xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision 31b90347)
1 /*
2  * Copyright (c) 2009-2010 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 "iw_cxgb4.h"
51 
52 static char *states[] = {
53 	"idle",
54 	"listen",
55 	"connecting",
56 	"mpa_wait_req",
57 	"mpa_req_sent",
58 	"mpa_req_rcvd",
59 	"mpa_rep_sent",
60 	"fpdu_mode",
61 	"aborting",
62 	"closing",
63 	"moribund",
64 	"dead",
65 	NULL,
66 };
67 
68 static int nocong;
69 module_param(nocong, int, 0644);
70 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
71 
72 static int enable_ecn;
73 module_param(enable_ecn, int, 0644);
74 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
75 
76 static int dack_mode = 1;
77 module_param(dack_mode, int, 0644);
78 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
79 
80 int c4iw_max_read_depth = 8;
81 module_param(c4iw_max_read_depth, int, 0644);
82 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
83 
84 static int enable_tcp_timestamps;
85 module_param(enable_tcp_timestamps, int, 0644);
86 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
87 
88 static int enable_tcp_sack;
89 module_param(enable_tcp_sack, int, 0644);
90 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
91 
92 static int enable_tcp_window_scaling = 1;
93 module_param(enable_tcp_window_scaling, int, 0644);
94 MODULE_PARM_DESC(enable_tcp_window_scaling,
95 		 "Enable tcp window scaling (default=1)");
96 
97 int c4iw_debug;
98 module_param(c4iw_debug, int, 0644);
99 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
100 
101 static int peer2peer;
102 module_param(peer2peer, int, 0644);
103 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
104 
105 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
106 module_param(p2p_type, int, 0644);
107 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
108 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
109 
110 static int ep_timeout_secs = 60;
111 module_param(ep_timeout_secs, int, 0644);
112 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
113 				   "in seconds (default=60)");
114 
115 static int mpa_rev = 1;
116 module_param(mpa_rev, int, 0644);
117 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
118 		"1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
119 		" compliant (default=1)");
120 
121 static int markers_enabled;
122 module_param(markers_enabled, int, 0644);
123 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
124 
125 static int crc_enabled = 1;
126 module_param(crc_enabled, int, 0644);
127 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
128 
129 static int rcv_win = 256 * 1024;
130 module_param(rcv_win, int, 0644);
131 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
132 
133 static int snd_win = 128 * 1024;
134 module_param(snd_win, int, 0644);
135 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
136 
137 static struct workqueue_struct *workq;
138 
139 static struct sk_buff_head rxq;
140 
141 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
142 static void ep_timeout(unsigned long arg);
143 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
144 
145 static LIST_HEAD(timeout_list);
146 static spinlock_t timeout_lock;
147 
148 static void deref_qp(struct c4iw_ep *ep)
149 {
150 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
151 	clear_bit(QP_REFERENCED, &ep->com.flags);
152 }
153 
154 static void ref_qp(struct c4iw_ep *ep)
155 {
156 	set_bit(QP_REFERENCED, &ep->com.flags);
157 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
158 }
159 
160 static void start_ep_timer(struct c4iw_ep *ep)
161 {
162 	PDBG("%s ep %p\n", __func__, ep);
163 	if (timer_pending(&ep->timer)) {
164 		pr_err("%s timer already started! ep %p\n",
165 		       __func__, ep);
166 		return;
167 	}
168 	clear_bit(TIMEOUT, &ep->com.flags);
169 	c4iw_get_ep(&ep->com);
170 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
171 	ep->timer.data = (unsigned long)ep;
172 	ep->timer.function = ep_timeout;
173 	add_timer(&ep->timer);
174 }
175 
176 static void stop_ep_timer(struct c4iw_ep *ep)
177 {
178 	PDBG("%s ep %p stopping\n", __func__, ep);
179 	del_timer_sync(&ep->timer);
180 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
181 		c4iw_put_ep(&ep->com);
182 }
183 
184 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
185 		  struct l2t_entry *l2e)
186 {
187 	int	error = 0;
188 
189 	if (c4iw_fatal_error(rdev)) {
190 		kfree_skb(skb);
191 		PDBG("%s - device in error state - dropping\n", __func__);
192 		return -EIO;
193 	}
194 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
195 	if (error < 0)
196 		kfree_skb(skb);
197 	return error < 0 ? error : 0;
198 }
199 
200 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
201 {
202 	int	error = 0;
203 
204 	if (c4iw_fatal_error(rdev)) {
205 		kfree_skb(skb);
206 		PDBG("%s - device in error state - dropping\n", __func__);
207 		return -EIO;
208 	}
209 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
210 	if (error < 0)
211 		kfree_skb(skb);
212 	return error < 0 ? error : 0;
213 }
214 
215 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
216 {
217 	struct cpl_tid_release *req;
218 
219 	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
220 	if (!skb)
221 		return;
222 	req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
223 	INIT_TP_WR(req, hwtid);
224 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
225 	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
226 	c4iw_ofld_send(rdev, skb);
227 	return;
228 }
229 
230 static void set_emss(struct c4iw_ep *ep, u16 opt)
231 {
232 	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
233 	ep->mss = ep->emss;
234 	if (GET_TCPOPT_TSTAMP(opt))
235 		ep->emss -= 12;
236 	if (ep->emss < 128)
237 		ep->emss = 128;
238 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
239 	     ep->mss, ep->emss);
240 }
241 
242 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
243 {
244 	enum c4iw_ep_state state;
245 
246 	mutex_lock(&epc->mutex);
247 	state = epc->state;
248 	mutex_unlock(&epc->mutex);
249 	return state;
250 }
251 
252 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
253 {
254 	epc->state = new;
255 }
256 
257 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
258 {
259 	mutex_lock(&epc->mutex);
260 	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
261 	__state_set(epc, new);
262 	mutex_unlock(&epc->mutex);
263 	return;
264 }
265 
266 static void *alloc_ep(int size, gfp_t gfp)
267 {
268 	struct c4iw_ep_common *epc;
269 
270 	epc = kzalloc(size, gfp);
271 	if (epc) {
272 		kref_init(&epc->kref);
273 		mutex_init(&epc->mutex);
274 		c4iw_init_wr_wait(&epc->wr_wait);
275 	}
276 	PDBG("%s alloc ep %p\n", __func__, epc);
277 	return epc;
278 }
279 
280 void _c4iw_free_ep(struct kref *kref)
281 {
282 	struct c4iw_ep *ep;
283 
284 	ep = container_of(kref, struct c4iw_ep, com.kref);
285 	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
286 	if (test_bit(QP_REFERENCED, &ep->com.flags))
287 		deref_qp(ep);
288 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
289 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
290 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
291 		dst_release(ep->dst);
292 		cxgb4_l2t_release(ep->l2t);
293 	}
294 	kfree(ep);
295 }
296 
297 static void release_ep_resources(struct c4iw_ep *ep)
298 {
299 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
300 	c4iw_put_ep(&ep->com);
301 }
302 
303 static int status2errno(int status)
304 {
305 	switch (status) {
306 	case CPL_ERR_NONE:
307 		return 0;
308 	case CPL_ERR_CONN_RESET:
309 		return -ECONNRESET;
310 	case CPL_ERR_ARP_MISS:
311 		return -EHOSTUNREACH;
312 	case CPL_ERR_CONN_TIMEDOUT:
313 		return -ETIMEDOUT;
314 	case CPL_ERR_TCAM_FULL:
315 		return -ENOMEM;
316 	case CPL_ERR_CONN_EXIST:
317 		return -EADDRINUSE;
318 	default:
319 		return -EIO;
320 	}
321 }
322 
323 /*
324  * Try and reuse skbs already allocated...
325  */
326 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
327 {
328 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
329 		skb_trim(skb, 0);
330 		skb_get(skb);
331 		skb_reset_transport_header(skb);
332 	} else {
333 		skb = alloc_skb(len, gfp);
334 	}
335 	t4_set_arp_err_handler(skb, NULL, NULL);
336 	return skb;
337 }
338 
339 static struct net_device *get_real_dev(struct net_device *egress_dev)
340 {
341 	struct net_device *phys_dev = egress_dev;
342 	if (egress_dev->priv_flags & IFF_802_1Q_VLAN)
343 		phys_dev = vlan_dev_real_dev(egress_dev);
344 	return phys_dev;
345 }
346 
347 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
348 {
349 	int i;
350 
351 	egress_dev = get_real_dev(egress_dev);
352 	for (i = 0; i < dev->rdev.lldi.nports; i++)
353 		if (dev->rdev.lldi.ports[i] == egress_dev)
354 			return 1;
355 	return 0;
356 }
357 
358 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
359 				     __u8 *peer_ip, __be16 local_port,
360 				     __be16 peer_port, u8 tos,
361 				     __u32 sin6_scope_id)
362 {
363 	struct dst_entry *dst = NULL;
364 
365 	if (IS_ENABLED(CONFIG_IPV6)) {
366 		struct flowi6 fl6;
367 
368 		memset(&fl6, 0, sizeof(fl6));
369 		memcpy(&fl6.daddr, peer_ip, 16);
370 		memcpy(&fl6.saddr, local_ip, 16);
371 		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
372 			fl6.flowi6_oif = sin6_scope_id;
373 		dst = ip6_route_output(&init_net, NULL, &fl6);
374 		if (!dst)
375 			goto out;
376 		if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
377 		    !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
378 			dst_release(dst);
379 			dst = NULL;
380 		}
381 	}
382 
383 out:
384 	return dst;
385 }
386 
387 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
388 				 __be32 peer_ip, __be16 local_port,
389 				 __be16 peer_port, u8 tos)
390 {
391 	struct rtable *rt;
392 	struct flowi4 fl4;
393 	struct neighbour *n;
394 
395 	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
396 				   peer_port, local_port, IPPROTO_TCP,
397 				   tos, 0);
398 	if (IS_ERR(rt))
399 		return NULL;
400 	n = dst_neigh_lookup(&rt->dst, &peer_ip);
401 	if (!n)
402 		return NULL;
403 	if (!our_interface(dev, n->dev)) {
404 		dst_release(&rt->dst);
405 		return NULL;
406 	}
407 	neigh_release(n);
408 	return &rt->dst;
409 }
410 
411 static void arp_failure_discard(void *handle, struct sk_buff *skb)
412 {
413 	PDBG("%s c4iw_dev %p\n", __func__, handle);
414 	kfree_skb(skb);
415 }
416 
417 /*
418  * Handle an ARP failure for an active open.
419  */
420 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
421 {
422 	printk(KERN_ERR MOD "ARP failure duing connect\n");
423 	kfree_skb(skb);
424 }
425 
426 /*
427  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
428  * and send it along.
