xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision 84d517f3)
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 = 1;
102 module_param(peer2peer, int, 0644);
103 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
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 int 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 		return 0;
183 	}
184 	return 1;
185 }
186 
187 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
188 		  struct l2t_entry *l2e)
189 {
190 	int	error = 0;
191 
192 	if (c4iw_fatal_error(rdev)) {
193 		kfree_skb(skb);
194 		PDBG("%s - device in error state - dropping\n", __func__);
195 		return -EIO;
196 	}
197 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
198 	if (error < 0)
199 		kfree_skb(skb);
200 	return error < 0 ? error : 0;
201 }
202 
203 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
204 {
205 	int	error = 0;
206 
207 	if (c4iw_fatal_error(rdev)) {
208 		kfree_skb(skb);
209 		PDBG("%s - device in error state - dropping\n", __func__);
210 		return -EIO;
211 	}
212 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
213 	if (error < 0)
214 		kfree_skb(skb);
215 	return error < 0 ? error : 0;
216 }
217 
218 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
219 {
220 	struct cpl_tid_release *req;
221 
222 	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
223 	if (!skb)
224 		return;
225 	req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
226 	INIT_TP_WR(req, hwtid);
227 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
228 	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
229 	c4iw_ofld_send(rdev, skb);
230 	return;
231 }
232 
233 static void set_emss(struct c4iw_ep *ep, u16 opt)
234 {
235 	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
236 	ep->mss = ep->emss;
237 	if (GET_TCPOPT_TSTAMP(opt))
238 		ep->emss -= 12;
239 	if (ep->emss < 128)
240 		ep->emss = 128;
241 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
242 	     ep->mss, ep->emss);
243 }
244 
245 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
246 {
247 	enum c4iw_ep_state state;
248 
249 	mutex_lock(&epc->mutex);
250 	state = epc->state;
251 	mutex_unlock(&epc->mutex);
252 	return state;
253 }
254 
255 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
256 {
257 	epc->state = new;
258 }
259 
260 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
261 {
262 	mutex_lock(&epc->mutex);
263 	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
264 	__state_set(epc, new);
265 	mutex_unlock(&epc->mutex);
266 	return;
267 }
268 
269 static void *alloc_ep(int size, gfp_t gfp)
270 {
271 	struct c4iw_ep_common *epc;
272 
273 	epc = kzalloc(size, gfp);
274 	if (epc) {
275 		kref_init(&epc->kref);
276 		mutex_init(&epc->mutex);
277 		c4iw_init_wr_wait(&epc->wr_wait);
278 	}
279 	PDBG("%s alloc ep %p\n", __func__, epc);
280 	return epc;
281 }
282 
283 void _c4iw_free_ep(struct kref *kref)
284 {
285 	struct c4iw_ep *ep;
286 
287 	ep = container_of(kref, struct c4iw_ep, com.kref);
288 	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
289 	if (test_bit(QP_REFERENCED, &ep->com.flags))
290 		deref_qp(ep);
291 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
292 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
293 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
294 		dst_release(ep->dst);
295 		cxgb4_l2t_release(ep->l2t);
296 	}
297 	kfree(ep);
298 }
299 
300 static void release_ep_resources(struct c4iw_ep *ep)
301 {
302 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
303 	c4iw_put_ep(&ep->com);
304 }
305 
306 static int status2errno(int status)
307 {
308 	switch (status) {
309 	case CPL_ERR_NONE:
310 		return 0;
311 	case CPL_ERR_CONN_RESET:
312 		return -ECONNRESET;
313 	case CPL_ERR_ARP_MISS:
314 		return -EHOSTUNREACH;
315 	case CPL_ERR_CONN_TIMEDOUT:
316 		return -ETIMEDOUT;
317 	case CPL_ERR_TCAM_FULL:
318 		return -ENOMEM;
319 	case CPL_ERR_CONN_EXIST:
320 		return -EADDRINUSE;
321 	default:
322 		return -EIO;
323 	}
324 }
325 
326 /*
327  * Try and reuse skbs already allocated...
328  */
329 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
330 {
331 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
332 		skb_trim(skb, 0);
333 		skb_get(skb);
334 		skb_reset_transport_header(skb);
335 	} else {
336 		skb = alloc_skb(len, gfp);
337 	}
338 	t4_set_arp_err_handler(skb, NULL, NULL);
339 	return skb;
340 }
341 
342 static struct net_device *get_real_dev(struct net_device *egress_dev)
343 {
344 	struct net_device *phys_dev = egress_dev;
345 	if (egress_dev->priv_flags & IFF_802_1Q_VLAN)
346 		phys_dev = vlan_dev_real_dev(egress_dev);
347 	return phys_dev;
348 }
349 
350 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
351 {
352 	int i;
353 
354 	egress_dev = get_real_dev(egress_dev);
355 	for (i = 0; i < dev->rdev.lldi.nports; i++)
356 		if (dev->rdev.lldi.ports[i] == egress_dev)
357 			return 1;
358 	return 0;
359 }
360 
361 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
362 				     __u8 *peer_ip, __be16 local_port,
363 				     __be16 peer_port, u8 tos,
364 				     __u32 sin6_scope_id)
365 {
366 	struct dst_entry *dst = NULL;
367 
368 	if (IS_ENABLED(CONFIG_IPV6)) {
369 		struct flowi6 fl6;
370 
371 		memset(&fl6, 0, sizeof(fl6));
372 		memcpy(&fl6.daddr, peer_ip, 16);
373 		memcpy(&fl6.saddr, local_ip, 16);
374 		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
375 			fl6.flowi6_oif = sin6_scope_id;
376 		dst = ip6_route_output(&init_net, NULL, &fl6);
377 		if (!dst)
378 			goto out;
379 		if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
380 		    !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
381 			dst_release(dst);
382 			dst = NULL;
383 		}
384 	}
385 
386 out:
387 	return dst;
388 }
389 
390 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
391 				 __be32 peer_ip, __be16 local_port,
392 				 __be16 peer_port, u8 tos)
393 {
394 	struct rtable *rt;
395 	struct flowi4 fl4;
396 	struct neighbour *n;
397 
398 	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
399 				   peer_port, local_port, IPPROTO_TCP,
400 				   tos, 0);
401 	if (IS_ERR(rt))
402 		return NULL;
403 	n = dst_neigh_lookup(&rt->dst, &peer_ip);
404 	if (!n)
405 		return NULL;
406 	if (!our_interface(dev, n->dev) &&
407 	    !(n->dev->flags & IFF_LOOPBACK)) {
408 		dst_release(&rt->dst);
409 		return NULL;
410 	}
411 	neigh_release(n);
412 	return &rt->dst;
413 }
414 
415 static void arp_failure_discard(void *handle, struct sk_buff *skb)
416 {
417 	PDBG("%s c4iw_dev %p\n", __func__, handle);
418 	kfree_skb(skb);
419 }
420 
421 /*
422  * Handle an ARP failure for an active open.
423  */
424 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
425 {
426 	printk(KERN_ERR MOD "ARP failure duing connect\n");
427 	kfree_skb(skb);
428 }
429 
430 /*
431  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
432  * and send it along.
433  */
434 static void abort_arp_failure(void *handle, struct sk_buff *skb)
435 {
436 	struct c4iw_rdev *rdev = handle;
437 	struct cpl_abort_req *req = cplhdr(skb);
438 
439 	PDBG("%s rdev %p\n", __func__, rdev);
440 	req->cmd = CPL_ABORT_NO_RST;
441 	c4iw_ofld_send(rdev, skb);
442 }
443 
444 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
445 {
446 	unsigned int flowclen = 80;
447 	struct fw_flowc_wr *flowc;
448 	int i;
449 
450 	skb = get_skb(skb, flowclen, GFP_KERNEL);
451 	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
452 
453 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
454 					   FW_FLOWC_WR_NPARAMS(8));
455 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
456 					  16)) | FW_WR_FLOWID(ep->hwtid));
457 
458 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
459 	flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
460 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
461 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
462 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
463 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
464 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
465 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
466 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
467 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
468 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
469 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
470 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
471 	flowc->mnemval[6].val = cpu_to_be32(snd_win);
472 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
473 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
474 	/* Pad WR to 16 byte boundary */
475 	flowc->mnemval[8].mnemonic = 0;
476 	flowc->mnemval[8].val = 0;
477 	for (i = 0; i < 9; i++) {
478 		flowc->mnemval[i].r4[0] = 0;
479 		flowc->mnemval[i].r4[1] = 0;
480 		flowc->mnemval[i].r4[2] = 0;
481 	}
482 
483 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
484 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
485 }
486 
487 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
488 {
489 	struct cpl_close_con_req *req;
490 	struct sk_buff *skb;
491 	int wrlen = roundup(sizeof *req, 16);
492 
493 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
494 	skb = get_skb(NULL, wrlen, gfp);
495 	if (!skb) {
496 		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
497 		return -ENOMEM;
498 	}
499 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
500 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
501 	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
502 	memset(req, 0, wrlen);
503 	INIT_TP_WR(req, ep->hwtid);
504 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
505 						    ep->hwtid));
506 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
507 }
508 
509 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
510 {
511 	struct cpl_abort_req *req;
512 	int wrlen = roundup(sizeof *req, 16);
513 
514 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
515 	skb = get_skb(skb, wrlen, gfp);
516 	if (!skb) {
517 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
518 		       __func__);
519 		return -ENOMEM;
520 	}
521 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
522 	t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
523 	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
524 	memset(req, 0, wrlen);
525 	INIT_TP_WR(req, ep->hwtid);
526 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
527 	req->cmd = CPL_ABORT_SEND_RST;
528 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
529 }
530 
531 static int send_connect(struct c4iw_ep *ep)
532 {
533 	struct cpl_act_open_req *req;
534 	struct cpl_t5_act_open_req *t5_req;
535 	struct cpl_act_open_req6 *req6;
536 	struct cpl_t5_act_open_req6 *t5_req6;
537 	struct sk_buff *skb;
538 	u64 opt0;
539 	u32 opt2;
540 	unsigned int mtu_idx;
541 	int wscale;
542 	int wrlen;
543 	int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
544 				sizeof(struct cpl_act_open_req) :
545 				sizeof(struct cpl_t5_act_open_req);
546 	int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
547 				sizeof(struct cpl_act_open_req6) :
548 				sizeof(struct cpl_t5_act_open_req6);
549 	struct sockaddr_in *la = (struct sockaddr_in *)&ep->com.local_addr;
550 	struct sockaddr_in *ra = (struct sockaddr_in *)&ep->com.remote_addr;
551 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
552 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
553 
554 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
555 			roundup(sizev4, 16) :
556 			roundup(sizev6, 16);
557 
558 	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
559 
560 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
561 	if (!skb) {
562 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
563 		       __func__);
564 		return -ENOMEM;
565 	}
566 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
567 
568 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
569 	wscale = compute_wscale(rcv_win);
570 	opt0 = (nocong ? NO_CONG(1) : 0) |
571 	       KEEP_ALIVE(1) |
572 	       DELACK(1) |
573 	       WND_SCALE(wscale) |
574 	       MSS_IDX(mtu_idx) |
575 	       L2T_IDX(ep->l2t->idx) |
576 	       TX_CHAN(ep->tx_chan) |
577 	       SMAC_SEL(ep->smac_idx) |
578 	       DSCP(ep->tos) |
579 	       ULP_MODE(ULP_MODE_TCPDDP) |
580 	       RCV_BUFSIZ(rcv_win>>10);
581 	opt2 = RX_CHANNEL(0) |
582 	       CCTRL_ECN(enable_ecn) |
583 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
584 	if (enable_tcp_timestamps)
585 		opt2 |= TSTAMPS_EN(1);
586 	if (enable_tcp_sack)
587 		opt2 |= SACK_EN(1);
