static char *neg_adv_str(unsigned int status)
1907 {
1908 	switch (status) {
1909 	case CPL_ERR_RTX_NEG_ADVICE:
1910 		return "Retransmit timeout";
1911 	case CPL_ERR_PERSIST_NEG_ADVICE:
1912 		return "Persist timeout";
1913 	case CPL_ERR_KEEPALV_NEG_ADVICE:
1914 		return "Keepalive timeout";
1915 	default:
1916 		return "Unknown";
1917 	}
1918 }
1919 
1920 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1921 {
1922 	ep->snd_win = snd_win;
1923 	ep->rcv_win = rcv_win;
1924 	PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1925 }
1926 
1927 #define ACT_OPEN_RETRY_COUNT 2
1928 
1929 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1930 		     struct dst_entry *dst, struct c4iw_dev *cdev,
1931 		     bool clear_mpa_v1, enum chip_type adapter_type)
1932 {
1933 	struct neighbour *n;
1934 	int err, step;
1935 	struct net_device *pdev;
1936 
1937 	n = dst_neigh_lookup(dst, peer_ip);
1938 	if (!n)
1939 		return -ENODEV;
1940 
1941 	rcu_read_lock();
1942 	err = -ENOMEM;
1943 	if (n->dev->flags & IFF_LOOPBACK) {
1944 		if (iptype == 4)
1945 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1946 		else if (IS_ENABLED(CONFIG_IPV6))
1947 			for_each_netdev(&init_net, pdev) {
1948 				if (ipv6_chk_addr(&init_net,
1949 						  (struct in6_addr *)peer_ip,
1950 						  pdev, 1))
1951 					break;
1952 			}
1953 		else
1954 			pdev = NULL;
1955 
1956 		if (!pdev) {
1957 			err = -ENODEV;
1958 			goto out;
1959 		}
1960 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1961 					n, pdev, 0);
1962 		if (!ep->l2t)
1963 			goto out;
1964 		ep->mtu = pdev->mtu;
1965 		ep->tx_chan = cxgb4_port_chan(pdev);
1966 		ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
1967 						cxgb4_port_viid(pdev));
1968 		step = cdev->rdev.lldi.ntxq /
1969 			cdev->rdev.lldi.nchan;
1970 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1971 		step = cdev->rdev.lldi.nrxq /
1972 			cdev->rdev.lldi.nchan;
1973 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1974 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1975 			cxgb4_port_idx(pdev) * step];
1976 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
1977 		dev_put(pdev);
1978 	} else {
1979 		pdev = get_real_dev(n->dev);
1980 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1981 					n, pdev, 0);
1982 		if (!ep->l2t)
1983 			goto out;
1984 		ep->mtu = dst_mtu(dst);
1985 		ep->tx_chan = cxgb4_port_chan(pdev);
1986 		ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
1987 						cxgb4_port_viid(pdev));
1988 		step = cdev->rdev.lldi.ntxq /
1989 			cdev->rdev.lldi.nchan;
1990 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1991 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1992 		step = cdev->rdev.lldi.nrxq /
1993 			cdev->rdev.lldi.nchan;
1994 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1995 			cxgb4_port_idx(pdev) * step];
1996 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
1997 
1998 		if (clear_mpa_v1) {
1999 			ep->retry_with_mpa_v1 = 0;
2000 			ep->tried_with_mpa_v1 = 0;
2001 		}
2002 	}
2003 	err = 0;
2004 out:
2005 	rcu_read_unlock();
2006 
2007 	neigh_release(n);
2008 
2009 	return err;
2010 }
2011 
2012 static int c4iw_reconnect(struct c4iw_ep *ep)
2013 {
2014 	int err = 0;
2015 	struct sockaddr_in *laddr = (struct sockaddr_in *)
2016 				    &ep->com.cm_id->local_addr;
2017 	struct sockaddr_in *raddr = (struct sockaddr_in *)
2018 				    &ep->com.cm_id->remote_addr;
2019 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2020 				      &ep->com.cm_id->local_addr;
2021 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2022 				      &ep->com.cm_id->remote_addr;
2023 	int iptype;
2024 	__u8 *ra;
2025 
2026 	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2027 	init_timer(&ep->timer);
2028 
2029 	/*
2030 	 * Allocate an active TID to initiate a TCP connection.
2031 	 */
2032 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2033 	if (ep->atid == -1) {
2034 		pr_err("%s - cannot alloc atid.\n", __func__);
2035 		err = -ENOMEM;
2036 		goto fail2;
2037 	}
2038 	insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2039 
2040 	/* find a route */
2041 	if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
2042 		ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
2043 				     raddr->sin_addr.s_addr, laddr->sin_port,
2044 				     raddr->sin_port, 0);
2045 		iptype = 4;
2046 		ra = (__u8 *)&raddr->sin_addr;
2047 	} else {
2048 		ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
2049 				      raddr6->sin6_addr.s6_addr,
2050 				      laddr6->sin6_port, raddr6->sin6_port, 0,
2051 				      raddr6->sin6_scope_id);
2052 		iptype = 6;
2053 		ra = (__u8 *)&raddr6->sin6_addr;
2054 	}
2055 	if (!ep->dst) {
2056 		pr_err("%s - cannot find route.\n", __func__);
2057 		err = -EHOSTUNREACH;
2058 		goto fail3;
2059 	}
2060 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2061 			ep->com.dev->rdev.lldi.adapter_type);
2062 	if (err) {
2063 		pr_err("%s - cannot alloc l2e.\n", __func__);
2064 		goto fail4;
2065 	}
2066 
2067 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2068 	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2069 	     ep->l2t->idx);
2070 
2071 	state_set(&ep->com, CONNECTING);
2072 	ep->tos = 0;
2073 
2074 	/* send connect request to rnic */
2075 	err = send_connect(ep);
2076 	if (!err)
2077 		goto out;
2078 
2079 	cxgb4_l2t_release(ep->l2t);
2080 fail4:
2081 	dst_release(ep->dst);
2082 fail3:
2083 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2084 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2085 fail2:
2086 	/*
2087 	 * remember to send notification to upper layer.
2088 	 * We are in here so the upper layer is not aware that this is
2089 	 * re-connect attempt and so, upper layer is still waiting for
2090 	 * response of 1st connect request.
2091 	 */
2092 	connect_reply_upcall(ep, -ECONNRESET);
2093 	c4iw_put_ep(&ep->com);
2094 out:
2095 	return err;
2096 }
2097 
2098 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2099 {
2100 	struct c4iw_ep *ep;
2101 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
2102 	unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2103 				      ntohl(rpl->atid_status)));
2104 	struct tid_info *t = dev->rdev.lldi.tids;
2105 	int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2106 	struct sockaddr_in *la;
2107 	struct sockaddr_in *ra;
2108 	struct sockaddr_in6 *la6;
2109 	struct sockaddr_in6 *ra6;
2110 
2111 	ep = lookup_atid(t, atid);
2112 	la = (struct sockaddr_in *)&ep->com.mapped_local_addr;
2113 	ra = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
2114 	la6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
2115 	ra6 = (struct sockaddr_in6 *)&ep->com.mapped_remote_addr;
2116 
2117 	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2118 	     status, status2errno(status));
2119 
2120 	if (is_neg_adv(status)) {
2121 		PDBG("%s Connection problems for atid %u status %u (%s)\n",
2122 		     __func__, atid, status, neg_adv_str(status));
2123 		ep->stats.connect_neg_adv++;
2124 		mutex_lock(&dev->rdev.stats.lock);
2125 		dev->rdev.stats.neg_adv++;
2126 		mutex_unlock(&dev->rdev.stats.lock);
2127 		return 0;
2128 	}
2129 
2130 	set_bit(ACT_OPEN_RPL, &ep->com.history);
2131 
2132 	/*
2133 	 * Log interesting failures.
