xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision 6774def6)
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 
54 static char *states[] = {
55 	"idle",
56 	"listen",
57 	"connecting",
58 	"mpa_wait_req",
59 	"mpa_req_sent",
60 	"mpa_req_rcvd",
61 	"mpa_rep_sent",
62 	"fpdu_mode",
63 	"aborting",
64 	"closing",
65 	"moribund",
66 	"dead",
67 	NULL,
68 };
69 
70 static int nocong;
71 module_param(nocong, int, 0644);
72 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
73 
74 static int enable_ecn;
75 module_param(enable_ecn, int, 0644);
76 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
77 
78 static int dack_mode = 1;
79 module_param(dack_mode, int, 0644);
80 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
81 
82 uint c4iw_max_read_depth = 32;
83 module_param(c4iw_max_read_depth, int, 0644);
84 MODULE_PARM_DESC(c4iw_max_read_depth,
85 		 "Per-connection max ORD/IRD (default=32)");
86 
87 static int enable_tcp_timestamps;
88 module_param(enable_tcp_timestamps, int, 0644);
89 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
90 
91 static int enable_tcp_sack;
92 module_param(enable_tcp_sack, int, 0644);
93 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
94 
95 static int enable_tcp_window_scaling = 1;
96 module_param(enable_tcp_window_scaling, int, 0644);
97 MODULE_PARM_DESC(enable_tcp_window_scaling,
98 		 "Enable tcp window scaling (default=1)");
99 
100 int c4iw_debug;
101 module_param(c4iw_debug, int, 0644);
102 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
103 
104 static int peer2peer = 1;
105 module_param(peer2peer, int, 0644);
106 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
107 
108 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
109 module_param(p2p_type, int, 0644);
110 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
111 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
112 
113 static int ep_timeout_secs = 60;
114 module_param(ep_timeout_secs, int, 0644);
115 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
116 				   "in seconds (default=60)");
117 
118 static int mpa_rev = 1;
119 module_param(mpa_rev, int, 0644);
120 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
121 		"1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
122 		" compliant (default=1)");
123 
124 static int markers_enabled;
125 module_param(markers_enabled, int, 0644);
126 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
127 
128 static int crc_enabled = 1;
129 module_param(crc_enabled, int, 0644);
130 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
131 
132 static int rcv_win = 256 * 1024;
133 module_param(rcv_win, int, 0644);
134 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
135 
136 static int snd_win = 128 * 1024;
137 module_param(snd_win, int, 0644);
138 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
139 
140 static struct workqueue_struct *workq;
141 
142 static struct sk_buff_head rxq;
143 
144 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
145 static void ep_timeout(unsigned long arg);
146 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
147 
148 static LIST_HEAD(timeout_list);
149 static spinlock_t timeout_lock;
150 
151 static void deref_qp(struct c4iw_ep *ep)
152 {
153 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
154 	clear_bit(QP_REFERENCED, &ep->com.flags);
155 }
156 
157 static void ref_qp(struct c4iw_ep *ep)
158 {
159 	set_bit(QP_REFERENCED, &ep->com.flags);
160 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
161 }
162 
163 static void start_ep_timer(struct c4iw_ep *ep)
164 {
165 	PDBG("%s ep %p\n", __func__, ep);
166 	if (timer_pending(&ep->timer)) {
167 		pr_err("%s timer already started! ep %p\n",
168 		       __func__, ep);
169 		return;
170 	}
171 	clear_bit(TIMEOUT, &ep->com.flags);
172 	c4iw_get_ep(&ep->com);
173 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
174 	ep->timer.data = (unsigned long)ep;
175 	ep->timer.function = ep_timeout;
176 	add_timer(&ep->timer);
177 }
178 
179 static int stop_ep_timer(struct c4iw_ep *ep)
180 {
181 	PDBG("%s ep %p stopping\n", __func__, ep);
182 	del_timer_sync(&ep->timer);
183 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
184 		c4iw_put_ep(&ep->com);
185 		return 0;
186 	}
187 	return 1;
188 }
189 
190 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
191 		  struct l2t_entry *l2e)
192 {
193 	int	error = 0;
194 
195 	if (c4iw_fatal_error(rdev)) {
196 		kfree_skb(skb);
197 		PDBG("%s - device in error state - dropping\n", __func__);
198 		return -EIO;
199 	}
200 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
201 	if (error < 0)
202 		kfree_skb(skb);
203 	return error < 0 ? error : 0;
204 }
205 
206 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
207 {
208 	int	error = 0;
209 
210 	if (c4iw_fatal_error(rdev)) {
211 		kfree_skb(skb);
212 		PDBG("%s - device in error state - dropping\n", __func__);
213 		return -EIO;
214 	}
215 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
216 	if (error < 0)
217 		kfree_skb(skb);
218 	return error < 0 ? error : 0;
219 }
220 
221 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
222 {
223 	struct cpl_tid_release *req;
224 
225 	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
226 	if (!skb)
227 		return;
228 	req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
229 	INIT_TP_WR(req, hwtid);
230 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
231 	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
232 	c4iw_ofld_send(rdev, skb);
233 	return;
234 }
235 
236 static void set_emss(struct c4iw_ep *ep, u16 opt)
237 {
238 	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] -
239 		   ((AF_INET == ep->com.remote_addr.ss_family) ?
240 		    sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
241 		   sizeof(struct tcphdr);
242 	ep->mss = ep->emss;
243 	if (GET_TCPOPT_TSTAMP(opt))
244 		ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
245 	if (ep->emss < 128)
246 		ep->emss = 128;
247 	if (ep->emss & 7)
248 		PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
249 		     GET_TCPOPT_MSS(opt), ep->mss, ep->emss);
250 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
251 	     ep->mss, ep->emss);
252 }
253 
254 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
255 {
256 	enum c4iw_ep_state state;
257 
258 	mutex_lock(&epc->mutex);
259 	state = epc->state;
260 	mutex_unlock(&epc->mutex);
261 	return state;
262 }
263 
264 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
265 {
266 	epc->state = new;
267 }
268 
269 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
270 {
271 	mutex_lock(&epc->mutex);
272 	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
273 	__state_set(epc, new);
274 	mutex_unlock(&epc->mutex);
275 	return;
276 }
277 
278 static void *alloc_ep(int size, gfp_t gfp)
279 {
280 	struct c4iw_ep_common *epc;
281 
282 	epc = kzalloc(size, gfp);
283 	if (epc) {
284 		kref_init(&epc->kref);
285 		mutex_init(&epc->mutex);
286 		c4iw_init_wr_wait(&epc->wr_wait);
287 	}
288 	PDBG("%s alloc ep %p\n", __func__, epc);
289 	return epc;
290 }
291 
292 void _c4iw_free_ep(struct kref *kref)
293 {
294 	struct c4iw_ep *ep;
295 
296 	ep = container_of(kref, struct c4iw_ep, com.kref);
297 	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
298 	if (test_bit(QP_REFERENCED, &ep->com.flags))
299 		deref_qp(ep);
300 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
301 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
302 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
303 		dst_release(ep->dst);
304 		cxgb4_l2t_release(ep->l2t);
305 	}
306 	if (test_bit(RELEASE_MAPINFO, &ep->com.flags)) {
307 		print_addr(&ep->com, __func__, "remove_mapinfo/mapping");
308 		iwpm_remove_mapinfo(&ep->com.local_addr,
309 				    &ep->com.mapped_local_addr);
310 		iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
311 	}
312 	kfree(ep);
313 }
314 
315 static void release_ep_resources(struct c4iw_ep *ep)
316 {
317 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
318 	c4iw_put_ep(&ep->com);
319 }
320 
321 static int status2errno(int status)
322 {
323 	switch (status) {
324 	case CPL_ERR_NONE:
325 		return 0;
326 	case CPL_ERR_CONN_RESET:
327 		return -ECONNRESET;
328 	case CPL_ERR_ARP_MISS:
329 		return -EHOSTUNREACH;
330 	case CPL_ERR_CONN_TIMEDOUT:
331 		return -ETIMEDOUT;
332 	case CPL_ERR_TCAM_FULL:
333 		return -ENOMEM;
334 	case CPL_ERR_CONN_EXIST:
335 		return -EADDRINUSE;
336 	default:
337 		return -EIO;
338 	}
339 }
340 
341 /*
342  * Try and reuse skbs already allocated...
343  */
344 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
345 {
346 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
347 		skb_trim(skb, 0);
348 		skb_get(skb);
349 		skb_reset_transport_header(skb);
350 	} else {
351 		skb = alloc_skb(len, gfp);
352 	}
353 	t4_set_arp_err_handler(skb, NULL, NULL);
354 	return skb;
355 }
356 
357 static struct net_device *get_real_dev(struct net_device *egress_dev)
358 {
359 	return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
360 }
361 
362 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
363 {
364 	int i;
365 
366 	egress_dev = get_real_dev(egress_dev);
367 	for (i = 0; i < dev->rdev.lldi.nports; i++)
368 		if (dev->rdev.lldi.ports[i] == egress_dev)
369 			return 1;
370 	return 0;
371 }
372 
373 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
374 				     __u8 *peer_ip, __be16 local_port,
375 				     __be16 peer_port, u8 tos,
376 				     __u32 sin6_scope_id)
377 {
378 	struct dst_entry *dst = NULL;
379 
380 	if (IS_ENABLED(CONFIG_IPV6)) {
381 		struct flowi6 fl6;
382 
383 		memset(&fl6, 0, sizeof(fl6));
384 		memcpy(&fl6.daddr, peer_ip, 16);
385 		memcpy(&fl6.saddr, local_ip, 16);
386 		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
387 			fl6.flowi6_oif = sin6_scope_id;
388 		dst = ip6_route_output(&init_net, NULL, &fl6);
389 		if (!dst)
390 			goto out;
391 		if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
392 		    !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
393 			dst_release(dst);
394 			dst = NULL;
395 		}
396 	}
397 
398 out:
399 	return dst;
400 }
401 
402 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
403 				 __be32 peer_ip, __be16 local_port,
404 				 __be16 peer_port, u8 tos)
405 {
406 	struct rtable *rt;
407 	struct flowi4 fl4;
408 	struct neighbour *n;
409 
410 	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
411 				   peer_port, local_port, IPPROTO_TCP,
412 				   tos, 0);
413 	if (IS_ERR(rt))
414 		return NULL;
415 	n = dst_neigh_lookup(&rt->dst, &peer_ip);
416 	if (!n)
417 		return NULL;
418 	if (!our_interface(dev, n->dev) &&
419 	    !(n->dev->flags & IFF_LOOPBACK)) {
420 		neigh_release(n);
421 		dst_release(&rt->dst);
422 		return NULL;
423 	}
424 	neigh_release(n);
425 	return &rt->dst;
426 }
427 
428 static void arp_failure_discard(void *handle, struct sk_buff *skb)
429 {
430 	PDBG("%s c4iw_dev %p\n", __func__, handle);
431 	kfree_skb(skb);
432 }
433 
434 /*
435  * Handle an ARP failure for an active open.
436  */
437 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
438 {
439 	struct c4iw_ep *ep = handle;
440 
441 	printk(KERN_ERR MOD "ARP failure duing connect\n");
442 	kfree_skb(skb);
443 	connect_reply_upcall(ep, -EHOSTUNREACH);
444 	state_set(&ep->com, DEAD);
445 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
446 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
447 	dst_release(ep->dst);
448 	cxgb4_l2t_release(ep->l2t);
449 	c4iw_put_ep(&ep->com);
450 }
451 
452 /*
453  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
454  * and send it along.
455  */
456 static void abort_arp_failure(void *handle, struct sk_buff *skb)
457 {
458 	struct c4iw_rdev *rdev = handle;
459 	struct cpl_abort_req *req = cplhdr(skb);
460 
461 	PDBG("%s rdev %p\n", __func__, rdev);
462 	req->cmd = CPL_ABORT_NO_RST;
463 	c4iw_ofld_send(rdev, skb);
464 }
465 
466 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
467 {
468 	unsigned int flowclen = 80;
469 	struct fw_flowc_wr *flowc;
470 	int i;
471 
472 	skb = get_skb(skb, flowclen, GFP_KERNEL);
473 	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
474 
475 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
476 					   FW_FLOWC_WR_NPARAMS(8));
477 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
478 					  16)) | FW_WR_FLOWID(ep->hwtid));
479 
480 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
481 	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN
482 					    (ep->com.dev->rdev.lldi.pf));
483 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
484 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
485 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
486 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
487 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
488 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
489 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
490 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
491 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
492 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
493 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
494 	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
495 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
496 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
497 	/* Pad WR to 16 byte boundary */
498 	flowc->mnemval[8].mnemonic = 0;
499 	flowc->mnemval[8].val = 0;
500 	for (i = 0; i < 9; i++) {
501 		flowc->mnemval[i].r4[0] = 0;
502 		flowc->mnemval[i].r4[1] = 0;
503 		flowc->mnemval[i].r4[2] = 0;
504 	}
505 
506 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
507 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
508 }
509 
510 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
511 {
512 	struct cpl_close_con_req *req;
513 	struct sk_buff *skb;
514 	int wrlen = roundup(sizeof *req, 16);
515 
516 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
517 	skb = get_skb(NULL, wrlen, gfp);
518 	if (!skb) {
519 		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
520 		return -ENOMEM;
521 	}
522 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
523 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
524 	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
525 	memset(req, 0, wrlen);
526 	INIT_TP_WR(req, ep->hwtid);
527 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
528 						    ep->hwtid));
529 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
530 }
531 
532 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
533 {
534 	struct cpl_abort_req *req;
535 	int wrlen = roundup(sizeof *req, 16);
536 
537 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
538 	skb = get_skb(skb, wrlen, gfp);
539 	if (!skb) {
540 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
541 		       __func__);
542 		return -ENOMEM;
543 	}
544 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
545 	t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
546 	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
547 	memset(req, 0, wrlen);
548 	INIT_TP_WR(req, ep->hwtid);
549 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
550 	req->cmd = CPL_ABORT_SEND_RST;
551 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
552 }
553 
554 /*
555  * c4iw_form_pm_msg - Form a port mapper message with mapping info
556  */
557 static void c4iw_form_pm_msg(struct c4iw_ep *ep,
558 				struct iwpm_sa_data *pm_msg)
559 {
560 	memcpy(&pm_msg->loc_addr, &ep->com.local_addr,
561 		sizeof(ep->com.local_addr));
562 	memcpy(&pm_msg->rem_addr, &ep->com.remote_addr,
563 		sizeof(ep->com.remote_addr));
564 }
565 
566 /*
567  * c4iw_form_reg_msg - Form a port mapper message with dev info
568  */
569 static void c4iw_form_reg_msg(struct c4iw_dev *dev,
570 				struct iwpm_dev_data *pm_msg)
571 {
572 	memcpy(pm_msg->dev_name, dev->ibdev.name, IWPM_DEVNAME_SIZE);
573 	memcpy(pm_msg->if_name, dev->rdev.lldi.ports[0]->name,
574 				IWPM_IFNAME_SIZE);
575 }
576 
577 static void c4iw_record_pm_msg(struct c4iw_ep *ep,
578 			struct iwpm_sa_data *pm_msg)
579 {
580 	memcpy(&ep->com.mapped_local_addr, &pm_msg->mapped_loc_addr,
581 		sizeof(ep->com.mapped_local_addr));
582 	memcpy(&ep->com.mapped_remote_addr, &pm_msg->mapped_rem_addr,
583 		sizeof(ep->com.mapped_remote_addr));
584 }
585 
586 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
587 		     unsigned int *idx, int use_ts, int ipv6)
588 {
589 	unsigned short hdr_size = (ipv6 ?
590 				   sizeof(struct ipv6hdr) :
591 				   sizeof(struct iphdr)) +
592 				  sizeof(struct tcphdr) +
593 				  (use_ts ?