429  */
430 static void abort_arp_failure(void *handle, struct sk_buff *skb)
431 {
432 	struct c4iw_rdev *rdev = handle;
433 	struct cpl_abort_req *req = cplhdr(skb);
434 
435 	PDBG("%s rdev %p\n", __func__, rdev);
436 	req->cmd = CPL_ABORT_NO_RST;
437 	c4iw_ofld_send(rdev, skb);
438 }
439 
440 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
441 {
442 	unsigned int flowclen = 80;
443 	struct fw_flowc_wr *flowc;
444 	int i;
445 
446 	skb = get_skb(skb, flowclen, GFP_KERNEL);
447 	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
448 
449 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
450 					   FW_FLOWC_WR_NPARAMS(8));
451 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
452 					  16)) | FW_WR_FLOWID(ep->hwtid));
453 
454 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
455 	flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
456 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
457 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
458 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
459 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
460 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
461 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
462 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
463 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
464 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
465 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
466 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
467 	flowc->mnemval[6].val = cpu_to_be32(snd_win);
468 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
469 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
470 	/* Pad WR to 16 byte boundary */
471 	flowc->mnemval[8].mnemonic = 0;
472 	flowc->mnemval[8].val = 0;
473 	for (i = 0; i < 9; i++) {
474 		flowc->mnemval[i].r4[0] = 0;
475 		flowc->mnemval[i].r4[1] = 0;
476 		flowc->mnemval[i].r4[2] = 0;
477 	}
478 
479 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
480 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
481 }
482 
483 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
484 {
485 	struct cpl_close_con_req *req;
486 	struct sk_buff *skb;
487 	int wrlen = roundup(sizeof *req, 16);
488 
489 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
490 	skb = get_skb(NULL, wrlen, gfp);
491 	if (!skb) {
492 		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
493 		return -ENOMEM;
494 	}
495 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
496 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
497 	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
498 	memset(req, 0, wrlen);
499 	INIT_TP_WR(req, ep->hwtid);
500 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
501 						    ep->hwtid));
502 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
503 }
504 
505 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
506 {
507 	struct cpl_abort_req *req;
508 	int wrlen = roundup(sizeof *req, 16);
509 
510 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
511 	skb = get_skb(skb, wrlen, gfp);
512 	if (!skb) {
513 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
514 		       __func__);
515 		return -ENOMEM;
516 	}
517 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
518 	t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
519 	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
520 	memset(req, 0, wrlen);
521 	INIT_TP_WR(req, ep->hwtid);
522 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
523 	req->cmd = CPL_ABORT_SEND_RST;
524 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
525 }
526 
527 static int send_connect(struct c4iw_ep *ep)
528 {
529 	struct cpl_act_open_req *req;
530 	struct cpl_t5_act_open_req *t5_req;
531 	struct cpl_act_open_req6 *req6;
532 	struct cpl_t5_act_open_req6 *t5_req6;
533 	struct sk_buff *skb;
534 	u64 opt0;
535 	u32 opt2;
536 	unsigned int mtu_idx;
537 	int wscale;
538 	int wrlen;
539 	int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
540 				sizeof(struct cpl_act_open_req) :
541 				sizeof(struct cpl_t5_act_open_req);
542 	int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
543 				sizeof(struct cpl_act_open_req6) :
544 				sizeof(struct cpl_t5_act_open_req6);
545 	struct sockaddr_in *la = (struct sockaddr_in *)&ep->com.local_addr;
546 	struct sockaddr_in *ra = (struct sockaddr_in *)&ep->com.remote_addr;
547 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
548 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
549 
550 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
551 			roundup(sizev4, 16) :
552 			roundup(sizev6, 16);
553 
554 	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
555 
556 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
557 	if (!skb) {
558 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
559 		       __func__);
560 		return -ENOMEM;
561 	}
562 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
563 
564 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
565 	wscale = compute_wscale(rcv_win);
566 	opt0 = (nocong ? NO_CONG(1) : 0) |
567 	       KEEP_ALIVE(1) |
568 	       DELACK(1) |
569 	       WND_SCALE(wscale) |
570 	       MSS_IDX(mtu_idx) |
571 	       L2T_IDX(ep->l2t->idx) |
572 	       TX_CHAN(ep->tx_chan) |
573 	       SMAC_SEL(ep->smac_idx) |
574 	       DSCP(ep->tos) |
575 	       ULP_MODE(ULP_MODE_TCPDDP) |
576 	       RCV_BUFSIZ(rcv_win>>10);
577 	opt2 = RX_CHANNEL(0) |
578 	       CCTRL_ECN(enable_ecn) |
579 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
580 	if (enable_tcp_timestamps)
581 		opt2 |= TSTAMPS_EN(1);
582 	if (enable_tcp_sack)
583 		opt2 |= SACK_EN(1);
584 	if (wscale && enable_tcp_window_scaling)
585 		opt2 |= WND_SCALE_EN(1);
586 	t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
587 
588 	if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
589 		if (ep->com.remote_addr.ss_family == AF_INET) {
590 			req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
591 			INIT_TP_WR(req, 0);
592 			OPCODE_TID(req) = cpu_to_be32(
593 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
594 					((ep->rss_qid << 14) | ep->atid)));
595 			req->local_port = la->sin_port;
596 			req->peer_port = ra->sin_port;
597 			req->local_ip = la->sin_addr.s_addr;
598 			req->peer_ip = ra->sin_addr.s_addr;
599 			req->opt0 = cpu_to_be64(opt0);
600 			req->params = cpu_to_be32(cxgb4_select_ntuple(
601 						ep->com.dev->rdev.lldi.ports[0],
602 						ep->l2t));
603 			req->opt2 = cpu_to_be32(opt2);
604 		} else {
605 			req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
606 
607 			INIT_TP_WR(req6, 0);
608 			OPCODE_TID(req6) = cpu_to_be32(
609 					   MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
610 					   ((ep->rss_qid<<14)|ep->atid)));
611 			req6->local_port = la6->sin6_port;
612 			req6->peer_port = ra6->sin6_port;
613 			req6->local_ip_hi = *((__be64 *)
614 						(la6->sin6_addr.s6_addr));
615 			req6->local_ip_lo = *((__be64 *)
616 						(la6->sin6_addr.s6_addr + 8));
617 			req6->peer_ip_hi = *((__be64 *)
618 						(ra6->sin6_addr.s6_addr));
619 			req6->peer_ip_lo = *((__be64 *)
620 						(ra6->sin6_addr.s6_addr + 8));
621 			req6->opt0 = cpu_to_be64(opt0);
622 			req6->params = cpu_to_be32(cxgb4_select_ntuple(
623 						ep->com.dev->rdev.lldi.ports[0],
624 						ep->l2t));
625 			req6->opt2 = cpu_to_be32(opt2);
626 		}
627 	} else {
628 		if (ep->com.remote_addr.ss_family == AF_INET) {
629 			t5_req = (struct cpl_t5_act_open_req *)
630 				 skb_put(skb, wrlen);
631 			INIT_TP_WR(t5_req, 0);
632 			OPCODE_TID(t5_req) = cpu_to_be32(
633 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
634 					((ep->rss_qid << 14) | ep->atid)));
635 			t5_req->local_port = la->sin_port;
636 			t5_req->peer_port = ra->sin_port;
637 			t5_req->local_ip = la->sin_addr.s_addr;
638 			t5_req->peer_ip = ra->sin_addr.s_addr;
639 			t5_req->opt0 = cpu_to_be64(opt0);
640 			t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
641 						     cxgb4_select_ntuple(
642 					     ep->com.dev->rdev.lldi.ports[0],
643 					     ep->l2t)));
644 			t5_req->opt2 = cpu_to_be32(opt2);
645 		} else {
646 			t5_req6 = (struct cpl_t5_act_open_req6 *)
647 				  skb_put(skb, wrlen);
648 			INIT_TP_WR(t5_req6, 0);
649 			OPCODE_TID(t5_req6) = cpu_to_be32(
650 					      MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
651 					      ((ep->rss_qid<<14)|ep->atid)));
652 			t5_req6->local_port = la6->sin6_port;
653 			t5_req6->peer_port = ra6->sin6_port;
654 			t5_req6->local_ip_hi = *((__be64 *)
655 						(la6->sin6_addr.s6_addr));
656 			t5_req6->local_ip_lo = *((__be64 *)
657 						(la6->sin6_addr.s6_addr + 8));
658 			t5_req6->peer_ip_hi = *((__be64 *)
659 						(ra6->sin6_addr.s6_addr));
660 			t5_req6->peer_ip_lo = *((__be64 *)
661 						(ra6->sin6_addr.s6_addr + 8));
662 			t5_req6->opt0 = cpu_to_be64(opt0);
663 			t5_req6->params = (__force __be64)cpu_to_be32(
664 							cxgb4_select_ntuple(
665 						ep->com.dev->rdev.lldi.ports[0],
666 						ep->l2t));
667 			t5_req6->opt2 = cpu_to_be32(opt2);
668 		}
669 	}
670 
671 	set_bit(ACT_OPEN_REQ, &ep->com.history);
672 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
673 }
674 
675 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
676 		u8 mpa_rev_to_use)
677 {
678 	int mpalen, wrlen;
679 	struct fw_ofld_tx_data_wr *req;
680 	struct mpa_message *mpa;
681 	struct mpa_v2_conn_params mpa_v2_params;
682 
683 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
684 
685 	BUG_ON(skb_cloned(skb));
686 
687 	mpalen = sizeof(*mpa) + ep->plen;
688 	if (mpa_rev_to_use == 2)
689 		mpalen += sizeof(struct mpa_v2_conn_params);
690 	wrlen = roundup(mpalen + sizeof *req, 16);
691 	skb = get_skb(skb, wrlen, GFP_KERNEL);
692 	if (!skb) {
693 		connect_reply_upcall(ep, -ENOMEM);
694 		return;
695 	}
696 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
697 
698 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
699 	memset(req, 0, wrlen);
700 	req->op_to_immdlen = cpu_to_be32(
701 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
702 		FW_WR_COMPL(1) |
703 		FW_WR_IMMDLEN(mpalen));
704 	req->flowid_len16 = cpu_to_be32(
705 		FW_WR_FLOWID(ep->hwtid) |
706 		FW_WR_LEN16(wrlen >> 4));
707 	req->plen = cpu_to_be32(mpalen);
708 	req->tunnel_to_proxy = cpu_to_be32(
709 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
710 		FW_OFLD_TX_DATA_WR_SHOVE(1));
711 
712 	mpa = (struct mpa_message *)(req + 1);
713 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
714 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
715 		     (markers_enabled ? MPA_MARKERS : 0) |
716 		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
717 	mpa->private_data_size = htons(ep->plen);
718 	mpa->revision = mpa_rev_to_use;
719 	if (mpa_rev_to_use == 1) {
720 		ep->tried_with_mpa_v1 = 1;
721 		ep->retry_with_mpa_v1 = 0;
722 	}
723 
724 	if (mpa_rev_to_use == 2) {
725 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
726 					       sizeof (struct mpa_v2_conn_params));
727 		mpa_v2_params.ird = htons((u16)ep->ird);
728 		mpa_v2_params.ord = htons((u16)ep->ord);
729 
730 		if (peer2peer) {
731 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
732 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
733 				mpa_v2_params.ord |=
734 					htons(MPA_V2_RDMA_WRITE_RTR);
735 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
736 				mpa_v2_params.ord |=
737 					htons(MPA_V2_RDMA_READ_RTR);
738 		}
739 		memcpy(mpa->private_data, &mpa_v2_params,
740 		       sizeof(struct mpa_v2_conn_params));
741 
742 		if (ep->plen)
743 			memcpy(mpa->private_data +
744 			       sizeof(struct mpa_v2_conn_params),
745 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
746 	} else
747 		if (ep->plen)
748 			memcpy(mpa->private_data,
749 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
750 
751 	/*
752 	 * Reference the mpa skb.  This ensures the data area
753 	 * will remain in memory until the hw acks the tx.
754 	 * Function fw4_ack() will deref it.
755 	 */
756 	skb_get(skb);
757 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
758 	BUG_ON(ep->mpa_skb);
759 	ep->mpa_skb = skb;
760 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
761 	start_ep_timer(ep);
762 	state_set(&ep->com, MPA_REQ_SENT);
763 	ep->mpa_attr.initiator = 1;
764 	return;
765 }
766 
767 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
768 {
769 	int mpalen, wrlen;
770 	struct fw_ofld_tx_data_wr *req;
771 	struct mpa_message *mpa;
772 	struct sk_buff *skb;
773 	struct mpa_v2_conn_params mpa_v2_params;
774 
775 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
776 
777 	mpalen = sizeof(*mpa) + plen;
778 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
779 		mpalen += sizeof(struct mpa_v2_conn_params);
780 	wrlen = roundup(mpalen + sizeof *req, 16);
781 
782 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
783 	if (!skb) {
784 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
785 		return -ENOMEM;
786 	}
787 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
788 
789 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
790 	memset(req, 0, wrlen);
791 	req->op_to_immdlen = cpu_to_be32(
792 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
793 		FW_WR_COMPL(1) |
794 		FW_WR_IMMDLEN(mpalen));
795 	req->flowid_len16 = cpu_to_be32(
796 		FW_WR_FLOWID(ep->hwtid) |
797 		FW_WR_LEN16(wrlen >> 4));
798 	req->plen = cpu_to_be32(mpalen);
799 	req->tunnel_to_proxy = cpu_to_be32(
800 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
801 		FW_OFLD_TX_DATA_WR_SHOVE(1));
802 
803 	mpa = (struct mpa_message *)(req + 1);
804 	memset(mpa, 0, sizeof(*mpa));
805 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
806 	mpa->flags = MPA_REJECT;
807 	mpa->revision = ep->mpa_attr.version;
808 	mpa->private_data_size = htons(plen);
809 
810 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
811 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
812 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
813 					       sizeof (struct mpa_v2_conn_params));
814 		mpa_v2_params.ird = htons(((u16)ep->ird) |
815 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
816 					   0));
817 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
818 					  (p2p_type ==
819 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
820 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
821 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
822 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
823 		memcpy(mpa->private_data, &mpa_v2_params,
824 		       sizeof(struct mpa_v2_conn_params));
825 
826 		if (ep->plen)
827 			memcpy(mpa->private_data +
828 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
829 	} else
830 		if (plen)
831 			memcpy(mpa->private_data, pdata, plen);
832 
833 	/*
834 	 * Reference the mpa skb again.  This ensures the data area
835 	 * will remain in memory until the hw acks the tx.
836 	 * Function fw4_ack() will deref it.
837 	 */
838 	skb_get(skb);
839 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
840 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
841 	BUG_ON(ep->mpa_skb);
842 	ep->mpa_skb = skb;
843 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
844 }
845 
846 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
847 {
848 	int mpalen, wrlen;
849 	struct fw_ofld_tx_data_wr *req;
850 	struct mpa_message *mpa;
851 	struct sk_buff *skb;
852 	struct mpa_v2_conn_params mpa_v2_params;
853 
854 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
855 
856 	mpalen = sizeof(*mpa) + plen;
857 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
858 		mpalen += sizeof(struct mpa_v2_conn_params);
859 	wrlen = roundup(mpalen + sizeof *req, 16);
860 
861 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
862 	if (!skb) {
863 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
864 		return -ENOMEM;
865 	}
866 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
867 
868 	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
869 	memset(req, 0, wrlen);
870 	req->op_to_immdlen = cpu_to_be32(
871 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
872 		FW_WR_COMPL(1) |
873 		FW_WR_IMMDLEN(mpalen));
874 	req->flowid_len16 = cpu_to_be32(
875 		FW_WR_FLOWID(ep->hwtid) |
876 		FW_WR_LEN16(wrlen >> 4));
877 	req->plen = cpu_to_be32(mpalen);
878 	req->tunnel_to_proxy = cpu_to_be32(
879 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
880 		FW_OFLD_TX_DATA_WR_SHOVE(1));
881 
882 	mpa = (struct mpa_message *)(req + 1);
883 	memset(mpa, 0, sizeof(*mpa));
884 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
885 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
886 		     (markers_enabled ? MPA_MARKERS : 0);
887 	mpa->revision = ep->mpa_attr.version;
888 	mpa->private_data_size = htons(plen);
889 
890 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
891 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
892 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
893 					       sizeof (struct mpa_v2_conn_params));
894 		mpa_v2_params.ird = htons((u16)ep->ird);
895 		mpa_v2_params.ord = htons((u16)ep->ord);
896 		if (peer2peer && (ep->mpa_attr.p2p_type !=
897 					FW_RI_INIT_P2PTYPE_DISABLED)) {
898 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
899 
900 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
901 				mpa_v2_params.ord |=
902 					htons(MPA_V2_RDMA_WRITE_RTR);
903 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
904 				mpa_v2_params.ord |=
905 					htons(MPA_V2_RDMA_READ_RTR);
906 		}
907 
908 		memcpy(mpa->private_data, &mpa_v2_params,
909 		       sizeof(struct mpa_v2_conn_params));
910 
911 		if (ep->plen)
912 			memcpy(mpa->private_data +
913 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
914 	} else
915 		if (plen)
916 			memcpy(mpa->private_data, pdata, plen);
917 
918 	/*
919 	 * Reference the mpa skb.  This ensures the data area
920 	 * will remain in memory until the hw acks the tx.
921 	 * Function fw4_ack() will deref it.