2134 	 */
2135 	switch (status) {
2136 	case CPL_ERR_CONN_RESET:
2137 	case CPL_ERR_CONN_TIMEDOUT:
2138 		break;
2139 	case CPL_ERR_TCAM_FULL:
2140 		mutex_lock(&dev->rdev.stats.lock);
2141 		dev->rdev.stats.tcam_full++;
2142 		mutex_unlock(&dev->rdev.stats.lock);
2143 		if (ep->com.local_addr.ss_family == AF_INET &&
2144 		    dev->rdev.lldi.enable_fw_ofld_conn) {
2145 			send_fw_act_open_req(ep,
2146 					     TID_TID_G(AOPEN_ATID_G(
2147 					     ntohl(rpl->atid_status))));
2148 			return 0;
2149 		}
2150 		break;
2151 	case CPL_ERR_CONN_EXIST:
2152 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2153 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
2154 			if (ep->com.remote_addr.ss_family == AF_INET6) {
2155 				struct sockaddr_in6 *sin6 =
2156 						(struct sockaddr_in6 *)
2157 						&ep->com.mapped_local_addr;
2158 				cxgb4_clip_release(
2159 						ep->com.dev->rdev.lldi.ports[0],
2160 						(const u32 *)
2161 						&sin6->sin6_addr.s6_addr, 1);
2162 			}
2163 			remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2164 					atid);
2165 			cxgb4_free_atid(t, atid);
2166 			dst_release(ep->dst);
2167 			cxgb4_l2t_release(ep->l2t);
2168 			c4iw_reconnect(ep);
2169 			return 0;
2170 		}
2171 		break;
2172 	default:
2173 		if (ep->com.local_addr.ss_family == AF_INET) {
2174 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2175 				atid, status, status2errno(status),
2176 				&la->sin_addr.s_addr, ntohs(la->sin_port),
2177 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
2178 		} else {
2179 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2180 				atid, status, status2errno(status),
2181 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2182 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2183 		}
2184 		break;
2185 	}
2186 
2187 	connect_reply_upcall(ep, status2errno(status));
2188 	state_set(&ep->com, DEAD);
2189 
2190 	if (ep->com.remote_addr.ss_family == AF_INET6) {
2191 		struct sockaddr_in6 *sin6 =
2192 			(struct sockaddr_in6 *)&ep->com.mapped_local_addr;
2193 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2194 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2195 	}
2196 	if (status && act_open_has_tid(status))
2197 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2198 
2199 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2200 	cxgb4_free_atid(t, atid);
2201 	dst_release(ep->dst);
2202 	cxgb4_l2t_release(ep->l2t);
2203 	c4iw_put_ep(&ep->com);
2204 
2205 	return 0;
2206 }
2207 
2208 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2209 {
2210 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2211 	struct tid_info *t = dev->rdev.lldi.tids;
2212 	unsigned int stid = GET_TID(rpl);
2213 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
2214 
2215 	if (!ep) {
2216 		PDBG("%s stid %d lookup failure!\n", __func__, stid);
2217 		goto out;
2218 	}
2219 	PDBG("%s ep %p status %d error %d\n", __func__, ep,
2220 	     rpl->status, status2errno(rpl->status));
2221 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2222 
2223 out:
2224 	return 0;
2225 }
2226 
2227 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2228 {
2229 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2230 	struct tid_info *t = dev->rdev.lldi.tids;
2231 	unsigned int stid = GET_TID(rpl);
2232 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
2233 
2234 	PDBG("%s ep %p\n", __func__, ep);
2235 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2236 	return 0;
2237 }
2238 
2239 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2240 		      struct cpl_pass_accept_req *req)
2241 {
2242 	struct cpl_pass_accept_rpl *rpl;
2243 	unsigned int mtu_idx;
2244 	u64 opt0;
2245 	u32 opt2;
2246 	int wscale;
2247 	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2248 	int win;
2249 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2250 
2251 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2252 	BUG_ON(skb_cloned(skb));
2253 
2254 	skb_get(skb);
2255 	rpl = cplhdr(skb);
2256 	if (!is_t4(adapter_type)) {
2257 		skb_trim(skb, roundup(sizeof(*rpl5), 16));
2258 		rpl5 = (void *)rpl;
2259 		INIT_TP_WR(rpl5, ep->hwtid);
2260 	} else {
2261 		skb_trim(skb, sizeof(*rpl));
2262 		INIT_TP_WR(rpl, ep->hwtid);
2263 	}
2264 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2265 						    ep->hwtid));
2266 
2267 	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2268 		 enable_tcp_timestamps && req->tcpopt.tstamp,
2269 		 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2270 	wscale = compute_wscale(rcv_win);
2271 
2272 	/*
2273 	 * Specify the largest window that will fit in opt0. The
2274 	 * remainder will be specified in the rx_data_ack.
2275 	 */
2276 	win = ep->rcv_win >> 10;
2277 	if (win > RCV_BUFSIZ_M)
2278 		win = RCV_BUFSIZ_M;
2279 	opt0 = (nocong ? NO_CONG_F : 0) |
2280 	       KEEP_ALIVE_F |
2281 	       DELACK_F |
2282 	       WND_SCALE_V(wscale) |
2283 	       MSS_IDX_V(mtu_idx) |
2284 	       L2T_IDX_V(ep->l2t->idx) |
2285 	       TX_CHAN_V(ep->tx_chan) |
2286 	       SMAC_SEL_V(ep->smac_idx) |
2287 	       DSCP_V(ep->tos >> 2) |
2288 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
2289 	       RCV_BUFSIZ_V(win);
2290 	opt2 = RX_CHANNEL_V(0) |
2291 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2292 
2293 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2294 		opt2 |= TSTAMPS_EN_F;
2295 	if (enable_tcp_sack && req->tcpopt.sack)
2296 		opt2 |= SACK_EN_F;
2297 	if (wscale && enable_tcp_window_scaling)
2298 		opt2 |= WND_SCALE_EN_F;
2299 	if (enable_ecn) {
2300 		const struct tcphdr *tcph;
2301 		u32 hlen = ntohl(req->hdr_len);
2302 
2303 		if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2304 			tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2305 				IP_HDR_LEN_G(hlen);
2306 		else
2307 			tcph = (const void *)(req + 1) +
2308 				T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2309 		if (tcph->ece && tcph->cwr)
2310 			opt2 |= CCTRL_ECN_V(1);
2311 	}
2312 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2313 		u32 isn = (prandom_u32() & ~7UL) - 1;
2314 		opt2 |= T5_OPT_2_VALID_F;
2315 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2316 		opt2 |= T5_ISS_F;
2317 		rpl5 = (void *)rpl;
2318 		memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2319 		if (peer2peer)
2320 			isn += 4;
2321 		rpl5->iss = cpu_to_be32(isn);
2322 		PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2323 	}
2324 
2325 	rpl->opt0 = cpu_to_be64(opt0);
2326 	rpl->opt2 = cpu_to_be32(opt2);
2327 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2328 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
2329 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2330 
2331 	return;
2332 }
2333 
2334 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2335 {
2336 	PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2337 	BUG_ON(skb_cloned(skb));
2338 	skb_trim(skb, sizeof(struct cpl_tid_release));
2339 	release_tid(&dev->rdev, hwtid, skb);
2340 	return;
2341 }
2342 
2343 static void get_4tuple(struct cpl_pass_accept_req *req, enum chip_type type,
2344 		       int *iptype, __u8 *local_ip, __u8 *peer_ip,
2345 		       __be16 *local_port, __be16 *peer_port)
2346 {
2347 	int eth_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2348 		      ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2349 		      T6_ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2350 	int ip_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2351 		     IP_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2352 		     T6_IP_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2353 	struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2354 	struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2355 	struct tcphdr *tcp = (struct tcphdr *)
2356 			     ((u8 *)(req + 1) + eth_len + ip_len);
2357 
2358 	if (ip->version == 4) {
2359 		PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2360 		     ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2361 		     ntohs(tcp->dest));
2362 		*iptype = 4;
2363 		memcpy(peer_ip, &ip->saddr, 4);
2364 		memcpy(local_ip, &ip->daddr, 4);
2365 	} else {
2366 		PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2367 		     ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2368 		     ntohs(tcp->dest));
2369 		*iptype = 6;
2370 		memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2371 		memcpy(local_ip, ip6->daddr.s6_addr, 16);
2372 	}
2373 	*peer_port = tcp->source;
2374 	*local_port = tcp->dest;
2375 
2376 	return;
2377 }
2378 
2379 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2380 {
2381 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2382 	struct cpl_pass_accept_req *req = cplhdr(skb);
2383 	unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2384 	struct tid_info *t = dev->rdev.lldi.tids;
2385 	unsigned int hwtid = GET_TID(req);
2386 	struct dst_entry *dst;
2387 	__u8 local_ip[16], peer_ip[16];
2388 	__be16 local_port, peer_port;
2389 	struct sockaddr_in6 *sin6;
2390 	int err;
2391 	u16 peer_mss = ntohs(req->tcpopt.mss);
2392 	int iptype;
2393 	unsigned short hdrs;
2394 
2395 	parent_ep = lookup_stid(t, stid);
2396 	if (!parent_ep) {
2397 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2398 		goto reject;
2399 	}
2400 
2401 	if (state_read(&parent_ep->com) != LISTEN) {
2402 		printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2403 		       __func__);
2404 		goto reject;
2405 	}
2406 
2407 	get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, &iptype,
2408 		   local_ip, peer_ip, &local_port, &peer_port);
2409 
2410 	/* Find output route */
2411 	if (iptype == 4)  {
2412 		PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2413 		     , __func__, parent_ep, hwtid,
2414 		     local_ip, peer_ip, ntohs(local_port),
2415 		     ntohs(peer_port), peer_mss);
2416 		dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2417 				 local_port, peer_port,
2418 				 PASS_OPEN_TOS_G(ntohl(req->tos_stid)));
2419 	} else {
2420 		PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2421 		     , __func__, parent_ep, hwtid,
2422 		     local_ip, peer_ip, ntohs(local_port),
2423 		     ntohs(peer_port), peer_mss);
2424 		dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2425 				  PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2426 				  ((struct sockaddr_in6 *)
2427 				  &parent_ep->com.local_addr)->sin6_scope_id);
2428 	}
2429 	if (!dst) {
2430 		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2431 		       __func__);
2432 		goto reject;
2433 	}
2434 
2435 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2436 	if (!child_ep) {
2437 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2438 		       __func__);
2439 		dst_release(dst);
2440 		goto reject;
2441 	}
2442 
2443 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2444 			parent_ep->com.dev->rdev.lldi.adapter_type);
2445 	if (err) {
2446 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2447 		       __func__);
2448 		dst_release(dst);
2449 		kfree(child_ep);
2450 		goto reject;
2451 	}
2452 
2453 	hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2454 	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2455 	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2456 		child_ep->mtu = peer_mss + hdrs;
2457 
2458 	state_set(&child_ep->com, CONNECTING);
2459 	child_ep->com.dev = dev;
2460 	child_ep->com.cm_id = NULL;
2461 
2462 	/*
2463 	 * The mapped_local and mapped_remote addresses get setup with
2464 	 * the actual 4-tuple.  The local address will be based on the
2465 	 * actual local address of the connection, but on the port number
2466 	 * of the parent listening endpoint.  The remote address is
2467 	 * setup based on a query to the IWPM since we don't know what it
2468 	 * originally was before mapping.  If no mapping was done, then
2469 	 * mapped_remote == remote, and mapped_local == local.