588 	if (wscale && enable_tcp_window_scaling)
589 		opt2 |= WND_SCALE_EN(1);
590 	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
591 		opt2 |= T5_OPT_2_VALID;
592 		opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
593 	}
594 	t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
595 
596 	if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
597 		if (ep->com.remote_addr.ss_family == AF_INET) {
598 			req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
599 			INIT_TP_WR(req, 0);
600 			OPCODE_TID(req) = cpu_to_be32(
601 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
602 					((ep->rss_qid << 14) | ep->atid)));
603 			req->local_port = la->sin_port;
604 			req->peer_port = ra->sin_port;
605 			req->local_ip = la->sin_addr.s_addr;
606 			req->peer_ip = ra->sin_addr.s_addr;
607 			req->opt0 = cpu_to_be64(opt0);
608 			req->params = cpu_to_be32(cxgb4_select_ntuple(
609 						ep->com.dev->rdev.lldi.ports[0],
610 						ep->l2t));
611 			req->opt2 = cpu_to_be32(opt2);
612 		} else {
613 			req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
614 
615 			INIT_TP_WR(req6, 0);
616 			OPCODE_TID(req6) = cpu_to_be32(
617 					   MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
618 					   ((ep->rss_qid<<14)|ep->atid)));
619 			req6->local_port = la6->sin6_port;
620 			req6->peer_port = ra6->sin6_port;
621 			req6->local_ip_hi = *((__be64 *)
622 						(la6->sin6_addr.s6_addr));
623 			req6->local_ip_lo = *((__be64 *)
624 						(la6->sin6_addr.s6_addr + 8));
625 			req6->peer_ip_hi = *((__be64 *)
626 						(ra6->sin6_addr.s6_addr));
627 			req6->peer_ip_lo = *((__be64 *)
628 						(ra6->sin6_addr.s6_addr + 8));
629 			req6->opt0 = cpu_to_be64(opt0);
630 			req6->params = cpu_to_be32(cxgb4_select_ntuple(
631 						ep->com.dev->rdev.lldi.ports[0],
632 						ep->l2t));
633 			req6->opt2 = cpu_to_be32(opt2);
634 		}
635 	} else {
636 		if (ep->com.remote_addr.ss_family == AF_INET) {
637 			t5_req = (struct cpl_t5_act_open_req *)
638 				 skb_put(skb, wrlen);
639 			INIT_TP_WR(t5_req, 0);
640 			OPCODE_TID(t5_req) = cpu_to_be32(
641 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
642 					((ep->rss_qid << 14) | ep->atid)));
643 			t5_req->local_port = la->sin_port;
644 			t5_req->peer_port = ra->sin_port;
645 			t5_req->local_ip = la->sin_addr.s_addr;
646 			t5_req->peer_ip = ra->sin_addr.s_addr;
647 			t5_req->opt0 = cpu_to_be64(opt0);
648 			t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
649 						     cxgb4_select_ntuple(
650 					     ep->com.dev->rdev.lldi.ports[0],
651 					     ep->l2t)));
652 			t5_req->opt2 = cpu_to_be32(opt2);
653 		} else {
654 			t5_req6 = (struct cpl_t5_act_open_req6 *)
655 				  skb_put(skb, wrlen);
656 			INIT_TP_WR(t5_req6, 0);
657 			OPCODE_TID(t5_req6) = cpu_to_be32(
658 					      MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
659 					      ((ep->rss_qid<<14)|ep->atid)));
660 			t5_req6->local_port = la6->sin6_port;
661 			t5_req6->peer_port = ra6->sin6_port;
662 			t5_req6->local_ip_hi = *((__be64 *)
663 						(la6->sin6_addr.s6_addr));
664 			t5_req6->local_ip_lo = *((__be64 *)
665 						(la6->sin6_addr.s6_addr + 8));
666 			t5_req6->peer_ip_hi = *((__be64 *)
667 						(ra6->sin6_addr.s6_addr));
668 			t5_req6->peer_ip_lo = *((__be64 *)
669 						(ra6->sin6_addr.s6_addr + 8));
670 			t5_req6->opt0 = cpu_to_be64(opt0);
671 			t5_req6->params = (__force __be64)cpu_to_be32(
672 							cxgb4_select_ntuple(
673 						ep->com.dev->rdev.lldi.ports[0],
674 						ep->l2t));
675 			t5_req6->opt2 = cpu_to_be32(opt2);
676 		}
677 	}
678 
679 	set_bit(ACT_OPEN_REQ, &ep->com.history);
680 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
681 }
682 
683 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
684 		u8 mpa_rev_to_use)
685 {
686 	int mpalen, wrlen;
687 	struct fw_ofld_tx_data_wr *req;
688 	struct mpa_message *mpa;
689 	struct mpa_v2_conn_params mpa_v2_params;
690 
691 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
692 
693 	BUG_ON(skb_cloned(skb));
694 
695 	mpalen = sizeof(*mpa) + ep->plen;
696 	if (mpa_rev_to_use == 2)
697 		mpalen += sizeof(struct mpa_v2_conn_params);
698 	wrlen = roundup(mpalen + sizeof *req, 16);
699 	skb = get_skb(skb, wrlen, GFP_KERNEL);
700 	if (!skb) {
701 		connect_reply_upcall(ep, -ENOMEM);
702 		return;
703 	}
704 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
705 
706 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
707 	memset(req, 0, wrlen);
708 	req->op_to_immdlen = cpu_to_be32(
709 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
710 		FW_WR_COMPL(1) |
711 		FW_WR_IMMDLEN(mpalen));
712 	req->flowid_len16 = cpu_to_be32(
713 		FW_WR_FLOWID(ep->hwtid) |
714 		FW_WR_LEN16(wrlen >> 4));
715 	req->plen = cpu_to_be32(mpalen);
716 	req->tunnel_to_proxy = cpu_to_be32(
717 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
718 		FW_OFLD_TX_DATA_WR_SHOVE(1));
719 
720 	mpa = (struct mpa_message *)(req + 1);
721 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
722 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
723 		     (markers_enabled ? MPA_MARKERS : 0) |
724 		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
725 	mpa->private_data_size = htons(ep->plen);
726 	mpa->revision = mpa_rev_to_use;
727 	if (mpa_rev_to_use == 1) {
728 		ep->tried_with_mpa_v1 = 1;
729 		ep->retry_with_mpa_v1 = 0;
730 	}
731 
732 	if (mpa_rev_to_use == 2) {
733 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
734 					       sizeof (struct mpa_v2_conn_params));
735 		mpa_v2_params.ird = htons((u16)ep->ird);
736 		mpa_v2_params.ord = htons((u16)ep->ord);
737 
738 		if (peer2peer) {
739 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
740 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
741 				mpa_v2_params.ord |=
742 					htons(MPA_V2_RDMA_WRITE_RTR);
743 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
744 				mpa_v2_params.ord |=
745 					htons(MPA_V2_RDMA_READ_RTR);
746 		}
747 		memcpy(mpa->private_data, &mpa_v2_params,
748 		       sizeof(struct mpa_v2_conn_params));
749 
750 		if (ep->plen)
751 			memcpy(mpa->private_data +
752 			       sizeof(struct mpa_v2_conn_params),
753 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
754 	} else
755 		if (ep->plen)
756 			memcpy(mpa->private_data,
757 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
758 
759 	/*
760 	 * Reference the mpa skb.  This ensures the data area
761 	 * will remain in memory until the hw acks the tx.
762 	 * Function fw4_ack() will deref it.
763 	 */
764 	skb_get(skb);
765 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
766 	BUG_ON(ep->mpa_skb);
767 	ep->mpa_skb = skb;
768 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
769 	start_ep_timer(ep);
770 	__state_set(&ep->com, MPA_REQ_SENT);
771 	ep->mpa_attr.initiator = 1;
772 	ep->snd_seq += mpalen;
773 	return;
774 }
775 
776 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
777 {
778 	int mpalen, wrlen;
779 	struct fw_ofld_tx_data_wr *req;
780 	struct mpa_message *mpa;
781 	struct sk_buff *skb;
782 	struct mpa_v2_conn_params mpa_v2_params;
783 
784 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
785 
786 	mpalen = sizeof(*mpa) + plen;
787 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
788 		mpalen += sizeof(struct mpa_v2_conn_params);
789 	wrlen = roundup(mpalen + sizeof *req, 16);
790 
791 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
792 	if (!skb) {
793 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
794 		return -ENOMEM;
795 	}
796 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
797 
798 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
799 	memset(req, 0, wrlen);
800 	req->op_to_immdlen = cpu_to_be32(
801 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
802 		FW_WR_COMPL(1) |
803 		FW_WR_IMMDLEN(mpalen));
804 	req->flowid_len16 = cpu_to_be32(
805 		FW_WR_FLOWID(ep->hwtid) |
806 		FW_WR_LEN16(wrlen >> 4));
807 	req->plen = cpu_to_be32(mpalen);
808 	req->tunnel_to_proxy = cpu_to_be32(
809 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
810 		FW_OFLD_TX_DATA_WR_SHOVE(1));
811 
812 	mpa = (struct mpa_message *)(req + 1);
813 	memset(mpa, 0, sizeof(*mpa));
814 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
815 	mpa->flags = MPA_REJECT;
816 	mpa->revision = ep->mpa_attr.version;
817 	mpa->private_data_size = htons(plen);
818 
819 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
820 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
821 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
822 					       sizeof (struct mpa_v2_conn_params));
823 		mpa_v2_params.ird = htons(((u16)ep->ird) |
824 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
825 					   0));
826 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
827 					  (p2p_type ==
828 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
829 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
830 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
831 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
832 		memcpy(mpa->private_data, &mpa_v2_params,
833 		       sizeof(struct mpa_v2_conn_params));
834 
835 		if (ep->plen)
836 			memcpy(mpa->private_data +
837 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
838 	} else
839 		if (plen)
840 			memcpy(mpa->private_data, pdata, plen);
841 
842 	/*
843 	 * Reference the mpa skb again.  This ensures the data area
844 	 * will remain in memory until the hw acks the tx.
845 	 * Function fw4_ack() will deref it.
846 	 */
847 	skb_get(skb);
848 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
849 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
850 	BUG_ON(ep->mpa_skb);
851 	ep->mpa_skb = skb;
852 	ep->snd_seq += mpalen;
853 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
854 }
855 
856 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
857 {
858 	int mpalen, wrlen;
859 	struct fw_ofld_tx_data_wr *req;
860 	struct mpa_message *mpa;
861 	struct sk_buff *skb;
862 	struct mpa_v2_conn_params mpa_v2_params;
863 
864 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
865 
866 	mpalen = sizeof(*mpa) + plen;
867 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
868 		mpalen += sizeof(struct mpa_v2_conn_params);
869 	wrlen = roundup(mpalen + sizeof *req, 16);
870 
871 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
872 	if (!skb) {
873 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
874 		return -ENOMEM;
875 	}
876 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
877 
878 	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
879 	memset(req, 0, wrlen);
880 	req->op_to_immdlen = cpu_to_be32(
881 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
882 		FW_WR_COMPL(1) |
883 		FW_WR_IMMDLEN(mpalen));
884 	req->flowid_len16 = cpu_to_be32(
885 		FW_WR_FLOWID(ep->hwtid) |
886 		FW_WR_LEN16(wrlen >> 4));
887 	req->plen = cpu_to_be32(mpalen);
888 	req->tunnel_to_proxy = cpu_to_be32(
889 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
890 		FW_OFLD_TX_DATA_WR_SHOVE(1));
891 
892 	mpa = (struct mpa_message *)(req + 1);
893 	memset(mpa, 0, sizeof(*mpa));
894 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
895 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
896 		     (markers_enabled ? MPA_MARKERS : 0);
897 	mpa->revision = ep->mpa_attr.version;
898 	mpa->private_data_size = htons(plen);
899 
900 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
901 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
902 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
903 					       sizeof (struct mpa_v2_conn_params));
904 		mpa_v2_params.ird = htons((u16)ep->ird);
905 		mpa_v2_params.ord = htons((u16)ep->ord);
906 		if (peer2peer && (ep->mpa_attr.p2p_type !=
907 					FW_RI_INIT_P2PTYPE_DISABLED)) {
908 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
909 
910 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
911 				mpa_v2_params.ord |=
912 					htons(MPA_V2_RDMA_WRITE_RTR);
913 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
914 				mpa_v2_params.ord |=
915 					htons(MPA_V2_RDMA_READ_RTR);
916 		}
917 
918 		memcpy(mpa->private_data, &mpa_v2_params,
919 		       sizeof(struct mpa_v2_conn_params));
920 
921 		if (ep->plen)
922 			memcpy(mpa->private_data +
923 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
924 	} else
925 		if (plen)
926 			memcpy(mpa->private_data, pdata, plen);
927 
928 	/*
929 	 * Reference the mpa skb.  This ensures the data area
930 	 * will remain in memory until the hw acks the tx.
931 	 * Function fw4_ack() will deref it.