922 	 */
923 	skb_get(skb);
924 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
925 	ep->mpa_skb = skb;
926 	state_set(&ep->com, MPA_REP_SENT);
927 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
928 }
929 
930 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
931 {
932 	struct c4iw_ep *ep;
933 	struct cpl_act_establish *req = cplhdr(skb);
934 	unsigned int tid = GET_TID(req);
935 	unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
936 	struct tid_info *t = dev->rdev.lldi.tids;
937 
938 	ep = lookup_atid(t, atid);
939 
940 	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
941 	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
942 
943 	dst_confirm(ep->dst);
944 
945 	/* setup the hwtid for this connection */
946 	ep->hwtid = tid;
947 	cxgb4_insert_tid(t, ep, tid);
948 	insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
949 
950 	ep->snd_seq = be32_to_cpu(req->snd_isn);
951 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
952 
953 	set_emss(ep, ntohs(req->tcp_opt));
954 
955 	/* dealloc the atid */
956 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
957 	cxgb4_free_atid(t, atid);
958 	set_bit(ACT_ESTAB, &ep->com.history);
959 
960 	/* start MPA negotiation */
961 	send_flowc(ep, NULL);
962 	if (ep->retry_with_mpa_v1)
963 		send_mpa_req(ep, skb, 1);
964 	else
965 		send_mpa_req(ep, skb, mpa_rev);
966 
967 	return 0;
968 }
969 
970 static void close_complete_upcall(struct c4iw_ep *ep)
971 {
972 	struct iw_cm_event event;
973 
974 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
975 	memset(&event, 0, sizeof(event));
976 	event.event = IW_CM_EVENT_CLOSE;
977 	if (ep->com.cm_id) {
978 		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
979 		     ep, ep->com.cm_id, ep->hwtid);
980 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
981 		ep->com.cm_id->rem_ref(ep->com.cm_id);
982 		ep->com.cm_id = NULL;
983 		set_bit(CLOSE_UPCALL, &ep->com.history);
984 	}
985 }
986 
987 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
988 {
989 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
990 	close_complete_upcall(ep);
991 	state_set(&ep->com, ABORTING);
992 	set_bit(ABORT_CONN, &ep->com.history);
993 	return send_abort(ep, skb, gfp);
994 }
995 
996 static void peer_close_upcall(struct c4iw_ep *ep)
997 {
998 	struct iw_cm_event event;
999 
1000 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1001 	memset(&event, 0, sizeof(event));
1002 	event.event = IW_CM_EVENT_DISCONNECT;
1003 	if (ep->com.cm_id) {
1004 		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1005 		     ep, ep->com.cm_id, ep->hwtid);
1006 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1007 		set_bit(DISCONN_UPCALL, &ep->com.history);
1008 	}
1009 }
1010 
1011 static void peer_abort_upcall(struct c4iw_ep *ep)
1012 {
1013 	struct iw_cm_event event;
1014 
1015 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1016 	memset(&event, 0, sizeof(event));
1017 	event.event = IW_CM_EVENT_CLOSE;
1018 	event.status = -ECONNRESET;
1019 	if (ep->com.cm_id) {
1020 		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1021 		     ep->com.cm_id, ep->hwtid);
1022 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1023 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1024 		ep->com.cm_id = NULL;
1025 		set_bit(ABORT_UPCALL, &ep->com.history);
1026 	}
1027 }
1028 
1029 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1030 {
1031 	struct iw_cm_event event;
1032 
1033 	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1034 	memset(&event, 0, sizeof(event));
1035 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1036 	event.status = status;
1037 	memcpy(&event.local_addr, &ep->com.local_addr,
1038 	       sizeof(ep->com.local_addr));
1039 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1040 	       sizeof(ep->com.remote_addr));
1041 
1042 	if ((status == 0) || (status == -ECONNREFUSED)) {
1043 		if (!ep->tried_with_mpa_v1) {
1044 			/* this means MPA_v2 is used */
1045 			event.private_data_len = ep->plen -
1046 				sizeof(struct mpa_v2_conn_params);
1047 			event.private_data = ep->mpa_pkt +
1048 				sizeof(struct mpa_message) +
1049 				sizeof(struct mpa_v2_conn_params);
1050 		} else {
1051 			/* this means MPA_v1 is used */
1052 			event.private_data_len = ep->plen;
1053 			event.private_data = ep->mpa_pkt +
1054 				sizeof(struct mpa_message);
1055 		}
1056 	}
1057 
1058 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1059 	     ep->hwtid, status);
1060 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1061 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1062 
1063 	if (status < 0) {
1064 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1065 		ep->com.cm_id = NULL;
1066 	}
1067 }
1068 
1069 static void connect_request_upcall(struct c4iw_ep *ep)
1070 {
1071 	struct iw_cm_event event;
1072 
1073 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1074 	memset(&event, 0, sizeof(event));
1075 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1076 	memcpy(&event.local_addr, &ep->com.local_addr,
1077 	       sizeof(ep->com.local_addr));
1078 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1079 	       sizeof(ep->com.remote_addr));
1080 	event.provider_data = ep;
1081 	if (!ep->tried_with_mpa_v1) {
1082 		/* this means MPA_v2 is used */
1083 		event.ord = ep->ord;
1084 		event.ird = ep->ird;
1085 		event.private_data_len = ep->plen -
1086 			sizeof(struct mpa_v2_conn_params);
1087 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1088 			sizeof(struct mpa_v2_conn_params);
1089 	} else {
1090 		/* this means MPA_v1 is used. Send max supported */
1091 		event.ord = c4iw_max_read_depth;
1092 		event.ird = c4iw_max_read_depth;
1093 		event.private_data_len = ep->plen;
1094 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1095 	}
1096 	if (state_read(&ep->parent_ep->com) != DEAD) {
1097 		c4iw_get_ep(&ep->com);
1098 		ep->parent_ep->com.cm_id->event_handler(
1099 						ep->parent_ep->com.cm_id,
1100 						&event);
1101 	}
1102 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1103 	c4iw_put_ep(&ep->parent_ep->com);
1104 	ep->parent_ep = NULL;
1105 }
1106 
1107 static void established_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_ESTABLISHED;
1114 	event.ird = ep->ird;
1115 	event.ord = ep->ord;
1116 	if (ep->com.cm_id) {
1117 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1118 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1119 		set_bit(ESTAB_UPCALL, &ep->com.history);
1120 	}
1121 }
1122 
1123 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1124 {
1125 	struct cpl_rx_data_ack *req;
1126 	struct sk_buff *skb;
1127 	int wrlen = roundup(sizeof *req, 16);
1128 
1129 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1130 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1131 	if (!skb) {
1132 		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1133 		return 0;
1134 	}
1135 
1136 	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1137 	memset(req, 0, wrlen);
1138 	INIT_TP_WR(req, ep->hwtid);
1139 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1140 						    ep->hwtid));
1141 	req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1142 				       F_RX_DACK_CHANGE |
1143 				       V_RX_DACK_MODE(dack_mode));
1144 	set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1145 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1146 	return credits;
1147 }
1148 
1149 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1150 {
1151 	struct mpa_message *mpa;
1152 	struct mpa_v2_conn_params *mpa_v2_params;
1153 	u16 plen;
1154 	u16 resp_ird, resp_ord;
1155 	u8 rtr_mismatch = 0, insuff_ird = 0;
1156 	struct c4iw_qp_attributes attrs;
1157 	enum c4iw_qp_attr_mask mask;
1158 	int err;
1159 
1160 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1161 
1162 	/*
1163 	 * Stop mpa timer.  If it expired, then the state has
1164 	 * changed and we bail since ep_timeout already aborted
1165 	 * the connection.
1166 	 */
1167 	stop_ep_timer(ep);
1168 	if (state_read(&ep->com) != MPA_REQ_SENT)
1169 		return;
1170 
1171 	/*
1172 	 * If we get more than the supported amount of private data
1173 	 * then we must fail this connection.
1174 	 */
1175 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1176 		err = -EINVAL;
1177 		goto err;
1178 	}
1179 
1180 	/*
1181 	 * copy the new data into our accumulation buffer.
1182 	 */
1183 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1184 				  skb->len);
1185 	ep->mpa_pkt_len += skb->len;
1186 
1187 	/*
1188 	 * if we don't even have the mpa message, then bail.
1189 	 */
1190 	if (ep->mpa_pkt_len < sizeof(*mpa))
1191 		return;
1192 	mpa = (struct mpa_message *) ep->mpa_pkt;
1193 
1194 	/* Validate MPA header. */
1195 	if (mpa->revision > mpa_rev) {
1196 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1197 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1198 		err = -EPROTO;
1199 		goto err;
1200 	}
1201 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1202 		err = -EPROTO;
1203 		goto err;
1204 	}
1205 
1206 	plen = ntohs(mpa->private_data_size);
1207 
1208 	/*
1209 	 * Fail if there's too much private data.
1210 	 */
1211 	if (plen > MPA_MAX_PRIVATE_DATA) {
1212 		err = -EPROTO;
1213 		goto err;
1214 	}
1215 
1216 	/*
1217 	 * If plen does not account for pkt size
1218 	 */
1219 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1220 		err = -EPROTO;
1221 		goto err;
1222 	}
1223 
1224 	ep->plen = (u8) plen;
1225 
1226 	/*
1227 	 * If we don't have all the pdata yet, then bail.
1228 	 * We'll continue process when more data arrives.
1229 	 */
1230 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1231 		return;
1232 
1233 	if (mpa->flags & MPA_REJECT) {
1234 		err = -ECONNREFUSED;
1235 		goto err;
1236 	}
1237 
1238 	/*
1239 	 * If we get here we have accumulated the entire mpa
1240 	 * start reply message including private data. And
1241 	 * the MPA header is valid.
1242 	 */
1243 	state_set(&ep->com, FPDU_MODE);
1244 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1245 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1246 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1247 	ep->mpa_attr.version = mpa->revision;
1248 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1249 
1250 	if (mpa->revision == 2) {
1251 		ep->mpa_attr.enhanced_rdma_conn =
1252 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1253 		if (ep->mpa_attr.enhanced_rdma_conn) {
1254 			mpa_v2_params = (struct mpa_v2_conn_params *)
1255 				(ep->mpa_pkt + sizeof(*mpa));
1256 			resp_ird = ntohs(mpa_v2_params->ird) &
1257 				MPA_V2_IRD_ORD_MASK;
1258 			resp_ord = ntohs(mpa_v2_params->ord) &
1259 				MPA_V2_IRD_ORD_MASK;
1260 
1261 			/*
1262 			 * This is a double-check. Ideally, below checks are
1263 			 * not required since ird/ord stuff has been taken
1264 			 * care of in c4iw_accept_cr
1265 			 */
1266 			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1267 				err = -ENOMEM;
1268 				ep->ird = resp_ord;
1269 				ep->ord = resp_ird;
1270 				insuff_ird = 1;
1271 			}
1272 
1273 			if (ntohs(mpa_v2_params->ird) &
1274 					MPA_V2_PEER2PEER_MODEL) {
1275 				if (ntohs(mpa_v2_params->ord) &
1276 						MPA_V2_RDMA_WRITE_RTR)
1277 					ep->mpa_attr.p2p_type =
1278 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1279 				else if (ntohs(mpa_v2_params->ord) &
1280 						MPA_V2_RDMA_READ_RTR)
1281 					ep->mpa_attr.p2p_type =
1282 						FW_RI_INIT_P2PTYPE_READ_REQ;
1283 			}
1284 		}
1285 	} else if (mpa->revision == 1)
1286 		if (peer2peer)
1287 			ep->mpa_attr.p2p_type = p2p_type;
1288 
1289 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1290 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1291 	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
1292 	     ep->mpa_attr.recv_marker_enabled,
1293 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1294 	     ep->mpa_attr.p2p_type, p2p_type);
1295 
1296 	/*
1297 	 * If responder's RTR does not match with that of initiator, assign
1298 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1299 	 * generated when moving QP to RTS state.
1300 	 * A TERM message will be sent after QP has moved to RTS state
1301 	 */
1302 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1303 			(ep->mpa_attr.p2p_type != p2p_type)) {
1304 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1305 		rtr_mismatch = 1;
1306 	}
1307 
1308 	attrs.mpa_attr = ep->mpa_attr;
1309 	attrs.max_ird = ep->ird;
1310 	attrs.max_ord = ep->ord;
1311 	attrs.llp_stream_handle = ep;
1312 	attrs.next_state = C4IW_QP_STATE_RTS;
1313 
1314 	mask = C4IW_QP_ATTR_NEXT_STATE |
1315 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1316 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1317 
1318 	/* bind QP and TID with INIT_WR */
1319 	err = c4iw_modify_qp(ep->com.qp->rhp,
1320 			     ep->com.qp, mask, &attrs, 1);
1321 	if (err)
1322 		goto err;
1323 
1324 	/*
1325 	 * If responder's RTR requirement did not match with what initiator
1326 	 * supports, generate TERM message
1327 	 */
1328 	if (rtr_mismatch) {
1329 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1330 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1331 		attrs.ecode = MPA_NOMATCH_RTR;
1332 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1333 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1334 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1335 		err = -ENOMEM;
1336 		goto out;
1337 	}
1338 
1339 	/*
1340 	 * Generate TERM if initiator IRD is not sufficient for responder
1341 	 * provided ORD. Currently, we do the same behaviour even when
1342 	 * responder provided IRD is also not sufficient as regards to
1343 	 * initiator ORD.
1344 	 */
1345 	if (insuff_ird) {
1346 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1347 				__func__);
1348 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1349 		attrs.ecode = MPA_INSUFF_IRD;
1350 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1351 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1352 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1353 		err = -ENOMEM;
1354 		goto out;
1355 	}
1356 	goto out;
1357 err:
1358 	state_set(&ep->com, ABORTING);
1359 	send_abort(ep, skb, GFP_KERNEL);
1360 out:
1361 	connect_reply_upcall(ep, err);
1362 	return;
1363 }
1364 
1365 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1366 {
1367 	struct mpa_message *mpa;
1368 	struct mpa_v2_conn_params *mpa_v2_params;
1369 	u16 plen;
1370 
1371 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1372 
1373 	if (state_read(&ep->com) != MPA_REQ_WAIT)
1374 		return;
1375 
1376 	/*
1377 	 * If we get more than the supported amount of private data
1378 	 * then we must fail this connection.
1379 	 */
1380 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1381 		stop_ep_timer(ep);
1382 		abort_connection(ep, skb, GFP_KERNEL);
1383 		return;
1384 	}
1385 
1386 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1387 
1388 	/*
1389 	 * Copy the new data into our accumulation buffer.
1390 	 */
1391 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1392 				  skb->len);
1393 	ep->mpa_pkt_len += skb->len;
1394 
1395 	/*
1396 	 * If we don't even have the mpa message, then bail.