594 				   round_up(TCPOLEN_TIMESTAMP, 4) : 0);
595 	unsigned short data_size = mtu - hdr_size;
596 
597 	cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
598 }
599 
600 static int send_connect(struct c4iw_ep *ep)
601 {
602 	struct cpl_act_open_req *req;
603 	struct cpl_t5_act_open_req *t5_req;
604 	struct cpl_act_open_req6 *req6;
605 	struct cpl_t5_act_open_req6 *t5_req6;
606 	struct sk_buff *skb;
607 	u64 opt0;
608 	u32 opt2;
609 	unsigned int mtu_idx;
610 	int wscale;
611 	int wrlen;
612 	int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
613 				sizeof(struct cpl_act_open_req) :
614 				sizeof(struct cpl_t5_act_open_req);
615 	int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
616 				sizeof(struct cpl_act_open_req6) :
617 				sizeof(struct cpl_t5_act_open_req6);
618 	struct sockaddr_in *la = (struct sockaddr_in *)
619 				 &ep->com.mapped_local_addr;
620 	struct sockaddr_in *ra = (struct sockaddr_in *)
621 				 &ep->com.mapped_remote_addr;
622 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
623 				   &ep->com.mapped_local_addr;
624 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
625 				   &ep->com.mapped_remote_addr;
626 	int win;
627 
628 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
629 			roundup(sizev4, 16) :
630 			roundup(sizev6, 16);
631 
632 	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
633 
634 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
635 	if (!skb) {
636 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
637 		       __func__);
638 		return -ENOMEM;
639 	}
640 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
641 
642 	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
643 		 enable_tcp_timestamps,
644 		 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
645 	wscale = compute_wscale(rcv_win);
646 
647 	/*
648 	 * Specify the largest window that will fit in opt0. The
649 	 * remainder will be specified in the rx_data_ack.
650 	 */
651 	win = ep->rcv_win >> 10;
652 	if (win > RCV_BUFSIZ_MASK)
653 		win = RCV_BUFSIZ_MASK;
654 
655 	opt0 = (nocong ? NO_CONG(1) : 0) |
656 	       KEEP_ALIVE(1) |
657 	       DELACK(1) |
658 	       WND_SCALE(wscale) |
659 	       MSS_IDX(mtu_idx) |
660 	       L2T_IDX(ep->l2t->idx) |
661 	       TX_CHAN(ep->tx_chan) |
662 	       SMAC_SEL(ep->smac_idx) |
663 	       DSCP(ep->tos) |
664 	       ULP_MODE(ULP_MODE_TCPDDP) |
665 	       RCV_BUFSIZ(win);
666 	opt2 = RX_CHANNEL(0) |
667 	       CCTRL_ECN(enable_ecn) |
668 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
669 	if (enable_tcp_timestamps)
670 		opt2 |= TSTAMPS_EN(1);
671 	if (enable_tcp_sack)
672 		opt2 |= SACK_EN(1);
673 	if (wscale && enable_tcp_window_scaling)
674 		opt2 |= WND_SCALE_EN(1);
675 	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
676 		opt2 |= T5_OPT_2_VALID;
677 		opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
678 		opt2 |= CONG_CNTRL_VALID; /* OPT_2_ISS for T5 */
679 	}
680 	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
681 
682 	if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
683 		if (ep->com.remote_addr.ss_family == AF_INET) {
684 			req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
685 			INIT_TP_WR(req, 0);
686 			OPCODE_TID(req) = cpu_to_be32(
687 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
688 					((ep->rss_qid << 14) | ep->atid)));
689 			req->local_port = la->sin_port;
690 			req->peer_port = ra->sin_port;
691 			req->local_ip = la->sin_addr.s_addr;
692 			req->peer_ip = ra->sin_addr.s_addr;
693 			req->opt0 = cpu_to_be64(opt0);
694 			req->params = cpu_to_be32(cxgb4_select_ntuple(
695 						ep->com.dev->rdev.lldi.ports[0],
696 						ep->l2t));
697 			req->opt2 = cpu_to_be32(opt2);
698 		} else {
699 			req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
700 
701 			INIT_TP_WR(req6, 0);
702 			OPCODE_TID(req6) = cpu_to_be32(
703 					   MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
704 					   ((ep->rss_qid<<14)|ep->atid)));
705 			req6->local_port = la6->sin6_port;
706 			req6->peer_port = ra6->sin6_port;
707 			req6->local_ip_hi = *((__be64 *)
708 						(la6->sin6_addr.s6_addr));
709 			req6->local_ip_lo = *((__be64 *)
710 						(la6->sin6_addr.s6_addr + 8));
711 			req6->peer_ip_hi = *((__be64 *)
712 						(ra6->sin6_addr.s6_addr));
713 			req6->peer_ip_lo = *((__be64 *)
714 						(ra6->sin6_addr.s6_addr + 8));
715 			req6->opt0 = cpu_to_be64(opt0);
716 			req6->params = cpu_to_be32(cxgb4_select_ntuple(
717 						ep->com.dev->rdev.lldi.ports[0],
718 						ep->l2t));
719 			req6->opt2 = cpu_to_be32(opt2);
720 		}
721 	} else {
722 		u32 isn = (prandom_u32() & ~7UL) - 1;
723 
724 		if (peer2peer)
725 			isn += 4;
726 
727 		if (ep->com.remote_addr.ss_family == AF_INET) {
728 			t5_req = (struct cpl_t5_act_open_req *)
729 				 skb_put(skb, wrlen);
730 			INIT_TP_WR(t5_req, 0);
731 			OPCODE_TID(t5_req) = cpu_to_be32(
732 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
733 					((ep->rss_qid << 14) | ep->atid)));
734 			t5_req->local_port = la->sin_port;
735 			t5_req->peer_port = ra->sin_port;
736 			t5_req->local_ip = la->sin_addr.s_addr;
737 			t5_req->peer_ip = ra->sin_addr.s_addr;
738 			t5_req->opt0 = cpu_to_be64(opt0);
739 			t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
740 						     cxgb4_select_ntuple(
741 					     ep->com.dev->rdev.lldi.ports[0],
742 					     ep->l2t)));
743 			t5_req->rsvd = cpu_to_be32(isn);
744 			PDBG("%s snd_isn %u\n", __func__,
745 			     be32_to_cpu(t5_req->rsvd));
746 			t5_req->opt2 = cpu_to_be32(opt2);
747 		} else {
748 			t5_req6 = (struct cpl_t5_act_open_req6 *)
749 				  skb_put(skb, wrlen);
750 			INIT_TP_WR(t5_req6, 0);
751 			OPCODE_TID(t5_req6) = cpu_to_be32(
752 					      MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
753 					      ((ep->rss_qid<<14)|ep->atid)));
754 			t5_req6->local_port = la6->sin6_port;
755 			t5_req6->peer_port = ra6->sin6_port;
756 			t5_req6->local_ip_hi = *((__be64 *)
757 						(la6->sin6_addr.s6_addr));
758 			t5_req6->local_ip_lo = *((__be64 *)
759 						(la6->sin6_addr.s6_addr + 8));
760 			t5_req6->peer_ip_hi = *((__be64 *)
761 						(ra6->sin6_addr.s6_addr));
762 			t5_req6->peer_ip_lo = *((__be64 *)
763 						(ra6->sin6_addr.s6_addr + 8));
764 			t5_req6->opt0 = cpu_to_be64(opt0);
765 			t5_req6->params = cpu_to_be64(V_FILTER_TUPLE(
766 							cxgb4_select_ntuple(
767 						ep->com.dev->rdev.lldi.ports[0],
768 						ep->l2t)));
769 			t5_req6->rsvd = cpu_to_be32(isn);
770 			PDBG("%s snd_isn %u\n", __func__,
771 			     be32_to_cpu(t5_req6->rsvd));
772 			t5_req6->opt2 = cpu_to_be32(opt2);
773 		}
774 	}
775 
776 	set_bit(ACT_OPEN_REQ, &ep->com.history);
777 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
778 }
779 
780 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
781 		u8 mpa_rev_to_use)
782 {
783 	int mpalen, wrlen;
784 	struct fw_ofld_tx_data_wr *req;
785 	struct mpa_message *mpa;
786 	struct mpa_v2_conn_params mpa_v2_params;
787 
788 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
789 
790 	BUG_ON(skb_cloned(skb));
791 
792 	mpalen = sizeof(*mpa) + ep->plen;
793 	if (mpa_rev_to_use == 2)
794 		mpalen += sizeof(struct mpa_v2_conn_params);
795 	wrlen = roundup(mpalen + sizeof *req, 16);
796 	skb = get_skb(skb, wrlen, GFP_KERNEL);
797 	if (!skb) {
798 		connect_reply_upcall(ep, -ENOMEM);
799 		return;
800 	}
801 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
802 
803 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
804 	memset(req, 0, wrlen);
805 	req->op_to_immdlen = cpu_to_be32(
806 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
807 		FW_WR_COMPL(1) |
808 		FW_WR_IMMDLEN(mpalen));
809 	req->flowid_len16 = cpu_to_be32(
810 		FW_WR_FLOWID(ep->hwtid) |
811 		FW_WR_LEN16(wrlen >> 4));
812 	req->plen = cpu_to_be32(mpalen);
813 	req->tunnel_to_proxy = cpu_to_be32(
814 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
815 		FW_OFLD_TX_DATA_WR_SHOVE(1));
816 
817 	mpa = (struct mpa_message *)(req + 1);
818 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
819 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
820 		     (markers_enabled ? MPA_MARKERS : 0) |
821 		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
822 	mpa->private_data_size = htons(ep->plen);
823 	mpa->revision = mpa_rev_to_use;
824 	if (mpa_rev_to_use == 1) {
825 		ep->tried_with_mpa_v1 = 1;
826 		ep->retry_with_mpa_v1 = 0;
827 	}
828 
829 	if (mpa_rev_to_use == 2) {
830 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
831 					       sizeof (struct mpa_v2_conn_params));
832 		PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
833 		     ep->ord);
834 		mpa_v2_params.ird = htons((u16)ep->ird);
835 		mpa_v2_params.ord = htons((u16)ep->ord);
836 
837 		if (peer2peer) {
838 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
839 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
840 				mpa_v2_params.ord |=
841 					htons(MPA_V2_RDMA_WRITE_RTR);
842 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
843 				mpa_v2_params.ord |=
844 					htons(MPA_V2_RDMA_READ_RTR);
845 		}
846 		memcpy(mpa->private_data, &mpa_v2_params,
847 		       sizeof(struct mpa_v2_conn_params));
848 
849 		if (ep->plen)
850 			memcpy(mpa->private_data +
851 			       sizeof(struct mpa_v2_conn_params),
852 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
853 	} else
854 		if (ep->plen)
855 			memcpy(mpa->private_data,
856 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
857 
858 	/*
859 	 * Reference the mpa skb.  This ensures the data area
860 	 * will remain in memory until the hw acks the tx.
861 	 * Function fw4_ack() will deref it.
862 	 */
863 	skb_get(skb);
864 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
865 	BUG_ON(ep->mpa_skb);
866 	ep->mpa_skb = skb;
867 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
868 	start_ep_timer(ep);
869 	__state_set(&ep->com, MPA_REQ_SENT);
870 	ep->mpa_attr.initiator = 1;
871 	ep->snd_seq += mpalen;
872 	return;
873 }
874 
875 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
876 {
877 	int mpalen, wrlen;
878 	struct fw_ofld_tx_data_wr *req;
879 	struct mpa_message *mpa;
880 	struct sk_buff *skb;
881 	struct mpa_v2_conn_params mpa_v2_params;
882 
883 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
884 
885 	mpalen = sizeof(*mpa) + plen;
886 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
887 		mpalen += sizeof(struct mpa_v2_conn_params);
888 	wrlen = roundup(mpalen + sizeof *req, 16);
889 
890 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
891 	if (!skb) {
892 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
893 		return -ENOMEM;
894 	}
895 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
896 
897 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
898 	memset(req, 0, wrlen);
899 	req->op_to_immdlen = cpu_to_be32(
900 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
901 		FW_WR_COMPL(1) |
902 		FW_WR_IMMDLEN(mpalen));
903 	req->flowid_len16 = cpu_to_be32(
904 		FW_WR_FLOWID(ep->hwtid) |
905 		FW_WR_LEN16(wrlen >> 4));
906 	req->plen = cpu_to_be32(mpalen);
907 	req->tunnel_to_proxy = cpu_to_be32(
908 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
909 		FW_OFLD_TX_DATA_WR_SHOVE(1));
910 
911 	mpa = (struct mpa_message *)(req + 1);
912 	memset(mpa, 0, sizeof(*mpa));
913 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
914 	mpa->flags = MPA_REJECT;
915 	mpa->revision = ep->mpa_attr.version;
916 	mpa->private_data_size = htons(plen);
917 
918 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
919 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
920 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
921 					       sizeof (struct mpa_v2_conn_params));
922 		mpa_v2_params.ird = htons(((u16)ep->ird) |
923 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
924 					   0));
925 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
926 					  (p2p_type ==
927 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
928 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
929 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
930 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
931 		memcpy(mpa->private_data, &mpa_v2_params,
932 		       sizeof(struct mpa_v2_conn_params));
933 
934 		if (ep->plen)
935 			memcpy(mpa->private_data +
936 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
937 	} else
938 		if (plen)
939 			memcpy(mpa->private_data, pdata, plen);
940 
941 	/*
942 	 * Reference the mpa skb again.  This ensures the data area
943 	 * will remain in memory until the hw acks the tx.
944 	 * Function fw4_ack() will deref it.
945 	 */
946 	skb_get(skb);
947 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
948 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
949 	BUG_ON(ep->mpa_skb);
950 	ep->mpa_skb = skb;
951 	ep->snd_seq += mpalen;
952 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
953 }
954 
955 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
956 {
957 	int mpalen, wrlen;
958 	struct fw_ofld_tx_data_wr *req;
959 	struct mpa_message *mpa;
960 	struct sk_buff *skb;
961 	struct mpa_v2_conn_params mpa_v2_params;
962 
963 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
964 
965 	mpalen = sizeof(*mpa) + plen;
966 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
967 		mpalen += sizeof(struct mpa_v2_conn_params);
968 	wrlen = roundup(mpalen + sizeof *req, 16);
969 
970 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
971 	if (!skb) {
972 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
973 		return -ENOMEM;
974 	}
975 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
976 
977 	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
978 	memset(req, 0, wrlen);
979 	req->op_to_immdlen = cpu_to_be32(
980 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
981 		FW_WR_COMPL(1) |
982 		FW_WR_IMMDLEN(mpalen));
983 	req->flowid_len16 = cpu_to_be32(
984 		FW_WR_FLOWID(ep->hwtid) |
985 		FW_WR_LEN16(wrlen >> 4));
986 	req->plen = cpu_to_be32(mpalen);
987 	req->tunnel_to_proxy = cpu_to_be32(
988 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
989 		FW_OFLD_TX_DATA_WR_SHOVE(1));
990 
991 	mpa = (struct mpa_message *)(req + 1);
992 	memset(mpa, 0, sizeof(*mpa));
993 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
994 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
995 		     (markers_enabled ? MPA_MARKERS : 0);
996 	mpa->revision = ep->mpa_attr.version;
997 	mpa->private_data_size = htons(plen);
998 
999 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1000 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1001 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1002 					       sizeof (struct mpa_v2_conn_params));
1003 		mpa_v2_params.ird = htons((u16)ep->ird);
1004 		mpa_v2_params.ord = htons((u16)ep->ord);
1005 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1006 					FW_RI_INIT_P2PTYPE_DISABLED)) {
1007 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1008 
1009 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1010 				mpa_v2_params.ord |=
1011 					htons(MPA_V2_RDMA_WRITE_RTR);
1012 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1013 				mpa_v2_params.ord |=
1014 					htons(MPA_V2_RDMA_READ_RTR);
1015 		}
1016 
1017 		memcpy(mpa->private_data, &mpa_v2_params,
1018 		       sizeof(struct mpa_v2_conn_params));
1019 
1020 		if (ep->plen)
1021 			memcpy(mpa->private_data +
1022 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1023 	} else
1024 		if (plen)
1025 			memcpy(mpa->private_data, pdata, plen);
1026 
1027 	/*
1028 	 * Reference the mpa skb.  This ensures the data area
1029 	 * will remain in memory until the hw acks the tx.