932 	 */
933 	skb_get(skb);
934 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
935 	ep->mpa_skb = skb;
936 	__state_set(&ep->com, MPA_REP_SENT);
937 	ep->snd_seq += mpalen;
938 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
939 }
940 
941 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
942 {
943 	struct c4iw_ep *ep;
944 	struct cpl_act_establish *req = cplhdr(skb);
945 	unsigned int tid = GET_TID(req);
946 	unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
947 	struct tid_info *t = dev->rdev.lldi.tids;
948 
949 	ep = lookup_atid(t, atid);
950 
951 	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
952 	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
953 
954 	mutex_lock(&ep->com.mutex);
955 	dst_confirm(ep->dst);
956 
957 	/* setup the hwtid for this connection */
958 	ep->hwtid = tid;
959 	cxgb4_insert_tid(t, ep, tid);
960 	insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
961 
962 	ep->snd_seq = be32_to_cpu(req->snd_isn);
963 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
964 
965 	set_emss(ep, ntohs(req->tcp_opt));
966 
967 	/* dealloc the atid */
968 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
969 	cxgb4_free_atid(t, atid);
970 	set_bit(ACT_ESTAB, &ep->com.history);
971 
972 	/* start MPA negotiation */
973 	send_flowc(ep, NULL);
974 	if (ep->retry_with_mpa_v1)
975 		send_mpa_req(ep, skb, 1);
976 	else
977 		send_mpa_req(ep, skb, mpa_rev);
978 	mutex_unlock(&ep->com.mutex);
979 	return 0;
980 }
981 
982 static void close_complete_upcall(struct c4iw_ep *ep, int status)
983 {
984 	struct iw_cm_event event;
985 
986 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
987 	memset(&event, 0, sizeof(event));
988 	event.event = IW_CM_EVENT_CLOSE;
989 	event.status = status;
990 	if (ep->com.cm_id) {
991 		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
992 		     ep, ep->com.cm_id, ep->hwtid);
993 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
994 		ep->com.cm_id->rem_ref(ep->com.cm_id);
995 		ep->com.cm_id = NULL;
996 		set_bit(CLOSE_UPCALL, &ep->com.history);
997 	}
998 }
999 
1000 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1001 {
1002 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1003 	__state_set(&ep->com, ABORTING);
1004 	set_bit(ABORT_CONN, &ep->com.history);
1005 	return send_abort(ep, skb, gfp);
1006 }
1007 
1008 static void peer_close_upcall(struct c4iw_ep *ep)
1009 {
1010 	struct iw_cm_event event;
1011 
1012 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1013 	memset(&event, 0, sizeof(event));
1014 	event.event = IW_CM_EVENT_DISCONNECT;
1015 	if (ep->com.cm_id) {
1016 		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1017 		     ep, ep->com.cm_id, ep->hwtid);
1018 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1019 		set_bit(DISCONN_UPCALL, &ep->com.history);
1020 	}
1021 }
1022 
1023 static void peer_abort_upcall(struct c4iw_ep *ep)
1024 {
1025 	struct iw_cm_event event;
1026 
1027 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1028 	memset(&event, 0, sizeof(event));
1029 	event.event = IW_CM_EVENT_CLOSE;
1030 	event.status = -ECONNRESET;
1031 	if (ep->com.cm_id) {
1032 		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1033 		     ep->com.cm_id, ep->hwtid);
1034 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1035 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1036 		ep->com.cm_id = NULL;
1037 		set_bit(ABORT_UPCALL, &ep->com.history);
1038 	}
1039 }
1040 
1041 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1042 {
1043 	struct iw_cm_event event;
1044 
1045 	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1046 	memset(&event, 0, sizeof(event));
1047 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1048 	event.status = status;
1049 	memcpy(&event.local_addr, &ep->com.local_addr,
1050 	       sizeof(ep->com.local_addr));
1051 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1052 	       sizeof(ep->com.remote_addr));
1053 
1054 	if ((status == 0) || (status == -ECONNREFUSED)) {
1055 		if (!ep->tried_with_mpa_v1) {
1056 			/* this means MPA_v2 is used */
1057 			event.private_data_len = ep->plen -
1058 				sizeof(struct mpa_v2_conn_params);
1059 			event.private_data = ep->mpa_pkt +
1060 				sizeof(struct mpa_message) +
1061 				sizeof(struct mpa_v2_conn_params);
1062 		} else {
1063 			/* this means MPA_v1 is used */
1064 			event.private_data_len = ep->plen;
1065 			event.private_data = ep->mpa_pkt +
1066 				sizeof(struct mpa_message);
1067 		}
1068 	}
1069 
1070 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1071 	     ep->hwtid, status);
1072 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1073 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1074 
1075 	if (status < 0) {
1076 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1077 		ep->com.cm_id = NULL;
1078 	}
1079 }
1080 
1081 static int connect_request_upcall(struct c4iw_ep *ep)
1082 {
1083 	struct iw_cm_event event;
1084 	int ret;
1085 
1086 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1087 	memset(&event, 0, sizeof(event));
1088 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1089 	memcpy(&event.local_addr, &ep->com.local_addr,
1090 	       sizeof(ep->com.local_addr));
1091 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1092 	       sizeof(ep->com.remote_addr));
1093 	event.provider_data = ep;
1094 	if (!ep->tried_with_mpa_v1) {
1095 		/* this means MPA_v2 is used */
1096 		event.ord = ep->ord;
1097 		event.ird = ep->ird;
1098 		event.private_data_len = ep->plen -
1099 			sizeof(struct mpa_v2_conn_params);
1100 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1101 			sizeof(struct mpa_v2_conn_params);
1102 	} else {
1103 		/* this means MPA_v1 is used. Send max supported */
1104 		event.ord = c4iw_max_read_depth;
1105 		event.ird = c4iw_max_read_depth;
1106 		event.private_data_len = ep->plen;
1107 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1108 	}
1109 	c4iw_get_ep(&ep->com);
1110 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1111 						      &event);
1112 	if (ret)
1113 		c4iw_put_ep(&ep->com);
1114 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1115 	c4iw_put_ep(&ep->parent_ep->com);
1116 	return ret;
1117 }
1118 
1119 static void established_upcall(struct c4iw_ep *ep)
1120 {
1121 	struct iw_cm_event event;
1122 
1123 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1124 	memset(&event, 0, sizeof(event));
1125 	event.event = IW_CM_EVENT_ESTABLISHED;
1126 	event.ird = ep->ird;
1127 	event.ord = ep->ord;
1128 	if (ep->com.cm_id) {
1129 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1130 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1131 		set_bit(ESTAB_UPCALL, &ep->com.history);
1132 	}
1133 }
1134 
1135 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1136 {
1137 	struct cpl_rx_data_ack *req;
1138 	struct sk_buff *skb;
1139 	int wrlen = roundup(sizeof *req, 16);
1140 
1141 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1142 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1143 	if (!skb) {
1144 		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1145 		return 0;
1146 	}
1147 
1148 	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1149 	memset(req, 0, wrlen);
1150 	INIT_TP_WR(req, ep->hwtid);
1151 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1152 						    ep->hwtid));
1153 	req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1154 				       F_RX_DACK_CHANGE |
1155 				       V_RX_DACK_MODE(dack_mode));
1156 	set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1157 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1158 	return credits;
1159 }
1160 
1161 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1162 {
1163 	struct mpa_message *mpa;
1164 	struct mpa_v2_conn_params *mpa_v2_params;
1165 	u16 plen;
1166 	u16 resp_ird, resp_ord;
1167 	u8 rtr_mismatch = 0, insuff_ird = 0;
1168 	struct c4iw_qp_attributes attrs;
1169 	enum c4iw_qp_attr_mask mask;
1170 	int err;
1171 	int disconnect = 0;
1172 
1173 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1174 
1175 	/*
1176 	 * Stop mpa timer.  If it expired, then
1177 	 * we ignore the MPA reply.  process_timeout()
1178 	 * will abort the connection.
1179 	 */
1180 	if (stop_ep_timer(ep))
1181 		return 0;
1182 
1183 	/*
1184 	 * If we get more than the supported amount of private data
1185 	 * then we must fail this connection.
1186 	 */
1187 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1188 		err = -EINVAL;
1189 		goto err;
1190 	}
1191 
1192 	/*
1193 	 * copy the new data into our accumulation buffer.
1194 	 */
1195 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1196 				  skb->len);
1197 	ep->mpa_pkt_len += skb->len;
1198 
1199 	/*
1200 	 * if we don't even have the mpa message, then bail.
1201 	 */
1202 	if (ep->mpa_pkt_len < sizeof(*mpa))
1203 		return 0;
1204 	mpa = (struct mpa_message *) ep->mpa_pkt;
1205 
1206 	/* Validate MPA header. */
1207 	if (mpa->revision > mpa_rev) {
1208 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1209 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1210 		err = -EPROTO;
1211 		goto err;
1212 	}
1213 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1214 		err = -EPROTO;
1215 		goto err;
1216 	}
1217 
1218 	plen = ntohs(mpa->private_data_size);
1219 
1220 	/*
1221 	 * Fail if there's too much private data.
1222 	 */
1223 	if (plen > MPA_MAX_PRIVATE_DATA) {
1224 		err = -EPROTO;
1225 		goto err;
1226 	}
1227 
1228 	/*
1229 	 * If plen does not account for pkt size
1230 	 */
1231 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1232 		err = -EPROTO;
1233 		goto err;
1234 	}
1235 
1236 	ep->plen = (u8) plen;
1237 
1238 	/*
1239 	 * If we don't have all the pdata yet, then bail.
1240 	 * We'll continue process when more data arrives.
1241 	 */
1242 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1243 		return 0;
1244 
1245 	if (mpa->flags & MPA_REJECT) {
1246 		err = -ECONNREFUSED;
1247 		goto err;
1248 	}
1249 
1250 	/*
1251 	 * If we get here we have accumulated the entire mpa
1252 	 * start reply message including private data. And
1253 	 * the MPA header is valid.
1254 	 */
1255 	__state_set(&ep->com, FPDU_MODE);
1256 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1257 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1258 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1259 	ep->mpa_attr.version = mpa->revision;
1260 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1261 
1262 	if (mpa->revision == 2) {
1263 		ep->mpa_attr.enhanced_rdma_conn =
1264 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1265 		if (ep->mpa_attr.enhanced_rdma_conn) {
1266 			mpa_v2_params = (struct mpa_v2_conn_params *)
1267 				(ep->mpa_pkt + sizeof(*mpa));
1268 			resp_ird = ntohs(mpa_v2_params->ird) &
1269 				MPA_V2_IRD_ORD_MASK;
1270 			resp_ord = ntohs(mpa_v2_params->ord) &
1271 				MPA_V2_IRD_ORD_MASK;
1272 
1273 			/*
1274 			 * This is a double-check. Ideally, below checks are
1275 			 * not required since ird/ord stuff has been taken
1276 			 * care of in c4iw_accept_cr
1277 			 */
1278 			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1279 				err = -ENOMEM;
1280 				ep->ird = resp_ord;
1281 				ep->ord = resp_ird;
1282 				insuff_ird = 1;
1283 			}
1284 
1285 			if (ntohs(mpa_v2_params->ird) &
1286 					MPA_V2_PEER2PEER_MODEL) {
1287 				if (ntohs(mpa_v2_params->ord) &
1288 						MPA_V2_RDMA_WRITE_RTR)
1289 					ep->mpa_attr.p2p_type =
1290 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1291 				else if (ntohs(mpa_v2_params->ord) &
1292 						MPA_V2_RDMA_READ_RTR)
1293 					ep->mpa_attr.p2p_type =
1294 						FW_RI_INIT_P2PTYPE_READ_REQ;
1295 			}
1296 		}
1297 	} else if (mpa->revision == 1)
1298 		if (peer2peer)
1299 			ep->mpa_attr.p2p_type = p2p_type;
1300 
1301 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1302 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1303 	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
1304 	     ep->mpa_attr.recv_marker_enabled,
1305 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1306 	     ep->mpa_attr.p2p_type, p2p_type);
1307 
1308 	/*
1309 	 * If responder's RTR does not match with that of initiator, assign
1310 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1311 	 * generated when moving QP to RTS state.
1312 	 * A TERM message will be sent after QP has moved to RTS state
1313 	 */
1314 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1315 			(ep->mpa_attr.p2p_type != p2p_type)) {
1316 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1317 		rtr_mismatch = 1;
1318 	}
1319 
1320 	attrs.mpa_attr = ep->mpa_attr;
1321 	attrs.max_ird = ep->ird;
1322 	attrs.max_ord = ep->ord;
1323 	attrs.llp_stream_handle = ep;
1324 	attrs.next_state = C4IW_QP_STATE_RTS;
1325 
1326 	mask = C4IW_QP_ATTR_NEXT_STATE |
1327 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1328 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1329 
1330 	/* bind QP and TID with INIT_WR */
1331 	err = c4iw_modify_qp(ep->com.qp->rhp,
1332 			     ep->com.qp, mask, &attrs, 1);
1333 	if (err)
1334 		goto err;
1335 
1336 	/*
1337 	 * If responder's RTR requirement did not match with what initiator
1338 	 * supports, generate TERM message
1339 	 */
1340 	if (rtr_mismatch) {
1341 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1342 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1343 		attrs.ecode = MPA_NOMATCH_RTR;
1344 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1345 		attrs.send_term = 1;
1346 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1347 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1348 		err = -ENOMEM;
1349 		disconnect = 1;
1350 		goto out;
1351 	}
1352 
1353 	/*
1354 	 * Generate TERM if initiator IRD is not sufficient for responder
1355 	 * provided ORD. Currently, we do the same behaviour even when
1356 	 * responder provided IRD is also not sufficient as regards to
1357 	 * initiator ORD.