2470 	 */
2471 	if (iptype == 4) {
2472 		struct sockaddr_in *sin = (struct sockaddr_in *)
2473 			&child_ep->com.mapped_local_addr;
2474 
2475 		sin->sin_family = PF_INET;
2476 		sin->sin_port = local_port;
2477 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2478 
2479 		sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2480 		sin->sin_family = PF_INET;
2481 		sin->sin_port = ((struct sockaddr_in *)
2482 				 &parent_ep->com.local_addr)->sin_port;
2483 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2484 
2485 		sin = (struct sockaddr_in *)&child_ep->com.mapped_remote_addr;
2486 		sin->sin_family = PF_INET;
2487 		sin->sin_port = peer_port;
2488 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2489 	} else {
2490 		sin6 = (struct sockaddr_in6 *)&child_ep->com.mapped_local_addr;
2491 		sin6->sin6_family = PF_INET6;
2492 		sin6->sin6_port = local_port;
2493 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2494 
2495 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2496 		sin6->sin6_family = PF_INET6;
2497 		sin6->sin6_port = ((struct sockaddr_in6 *)
2498 				   &parent_ep->com.local_addr)->sin6_port;
2499 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2500 
2501 		sin6 = (struct sockaddr_in6 *)&child_ep->com.mapped_remote_addr;
2502 		sin6->sin6_family = PF_INET6;
2503 		sin6->sin6_port = peer_port;
2504 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2505 	}
2506 	memcpy(&child_ep->com.remote_addr, &child_ep->com.mapped_remote_addr,
2507 	       sizeof(child_ep->com.remote_addr));
2508 	get_remote_addr(parent_ep, child_ep);
2509 
2510 	c4iw_get_ep(&parent_ep->com);
2511 	child_ep->parent_ep = parent_ep;
2512 	child_ep->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2513 	child_ep->dst = dst;
2514 	child_ep->hwtid = hwtid;
2515 
2516 	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2517 	     child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2518 
2519 	init_timer(&child_ep->timer);
2520 	cxgb4_insert_tid(t, child_ep, hwtid);
2521 	insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2522 	accept_cr(child_ep, skb, req);
2523 	set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2524 	if (iptype == 6) {
2525 		sin6 = (struct sockaddr_in6 *)&child_ep->com.mapped_local_addr;
2526 		cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2527 			       (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2528 	}
2529 	goto out;
2530 reject:
2531 	reject_cr(dev, hwtid, skb);
2532 out:
2533 	return 0;
2534 }
2535 
2536 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2537 {
2538 	struct c4iw_ep *ep;
2539 	struct cpl_pass_establish *req = cplhdr(skb);
2540 	struct tid_info *t = dev->rdev.lldi.tids;
2541 	unsigned int tid = GET_TID(req);
2542 
2543 	ep = lookup_tid(t, tid);
2544 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2545 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2546 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2547 
2548 	PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2549 	     ntohs(req->tcp_opt));
2550 
2551 	set_emss(ep, ntohs(req->tcp_opt));
2552 
2553 	dst_confirm(ep->dst);
2554 	state_set(&ep->com, MPA_REQ_WAIT);
2555 	start_ep_timer(ep);
2556 	send_flowc(ep, skb);
2557 	set_bit(PASS_ESTAB, &ep->com.history);
2558 
2559 	return 0;
2560 }
2561 
2562 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2563 {
2564 	struct cpl_peer_close *hdr = cplhdr(skb);
2565 	struct c4iw_ep *ep;
2566 	struct c4iw_qp_attributes attrs;
2567 	int disconnect = 1;
2568 	int release = 0;
2569 	struct tid_info *t = dev->rdev.lldi.tids;
2570 	unsigned int tid = GET_TID(hdr);
2571 	int ret;
2572 
2573 	ep = lookup_tid(t, tid);
2574 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2575 	dst_confirm(ep->dst);
2576 
2577 	set_bit(PEER_CLOSE, &ep->com.history);
2578 	mutex_lock(&ep->com.mutex);
2579 	switch (ep->com.state) {
2580 	case MPA_REQ_WAIT:
2581 		__state_set(&ep->com, CLOSING);
2582 		break;
2583 	case MPA_REQ_SENT:
2584 		__state_set(&ep->com, CLOSING);
2585 		connect_reply_upcall(ep, -ECONNRESET);
2586 		break;
2587 	case MPA_REQ_RCVD:
2588 
2589 		/*
2590 		 * We're gonna mark this puppy DEAD, but keep
2591 		 * the reference on it until the ULP accepts or
2592 		 * rejects the CR. Also wake up anyone waiting
2593 		 * in rdma connection migration (see c4iw_accept_cr()).
2594 		 */
2595 		__state_set(&ep->com, CLOSING);
2596 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2597 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2598 		break;
2599 	case MPA_REP_SENT:
2600 		__state_set(&ep->com, CLOSING);
2601 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2602 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2603 		break;
2604 	case FPDU_MODE:
2605 		start_ep_timer(ep);
2606 		__state_set(&ep->com, CLOSING);
2607 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2608 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2609 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2610 		if (ret != -ECONNRESET) {
2611 			peer_close_upcall(ep);
2612 			disconnect = 1;
2613 		}
2614 		break;
2615 	case ABORTING:
2616 		disconnect = 0;
2617 		break;
2618 	case CLOSING:
2619 		__state_set(&ep->com, MORIBUND);
2620 		disconnect = 0;
2621 		break;
2622 	case MORIBUND:
2623 		(void)stop_ep_timer(ep);
2624 		if (ep->com.cm_id && ep->com.qp) {
2625 			attrs.next_state = C4IW_QP_STATE_IDLE;
2626 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2627 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2628 		}
2629 		close_complete_upcall(ep, 0);
2630 		__state_set(&ep->com, DEAD);
2631 		release = 1;
2632 		disconnect = 0;
2633 		break;
2634 	case DEAD:
2635 		disconnect = 0;
2636 		break;
2637 	default:
2638 		BUG_ON(1);
2639 	}
2640 	mutex_unlock(&ep->com.mutex);
2641 	if (disconnect)
2642 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2643 	if (release)
2644 		release_ep_resources(ep);
2645 	return 0;
2646 }
2647 
2648 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2649 {
2650 	struct cpl_abort_req_rss *req = cplhdr(skb);
2651 	struct c4iw_ep *ep;
2652 	struct cpl_abort_rpl *rpl;
2653 	struct sk_buff *rpl_skb;
2654 	struct c4iw_qp_attributes attrs;
2655 	int ret;
2656 	int release = 0;
2657 	struct tid_info *t = dev->rdev.lldi.tids;
2658 	unsigned int tid = GET_TID(req);
2659 
2660 	ep = lookup_tid(t, tid);
2661 	if (is_neg_adv(req->status)) {
2662 		PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2663 		     __func__, ep->hwtid, req->status,
2664 		     neg_adv_str(req->status));
2665 		ep->stats.abort_neg_adv++;
2666 		mutex_lock(&dev->rdev.stats.lock);
2667 		dev->rdev.stats.neg_adv++;
2668 		mutex_unlock(&dev->rdev.stats.lock);
2669 		return 0;
2670 	}
2671 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2672 	     ep->com.state);
2673 	set_bit(PEER_ABORT, &ep->com.history);
2674 
2675 	/*
2676 	 * Wake up any threads in rdma_init() or rdma_fini().
2677 	 * However, this is not needed if com state is just
2678 	 * MPA_REQ_SENT
2679 	 */
2680 	if (ep->com.state != MPA_REQ_SENT)
2681 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2682 
2683 	mutex_lock(&ep->com.mutex);
2684 	switch (ep->com.state) {
2685 	case CONNECTING:
2686 		break;
2687 	case MPA_REQ_WAIT:
2688 		(void)stop_ep_timer(ep);
2689 		break;
2690 	case MPA_REQ_SENT:
2691 		(void)stop_ep_timer(ep);
2692 		if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2693 			connect_reply_upcall(ep, -ECONNRESET);
2694 		else {
2695 			/*
2696 			 * we just don't send notification upwards because we
2697 			 * want to retry with mpa_v1 without upper layers even
2698 			 * knowing it.