1030 	 * Function fw4_ack() will deref it.
1031 	 */
1032 	skb_get(skb);
1033 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1034 	ep->mpa_skb = skb;
1035 	__state_set(&ep->com, MPA_REP_SENT);
1036 	ep->snd_seq += mpalen;
1037 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1038 }
1039 
1040 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1041 {
1042 	struct c4iw_ep *ep;
1043 	struct cpl_act_establish *req = cplhdr(skb);
1044 	unsigned int tid = GET_TID(req);
1045 	unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
1046 	struct tid_info *t = dev->rdev.lldi.tids;
1047 
1048 	ep = lookup_atid(t, atid);
1049 
1050 	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1051 	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1052 
1053 	mutex_lock(&ep->com.mutex);
1054 	dst_confirm(ep->dst);
1055 
1056 	/* setup the hwtid for this connection */
1057 	ep->hwtid = tid;
1058 	cxgb4_insert_tid(t, ep, tid);
1059 	insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
1060 
1061 	ep->snd_seq = be32_to_cpu(req->snd_isn);
1062 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1063 
1064 	set_emss(ep, ntohs(req->tcp_opt));
1065 
1066 	/* dealloc the atid */
1067 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1068 	cxgb4_free_atid(t, atid);
1069 	set_bit(ACT_ESTAB, &ep->com.history);
1070 
1071 	/* start MPA negotiation */
1072 	send_flowc(ep, NULL);
1073 	if (ep->retry_with_mpa_v1)
1074 		send_mpa_req(ep, skb, 1);
1075 	else
1076 		send_mpa_req(ep, skb, mpa_rev);
1077 	mutex_unlock(&ep->com.mutex);
1078 	return 0;
1079 }
1080 
1081 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1082 {
1083 	struct iw_cm_event event;
1084 
1085 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1086 	memset(&event, 0, sizeof(event));
1087 	event.event = IW_CM_EVENT_CLOSE;
1088 	event.status = status;
1089 	if (ep->com.cm_id) {
1090 		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1091 		     ep, ep->com.cm_id, ep->hwtid);
1092 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1093 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1094 		ep->com.cm_id = NULL;
1095 		set_bit(CLOSE_UPCALL, &ep->com.history);
1096 	}
1097 }
1098 
1099 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1100 {
1101 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1102 	__state_set(&ep->com, ABORTING);
1103 	set_bit(ABORT_CONN, &ep->com.history);
1104 	return send_abort(ep, skb, gfp);
1105 }
1106 
1107 static void peer_close_upcall(struct c4iw_ep *ep)
1108 {
1109 	struct iw_cm_event event;
1110 
1111 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1112 	memset(&event, 0, sizeof(event));
1113 	event.event = IW_CM_EVENT_DISCONNECT;
1114 	if (ep->com.cm_id) {
1115 		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1116 		     ep, ep->com.cm_id, ep->hwtid);
1117 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1118 		set_bit(DISCONN_UPCALL, &ep->com.history);
1119 	}
1120 }
1121 
1122 static void peer_abort_upcall(struct c4iw_ep *ep)
1123 {
1124 	struct iw_cm_event event;
1125 
1126 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1127 	memset(&event, 0, sizeof(event));
1128 	event.event = IW_CM_EVENT_CLOSE;
1129 	event.status = -ECONNRESET;
1130 	if (ep->com.cm_id) {
1131 		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1132 		     ep->com.cm_id, ep->hwtid);
1133 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1134 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1135 		ep->com.cm_id = NULL;
1136 		set_bit(ABORT_UPCALL, &ep->com.history);
1137 	}
1138 }
1139 
1140 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1141 {
1142 	struct iw_cm_event event;
1143 
1144 	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1145 	memset(&event, 0, sizeof(event));
1146 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1147 	event.status = status;
1148 	memcpy(&event.local_addr, &ep->com.local_addr,
1149 	       sizeof(ep->com.local_addr));
1150 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1151 	       sizeof(ep->com.remote_addr));
1152 
1153 	if ((status == 0) || (status == -ECONNREFUSED)) {
1154 		if (!ep->tried_with_mpa_v1) {
1155 			/* this means MPA_v2 is used */
1156 			event.private_data_len = ep->plen -
1157 				sizeof(struct mpa_v2_conn_params);
1158 			event.private_data = ep->mpa_pkt +
1159 				sizeof(struct mpa_message) +
1160 				sizeof(struct mpa_v2_conn_params);
1161 		} else {
1162 			/* this means MPA_v1 is used */
1163 			event.private_data_len = ep->plen;
1164 			event.private_data = ep->mpa_pkt +
1165 				sizeof(struct mpa_message);
1166 		}
1167 	}
1168 
1169 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1170 	     ep->hwtid, status);
1171 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1172 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1173 
1174 	if (status < 0) {
1175 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1176 		ep->com.cm_id = NULL;
1177 	}
1178 }
1179 
1180 static int connect_request_upcall(struct c4iw_ep *ep)
1181 {
1182 	struct iw_cm_event event;
1183 	int ret;
1184 
1185 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1186 	memset(&event, 0, sizeof(event));
1187 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1188 	memcpy(&event.local_addr, &ep->com.local_addr,
1189 	       sizeof(ep->com.local_addr));
1190 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1191 	       sizeof(ep->com.remote_addr));
1192 	event.provider_data = ep;
1193 	if (!ep->tried_with_mpa_v1) {
1194 		/* this means MPA_v2 is used */
1195 		event.ord = ep->ord;
1196 		event.ird = ep->ird;
1197 		event.private_data_len = ep->plen -
1198 			sizeof(struct mpa_v2_conn_params);
1199 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1200 			sizeof(struct mpa_v2_conn_params);
1201 	} else {
1202 		/* this means MPA_v1 is used. Send max supported */
1203 		event.ord = cur_max_read_depth(ep->com.dev);
1204 		event.ird = cur_max_read_depth(ep->com.dev);
1205 		event.private_data_len = ep->plen;
1206 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1207 	}
1208 	c4iw_get_ep(&ep->com);
1209 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1210 						      &event);
1211 	if (ret)
1212 		c4iw_put_ep(&ep->com);
1213 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1214 	c4iw_put_ep(&ep->parent_ep->com);
1215 	return ret;
1216 }
1217 
1218 static void established_upcall(struct c4iw_ep *ep)
1219 {
1220 	struct iw_cm_event event;
1221 
1222 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1223 	memset(&event, 0, sizeof(event));
1224 	event.event = IW_CM_EVENT_ESTABLISHED;
1225 	event.ird = ep->ird;
1226 	event.ord = ep->ord;
1227 	if (ep->com.cm_id) {
1228 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1229 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1230 		set_bit(ESTAB_UPCALL, &ep->com.history);
1231 	}
1232 }
1233 
1234 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1235 {
1236 	struct cpl_rx_data_ack *req;
1237 	struct sk_buff *skb;
1238 	int wrlen = roundup(sizeof *req, 16);
1239 
1240 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1241 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1242 	if (!skb) {
1243 		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1244 		return 0;
1245 	}
1246 
1247 	/*
1248 	 * If we couldn't specify the entire rcv window at connection setup
1249 	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1250 	 * then add the overage in to the credits returned.
1251 	 */
1252 	if (ep->rcv_win > RCV_BUFSIZ_MASK * 1024)
1253 		credits += ep->rcv_win - RCV_BUFSIZ_MASK * 1024;
1254 
1255 	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1256 	memset(req, 0, wrlen);
1257 	INIT_TP_WR(req, ep->hwtid);
1258 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1259 						    ep->hwtid));
1260 	req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1261 				       F_RX_DACK_CHANGE |
1262 				       V_RX_DACK_MODE(dack_mode));
1263 	set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1264 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1265 	return credits;
1266 }
1267 
1268 #define RELAXED_IRD_NEGOTIATION 1
1269 
1270 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1271 {
1272 	struct mpa_message *mpa;
1273 	struct mpa_v2_conn_params *mpa_v2_params;
1274 	u16 plen;
1275 	u16 resp_ird, resp_ord;
1276 	u8 rtr_mismatch = 0, insuff_ird = 0;
1277 	struct c4iw_qp_attributes attrs;
1278 	enum c4iw_qp_attr_mask mask;
1279 	int err;
1280 	int disconnect = 0;
1281 
1282 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1283 
1284 	/*
1285 	 * Stop mpa timer.  If it expired, then
1286 	 * we ignore the MPA reply.  process_timeout()
1287 	 * will abort the connection.
1288 	 */
1289 	if (stop_ep_timer(ep))
1290 		return 0;
1291 
1292 	/*
1293 	 * If we get more than the supported amount of private data
1294 	 * then we must fail this connection.
1295 	 */
1296 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1297 		err = -EINVAL;
1298 		goto err;
1299 	}
1300 
1301 	/*
1302 	 * copy the new data into our accumulation buffer.
1303 	 */
1304 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1305 				  skb->len);
1306 	ep->mpa_pkt_len += skb->len;
1307 
1308 	/*
1309 	 * if we don't even have the mpa message, then bail.
1310 	 */
1311 	if (ep->mpa_pkt_len < sizeof(*mpa))
1312 		return 0;
1313 	mpa = (struct mpa_message *) ep->mpa_pkt;
1314 
1315 	/* Validate MPA header. */
1316 	if (mpa->revision > mpa_rev) {
1317 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1318 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1319 		err = -EPROTO;
1320 		goto err;
1321 	}
1322 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1323 		err = -EPROTO;
1324 		goto err;
1325 	}
1326 
1327 	plen = ntohs(mpa->private_data_size);
1328 
1329 	/*
1330 	 * Fail if there's too much private data.
1331 	 */
1332 	if (plen > MPA_MAX_PRIVATE_DATA) {
1333 		err = -EPROTO;
1334 		goto err;
1335 	}
1336 
1337 	/*
1338 	 * If plen does not account for pkt size
1339 	 */
1340 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1341 		err = -EPROTO;
1342 		goto err;
1343 	}
1344 
1345 	ep->plen = (u8) plen;
1346 
1347 	/*
1348 	 * If we don't have all the pdata yet, then bail.
1349 	 * We'll continue process when more data arrives.
1350 	 */
1351 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1352 		return 0;
1353 
1354 	if (mpa->flags & MPA_REJECT) {
1355 		err = -ECONNREFUSED;
1356 		goto err;
1357 	}
1358 
1359 	/*
1360 	 * If we get here we have accumulated the entire mpa
1361 	 * start reply message including private data. And
1362 	 * the MPA header is valid.
1363 	 */
1364 	__state_set(&ep->com, FPDU_MODE);
1365 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1366 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1367 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1368 	ep->mpa_attr.version = mpa->revision;
1369 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1370 
1371 	if (mpa->revision == 2) {
1372 		ep->mpa_attr.enhanced_rdma_conn =
1373 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1374 		if (ep->mpa_attr.enhanced_rdma_conn) {
1375 			mpa_v2_params = (struct mpa_v2_conn_params *)
1376 				(ep->mpa_pkt + sizeof(*mpa));
1377 			resp_ird = ntohs(mpa_v2_params->ird) &
1378 				MPA_V2_IRD_ORD_MASK;
1379 			resp_ord = ntohs(mpa_v2_params->ord) &
1380 				MPA_V2_IRD_ORD_MASK;
1381 			PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1382 			     __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1383 
1384 			/*
1385 			 * This is a double-check. Ideally, below checks are
1386 			 * not required since ird/ord stuff has been taken
1387 			 * care of in c4iw_accept_cr
1388 			 */
1389 			if (ep->ird < resp_ord) {
1390 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1391 				    ep->com.dev->rdev.lldi.max_ordird_qp)
1392 					ep->ird = resp_ord;
1393 				else
1394 					insuff_ird = 1;
1395 			} else if (ep->ird > resp_ord) {
1396 				ep->ird = resp_ord;
1397 			}
1398 			if (ep->ord > resp_ird) {
1399 				if (RELAXED_IRD_NEGOTIATION)
1400 					ep->ord = resp_ird;
1401 				else
1402 					insuff_ird = 1;
1403 			}
1404 			if (insuff_ird) {
1405 				err = -ENOMEM;
1406 				ep->ird = resp_ord;
1407 				ep->ord = resp_ird;
1408 			}
1409 
1410 			if (ntohs(mpa_v2_params->ird) &
1411 					MPA_V2_PEER2PEER_MODEL) {
1412 				if (ntohs(mpa_v2_params->ord) &
1413 						MPA_V2_RDMA_WRITE_RTR)
1414 					ep->mpa_attr.p2p_type =
1415 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1416 				else if (ntohs(mpa_v2_params->ord) &
1417 						MPA_V2_RDMA_READ_RTR)
1418 					ep->mpa_attr.p2p_type =
1419 						FW_RI_INIT_P2PTYPE_READ_REQ;
1420 			}
1421 		}
1422 	} else if (mpa->revision == 1)
1423 		if (peer2peer)
1424 			ep->mpa_attr.p2p_type = p2p_type;
1425 
1426 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1427 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1428 	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
1429 	     ep->mpa_attr.recv_marker_enabled,
1430 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1431 	     ep->mpa_attr.p2p_type, p2p_type);
1432 
1433 	/*
1434 	 * If responder's RTR does not match with that of initiator, assign
1435 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1436 	 * generated when moving QP to RTS state.
1437 	 * A TERM message will be sent after QP has moved to RTS state
1438 	 */
1439 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1440 			(ep->mpa_attr.p2p_type != p2p_type)) {
1441 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1442 		rtr_mismatch = 1;
1443 	}
1444 
1445 	attrs.mpa_attr = ep->mpa_attr;
1446 	attrs.max_ird = ep->ird;
1447 	attrs.max_ord = ep->ord;
1448 	attrs.llp_stream_handle = ep;
1449 	attrs.next_state = C4IW_QP_STATE_RTS;
1450 
1451 	mask = C4IW_QP_ATTR_NEXT_STATE |
1452 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1453 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1454 
1455 	/* bind QP and TID with INIT_WR */
1456 	err = c4iw_modify_qp(ep->com.qp->rhp,
1457 			     ep->com.qp, mask, &attrs, 1);
1458 	if (err)
1459 		goto err;
1460 
1461 	/*
1462 	 * If responder's RTR requirement did not match with what initiator
1463 	 * supports, generate TERM message
1464 	 */
1465 	if (rtr_mismatch) {
1466 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1467 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1468 		attrs.ecode = MPA_NOMATCH_RTR;
1469 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1470 		attrs.send_term = 1;
1471 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1472 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1473 		err = -ENOMEM;
1474 		disconnect = 1;
1475 		goto out;
1476 	}
1477 
1478 	/*
1479 	 * Generate TERM if initiator IRD is not sufficient for responder
1480 	 * provided ORD. Currently, we do the same behaviour even when
1481 	 * responder provided IRD is also not sufficient as regards to
1482 	 * initiator ORD.
1483 	 */
1484 	if (insuff_ird) {
1485 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1486 				__func__);
1487 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1488 		attrs.ecode = MPA_INSUFF_IRD;
1489 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1490 		attrs.send_term = 1;
1491 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1492 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1493 		err = -ENOMEM;
1494 		disconnect = 1;
1495 		goto out;
1496 	}
1497 	goto out;
1498 err:
1499 	__state_set(&ep->com, ABORTING);
1500 	send_abort(ep, skb, GFP_KERNEL);
1501 out:
1502 	connect_reply_upcall(ep, err);
1503 	return disconnect;
1504 }
1505 
1506 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1507 {
1508 	struct mpa_message *mpa;
1509 	struct mpa_v2_conn_params *mpa_v2_params;
1510 	u16 plen;
1511 
1512 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1513 
1514 	/*
1515 	 * If we get more than the supported amount of private data
1516 	 * then we must fail this connection.
1517 	 */
1518 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1519 		(void)stop_ep_timer(ep);
1520 		abort_connection(ep, skb, GFP_KERNEL);
1521 		return;
1522 	}
1523 
1524 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1525 
1526 	/*
1527 	 * Copy the new data into our accumulation buffer.
1528 	 */
1529 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1530 				  skb->len);
1531 	ep->mpa_pkt_len += skb->len;
1532 
1533 	/*
1534 	 * If we don't even have the mpa message, then bail.