1397 	 * We'll continue process when more data arrives.
1398 	 */
1399 	if (ep->mpa_pkt_len < sizeof(*mpa))
1400 		return;
1401 
1402 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1403 	stop_ep_timer(ep);
1404 	mpa = (struct mpa_message *) ep->mpa_pkt;
1405 
1406 	/*
1407 	 * Validate MPA Header.
1408 	 */
1409 	if (mpa->revision > mpa_rev) {
1410 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1411 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1412 		stop_ep_timer(ep);
1413 		abort_connection(ep, skb, GFP_KERNEL);
1414 		return;
1415 	}
1416 
1417 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1418 		stop_ep_timer(ep);
1419 		abort_connection(ep, skb, GFP_KERNEL);
1420 		return;
1421 	}
1422 
1423 	plen = ntohs(mpa->private_data_size);
1424 
1425 	/*
1426 	 * Fail if there's too much private data.
1427 	 */
1428 	if (plen > MPA_MAX_PRIVATE_DATA) {
1429 		stop_ep_timer(ep);
1430 		abort_connection(ep, skb, GFP_KERNEL);
1431 		return;
1432 	}
1433 
1434 	/*
1435 	 * If plen does not account for pkt size
1436 	 */
1437 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1438 		stop_ep_timer(ep);
1439 		abort_connection(ep, skb, GFP_KERNEL);
1440 		return;
1441 	}
1442 	ep->plen = (u8) plen;
1443 
1444 	/*
1445 	 * If we don't have all the pdata yet, then bail.
1446 	 */
1447 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1448 		return;
1449 
1450 	/*
1451 	 * If we get here we have accumulated the entire mpa
1452 	 * start reply message including private data.
1453 	 */
1454 	ep->mpa_attr.initiator = 0;
1455 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1456 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1457 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1458 	ep->mpa_attr.version = mpa->revision;
1459 	if (mpa->revision == 1)
1460 		ep->tried_with_mpa_v1 = 1;
1461 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1462 
1463 	if (mpa->revision == 2) {
1464 		ep->mpa_attr.enhanced_rdma_conn =
1465 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1466 		if (ep->mpa_attr.enhanced_rdma_conn) {
1467 			mpa_v2_params = (struct mpa_v2_conn_params *)
1468 				(ep->mpa_pkt + sizeof(*mpa));
1469 			ep->ird = ntohs(mpa_v2_params->ird) &
1470 				MPA_V2_IRD_ORD_MASK;
1471 			ep->ord = ntohs(mpa_v2_params->ord) &
1472 				MPA_V2_IRD_ORD_MASK;
1473 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1474 				if (peer2peer) {
1475 					if (ntohs(mpa_v2_params->ord) &
1476 							MPA_V2_RDMA_WRITE_RTR)
1477 						ep->mpa_attr.p2p_type =
1478 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1479 					else if (ntohs(mpa_v2_params->ord) &
1480 							MPA_V2_RDMA_READ_RTR)
1481 						ep->mpa_attr.p2p_type =
1482 						FW_RI_INIT_P2PTYPE_READ_REQ;
1483 				}
1484 		}
1485 	} else if (mpa->revision == 1)
1486 		if (peer2peer)
1487 			ep->mpa_attr.p2p_type = p2p_type;
1488 
1489 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1490 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1491 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1492 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1493 	     ep->mpa_attr.p2p_type);
1494 
1495 	state_set(&ep->com, MPA_REQ_RCVD);
1496 
1497 	/* drive upcall */
1498 	connect_request_upcall(ep);
1499 	return;
1500 }
1501 
1502 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1503 {
1504 	struct c4iw_ep *ep;
1505 	struct cpl_rx_data *hdr = cplhdr(skb);
1506 	unsigned int dlen = ntohs(hdr->len);
1507 	unsigned int tid = GET_TID(hdr);
1508 	struct tid_info *t = dev->rdev.lldi.tids;
1509 	__u8 status = hdr->status;
1510 
1511 	ep = lookup_tid(t, tid);
1512 	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1513 	skb_pull(skb, sizeof(*hdr));
1514 	skb_trim(skb, dlen);
1515 
1516 	/* update RX credits */
1517 	update_rx_credits(ep, dlen);
1518 
1519 	switch (state_read(&ep->com)) {
1520 	case MPA_REQ_SENT:
1521 		ep->rcv_seq += dlen;
1522 		process_mpa_reply(ep, skb);
1523 		break;
1524 	case MPA_REQ_WAIT:
1525 		ep->rcv_seq += dlen;
1526 		process_mpa_request(ep, skb);
1527 		break;
1528 	case FPDU_MODE: {
1529 		struct c4iw_qp_attributes attrs;
1530 		BUG_ON(!ep->com.qp);
1531 		if (status)
1532 			pr_err("%s Unexpected streaming data." \
1533 			       " qpid %u ep %p state %d tid %u status %d\n",
1534 			       __func__, ep->com.qp->wq.sq.qid, ep,
1535 			       state_read(&ep->com), ep->hwtid, status);
1536 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1537 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1538 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1539 		break;
1540 	}
1541 	default:
1542 		break;
1543 	}
1544 	return 0;
1545 }
1546 
1547 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1548 {
1549 	struct c4iw_ep *ep;
1550 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1551 	int release = 0;
1552 	unsigned int tid = GET_TID(rpl);
1553 	struct tid_info *t = dev->rdev.lldi.tids;
1554 
1555 	ep = lookup_tid(t, tid);
1556 	if (!ep) {
1557 		printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1558 		return 0;
1559 	}
1560 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1561 	mutex_lock(&ep->com.mutex);
1562 	switch (ep->com.state) {
1563 	case ABORTING:
1564 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1565 		__state_set(&ep->com, DEAD);
1566 		release = 1;
1567 		break;
1568 	default:
1569 		printk(KERN_ERR "%s ep %p state %d\n",
1570 		     __func__, ep, ep->com.state);
1571 		break;
1572 	}
1573 	mutex_unlock(&ep->com.mutex);
1574 
1575 	if (release)
1576 		release_ep_resources(ep);
1577 	return 0;
1578 }
1579 
1580 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1581 {
1582 	struct sk_buff *skb;
1583 	struct fw_ofld_connection_wr *req;
1584 	unsigned int mtu_idx;
1585 	int wscale;
1586 	struct sockaddr_in *sin;
1587 
1588 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1589 	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1590 	memset(req, 0, sizeof(*req));
1591 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1592 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1593 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1594 				     ep->com.dev->rdev.lldi.ports[0],
1595 				     ep->l2t));
1596 	sin = (struct sockaddr_in *)&ep->com.local_addr;
1597 	req->le.lport = sin->sin_port;
1598 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1599 	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1600 	req->le.pport = sin->sin_port;
1601 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1602 	req->tcb.t_state_to_astid =
1603 			htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1604 			V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1605 	req->tcb.cplrxdataack_cplpassacceptrpl =
1606 			htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1607 	req->tcb.tx_max = (__force __be32) jiffies;
1608 	req->tcb.rcv_adv = htons(1);
1609 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1610 	wscale = compute_wscale(rcv_win);
1611 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1612 		(nocong ? NO_CONG(1) : 0) |
1613 		KEEP_ALIVE(1) |
1614 		DELACK(1) |
1615 		WND_SCALE(wscale) |
1616 		MSS_IDX(mtu_idx) |
1617 		L2T_IDX(ep->l2t->idx) |
1618 		TX_CHAN(ep->tx_chan) |
1619 		SMAC_SEL(ep->smac_idx) |
1620 		DSCP(ep->tos) |
1621 		ULP_MODE(ULP_MODE_TCPDDP) |
1622 		RCV_BUFSIZ(rcv_win >> 10));
1623 	req->tcb.opt2 = (__force __be32) (PACE(1) |
1624 		TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1625 		RX_CHANNEL(0) |
1626 		CCTRL_ECN(enable_ecn) |
1627 		RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1628 	if (enable_tcp_timestamps)
1629 		req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1630 	if (enable_tcp_sack)
1631 		req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1632 	if (wscale && enable_tcp_window_scaling)
1633 		req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1634 	req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1635 	req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1636 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1637 	set_bit(ACT_OFLD_CONN, &ep->com.history);
1638 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1639 }
1640 
1641 /*
1642  * Return whether a failed active open has allocated a TID
1643  */
1644 static inline int act_open_has_tid(int status)
1645 {
1646 	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1647 	       status != CPL_ERR_ARP_MISS;
1648 }
1649 
1650 #define ACT_OPEN_RETRY_COUNT 2
1651 
1652 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1653 		     struct dst_entry *dst, struct c4iw_dev *cdev,
1654 		     bool clear_mpa_v1)
1655 {
1656 	struct neighbour *n;
1657 	int err, step;
1658 	struct net_device *pdev;
1659 
1660 	n = dst_neigh_lookup(dst, peer_ip);
1661 	if (!n)
1662 		return -ENODEV;
1663 
1664 	rcu_read_lock();
1665 	err = -ENOMEM;
1666 	if (n->dev->flags & IFF_LOOPBACK) {
1667 		if (iptype == 4)
1668 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1669 		else if (IS_ENABLED(CONFIG_IPV6))
1670 			for_each_netdev(&init_net, pdev) {
1671 				if (ipv6_chk_addr(&init_net,
1672 						  (struct in6_addr *)peer_ip,
1673 						  pdev, 1))
1674 					break;
1675 			}
1676 		else
1677 			pdev = NULL;
1678 
1679 		if (!pdev) {
1680 			err = -ENODEV;
1681 			goto out;
1682 		}
1683 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1684 					n, pdev, 0);
1685 		if (!ep->l2t)
1686 			goto out;
1687 		ep->mtu = pdev->mtu;
1688 		ep->tx_chan = cxgb4_port_chan(pdev);
1689 		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1690 		step = cdev->rdev.lldi.ntxq /
1691 			cdev->rdev.lldi.nchan;
1692 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1693 		step = cdev->rdev.lldi.nrxq /
1694 			cdev->rdev.lldi.nchan;
1695 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1696 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1697 			cxgb4_port_idx(pdev) * step];
1698 		dev_put(pdev);
1699 	} else {
1700 		pdev = get_real_dev(n->dev);
1701 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1702 					n, pdev, 0);
1703 		if (!ep->l2t)
1704 			goto out;
1705 		ep->mtu = dst_mtu(dst);
1706 		ep->tx_chan = cxgb4_port_chan(n->dev);
1707 		ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1708 		step = cdev->rdev.lldi.ntxq /
1709 			cdev->rdev.lldi.nchan;
1710 		ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1711 		ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1712 		step = cdev->rdev.lldi.nrxq /
1713 			cdev->rdev.lldi.nchan;
1714 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1715 			cxgb4_port_idx(n->dev) * step];
1716 
1717 		if (clear_mpa_v1) {
1718 			ep->retry_with_mpa_v1 = 0;
1719 			ep->tried_with_mpa_v1 = 0;
1720 		}
1721 	}
1722 	err = 0;
1723 out:
1724 	rcu_read_unlock();
1725 
1726 	neigh_release(n);
1727 
1728 	return err;
1729 }
1730 
1731 static int c4iw_reconnect(struct c4iw_ep *ep)
1732 {
1733 	int err = 0;
1734 	struct sockaddr_in *laddr = (struct sockaddr_in *)
1735 				    &ep->com.cm_id->local_addr;
1736 	struct sockaddr_in *raddr = (struct sockaddr_in *)
1737 				    &ep->com.cm_id->remote_addr;
1738 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1739 				      &ep->com.cm_id->local_addr;
1740 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1741 				      &ep->com.cm_id->remote_addr;
1742 	int iptype;
1743 	__u8 *ra;
1744 
1745 	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1746 	init_timer(&ep->timer);
1747 
1748 	/*
1749 	 * Allocate an active TID to initiate a TCP connection.
1750 	 */
1751 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1752 	if (ep->atid == -1) {
1753 		pr_err("%s - cannot alloc atid.\n", __func__);
1754 		err = -ENOMEM;
1755 		goto fail2;
1756 	}
1757 	insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1758 
1759 	/* find a route */
1760 	if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1761 		ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1762 				     raddr->sin_addr.s_addr, laddr->sin_port,
1763 				     raddr->sin_port, 0);
1764 		iptype = 4;
1765 		ra = (__u8 *)&raddr->sin_addr;
1766 	} else {
1767 		ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1768 				      raddr6->sin6_addr.s6_addr,
1769 				      laddr6->sin6_port, raddr6->sin6_port, 0,
1770 				      raddr6->sin6_scope_id);
1771 		iptype = 6;
1772 		ra = (__u8 *)&raddr6->sin6_addr;
1773 	}
1774 	if (!ep->dst) {
1775 		pr_err("%s - cannot find route.\n", __func__);
1776 		err = -EHOSTUNREACH;
1777 		goto fail3;
1778 	}
1779 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1780 	if (err) {
1781 		pr_err("%s - cannot alloc l2e.\n", __func__);
1782 		goto fail4;
1783 	}
1784 
1785 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1786 	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1787 	     ep->l2t->idx);
1788 
1789 	state_set(&ep->com, CONNECTING);
1790 	ep->tos = 0;
1791 
1792 	/* send connect request to rnic */
1793 	err = send_connect(ep);
1794 	if (!err)
1795 		goto out;
1796 
1797 	cxgb4_l2t_release(ep->l2t);
1798 fail4:
1799 	dst_release(ep->dst);
1800 fail3:
1801 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1802 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1803 fail2:
1804 	/*
1805 	 * remember to send notification to upper layer.
1806 	 * We are in here so the upper layer is not aware that this is
1807 	 * re-connect attempt and so, upper layer is still waiting for
1808 	 * response of 1st connect request.