1358 	 */
1359 	if (insuff_ird) {
1360 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1361 				__func__);
1362 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1363 		attrs.ecode = MPA_INSUFF_IRD;
1364 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1365 		attrs.send_term = 1;
1366 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1367 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1368 		err = -ENOMEM;
1369 		disconnect = 1;
1370 		goto out;
1371 	}
1372 	goto out;
1373 err:
1374 	__state_set(&ep->com, ABORTING);
1375 	send_abort(ep, skb, GFP_KERNEL);
1376 out:
1377 	connect_reply_upcall(ep, err);
1378 	return disconnect;
1379 }
1380 
1381 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1382 {
1383 	struct mpa_message *mpa;
1384 	struct mpa_v2_conn_params *mpa_v2_params;
1385 	u16 plen;
1386 
1387 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1388 
1389 	/*
1390 	 * If we get more than the supported amount of private data
1391 	 * then we must fail this connection.
1392 	 */
1393 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1394 		(void)stop_ep_timer(ep);
1395 		abort_connection(ep, skb, GFP_KERNEL);
1396 		return;
1397 	}
1398 
1399 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1400 
1401 	/*
1402 	 * Copy the new data into our accumulation buffer.
1403 	 */
1404 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1405 				  skb->len);
1406 	ep->mpa_pkt_len += skb->len;
1407 
1408 	/*
1409 	 * If we don't even have the mpa message, then bail.
1410 	 * We'll continue process when more data arrives.
1411 	 */
1412 	if (ep->mpa_pkt_len < sizeof(*mpa))
1413 		return;
1414 
1415 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1416 	mpa = (struct mpa_message *) ep->mpa_pkt;
1417 
1418 	/*
1419 	 * Validate MPA Header.
1420 	 */
1421 	if (mpa->revision > mpa_rev) {
1422 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1423 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1424 		(void)stop_ep_timer(ep);
1425 		abort_connection(ep, skb, GFP_KERNEL);
1426 		return;
1427 	}
1428 
1429 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1430 		(void)stop_ep_timer(ep);
1431 		abort_connection(ep, skb, GFP_KERNEL);
1432 		return;
1433 	}
1434 
1435 	plen = ntohs(mpa->private_data_size);
1436 
1437 	/*
1438 	 * Fail if there's too much private data.
1439 	 */
1440 	if (plen > MPA_MAX_PRIVATE_DATA) {
1441 		(void)stop_ep_timer(ep);
1442 		abort_connection(ep, skb, GFP_KERNEL);
1443 		return;
1444 	}
1445 
1446 	/*
1447 	 * If plen does not account for pkt size
1448 	 */
1449 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1450 		(void)stop_ep_timer(ep);
1451 		abort_connection(ep, skb, GFP_KERNEL);
1452 		return;
1453 	}
1454 	ep->plen = (u8) plen;
1455 
1456 	/*
1457 	 * If we don't have all the pdata yet, then bail.
1458 	 */
1459 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1460 		return;
1461 
1462 	/*
1463 	 * If we get here we have accumulated the entire mpa
1464 	 * start reply message including private data.
1465 	 */
1466 	ep->mpa_attr.initiator = 0;
1467 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1468 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1469 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1470 	ep->mpa_attr.version = mpa->revision;
1471 	if (mpa->revision == 1)
1472 		ep->tried_with_mpa_v1 = 1;
1473 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1474 
1475 	if (mpa->revision == 2) {
1476 		ep->mpa_attr.enhanced_rdma_conn =
1477 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1478 		if (ep->mpa_attr.enhanced_rdma_conn) {
1479 			mpa_v2_params = (struct mpa_v2_conn_params *)
1480 				(ep->mpa_pkt + sizeof(*mpa));
1481 			ep->ird = ntohs(mpa_v2_params->ird) &
1482 				MPA_V2_IRD_ORD_MASK;
1483 			ep->ord = ntohs(mpa_v2_params->ord) &
1484 				MPA_V2_IRD_ORD_MASK;
1485 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1486 				if (peer2peer) {
1487 					if (ntohs(mpa_v2_params->ord) &
1488 							MPA_V2_RDMA_WRITE_RTR)
1489 						ep->mpa_attr.p2p_type =
1490 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1491 					else if (ntohs(mpa_v2_params->ord) &
1492 							MPA_V2_RDMA_READ_RTR)
1493 						ep->mpa_attr.p2p_type =
1494 						FW_RI_INIT_P2PTYPE_READ_REQ;
1495 				}
1496 		}
1497 	} else if (mpa->revision == 1)
1498 		if (peer2peer)
1499 			ep->mpa_attr.p2p_type = p2p_type;
1500 
1501 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1502 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1503 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1504 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1505 	     ep->mpa_attr.p2p_type);
1506 
1507 	/*
1508 	 * If the endpoint timer already expired, then we ignore
1509 	 * the start request.  process_timeout() will abort
1510 	 * the connection.
1511 	 */
1512 	if (!stop_ep_timer(ep)) {
1513 		__state_set(&ep->com, MPA_REQ_RCVD);
1514 
1515 		/* drive upcall */
1516 		mutex_lock(&ep->parent_ep->com.mutex);
1517 		if (ep->parent_ep->com.state != DEAD) {
1518 			if (connect_request_upcall(ep))
1519 				abort_connection(ep, skb, GFP_KERNEL);
1520 		} else {
1521 			abort_connection(ep, skb, GFP_KERNEL);
1522 		}
1523 		mutex_unlock(&ep->parent_ep->com.mutex);
1524 	}
1525 	return;
1526 }
1527 
1528 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1529 {
1530 	struct c4iw_ep *ep;
1531 	struct cpl_rx_data *hdr = cplhdr(skb);
1532 	unsigned int dlen = ntohs(hdr->len);
1533 	unsigned int tid = GET_TID(hdr);
1534 	struct tid_info *t = dev->rdev.lldi.tids;
1535 	__u8 status = hdr->status;
1536 	int disconnect = 0;
1537 
1538 	ep = lookup_tid(t, tid);
1539 	if (!ep)
1540 		return 0;
1541 	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1542 	skb_pull(skb, sizeof(*hdr));
1543 	skb_trim(skb, dlen);
1544 	mutex_lock(&ep->com.mutex);
1545 
1546 	/* update RX credits */
1547 	update_rx_credits(ep, dlen);
1548 
1549 	switch (ep->com.state) {
1550 	case MPA_REQ_SENT:
1551 		ep->rcv_seq += dlen;
1552 		disconnect = process_mpa_reply(ep, skb);
1553 		break;
1554 	case MPA_REQ_WAIT:
1555 		ep->rcv_seq += dlen;
1556 		process_mpa_request(ep, skb);
1557 		break;
1558 	case FPDU_MODE: {
1559 		struct c4iw_qp_attributes attrs;
1560 		BUG_ON(!ep->com.qp);
1561 		if (status)
1562 			pr_err("%s Unexpected streaming data." \
1563 			       " qpid %u ep %p state %d tid %u status %d\n",
1564 			       __func__, ep->com.qp->wq.sq.qid, ep,
1565 			       ep->com.state, ep->hwtid, status);
1566 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1567 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1568 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1569 		disconnect = 1;
1570 		break;
1571 	}
1572 	default:
1573 		break;
1574 	}
1575 	mutex_unlock(&ep->com.mutex);
1576 	if (disconnect)
1577 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1578 	return 0;
1579 }
1580 
1581 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1582 {
1583 	struct c4iw_ep *ep;
1584 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1585 	int release = 0;
1586 	unsigned int tid = GET_TID(rpl);
1587 	struct tid_info *t = dev->rdev.lldi.tids;
1588 
1589 	ep = lookup_tid(t, tid);
1590 	if (!ep) {
1591 		printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1592 		return 0;
1593 	}
1594 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1595 	mutex_lock(&ep->com.mutex);
1596 	switch (ep->com.state) {
1597 	case ABORTING:
1598 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1599 		__state_set(&ep->com, DEAD);
1600 		release = 1;
1601 		break;
1602 	default:
1603 		printk(KERN_ERR "%s ep %p state %d\n",
1604 		     __func__, ep, ep->com.state);
1605 		break;
1606 	}
1607 	mutex_unlock(&ep->com.mutex);
1608 
1609 	if (release)
1610 		release_ep_resources(ep);
1611 	return 0;
1612 }
1613 
1614 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1615 {
1616 	struct sk_buff *skb;
1617 	struct fw_ofld_connection_wr *req;
1618 	unsigned int mtu_idx;
1619 	int wscale;
1620 	struct sockaddr_in *sin;
1621 
1622 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1623 	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1624 	memset(req, 0, sizeof(*req));
1625 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1626 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1627 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1628 				     ep->com.dev->rdev.lldi.ports[0],
1629 				     ep->l2t));
1630 	sin = (struct sockaddr_in *)&ep->com.local_addr;
1631 	req->le.lport = sin->sin_port;
1632 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1633 	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1634 	req->le.pport = sin->sin_port;
1635 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1636 	req->tcb.t_state_to_astid =
1637 			htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1638 			V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1639 	req->tcb.cplrxdataack_cplpassacceptrpl =
1640 			htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1641 	req->tcb.tx_max = (__force __be32) jiffies;
1642 	req->tcb.rcv_adv = htons(1);
1643 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1644 	wscale = compute_wscale(rcv_win);
1645 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1646 		(nocong ? NO_CONG(1) : 0) |
1647 		KEEP_ALIVE(1) |
1648 		DELACK(1) |
1649 		WND_SCALE(wscale) |
1650 		MSS_IDX(mtu_idx) |
1651 		L2T_IDX(ep->l2t->idx) |
1652 		TX_CHAN(ep->tx_chan) |
1653 		SMAC_SEL(ep->smac_idx) |
1654 		DSCP(ep->tos) |
1655 		ULP_MODE(ULP_MODE_TCPDDP) |
1656 		RCV_BUFSIZ(rcv_win >> 10));
1657 	req->tcb.opt2 = (__force __be32) (PACE(1) |
1658 		TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1659 		RX_CHANNEL(0) |
1660 		CCTRL_ECN(enable_ecn) |
1661 		RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1662 	if (enable_tcp_timestamps)
1663 		req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1664 	if (enable_tcp_sack)
1665 		req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1666 	if (wscale && enable_tcp_window_scaling)
1667 		req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1668 	req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1669 	req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1670 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1671 	set_bit(ACT_OFLD_CONN, &ep->com.history);
1672 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1673 }
1674 
1675 /*
1676  * Return whether a failed active open has allocated a TID
1677  */
1678 static inline int act_open_has_tid(int status)
1679 {
1680 	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1681 	       status != CPL_ERR_ARP_MISS;
1682 }
1683 
1684 /* Returns whether a CPL status conveys negative advice.
1685  */
1686 static int is_neg_adv(unsigned int status)
1687 {
1688 	return status == CPL_ERR_RTX_NEG_ADVICE ||
1689 	       status == CPL_ERR_PERSIST_NEG_ADVICE ||
1690 	       status == CPL_ERR_KEEPALV_NEG_ADVICE;
1691 }
1692 
1693 #define ACT_OPEN_RETRY_COUNT 2
1694 
1695 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1696 		     struct dst_entry *dst, struct c4iw_dev *cdev,
1697 		     bool clear_mpa_v1)
1698 {
1699 	struct neighbour *n;
1700 	int err, step;
1701 	struct net_device *pdev;
1702 
1703 	n = dst_neigh_lookup(dst, peer_ip);
1704 	if (!n)
1705 		return -ENODEV;
1706 
1707 	rcu_read_lock();
1708 	err = -ENOMEM;
1709 	if (n->dev->flags & IFF_LOOPBACK) {
1710 		if (iptype == 4)
1711 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1712 		else if (IS_ENABLED(CONFIG_IPV6))
1713 			for_each_netdev(&init_net, pdev) {
1714 				if (ipv6_chk_addr(&init_net,
1715 						  (struct in6_addr *)peer_ip,
1716 						  pdev, 1))
1717 					break;
1718 			}
1719 		else
1720 			pdev = NULL;
1721 
1722 		if (!pdev) {
1723 			err = -ENODEV;
1724 			goto out;
1725 		}
1726 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1727 					n, pdev, 0);
1728 		if (!ep->l2t)
1729 			goto out;
1730 		ep->mtu = pdev->mtu;
1731 		ep->tx_chan = cxgb4_port_chan(pdev);
1732 		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1733 		step = cdev->rdev.lldi.ntxq /
1734 			cdev->rdev.lldi.nchan;
1735 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1736 		step = cdev->rdev.lldi.nrxq /
1737 			cdev->rdev.lldi.nchan;
1738 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1739 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1740 			cxgb4_port_idx(pdev) * step];
1741 		dev_put(pdev);
1742 	} else {
1743 		pdev = get_real_dev(n->dev);
1744 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1745 					n, pdev, 0);
1746 		if (!ep->l2t)
1747 			goto out;
1748 		ep->mtu = dst_mtu(dst);
1749 		ep->tx_chan = cxgb4_port_chan(n->dev);
1750 		ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1751 		step = cdev->rdev.lldi.ntxq /
1752 			cdev->rdev.lldi.nchan;
1753 		ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1754 		ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1755 		step = cdev->rdev.lldi.nrxq /
1756 			cdev->rdev.lldi.nchan;
1757 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1758 			cxgb4_port_idx(n->dev) * step];
1759 
1760 		if (clear_mpa_v1) {
1761 			ep->retry_with_mpa_v1 = 0;
1762 			ep->tried_with_mpa_v1 = 0;
1763 		}
1764 	}
1765 	err = 0;
1766 out:
1767 	rcu_read_unlock();
1768 
1769 	neigh_release(n);
1770 
1771 	return err;
1772 }
1773 
1774 static int c4iw_reconnect(struct c4iw_ep *ep)
1775 {
1776 	int err = 0;
1777 	struct sockaddr_in *laddr = (struct sockaddr_in *)
1778 				    &ep->com.cm_id->local_addr;
1779 	struct sockaddr_in *raddr = (struct sockaddr_in *)
1780 				    &ep->com.cm_id->remote_addr;
1781 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1782 				      &ep->com.cm_id->local_addr;
1783 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1784 				      &ep->com.cm_id->remote_addr;
1785 	int iptype;
1786 	__u8 *ra;
1787 
1788 	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1789 	init_timer(&ep->timer);
1790 
1791 	/*
1792 	 * Allocate an active TID to initiate a TCP connection.