1535 	 * We'll continue process when more data arrives.
1536 	 */
1537 	if (ep->mpa_pkt_len < sizeof(*mpa))
1538 		return;
1539 
1540 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1541 	mpa = (struct mpa_message *) ep->mpa_pkt;
1542 
1543 	/*
1544 	 * Validate MPA Header.
1545 	 */
1546 	if (mpa->revision > mpa_rev) {
1547 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1548 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1549 		(void)stop_ep_timer(ep);
1550 		abort_connection(ep, skb, GFP_KERNEL);
1551 		return;
1552 	}
1553 
1554 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1555 		(void)stop_ep_timer(ep);
1556 		abort_connection(ep, skb, GFP_KERNEL);
1557 		return;
1558 	}
1559 
1560 	plen = ntohs(mpa->private_data_size);
1561 
1562 	/*
1563 	 * Fail if there's too much private data.
1564 	 */
1565 	if (plen > MPA_MAX_PRIVATE_DATA) {
1566 		(void)stop_ep_timer(ep);
1567 		abort_connection(ep, skb, GFP_KERNEL);
1568 		return;
1569 	}
1570 
1571 	/*
1572 	 * If plen does not account for pkt size
1573 	 */
1574 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1575 		(void)stop_ep_timer(ep);
1576 		abort_connection(ep, skb, GFP_KERNEL);
1577 		return;
1578 	}
1579 	ep->plen = (u8) plen;
1580 
1581 	/*
1582 	 * If we don't have all the pdata yet, then bail.
1583 	 */
1584 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1585 		return;
1586 
1587 	/*
1588 	 * If we get here we have accumulated the entire mpa
1589 	 * start reply message including private data.
1590 	 */
1591 	ep->mpa_attr.initiator = 0;
1592 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1593 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1594 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1595 	ep->mpa_attr.version = mpa->revision;
1596 	if (mpa->revision == 1)
1597 		ep->tried_with_mpa_v1 = 1;
1598 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1599 
1600 	if (mpa->revision == 2) {
1601 		ep->mpa_attr.enhanced_rdma_conn =
1602 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1603 		if (ep->mpa_attr.enhanced_rdma_conn) {
1604 			mpa_v2_params = (struct mpa_v2_conn_params *)
1605 				(ep->mpa_pkt + sizeof(*mpa));
1606 			ep->ird = ntohs(mpa_v2_params->ird) &
1607 				MPA_V2_IRD_ORD_MASK;
1608 			ep->ord = ntohs(mpa_v2_params->ord) &
1609 				MPA_V2_IRD_ORD_MASK;
1610 			PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1611 			     ep->ord);
1612 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1613 				if (peer2peer) {
1614 					if (ntohs(mpa_v2_params->ord) &
1615 							MPA_V2_RDMA_WRITE_RTR)
1616 						ep->mpa_attr.p2p_type =
1617 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1618 					else if (ntohs(mpa_v2_params->ord) &
1619 							MPA_V2_RDMA_READ_RTR)
1620 						ep->mpa_attr.p2p_type =
1621 						FW_RI_INIT_P2PTYPE_READ_REQ;
1622 				}
1623 		}
1624 	} else if (mpa->revision == 1)
1625 		if (peer2peer)
1626 			ep->mpa_attr.p2p_type = p2p_type;
1627 
1628 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1629 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1630 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1631 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1632 	     ep->mpa_attr.p2p_type);
1633 
1634 	/*
1635 	 * If the endpoint timer already expired, then we ignore
1636 	 * the start request.  process_timeout() will abort
1637 	 * the connection.
1638 	 */
1639 	if (!stop_ep_timer(ep)) {
1640 		__state_set(&ep->com, MPA_REQ_RCVD);
1641 
1642 		/* drive upcall */
1643 		mutex_lock(&ep->parent_ep->com.mutex);
1644 		if (ep->parent_ep->com.state != DEAD) {
1645 			if (connect_request_upcall(ep))
1646 				abort_connection(ep, skb, GFP_KERNEL);
1647 		} else {
1648 			abort_connection(ep, skb, GFP_KERNEL);
1649 		}
1650 		mutex_unlock(&ep->parent_ep->com.mutex);
1651 	}
1652 	return;
1653 }
1654 
1655 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1656 {
1657 	struct c4iw_ep *ep;
1658 	struct cpl_rx_data *hdr = cplhdr(skb);
1659 	unsigned int dlen = ntohs(hdr->len);
1660 	unsigned int tid = GET_TID(hdr);
1661 	struct tid_info *t = dev->rdev.lldi.tids;
1662 	__u8 status = hdr->status;
1663 	int disconnect = 0;
1664 
1665 	ep = lookup_tid(t, tid);
1666 	if (!ep)
1667 		return 0;
1668 	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1669 	skb_pull(skb, sizeof(*hdr));
1670 	skb_trim(skb, dlen);
1671 	mutex_lock(&ep->com.mutex);
1672 
1673 	/* update RX credits */
1674 	update_rx_credits(ep, dlen);
1675 
1676 	switch (ep->com.state) {
1677 	case MPA_REQ_SENT:
1678 		ep->rcv_seq += dlen;
1679 		disconnect = process_mpa_reply(ep, skb);
1680 		break;
1681 	case MPA_REQ_WAIT:
1682 		ep->rcv_seq += dlen;
1683 		process_mpa_request(ep, skb);
1684 		break;
1685 	case FPDU_MODE: {
1686 		struct c4iw_qp_attributes attrs;
1687 		BUG_ON(!ep->com.qp);
1688 		if (status)
1689 			pr_err("%s Unexpected streaming data." \
1690 			       " qpid %u ep %p state %d tid %u status %d\n",
1691 			       __func__, ep->com.qp->wq.sq.qid, ep,
1692 			       ep->com.state, ep->hwtid, status);
1693 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1694 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1695 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1696 		disconnect = 1;
1697 		break;
1698 	}
1699 	default:
1700 		break;
1701 	}
1702 	mutex_unlock(&ep->com.mutex);
1703 	if (disconnect)
1704 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1705 	return 0;
1706 }
1707 
1708 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1709 {
1710 	struct c4iw_ep *ep;
1711 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1712 	int release = 0;
1713 	unsigned int tid = GET_TID(rpl);
1714 	struct tid_info *t = dev->rdev.lldi.tids;
1715 
1716 	ep = lookup_tid(t, tid);
1717 	if (!ep) {
1718 		printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1719 		return 0;
1720 	}
1721 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1722 	mutex_lock(&ep->com.mutex);
1723 	switch (ep->com.state) {
1724 	case ABORTING:
1725 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1726 		__state_set(&ep->com, DEAD);
1727 		release = 1;
1728 		break;
1729 	default:
1730 		printk(KERN_ERR "%s ep %p state %d\n",
1731 		     __func__, ep, ep->com.state);
1732 		break;
1733 	}
1734 	mutex_unlock(&ep->com.mutex);
1735 
1736 	if (release)
1737 		release_ep_resources(ep);
1738 	return 0;
1739 }
1740 
1741 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1742 {
1743 	struct sk_buff *skb;
1744 	struct fw_ofld_connection_wr *req;
1745 	unsigned int mtu_idx;
1746 	int wscale;
1747 	struct sockaddr_in *sin;
1748 	int win;
1749 
1750 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1751 	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1752 	memset(req, 0, sizeof(*req));
1753 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1754 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1755 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1756 				     ep->com.dev->rdev.lldi.ports[0],
1757 				     ep->l2t));
1758 	sin = (struct sockaddr_in *)&ep->com.mapped_local_addr;
1759 	req->le.lport = sin->sin_port;
1760 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1761 	sin = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
1762 	req->le.pport = sin->sin_port;
1763 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1764 	req->tcb.t_state_to_astid =
1765 			htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1766 			V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1767 	req->tcb.cplrxdataack_cplpassacceptrpl =
1768 			htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1769 	req->tcb.tx_max = (__force __be32) jiffies;
1770 	req->tcb.rcv_adv = htons(1);
1771 	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1772 		 enable_tcp_timestamps,
1773 		 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
1774 	wscale = compute_wscale(rcv_win);
1775 
1776 	/*
1777 	 * Specify the largest window that will fit in opt0. The
1778 	 * remainder will be specified in the rx_data_ack.
1779 	 */
1780 	win = ep->rcv_win >> 10;
1781 	if (win > RCV_BUFSIZ_MASK)
1782 		win = RCV_BUFSIZ_MASK;
1783 
1784 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1785 		(nocong ? NO_CONG(1) : 0) |
1786 		KEEP_ALIVE(1) |
1787 		DELACK(1) |
1788 		WND_SCALE(wscale) |
1789 		MSS_IDX(mtu_idx) |
1790 		L2T_IDX(ep->l2t->idx) |
1791 		TX_CHAN(ep->tx_chan) |
1792 		SMAC_SEL(ep->smac_idx) |
1793 		DSCP(ep->tos) |
1794 		ULP_MODE(ULP_MODE_TCPDDP) |
1795 		RCV_BUFSIZ(win));
1796 	req->tcb.opt2 = (__force __be32) (PACE(1) |
1797 		TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1798 		RX_CHANNEL(0) |
1799 		CCTRL_ECN(enable_ecn) |
1800 		RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1801 	if (enable_tcp_timestamps)
1802 		req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1803 	if (enable_tcp_sack)
1804 		req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1805 	if (wscale && enable_tcp_window_scaling)
1806 		req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1807 	req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1808 	req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1809 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1810 	set_bit(ACT_OFLD_CONN, &ep->com.history);
1811 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1812 }
1813 
1814 /*
1815  * Return whether a failed active open has allocated a TID
1816  */
1817 static inline int act_open_has_tid(int status)
1818 {
1819 	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1820 	       status != CPL_ERR_ARP_MISS;
1821 }
1822 
1823 /* Returns whether a CPL status conveys negative advice.
1824  */
1825 static int is_neg_adv(unsigned int status)
1826 {
1827 	return status == CPL_ERR_RTX_NEG_ADVICE ||
1828 	       status == CPL_ERR_PERSIST_NEG_ADVICE ||
1829 	       status == CPL_ERR_KEEPALV_NEG_ADVICE;
1830 }
1831 
1832 static char *neg_adv_str(unsigned int status)
1833 {
1834 	switch (status) {
1835 	case CPL_ERR_RTX_NEG_ADVICE:
1836 		return "Retransmit timeout";
1837 	case CPL_ERR_PERSIST_NEG_ADVICE:
1838 		return "Persist timeout";
1839 	case CPL_ERR_KEEPALV_NEG_ADVICE:
1840 		return "Keepalive timeout";
1841 	default:
1842 		return "Unknown";
1843 	}
1844 }
1845 
1846 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1847 {
1848 	ep->snd_win = snd_win;
1849 	ep->rcv_win = rcv_win;
1850 	PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1851 }
1852 
1853 #define ACT_OPEN_RETRY_COUNT 2
1854 
1855 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1856 		     struct dst_entry *dst, struct c4iw_dev *cdev,
1857 		     bool clear_mpa_v1)
1858 {
1859 	struct neighbour *n;
1860 	int err, step;
1861 	struct net_device *pdev;
1862 
1863 	n = dst_neigh_lookup(dst, peer_ip);
1864 	if (!n)
1865 		return -ENODEV;
1866 
1867 	rcu_read_lock();
1868 	err = -ENOMEM;
1869 	if (n->dev->flags & IFF_LOOPBACK) {
1870 		if (iptype == 4)
1871 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1872 		else if (IS_ENABLED(CONFIG_IPV6))
1873 			for_each_netdev(&init_net, pdev) {
1874 				if (ipv6_chk_addr(&init_net,
1875 						  (struct in6_addr *)peer_ip,
1876 						  pdev, 1))
1877 					break;
1878 			}
1879 		else
1880 			pdev = NULL;
1881 
1882 		if (!pdev) {
1883 			err = -ENODEV;
1884 			goto out;
1885 		}
1886 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1887 					n, pdev, 0);
1888 		if (!ep->l2t)
1889 			goto out;
1890 		ep->mtu = pdev->mtu;
1891 		ep->tx_chan = cxgb4_port_chan(pdev);
1892 		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1893 		step = cdev->rdev.lldi.ntxq /
1894 			cdev->rdev.lldi.nchan;
1895 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1896 		step = cdev->rdev.lldi.nrxq /
1897 			cdev->rdev.lldi.nchan;
1898 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1899 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1900 			cxgb4_port_idx(pdev) * step];
1901 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
1902 		dev_put(pdev);
1903 	} else {
1904 		pdev = get_real_dev(n->dev);
1905 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1906 					n, pdev, 0);
1907 		if (!ep->l2t)
1908 			goto out;
1909 		ep->mtu = dst_mtu(dst);
1910 		ep->tx_chan = cxgb4_port_chan(pdev);
1911 		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1912 		step = cdev->rdev.lldi.ntxq /
1913 			cdev->rdev.lldi.nchan;
1914 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1915 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1916 		step = cdev->rdev.lldi.nrxq /
1917 			cdev->rdev.lldi.nchan;
1918 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1919 			cxgb4_port_idx(pdev) * step];
1920 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
1921 
1922 		if (clear_mpa_v1) {
1923 			ep->retry_with_mpa_v1 = 0;
1924 			ep->tried_with_mpa_v1 = 0;
1925 		}
1926 	}
1927 	err = 0;
1928 out:
1929 	rcu_read_unlock();
1930 
1931 	neigh_release(n);
1932 
1933 	return err;
1934 }
1935 
1936 static int c4iw_reconnect(struct c4iw_ep *ep)
1937 {
1938 	int err = 0;
1939 	struct sockaddr_in *laddr = (struct sockaddr_in *)
1940 				    &ep->com.cm_id->local_addr;
1941 	struct sockaddr_in *raddr = (struct sockaddr_in *)
1942 				    &ep->com.cm_id->remote_addr;
1943 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1944 				      &ep->com.cm_id->local_addr;
1945 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1946 				      &ep->com.cm_id->remote_addr;
1947 	int iptype;
1948 	__u8 *ra;
1949 
1950 	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1951 	init_timer(&ep->timer);
1952 
1953 	/*
1954 	 * Allocate an active TID to initiate a TCP connection.
1955 	 */
1956 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1957 	if (ep->atid == -1) {
1958 		pr_err("%s - cannot alloc atid.\n", __func__);
1959 		err = -ENOMEM;
1960 		goto fail2;
1961 	}
1962 	insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1963 
1964 	/* find a route */
1965 	if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1966 		ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1967 				     raddr->sin_addr.s_addr, laddr->sin_port,
1968 				     raddr->sin_port, 0);
1969 		iptype = 4;
1970 		ra = (__u8 *)&raddr->sin_addr;
1971 	} else {
1972 		ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1973 				      raddr6->sin6_addr.s6_addr,
1974 				      laddr6->sin6_port, raddr6->sin6_port, 0,
1975 				      raddr6->sin6_scope_id);
1976 		iptype = 6;
1977 		ra = (__u8 *)&raddr6->sin6_addr;
1978 	}
1979 	if (!ep->dst) {
1980 		pr_err("%s - cannot find route.\n", __func__);
1981 		err = -EHOSTUNREACH;
1982 		goto fail3;
1983 	}
1984 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1985 	if (err) {
1986 		pr_err("%s - cannot alloc l2e.\n", __func__);
1987 		goto fail4;
1988 	}
1989 
1990 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1991 	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1992 	     ep->l2t->idx);
1993 
1994 	state_set(&ep->com, CONNECTING);
1995 	ep->tos = 0;
1996 
1997 	/* send connect request to rnic */
1998 	err = send_connect(ep);
1999 	if (!err)
2000 		goto out;
2001 
2002 	cxgb4_l2t_release(ep->l2t);
2003 fail4:
2004 	dst_release(ep->dst);
2005 fail3:
2006 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2007 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2008 fail2:
2009 	/*
2010 	 * remember to send notification to upper layer.
2011 	 * We are in here so the upper layer is not aware that this is
2012 	 * re-connect attempt and so, upper layer is still waiting for
2013 	 * response of 1st connect request.