1809 	 */
1810 	connect_reply_upcall(ep, -ECONNRESET);
1811 	c4iw_put_ep(&ep->com);
1812 out:
1813 	return err;
1814 }
1815 
1816 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1817 {
1818 	struct c4iw_ep *ep;
1819 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
1820 	unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1821 					ntohl(rpl->atid_status)));
1822 	struct tid_info *t = dev->rdev.lldi.tids;
1823 	int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1824 	struct sockaddr_in *la;
1825 	struct sockaddr_in *ra;
1826 	struct sockaddr_in6 *la6;
1827 	struct sockaddr_in6 *ra6;
1828 
1829 	ep = lookup_atid(t, atid);
1830 	la = (struct sockaddr_in *)&ep->com.local_addr;
1831 	ra = (struct sockaddr_in *)&ep->com.remote_addr;
1832 	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1833 	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1834 
1835 	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1836 	     status, status2errno(status));
1837 
1838 	if (status == CPL_ERR_RTX_NEG_ADVICE) {
1839 		printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1840 			atid);
1841 		return 0;
1842 	}
1843 
1844 	set_bit(ACT_OPEN_RPL, &ep->com.history);
1845 
1846 	/*
1847 	 * Log interesting failures.
1848 	 */
1849 	switch (status) {
1850 	case CPL_ERR_CONN_RESET:
1851 	case CPL_ERR_CONN_TIMEDOUT:
1852 		break;
1853 	case CPL_ERR_TCAM_FULL:
1854 		mutex_lock(&dev->rdev.stats.lock);
1855 		dev->rdev.stats.tcam_full++;
1856 		mutex_unlock(&dev->rdev.stats.lock);
1857 		if (ep->com.local_addr.ss_family == AF_INET &&
1858 		    dev->rdev.lldi.enable_fw_ofld_conn) {
1859 			send_fw_act_open_req(ep,
1860 					     GET_TID_TID(GET_AOPEN_ATID(
1861 					     ntohl(rpl->atid_status))));
1862 			return 0;
1863 		}
1864 		break;
1865 	case CPL_ERR_CONN_EXIST:
1866 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1867 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
1868 			remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1869 					atid);
1870 			cxgb4_free_atid(t, atid);
1871 			dst_release(ep->dst);
1872 			cxgb4_l2t_release(ep->l2t);
1873 			c4iw_reconnect(ep);
1874 			return 0;
1875 		}
1876 		break;
1877 	default:
1878 		if (ep->com.local_addr.ss_family == AF_INET) {
1879 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1880 				atid, status, status2errno(status),
1881 				&la->sin_addr.s_addr, ntohs(la->sin_port),
1882 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
1883 		} else {
1884 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1885 				atid, status, status2errno(status),
1886 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1887 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1888 		}
1889 		break;
1890 	}
1891 
1892 	connect_reply_upcall(ep, status2errno(status));
1893 	state_set(&ep->com, DEAD);
1894 
1895 	if (status && act_open_has_tid(status))
1896 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1897 
1898 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1899 	cxgb4_free_atid(t, atid);
1900 	dst_release(ep->dst);
1901 	cxgb4_l2t_release(ep->l2t);
1902 	c4iw_put_ep(&ep->com);
1903 
1904 	return 0;
1905 }
1906 
1907 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1908 {
1909 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1910 	struct tid_info *t = dev->rdev.lldi.tids;
1911 	unsigned int stid = GET_TID(rpl);
1912 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1913 
1914 	if (!ep) {
1915 		PDBG("%s stid %d lookup failure!\n", __func__, stid);
1916 		goto out;
1917 	}
1918 	PDBG("%s ep %p status %d error %d\n", __func__, ep,
1919 	     rpl->status, status2errno(rpl->status));
1920 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1921 
1922 out:
1923 	return 0;
1924 }
1925 
1926 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1927 {
1928 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1929 	struct tid_info *t = dev->rdev.lldi.tids;
1930 	unsigned int stid = GET_TID(rpl);
1931 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1932 
1933 	PDBG("%s ep %p\n", __func__, ep);
1934 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1935 	return 0;
1936 }
1937 
1938 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1939 		      struct cpl_pass_accept_req *req)
1940 {
1941 	struct cpl_pass_accept_rpl *rpl;
1942 	unsigned int mtu_idx;
1943 	u64 opt0;
1944 	u32 opt2;
1945 	int wscale;
1946 
1947 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1948 	BUG_ON(skb_cloned(skb));
1949 	skb_trim(skb, sizeof(*rpl));
1950 	skb_get(skb);
1951 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1952 	wscale = compute_wscale(rcv_win);
1953 	opt0 = (nocong ? NO_CONG(1) : 0) |
1954 	       KEEP_ALIVE(1) |
1955 	       DELACK(1) |
1956 	       WND_SCALE(wscale) |
1957 	       MSS_IDX(mtu_idx) |
1958 	       L2T_IDX(ep->l2t->idx) |
1959 	       TX_CHAN(ep->tx_chan) |
1960 	       SMAC_SEL(ep->smac_idx) |
1961 	       DSCP(ep->tos >> 2) |
1962 	       ULP_MODE(ULP_MODE_TCPDDP) |
1963 	       RCV_BUFSIZ(rcv_win>>10);
1964 	opt2 = RX_CHANNEL(0) |
1965 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1966 
1967 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
1968 		opt2 |= TSTAMPS_EN(1);
1969 	if (enable_tcp_sack && req->tcpopt.sack)
1970 		opt2 |= SACK_EN(1);
1971 	if (wscale && enable_tcp_window_scaling)
1972 		opt2 |= WND_SCALE_EN(1);
1973 	if (enable_ecn) {
1974 		const struct tcphdr *tcph;
1975 		u32 hlen = ntohl(req->hdr_len);
1976 
1977 		tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
1978 			G_IP_HDR_LEN(hlen);
1979 		if (tcph->ece && tcph->cwr)
1980 			opt2 |= CCTRL_ECN(1);
1981 	}
1982 
1983 	rpl = cplhdr(skb);
1984 	INIT_TP_WR(rpl, ep->hwtid);
1985 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1986 				      ep->hwtid));
1987 	rpl->opt0 = cpu_to_be64(opt0);
1988 	rpl->opt2 = cpu_to_be32(opt2);
1989 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
1990 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1991 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1992 
1993 	return;
1994 }
1995 
1996 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
1997 {
1998 	PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
1999 	BUG_ON(skb_cloned(skb));
2000 	skb_trim(skb, sizeof(struct cpl_tid_release));
2001 	skb_get(skb);
2002 	release_tid(&dev->rdev, hwtid, skb);
2003 	return;
2004 }
2005 
2006 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2007 		       __u8 *local_ip, __u8 *peer_ip,
2008 		       __be16 *local_port, __be16 *peer_port)
2009 {
2010 	int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2011 	int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2012 	struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2013 	struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2014 	struct tcphdr *tcp = (struct tcphdr *)
2015 			     ((u8 *)(req + 1) + eth_len + ip_len);
2016 
2017 	if (ip->version == 4) {
2018 		PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2019 		     ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2020 		     ntohs(tcp->dest));
2021 		*iptype = 4;
2022 		memcpy(peer_ip, &ip->saddr, 4);
2023 		memcpy(local_ip, &ip->daddr, 4);
2024 	} else {
2025 		PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2026 		     ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2027 		     ntohs(tcp->dest));
2028 		*iptype = 6;
2029 		memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2030 		memcpy(local_ip, ip6->daddr.s6_addr, 16);
2031 	}
2032 	*peer_port = tcp->source;
2033 	*local_port = tcp->dest;
2034 
2035 	return;
2036 }
2037 
2038 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2039 {
2040 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2041 	struct cpl_pass_accept_req *req = cplhdr(skb);
2042 	unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2043 	struct tid_info *t = dev->rdev.lldi.tids;
2044 	unsigned int hwtid = GET_TID(req);
2045 	struct dst_entry *dst;
2046 	__u8 local_ip[16], peer_ip[16];
2047 	__be16 local_port, peer_port;
2048 	int err;
2049 	u16 peer_mss = ntohs(req->tcpopt.mss);
2050 	int iptype;
2051 
2052 	parent_ep = lookup_stid(t, stid);
2053 	if (!parent_ep) {
2054 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2055 		goto reject;
2056 	}
2057 
2058 	if (state_read(&parent_ep->com) != LISTEN) {
2059 		printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2060 		       __func__);
2061 		goto reject;
2062 	}
2063 
2064 	get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2065 
2066 	/* Find output route */
2067 	if (iptype == 4)  {
2068 		PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2069 		     , __func__, parent_ep, hwtid,
2070 		     local_ip, peer_ip, ntohs(local_port),
2071 		     ntohs(peer_port), peer_mss);
2072 		dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2073 				 local_port, peer_port,
2074 				 GET_POPEN_TOS(ntohl(req->tos_stid)));
2075 	} else {
2076 		PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2077 		     , __func__, parent_ep, hwtid,
2078 		     local_ip, peer_ip, ntohs(local_port),
2079 		     ntohs(peer_port), peer_mss);
2080 		dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2081 				  PASS_OPEN_TOS(ntohl(req->tos_stid)),
2082 				  ((struct sockaddr_in6 *)
2083 				  &parent_ep->com.local_addr)->sin6_scope_id);
2084 	}
2085 	if (!dst) {
2086 		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2087 		       __func__);
2088 		goto reject;
2089 	}
2090 
2091 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2092 	if (!child_ep) {
2093 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2094 		       __func__);
2095 		dst_release(dst);
2096 		goto reject;
2097 	}
2098 
2099 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2100 	if (err) {
2101 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2102 		       __func__);
2103 		dst_release(dst);
2104 		kfree(child_ep);
2105 		goto reject;
2106 	}
2107 
2108 	if (peer_mss && child_ep->mtu > (peer_mss + 40))
2109 		child_ep->mtu = peer_mss + 40;
2110 
2111 	state_set(&child_ep->com, CONNECTING);
2112 	child_ep->com.dev = dev;
2113 	child_ep->com.cm_id = NULL;
2114 	if (iptype == 4) {
2115 		struct sockaddr_in *sin = (struct sockaddr_in *)
2116 			&child_ep->com.local_addr;
2117 		sin->sin_family = PF_INET;
2118 		sin->sin_port = local_port;
2119 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2120 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2121 		sin->sin_family = PF_INET;
2122 		sin->sin_port = peer_port;
2123 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2124 	} else {
2125 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2126 			&child_ep->com.local_addr;
2127 		sin6->sin6_family = PF_INET6;
2128 		sin6->sin6_port = local_port;
2129 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2130 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2131 		sin6->sin6_family = PF_INET6;
2132 		sin6->sin6_port = peer_port;
2133 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2134 	}
2135 	c4iw_get_ep(&parent_ep->com);
2136 	child_ep->parent_ep = parent_ep;
2137 	child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2138 	child_ep->dst = dst;
2139 	child_ep->hwtid = hwtid;
2140 
2141 	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2142 	     child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2143 
2144 	init_timer(&child_ep->timer);
2145 	cxgb4_insert_tid(t, child_ep, hwtid);
2146 	insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2147 	accept_cr(child_ep, skb, req);
2148 	set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2149 	goto out;
2150 reject:
2151 	reject_cr(dev, hwtid, skb);
2152 out:
2153 	return 0;
2154 }
2155 
2156 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2157 {
2158 	struct c4iw_ep *ep;
2159 	struct cpl_pass_establish *req = cplhdr(skb);
2160 	struct tid_info *t = dev->rdev.lldi.tids;
2161 	unsigned int tid = GET_TID(req);
2162 
2163 	ep = lookup_tid(t, tid);
2164 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2165 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2166 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2167 
2168 	PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2169 	     ntohs(req->tcp_opt));
2170 
2171 	set_emss(ep, ntohs(req->tcp_opt));
2172 
2173 	dst_confirm(ep->dst);
2174 	state_set(&ep->com, MPA_REQ_WAIT);
2175 	start_ep_timer(ep);
2176 	send_flowc(ep, skb);
2177 	set_bit(PASS_ESTAB, &ep->com.history);
2178 
2179 	return 0;
2180 }
2181 
2182 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2183 {
2184 	struct cpl_peer_close *hdr = cplhdr(skb);
2185 	struct c4iw_ep *ep;
2186 	struct c4iw_qp_attributes attrs;
2187 	int disconnect = 1;
2188 	int release = 0;
2189 	struct tid_info *t = dev->rdev.lldi.tids;
2190 	unsigned int tid = GET_TID(hdr);
2191 	int ret;
2192 
2193 	ep = lookup_tid(t, tid);
2194 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2195 	dst_confirm(ep->dst);
2196 
2197 	set_bit(PEER_CLOSE, &ep->com.history);
2198 	mutex_lock(&ep->com.mutex);
2199 	switch (ep->com.state) {
2200 	case MPA_REQ_WAIT:
2201 		__state_set(&ep->com, CLOSING);
2202 		break;
2203 	case MPA_REQ_SENT:
2204 		__state_set(&ep->com, CLOSING);
2205 		connect_reply_upcall(ep, -ECONNRESET);
2206 		break;
2207 	case MPA_REQ_RCVD:
2208 
2209 		/*
2210 		 * We're gonna mark this puppy DEAD, but keep
2211 		 * the reference on it until the ULP accepts or
2212 		 * rejects the CR. Also wake up anyone waiting
2213 		 * in rdma connection migration (see c4iw_accept_cr()).
2214 		 */
2215 		__state_set(&ep->com, CLOSING);
2216 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2217 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2218 		break;
2219 	case MPA_REP_SENT:
2220 		__state_set(&ep->com, CLOSING);
2221 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2222 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2223 		break;
2224 	case FPDU_MODE:
2225 		start_ep_timer(ep);
2226 		__state_set(&ep->com, CLOSING);
2227 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2228 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2229 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2230 		if (ret != -ECONNRESET) {
2231 			peer_close_upcall(ep);
2232 			disconnect = 1;
2233 		}
2234 		break;
2235 	case ABORTING:
2236 		disconnect = 0;
2237 		break;
2238 	case CLOSING:
2239 		__state_set(&ep->com, MORIBUND);
2240 		disconnect = 0;
2241 		break;
2242 	case MORIBUND:
2243 		stop_ep_timer(ep);
2244 		if (ep->com.cm_id && ep->com.qp) {
2245 			attrs.next_state = C4IW_QP_STATE_IDLE;
2246 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2247 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2248 		}
2249 		close_complete_upcall(ep);
2250 		__state_set(&ep->com, DEAD);
2251 		release = 1;
2252 		disconnect = 0;
2253 		break;
2254 	case DEAD:
2255 		disconnect = 0;
2256 		break;
2257 	default:
2258 		BUG_ON(1);
2259 	}
2260 	mutex_unlock(&ep->com.mutex);
2261 	if (disconnect)
2262 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2263 	if (release)
2264 		release_ep_resources(ep);
2265 	return 0;
2266 }
2267 
2268 /*
2269  * Returns whether an ABORT_REQ_RSS message is a negative advice.