1793 	 */
1794 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1795 	if (ep->atid == -1) {
1796 		pr_err("%s - cannot alloc atid.\n", __func__);
1797 		err = -ENOMEM;
1798 		goto fail2;
1799 	}
1800 	insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1801 
1802 	/* find a route */
1803 	if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1804 		ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1805 				     raddr->sin_addr.s_addr, laddr->sin_port,
1806 				     raddr->sin_port, 0);
1807 		iptype = 4;
1808 		ra = (__u8 *)&raddr->sin_addr;
1809 	} else {
1810 		ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1811 				      raddr6->sin6_addr.s6_addr,
1812 				      laddr6->sin6_port, raddr6->sin6_port, 0,
1813 				      raddr6->sin6_scope_id);
1814 		iptype = 6;
1815 		ra = (__u8 *)&raddr6->sin6_addr;
1816 	}
1817 	if (!ep->dst) {
1818 		pr_err("%s - cannot find route.\n", __func__);
1819 		err = -EHOSTUNREACH;
1820 		goto fail3;
1821 	}
1822 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1823 	if (err) {
1824 		pr_err("%s - cannot alloc l2e.\n", __func__);
1825 		goto fail4;
1826 	}
1827 
1828 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1829 	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1830 	     ep->l2t->idx);
1831 
1832 	state_set(&ep->com, CONNECTING);
1833 	ep->tos = 0;
1834 
1835 	/* send connect request to rnic */
1836 	err = send_connect(ep);
1837 	if (!err)
1838 		goto out;
1839 
1840 	cxgb4_l2t_release(ep->l2t);
1841 fail4:
1842 	dst_release(ep->dst);
1843 fail3:
1844 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1845 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1846 fail2:
1847 	/*
1848 	 * remember to send notification to upper layer.
1849 	 * We are in here so the upper layer is not aware that this is
1850 	 * re-connect attempt and so, upper layer is still waiting for
1851 	 * response of 1st connect request.
1852 	 */
1853 	connect_reply_upcall(ep, -ECONNRESET);
1854 	c4iw_put_ep(&ep->com);
1855 out:
1856 	return err;
1857 }
1858 
1859 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1860 {
1861 	struct c4iw_ep *ep;
1862 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
1863 	unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1864 					ntohl(rpl->atid_status)));
1865 	struct tid_info *t = dev->rdev.lldi.tids;
1866 	int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1867 	struct sockaddr_in *la;
1868 	struct sockaddr_in *ra;
1869 	struct sockaddr_in6 *la6;
1870 	struct sockaddr_in6 *ra6;
1871 
1872 	ep = lookup_atid(t, atid);
1873 	la = (struct sockaddr_in *)&ep->com.local_addr;
1874 	ra = (struct sockaddr_in *)&ep->com.remote_addr;
1875 	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1876 	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1877 
1878 	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1879 	     status, status2errno(status));
1880 
1881 	if (is_neg_adv(status)) {
1882 		printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1883 			atid);
1884 		return 0;
1885 	}
1886 
1887 	set_bit(ACT_OPEN_RPL, &ep->com.history);
1888 
1889 	/*
1890 	 * Log interesting failures.
1891 	 */
1892 	switch (status) {
1893 	case CPL_ERR_CONN_RESET:
1894 	case CPL_ERR_CONN_TIMEDOUT:
1895 		break;
1896 	case CPL_ERR_TCAM_FULL:
1897 		mutex_lock(&dev->rdev.stats.lock);
1898 		dev->rdev.stats.tcam_full++;
1899 		mutex_unlock(&dev->rdev.stats.lock);
1900 		if (ep->com.local_addr.ss_family == AF_INET &&
1901 		    dev->rdev.lldi.enable_fw_ofld_conn) {
1902 			send_fw_act_open_req(ep,
1903 					     GET_TID_TID(GET_AOPEN_ATID(
1904 					     ntohl(rpl->atid_status))));
1905 			return 0;
1906 		}
1907 		break;
1908 	case CPL_ERR_CONN_EXIST:
1909 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1910 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
1911 			remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1912 					atid);
1913 			cxgb4_free_atid(t, atid);
1914 			dst_release(ep->dst);
1915 			cxgb4_l2t_release(ep->l2t);
1916 			c4iw_reconnect(ep);
1917 			return 0;
1918 		}
1919 		break;
1920 	default:
1921 		if (ep->com.local_addr.ss_family == AF_INET) {
1922 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1923 				atid, status, status2errno(status),
1924 				&la->sin_addr.s_addr, ntohs(la->sin_port),
1925 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
1926 		} else {
1927 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1928 				atid, status, status2errno(status),
1929 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1930 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1931 		}
1932 		break;
1933 	}
1934 
1935 	connect_reply_upcall(ep, status2errno(status));
1936 	state_set(&ep->com, DEAD);
1937 
1938 	if (status && act_open_has_tid(status))
1939 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1940 
1941 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1942 	cxgb4_free_atid(t, atid);
1943 	dst_release(ep->dst);
1944 	cxgb4_l2t_release(ep->l2t);
1945 	c4iw_put_ep(&ep->com);
1946 
1947 	return 0;
1948 }
1949 
1950 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1951 {
1952 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1953 	struct tid_info *t = dev->rdev.lldi.tids;
1954 	unsigned int stid = GET_TID(rpl);
1955 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1956 
1957 	if (!ep) {
1958 		PDBG("%s stid %d lookup failure!\n", __func__, stid);
1959 		goto out;
1960 	}
1961 	PDBG("%s ep %p status %d error %d\n", __func__, ep,
1962 	     rpl->status, status2errno(rpl->status));
1963 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1964 
1965 out:
1966 	return 0;
1967 }
1968 
1969 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1970 {
1971 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1972 	struct tid_info *t = dev->rdev.lldi.tids;
1973 	unsigned int stid = GET_TID(rpl);
1974 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1975 
1976 	PDBG("%s ep %p\n", __func__, ep);
1977 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1978 	return 0;
1979 }
1980 
1981 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1982 		      struct cpl_pass_accept_req *req)
1983 {
1984 	struct cpl_pass_accept_rpl *rpl;
1985 	unsigned int mtu_idx;
1986 	u64 opt0;
1987 	u32 opt2;
1988 	int wscale;
1989 
1990 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1991 	BUG_ON(skb_cloned(skb));
1992 	skb_trim(skb, sizeof(*rpl));
1993 	skb_get(skb);
1994 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1995 	wscale = compute_wscale(rcv_win);
1996 	opt0 = (nocong ? NO_CONG(1) : 0) |
1997 	       KEEP_ALIVE(1) |
1998 	       DELACK(1) |
1999 	       WND_SCALE(wscale) |
2000 	       MSS_IDX(mtu_idx) |
2001 	       L2T_IDX(ep->l2t->idx) |
2002 	       TX_CHAN(ep->tx_chan) |
2003 	       SMAC_SEL(ep->smac_idx) |
2004 	       DSCP(ep->tos >> 2) |
2005 	       ULP_MODE(ULP_MODE_TCPDDP) |
2006 	       RCV_BUFSIZ(rcv_win>>10);
2007 	opt2 = RX_CHANNEL(0) |
2008 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
2009 
2010 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2011 		opt2 |= TSTAMPS_EN(1);
2012 	if (enable_tcp_sack && req->tcpopt.sack)
2013 		opt2 |= SACK_EN(1);
2014 	if (wscale && enable_tcp_window_scaling)
2015 		opt2 |= WND_SCALE_EN(1);
2016 	if (enable_ecn) {
2017 		const struct tcphdr *tcph;
2018 		u32 hlen = ntohl(req->hdr_len);
2019 
2020 		tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2021 			G_IP_HDR_LEN(hlen);
2022 		if (tcph->ece && tcph->cwr)
2023 			opt2 |= CCTRL_ECN(1);
2024 	}
2025 	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
2026 		opt2 |= T5_OPT_2_VALID;
2027 		opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
2028 	}
2029 
2030 	rpl = cplhdr(skb);
2031 	INIT_TP_WR(rpl, ep->hwtid);
2032 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2033 				      ep->hwtid));
2034 	rpl->opt0 = cpu_to_be64(opt0);
2035 	rpl->opt2 = cpu_to_be32(opt2);
2036 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2037 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
2038 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2039 
2040 	return;
2041 }
2042 
2043 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2044 {
2045 	PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2046 	BUG_ON(skb_cloned(skb));
2047 	skb_trim(skb, sizeof(struct cpl_tid_release));
2048 	skb_get(skb);
2049 	release_tid(&dev->rdev, hwtid, skb);
2050 	return;
2051 }
2052 
2053 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2054 		       __u8 *local_ip, __u8 *peer_ip,
2055 		       __be16 *local_port, __be16 *peer_port)
2056 {
2057 	int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2058 	int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2059 	struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2060 	struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2061 	struct tcphdr *tcp = (struct tcphdr *)
2062 			     ((u8 *)(req + 1) + eth_len + ip_len);
2063 
2064 	if (ip->version == 4) {
2065 		PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2066 		     ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2067 		     ntohs(tcp->dest));
2068 		*iptype = 4;
2069 		memcpy(peer_ip, &ip->saddr, 4);
2070 		memcpy(local_ip, &ip->daddr, 4);
2071 	} else {
2072 		PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2073 		     ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2074 		     ntohs(tcp->dest));
2075 		*iptype = 6;
2076 		memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2077 		memcpy(local_ip, ip6->daddr.s6_addr, 16);
2078 	}
2079 	*peer_port = tcp->source;
2080 	*local_port = tcp->dest;
2081 
2082 	return;
2083 }
2084 
2085 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2086 {
2087 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2088 	struct cpl_pass_accept_req *req = cplhdr(skb);
2089 	unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2090 	struct tid_info *t = dev->rdev.lldi.tids;
2091 	unsigned int hwtid = GET_TID(req);
2092 	struct dst_entry *dst;
2093 	__u8 local_ip[16], peer_ip[16];
2094 	__be16 local_port, peer_port;
2095 	int err;
2096 	u16 peer_mss = ntohs(req->tcpopt.mss);
2097 	int iptype;
2098 
2099 	parent_ep = lookup_stid(t, stid);
2100 	if (!parent_ep) {
2101 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2102 		goto reject;
2103 	}
2104 
2105 	if (state_read(&parent_ep->com) != LISTEN) {
2106 		printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2107 		       __func__);
2108 		goto reject;
2109 	}
2110 
2111 	get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2112 
2113 	/* Find output route */
2114 	if (iptype == 4)  {
2115 		PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2116 		     , __func__, parent_ep, hwtid,
2117 		     local_ip, peer_ip, ntohs(local_port),
2118 		     ntohs(peer_port), peer_mss);
2119 		dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2120 				 local_port, peer_port,
2121 				 GET_POPEN_TOS(ntohl(req->tos_stid)));
2122 	} else {
2123 		PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2124 		     , __func__, parent_ep, hwtid,
2125 		     local_ip, peer_ip, ntohs(local_port),
2126 		     ntohs(peer_port), peer_mss);
2127 		dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2128 				  PASS_OPEN_TOS(ntohl(req->tos_stid)),
2129 				  ((struct sockaddr_in6 *)
2130 				  &parent_ep->com.local_addr)->sin6_scope_id);
2131 	}
2132 	if (!dst) {
2133 		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2134 		       __func__);
2135 		goto reject;
2136 	}
2137 
2138 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2139 	if (!child_ep) {
2140 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2141 		       __func__);
2142 		dst_release(dst);
2143 		goto reject;
2144 	}
2145 
2146 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2147 	if (err) {
2148 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2149 		       __func__);
2150 		dst_release(dst);
2151 		kfree(child_ep);
2152 		goto reject;
2153 	}
2154 
2155 	if (peer_mss && child_ep->mtu > (peer_mss + 40))
2156 		child_ep->mtu = peer_mss + 40;
2157 
2158 	state_set(&child_ep->com, CONNECTING);
2159 	child_ep->com.dev = dev;
2160 	child_ep->com.cm_id = NULL;
2161 	if (iptype == 4) {
2162 		struct sockaddr_in *sin = (struct sockaddr_in *)
2163 			&child_ep->com.local_addr;
2164 		sin->sin_family = PF_INET;
2165 		sin->sin_port = local_port;
2166 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2167 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2168 		sin->sin_family = PF_INET;
2169 		sin->sin_port = peer_port;
2170 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2171 	} else {
2172 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2173 			&child_ep->com.local_addr;
2174 		sin6->sin6_family = PF_INET6;
2175 		sin6->sin6_port = local_port;
2176 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2177 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2178 		sin6->sin6_family = PF_INET6;
2179 		sin6->sin6_port = peer_port;
2180 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2181 	}
2182 	c4iw_get_ep(&parent_ep->com);
2183 	child_ep->parent_ep = parent_ep;
2184 	child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2185 	child_ep->dst = dst;
2186 	child_ep->hwtid = hwtid;
2187 
2188 	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2189 	     child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2190 
2191 	init_timer(&child_ep->timer);
2192 	cxgb4_insert_tid(t, child_ep, hwtid);
2193 	insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2194 	accept_cr(child_ep, skb, req);
2195 	set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2196 	goto out;
2197 reject:
2198 	reject_cr(dev, hwtid, skb);
2199 out:
2200 	return 0;
2201 }
2202 
2203 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2204 {
2205 	struct c4iw_ep *ep;
2206 	struct cpl_pass_establish *req = cplhdr(skb);
2207 	struct tid_info *t = dev->rdev.lldi.tids;
2208 	unsigned int tid = GET_TID(req);
2209 
2210 	ep = lookup_tid(t, tid);
2211 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2212 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2213 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2214 
2215 	PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2216 	     ntohs(req->tcp_opt));
2217 
2218 	set_emss(ep, ntohs(req->tcp_opt));
2219 
2220 	dst_confirm(ep->dst);
2221 	state_set(&ep->com, MPA_REQ_WAIT);
2222 	start_ep_timer(ep);
2223 	send_flowc(ep, skb);
2224 	set_bit(PASS_ESTAB, &ep->com.history);
2225 
2226 	return 0;
2227 }
2228 
2229 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2230 {
2231 	struct cpl_peer_close *hdr = cplhdr(skb);
2232 	struct c4iw_ep *ep;
2233 	struct c4iw_qp_attributes attrs;
2234 	int disconnect = 1;
2235 	int release = 0;
2236 	struct tid_info *t = dev->rdev.lldi.tids;
2237 	unsigned int tid = GET_TID(hdr);
2238 	int ret;
2239 
2240 	ep = lookup_tid(t, tid);
2241 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2242 	dst_confirm(ep->dst);
2243 
2244 	set_bit(PEER_CLOSE, &ep->com.history);
2245 	mutex_lock(&ep->com.mutex);
2246 	switch (ep->com.state) {
2247 	case MPA_REQ_WAIT:
2248 		__state_set(&ep->com, CLOSING);
2249 		break;
2250 	case MPA_REQ_SENT:
2251 		__state_set(&ep->com, CLOSING);
2252 		connect_reply_upcall(ep, -ECONNRESET);
2253 		break;
2254 	case MPA_REQ_RCVD:
2255 
2256 		/*
2257 		 * We're gonna mark this puppy DEAD, but keep
2258 		 * the reference on it until the ULP accepts or
2259 		 * rejects the CR. Also wake up anyone waiting
2260 		 * in rdma connection migration (see c4iw_accept_cr()).