2699 			 *
2700 			 * do some housekeeping so as to re-initiate the
2701 			 * connection
2702 			 */
2703 			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2704 			     mpa_rev);
2705 			ep->retry_with_mpa_v1 = 1;
2706 		}
2707 		break;
2708 	case MPA_REP_SENT:
2709 		break;
2710 	case MPA_REQ_RCVD:
2711 		break;
2712 	case MORIBUND:
2713 	case CLOSING:
2714 		stop_ep_timer(ep);
2715 		/*FALLTHROUGH*/
2716 	case FPDU_MODE:
2717 		if (ep->com.cm_id && ep->com.qp) {
2718 			attrs.next_state = C4IW_QP_STATE_ERROR;
2719 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2720 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2721 				     &attrs, 1);
2722 			if (ret)
2723 				printk(KERN_ERR MOD
2724 				       "%s - qp <- error failed!\n",
2725 				       __func__);
2726 		}
2727 		peer_abort_upcall(ep);
2728 		break;
2729 	case ABORTING:
2730 		break;
2731 	case DEAD:
2732 		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2733 		mutex_unlock(&ep->com.mutex);
2734 		return 0;
2735 	default:
2736 		BUG_ON(1);
2737 		break;
2738 	}
2739 	dst_confirm(ep->dst);
2740 	if (ep->com.state != ABORTING) {
2741 		__state_set(&ep->com, DEAD);
2742 		/* we don't release if we want to retry with mpa_v1 */
2743 		if (!ep->retry_with_mpa_v1)
2744 			release = 1;
2745 	}
2746 	mutex_unlock(&ep->com.mutex);
2747 
2748 	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2749 	if (!rpl_skb) {
2750 		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2751 		       __func__);
2752 		release = 1;
2753 		goto out;
2754 	}
2755 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2756 	rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2757 	INIT_TP_WR(rpl, ep->hwtid);
2758 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2759 	rpl->cmd = CPL_ABORT_NO_RST;
2760 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2761 out:
2762 	if (release)
2763 		release_ep_resources(ep);
2764 	else if (ep->retry_with_mpa_v1) {
2765 		if (ep->com.remote_addr.ss_family == AF_INET6) {
2766 			struct sockaddr_in6 *sin6 =
2767 					(struct sockaddr_in6 *)
2768 					&ep->com.mapped_local_addr;
2769 			cxgb4_clip_release(
2770 					ep->com.dev->rdev.lldi.ports[0],
2771 					(const u32 *)&sin6->sin6_addr.s6_addr,
2772 					1);
2773 		}
2774 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2775 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2776 		dst_release(ep->dst);
2777 		cxgb4_l2t_release(ep->l2t);
2778 		c4iw_reconnect(ep);
2779 	}
2780 
2781 	return 0;
2782 }
2783 
2784 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2785 {
2786 	struct c4iw_ep *ep;
2787 	struct c4iw_qp_attributes attrs;
2788 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2789 	int release = 0;
2790 	struct tid_info *t = dev->rdev.lldi.tids;
2791 	unsigned int tid = GET_TID(rpl);
2792 
2793 	ep = lookup_tid(t, tid);
2794 
2795 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2796 	BUG_ON(!ep);
2797 
2798 	/* The cm_id may be null if we failed to connect */
2799 	mutex_lock(&ep->com.mutex);
2800 	switch (ep->com.state) {
2801 	case CLOSING:
2802 		__state_set(&ep->com, MORIBUND);
2803 		break;
2804 	case MORIBUND:
2805 		(void)stop_ep_timer(ep);
2806 		if ((ep->com.cm_id) && (ep->com.qp)) {
2807 			attrs.next_state = C4IW_QP_STATE_IDLE;
2808 			c4iw_modify_qp(ep->com.qp->rhp,
2809 					     ep->com.qp,
2810 					     C4IW_QP_ATTR_NEXT_STATE,
2811 					     &attrs, 1);
2812 		}
2813 		close_complete_upcall(ep, 0);
2814 		__state_set(&ep->com, DEAD);
2815 		release = 1;
2816 		break;
2817 	case ABORTING:
2818 	case DEAD:
2819 		break;
2820 	default:
2821 		BUG_ON(1);
2822 		break;
2823 	}
2824 	mutex_unlock(&ep->com.mutex);
2825 	if (release)
2826 		release_ep_resources(ep);
2827 	return 0;
2828 }
2829 
2830 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2831 {
2832 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
2833 	struct tid_info *t = dev->rdev.lldi.tids;
2834 	unsigned int tid = GET_TID(rpl);
2835 	struct c4iw_ep *ep;
2836 	struct c4iw_qp_attributes attrs;
2837 
2838 	ep = lookup_tid(t, tid);
2839 	BUG_ON(!ep);
2840 
2841 	if (ep && ep->com.qp) {
2842 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2843 		       ep->com.qp->wq.sq.qid);
2844 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2845 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2846 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2847 	} else
2848 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2849 
2850 	return 0;
2851 }
2852 
2853 /*
2854  * Upcall from the adapter indicating data has been transmitted.
2855  * For us its just the single MPA request or reply.  We can now free
2856  * the skb holding the mpa message.
2857  */
2858 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2859 {
2860 	struct c4iw_ep *ep;
2861 	struct cpl_fw4_ack *hdr = cplhdr(skb);
2862 	u8 credits = hdr->credits;
2863 	unsigned int tid = GET_TID(hdr);
2864 	struct tid_info *t = dev->rdev.lldi.tids;
2865 
2866 
2867 	ep = lookup_tid(t, tid);
2868 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2869 	if (credits == 0) {
2870 		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2871 		     __func__, ep, ep->hwtid, state_read(&ep->com));
2872 		return 0;
2873 	}
2874 
2875 	dst_confirm(ep->dst);
2876 	if (ep->mpa_skb) {
2877 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
2878 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2879 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2880 		kfree_skb(ep->mpa_skb);
2881 		ep->mpa_skb = NULL;
2882 	}
2883 	return 0;
2884 }
2885 
2886 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2887 {
2888 	int err = 0;
2889 	int disconnect = 0;
2890 	struct c4iw_ep *ep = to_ep(cm_id);
2891 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2892 
2893 	mutex_lock(&ep->com.mutex);
2894 	if (ep->com.state == DEAD) {
2895 		mutex_unlock(&ep->com.mutex);
2896 		c4iw_put_ep(&ep->com);
2897 		return -ECONNRESET;
2898 	}
2899 	set_bit(ULP_REJECT, &ep->com.history);
2900 	BUG_ON(ep->com.state != MPA_REQ_RCVD);
2901 	if (mpa_rev == 0)
2902 		abort_connection(ep, NULL, GFP_KERNEL);
2903 	else {
2904 		err = send_mpa_reject(ep, pdata, pdata_len);
2905 		disconnect = 1;
2906 	}
2907 	mutex_unlock(&ep->com.mutex);
2908 	if (disconnect)
2909 		err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2910 	c4iw_put_ep(&ep->com);
2911 	return 0;
2912 }
2913 
2914 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2915 {
2916 	int err;
2917 	struct c4iw_qp_attributes attrs;
2918 	enum c4iw_qp_attr_mask mask;
2919 	struct c4iw_ep *ep = to_ep(cm_id);
2920 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2921 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2922 
2923 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2924 
2925 	mutex_lock(&ep->com.mutex);
2926 	if (ep->com.state == DEAD) {
2927 		err = -ECONNRESET;
2928 		goto err;
2929 	}
2930 
2931 	BUG_ON(ep->com.state != MPA_REQ_RCVD);
2932 	BUG_ON(!qp);
2933 
2934 	set_bit(ULP_ACCEPT, &ep->com.history);
2935 	if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
2936 	    (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
2937 		abort_connection(ep, NULL, GFP_KERNEL);
2938 		err = -EINVAL;
2939 		goto err;
2940 	}
2941 
2942 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2943 		if (conn_param->ord > ep->ird) {
2944 			if (RELAXED_IRD_NEGOTIATION) {
2945 				ep->ord = ep->ird;
2946 			} else {
2947 				ep->ird = conn_param->ird;
2948 				ep->ord = conn_param->ord;
2949 				send_mpa_reject(ep, conn_param->private_data,
2950 						conn_param->private_data_len);
2951 				abort_connection(ep, NULL, GFP_KERNEL);
2952 				err = -ENOMEM;
2953 				goto err;
2954 			}
2955 		}
2956 		if (conn_param->ird < ep->ord) {
2957 			if (RELAXED_IRD_NEGOTIATION &&
2958 			    ep->ord <= h->rdev.lldi.max_ordird_qp) {
2959 				conn_param->ird = ep->ord;
2960 			} else {
2961 				abort_connection(ep, NULL, GFP_KERNEL);
2962 				err = -ENOMEM;
2963 				goto err;
2964 			}
2965 		}
2966 	}
2967 	ep->ird = conn_param->ird;
2968 	ep->ord = conn_param->ord;
2969 
2970 	if (ep->mpa_attr.version == 1) {
2971 		if (peer2peer && ep->ird == 0)
2972 			ep->ird = 1;
2973 	} else {
2974 		if (peer2peer &&
2975 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
2976 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
2977 			ep->ird = 1;
2978 	}
2979 
2980 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2981 
2982 	cm_id->add_ref(cm_id);
2983 	ep->com.cm_id = cm_id;
2984 	ep->com.qp = qp;
2985 	ref_qp(ep);
2986 
2987 	/* bind QP to EP and move to RTS */
2988 	attrs.mpa_attr = ep->mpa_attr;
2989 	attrs.max_ird = ep->ird;
2990 	attrs.max_ord = ep->ord;
2991 	attrs.llp_stream_handle = ep;
2992 	attrs.next_state = C4IW_QP_STATE_RTS;
2993 
2994 	/* bind QP and TID with INIT_WR */
2995 	mask = C4IW_QP_ATTR_NEXT_STATE |
2996 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2997 			     C4IW_QP_ATTR_MPA_ATTR |
2998 			     C4IW_QP_ATTR_MAX_IRD |
2999 			     C4IW_QP_ATTR_MAX_ORD;
3000 
3001 	err = c4iw_modify_qp(ep->com.qp->rhp,
3002 			     ep->com.qp, mask, &attrs, 1);
3003 	if (err)
3004 		goto err1;
3005 	err = send_mpa_reply(ep, conn_param->private_data,
3006 			     conn_param->private_data_len);
3007 	if (err)
3008 		goto err1;
3009 
3010 	__state_set(&ep->com, FPDU_MODE);
3011 	established_upcall(ep);
3012 	mutex_unlock(&ep->com.mutex);
3013 	c4iw_put_ep(&ep->com);
3014 	return 0;
3015 err1:
3016 	ep->com.cm_id = NULL;
3017 	abort_connection(ep, NULL, GFP_KERNEL);
3018 	cm_id->rem_ref(cm_id);
3019 err:
3020 	mutex_unlock(&ep->com.mutex);
3021 	c4iw_put_ep(&ep->com);
3022 	return err;
3023 }
3024 
3025 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3026 {
3027 	struct in_device *ind;
3028 	int found = 0;
3029 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
3030 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
3031 
3032 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
3033 	if (!ind)
3034 		return -EADDRNOTAVAIL;
3035 	for_primary_ifa(ind) {
3036 		laddr->sin_addr.s_addr = ifa->ifa_address;
3037 		raddr->sin_addr.s_addr = ifa->ifa_address;
3038 		found = 1;
3039 		break;
3040 	}
3041 	endfor_ifa(ind);
3042 	in_dev_put(ind);
3043 	return found ? 0 : -EADDRNOTAVAIL;
3044 }
3045 
3046 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3047 		      unsigned char banned_flags)