2014 	 */
2015 	connect_reply_upcall(ep, -ECONNRESET);
2016 	c4iw_put_ep(&ep->com);
2017 out:
2018 	return err;
2019 }
2020 
2021 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2022 {
2023 	struct c4iw_ep *ep;
2024 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
2025 	unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
2026 					ntohl(rpl->atid_status)));
2027 	struct tid_info *t = dev->rdev.lldi.tids;
2028 	int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
2029 	struct sockaddr_in *la;
2030 	struct sockaddr_in *ra;
2031 	struct sockaddr_in6 *la6;
2032 	struct sockaddr_in6 *ra6;
2033 
2034 	ep = lookup_atid(t, atid);
2035 	la = (struct sockaddr_in *)&ep->com.mapped_local_addr;
2036 	ra = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
2037 	la6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
2038 	ra6 = (struct sockaddr_in6 *)&ep->com.mapped_remote_addr;
2039 
2040 	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2041 	     status, status2errno(status));
2042 
2043 	if (is_neg_adv(status)) {
2044 		dev_warn(&dev->rdev.lldi.pdev->dev,
2045 			 "Connection problems for atid %u status %u (%s)\n",
2046 			 atid, status, neg_adv_str(status));
2047 		return 0;
2048 	}
2049 
2050 	set_bit(ACT_OPEN_RPL, &ep->com.history);
2051 
2052 	/*
2053 	 * Log interesting failures.
2054 	 */
2055 	switch (status) {
2056 	case CPL_ERR_CONN_RESET:
2057 	case CPL_ERR_CONN_TIMEDOUT:
2058 		break;
2059 	case CPL_ERR_TCAM_FULL:
2060 		mutex_lock(&dev->rdev.stats.lock);
2061 		dev->rdev.stats.tcam_full++;
2062 		mutex_unlock(&dev->rdev.stats.lock);
2063 		if (ep->com.local_addr.ss_family == AF_INET &&
2064 		    dev->rdev.lldi.enable_fw_ofld_conn) {
2065 			send_fw_act_open_req(ep,
2066 					     GET_TID_TID(GET_AOPEN_ATID(
2067 					     ntohl(rpl->atid_status))));
2068 			return 0;
2069 		}
2070 		break;
2071 	case CPL_ERR_CONN_EXIST:
2072 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2073 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
2074 			remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2075 					atid);
2076 			cxgb4_free_atid(t, atid);
2077 			dst_release(ep->dst);
2078 			cxgb4_l2t_release(ep->l2t);
2079 			c4iw_reconnect(ep);
2080 			return 0;
2081 		}
2082 		break;
2083 	default:
2084 		if (ep->com.local_addr.ss_family == AF_INET) {
2085 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2086 				atid, status, status2errno(status),
2087 				&la->sin_addr.s_addr, ntohs(la->sin_port),
2088 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
2089 		} else {
2090 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2091 				atid, status, status2errno(status),
2092 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2093 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2094 		}
2095 		break;
2096 	}
2097 
2098 	connect_reply_upcall(ep, status2errno(status));
2099 	state_set(&ep->com, DEAD);
2100 
2101 	if (status && act_open_has_tid(status))
2102 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2103 
2104 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2105 	cxgb4_free_atid(t, atid);
2106 	dst_release(ep->dst);
2107 	cxgb4_l2t_release(ep->l2t);
2108 	c4iw_put_ep(&ep->com);
2109 
2110 	return 0;
2111 }
2112 
2113 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2114 {
2115 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2116 	struct tid_info *t = dev->rdev.lldi.tids;
2117 	unsigned int stid = GET_TID(rpl);
2118 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
2119 
2120 	if (!ep) {
2121 		PDBG("%s stid %d lookup failure!\n", __func__, stid);
2122 		goto out;
2123 	}
2124 	PDBG("%s ep %p status %d error %d\n", __func__, ep,
2125 	     rpl->status, status2errno(rpl->status));
2126 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2127 
2128 out:
2129 	return 0;
2130 }
2131 
2132 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2133 {
2134 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2135 	struct tid_info *t = dev->rdev.lldi.tids;
2136 	unsigned int stid = GET_TID(rpl);
2137 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
2138 
2139 	PDBG("%s ep %p\n", __func__, ep);
2140 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2141 	return 0;
2142 }
2143 
2144 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2145 		      struct cpl_pass_accept_req *req)
2146 {
2147 	struct cpl_pass_accept_rpl *rpl;
2148 	unsigned int mtu_idx;
2149 	u64 opt0;
2150 	u32 opt2;
2151 	int wscale;
2152 	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2153 	int win;
2154 
2155 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2156 	BUG_ON(skb_cloned(skb));
2157 
2158 	skb_get(skb);
2159 	rpl = cplhdr(skb);
2160 	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
2161 		skb_trim(skb, roundup(sizeof(*rpl5), 16));
2162 		rpl5 = (void *)rpl;
2163 		INIT_TP_WR(rpl5, ep->hwtid);
2164 	} else {
2165 		skb_trim(skb, sizeof(*rpl));
2166 		INIT_TP_WR(rpl, ep->hwtid);
2167 	}
2168 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2169 						    ep->hwtid));
2170 
2171 	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2172 		 enable_tcp_timestamps && req->tcpopt.tstamp,
2173 		 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2174 	wscale = compute_wscale(rcv_win);
2175 
2176 	/*
2177 	 * Specify the largest window that will fit in opt0. The
2178 	 * remainder will be specified in the rx_data_ack.
2179 	 */
2180 	win = ep->rcv_win >> 10;
2181 	if (win > RCV_BUFSIZ_MASK)
2182 		win = RCV_BUFSIZ_MASK;
2183 	opt0 = (nocong ? NO_CONG(1) : 0) |
2184 	       KEEP_ALIVE(1) |
2185 	       DELACK(1) |
2186 	       WND_SCALE(wscale) |
2187 	       MSS_IDX(mtu_idx) |
2188 	       L2T_IDX(ep->l2t->idx) |
2189 	       TX_CHAN(ep->tx_chan) |
2190 	       SMAC_SEL(ep->smac_idx) |
2191 	       DSCP(ep->tos >> 2) |
2192 	       ULP_MODE(ULP_MODE_TCPDDP) |
2193 	       RCV_BUFSIZ(win);
2194 	opt2 = RX_CHANNEL(0) |
2195 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
2196 
2197 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2198 		opt2 |= TSTAMPS_EN(1);
2199 	if (enable_tcp_sack && req->tcpopt.sack)
2200 		opt2 |= SACK_EN(1);
2201 	if (wscale && enable_tcp_window_scaling)
2202 		opt2 |= WND_SCALE_EN(1);
2203 	if (enable_ecn) {
2204 		const struct tcphdr *tcph;
2205 		u32 hlen = ntohl(req->hdr_len);
2206 
2207 		tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2208 			G_IP_HDR_LEN(hlen);
2209 		if (tcph->ece && tcph->cwr)
2210 			opt2 |= CCTRL_ECN(1);
2211 	}
2212 	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
2213 		u32 isn = (prandom_u32() & ~7UL) - 1;
2214 		opt2 |= T5_OPT_2_VALID;
2215 		opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
2216 		opt2 |= CONG_CNTRL_VALID; /* OPT_2_ISS for T5 */
2217 		rpl5 = (void *)rpl;
2218 		memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2219 		if (peer2peer)
2220 			isn += 4;
2221 		rpl5->iss = cpu_to_be32(isn);
2222 		PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2223 	}
2224 
2225 	rpl->opt0 = cpu_to_be64(opt0);
2226 	rpl->opt2 = cpu_to_be32(opt2);
2227 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2228 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
2229 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2230 
2231 	return;
2232 }
2233 
2234 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2235 {
2236 	PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2237 	BUG_ON(skb_cloned(skb));
2238 	skb_trim(skb, sizeof(struct cpl_tid_release));
2239 	release_tid(&dev->rdev, hwtid, skb);
2240 	return;
2241 }
2242 
2243 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2244 		       __u8 *local_ip, __u8 *peer_ip,
2245 		       __be16 *local_port, __be16 *peer_port)
2246 {
2247 	int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2248 	int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2249 	struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2250 	struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2251 	struct tcphdr *tcp = (struct tcphdr *)
2252 			     ((u8 *)(req + 1) + eth_len + ip_len);
2253 
2254 	if (ip->version == 4) {
2255 		PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2256 		     ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2257 		     ntohs(tcp->dest));
2258 		*iptype = 4;
2259 		memcpy(peer_ip, &ip->saddr, 4);
2260 		memcpy(local_ip, &ip->daddr, 4);
2261 	} else {
2262 		PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2263 		     ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2264 		     ntohs(tcp->dest));
2265 		*iptype = 6;
2266 		memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2267 		memcpy(local_ip, ip6->daddr.s6_addr, 16);
2268 	}
2269 	*peer_port = tcp->source;
2270 	*local_port = tcp->dest;
2271 
2272 	return;
2273 }
2274 
2275 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2276 {
2277 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2278 	struct cpl_pass_accept_req *req = cplhdr(skb);
2279 	unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2280 	struct tid_info *t = dev->rdev.lldi.tids;
2281 	unsigned int hwtid = GET_TID(req);
2282 	struct dst_entry *dst;
2283 	__u8 local_ip[16], peer_ip[16];
2284 	__be16 local_port, peer_port;
2285 	int err;
2286 	u16 peer_mss = ntohs(req->tcpopt.mss);
2287 	int iptype;
2288 	unsigned short hdrs;
2289 
2290 	parent_ep = lookup_stid(t, stid);
2291 	if (!parent_ep) {
2292 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2293 		goto reject;
2294 	}
2295 
2296 	if (state_read(&parent_ep->com) != LISTEN) {
2297 		printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2298 		       __func__);
2299 		goto reject;
2300 	}
2301 
2302 	get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2303 
2304 	/* Find output route */
2305 	if (iptype == 4)  {
2306 		PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2307 		     , __func__, parent_ep, hwtid,
2308 		     local_ip, peer_ip, ntohs(local_port),
2309 		     ntohs(peer_port), peer_mss);
2310 		dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2311 				 local_port, peer_port,
2312 				 GET_POPEN_TOS(ntohl(req->tos_stid)));
2313 	} else {
2314 		PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2315 		     , __func__, parent_ep, hwtid,
2316 		     local_ip, peer_ip, ntohs(local_port),
2317 		     ntohs(peer_port), peer_mss);
2318 		dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2319 				  PASS_OPEN_TOS(ntohl(req->tos_stid)),
2320 				  ((struct sockaddr_in6 *)
2321 				  &parent_ep->com.local_addr)->sin6_scope_id);
2322 	}
2323 	if (!dst) {
2324 		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2325 		       __func__);
2326 		goto reject;
2327 	}
2328 
2329 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2330 	if (!child_ep) {
2331 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2332 		       __func__);
2333 		dst_release(dst);
2334 		goto reject;
2335 	}
2336 
2337 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2338 	if (err) {
2339 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2340 		       __func__);
2341 		dst_release(dst);
2342 		kfree(child_ep);
2343 		goto reject;
2344 	}
2345 
2346 	hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2347 	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2348 	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2349 		child_ep->mtu = peer_mss + hdrs;
2350 
2351 	state_set(&child_ep->com, CONNECTING);
2352 	child_ep->com.dev = dev;
2353 	child_ep->com.cm_id = NULL;
2354 	if (iptype == 4) {
2355 		struct sockaddr_in *sin = (struct sockaddr_in *)
2356 			&child_ep->com.local_addr;
2357 		sin->sin_family = PF_INET;
2358 		sin->sin_port = local_port;
2359 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2360 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2361 		sin->sin_family = PF_INET;
2362 		sin->sin_port = peer_port;
2363 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2364 	} else {
2365 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2366 			&child_ep->com.local_addr;
2367 		sin6->sin6_family = PF_INET6;
2368 		sin6->sin6_port = local_port;
2369 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2370 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2371 		sin6->sin6_family = PF_INET6;
2372 		sin6->sin6_port = peer_port;
2373 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2374 	}
2375 	c4iw_get_ep(&parent_ep->com);
2376 	child_ep->parent_ep = parent_ep;
2377 	child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2378 	child_ep->dst = dst;
2379 	child_ep->hwtid = hwtid;
2380 
2381 	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2382 	     child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2383 
2384 	init_timer(&child_ep->timer);
2385 	cxgb4_insert_tid(t, child_ep, hwtid);
2386 	insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2387 	accept_cr(child_ep, skb, req);
2388 	set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2389 	goto out;
2390 reject:
2391 	reject_cr(dev, hwtid, skb);
2392 out:
2393 	return 0;
2394 }
2395 
2396 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2397 {
2398 	struct c4iw_ep *ep;
2399 	struct cpl_pass_establish *req = cplhdr(skb);
2400 	struct tid_info *t = dev->rdev.lldi.tids;
2401 	unsigned int tid = GET_TID(req);
2402 
2403 	ep = lookup_tid(t, tid);
2404 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2405 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2406 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2407 
2408 	PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2409 	     ntohs(req->tcp_opt));
2410 
2411 	set_emss(ep, ntohs(req->tcp_opt));
2412 
2413 	dst_confirm(ep->dst);
2414 	state_set(&ep->com, MPA_REQ_WAIT);
2415 	start_ep_timer(ep);
2416 	send_flowc(ep, skb);
2417 	set_bit(PASS_ESTAB, &ep->com.history);
2418 
2419 	return 0;
2420 }
2421 
2422 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2423 {
2424 	struct cpl_peer_close *hdr = cplhdr(skb);
2425 	struct c4iw_ep *ep;
2426 	struct c4iw_qp_attributes attrs;
2427 	int disconnect = 1;
2428 	int release = 0;
2429 	struct tid_info *t = dev->rdev.lldi.tids;
2430 	unsigned int tid = GET_TID(hdr);
2431 	int ret;
2432 
2433 	ep = lookup_tid(t, tid);
2434 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2435 	dst_confirm(ep->dst);
2436 
2437 	set_bit(PEER_CLOSE, &ep->com.history);
2438 	mutex_lock(&ep->com.mutex);
2439 	switch (ep->com.state) {
2440 	case MPA_REQ_WAIT:
2441 		__state_set(&ep->com, CLOSING);
2442 		break;
2443 	case MPA_REQ_SENT:
2444 		__state_set(&ep->com, CLOSING);
2445 		connect_reply_upcall(ep, -ECONNRESET);
2446 		break;
2447 	case MPA_REQ_RCVD:
2448 
2449 		/*
2450 		 * We're gonna mark this puppy DEAD, but keep
2451 		 * the reference on it until the ULP accepts or
2452 		 * rejects the CR. Also wake up anyone waiting
2453 		 * in rdma connection migration (see c4iw_accept_cr()).
2454 		 */
2455 		__state_set(&ep->com, CLOSING);
2456 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2457 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2458 		break;
2459 	case MPA_REP_SENT:
2460 		__state_set(&ep->com, CLOSING);
2461 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2462 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2463 		break;
2464 	case FPDU_MODE:
2465 		start_ep_timer(ep);
2466 		__state_set(&ep->com, CLOSING);
2467 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2468 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2469 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2470 		if (ret != -ECONNRESET) {
2471 			peer_close_upcall(ep);
2472 			disconnect = 1;
2473 		}
2474 		break;
2475 	case ABORTING:
2476 		disconnect = 0;
2477 		break;
2478 	case CLOSING:
2479 		__state_set(&ep->com, MORIBUND);
2480 		disconnect = 0;
2481 		break;
2482 	case MORIBUND:
2483 		(void)stop_ep_timer(ep);
2484 		if (ep->com.cm_id && ep->com.qp) {
2485 			attrs.next_state = C4IW_QP_STATE_IDLE;
2486 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2487 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2488 		}
2489 		close_complete_upcall(ep, 0);
2490 		__state_set(&ep->com, DEAD);
2491 		release = 1;
2492 		disconnect = 0;
2493 		break;
2494 	case DEAD:
2495 		disconnect = 0;
2496 		break;
2497 	default:
2498 		BUG_ON(1);
2499 	}
2500 	mutex_unlock(&ep->com.mutex);
2501 	if (disconnect)
2502 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2503 	if (release)
2504 		release_ep_resources(ep);
2505 	return 0;
2506 }
2507 
2508 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2509 {
2510 	struct cpl_abort_req_rss *req = cplhdr(skb);
2511 	struct c4iw_ep *ep;
2512 	struct cpl_abort_rpl *rpl;
2513 	struct sk_buff *rpl_skb;
2514 	struct c4iw_qp_attributes attrs;
2515 	int ret;
2516 	int release = 0;
2517 	struct tid_info *t = dev->rdev.lldi.tids;
2518 	unsigned int tid = GET_TID(req);
2519 
2520 	ep = lookup_tid(t, tid);
2521 	if (is_neg_adv(req->status)) {
2522 		dev_warn(&dev->rdev.lldi.pdev->dev,
2523 			 "Negative advice on abort - tid %u status %d (%s)\n",
2524 			 ep->hwtid, req->status, neg_adv_str(req->status));
2525 		return 0;
2526 	}
2527 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2528 	     ep->com.state);
2529 	set_bit(PEER_ABORT, &ep->com.history);
2530 
2531 	/*
2532 	 * Wake up any threads in rdma_init() or rdma_fini().