2261 		 */
2262 		__state_set(&ep->com, CLOSING);
2263 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2264 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2265 		break;
2266 	case MPA_REP_SENT:
2267 		__state_set(&ep->com, CLOSING);
2268 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2269 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2270 		break;
2271 	case FPDU_MODE:
2272 		start_ep_timer(ep);
2273 		__state_set(&ep->com, CLOSING);
2274 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2275 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2276 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2277 		if (ret != -ECONNRESET) {
2278 			peer_close_upcall(ep);
2279 			disconnect = 1;
2280 		}
2281 		break;
2282 	case ABORTING:
2283 		disconnect = 0;
2284 		break;
2285 	case CLOSING:
2286 		__state_set(&ep->com, MORIBUND);
2287 		disconnect = 0;
2288 		break;
2289 	case MORIBUND:
2290 		(void)stop_ep_timer(ep);
2291 		if (ep->com.cm_id && ep->com.qp) {
2292 			attrs.next_state = C4IW_QP_STATE_IDLE;
2293 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2294 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2295 		}
2296 		close_complete_upcall(ep, 0);
2297 		__state_set(&ep->com, DEAD);
2298 		release = 1;
2299 		disconnect = 0;
2300 		break;
2301 	case DEAD:
2302 		disconnect = 0;
2303 		break;
2304 	default:
2305 		BUG_ON(1);
2306 	}
2307 	mutex_unlock(&ep->com.mutex);
2308 	if (disconnect)
2309 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2310 	if (release)
2311 		release_ep_resources(ep);
2312 	return 0;
2313 }
2314 
2315 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2316 {
2317 	struct cpl_abort_req_rss *req = cplhdr(skb);
2318 	struct c4iw_ep *ep;
2319 	struct cpl_abort_rpl *rpl;
2320 	struct sk_buff *rpl_skb;
2321 	struct c4iw_qp_attributes attrs;
2322 	int ret;
2323 	int release = 0;
2324 	struct tid_info *t = dev->rdev.lldi.tids;
2325 	unsigned int tid = GET_TID(req);
2326 
2327 	ep = lookup_tid(t, tid);
2328 	if (is_neg_adv(req->status)) {
2329 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2330 		     ep->hwtid);
2331 		return 0;
2332 	}
2333 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2334 	     ep->com.state);
2335 	set_bit(PEER_ABORT, &ep->com.history);
2336 
2337 	/*
2338 	 * Wake up any threads in rdma_init() or rdma_fini().
2339 	 * However, this is not needed if com state is just
2340 	 * MPA_REQ_SENT
2341 	 */
2342 	if (ep->com.state != MPA_REQ_SENT)
2343 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2344 
2345 	mutex_lock(&ep->com.mutex);
2346 	switch (ep->com.state) {
2347 	case CONNECTING:
2348 		break;
2349 	case MPA_REQ_WAIT:
2350 		(void)stop_ep_timer(ep);
2351 		break;
2352 	case MPA_REQ_SENT:
2353 		(void)stop_ep_timer(ep);
2354 		if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2355 			connect_reply_upcall(ep, -ECONNRESET);
2356 		else {
2357 			/*
2358 			 * we just don't send notification upwards because we
2359 			 * want to retry with mpa_v1 without upper layers even
2360 			 * knowing it.
2361 			 *
2362 			 * do some housekeeping so as to re-initiate the
2363 			 * connection
2364 			 */
2365 			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2366 			     mpa_rev);
2367 			ep->retry_with_mpa_v1 = 1;
2368 		}
2369 		break;
2370 	case MPA_REP_SENT:
2371 		break;
2372 	case MPA_REQ_RCVD:
2373 		break;
2374 	case MORIBUND:
2375 	case CLOSING:
2376 		stop_ep_timer(ep);
2377 		/*FALLTHROUGH*/
2378 	case FPDU_MODE:
2379 		if (ep->com.cm_id && ep->com.qp) {
2380 			attrs.next_state = C4IW_QP_STATE_ERROR;
2381 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2382 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2383 				     &attrs, 1);
2384 			if (ret)
2385 				printk(KERN_ERR MOD
2386 				       "%s - qp <- error failed!\n",
2387 				       __func__);
2388 		}
2389 		peer_abort_upcall(ep);
2390 		break;
2391 	case ABORTING:
2392 		break;
2393 	case DEAD:
2394 		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2395 		mutex_unlock(&ep->com.mutex);
2396 		return 0;
2397 	default:
2398 		BUG_ON(1);
2399 		break;
2400 	}
2401 	dst_confirm(ep->dst);
2402 	if (ep->com.state != ABORTING) {
2403 		__state_set(&ep->com, DEAD);
2404 		/* we don't release if we want to retry with mpa_v1 */
2405 		if (!ep->retry_with_mpa_v1)
2406 			release = 1;
2407 	}
2408 	mutex_unlock(&ep->com.mutex);
2409 
2410 	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2411 	if (!rpl_skb) {
2412 		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2413 		       __func__);
2414 		release = 1;
2415 		goto out;
2416 	}
2417 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2418 	rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2419 	INIT_TP_WR(rpl, ep->hwtid);
2420 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2421 	rpl->cmd = CPL_ABORT_NO_RST;
2422 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2423 out:
2424 	if (release)
2425 		release_ep_resources(ep);
2426 	else if (ep->retry_with_mpa_v1) {
2427 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2428 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2429 		dst_release(ep->dst);
2430 		cxgb4_l2t_release(ep->l2t);
2431 		c4iw_reconnect(ep);
2432 	}
2433 
2434 	return 0;
2435 }
2436 
2437 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2438 {
2439 	struct c4iw_ep *ep;
2440 	struct c4iw_qp_attributes attrs;
2441 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2442 	int release = 0;
2443 	struct tid_info *t = dev->rdev.lldi.tids;
2444 	unsigned int tid = GET_TID(rpl);
2445 
2446 	ep = lookup_tid(t, tid);
2447 
2448 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2449 	BUG_ON(!ep);
2450 
2451 	/* The cm_id may be null if we failed to connect */
2452 	mutex_lock(&ep->com.mutex);
2453 	switch (ep->com.state) {
2454 	case CLOSING:
2455 		__state_set(&ep->com, MORIBUND);
2456 		break;
2457 	case MORIBUND:
2458 		(void)stop_ep_timer(ep);
2459 		if ((ep->com.cm_id) && (ep->com.qp)) {
2460 			attrs.next_state = C4IW_QP_STATE_IDLE;
2461 			c4iw_modify_qp(ep->com.qp->rhp,
2462 					     ep->com.qp,
2463 					     C4IW_QP_ATTR_NEXT_STATE,
2464 					     &attrs, 1);
2465 		}
2466 		close_complete_upcall(ep, 0);
2467 		__state_set(&ep->com, DEAD);
2468 		release = 1;
2469 		break;
2470 	case ABORTING:
2471 	case DEAD:
2472 		break;
2473 	default:
2474 		BUG_ON(1);
2475 		break;
2476 	}
2477 	mutex_unlock(&ep->com.mutex);
2478 	if (release)
2479 		release_ep_resources(ep);
2480 	return 0;
2481 }
2482 
2483 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2484 {
2485 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
2486 	struct tid_info *t = dev->rdev.lldi.tids;
2487 	unsigned int tid = GET_TID(rpl);
2488 	struct c4iw_ep *ep;
2489 	struct c4iw_qp_attributes attrs;
2490 
2491 	ep = lookup_tid(t, tid);
2492 	BUG_ON(!ep);
2493 
2494 	if (ep && ep->com.qp) {
2495 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2496 		       ep->com.qp->wq.sq.qid);
2497 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2498 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2499 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2500 	} else
2501 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2502 
2503 	return 0;
2504 }
2505 
2506 /*
2507  * Upcall from the adapter indicating data has been transmitted.
2508  * For us its just the single MPA request or reply.  We can now free
2509  * the skb holding the mpa message.