3048 {
3049 	struct inet6_dev *idev;
3050 	int err = -EADDRNOTAVAIL;
3051 
3052 	rcu_read_lock();
3053 	idev = __in6_dev_get(dev);
3054 	if (idev != NULL) {
3055 		struct inet6_ifaddr *ifp;
3056 
3057 		read_lock_bh(&idev->lock);
3058 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
3059 			if (ifp->scope == IFA_LINK &&
3060 			    !(ifp->flags & banned_flags)) {
3061 				memcpy(addr, &ifp->addr, 16);
3062 				err = 0;
3063 				break;
3064 			}
3065 		}
3066 		read_unlock_bh(&idev->lock);
3067 	}
3068 	rcu_read_unlock();
3069 	return err;
3070 }
3071 
3072 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3073 {
3074 	struct in6_addr uninitialized_var(addr);
3075 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
3076 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
3077 
3078 	if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3079 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3080 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3081 		return 0;
3082 	}
3083 	return -EADDRNOTAVAIL;
3084 }
3085 
3086 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3087 {
3088 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3089 	struct c4iw_ep *ep;
3090 	int err = 0;
3091 	struct sockaddr_in *laddr;
3092 	struct sockaddr_in *raddr;
3093 	struct sockaddr_in6 *laddr6;
3094 	struct sockaddr_in6 *raddr6;
3095 	struct iwpm_dev_data pm_reg_msg;
3096 	struct iwpm_sa_data pm_msg;
3097 	__u8 *ra;
3098 	int iptype;
3099 	int iwpm_err = 0;
3100 
3101 	if ((conn_param->ord > cur_max_read_depth(dev)) ||
3102 	    (conn_param->ird > cur_max_read_depth(dev))) {
3103 		err = -EINVAL;
3104 		goto out;
3105 	}
3106 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3107 	if (!ep) {
3108 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3109 		err = -ENOMEM;
3110 		goto out;
3111 	}
3112 	init_timer(&ep->timer);
3113 	ep->plen = conn_param->private_data_len;
3114 	if (ep->plen)
3115 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3116 		       conn_param->private_data, ep->plen);
3117 	ep->ird = conn_param->ird;
3118 	ep->ord = conn_param->ord;
3119 
3120 	if (peer2peer && ep->ord == 0)
3121 		ep->ord = 1;
3122 
3123 	cm_id->add_ref(cm_id);
3124 	ep->com.dev = dev;
3125 	ep->com.cm_id = cm_id;
3126 	ep->com.qp = get_qhp(dev, conn_param->qpn);
3127 	if (!ep->com.qp) {
3128 		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3129 		err = -EINVAL;
3130 		goto fail1;
3131 	}
3132 	ref_qp(ep);
3133 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3134 	     ep->com.qp, cm_id);
3135 
3136 	/*
3137 	 * Allocate an active TID to initiate a TCP connection.
3138 	 */
3139 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3140 	if (ep->atid == -1) {
3141 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3142 		err = -ENOMEM;
3143 		goto fail1;
3144 	}
3145 	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3146 
3147 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
3148 	       sizeof(ep->com.local_addr));
3149 	memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
3150 	       sizeof(ep->com.remote_addr));
3151 
3152 	/* No port mapper available, go with the specified peer information */
3153 	memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
3154 	       sizeof(ep->com.mapped_local_addr));
3155 	memcpy(&ep->com.mapped_remote_addr, &cm_id->remote_addr,
3156 	       sizeof(ep->com.mapped_remote_addr));
3157 
3158 	c4iw_form_reg_msg(dev, &pm_reg_msg);
3159 	iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
3160 	if (iwpm_err) {
3161 		PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
3162 			__func__, iwpm_err);
3163 	}
3164 	if (iwpm_valid_pid() && !iwpm_err) {
3165 		c4iw_form_pm_msg(ep, &pm_msg);
3166 		iwpm_err = iwpm_add_and_query_mapping(&pm_msg, RDMA_NL_C4IW);
3167 		if (iwpm_err)
3168 			PDBG("%s: Port Mapper query fail (err = %d).\n",
3169 				__func__, iwpm_err);
3170 		else
3171 			c4iw_record_pm_msg(ep, &pm_msg);
3172 	}
3173 	if (iwpm_create_mapinfo(&ep->com.local_addr,
3174 				&ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
3175 		iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
3176 		err = -ENOMEM;
3177 		goto fail1;
3178 	}
3179 	print_addr(&ep->com, __func__, "add_query/create_mapinfo");
3180 	set_bit(RELEASE_MAPINFO, &ep->com.flags);
3181 
3182 	laddr = (struct sockaddr_in *)&ep->com.mapped_local_addr;
3183 	raddr = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
3184 	laddr6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
3185 	raddr6 = (struct sockaddr_in6 *) &ep->com.mapped_remote_addr;
3186 
3187 	if (cm_id->remote_addr.ss_family == AF_INET) {
3188 		iptype = 4;
3189 		ra = (__u8 *)&raddr->sin_addr;
3190 
3191 		/*
3192 		 * Handle loopback requests to INADDR_ANY.
3193 		 */
3194 		if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
3195 			err = pick_local_ipaddrs(dev, cm_id);
3196 			if (err)
3197 				goto fail1;
3198 		}
3199 
3200 		/* find a route */
3201 		PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3202 		     __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3203 		     ra, ntohs(raddr->sin_port));
3204 		ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3205 				     raddr->sin_addr.s_addr, laddr->sin_port,
3206 				     raddr->sin_port, 0);
3207 	} else {
3208 		iptype = 6;
3209 		ra = (__u8 *)&raddr6->sin6_addr;
3210 
3211 		/*
3212 		 * Handle loopback requests to INADDR_ANY.
3213 		 */
3214 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3215 			err = pick_local_ip6addrs(dev, cm_id);
3216 			if (err)
3217 				goto fail1;
3218 		}
3219 
3220 		/* find a route */
3221 		PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3222 		     __func__, laddr6->sin6_addr.s6_addr,
3223 		     ntohs(laddr6->sin6_port),
3224 		     raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3225 		ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3226 				      raddr6->sin6_addr.s6_addr,
3227 				      laddr6->sin6_port, raddr6->sin6_port, 0,
3228 				      raddr6->sin6_scope_id);
3229 	}
3230 	if (!ep->dst) {
3231 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3232 		err = -EHOSTUNREACH;
3233 		goto fail2;
3234 	}
3235 
3236 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3237 			ep->com.dev->rdev.lldi.adapter_type);
3238 	if (err) {
3239 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3240 		goto fail3;
3241 	}
3242 
3243 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3244 		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3245 		ep->l2t->idx);
3246 
3247 	state_set(&ep->com, CONNECTING);
3248 	ep->tos = 0;
3249 
3250 	/* send connect request to rnic */
3251 	err = send_connect(ep);
3252 	if (!err)
3253 		goto out;
3254 
3255 	cxgb4_l2t_release(ep->l2t);
3256 fail3:
3257 	dst_release(ep->dst);
3258 fail2:
3259 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3260 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3261 fail1:
3262 	cm_id->rem_ref(cm_id);
3263 	c4iw_put_ep(&ep->com);
3264 out:
3265 	return err;
3266 }
3267 
3268 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3269 {
3270 	int err;
3271 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3272 				    &ep->com.mapped_local_addr;
3273 
3274 	if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3275 		err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3276 				     (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3277 		if (err)
3278 			return err;
3279 	}
3280 	c4iw_init_wr_wait(&ep->com.wr_wait);
3281 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3282 				   ep->stid, &sin6->sin6_addr,
3283 				   sin6->sin6_port,
3284 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
3285 	if (!err)
3286 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3287 					  &ep->com.wr_wait,
3288 					  0, 0, __func__);
3289 	else if (err > 0)
3290 		err = net_xmit_errno(err);
3291 	if (err) {
3292 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3293 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3294 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3295 		       err, ep->stid,
3296 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3297 	}
3298 	return err;
3299 }
3300 
3301 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3302 {
3303 	int err;
3304 	struct sockaddr_in *sin = (struct sockaddr_in *)
3305 				  &ep->com.mapped_local_addr;
3306 
3307 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
3308 		do {
3309 			err = cxgb4_create_server_filter(
3310 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3311 				sin->sin_addr.s_addr, sin->sin_port, 0,
3312 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3313 			if (err == -EBUSY) {
3314 				if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3315 					err = -EIO;
3316 					break;
3317 				}
3318 				set_current_state(TASK_UNINTERRUPTIBLE);
3319 				schedule_timeout(usecs_to_jiffies(100));
3320 			}
3321 		} while (err == -EBUSY);
3322 	} else {
3323 		c4iw_init_wr_wait(&ep->com.wr_wait);
3324 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3325 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3326 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3327 		if (!err)
3328 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3329 						  &ep->com.wr_wait,
3330 						  0, 0, __func__);
3331 		else if (err > 0)
3332 			err = net_xmit_errno(err);
3333 	}
3334 	if (err)
3335 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3336 		       , err, ep->stid,
3337 		       &sin->sin_addr, ntohs(sin->sin_port));
3338 	return err;
3339 }
3340 
3341 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3342 {
3343 	int err = 0;
3344 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3345 	struct c4iw_listen_ep *ep;
3346 	struct iwpm_dev_data pm_reg_msg;
3347 	struct iwpm_sa_data pm_msg;
3348 	int iwpm_err = 0;
3349 
3350 	might_sleep();
3351 
3352 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3353 	if (!ep) {
3354 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3355 		err = -ENOMEM;
3356 		goto fail1;
3357 	}
3358 	PDBG("%s ep %p\n", __func__, ep);
3359 	cm_id->add_ref(cm_id);
3360 	ep->com.cm_id = cm_id;
3361 	ep->com.dev = dev;
3362 	ep->backlog = backlog;
3363 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
3364 	       sizeof(ep->com.local_addr));
3365 
3366 	/*
3367 	 * Allocate a server TID.