2533 	 * However, this is not needed if com state is just
2534 	 * MPA_REQ_SENT
2535 	 */
2536 	if (ep->com.state != MPA_REQ_SENT)
2537 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2538 
2539 	mutex_lock(&ep->com.mutex);
2540 	switch (ep->com.state) {
2541 	case CONNECTING:
2542 		break;
2543 	case MPA_REQ_WAIT:
2544 		(void)stop_ep_timer(ep);
2545 		break;
2546 	case MPA_REQ_SENT:
2547 		(void)stop_ep_timer(ep);
2548 		if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2549 			connect_reply_upcall(ep, -ECONNRESET);
2550 		else {
2551 			/*
2552 			 * we just don't send notification upwards because we
2553 			 * want to retry with mpa_v1 without upper layers even
2554 			 * knowing it.
2555 			 *
2556 			 * do some housekeeping so as to re-initiate the
2557 			 * connection
2558 			 */
2559 			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2560 			     mpa_rev);
2561 			ep->retry_with_mpa_v1 = 1;
2562 		}
2563 		break;
2564 	case MPA_REP_SENT:
2565 		break;
2566 	case MPA_REQ_RCVD:
2567 		break;
2568 	case MORIBUND:
2569 	case CLOSING:
2570 		stop_ep_timer(ep);
2571 		/*FALLTHROUGH*/
2572 	case FPDU_MODE:
2573 		if (ep->com.cm_id && ep->com.qp) {
2574 			attrs.next_state = C4IW_QP_STATE_ERROR;
2575 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2576 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2577 				     &attrs, 1);
2578 			if (ret)
2579 				printk(KERN_ERR MOD
2580 				       "%s - qp <- error failed!\n",
2581 				       __func__);
2582 		}
2583 		peer_abort_upcall(ep);
2584 		break;
2585 	case ABORTING:
2586 		break;
2587 	case DEAD:
2588 		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2589 		mutex_unlock(&ep->com.mutex);
2590 		return 0;
2591 	default:
2592 		BUG_ON(1);
2593 		break;
2594 	}
2595 	dst_confirm(ep->dst);
2596 	if (ep->com.state != ABORTING) {
2597 		__state_set(&ep->com, DEAD);
2598 		/* we don't release if we want to retry with mpa_v1 */
2599 		if (!ep->retry_with_mpa_v1)
2600 			release = 1;
2601 	}
2602 	mutex_unlock(&ep->com.mutex);
2603 
2604 	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2605 	if (!rpl_skb) {
2606 		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2607 		       __func__);
2608 		release = 1;
2609 		goto out;
2610 	}
2611 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2612 	rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2613 	INIT_TP_WR(rpl, ep->hwtid);
2614 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2615 	rpl->cmd = CPL_ABORT_NO_RST;
2616 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2617 out:
2618 	if (release)
2619 		release_ep_resources(ep);
2620 	else if (ep->retry_with_mpa_v1) {
2621 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2622 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2623 		dst_release(ep->dst);
2624 		cxgb4_l2t_release(ep->l2t);
2625 		c4iw_reconnect(ep);
2626 	}
2627 
2628 	return 0;
2629 }
2630 
2631 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2632 {
2633 	struct c4iw_ep *ep;
2634 	struct c4iw_qp_attributes attrs;
2635 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2636 	int release = 0;
2637 	struct tid_info *t = dev->rdev.lldi.tids;
2638 	unsigned int tid = GET_TID(rpl);
2639 
2640 	ep = lookup_tid(t, tid);
2641 
2642 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2643 	BUG_ON(!ep);
2644 
2645 	/* The cm_id may be null if we failed to connect */
2646 	mutex_lock(&ep->com.mutex);
2647 	switch (ep->com.state) {
2648 	case CLOSING:
2649 		__state_set(&ep->com, MORIBUND);
2650 		break;
2651 	case MORIBUND:
2652 		(void)stop_ep_timer(ep);
2653 		if ((ep->com.cm_id) && (ep->com.qp)) {
2654 			attrs.next_state = C4IW_QP_STATE_IDLE;
2655 			c4iw_modify_qp(ep->com.qp->rhp,
2656 					     ep->com.qp,
2657 					     C4IW_QP_ATTR_NEXT_STATE,
2658 					     &attrs, 1);
2659 		}
2660 		close_complete_upcall(ep, 0);
2661 		__state_set(&ep->com, DEAD);
2662 		release = 1;
2663 		break;
2664 	case ABORTING:
2665 	case DEAD:
2666 		break;
2667 	default:
2668 		BUG_ON(1);
2669 		break;
2670 	}
2671 	mutex_unlock(&ep->com.mutex);
2672 	if (release)
2673 		release_ep_resources(ep);
2674 	return 0;
2675 }
2676 
2677 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2678 {
2679 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
2680 	struct tid_info *t = dev->rdev.lldi.tids;
2681 	unsigned int tid = GET_TID(rpl);
2682 	struct c4iw_ep *ep;
2683 	struct c4iw_qp_attributes attrs;
2684 
2685 	ep = lookup_tid(t, tid);
2686 	BUG_ON(!ep);
2687 
2688 	if (ep && ep->com.qp) {
2689 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2690 		       ep->com.qp->wq.sq.qid);
2691 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2692 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2693 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2694 	} else
2695 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2696 
2697 	return 0;
2698 }
2699 
2700 /*
2701  * Upcall from the adapter indicating data has been transmitted.
2702  * For us its just the single MPA request or reply.  We can now free
2703  * the skb holding the mpa message.
2704  */
2705 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2706 {
2707 	struct c4iw_ep *ep;
2708 	struct cpl_fw4_ack *hdr = cplhdr(skb);
2709 	u8 credits = hdr->credits;
2710 	unsigned int tid = GET_TID(hdr);
2711 	struct tid_info *t = dev->rdev.lldi.tids;
2712 
2713 
2714 	ep = lookup_tid(t, tid);
2715 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2716 	if (credits == 0) {
2717 		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2718 		     __func__, ep, ep->hwtid, state_read(&ep->com));
2719 		return 0;
2720 	}
2721 
2722 	dst_confirm(ep->dst);
2723 	if (ep->mpa_skb) {
2724 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
2725 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2726 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2727 		kfree_skb(ep->mpa_skb);
2728 		ep->mpa_skb = NULL;
2729 	}
2730 	return 0;
2731 }
2732 
2733 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2734 {
2735 	int err = 0;
2736 	int disconnect = 0;
2737 	struct c4iw_ep *ep = to_ep(cm_id);
2738 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2739 
2740 	mutex_lock(&ep->com.mutex);
2741 	if (ep->com.state == DEAD) {
2742 		mutex_unlock(&ep->com.mutex);
2743 		c4iw_put_ep(&ep->com);
2744 		return -ECONNRESET;
2745 	}
2746 	set_bit(ULP_REJECT, &ep->com.history);
2747 	BUG_ON(ep->com.state != MPA_REQ_RCVD);
2748 	if (mpa_rev == 0)
2749 		abort_connection(ep, NULL, GFP_KERNEL);
2750 	else {
2751 		err = send_mpa_reject(ep, pdata, pdata_len);
2752 		disconnect = 1;
2753 	}
2754 	mutex_unlock(&ep->com.mutex);
2755 	if (disconnect)
2756 		err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2757 	c4iw_put_ep(&ep->com);
2758 	return 0;
2759 }
2760 
2761 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2762 {
2763 	int err;
2764 	struct c4iw_qp_attributes attrs;
2765 	enum c4iw_qp_attr_mask mask;
2766 	struct c4iw_ep *ep = to_ep(cm_id);
2767 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2768 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2769 
2770 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2771 
2772 	mutex_lock(&ep->com.mutex);
2773 	if (ep->com.state == DEAD) {
2774 		err = -ECONNRESET;
2775 		goto err;
2776 	}
2777 
2778 	BUG_ON(ep->com.state != MPA_REQ_RCVD);
2779 	BUG_ON(!qp);
2780 
2781 	set_bit(ULP_ACCEPT, &ep->com.history);
2782 	if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
2783 	    (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
2784 		abort_connection(ep, NULL, GFP_KERNEL);
2785 		err = -EINVAL;
2786 		goto err;
2787 	}
2788 
2789 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2790 		if (conn_param->ord > ep->ird) {
2791 			if (RELAXED_IRD_NEGOTIATION) {
2792 				ep->ord = ep->ird;
2793 			} else {
2794 				ep->ird = conn_param->ird;
2795 				ep->ord = conn_param->ord;
2796 				send_mpa_reject(ep, conn_param->private_data,
2797 						conn_param->private_data_len);
2798 				abort_connection(ep, NULL, GFP_KERNEL);
2799 				err = -ENOMEM;
2800 				goto err;
2801 			}
2802 		}
2803 		if (conn_param->ird < ep->ord) {
2804 			if (RELAXED_IRD_NEGOTIATION &&
2805 			    ep->ord <= h->rdev.lldi.max_ordird_qp) {
2806 				conn_param->ird = ep->ord;
2807 			} else {
2808 				abort_connection(ep, NULL, GFP_KERNEL);
2809 				err = -ENOMEM;
2810 				goto err;
2811 			}
2812 		}
2813 	}
2814 	ep->ird = conn_param->ird;
2815 	ep->ord = conn_param->ord;
2816 
2817 	if (ep->mpa_attr.version == 1) {
2818 		if (peer2peer && ep->ird == 0)
2819 			ep->ird = 1;
2820 	} else {
2821 		if (peer2peer &&
2822 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
2823 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ord == 0)
2824 			ep->ird = 1;
2825 	}
2826 
2827 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2828 
2829 	cm_id->add_ref(cm_id);
2830 	ep->com.cm_id = cm_id;
2831 	ep->com.qp = qp;
2832 	ref_qp(ep);
2833 
2834 	/* bind QP to EP and move to RTS */
2835 	attrs.mpa_attr = ep->mpa_attr;
2836 	attrs.max_ird = ep->ird;
2837 	attrs.max_ord = ep->ord;
2838 	attrs.llp_stream_handle = ep;
2839 	attrs.next_state = C4IW_QP_STATE_RTS;
2840 
2841 	/* bind QP and TID with INIT_WR */
2842 	mask = C4IW_QP_ATTR_NEXT_STATE |
2843 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2844 			     C4IW_QP_ATTR_MPA_ATTR |
2845 			     C4IW_QP_ATTR_MAX_IRD |
2846 			     C4IW_QP_ATTR_MAX_ORD;
2847 
2848 	err = c4iw_modify_qp(ep->com.qp->rhp,
2849 			     ep->com.qp, mask, &attrs, 1);
2850 	if (err)
2851 		goto err1;
2852 	err = send_mpa_reply(ep, conn_param->private_data,
2853 			     conn_param->private_data_len);
2854 	if (err)
2855 		goto err1;
2856 
2857 	__state_set(&ep->com, FPDU_MODE);
2858 	established_upcall(ep);
2859 	mutex_unlock(&ep->com.mutex);
2860 	c4iw_put_ep(&ep->com);
2861 	return 0;
2862 err1:
2863 	ep->com.cm_id = NULL;
2864 	abort_connection(ep, NULL, GFP_KERNEL);
2865 	cm_id->rem_ref(cm_id);
2866 err:
2867 	mutex_unlock(&ep->com.mutex);
2868 	c4iw_put_ep(&ep->com);
2869 	return err;
2870 }
2871 
2872 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2873 {
2874 	struct in_device *ind;
2875 	int found = 0;
2876 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2877 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2878 
2879 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
2880 	if (!ind)
2881 		return -EADDRNOTAVAIL;
2882 	for_primary_ifa(ind) {
2883 		laddr->sin_addr.s_addr = ifa->ifa_address;
2884 		raddr->sin_addr.s_addr = ifa->ifa_address;
2885 		found = 1;
2886 		break;
2887 	}
2888 	endfor_ifa(ind);
2889 	in_dev_put(ind);
2890 	return found ? 0 : -EADDRNOTAVAIL;
2891 }
2892 
2893 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2894 		      unsigned char banned_flags)
2895 {
2896 	struct inet6_dev *idev;
2897 	int err = -EADDRNOTAVAIL;
2898 
2899 	rcu_read_lock();
2900 	idev = __in6_dev_get(dev);
2901 	if (idev != NULL) {
2902 		struct inet6_ifaddr *ifp;
2903 
2904 		read_lock_bh(&idev->lock);
2905 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
2906 			if (ifp->scope == IFA_LINK &&
2907 			    !(ifp->flags & banned_flags)) {
2908 				memcpy(addr, &ifp->addr, 16);
2909 				err = 0;
2910 				break;
2911 			}
2912 		}
2913 		read_unlock_bh(&idev->lock);
2914 	}
2915 	rcu_read_unlock();
2916 	return err;
2917 }
2918 
2919 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2920 {
2921 	struct in6_addr uninitialized_var(addr);
2922 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2923 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2924 
2925 	if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2926 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2927 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2928 		return 0;
2929 	}
2930 	return -EADDRNOTAVAIL;
2931 }
2932 
2933 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2934 {
2935 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2936 	struct c4iw_ep *ep;
2937 	int err = 0;
2938 	struct sockaddr_in *laddr;
2939 	struct sockaddr_in *raddr;
2940 	struct sockaddr_in6 *laddr6;
2941 	struct sockaddr_in6 *raddr6;
2942 	struct iwpm_dev_data pm_reg_msg;
2943 	struct iwpm_sa_data pm_msg;
2944 	__u8 *ra;
2945 	int iptype;
2946 	int iwpm_err = 0;
2947 
2948 	if ((conn_param->ord > cur_max_read_depth(dev)) ||
2949 	    (conn_param->ird > cur_max_read_depth(dev))) {
2950 		err = -EINVAL;
2951 		goto out;
2952 	}
2953 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2954 	if (!ep) {
2955 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2956 		err = -ENOMEM;
2957 		goto out;
2958 	}
2959 	init_timer(&ep->timer);
2960 	ep->plen = conn_param->private_data_len;
2961 	if (ep->plen)
2962 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2963 		       conn_param->private_data, ep->plen);
2964 	ep->ird = conn_param->ird;
2965 	ep->ord = conn_param->ord;
2966 
2967 	if (peer2peer && ep->ord == 0)
2968 		ep->ord = 1;
2969 
2970 	cm_id->add_ref(cm_id);
2971 	ep->com.dev = dev;
2972 	ep->com.cm_id = cm_id;
2973 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2974 	if (!ep->com.qp) {
2975 		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2976 		err = -EINVAL;
2977 		goto fail1;
2978 	}
2979 	ref_qp(ep);
2980 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2981 	     ep->com.qp, cm_id);
2982 
2983 	/*
2984 	 * Allocate an active TID to initiate a TCP connection.