2510  */
2511 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2512 {
2513 	struct c4iw_ep *ep;
2514 	struct cpl_fw4_ack *hdr = cplhdr(skb);
2515 	u8 credits = hdr->credits;
2516 	unsigned int tid = GET_TID(hdr);
2517 	struct tid_info *t = dev->rdev.lldi.tids;
2518 
2519 
2520 	ep = lookup_tid(t, tid);
2521 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2522 	if (credits == 0) {
2523 		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2524 		     __func__, ep, ep->hwtid, state_read(&ep->com));
2525 		return 0;
2526 	}
2527 
2528 	dst_confirm(ep->dst);
2529 	if (ep->mpa_skb) {
2530 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
2531 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2532 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2533 		kfree_skb(ep->mpa_skb);
2534 		ep->mpa_skb = NULL;
2535 	}
2536 	return 0;
2537 }
2538 
2539 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2540 {
2541 	int err = 0;
2542 	int disconnect = 0;
2543 	struct c4iw_ep *ep = to_ep(cm_id);
2544 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2545 
2546 	mutex_lock(&ep->com.mutex);
2547 	if (ep->com.state == DEAD) {
2548 		mutex_unlock(&ep->com.mutex);
2549 		c4iw_put_ep(&ep->com);
2550 		return -ECONNRESET;
2551 	}
2552 	set_bit(ULP_REJECT, &ep->com.history);
2553 	BUG_ON(ep->com.state != MPA_REQ_RCVD);
2554 	if (mpa_rev == 0)
2555 		abort_connection(ep, NULL, GFP_KERNEL);
2556 	else {
2557 		err = send_mpa_reject(ep, pdata, pdata_len);
2558 		disconnect = 1;
2559 	}
2560 	mutex_unlock(&ep->com.mutex);
2561 	if (disconnect)
2562 		err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2563 	c4iw_put_ep(&ep->com);
2564 	return 0;
2565 }
2566 
2567 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2568 {
2569 	int err;
2570 	struct c4iw_qp_attributes attrs;
2571 	enum c4iw_qp_attr_mask mask;
2572 	struct c4iw_ep *ep = to_ep(cm_id);
2573 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2574 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2575 
2576 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2577 
2578 	mutex_lock(&ep->com.mutex);
2579 	if (ep->com.state == DEAD) {
2580 		err = -ECONNRESET;
2581 		goto err;
2582 	}
2583 
2584 	BUG_ON(ep->com.state != MPA_REQ_RCVD);
2585 	BUG_ON(!qp);
2586 
2587 	set_bit(ULP_ACCEPT, &ep->com.history);
2588 	if ((conn_param->ord > c4iw_max_read_depth) ||
2589 	    (conn_param->ird > c4iw_max_read_depth)) {
2590 		abort_connection(ep, NULL, GFP_KERNEL);
2591 		err = -EINVAL;
2592 		goto err;
2593 	}
2594 
2595 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2596 		if (conn_param->ord > ep->ird) {
2597 			ep->ird = conn_param->ird;
2598 			ep->ord = conn_param->ord;
2599 			send_mpa_reject(ep, conn_param->private_data,
2600 					conn_param->private_data_len);
2601 			abort_connection(ep, NULL, GFP_KERNEL);
2602 			err = -ENOMEM;
2603 			goto err;
2604 		}
2605 		if (conn_param->ird > ep->ord) {
2606 			if (!ep->ord)
2607 				conn_param->ird = 1;
2608 			else {
2609 				abort_connection(ep, NULL, GFP_KERNEL);
2610 				err = -ENOMEM;
2611 				goto err;
2612 			}
2613 		}
2614 
2615 	}
2616 	ep->ird = conn_param->ird;
2617 	ep->ord = conn_param->ord;
2618 
2619 	if (ep->mpa_attr.version != 2)
2620 		if (peer2peer && ep->ird == 0)
2621 			ep->ird = 1;
2622 
2623 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2624 
2625 	cm_id->add_ref(cm_id);
2626 	ep->com.cm_id = cm_id;
2627 	ep->com.qp = qp;
2628 	ref_qp(ep);
2629 
2630 	/* bind QP to EP and move to RTS */
2631 	attrs.mpa_attr = ep->mpa_attr;
2632 	attrs.max_ird = ep->ird;
2633 	attrs.max_ord = ep->ord;
2634 	attrs.llp_stream_handle = ep;
2635 	attrs.next_state = C4IW_QP_STATE_RTS;
2636 
2637 	/* bind QP and TID with INIT_WR */
2638 	mask = C4IW_QP_ATTR_NEXT_STATE |
2639 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2640 			     C4IW_QP_ATTR_MPA_ATTR |
2641 			     C4IW_QP_ATTR_MAX_IRD |
2642 			     C4IW_QP_ATTR_MAX_ORD;
2643 
2644 	err = c4iw_modify_qp(ep->com.qp->rhp,
2645 			     ep->com.qp, mask, &attrs, 1);
2646 	if (err)
2647 		goto err1;
2648 	err = send_mpa_reply(ep, conn_param->private_data,
2649 			     conn_param->private_data_len);
2650 	if (err)
2651 		goto err1;
2652 
2653 	__state_set(&ep->com, FPDU_MODE);
2654 	established_upcall(ep);
2655 	mutex_unlock(&ep->com.mutex);
2656 	c4iw_put_ep(&ep->com);
2657 	return 0;
2658 err1:
2659 	ep->com.cm_id = NULL;
2660 	cm_id->rem_ref(cm_id);
2661 err:
2662 	mutex_unlock(&ep->com.mutex);
2663 	c4iw_put_ep(&ep->com);
2664 	return err;
2665 }
2666 
2667 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2668 {
2669 	struct in_device *ind;
2670 	int found = 0;
2671 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2672 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2673 
2674 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
2675 	if (!ind)
2676 		return -EADDRNOTAVAIL;
2677 	for_primary_ifa(ind) {
2678 		laddr->sin_addr.s_addr = ifa->ifa_address;
2679 		raddr->sin_addr.s_addr = ifa->ifa_address;
2680 		found = 1;
2681 		break;
2682 	}
2683 	endfor_ifa(ind);
2684 	in_dev_put(ind);
2685 	return found ? 0 : -EADDRNOTAVAIL;
2686 }
2687 
2688 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2689 		      unsigned char banned_flags)
2690 {
2691 	struct inet6_dev *idev;
2692 	int err = -EADDRNOTAVAIL;
2693 
2694 	rcu_read_lock();
2695 	idev = __in6_dev_get(dev);
2696 	if (idev != NULL) {
2697 		struct inet6_ifaddr *ifp;
2698 
2699 		read_lock_bh(&idev->lock);
2700 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
2701 			if (ifp->scope == IFA_LINK &&
2702 			    !(ifp->flags & banned_flags)) {
2703 				memcpy(addr, &ifp->addr, 16);
2704 				err = 0;
2705 				break;
2706 			}
2707 		}
2708 		read_unlock_bh(&idev->lock);
2709 	}
2710 	rcu_read_unlock();
2711 	return err;
2712 }
2713 
2714 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2715 {
2716 	struct in6_addr uninitialized_var(addr);
2717 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2718 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2719 
2720 	if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2721 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2722 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2723 		return 0;
2724 	}
2725 	return -EADDRNOTAVAIL;
2726 }
2727 
2728 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2729 {
2730 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2731 	struct c4iw_ep *ep;
2732 	int err = 0;
2733 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2734 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2735 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2736 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2737 				      &cm_id->remote_addr;
2738 	__u8 *ra;
2739 	int iptype;
2740 
2741 	if ((conn_param->ord > c4iw_max_read_depth) ||
2742 	    (conn_param->ird > c4iw_max_read_depth)) {
2743 		err = -EINVAL;
2744 		goto out;
2745 	}
2746 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2747 	if (!ep) {
2748 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2749 		err = -ENOMEM;
2750 		goto out;
2751 	}
2752 	init_timer(&ep->timer);
2753 	ep->plen = conn_param->private_data_len;
2754 	if (ep->plen)
2755 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2756 		       conn_param->private_data, ep->plen);
2757 	ep->ird = conn_param->ird;
2758 	ep->ord = conn_param->ord;
2759 
2760 	if (peer2peer && ep->ord == 0)
2761 		ep->ord = 1;
2762 
2763 	cm_id->add_ref(cm_id);
2764 	ep->com.dev = dev;
2765 	ep->com.cm_id = cm_id;
2766 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2767 	if (!ep->com.qp) {
2768 		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2769 		err = -EINVAL;
2770 		goto fail2;
2771 	}
2772 	ref_qp(ep);
2773 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2774 	     ep->com.qp, cm_id);
2775 
2776 	/*
2777 	 * Allocate an active TID to initiate a TCP connection.
2778 	 */
2779 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2780 	if (ep->atid == -1) {
2781 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2782 		err = -ENOMEM;
2783 		goto fail2;
2784 	}
2785 	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2786 
2787 	if (cm_id->remote_addr.ss_family == AF_INET) {
2788 		iptype = 4;
2789 		ra = (__u8 *)&raddr->sin_addr;
2790 
2791 		/*
2792 		 * Handle loopback requests to INADDR_ANY.
2793 		 */
2794 		if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2795 			err = pick_local_ipaddrs(dev, cm_id);
2796 			if (err)
2797 				goto fail2;
2798 		}
2799 
2800 		/* find a route */
2801 		PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2802 		     __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2803 		     ra, ntohs(raddr->sin_port));
2804 		ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2805 				     raddr->sin_addr.s_addr, laddr->sin_port,
2806 				     raddr->sin_port, 0);
2807 	} else {
2808 		iptype = 6;
2809 		ra = (__u8 *)&raddr6->sin6_addr;
2810 
2811 		/*
2812 		 * Handle loopback requests to INADDR_ANY.
2813 		 */
2814 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2815 			err = pick_local_ip6addrs(dev, cm_id);
2816 			if (err)
2817 				goto fail2;
2818 		}
2819 
2820 		/* find a route */
2821 		PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2822 		     __func__, laddr6->sin6_addr.s6_addr,
2823 		     ntohs(laddr6->sin6_port),
2824 		     raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2825 		ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2826 				      raddr6->sin6_addr.s6_addr,
2827 				      laddr6->sin6_port, raddr6->sin6_port, 0,
2828 				      raddr6->sin6_scope_id);
2829 	}
2830 	if (!ep->dst) {
2831 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2832 		err = -EHOSTUNREACH;
2833 		goto fail3;
2834 	}
2835 
2836 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2837 	if (err) {
2838 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2839 		goto fail4;
2840 	}
2841 
2842 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2843 		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2844 		ep->l2t->idx);
2845 
2846 	state_set(&ep->com, CONNECTING);
2847 	ep->tos = 0;
2848 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
2849 	       sizeof(ep->com.local_addr));
2850 	memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2851 	       sizeof(ep->com.remote_addr));
2852 
2853 	/* send connect request to rnic */
2854 	err = send_connect(ep);
2855 	if (!err)
2856 		goto out;
2857 
2858 	cxgb4_l2t_release(ep->l2t);
2859 fail4:
2860 	dst_release(ep->dst);
2861 fail3:
2862 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2863 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2864 fail2:
2865 	cm_id->rem_ref(cm_id);
2866 	c4iw_put_ep(&ep->com);
2867 out:
2868 	return err;
2869 }
2870 
2871 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2872 {
2873 	int err;
2874 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2875 
2876 	c4iw_init_wr_wait(&ep->com.wr_wait);
2877 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2878 				   ep->stid, &sin6->sin6_addr,
2879 				   sin6->sin6_port,
2880 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
2881 	if (!err)
2882 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2883 					  &ep->com.wr_wait,
2884 					  0, 0, __func__);
2885 	if (err)
2886 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2887 		       err, ep->stid,
2888 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2889 	return err;
2890 }
2891 
2892 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2893 {
2894 	int err;
2895 	struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2896 
2897 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
2898 		do {
2899 			err = cxgb4_create_server_filter(
2900 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
2901 				sin->sin_addr.s_addr, sin->sin_port, 0,
2902 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2903 			if (err == -EBUSY) {
2904 				set_current_state(TASK_UNINTERRUPTIBLE);
2905 				schedule_timeout(usecs_to_jiffies(100));
2906 			}
2907 		} while (err == -EBUSY);
2908 	} else {
2909 		c4iw_init_wr_wait(&ep->com.wr_wait);
2910 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2911 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2912 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2913 		if (!err)
2914 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2915 						  &ep->com.wr_wait,
2916 						  0, 0, __func__);
2917 	}
2918 	if (err)
2919 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2920 		       , err, ep->stid,
2921 		       &sin->sin_addr, ntohs(sin->sin_port));
2922 	return err;
2923 }
2924 
2925 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2926 {
2927 	int err = 0;
2928 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2929 	struct c4iw_listen_ep *ep;
2930 
2931 	might_sleep();
2932 
2933 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2934 	if (!ep) {
2935 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2936 		err = -ENOMEM;
2937 		goto fail1;
2938 	}
2939 	PDBG("%s ep %p\n", __func__, ep);
2940 	cm_id->add_ref(cm_id);
2941 	ep->com.cm_id = cm_id;
2942 	ep->com.dev = dev;
2943 	ep->backlog = backlog;
2944 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
2945 	       sizeof(ep->com.local_addr));
2946 
2947 	/*
2948 	 * Allocate a server TID.