3368 	 */
3369 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
3370 	    ep->com.local_addr.ss_family == AF_INET)
3371 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3372 					     cm_id->local_addr.ss_family, ep);
3373 	else
3374 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3375 					    cm_id->local_addr.ss_family, ep);
3376 
3377 	if (ep->stid == -1) {
3378 		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3379 		err = -ENOMEM;
3380 		goto fail2;
3381 	}
3382 	insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3383 
3384 	/* No port mapper available, go with the specified info */
3385 	memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
3386 	       sizeof(ep->com.mapped_local_addr));
3387 
3388 	c4iw_form_reg_msg(dev, &pm_reg_msg);
3389 	iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
3390 	if (iwpm_err) {
3391 		PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
3392 			__func__, iwpm_err);
3393 	}
3394 	if (iwpm_valid_pid() && !iwpm_err) {
3395 		memcpy(&pm_msg.loc_addr, &ep->com.local_addr,
3396 				sizeof(ep->com.local_addr));
3397 		iwpm_err = iwpm_add_mapping(&pm_msg, RDMA_NL_C4IW);
3398 		if (iwpm_err)
3399 			PDBG("%s: Port Mapper query fail (err = %d).\n",
3400 				__func__, iwpm_err);
3401 		else
3402 			memcpy(&ep->com.mapped_local_addr,
3403 				&pm_msg.mapped_loc_addr,
3404 				sizeof(ep->com.mapped_local_addr));
3405 	}
3406 	if (iwpm_create_mapinfo(&ep->com.local_addr,
3407 				&ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
3408 		err = -ENOMEM;
3409 		goto fail3;
3410 	}
3411 	print_addr(&ep->com, __func__, "add_mapping/create_mapinfo");
3412 
3413 	set_bit(RELEASE_MAPINFO, &ep->com.flags);
3414 	state_set(&ep->com, LISTEN);
3415 	if (ep->com.local_addr.ss_family == AF_INET)
3416 		err = create_server4(dev, ep);
3417 	else
3418 		err = create_server6(dev, ep);
3419 	if (!err) {
3420 		cm_id->provider_data = ep;
3421 		goto out;
3422 	}
3423 
3424 fail3:
3425 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3426 			ep->com.local_addr.ss_family);
3427 fail2:
3428 	cm_id->rem_ref(cm_id);
3429 	c4iw_put_ep(&ep->com);
3430 fail1:
3431 out:
3432 	return err;
3433 }
3434 
3435 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3436 {
3437 	int err;
3438 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3439 
3440 	PDBG("%s ep %p\n", __func__, ep);
3441 
3442 	might_sleep();
3443 	state_set(&ep->com, DEAD);
3444 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3445 	    ep->com.local_addr.ss_family == AF_INET) {
3446 		err = cxgb4_remove_server_filter(
3447 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
3448 			ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3449 	} else {
3450 		struct sockaddr_in6 *sin6;
3451 		c4iw_init_wr_wait(&ep->com.wr_wait);
3452 		err = cxgb4_remove_server(
3453 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3454 				ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3455 		if (err)
3456 			goto done;
3457 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3458 					  0, 0, __func__);
3459 		sin6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
3460 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3461 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3462 	}
3463 	remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3464 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3465 			ep->com.local_addr.ss_family);
3466 done:
3467 	cm_id->rem_ref(cm_id);
3468 	c4iw_put_ep(&ep->com);
3469 	return err;
3470 }
3471 
3472 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3473 {
3474 	int ret = 0;
3475 	int close = 0;
3476 	int fatal = 0;
3477 	struct c4iw_rdev *rdev;
3478 
3479 	mutex_lock(&ep->com.mutex);
3480 
3481 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3482 	     states[ep->com.state], abrupt);
3483 
3484 	rdev = &ep->com.dev->rdev;
3485 	if (c4iw_fatal_error(rdev)) {
3486 		fatal = 1;
3487 		close_complete_upcall(ep, -EIO);
3488 		ep->com.state = DEAD;
3489 	}
3490 	switch (ep->com.state) {
3491 	case MPA_REQ_WAIT:
3492 	case MPA_REQ_SENT:
3493 	case MPA_REQ_RCVD:
3494 	case MPA_REP_SENT:
3495 	case FPDU_MODE:
3496 		close = 1;
3497 		if (abrupt)
3498 			ep->com.state = ABORTING;
3499 		else {
3500 			ep->com.state = CLOSING;
3501 			start_ep_timer(ep);
3502 		}
3503 		set_bit(CLOSE_SENT, &ep->com.flags);
3504 		break;
3505 	case CLOSING:
3506 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3507 			close = 1;
3508 			if (abrupt) {
3509 				(void)stop_ep_timer(ep);
3510 				ep->com.state = ABORTING;
3511 			} else
3512 				ep->com.state = MORIBUND;
3513 		}
3514 		break;
3515 	case MORIBUND:
3516 	case ABORTING:
3517 	case DEAD:
3518 		PDBG("%s ignoring disconnect ep %p state %u\n",
3519 		     __func__, ep, ep->com.state);
3520 		break;
3521 	default:
3522 		BUG();
3523 		break;
3524 	}
3525 
3526 	if (close) {
3527 		if (abrupt) {
3528 			set_bit(EP_DISC_ABORT, &ep->com.history);
3529 			close_complete_upcall(ep, -ECONNRESET);
3530 			ret = send_abort(ep, NULL, gfp);
3531 		} else {
3532 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3533 			ret = send_halfclose(ep, gfp);
3534 		}
3535 		if (ret)
3536 			fatal = 1;
3537 	}
3538 	mutex_unlock(&ep->com.mutex);
3539 	if (fatal)
3540 		release_ep_resources(ep);
3541 	return ret;
3542 }
3543 
3544 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3545 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3546 {
3547 	struct c4iw_ep *ep;
3548 	int atid = be32_to_cpu(req->tid);
3549 
3550 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3551 					   (__force u32) req->tid);
3552 	if (!ep)
3553 		return;
3554 
3555 	switch (req->retval) {
3556 	case FW_ENOMEM:
3557 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3558 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3559 			send_fw_act_open_req(ep, atid);
3560 			return;
3561 		}
3562 	case FW_EADDRINUSE:
3563 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3564 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3565 			send_fw_act_open_req(ep, atid);
3566 			return;
3567 		}
3568 		break;
3569 	default:
3570 		pr_info("%s unexpected ofld conn wr retval %d\n",
3571 		       __func__, req->retval);
3572 		break;
3573 	}
3574 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3575 	       req->retval, atid);
3576 	mutex_lock(&dev->rdev.stats.lock);
3577 	dev->rdev.stats.act_ofld_conn_fails++;
3578 	mutex_unlock(&dev->rdev.stats.lock);
3579 	connect_reply_upcall(ep, status2errno(req->retval));
3580 	state_set(&ep->com, DEAD);
3581 	if (ep->com.remote_addr.ss_family == AF_INET6) {
3582 		struct sockaddr_in6 *sin6 =
3583 			(struct sockaddr_in6 *)&ep->com.mapped_local_addr;
3584 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3585 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3586 	}
3587 	remove_handle(dev, &dev->atid_idr, atid);
3588 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3589 	dst_release(ep->dst);
3590 	cxgb4_l2t_release(ep->l2t);
3591 	c4iw_put_ep(&ep->com);
3592 }
3593 
3594 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3595 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3596 {
3597 	struct sk_buff *rpl_skb;
3598 	struct cpl_pass_accept_req *cpl;
3599 	int ret;
3600 
3601 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3602 	BUG_ON(!rpl_skb);
3603 	if (req->retval) {
3604 		PDBG("%s passive open failure %d\n", __func__, req->retval);
3605 		mutex_lock(&dev->rdev.stats.lock);
3606 		dev->rdev.stats.pas_ofld_conn_fails++;
3607 		mutex_unlock(&dev->rdev.stats.lock);
3608 		kfree_skb(rpl_skb);
3609 	} else {
3610 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3611 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3612 					(__force u32) htonl(
3613 					(__force u32) req->tid)));
3614 		ret = pass_accept_req(dev, rpl_skb);
3615 		if (!ret)
3616 			kfree_skb(rpl_skb);
3617 	}
3618 	return;
3619 }
3620 
3621 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3622 {
3623 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3624 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3625 
3626 	switch (rpl->type) {
3627 	case FW6_TYPE_CQE:
3628 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3629 		break;
3630 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3631 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3632 		switch (req->t_state) {
3633 		case TCP_SYN_SENT:
3634 			active_ofld_conn_reply(dev, skb, req);
3635 			break;
3636 		case TCP_SYN_RECV:
3637 			passive_ofld_conn_reply(dev, skb, req);
3638 			break;
3639 		default:
3640 			pr_err("%s unexpected ofld conn wr state %d\n",
3641 			       __func__, req->t_state);
3642 			break;
3643 		}
3644 		break;
3645 	}
3646 	return 0;
3647 }
3648 
3649 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3650 {
3651 	__be32 l2info;
3652 	__be16 hdr_len, vlantag, len;
3653 	u16 eth_hdr_len;
3654 	int tcp_hdr_len, ip_hdr_len;
3655 	u8 intf;
3656 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3657 	struct cpl_pass_accept_req *req;
3658 	struct tcp_options_received tmp_opt;
3659 	struct c4iw_dev *dev;
3660 	enum chip_type type;
3661 
3662 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3663 	/* Store values from cpl_rx_pkt in temporary location. */
3664 	vlantag = cpl->vlan;
3665 	len = cpl->len;
3666 	l2info  = cpl->l2info;
3667 	hdr_len = cpl->hdr_len;
3668 	intf = cpl->iff;
3669 
3670 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3671 
3672 	/*
3673 	 * We need to parse the TCP options from SYN packet.