2985 	 */
2986 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2987 	if (ep->atid == -1) {
2988 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2989 		err = -ENOMEM;
2990 		goto fail1;
2991 	}
2992 	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2993 
2994 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
2995 	       sizeof(ep->com.local_addr));
2996 	memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2997 	       sizeof(ep->com.remote_addr));
2998 
2999 	/* No port mapper available, go with the specified peer information */
3000 	memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
3001 	       sizeof(ep->com.mapped_local_addr));
3002 	memcpy(&ep->com.mapped_remote_addr, &cm_id->remote_addr,
3003 	       sizeof(ep->com.mapped_remote_addr));
3004 
3005 	c4iw_form_reg_msg(dev, &pm_reg_msg);
3006 	iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
3007 	if (iwpm_err) {
3008 		PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
3009 			__func__, iwpm_err);
3010 	}
3011 	if (iwpm_valid_pid() && !iwpm_err) {
3012 		c4iw_form_pm_msg(ep, &pm_msg);
3013 		iwpm_err = iwpm_add_and_query_mapping(&pm_msg, RDMA_NL_C4IW);
3014 		if (iwpm_err)
3015 			PDBG("%s: Port Mapper query fail (err = %d).\n",
3016 				__func__, iwpm_err);
3017 		else
3018 			c4iw_record_pm_msg(ep, &pm_msg);
3019 	}
3020 	if (iwpm_create_mapinfo(&ep->com.local_addr,
3021 				&ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
3022 		iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
3023 		err = -ENOMEM;
3024 		goto fail1;
3025 	}
3026 	print_addr(&ep->com, __func__, "add_query/create_mapinfo");
3027 	set_bit(RELEASE_MAPINFO, &ep->com.flags);
3028 
3029 	laddr = (struct sockaddr_in *)&ep->com.mapped_local_addr;
3030 	raddr = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
3031 	laddr6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
3032 	raddr6 = (struct sockaddr_in6 *) &ep->com.mapped_remote_addr;
3033 
3034 	if (cm_id->remote_addr.ss_family == AF_INET) {
3035 		iptype = 4;
3036 		ra = (__u8 *)&raddr->sin_addr;
3037 
3038 		/*
3039 		 * Handle loopback requests to INADDR_ANY.
3040 		 */
3041 		if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
3042 			err = pick_local_ipaddrs(dev, cm_id);
3043 			if (err)
3044 				goto fail1;
3045 		}
3046 
3047 		/* find a route */
3048 		PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3049 		     __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3050 		     ra, ntohs(raddr->sin_port));
3051 		ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3052 				     raddr->sin_addr.s_addr, laddr->sin_port,
3053 				     raddr->sin_port, 0);
3054 	} else {
3055 		iptype = 6;
3056 		ra = (__u8 *)&raddr6->sin6_addr;
3057 
3058 		/*
3059 		 * Handle loopback requests to INADDR_ANY.
3060 		 */
3061 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3062 			err = pick_local_ip6addrs(dev, cm_id);
3063 			if (err)
3064 				goto fail1;
3065 		}
3066 
3067 		/* find a route */
3068 		PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3069 		     __func__, laddr6->sin6_addr.s6_addr,
3070 		     ntohs(laddr6->sin6_port),
3071 		     raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3072 		ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3073 				      raddr6->sin6_addr.s6_addr,
3074 				      laddr6->sin6_port, raddr6->sin6_port, 0,
3075 				      raddr6->sin6_scope_id);
3076 	}
3077 	if (!ep->dst) {
3078 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3079 		err = -EHOSTUNREACH;
3080 		goto fail2;
3081 	}
3082 
3083 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
3084 	if (err) {
3085 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3086 		goto fail3;
3087 	}
3088 
3089 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3090 		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3091 		ep->l2t->idx);
3092 
3093 	state_set(&ep->com, CONNECTING);
3094 	ep->tos = 0;
3095 
3096 	/* send connect request to rnic */
3097 	err = send_connect(ep);
3098 	if (!err)
3099 		goto out;
3100 
3101 	cxgb4_l2t_release(ep->l2t);
3102 fail3:
3103 	dst_release(ep->dst);
3104 fail2:
3105 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3106 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3107 fail1:
3108 	cm_id->rem_ref(cm_id);
3109 	c4iw_put_ep(&ep->com);
3110 out:
3111 	return err;
3112 }
3113 
3114 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3115 {
3116 	int err;
3117 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3118 				    &ep->com.mapped_local_addr;
3119 
3120 	c4iw_init_wr_wait(&ep->com.wr_wait);
3121 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3122 				   ep->stid, &sin6->sin6_addr,
3123 				   sin6->sin6_port,
3124 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
3125 	if (!err)
3126 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3127 					  &ep->com.wr_wait,
3128 					  0, 0, __func__);
3129 	if (err)
3130 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3131 		       err, ep->stid,
3132 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3133 	return err;
3134 }
3135 
3136 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3137 {
3138 	int err;
3139 	struct sockaddr_in *sin = (struct sockaddr_in *)
3140 				  &ep->com.mapped_local_addr;
3141 
3142 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
3143 		do {
3144 			err = cxgb4_create_server_filter(
3145 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3146 				sin->sin_addr.s_addr, sin->sin_port, 0,
3147 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3148 			if (err == -EBUSY) {
3149 				set_current_state(TASK_UNINTERRUPTIBLE);
3150 				schedule_timeout(usecs_to_jiffies(100));
3151 			}
3152 		} while (err == -EBUSY);
3153 	} else {
3154 		c4iw_init_wr_wait(&ep->com.wr_wait);
3155 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3156 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3157 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3158 		if (!err)
3159 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3160 						  &ep->com.wr_wait,
3161 						  0, 0, __func__);
3162 	}
3163 	if (err)
3164 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3165 		       , err, ep->stid,
3166 		       &sin->sin_addr, ntohs(sin->sin_port));
3167 	return err;
3168 }
3169 
3170 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3171 {
3172 	int err = 0;
3173 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3174 	struct c4iw_listen_ep *ep;
3175 	struct iwpm_dev_data pm_reg_msg;
3176 	struct iwpm_sa_data pm_msg;
3177 	int iwpm_err = 0;
3178 
3179 	might_sleep();
3180 
3181 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3182 	if (!ep) {
3183 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3184 		err = -ENOMEM;
3185 		goto fail1;
3186 	}
3187 	PDBG("%s ep %p\n", __func__, ep);
3188 	cm_id->add_ref(cm_id);
3189 	ep->com.cm_id = cm_id;
3190 	ep->com.dev = dev;
3191 	ep->backlog = backlog;
3192 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
3193 	       sizeof(ep->com.local_addr));
3194 
3195 	/*
3196 	 * Allocate a server TID.
3197 	 */
3198 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
3199 	    ep->com.local_addr.ss_family == AF_INET)
3200 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3201 					     cm_id->local_addr.ss_family, ep);
3202 	else
3203 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3204 					    cm_id->local_addr.ss_family, ep);
3205 
3206 	if (ep->stid == -1) {
3207 		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3208 		err = -ENOMEM;
3209 		goto fail2;
3210 	}
3211 	insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3212 
3213 	/* No port mapper available, go with the specified info */
3214 	memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
3215 	       sizeof(ep->com.mapped_local_addr));
3216 
3217 	c4iw_form_reg_msg(dev, &pm_reg_msg);
3218 	iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
3219 	if (iwpm_err) {
3220 		PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
3221 			__func__, iwpm_err);
3222 	}
3223 	if (iwpm_valid_pid() && !iwpm_err) {
3224 		memcpy(&pm_msg.loc_addr, &ep->com.local_addr,
3225 				sizeof(ep->com.local_addr));
3226 		iwpm_err = iwpm_add_mapping(&pm_msg, RDMA_NL_C4IW);
3227 		if (iwpm_err)
3228 			PDBG("%s: Port Mapper query fail (err = %d).\n",
3229 				__func__, iwpm_err);
3230 		else
3231 			memcpy(&ep->com.mapped_local_addr,
3232 				&pm_msg.mapped_loc_addr,
3233 				sizeof(ep->com.mapped_local_addr));
3234 	}
3235 	if (iwpm_create_mapinfo(&ep->com.local_addr,
3236 				&ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
3237 		err = -ENOMEM;
3238 		goto fail3;
3239 	}
3240 	print_addr(&ep->com, __func__, "add_mapping/create_mapinfo");
3241 
3242 	set_bit(RELEASE_MAPINFO, &ep->com.flags);
3243 	state_set(&ep->com, LISTEN);
3244 	if (ep->com.local_addr.ss_family == AF_INET)
3245 		err = create_server4(dev, ep);
3246 	else
3247 		err = create_server6(dev, ep);
3248 	if (!err) {
3249 		cm_id->provider_data = ep;
3250 		goto out;
3251 	}
3252 
3253 fail3:
3254 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3255 			ep->com.local_addr.ss_family);
3256 fail2:
3257 	cm_id->rem_ref(cm_id);
3258 	c4iw_put_ep(&ep->com);
3259 fail1:
3260 out:
3261 	return err;
3262 }
3263 
3264 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3265 {
3266 	int err;
3267 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3268 
3269 	PDBG("%s ep %p\n", __func__, ep);
3270 
3271 	might_sleep();
3272 	state_set(&ep->com, DEAD);
3273 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3274 	    ep->com.local_addr.ss_family == AF_INET) {
3275 		err = cxgb4_remove_server_filter(
3276 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
3277 			ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3278 	} else {
3279 		c4iw_init_wr_wait(&ep->com.wr_wait);
3280 		err = cxgb4_remove_server(
3281 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3282 				ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3283 		if (err)
3284 			goto done;
3285 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3286 					  0, 0, __func__);
3287 	}
3288 	remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3289 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3290 			ep->com.local_addr.ss_family);
3291 done:
3292 	cm_id->rem_ref(cm_id);
3293 	c4iw_put_ep(&ep->com);
3294 	return err;
3295 }
3296 
3297 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3298 {
3299 	int ret = 0;
3300 	int close = 0;
3301 	int fatal = 0;
3302 	struct c4iw_rdev *rdev;
3303 
3304 	mutex_lock(&ep->com.mutex);
3305 
3306 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3307 	     states[ep->com.state], abrupt);
3308 
3309 	rdev = &ep->com.dev->rdev;
3310 	if (c4iw_fatal_error(rdev)) {
3311 		fatal = 1;
3312 		close_complete_upcall(ep, -EIO);
3313 		ep->com.state = DEAD;
3314 	}
3315 	switch (ep->com.state) {
3316 	case MPA_REQ_WAIT:
3317 	case MPA_REQ_SENT:
3318 	case MPA_REQ_RCVD:
3319 	case MPA_REP_SENT:
3320 	case FPDU_MODE:
3321 		close = 1;
3322 		if (abrupt)
3323 			ep->com.state = ABORTING;
3324 		else {
3325 			ep->com.state = CLOSING;
3326 			start_ep_timer(ep);
3327 		}
3328 		set_bit(CLOSE_SENT, &ep->com.flags);
3329 		break;
3330 	case CLOSING:
3331 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3332 			close = 1;
3333 			if (abrupt) {
3334 				(void)stop_ep_timer(ep);
3335 				ep->com.state = ABORTING;
3336 			} else
3337 				ep->com.state = MORIBUND;
3338 		}
3339 		break;
3340 	case MORIBUND:
3341 	case ABORTING:
3342 	case DEAD:
3343 		PDBG("%s ignoring disconnect ep %p state %u\n",
3344 		     __func__, ep, ep->com.state);
3345 		break;
3346 	default:
3347 		BUG();
3348 		break;
3349 	}
3350 
3351 	if (close) {
3352 		if (abrupt) {
3353 			set_bit(EP_DISC_ABORT, &ep->com.history);
3354 			close_complete_upcall(ep, -ECONNRESET);
3355 			ret = send_abort(ep, NULL, gfp);
3356 		} else {
3357 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3358 			ret = send_halfclose(ep, gfp);
3359 		}
3360 		if (ret)
3361 			fatal = 1;
3362 	}
3363 	mutex_unlock(&ep->com.mutex);
3364 	if (fatal)
3365 		release_ep_resources(ep);
3366 	return ret;
3367 }
3368 
3369 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3370 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3371 {
3372 	struct c4iw_ep *ep;
3373 	int atid = be32_to_cpu(req->tid);
3374 
3375 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3376 					   (__force u32) req->tid);
3377 	if (!ep)
3378 		return;
3379 
3380 	switch (req->retval) {
3381 	case FW_ENOMEM:
3382 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3383 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3384 			send_fw_act_open_req(ep, atid);
3385 			return;
3386 		}
3387 	case FW_EADDRINUSE:
3388 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3389 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3390 			send_fw_act_open_req(ep, atid);
3391 			return;
3392 		}
3393 		break;
3394 	default:
3395 		pr_info("%s unexpected ofld conn wr retval %d\n",
3396 		       __func__, req->retval);
3397 		break;
3398 	}
3399 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3400 	       req->retval, atid);
3401 	mutex_lock(&dev->rdev.stats.lock);
3402 	dev->rdev.stats.act_ofld_conn_fails++;
3403 	mutex_unlock(&dev->rdev.stats.lock);
3404 	connect_reply_upcall(ep, status2errno(req->retval));
3405 	state_set(&ep->com, DEAD);
3406 	remove_handle(dev, &dev->atid_idr, atid);
3407 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3408 	dst_release(ep->dst);
3409 	cxgb4_l2t_release(ep->l2t);
3410 	c4iw_put_ep(&ep->com);
3411 }
3412 
3413 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3414 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3415 {
3416 	struct sk_buff *rpl_skb;
3417 	struct cpl_pass_accept_req *cpl;
3418 	int ret;
3419 
3420 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3421 	BUG_ON(!rpl_skb);
3422 	if (req->retval) {
3423 		PDBG("%s passive open failure %d\n", __func__, req->retval);
3424 		mutex_lock(&dev->rdev.stats.lock);
3425 		dev->rdev.stats.pas_ofld_conn_fails++;
3426 		mutex_unlock(&dev->rdev.stats.lock);
3427 		kfree_skb(rpl_skb);
3428 	} else {
3429 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3430 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3431 					(__force u32) htonl(
3432 					(__force u32) req->tid)));
3433 		ret = pass_accept_req(dev, rpl_skb);
3434 		if (!ret)
3435 			kfree_skb(rpl_skb);
3436 	}
3437 	return;
3438 }
3439 
3440 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3441 {
3442 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3443 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3444 
3445 	switch (rpl->type) {
3446 	case FW6_TYPE_CQE:
3447 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3448 		break;
3449 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3450 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3451 		switch (req->t_state) {
3452 		case TCP_SYN_SENT:
3453 			active_ofld_conn_reply(dev, skb, req);
3454 			break;
3455 		case TCP_SYN_RECV:
3456 			passive_ofld_conn_reply(dev, skb, req);
3457 			break;
3458 		default:
3459 			pr_err("%s unexpected ofld conn wr state %d\n",
3460 			       __func__, req->t_state);
3461 			break;
3462 		}
3463 		break;
3464 	}
3465 	return 0;
3466 }
3467 
3468 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3469 {
3470 	u32 l2info;
3471 	u16 vlantag, len, hdr_len, eth_hdr_len;
3472 	u8 intf;
3473 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3474 	struct cpl_pass_accept_req *req;
3475 	struct tcp_options_received tmp_opt;
3476 	struct c4iw_dev *dev;
3477 
3478 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3479 	/* Store values from cpl_rx_pkt in temporary location. */
3480 	vlantag = (__force u16) cpl->vlan;
3481 	len = (__force u16) cpl->len;
3482 	l2info  = (__force u32) cpl->l2info;
3483 	hdr_len = (__force u16) cpl->hdr_len;
3484 	intf = cpl->iff;
3485 
3486 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3487 
3488 	/*
3489 	 * We need to parse the TCP options from SYN packet.
3490 	 * to generate cpl_pass_accept_req.