2270  */
2271 static int is_neg_adv_abort(unsigned int status)
2272 {
2273 	return status == CPL_ERR_RTX_NEG_ADVICE ||
2274 	       status == CPL_ERR_PERSIST_NEG_ADVICE;
2275 }
2276 
2277 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2278 {
2279 	struct cpl_abort_req_rss *req = cplhdr(skb);
2280 	struct c4iw_ep *ep;
2281 	struct cpl_abort_rpl *rpl;
2282 	struct sk_buff *rpl_skb;
2283 	struct c4iw_qp_attributes attrs;
2284 	int ret;
2285 	int release = 0;
2286 	struct tid_info *t = dev->rdev.lldi.tids;
2287 	unsigned int tid = GET_TID(req);
2288 
2289 	ep = lookup_tid(t, tid);
2290 	if (is_neg_adv_abort(req->status)) {
2291 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2292 		     ep->hwtid);
2293 		return 0;
2294 	}
2295 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2296 	     ep->com.state);
2297 	set_bit(PEER_ABORT, &ep->com.history);
2298 
2299 	/*
2300 	 * Wake up any threads in rdma_init() or rdma_fini().
2301 	 * However, this is not needed if com state is just
2302 	 * MPA_REQ_SENT
2303 	 */
2304 	if (ep->com.state != MPA_REQ_SENT)
2305 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2306 
2307 	mutex_lock(&ep->com.mutex);
2308 	switch (ep->com.state) {
2309 	case CONNECTING:
2310 		break;
2311 	case MPA_REQ_WAIT:
2312 		stop_ep_timer(ep);
2313 		break;
2314 	case MPA_REQ_SENT:
2315 		stop_ep_timer(ep);
2316 		if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2317 			connect_reply_upcall(ep, -ECONNRESET);
2318 		else {
2319 			/*
2320 			 * we just don't send notification upwards because we
2321 			 * want to retry with mpa_v1 without upper layers even
2322 			 * knowing it.
2323 			 *
2324 			 * do some housekeeping so as to re-initiate the
2325 			 * connection
2326 			 */
2327 			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2328 			     mpa_rev);
2329 			ep->retry_with_mpa_v1 = 1;
2330 		}
2331 		break;
2332 	case MPA_REP_SENT:
2333 		break;
2334 	case MPA_REQ_RCVD:
2335 		break;
2336 	case MORIBUND:
2337 	case CLOSING:
2338 		stop_ep_timer(ep);
2339 		/*FALLTHROUGH*/
2340 	case FPDU_MODE:
2341 		if (ep->com.cm_id && ep->com.qp) {
2342 			attrs.next_state = C4IW_QP_STATE_ERROR;
2343 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2344 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2345 				     &attrs, 1);
2346 			if (ret)
2347 				printk(KERN_ERR MOD
2348 				       "%s - qp <- error failed!\n",
2349 				       __func__);
2350 		}
2351 		peer_abort_upcall(ep);
2352 		break;
2353 	case ABORTING:
2354 		break;
2355 	case DEAD:
2356 		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2357 		mutex_unlock(&ep->com.mutex);
2358 		return 0;
2359 	default:
2360 		BUG_ON(1);
2361 		break;
2362 	}
2363 	dst_confirm(ep->dst);
2364 	if (ep->com.state != ABORTING) {
2365 		__state_set(&ep->com, DEAD);
2366 		/* we don't release if we want to retry with mpa_v1 */
2367 		if (!ep->retry_with_mpa_v1)
2368 			release = 1;
2369 	}
2370 	mutex_unlock(&ep->com.mutex);
2371 
2372 	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2373 	if (!rpl_skb) {
2374 		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2375 		       __func__);
2376 		release = 1;
2377 		goto out;
2378 	}
2379 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2380 	rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2381 	INIT_TP_WR(rpl, ep->hwtid);
2382 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2383 	rpl->cmd = CPL_ABORT_NO_RST;
2384 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2385 out:
2386 	if (release)
2387 		release_ep_resources(ep);
2388 	else if (ep->retry_with_mpa_v1) {
2389 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2390 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2391 		dst_release(ep->dst);
2392 		cxgb4_l2t_release(ep->l2t);
2393 		c4iw_reconnect(ep);
2394 	}
2395 
2396 	return 0;
2397 }
2398 
2399 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2400 {
2401 	struct c4iw_ep *ep;
2402 	struct c4iw_qp_attributes attrs;
2403 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2404 	int release = 0;
2405 	struct tid_info *t = dev->rdev.lldi.tids;
2406 	unsigned int tid = GET_TID(rpl);
2407 
2408 	ep = lookup_tid(t, tid);
2409 
2410 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2411 	BUG_ON(!ep);
2412 
2413 	/* The cm_id may be null if we failed to connect */
2414 	mutex_lock(&ep->com.mutex);
2415 	switch (ep->com.state) {
2416 	case CLOSING:
2417 		__state_set(&ep->com, MORIBUND);
2418 		break;
2419 	case MORIBUND:
2420 		stop_ep_timer(ep);
2421 		if ((ep->com.cm_id) && (ep->com.qp)) {
2422 			attrs.next_state = C4IW_QP_STATE_IDLE;
2423 			c4iw_modify_qp(ep->com.qp->rhp,
2424 					     ep->com.qp,
2425 					     C4IW_QP_ATTR_NEXT_STATE,
2426 					     &attrs, 1);
2427 		}
2428 		close_complete_upcall(ep);
2429 		__state_set(&ep->com, DEAD);
2430 		release = 1;
2431 		break;
2432 	case ABORTING:
2433 	case DEAD:
2434 		break;
2435 	default:
2436 		BUG_ON(1);
2437 		break;
2438 	}
2439 	mutex_unlock(&ep->com.mutex);
2440 	if (release)
2441 		release_ep_resources(ep);
2442 	return 0;
2443 }
2444 
2445 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2446 {
2447 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
2448 	struct tid_info *t = dev->rdev.lldi.tids;
2449 	unsigned int tid = GET_TID(rpl);
2450 	struct c4iw_ep *ep;
2451 	struct c4iw_qp_attributes attrs;
2452 
2453 	ep = lookup_tid(t, tid);
2454 	BUG_ON(!ep);
2455 
2456 	if (ep && ep->com.qp) {
2457 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2458 		       ep->com.qp->wq.sq.qid);
2459 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2460 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2461 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2462 	} else
2463 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2464 
2465 	return 0;
2466 }
2467 
2468 /*
2469  * Upcall from the adapter indicating data has been transmitted.
2470  * For us its just the single MPA request or reply.  We can now free
2471  * the skb holding the mpa message.
2472  */
2473 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2474 {
2475 	struct c4iw_ep *ep;
2476 	struct cpl_fw4_ack *hdr = cplhdr(skb);
2477 	u8 credits = hdr->credits;
2478 	unsigned int tid = GET_TID(hdr);
2479 	struct tid_info *t = dev->rdev.lldi.tids;
2480 
2481 
2482 	ep = lookup_tid(t, tid);
2483 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2484 	if (credits == 0) {
2485 		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2486 		     __func__, ep, ep->hwtid, state_read(&ep->com));
2487 		return 0;
2488 	}
2489 
2490 	dst_confirm(ep->dst);
2491 	if (ep->mpa_skb) {
2492 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
2493 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2494 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2495 		kfree_skb(ep->mpa_skb);
2496 		ep->mpa_skb = NULL;
2497 	}
2498 	return 0;
2499 }
2500 
2501 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2502 {
2503 	int err;
2504 	struct c4iw_ep *ep = to_ep(cm_id);
2505 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2506 
2507 	if (state_read(&ep->com) == DEAD) {
2508 		c4iw_put_ep(&ep->com);
2509 		return -ECONNRESET;
2510 	}
2511 	set_bit(ULP_REJECT, &ep->com.history);
2512 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2513 	if (mpa_rev == 0)
2514 		abort_connection(ep, NULL, GFP_KERNEL);
2515 	else {
2516 		err = send_mpa_reject(ep, pdata, pdata_len);
2517 		err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2518 	}
2519 	c4iw_put_ep(&ep->com);
2520 	return 0;
2521 }
2522 
2523 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2524 {
2525 	int err;
2526 	struct c4iw_qp_attributes attrs;
2527 	enum c4iw_qp_attr_mask mask;
2528 	struct c4iw_ep *ep = to_ep(cm_id);
2529 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2530 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2531 
2532 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2533 	if (state_read(&ep->com) == DEAD) {
2534 		err = -ECONNRESET;
2535 		goto err;
2536 	}
2537 
2538 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2539 	BUG_ON(!qp);
2540 
2541 	set_bit(ULP_ACCEPT, &ep->com.history);
2542 	if ((conn_param->ord > c4iw_max_read_depth) ||
2543 	    (conn_param->ird > c4iw_max_read_depth)) {
2544 		abort_connection(ep, NULL, GFP_KERNEL);
2545 		err = -EINVAL;
2546 		goto err;
2547 	}
2548 
2549 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2550 		if (conn_param->ord > ep->ird) {
2551 			ep->ird = conn_param->ird;
2552 			ep->ord = conn_param->ord;
2553 			send_mpa_reject(ep, conn_param->private_data,
2554 					conn_param->private_data_len);
2555 			abort_connection(ep, NULL, GFP_KERNEL);
2556 			err = -ENOMEM;
2557 			goto err;
2558 		}
2559 		if (conn_param->ird > ep->ord) {
2560 			if (!ep->ord)
2561 				conn_param->ird = 1;
2562 			else {
2563 				abort_connection(ep, NULL, GFP_KERNEL);
2564 				err = -ENOMEM;
2565 				goto err;
2566 			}
2567 		}
2568 
2569 	}
2570 	ep->ird = conn_param->ird;
2571 	ep->ord = conn_param->ord;
2572 
2573 	if (ep->mpa_attr.version != 2)
2574 		if (peer2peer && ep->ird == 0)
2575 			ep->ird = 1;
2576 
2577 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2578 
2579 	cm_id->add_ref(cm_id);
2580 	ep->com.cm_id = cm_id;
2581 	ep->com.qp = qp;
2582 	ref_qp(ep);
2583 
2584 	/* bind QP to EP and move to RTS */
2585 	attrs.mpa_attr = ep->mpa_attr;
2586 	attrs.max_ird = ep->ird;
2587 	attrs.max_ord = ep->ord;
2588 	attrs.llp_stream_handle = ep;
2589 	attrs.next_state = C4IW_QP_STATE_RTS;
2590 
2591 	/* bind QP and TID with INIT_WR */
2592 	mask = C4IW_QP_ATTR_NEXT_STATE |
2593 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2594 			     C4IW_QP_ATTR_MPA_ATTR |
2595 			     C4IW_QP_ATTR_MAX_IRD |
2596 			     C4IW_QP_ATTR_MAX_ORD;
2597 
2598 	err = c4iw_modify_qp(ep->com.qp->rhp,
2599 			     ep->com.qp, mask, &attrs, 1);
2600 	if (err)
2601 		goto err1;
2602 	err = send_mpa_reply(ep, conn_param->private_data,
2603 			     conn_param->private_data_len);
2604 	if (err)
2605 		goto err1;
2606 
2607 	state_set(&ep->com, FPDU_MODE);
2608 	established_upcall(ep);
2609 	c4iw_put_ep(&ep->com);
2610 	return 0;
2611 err1:
2612 	ep->com.cm_id = NULL;
2613 	cm_id->rem_ref(cm_id);
2614 err:
2615 	c4iw_put_ep(&ep->com);
2616 	return err;
2617 }
2618 
2619 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2620 {
2621 	struct in_device *ind;
2622 	int found = 0;
2623 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2624 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2625 
2626 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
2627 	if (!ind)
2628 		return -EADDRNOTAVAIL;
2629 	for_primary_ifa(ind) {
2630 		laddr->sin_addr.s_addr = ifa->ifa_address;
2631 		raddr->sin_addr.s_addr = ifa->ifa_address;
2632 		found = 1;
2633 		break;
2634 	}
2635 	endfor_ifa(ind);
2636 	in_dev_put(ind);
2637 	return found ? 0 : -EADDRNOTAVAIL;
2638 }
2639 
2640 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2641 		      unsigned char banned_flags)
2642 {
2643 	struct inet6_dev *idev;
2644 	int err = -EADDRNOTAVAIL;
2645 
2646 	rcu_read_lock();
2647 	idev = __in6_dev_get(dev);
2648 	if (idev != NULL) {
2649 		struct inet6_ifaddr *ifp;
2650 
2651 		read_lock_bh(&idev->lock);
2652 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
2653 			if (ifp->scope == IFA_LINK &&
2654 			    !(ifp->flags & banned_flags)) {
2655 				memcpy(addr, &ifp->addr, 16);
2656 				err = 0;
2657 				break;
2658 			}
2659 		}
2660 		read_unlock_bh(&idev->lock);
2661 	}
2662 	rcu_read_unlock();
2663 	return err;
2664 }
2665 
2666 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2667 {
2668 	struct in6_addr uninitialized_var(addr);
2669 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2670 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2671 
2672 	if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2673 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2674 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2675 		return 0;
2676 	}
2677 	return -EADDRNOTAVAIL;
2678 }
2679 
2680 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2681 {
2682 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2683 	struct c4iw_ep *ep;
2684 	int err = 0;
2685 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2686 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2687 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2688 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2689 				      &cm_id->remote_addr;
2690 	__u8 *ra;
2691 	int iptype;
2692 
2693 	if ((conn_param->ord > c4iw_max_read_depth) ||
2694 	    (conn_param->ird > c4iw_max_read_depth)) {
2695 		err = -EINVAL;
2696 		goto out;
2697 	}
2698 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2699 	if (!ep) {
2700 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2701 		err = -ENOMEM;
2702 		goto out;
2703 	}
2704 	init_timer(&ep->timer);
2705 	ep->plen = conn_param->private_data_len;
2706 	if (ep->plen)
2707 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2708 		       conn_param->private_data, ep->plen);
2709 	ep->ird = conn_param->ird;
2710 	ep->ord = conn_param->ord;
2711 
2712 	if (peer2peer && ep->ord == 0)
2713 		ep->ord = 1;
2714 
2715 	cm_id->add_ref(cm_id);
2716 	ep->com.dev = dev;
2717 	ep->com.cm_id = cm_id;
2718 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2719 	if (!ep->com.qp) {
2720 		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2721 		err = -EINVAL;
2722 		goto fail2;
2723 	}
2724 	ref_qp(ep);
2725 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2726 	     ep->com.qp, cm_id);
2727 
2728 	/*
2729 	 * Allocate an active TID to initiate a TCP connection.