3674 	 * to generate cpl_pass_accept_req.
3675 	 */
3676 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3677 	tcp_clear_options(&tmp_opt);
3678 	tcp_parse_options(skb, &tmp_opt, 0, NULL);
3679 
3680 	req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3681 	memset(req, 0, sizeof(*req));
3682 	req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3683 			 SYN_MAC_IDX_V(RX_MACIDX_G(
3684 			 be32_to_cpu(l2info))) |
3685 			 SYN_XACT_MATCH_F);
3686 	type = dev->rdev.lldi.adapter_type;
3687 	tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3688 	ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3689 	req->hdr_len =
3690 		cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3691 	if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3692 		eth_hdr_len = is_t4(type) ?
3693 				RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3694 				RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3695 		req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3696 					    IP_HDR_LEN_V(ip_hdr_len) |
3697 					    ETH_HDR_LEN_V(eth_hdr_len));
3698 	} else { /* T6 and later */
3699 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3700 		req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3701 					    T6_IP_HDR_LEN_V(ip_hdr_len) |
3702 					    T6_ETH_HDR_LEN_V(eth_hdr_len));
3703 	}
3704 	req->vlan = vlantag;
3705 	req->len = len;
3706 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3707 				    PASS_OPEN_TOS_V(tos));
3708 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3709 	if (tmp_opt.wscale_ok)
3710 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3711 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3712 	if (tmp_opt.sack_ok)
3713 		req->tcpopt.sack = 1;
3714 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3715 	return;
3716 }
3717 
3718 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3719 				  __be32 laddr, __be16 lport,
3720 				  __be32 raddr, __be16 rport,
3721 				  u32 rcv_isn, u32 filter, u16 window,
3722 				  u32 rss_qid, u8 port_id)
3723 {
3724 	struct sk_buff *req_skb;
3725 	struct fw_ofld_connection_wr *req;
3726 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
3727 	int ret;
3728 
3729 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3730 	req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3731 	memset(req, 0, sizeof(*req));
3732 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3733 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3734 	req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3735 	req->le.filter = (__force __be32) filter;
3736 	req->le.lport = lport;
3737 	req->le.pport = rport;
3738 	req->le.u.ipv4.lip = laddr;
3739 	req->le.u.ipv4.pip = raddr;
3740 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3741 	req->tcb.rcv_adv = htons(window);
3742 	req->tcb.t_state_to_astid =
3743 		 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3744 			FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3745 			FW_OFLD_CONNECTION_WR_ASTID_V(
3746 			PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3747 
3748 	/*
3749 	 * We store the qid in opt2 which will be used by the firmware
3750 	 * to send us the wr response.
3751 	 */
3752 	req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3753 
3754 	/*
3755 	 * We initialize the MSS index in TCB to 0xF.
3756 	 * So that when driver sends cpl_pass_accept_rpl
3757 	 * TCB picks up the correct value. If this was 0
3758 	 * TP will ignore any value > 0 for MSS index.
3759 	 */
3760 	req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3761 	req->cookie = (uintptr_t)skb;
3762 
3763 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3764 	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3765 	if (ret < 0) {
3766 		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3767 		       ret);
3768 		kfree_skb(skb);
3769 		kfree_skb(req_skb);
3770 	}
3771 }
3772 
3773 /*
3774  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3775  * messages when a filter is being used instead of server to
3776  * redirect a syn packet. When packets hit filter they are redirected
3777  * to the offload queue and driver tries to establish the connection
3778  * using firmware work request.
3779  */
3780 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3781 {
3782 	int stid;
3783 	unsigned int filter;
3784 	struct ethhdr *eh = NULL;
3785 	struct vlan_ethhdr *vlan_eh = NULL;
3786 	struct iphdr *iph;
3787 	struct tcphdr *tcph;
3788 	struct rss_header *rss = (void *)skb->data;
3789 	struct cpl_rx_pkt *cpl = (void *)skb->data;
3790 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
3791 	struct l2t_entry *e;
3792 	struct dst_entry *dst;
3793 	struct c4iw_ep *lep;
3794 	u16 window;
3795 	struct port_info *pi;
3796 	struct net_device *pdev;
3797 	u16 rss_qid, eth_hdr_len;
3798 	int step;
3799 	u32 tx_chan;
3800 	struct neighbour *neigh;
3801 
3802 	/* Drop all non-SYN packets */
3803 	if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3804 		goto reject;
3805 
3806 	/*
3807 	 * Drop all packets which did not hit the filter.
3808 	 * Unlikely to happen.
3809 	 */
3810 	if (!(rss->filter_hit && rss->filter_tid))
3811 		goto reject;
3812 
3813 	/*
3814 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3815 	 */
3816 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3817 
3818 	lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3819 	if (!lep) {
3820 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3821 		goto reject;
3822 	}
3823 
3824 	switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3825 	case CHELSIO_T4:
3826 		eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3827 		break;
3828 	case CHELSIO_T5:
3829 		eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3830 		break;
3831 	case CHELSIO_T6:
3832 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3833 		break;
3834 	default:
3835 		pr_err("T%d Chip is not supported\n",
3836 		       CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3837 		goto reject;
3838 	}
3839 
3840 	if (eth_hdr_len == ETH_HLEN) {
3841 		eh = (struct ethhdr *)(req + 1);
3842 		iph = (struct iphdr *)(eh + 1);
3843 	} else {
3844 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
3845 		iph = (struct iphdr *)(vlan_eh + 1);
3846 		skb->vlan_tci = ntohs(cpl->vlan);
3847 	}
3848 
3849 	if (iph->version != 0x4)
3850 		goto reject;
3851 
3852 	tcph = (struct tcphdr *)(iph + 1);
3853 	skb_set_network_header(skb, (void *)iph - (void *)rss);
3854 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3855 	skb_get(skb);
3856 
3857 	PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3858 	     ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3859 	     ntohs(tcph->source), iph->tos);
3860 
3861 	dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3862 			 iph->tos);
3863 	if (!dst) {
3864 		pr_err("%s - failed to find dst entry!\n",
3865 		       __func__);
3866 		goto reject;
3867 	}
3868 	neigh = dst_neigh_lookup_skb(dst, skb);
3869 
3870 	if (!neigh) {
3871 		pr_err("%s - failed to allocate neigh!\n",
3872 		       __func__);
3873 		goto free_dst;
3874 	}
3875 
3876 	if (neigh->dev->flags & IFF_LOOPBACK) {
3877 		pdev = ip_dev_find(&init_net, iph->daddr);
3878 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3879 				    pdev, 0);
3880 		pi = (struct port_info *)netdev_priv(pdev);
3881 		tx_chan = cxgb4_port_chan(pdev);
3882 		dev_put(pdev);
3883 	} else {
3884 		pdev = get_real_dev(neigh->dev);
3885 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3886 					pdev, 0);
3887 		pi = (struct port_info *)netdev_priv(pdev);
3888 		tx_chan = cxgb4_port_chan(pdev);
3889 	}
3890 	neigh_release(neigh);
3891 	if (!e) {
3892 		pr_err("%s - failed to allocate l2t entry!\n",
3893 		       __func__);
3894 		goto free_dst;
3895 	}
3896 
3897 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3898 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3899 	window = (__force u16) htons((__force u16)tcph->window);
3900 
3901 	/* Calcuate filter portion for LE region. */
3902 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3903 						    dev->rdev.lldi.ports[0],
3904 						    e));
3905 
3906 	/*
3907 	 * Synthesize the cpl_pass_accept_req. We have everything except the
3908 	 * TID. Once firmware sends a reply with TID we update the TID field
3909 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
3910 	 */
3911 	build_cpl_pass_accept_req(skb, stid, iph->tos);
3912 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3913 			      tcph->source, ntohl(tcph->seq), filter, window,
3914 			      rss_qid, pi->port_id);
3915 	cxgb4_l2t_release(e);
3916 free_dst:
3917 	dst_release(dst);
3918 reject:
3919 	return 0;
3920 }
3921 
3922 /*
3923  * These are the real handlers that are called from a
3924  * work queue.