3491 	 */
3492 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3493 	tcp_clear_options(&tmp_opt);
3494 	tcp_parse_options(skb, &tmp_opt, 0, NULL);
3495 
3496 	req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3497 	memset(req, 0, sizeof(*req));
3498 	req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3499 			 V_SYN_MAC_IDX(G_RX_MACIDX(
3500 			 (__force int) htonl(l2info))) |
3501 			 F_SYN_XACT_MATCH);
3502 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3503 			    G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3504 			    G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3505 	req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3506 					(__force int) htonl(l2info))) |
3507 				   V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3508 					(__force int) htons(hdr_len))) |
3509 				   V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3510 					(__force int) htons(hdr_len))) |
3511 				   V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3512 	req->vlan = (__force __be16) vlantag;
3513 	req->len = (__force __be16) len;
3514 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3515 				    PASS_OPEN_TOS(tos));
3516 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3517 	if (tmp_opt.wscale_ok)
3518 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3519 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3520 	if (tmp_opt.sack_ok)
3521 		req->tcpopt.sack = 1;
3522 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3523 	return;
3524 }
3525 
3526 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3527 				  __be32 laddr, __be16 lport,
3528 				  __be32 raddr, __be16 rport,
3529 				  u32 rcv_isn, u32 filter, u16 window,
3530 				  u32 rss_qid, u8 port_id)
3531 {
3532 	struct sk_buff *req_skb;
3533 	struct fw_ofld_connection_wr *req;
3534 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
3535 	int ret;
3536 
3537 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3538 	req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3539 	memset(req, 0, sizeof(*req));
3540 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3541 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3542 	req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3543 	req->le.filter = (__force __be32) filter;
3544 	req->le.lport = lport;
3545 	req->le.pport = rport;
3546 	req->le.u.ipv4.lip = laddr;
3547 	req->le.u.ipv4.pip = raddr;
3548 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3549 	req->tcb.rcv_adv = htons(window);
3550 	req->tcb.t_state_to_astid =
3551 		 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3552 			V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3553 			V_FW_OFLD_CONNECTION_WR_ASTID(
3554 			GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3555 
3556 	/*
3557 	 * We store the qid in opt2 which will be used by the firmware
3558 	 * to send us the wr response.
3559 	 */
3560 	req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3561 
3562 	/*
3563 	 * We initialize the MSS index in TCB to 0xF.
3564 	 * So that when driver sends cpl_pass_accept_rpl
3565 	 * TCB picks up the correct value. If this was 0
3566 	 * TP will ignore any value > 0 for MSS index.
3567 	 */
3568 	req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3569 	req->cookie = (unsigned long)skb;
3570 
3571 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3572 	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3573 	if (ret < 0) {
3574 		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3575 		       ret);
3576 		kfree_skb(skb);
3577 		kfree_skb(req_skb);
3578 	}
3579 }
3580 
3581 /*
3582  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3583  * messages when a filter is being used instead of server to
3584  * redirect a syn packet. When packets hit filter they are redirected
3585  * to the offload queue and driver tries to establish the connection
3586  * using firmware work request.
3587  */
3588 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3589 {
3590 	int stid;
3591 	unsigned int filter;
3592 	struct ethhdr *eh = NULL;
3593 	struct vlan_ethhdr *vlan_eh = NULL;
3594 	struct iphdr *iph;
3595 	struct tcphdr *tcph;
3596 	struct rss_header *rss = (void *)skb->data;
3597 	struct cpl_rx_pkt *cpl = (void *)skb->data;
3598 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
3599 	struct l2t_entry *e;
3600 	struct dst_entry *dst;
3601 	struct c4iw_ep *lep;
3602 	u16 window;
3603 	struct port_info *pi;
3604 	struct net_device *pdev;
3605 	u16 rss_qid, eth_hdr_len;
3606 	int step;
3607 	u32 tx_chan;
3608 	struct neighbour *neigh;
3609 
3610 	/* Drop all non-SYN packets */
3611 	if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3612 		goto reject;
3613 
3614 	/*
3615 	 * Drop all packets which did not hit the filter.
3616 	 * Unlikely to happen.
3617 	 */
3618 	if (!(rss->filter_hit && rss->filter_tid))
3619 		goto reject;
3620 
3621 	/*
3622 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3623 	 */
3624 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3625 
3626 	lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3627 	if (!lep) {
3628 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3629 		goto reject;
3630 	}
3631 
3632 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3633 			    G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3634 			    G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3635 	if (eth_hdr_len == ETH_HLEN) {
3636 		eh = (struct ethhdr *)(req + 1);
3637 		iph = (struct iphdr *)(eh + 1);
3638 	} else {
3639 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
3640 		iph = (struct iphdr *)(vlan_eh + 1);
3641 		skb->vlan_tci = ntohs(cpl->vlan);
3642 	}
3643 
3644 	if (iph->version != 0x4)
3645 		goto reject;
3646 
3647 	tcph = (struct tcphdr *)(iph + 1);
3648 	skb_set_network_header(skb, (void *)iph - (void *)rss);
3649 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3650 	skb_get(skb);
3651 
3652 	PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3653 	     ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3654 	     ntohs(tcph->source), iph->tos);
3655 
3656 	dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3657 			 iph->tos);
3658 	if (!dst) {
3659 		pr_err("%s - failed to find dst entry!\n",
3660 		       __func__);
3661 		goto reject;
3662 	}
3663 	neigh = dst_neigh_lookup_skb(dst, skb);
3664 
3665 	if (!neigh) {
3666 		pr_err("%s - failed to allocate neigh!\n",
3667 		       __func__);
3668 		goto free_dst;
3669 	}
3670 
3671 	if (neigh->dev->flags & IFF_LOOPBACK) {
3672 		pdev = ip_dev_find(&init_net, iph->daddr);
3673 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3674 				    pdev, 0);
3675 		pi = (struct port_info *)netdev_priv(pdev);
3676 		tx_chan = cxgb4_port_chan(pdev);
3677 		dev_put(pdev);
3678 	} else {
3679 		pdev = get_real_dev(neigh->dev);
3680 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3681 					pdev, 0);
3682 		pi = (struct port_info *)netdev_priv(pdev);
3683 		tx_chan = cxgb4_port_chan(pdev);
3684 	}
3685 	neigh_release(neigh);
3686 	if (!e) {
3687 		pr_err("%s - failed to allocate l2t entry!\n",
3688 		       __func__);
3689 		goto free_dst;
3690 	}
3691 
3692 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3693 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3694 	window = (__force u16) htons((__force u16)tcph->window);
3695 
3696 	/* Calcuate filter portion for LE region. */
3697 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3698 						    dev->rdev.lldi.ports[0],
3699 						    e));
3700 
3701 	/*
3702 	 * Synthesize the cpl_pass_accept_req. We have everything except the
3703 	 * TID. Once firmware sends a reply with TID we update the TID field
3704 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
3705 	 */
3706 	build_cpl_pass_accept_req(skb, stid, iph->tos);
3707 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3708 			      tcph->source, ntohl(tcph->seq), filter, window,
3709 			      rss_qid, pi->port_id);
3710 	cxgb4_l2t_release(e);
3711 free_dst:
3712 	dst_release(dst);
3713 reject:
3714 	return 0;
3715 }
3716 
3717 /*
3718  * These are the real handlers that are called from a
3719  * work queue.
3720  */
3721 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3722 	[CPL_ACT_ESTABLISH] = act_establish,
3723 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
3724 	[CPL_RX_DATA] = rx_data,
3725 	[CPL_ABORT_RPL_RSS] = abort_rpl,
3726 	[CPL_ABORT_RPL] = abort_rpl,
3727 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
3728 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3729 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3730 	[CPL_PASS_ESTABLISH] = pass_establish,
3731 	[CPL_PEER_CLOSE] = peer_close,
3732 	[CPL_ABORT_REQ_RSS] = peer_abort,
3733 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
3734 	[CPL_RDMA_TERMINATE] = terminate,
3735 	[CPL_FW4_ACK] = fw4_ack,
3736 	[CPL_FW6_MSG] = deferred_fw6_msg,
3737 	[CPL_RX_PKT] = rx_pkt
3738 };
3739 
3740 static void process_timeout(struct c4iw_ep *ep)
3741 {
3742 	struct c4iw_qp_attributes attrs;
3743 	int abort = 1;
3744 
3745 	mutex_lock(&ep->com.mutex);
3746 	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3747 	     ep->com.state);
3748 	set_bit(TIMEDOUT, &ep->com.history);
3749 	switch (ep->com.state) {
3750 	case MPA_REQ_SENT:
3751 		__state_set(&ep->com, ABORTING);
3752 		connect_reply_upcall(ep, -ETIMEDOUT);
3753 		break;
3754 	case MPA_REQ_WAIT:
3755 		__state_set(&ep->com, ABORTING);
3756 		break;
3757 	case CLOSING:
3758 	case MORIBUND:
3759 		if (ep->com.cm_id && ep->com.qp) {
3760 			attrs.next_state = C4IW_QP_STATE_ERROR;
3761 			c4iw_modify_qp(ep->com.qp->rhp,
3762 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3763 				     &attrs, 1);
3764 		}
3765 		__state_set(&ep->com, ABORTING);
3766 		close_complete_upcall(ep, -ETIMEDOUT);
3767 		break;
3768 	case ABORTING:
3769 	case DEAD:
3770 
3771 		/*
3772 		 * These states are expected if the ep timed out at the same
3773 		 * time as another thread was calling stop_ep_timer().
3774 		 * So we silently do nothing for these states.
3775 		 */
3776 		abort = 0;
3777 		break;
3778 	default:
3779 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3780 			__func__, ep, ep->hwtid, ep->com.state);
3781 		abort = 0;
3782 	}
3783 	if (abort)
3784 		abort_connection(ep, NULL, GFP_KERNEL);
3785 	mutex_unlock(&ep->com.mutex);
3786 	c4iw_put_ep(&ep->com);
3787 }
3788 
3789 static void process_timedout_eps(void)
3790 {
3791 	struct c4iw_ep *ep;
3792 
3793 	spin_lock_irq(&timeout_lock);
3794 	while (!list_empty(&timeout_list)) {
3795 		struct list_head *tmp;
3796 
3797 		tmp = timeout_list.next;
3798 		list_del(tmp);
3799 		tmp->next = NULL;
3800 		tmp->prev = NULL;
3801 		spin_unlock_irq(&timeout_lock);
3802 		ep = list_entry(tmp, struct c4iw_ep, entry);
3803 		process_timeout(ep);
3804 		spin_lock_irq(&timeout_lock);
3805 	}
3806 	spin_unlock_irq(&timeout_lock);
3807 }
3808 
3809 static void process_work(struct work_struct *work)
3810 {
3811 	struct sk_buff *skb = NULL;
3812 	struct c4iw_dev *dev;
3813 	struct cpl_act_establish *rpl;
3814 	unsigned int opcode;
3815 	int ret;
3816 
3817 	process_timedout_eps();
3818 	while ((skb = skb_dequeue(&rxq))) {
3819 		rpl = cplhdr(skb);
3820 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3821 		opcode = rpl->ot.opcode;
3822 
3823 		BUG_ON(!work_handlers[opcode]);
3824 		ret = work_handlers[opcode](dev, skb);
3825 		if (!ret)
3826 			kfree_skb(skb);
3827 		process_timedout_eps();
3828 	}
3829 }
3830 
3831 static DECLARE_WORK(skb_work, process_work);
3832 
3833 static void ep_timeout(unsigned long arg)
3834 {
3835 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3836 	int kickit = 0;
3837 
3838 	spin_lock(&timeout_lock);
3839 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3840 		/*
3841 		 * Only insert if it is not already on the list.
3842 		 */
3843 		if (!ep->entry.next) {
3844 			list_add_tail(&ep->entry, &timeout_list);
3845 			kickit = 1;
3846 		}
3847 	}
3848 	spin_unlock(&timeout_lock);
3849 	if (kickit)
3850 		queue_work(workq, &skb_work);
3851 }
3852 
3853 /*
3854  * All the CM events are handled on a work queue to have a safe context.
3855  */
3856 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3857 {
3858 
3859 	/*
3860 	 * Save dev in the skb->cb area.
3861 	 */
3862 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3863 
3864 	/*
3865 	 * Queue the skb and schedule the worker thread.
3866 	 */
3867 	skb_queue_tail(&rxq, skb);
3868 	queue_work(workq, &skb_work);
3869 	return 0;
3870 }
3871 
3872 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3873 {
3874 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3875 
3876 	if (rpl->status != CPL_ERR_NONE) {
3877 		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3878 		       "for tid %u\n", rpl->status, GET_TID(rpl));
3879 	}
3880 	kfree_skb(skb);
3881 	return 0;
3882 }
3883 
3884 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3885 {
3886 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3887 	struct c4iw_wr_wait *wr_waitp;
3888 	int ret;
3889 
3890 	PDBG("%s type %u\n", __func__, rpl->type);
3891 
3892 	switch (rpl->type) {
3893 	case FW6_TYPE_WR_RPL:
3894 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3895 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3896 		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3897 		if (wr_waitp)
3898 			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3899 		kfree_skb(skb);
3900 		break;
3901 	case FW6_TYPE_CQE:
3902 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3903 		sched(dev, skb);
3904 		break;
3905 	default:
3906 		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3907 		       rpl->type);
3908 		kfree_skb(skb);
3909 		break;
3910 	}
3911 	return 0;
3912 }
3913 
3914 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3915 {
3916 	struct cpl_abort_req_rss *req = cplhdr(skb);
3917 	struct c4iw_ep *ep;
3918 	struct tid_info *t = dev->rdev.lldi.tids;
3919 	unsigned int tid = GET_TID(req);
3920 
3921 	ep = lookup_tid(t, tid);
3922 	if (!ep) {
3923 		printk(KERN_WARNING MOD
3924 		       "Abort on non-existent endpoint, tid %d\n", tid);
3925 		kfree_skb(skb);
3926 		return 0;
3927 	}
3928 	if (is_neg_adv(req->status)) {
3929 		dev_warn(&dev->rdev.lldi.pdev->dev,
3930 			 "Negative advice on abort - tid %u status %d (%s)\n",
3931 			 ep->hwtid, req->status, neg_adv_str(req->status));
3932 		kfree_skb(skb);
3933 		return 0;
3934 	}
3935 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3936 	     ep->com.state);
3937 
3938 	/*
3939 	 * Wake up any threads in rdma_init() or rdma_fini().
3940 	 * However, if we are on MPAv2 and want to retry with MPAv1
3941 	 * then, don't wake up yet.
3942 	 */
3943 	if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3944 		if (ep->com.state != MPA_REQ_SENT)
3945 			c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3946 	} else
3947 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3948 	sched(dev, skb);
3949 	return 0;
3950 }
3951 
3952 /*
3953  * Most upcalls from the T4 Core go to sched() to
3954  * schedule the processing on a work queue.
3955  */
3956 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3957 	[CPL_ACT_ESTABLISH] = sched,
3958 	[CPL_ACT_OPEN_RPL] = sched,
3959 	[CPL_RX_DATA] = sched,
3960 	[CPL_ABORT_RPL_RSS] = sched,
3961 	[CPL_ABORT_RPL] = sched,
3962 	[CPL_PASS_OPEN_RPL] = sched,
3963 	[CPL_CLOSE_LISTSRV_RPL] = sched,
3964 	[CPL_PASS_ACCEPT_REQ] = sched,
3965 	[CPL_PASS_ESTABLISH] = sched,
3966 	[CPL_PEER_CLOSE] = sched,
3967 	[CPL_CLOSE_CON_RPL] = sched,
3968 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
3969 	[CPL_RDMA_TERMINATE] = sched,
3970 	[CPL_FW4_ACK] = sched,
3971 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
3972 	[CPL_FW6_MSG] = fw6_msg,
3973 	[CPL_RX_PKT] = sched
3974 };
3975 
3976 int __init c4iw_cm_init(void)
3977 {
3978 	spin_lock_init(&timeout_lock);
3979 	skb_queue_head_init(&rxq);
3980 
3981 	workq = create_singlethread_workqueue("iw_cxgb4");
3982 	if (!workq)
3983 		return -ENOMEM;
3984 
3985 	return 0;
3986 }
3987 
3988 void c4iw_cm_term(void)
3989 {
3990 	WARN_ON(!list_empty(&timeout_list));
3991 	flush_workqueue(workq);
3992 	destroy_workqueue(workq);
3993 }
3994