2730 	 */
2731 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2732 	if (ep->atid == -1) {
2733 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2734 		err = -ENOMEM;
2735 		goto fail2;
2736 	}
2737 	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2738 
2739 	if (cm_id->remote_addr.ss_family == AF_INET) {
2740 		iptype = 4;
2741 		ra = (__u8 *)&raddr->sin_addr;
2742 
2743 		/*
2744 		 * Handle loopback requests to INADDR_ANY.
2745 		 */
2746 		if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2747 			err = pick_local_ipaddrs(dev, cm_id);
2748 			if (err)
2749 				goto fail2;
2750 		}
2751 
2752 		/* find a route */
2753 		PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2754 		     __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2755 		     ra, ntohs(raddr->sin_port));
2756 		ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2757 				     raddr->sin_addr.s_addr, laddr->sin_port,
2758 				     raddr->sin_port, 0);
2759 	} else {
2760 		iptype = 6;
2761 		ra = (__u8 *)&raddr6->sin6_addr;
2762 
2763 		/*
2764 		 * Handle loopback requests to INADDR_ANY.
2765 		 */
2766 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2767 			err = pick_local_ip6addrs(dev, cm_id);
2768 			if (err)
2769 				goto fail2;
2770 		}
2771 
2772 		/* find a route */
2773 		PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2774 		     __func__, laddr6->sin6_addr.s6_addr,
2775 		     ntohs(laddr6->sin6_port),
2776 		     raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2777 		ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2778 				      raddr6->sin6_addr.s6_addr,
2779 				      laddr6->sin6_port, raddr6->sin6_port, 0,
2780 				      raddr6->sin6_scope_id);
2781 	}
2782 	if (!ep->dst) {
2783 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2784 		err = -EHOSTUNREACH;
2785 		goto fail3;
2786 	}
2787 
2788 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2789 	if (err) {
2790 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2791 		goto fail4;
2792 	}
2793 
2794 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2795 		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2796 		ep->l2t->idx);
2797 
2798 	state_set(&ep->com, CONNECTING);
2799 	ep->tos = 0;
2800 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
2801 	       sizeof(ep->com.local_addr));
2802 	memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2803 	       sizeof(ep->com.remote_addr));
2804 
2805 	/* send connect request to rnic */
2806 	err = send_connect(ep);
2807 	if (!err)
2808 		goto out;
2809 
2810 	cxgb4_l2t_release(ep->l2t);
2811 fail4:
2812 	dst_release(ep->dst);
2813 fail3:
2814 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2815 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2816 fail2:
2817 	cm_id->rem_ref(cm_id);
2818 	c4iw_put_ep(&ep->com);
2819 out:
2820 	return err;
2821 }
2822 
2823 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2824 {
2825 	int err;
2826 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2827 
2828 	c4iw_init_wr_wait(&ep->com.wr_wait);
2829 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2830 				   ep->stid, &sin6->sin6_addr,
2831 				   sin6->sin6_port,
2832 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
2833 	if (!err)
2834 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2835 					  &ep->com.wr_wait,
2836 					  0, 0, __func__);
2837 	if (err)
2838 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2839 		       err, ep->stid,
2840 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2841 	return err;
2842 }
2843 
2844 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2845 {
2846 	int err;
2847 	struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2848 
2849 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
2850 		do {
2851 			err = cxgb4_create_server_filter(
2852 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
2853 				sin->sin_addr.s_addr, sin->sin_port, 0,
2854 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2855 			if (err == -EBUSY) {
2856 				set_current_state(TASK_UNINTERRUPTIBLE);
2857 				schedule_timeout(usecs_to_jiffies(100));
2858 			}
2859 		} while (err == -EBUSY);
2860 	} else {
2861 		c4iw_init_wr_wait(&ep->com.wr_wait);
2862 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2863 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2864 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2865 		if (!err)
2866 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2867 						  &ep->com.wr_wait,
2868 						  0, 0, __func__);
2869 	}
2870 	if (err)
2871 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2872 		       , err, ep->stid,
2873 		       &sin->sin_addr, ntohs(sin->sin_port));
2874 	return err;
2875 }
2876 
2877 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2878 {
2879 	int err = 0;
2880 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2881 	struct c4iw_listen_ep *ep;
2882 
2883 	might_sleep();
2884 
2885 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2886 	if (!ep) {
2887 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2888 		err = -ENOMEM;
2889 		goto fail1;
2890 	}
2891 	PDBG("%s ep %p\n", __func__, ep);
2892 	cm_id->add_ref(cm_id);
2893 	ep->com.cm_id = cm_id;
2894 	ep->com.dev = dev;
2895 	ep->backlog = backlog;
2896 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
2897 	       sizeof(ep->com.local_addr));
2898 
2899 	/*
2900 	 * Allocate a server TID.
2901 	 */
2902 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
2903 	    ep->com.local_addr.ss_family == AF_INET)
2904 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2905 					     cm_id->local_addr.ss_family, ep);
2906 	else
2907 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2908 					    cm_id->local_addr.ss_family, ep);
2909 
2910 	if (ep->stid == -1) {
2911 		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2912 		err = -ENOMEM;
2913 		goto fail2;
2914 	}
2915 	insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2916 	state_set(&ep->com, LISTEN);
2917 	if (ep->com.local_addr.ss_family == AF_INET)
2918 		err = create_server4(dev, ep);
2919 	else
2920 		err = create_server6(dev, ep);
2921 	if (!err) {
2922 		cm_id->provider_data = ep;
2923 		goto out;
2924 	}
2925 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2926 			ep->com.local_addr.ss_family);
2927 fail2:
2928 	cm_id->rem_ref(cm_id);
2929 	c4iw_put_ep(&ep->com);
2930 fail1:
2931 out:
2932 	return err;
2933 }
2934 
2935 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2936 {
2937 	int err;
2938 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2939 
2940 	PDBG("%s ep %p\n", __func__, ep);
2941 
2942 	might_sleep();
2943 	state_set(&ep->com, DEAD);
2944 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2945 	    ep->com.local_addr.ss_family == AF_INET) {
2946 		err = cxgb4_remove_server_filter(
2947 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
2948 			ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2949 	} else {
2950 		c4iw_init_wr_wait(&ep->com.wr_wait);
2951 		err = cxgb4_remove_server(
2952 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
2953 				ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2954 		if (err)
2955 			goto done;
2956 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2957 					  0, 0, __func__);
2958 	}
2959 	remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
2960 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2961 			ep->com.local_addr.ss_family);
2962 done:
2963 	cm_id->rem_ref(cm_id);
2964 	c4iw_put_ep(&ep->com);
2965 	return err;
2966 }
2967 
2968 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2969 {
2970 	int ret = 0;
2971 	int close = 0;
2972 	int fatal = 0;
2973 	struct c4iw_rdev *rdev;
2974 
2975 	mutex_lock(&ep->com.mutex);
2976 
2977 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2978 	     states[ep->com.state], abrupt);
2979 
2980 	rdev = &ep->com.dev->rdev;
2981 	if (c4iw_fatal_error(rdev)) {
2982 		fatal = 1;
2983 		close_complete_upcall(ep);
2984 		ep->com.state = DEAD;
2985 	}
2986 	switch (ep->com.state) {
2987 	case MPA_REQ_WAIT:
2988 	case MPA_REQ_SENT:
2989 	case MPA_REQ_RCVD:
2990 	case MPA_REP_SENT:
2991 	case FPDU_MODE:
2992 		close = 1;
2993 		if (abrupt)
2994 			ep->com.state = ABORTING;
2995 		else {
2996 			ep->com.state = CLOSING;
2997 			start_ep_timer(ep);
2998 		}
2999 		set_bit(CLOSE_SENT, &ep->com.flags);
3000 		break;
3001 	case CLOSING:
3002 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3003 			close = 1;
3004 			if (abrupt) {
3005 				stop_ep_timer(ep);
3006 				ep->com.state = ABORTING;
3007 			} else
3008 				ep->com.state = MORIBUND;
3009 		}
3010 		break;
3011 	case MORIBUND:
3012 	case ABORTING:
3013 	case DEAD:
3014 		PDBG("%s ignoring disconnect ep %p state %u\n",
3015 		     __func__, ep, ep->com.state);
3016 		break;
3017 	default:
3018 		BUG();
3019 		break;
3020 	}
3021 
3022 	if (close) {
3023 		if (abrupt) {
3024 			set_bit(EP_DISC_ABORT, &ep->com.history);
3025 			close_complete_upcall(ep);
3026 			ret = send_abort(ep, NULL, gfp);
3027 		} else {
3028 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3029 			ret = send_halfclose(ep, gfp);
3030 		}
3031 		if (ret)
3032 			fatal = 1;
3033 	}
3034 	mutex_unlock(&ep->com.mutex);
3035 	if (fatal)
3036 		release_ep_resources(ep);
3037 	return ret;
3038 }
3039 
3040 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3041 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3042 {
3043 	struct c4iw_ep *ep;
3044 	int atid = be32_to_cpu(req->tid);
3045 
3046 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3047 					   (__force u32) req->tid);
3048 	if (!ep)
3049 		return;
3050 
3051 	switch (req->retval) {
3052 	case FW_ENOMEM:
3053 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3054 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3055 			send_fw_act_open_req(ep, atid);
3056 			return;
3057 		}
3058 	case FW_EADDRINUSE:
3059 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3060 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3061 			send_fw_act_open_req(ep, atid);
3062 			return;
3063 		}
3064 		break;
3065 	default:
3066 		pr_info("%s unexpected ofld conn wr retval %d\n",
3067 		       __func__, req->retval);
3068 		break;
3069 	}
3070 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3071 	       req->retval, atid);
3072 	mutex_lock(&dev->rdev.stats.lock);
3073 	dev->rdev.stats.act_ofld_conn_fails++;
3074 	mutex_unlock(&dev->rdev.stats.lock);
3075 	connect_reply_upcall(ep, status2errno(req->retval));
3076 	state_set(&ep->com, DEAD);
3077 	remove_handle(dev, &dev->atid_idr, atid);
3078 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3079 	dst_release(ep->dst);
3080 	cxgb4_l2t_release(ep->l2t);
3081 	c4iw_put_ep(&ep->com);
3082 }
3083 
3084 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3085 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3086 {
3087 	struct sk_buff *rpl_skb;
3088 	struct cpl_pass_accept_req *cpl;
3089 	int ret;
3090 
3091 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3092 	BUG_ON(!rpl_skb);
3093 	if (req->retval) {
3094 		PDBG("%s passive open failure %d\n", __func__, req->retval);
3095 		mutex_lock(&dev->rdev.stats.lock);
3096 		dev->rdev.stats.pas_ofld_conn_fails++;
3097 		mutex_unlock(&dev->rdev.stats.lock);
3098 		kfree_skb(rpl_skb);
3099 	} else {
3100 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3101 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3102 					(__force u32) htonl(
3103 					(__force u32) req->tid)));
3104 		ret = pass_accept_req(dev, rpl_skb);
3105 		if (!ret)
3106 			kfree_skb(rpl_skb);
3107 	}
3108 	return;
3109 }
3110 
3111 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3112 {
3113 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3114 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3115 
3116 	switch (rpl->type) {
3117 	case FW6_TYPE_CQE:
3118 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3119 		break;
3120 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3121 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3122 		switch (req->t_state) {
3123 		case TCP_SYN_SENT:
3124 			active_ofld_conn_reply(dev, skb, req);
3125 			break;
3126 		case TCP_SYN_RECV:
3127 			passive_ofld_conn_reply(dev, skb, req);
3128 			break;
3129 		default:
3130 			pr_err("%s unexpected ofld conn wr state %d\n",
3131 			       __func__, req->t_state);
3132 			break;
3133 		}
3134 		break;
3135 	}
3136 	return 0;
3137 }
3138 
3139 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3140 {
3141 	u32 l2info;
3142 	u16 vlantag, len, hdr_len, eth_hdr_len;
3143 	u8 intf;
3144 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3145 	struct cpl_pass_accept_req *req;
3146 	struct tcp_options_received tmp_opt;
3147 	struct c4iw_dev *dev;
3148 
3149 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3150 	/* Store values from cpl_rx_pkt in temporary location. */
3151 	vlantag = (__force u16) cpl->vlan;
3152 	len = (__force u16) cpl->len;
3153 	l2info  = (__force u32) cpl->l2info;
3154 	hdr_len = (__force u16) cpl->hdr_len;
3155 	intf = cpl->iff;
3156 
3157 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3158 
3159 	/*
3160 	 * We need to parse the TCP options from SYN packet.
3161 	 * to generate cpl_pass_accept_req.
3162 	 */
3163 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3164 	tcp_clear_options(&tmp_opt);
3165 	tcp_parse_options(skb, &tmp_opt, 0, NULL);
3166 
3167 	req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3168 	memset(req, 0, sizeof(*req));
3169 	req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3170 			 V_SYN_MAC_IDX(G_RX_MACIDX(
3171 			 (__force int) htonl(l2info))) |
3172 			 F_SYN_XACT_MATCH);
3173 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3174 			    G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3175 			    G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3176 	req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3177 					(__force int) htonl(l2info))) |
3178 				   V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3179 					(__force int) htons(hdr_len))) |
3180 				   V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3181 					(__force int) htons(hdr_len))) |
3182 				   V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3183 	req->vlan = (__force __be16) vlantag;
3184 	req->len = (__force __be16) len;
3185 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3186 				    PASS_OPEN_TOS(tos));
3187 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3188 	if (tmp_opt.wscale_ok)
3189 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3190 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3191 	if (tmp_opt.sack_ok)
3192 		req->tcpopt.sack = 1;
3193 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3194 	return;
3195 }
3196 
3197 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3198 				  __be32 laddr, __be16 lport,
3199 				  __be32 raddr, __be16 rport,
3200 				  u32 rcv_isn, u32 filter, u16 window,
3201 				  u32 rss_qid, u8 port_id)
3202 {
3203 	struct sk_buff *req_skb;
3204 	struct fw_ofld_connection_wr *req;
3205 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
3206 
3207 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3208 	req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3209 	memset(req, 0, sizeof(*req));
3210 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3211 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3212 	req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3213 	req->le.filter = (__force __be32) filter;
3214 	req->le.lport = lport;
3215 	req->le.pport = rport;
3216 	req->le.u.ipv4.lip = laddr;
3217 	req->le.u.ipv4.pip = raddr;
3218 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3219 	req->tcb.rcv_adv = htons(window);
3220 	req->tcb.t_state_to_astid =
3221 		 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3222 			V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3223 			V_FW_OFLD_CONNECTION_WR_ASTID(
3224 			GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3225 
3226 	/*
3227 	 * We store the qid in opt2 which will be used by the firmware
3228 	 * to send us the wr response.