3925  */
3926 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3927 	[CPL_ACT_ESTABLISH] = act_establish,
3928 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
3929 	[CPL_RX_DATA] = rx_data,
3930 	[CPL_ABORT_RPL_RSS] = abort_rpl,
3931 	[CPL_ABORT_RPL] = abort_rpl,
3932 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
3933 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3934 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3935 	[CPL_PASS_ESTABLISH] = pass_establish,
3936 	[CPL_PEER_CLOSE] = peer_close,
3937 	[CPL_ABORT_REQ_RSS] = peer_abort,
3938 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
3939 	[CPL_RDMA_TERMINATE] = terminate,
3940 	[CPL_FW4_ACK] = fw4_ack,
3941 	[CPL_FW6_MSG] = deferred_fw6_msg,
3942 	[CPL_RX_PKT] = rx_pkt
3943 };
3944 
3945 static void process_timeout(struct c4iw_ep *ep)
3946 {
3947 	struct c4iw_qp_attributes attrs;
3948 	int abort = 1;
3949 
3950 	mutex_lock(&ep->com.mutex);
3951 	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3952 	     ep->com.state);
3953 	set_bit(TIMEDOUT, &ep->com.history);
3954 	switch (ep->com.state) {
3955 	case MPA_REQ_SENT:
3956 		__state_set(&ep->com, ABORTING);
3957 		connect_reply_upcall(ep, -ETIMEDOUT);
3958 		break;
3959 	case MPA_REQ_WAIT:
3960 		__state_set(&ep->com, ABORTING);
3961 		break;
3962 	case CLOSING:
3963 	case MORIBUND:
3964 		if (ep->com.cm_id && ep->com.qp) {
3965 			attrs.next_state = C4IW_QP_STATE_ERROR;
3966 			c4iw_modify_qp(ep->com.qp->rhp,
3967 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3968 				     &attrs, 1);
3969 		}
3970 		__state_set(&ep->com, ABORTING);
3971 		close_complete_upcall(ep, -ETIMEDOUT);
3972 		break;
3973 	case ABORTING:
3974 	case DEAD:
3975 
3976 		/*
3977 		 * These states are expected if the ep timed out at the same
3978 		 * time as another thread was calling stop_ep_timer().
3979 		 * So we silently do nothing for these states.
3980 		 */
3981 		abort = 0;
3982 		break;
3983 	default:
3984 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3985 			__func__, ep, ep->hwtid, ep->com.state);
3986 		abort = 0;
3987 	}
3988 	if (abort)
3989 		abort_connection(ep, NULL, GFP_KERNEL);
3990 	mutex_unlock(&ep->com.mutex);
3991 	c4iw_put_ep(&ep->com);
3992 }
3993 
3994 static void process_timedout_eps(void)
3995 {
3996 	struct c4iw_ep *ep;
3997 
3998 	spin_lock_irq(&timeout_lock);
3999 	while (!list_empty(&timeout_list)) {
4000 		struct list_head *tmp;
4001 
4002 		tmp = timeout_list.next;
4003 		list_del(tmp);
4004 		tmp->next = NULL;
4005 		tmp->prev = NULL;
4006 		spin_unlock_irq(&timeout_lock);
4007 		ep = list_entry(tmp, struct c4iw_ep, entry);
4008 		process_timeout(ep);
4009 		spin_lock_irq(&timeout_lock);
4010 	}
4011 	spin_unlock_irq(&timeout_lock);
4012 }
4013 
4014 static void process_work(struct work_struct *work)
4015 {
4016 	struct sk_buff *skb = NULL;
4017 	struct c4iw_dev *dev;
4018 	struct cpl_act_establish *rpl;
4019 	unsigned int opcode;
4020 	int ret;
4021 
4022 	process_timedout_eps();
4023 	while ((skb = skb_dequeue(&rxq))) {
4024 		rpl = cplhdr(skb);
4025 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4026 		opcode = rpl->ot.opcode;
4027 
4028 		BUG_ON(!work_handlers[opcode]);
4029 		ret = work_handlers[opcode](dev, skb);
4030 		if (!ret)
4031 			kfree_skb(skb);
4032 		process_timedout_eps();
4033 	}
4034 }
4035 
4036 static DECLARE_WORK(skb_work, process_work);
4037 
4038 static void ep_timeout(unsigned long arg)
4039 {
4040 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4041 	int kickit = 0;
4042 
4043 	spin_lock(&timeout_lock);
4044 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4045 		/*
4046 		 * Only insert if it is not already on the list.
4047 		 */
4048 		if (!ep->entry.next) {
4049 			list_add_tail(&ep->entry, &timeout_list);
4050 			kickit = 1;
4051 		}
4052 	}
4053 	spin_unlock(&timeout_lock);
4054 	if (kickit)
4055 		queue_work(workq, &skb_work);
4056 }
4057 
4058 /*
4059  * All the CM events are handled on a work queue to have a safe context.
4060  */
4061 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4062 {
4063 
4064 	/*
4065 	 * Save dev in the skb->cb area.
4066 	 */
4067 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4068 
4069 	/*
4070 	 * Queue the skb and schedule the worker thread.
4071 	 */
4072 	skb_queue_tail(&rxq, skb);
4073 	queue_work(workq, &skb_work);
4074 	return 0;
4075 }
4076 
4077 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4078 {
4079 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4080 
4081 	if (rpl->status != CPL_ERR_NONE) {
4082 		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4083 		       "for tid %u\n", rpl->status, GET_TID(rpl));
4084 	}
4085 	kfree_skb(skb);
4086 	return 0;
4087 }
4088 
4089 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4090 {
4091 	struct cpl_fw6_msg *rpl = cplhdr(skb);
4092 	struct c4iw_wr_wait *wr_waitp;
4093 	int ret;
4094 
4095 	PDBG("%s type %u\n", __func__, rpl->type);
4096 
4097 	switch (rpl->type) {
4098 	case FW6_TYPE_WR_RPL:
4099 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4100 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4101 		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4102 		if (wr_waitp)
4103 			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4104 		kfree_skb(skb);
4105 		break;
4106 	case FW6_TYPE_CQE:
4107 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4108 		sched(dev, skb);
4109 		break;
4110 	default:
4111 		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4112 		       rpl->type);
4113 		kfree_skb(skb);
4114 		break;
4115 	}
4116 	return 0;
4117 }
4118 
4119 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4120 {
4121 	struct cpl_abort_req_rss *req = cplhdr(skb);
4122 	struct c4iw_ep *ep;
4123 	struct tid_info *t = dev->rdev.lldi.tids;
4124 	unsigned int tid = GET_TID(req);
4125 
4126 	ep = lookup_tid(t, tid);
4127 	if (!ep) {
4128 		printk(KERN_WARNING MOD
4129 		       "Abort on non-existent endpoint, tid %d\n", tid);
4130 		kfree_skb(skb);
4131 		return 0;
4132 	}
4133 	if (is_neg_adv(req->status)) {
4134 		PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4135 		     __func__, ep->hwtid, req->status,
4136 		     neg_adv_str(req->status));
4137 		ep->stats.abort_neg_adv++;
4138 		dev->rdev.stats.neg_adv++;
4139 		kfree_skb(skb);
4140 		return 0;
4141 	}
4142 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4143 	     ep->com.state);
4144 
4145 	/*
4146 	 * Wake up any threads in rdma_init() or rdma_fini().
4147 	 * However, if we are on MPAv2 and want to retry with MPAv1
4148 	 * then, don't wake up yet.
4149 	 */
4150 	if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
4151 		if (ep->com.state != MPA_REQ_SENT)
4152 			c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4153 	} else
4154 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4155 	sched(dev, skb);
4156 	return 0;
4157 }
4158 
4159 /*
4160  * Most upcalls from the T4 Core go to sched() to
4161  * schedule the processing on a work queue.
4162  */
4163 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4164 	[CPL_ACT_ESTABLISH] = sched,
4165 	[CPL_ACT_OPEN_RPL] = sched,
4166 	[CPL_RX_DATA] = sched,
4167 	[CPL_ABORT_RPL_RSS] = sched,
4168 	[CPL_ABORT_RPL] = sched,
4169 	[CPL_PASS_OPEN_RPL] = sched,
4170 	[CPL_CLOSE_LISTSRV_RPL] = sched,
4171 	[CPL_PASS_ACCEPT_REQ] = sched,
4172 	[CPL_PASS_ESTABLISH] = sched,
4173 	[CPL_PEER_CLOSE] = sched,
4174 	[CPL_CLOSE_CON_RPL] = sched,
4175 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
4176 	[CPL_RDMA_TERMINATE] = sched,
4177 	[CPL_FW4_ACK] = sched,
4178 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
4179 	[CPL_FW6_MSG] = fw6_msg,
4180 	[CPL_RX_PKT] = sched
4181 };
4182 
4183 int __init c4iw_cm_init(void)
4184 {
4185 	spin_lock_init(&timeout_lock);
4186 	skb_queue_head_init(&rxq);
4187 
4188 	workq = create_singlethread_workqueue("iw_cxgb4");
4189 	if (!workq)
4190 		return -ENOMEM;
4191 
4192 	return 0;
4193 }
4194 
4195 void c4iw_cm_term(void)
4196 {
4197 	WARN_ON(!list_empty(&timeout_list));
4198 	flush_workqueue(workq);
4199 	destroy_workqueue(workq);
4200 }
4201