3229 	 */
3230 	req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3231 
3232 	/*
3233 	 * We initialize the MSS index in TCB to 0xF.
3234 	 * So that when driver sends cpl_pass_accept_rpl
3235 	 * TCB picks up the correct value. If this was 0
3236 	 * TP will ignore any value > 0 for MSS index.
3237 	 */
3238 	req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3239 	req->cookie = (unsigned long)skb;
3240 
3241 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3242 	cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3243 }
3244 
3245 /*
3246  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3247  * messages when a filter is being used instead of server to
3248  * redirect a syn packet. When packets hit filter they are redirected
3249  * to the offload queue and driver tries to establish the connection
3250  * using firmware work request.
3251  */
3252 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3253 {
3254 	int stid;
3255 	unsigned int filter;
3256 	struct ethhdr *eh = NULL;
3257 	struct vlan_ethhdr *vlan_eh = NULL;
3258 	struct iphdr *iph;
3259 	struct tcphdr *tcph;
3260 	struct rss_header *rss = (void *)skb->data;
3261 	struct cpl_rx_pkt *cpl = (void *)skb->data;
3262 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
3263 	struct l2t_entry *e;
3264 	struct dst_entry *dst;
3265 	struct c4iw_ep *lep;
3266 	u16 window;
3267 	struct port_info *pi;
3268 	struct net_device *pdev;
3269 	u16 rss_qid, eth_hdr_len;
3270 	int step;
3271 	u32 tx_chan;
3272 	struct neighbour *neigh;
3273 
3274 	/* Drop all non-SYN packets */
3275 	if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3276 		goto reject;
3277 
3278 	/*
3279 	 * Drop all packets which did not hit the filter.
3280 	 * Unlikely to happen.
3281 	 */
3282 	if (!(rss->filter_hit && rss->filter_tid))
3283 		goto reject;
3284 
3285 	/*
3286 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3287 	 */
3288 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3289 
3290 	lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3291 	if (!lep) {
3292 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3293 		goto reject;
3294 	}
3295 
3296 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3297 			    G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3298 			    G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3299 	if (eth_hdr_len == ETH_HLEN) {
3300 		eh = (struct ethhdr *)(req + 1);
3301 		iph = (struct iphdr *)(eh + 1);
3302 	} else {
3303 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
3304 		iph = (struct iphdr *)(vlan_eh + 1);
3305 		skb->vlan_tci = ntohs(cpl->vlan);
3306 	}
3307 
3308 	if (iph->version != 0x4)
3309 		goto reject;
3310 
3311 	tcph = (struct tcphdr *)(iph + 1);
3312 	skb_set_network_header(skb, (void *)iph - (void *)rss);
3313 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3314 	skb_get(skb);
3315 
3316 	PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3317 	     ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3318 	     ntohs(tcph->source), iph->tos);
3319 
3320 	dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3321 			 iph->tos);
3322 	if (!dst) {
3323 		pr_err("%s - failed to find dst entry!\n",
3324 		       __func__);
3325 		goto reject;
3326 	}
3327 	neigh = dst_neigh_lookup_skb(dst, skb);
3328 
3329 	if (!neigh) {
3330 		pr_err("%s - failed to allocate neigh!\n",
3331 		       __func__);
3332 		goto free_dst;
3333 	}
3334 
3335 	if (neigh->dev->flags & IFF_LOOPBACK) {
3336 		pdev = ip_dev_find(&init_net, iph->daddr);
3337 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3338 				    pdev, 0);
3339 		pi = (struct port_info *)netdev_priv(pdev);
3340 		tx_chan = cxgb4_port_chan(pdev);
3341 		dev_put(pdev);
3342 	} else {
3343 		pdev = get_real_dev(neigh->dev);
3344 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3345 					pdev, 0);
3346 		pi = (struct port_info *)netdev_priv(pdev);
3347 		tx_chan = cxgb4_port_chan(pdev);
3348 	}
3349 	if (!e) {
3350 		pr_err("%s - failed to allocate l2t entry!\n",
3351 		       __func__);
3352 		goto free_dst;
3353 	}
3354 
3355 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3356 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3357 	window = (__force u16) htons((__force u16)tcph->window);
3358 
3359 	/* Calcuate filter portion for LE region. */
3360 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3361 						    dev->rdev.lldi.ports[0],
3362 						    e));
3363 
3364 	/*
3365 	 * Synthesize the cpl_pass_accept_req. We have everything except the
3366 	 * TID. Once firmware sends a reply with TID we update the TID field
3367 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
3368 	 */
3369 	build_cpl_pass_accept_req(skb, stid, iph->tos);
3370 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3371 			      tcph->source, ntohl(tcph->seq), filter, window,
3372 			      rss_qid, pi->port_id);
3373 	cxgb4_l2t_release(e);
3374 free_dst:
3375 	dst_release(dst);
3376 reject:
3377 	return 0;
3378 }
3379 
3380 /*
3381  * These are the real handlers that are called from a
3382  * work queue.
3383  */
3384 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3385 	[CPL_ACT_ESTABLISH] = act_establish,
3386 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
3387 	[CPL_RX_DATA] = rx_data,
3388 	[CPL_ABORT_RPL_RSS] = abort_rpl,
3389 	[CPL_ABORT_RPL] = abort_rpl,
3390 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
3391 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3392 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3393 	[CPL_PASS_ESTABLISH] = pass_establish,
3394 	[CPL_PEER_CLOSE] = peer_close,
3395 	[CPL_ABORT_REQ_RSS] = peer_abort,
3396 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
3397 	[CPL_RDMA_TERMINATE] = terminate,
3398 	[CPL_FW4_ACK] = fw4_ack,
3399 	[CPL_FW6_MSG] = deferred_fw6_msg,
3400 	[CPL_RX_PKT] = rx_pkt
3401 };
3402 
3403 static void process_timeout(struct c4iw_ep *ep)
3404 {
3405 	struct c4iw_qp_attributes attrs;
3406 	int abort = 1;
3407 
3408 	mutex_lock(&ep->com.mutex);
3409 	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3410 	     ep->com.state);
3411 	set_bit(TIMEDOUT, &ep->com.history);
3412 	switch (ep->com.state) {
3413 	case MPA_REQ_SENT:
3414 		__state_set(&ep->com, ABORTING);
3415 		connect_reply_upcall(ep, -ETIMEDOUT);
3416 		break;
3417 	case MPA_REQ_WAIT:
3418 		__state_set(&ep->com, ABORTING);
3419 		break;
3420 	case CLOSING:
3421 	case MORIBUND:
3422 		if (ep->com.cm_id && ep->com.qp) {
3423 			attrs.next_state = C4IW_QP_STATE_ERROR;
3424 			c4iw_modify_qp(ep->com.qp->rhp,
3425 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3426 				     &attrs, 1);
3427 		}
3428 		__state_set(&ep->com, ABORTING);
3429 		break;
3430 	default:
3431 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3432 			__func__, ep, ep->hwtid, ep->com.state);
3433 		abort = 0;
3434 	}
3435 	mutex_unlock(&ep->com.mutex);
3436 	if (abort)
3437 		abort_connection(ep, NULL, GFP_KERNEL);
3438 	c4iw_put_ep(&ep->com);
3439 }
3440 
3441 static void process_timedout_eps(void)
3442 {
3443 	struct c4iw_ep *ep;
3444 
3445 	spin_lock_irq(&timeout_lock);
3446 	while (!list_empty(&timeout_list)) {
3447 		struct list_head *tmp;
3448 
3449 		tmp = timeout_list.next;
3450 		list_del(tmp);
3451 		spin_unlock_irq(&timeout_lock);
3452 		ep = list_entry(tmp, struct c4iw_ep, entry);
3453 		process_timeout(ep);
3454 		spin_lock_irq(&timeout_lock);
3455 	}
3456 	spin_unlock_irq(&timeout_lock);
3457 }
3458 
3459 static void process_work(struct work_struct *work)
3460 {
3461 	struct sk_buff *skb = NULL;
3462 	struct c4iw_dev *dev;
3463 	struct cpl_act_establish *rpl;
3464 	unsigned int opcode;
3465 	int ret;
3466 
3467 	while ((skb = skb_dequeue(&rxq))) {
3468 		rpl = cplhdr(skb);
3469 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3470 		opcode = rpl->ot.opcode;
3471 
3472 		BUG_ON(!work_handlers[opcode]);
3473 		ret = work_handlers[opcode](dev, skb);
3474 		if (!ret)
3475 			kfree_skb(skb);
3476 	}
3477 	process_timedout_eps();
3478 }
3479 
3480 static DECLARE_WORK(skb_work, process_work);
3481 
3482 static void ep_timeout(unsigned long arg)
3483 {
3484 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3485 	int kickit = 0;
3486 
3487 	spin_lock(&timeout_lock);
3488 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3489 		list_add_tail(&ep->entry, &timeout_list);
3490 		kickit = 1;
3491 	}
3492 	spin_unlock(&timeout_lock);
3493 	if (kickit)
3494 		queue_work(workq, &skb_work);
3495 }
3496 
3497 /*
3498  * All the CM events are handled on a work queue to have a safe context.
3499  */
3500 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3501 {
3502 
3503 	/*
3504 	 * Save dev in the skb->cb area.
3505 	 */
3506 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3507 
3508 	/*
3509 	 * Queue the skb and schedule the worker thread.
3510 	 */
3511 	skb_queue_tail(&rxq, skb);
3512 	queue_work(workq, &skb_work);
3513 	return 0;
3514 }
3515 
3516 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3517 {
3518 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3519 
3520 	if (rpl->status != CPL_ERR_NONE) {
3521 		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3522 		       "for tid %u\n", rpl->status, GET_TID(rpl));
3523 	}
3524 	kfree_skb(skb);
3525 	return 0;
3526 }
3527 
3528 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3529 {
3530 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3531 	struct c4iw_wr_wait *wr_waitp;
3532 	int ret;
3533 
3534 	PDBG("%s type %u\n", __func__, rpl->type);
3535 
3536 	switch (rpl->type) {
3537 	case FW6_TYPE_WR_RPL:
3538 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3539 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3540 		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3541 		if (wr_waitp)
3542 			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3543 		kfree_skb(skb);
3544 		break;
3545 	case FW6_TYPE_CQE:
3546 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3547 		sched(dev, skb);
3548 		break;
3549 	default:
3550 		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3551 		       rpl->type);
3552 		kfree_skb(skb);
3553 		break;
3554 	}
3555 	return 0;
3556 }
3557 
3558 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3559 {
3560 	struct cpl_abort_req_rss *req = cplhdr(skb);
3561 	struct c4iw_ep *ep;
3562 	struct tid_info *t = dev->rdev.lldi.tids;
3563 	unsigned int tid = GET_TID(req);
3564 
3565 	ep = lookup_tid(t, tid);
3566 	if (!ep) {
3567 		printk(KERN_WARNING MOD
3568 		       "Abort on non-existent endpoint, tid %d\n", tid);
3569 		kfree_skb(skb);
3570 		return 0;
3571 	}
3572 	if (is_neg_adv_abort(req->status)) {
3573 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3574 		     ep->hwtid);
3575 		kfree_skb(skb);
3576 		return 0;
3577 	}
3578 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3579 	     ep->com.state);
3580 
3581 	/*
3582 	 * Wake up any threads in rdma_init() or rdma_fini().
3583 	 * However, if we are on MPAv2 and want to retry with MPAv1
3584 	 * then, don't wake up yet.
3585 	 */
3586 	if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3587 		if (ep->com.state != MPA_REQ_SENT)
3588 			c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3589 	} else
3590 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3591 	sched(dev, skb);
3592 	return 0;
3593 }
3594 
3595 /*
3596  * Most upcalls from the T4 Core go to sched() to
3597  * schedule the processing on a work queue.
3598  */
3599 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3600 	[CPL_ACT_ESTABLISH] = sched,
3601 	[CPL_ACT_OPEN_RPL] = sched,
3602 	[CPL_RX_DATA] = sched,
3603 	[CPL_ABORT_RPL_RSS] = sched,
3604 	[CPL_ABORT_RPL] = sched,
3605 	[CPL_PASS_OPEN_RPL] = sched,
3606 	[CPL_CLOSE_LISTSRV_RPL] = sched,
3607 	[CPL_PASS_ACCEPT_REQ] = sched,
3608 	[CPL_PASS_ESTABLISH] = sched,
3609 	[CPL_PEER_CLOSE] = sched,
3610 	[CPL_CLOSE_CON_RPL] = sched,
3611 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
3612 	[CPL_RDMA_TERMINATE] = sched,
3613 	[CPL_FW4_ACK] = sched,
3614 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
3615 	[CPL_FW6_MSG] = fw6_msg,
3616 	[CPL_RX_PKT] = sched
3617 };
3618 
3619 int __init c4iw_cm_init(void)
3620 {
3621 	spin_lock_init(&timeout_lock);
3622 	skb_queue_head_init(&rxq);
3623 
3624 	workq = create_singlethread_workqueue("iw_cxgb4");
3625 	if (!workq)
3626 		return -ENOMEM;
3627 
3628 	return 0;
3629 }
3630 
3631 void __exit c4iw_cm_term(void)
3632 {
3633 	WARN_ON(!list_empty(&timeout_list));
3634 	flush_workqueue(workq);
3635 	destroy_workqueue(workq);
3636 }
3637