2949 	 */
2950 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
2951 	    ep->com.local_addr.ss_family == AF_INET)
2952 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2953 					     cm_id->local_addr.ss_family, ep);
2954 	else
2955 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2956 					    cm_id->local_addr.ss_family, ep);
2957 
2958 	if (ep->stid == -1) {
2959 		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2960 		err = -ENOMEM;
2961 		goto fail2;
2962 	}
2963 	insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2964 	state_set(&ep->com, LISTEN);
2965 	if (ep->com.local_addr.ss_family == AF_INET)
2966 		err = create_server4(dev, ep);
2967 	else
2968 		err = create_server6(dev, ep);
2969 	if (!err) {
2970 		cm_id->provider_data = ep;
2971 		goto out;
2972 	}
2973 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2974 			ep->com.local_addr.ss_family);
2975 fail2:
2976 	cm_id->rem_ref(cm_id);
2977 	c4iw_put_ep(&ep->com);
2978 fail1:
2979 out:
2980 	return err;
2981 }
2982 
2983 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2984 {
2985 	int err;
2986 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2987 
2988 	PDBG("%s ep %p\n", __func__, ep);
2989 
2990 	might_sleep();
2991 	state_set(&ep->com, DEAD);
2992 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2993 	    ep->com.local_addr.ss_family == AF_INET) {
2994 		err = cxgb4_remove_server_filter(
2995 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
2996 			ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2997 	} else {
2998 		c4iw_init_wr_wait(&ep->com.wr_wait);
2999 		err = cxgb4_remove_server(
3000 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3001 				ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3002 		if (err)
3003 			goto done;
3004 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3005 					  0, 0, __func__);
3006 	}
3007 	remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3008 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3009 			ep->com.local_addr.ss_family);
3010 done:
3011 	cm_id->rem_ref(cm_id);
3012 	c4iw_put_ep(&ep->com);
3013 	return err;
3014 }
3015 
3016 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3017 {
3018 	int ret = 0;
3019 	int close = 0;
3020 	int fatal = 0;
3021 	struct c4iw_rdev *rdev;
3022 
3023 	mutex_lock(&ep->com.mutex);
3024 
3025 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3026 	     states[ep->com.state], abrupt);
3027 
3028 	rdev = &ep->com.dev->rdev;
3029 	if (c4iw_fatal_error(rdev)) {
3030 		fatal = 1;
3031 		close_complete_upcall(ep, -EIO);
3032 		ep->com.state = DEAD;
3033 	}
3034 	switch (ep->com.state) {
3035 	case MPA_REQ_WAIT:
3036 	case MPA_REQ_SENT:
3037 	case MPA_REQ_RCVD:
3038 	case MPA_REP_SENT:
3039 	case FPDU_MODE:
3040 		close = 1;
3041 		if (abrupt)
3042 			ep->com.state = ABORTING;
3043 		else {
3044 			ep->com.state = CLOSING;
3045 			start_ep_timer(ep);
3046 		}
3047 		set_bit(CLOSE_SENT, &ep->com.flags);
3048 		break;
3049 	case CLOSING:
3050 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3051 			close = 1;
3052 			if (abrupt) {
3053 				(void)stop_ep_timer(ep);
3054 				ep->com.state = ABORTING;
3055 			} else
3056 				ep->com.state = MORIBUND;
3057 		}
3058 		break;
3059 	case MORIBUND:
3060 	case ABORTING:
3061 	case DEAD:
3062 		PDBG("%s ignoring disconnect ep %p state %u\n",
3063 		     __func__, ep, ep->com.state);
3064 		break;
3065 	default:
3066 		BUG();
3067 		break;
3068 	}
3069 
3070 	if (close) {
3071 		if (abrupt) {
3072 			set_bit(EP_DISC_ABORT, &ep->com.history);
3073 			close_complete_upcall(ep, -ECONNRESET);
3074 			ret = send_abort(ep, NULL, gfp);
3075 		} else {
3076 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3077 			ret = send_halfclose(ep, gfp);
3078 		}
3079 		if (ret)
3080 			fatal = 1;
3081 	}
3082 	mutex_unlock(&ep->com.mutex);
3083 	if (fatal)
3084 		release_ep_resources(ep);
3085 	return ret;
3086 }
3087 
3088 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3089 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3090 {
3091 	struct c4iw_ep *ep;
3092 	int atid = be32_to_cpu(req->tid);
3093 
3094 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3095 					   (__force u32) req->tid);
3096 	if (!ep)
3097 		return;
3098 
3099 	switch (req->retval) {
3100 	case FW_ENOMEM:
3101 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3102 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3103 			send_fw_act_open_req(ep, atid);
3104 			return;
3105 		}
3106 	case FW_EADDRINUSE:
3107 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3108 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3109 			send_fw_act_open_req(ep, atid);
3110 			return;
3111 		}
3112 		break;
3113 	default:
3114 		pr_info("%s unexpected ofld conn wr retval %d\n",
3115 		       __func__, req->retval);
3116 		break;
3117 	}
3118 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3119 	       req->retval, atid);
3120 	mutex_lock(&dev->rdev.stats.lock);
3121 	dev->rdev.stats.act_ofld_conn_fails++;
3122 	mutex_unlock(&dev->rdev.stats.lock);
3123 	connect_reply_upcall(ep, status2errno(req->retval));
3124 	state_set(&ep->com, DEAD);
3125 	remove_handle(dev, &dev->atid_idr, atid);
3126 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3127 	dst_release(ep->dst);
3128 	cxgb4_l2t_release(ep->l2t);
3129 	c4iw_put_ep(&ep->com);
3130 }
3131 
3132 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3133 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3134 {
3135 	struct sk_buff *rpl_skb;
3136 	struct cpl_pass_accept_req *cpl;
3137 	int ret;
3138 
3139 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3140 	BUG_ON(!rpl_skb);
3141 	if (req->retval) {
3142 		PDBG("%s passive open failure %d\n", __func__, req->retval);
3143 		mutex_lock(&dev->rdev.stats.lock);
3144 		dev->rdev.stats.pas_ofld_conn_fails++;
3145 		mutex_unlock(&dev->rdev.stats.lock);
3146 		kfree_skb(rpl_skb);
3147 	} else {
3148 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3149 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3150 					(__force u32) htonl(
3151 					(__force u32) req->tid)));
3152 		ret = pass_accept_req(dev, rpl_skb);
3153 		if (!ret)
3154 			kfree_skb(rpl_skb);
3155 	}
3156 	return;
3157 }
3158 
3159 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3160 {
3161 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3162 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3163 
3164 	switch (rpl->type) {
3165 	case FW6_TYPE_CQE:
3166 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3167 		break;
3168 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3169 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3170 		switch (req->t_state) {
3171 		case TCP_SYN_SENT:
3172 			active_ofld_conn_reply(dev, skb, req);
3173 			break;
3174 		case TCP_SYN_RECV:
3175 			passive_ofld_conn_reply(dev, skb, req);
3176 			break;
3177 		default:
3178 			pr_err("%s unexpected ofld conn wr state %d\n",
3179 			       __func__, req->t_state);
3180 			break;
3181 		}
3182 		break;
3183 	}
3184 	return 0;
3185 }
3186 
3187 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3188 {
3189 	u32 l2info;
3190 	u16 vlantag, len, hdr_len, eth_hdr_len;
3191 	u8 intf;
3192 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3193 	struct cpl_pass_accept_req *req;
3194 	struct tcp_options_received tmp_opt;
3195 	struct c4iw_dev *dev;
3196 
3197 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3198 	/* Store values from cpl_rx_pkt in temporary location. */
3199 	vlantag = (__force u16) cpl->vlan;
3200 	len = (__force u16) cpl->len;
3201 	l2info  = (__force u32) cpl->l2info;
3202 	hdr_len = (__force u16) cpl->hdr_len;
3203 	intf = cpl->iff;
3204 
3205 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3206 
3207 	/*
3208 	 * We need to parse the TCP options from SYN packet.
3209 	 * to generate cpl_pass_accept_req.
3210 	 */
3211 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3212 	tcp_clear_options(&tmp_opt);
3213 	tcp_parse_options(skb, &tmp_opt, 0, NULL);
3214 
3215 	req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3216 	memset(req, 0, sizeof(*req));
3217 	req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3218 			 V_SYN_MAC_IDX(G_RX_MACIDX(
3219 			 (__force int) htonl(l2info))) |
3220 			 F_SYN_XACT_MATCH);
3221 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3222 			    G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3223 			    G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3224 	req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3225 					(__force int) htonl(l2info))) |
3226 				   V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3227 					(__force int) htons(hdr_len))) |
3228 				   V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3229 					(__force int) htons(hdr_len))) |
3230 				   V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3231 	req->vlan = (__force __be16) vlantag;
3232 	req->len = (__force __be16) len;
3233 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3234 				    PASS_OPEN_TOS(tos));
3235 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3236 	if (tmp_opt.wscale_ok)
3237 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3238 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3239 	if (tmp_opt.sack_ok)
3240 		req->tcpopt.sack = 1;
3241 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3242 	return;
3243 }
3244 
3245 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3246 				  __be32 laddr, __be16 lport,
3247 				  __be32 raddr, __be16 rport,
3248 				  u32 rcv_isn, u32 filter, u16 window,
3249 				  u32 rss_qid, u8 port_id)
3250 {
3251 	struct sk_buff *req_skb;
3252 	struct fw_ofld_connection_wr *req;
3253 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
3254 	int ret;
3255 
3256 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3257 	req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3258 	memset(req, 0, sizeof(*req));
3259 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3260 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3261 	req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3262 	req->le.filter = (__force __be32) filter;
3263 	req->le.lport = lport;
3264 	req->le.pport = rport;
3265 	req->le.u.ipv4.lip = laddr;
3266 	req->le.u.ipv4.pip = raddr;
3267 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3268 	req->tcb.rcv_adv = htons(window);
3269 	req->tcb.t_state_to_astid =
3270 		 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3271 			V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3272 			V_FW_OFLD_CONNECTION_WR_ASTID(
3273 			GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3274 
3275 	/*
3276 	 * We store the qid in opt2 which will be used by the firmware
3277 	 * to send us the wr response.
3278 	 */
3279 	req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3280 
3281 	/*
3282 	 * We initialize the MSS index in TCB to 0xF.
3283 	 * So that when driver sends cpl_pass_accept_rpl
3284 	 * TCB picks up the correct value. If this was 0
3285 	 * TP will ignore any value > 0 for MSS index.
3286 	 */
3287 	req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3288 	req->cookie = (unsigned long)skb;
3289 
3290 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3291 	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3292 	if (ret < 0) {
3293 		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3294 		       ret);
3295 		kfree_skb(skb);
3296 		kfree_skb(req_skb);
3297 	}
3298 }
3299 
3300 /*
3301  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3302  * messages when a filter is being used instead of server to
3303  * redirect a syn packet. When packets hit filter they are redirected
3304  * to the offload queue and driver tries to establish the connection
3305  * using firmware work request.
3306  */
3307 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3308 {
3309 	int stid;
3310 	unsigned int filter;
3311 	struct ethhdr *eh = NULL;
3312 	struct vlan_ethhdr *vlan_eh = NULL;
3313 	struct iphdr *iph;
3314 	struct tcphdr *tcph;
3315 	struct rss_header *rss = (void *)skb->data;
3316 	struct cpl_rx_pkt *cpl = (void *)skb->data;
3317 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
3318 	struct l2t_entry *e;
3319 	struct dst_entry *dst;
3320 	struct c4iw_ep *lep;
3321 	u16 window;
3322 	struct port_info *pi;
3323 	struct net_device *pdev;
3324 	u16 rss_qid, eth_hdr_len;
3325 	int step;
3326 	u32 tx_chan;
3327 	struct neighbour *neigh;
3328 
3329 	/* Drop all non-SYN packets */
3330 	if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3331 		goto reject;
3332 
3333 	/*
3334 	 * Drop all packets which did not hit the filter.
3335 	 * Unlikely to happen.
3336 	 */
3337 	if (!(rss->filter_hit && rss->filter_tid))
3338 		goto reject;
3339 
3340 	/*
3341 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3342 	 */
3343 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3344 
3345 	lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3346 	if (!lep) {
3347 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3348 		goto reject;
3349 	}
3350 
3351 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3352 			    G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3353 			    G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3354 	if (eth_hdr_len == ETH_HLEN) {
3355 		eh = (struct ethhdr *)(req + 1);
3356 		iph = (struct iphdr *)(eh + 1);
3357 	} else {
3358 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
3359 		iph = (struct iphdr *)(vlan_eh + 1);
3360 		skb->vlan_tci = ntohs(cpl->vlan);
3361 	}
3362 
3363 	if (iph->version != 0x4)
3364 		goto reject;
3365 
3366 	tcph = (struct tcphdr *)(iph + 1);
3367 	skb_set_network_header(skb, (void *)iph - (void *)rss);
3368 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3369 	skb_get(skb);
3370 
3371 	PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3372 	     ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3373 	     ntohs(tcph->source), iph->tos);
3374 
3375 	dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3376 			 iph->tos);
3377 	if (!dst) {
3378 		pr_err("%s - failed to find dst entry!\n",
3379 		       __func__);
3380 		goto reject;
3381 	}
3382 	neigh = dst_neigh_lookup_skb(dst, skb);
3383 
3384 	if (!neigh) {
3385 		pr_err("%s - failed to allocate neigh!\n",
3386 		       __func__);
3387 		goto free_dst;
3388 	}
3389 
3390 	if (neigh->dev->flags & IFF_LOOPBACK) {
3391 		pdev = ip_dev_find(&init_net, iph->daddr);
3392 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3393 				    pdev, 0);
3394 		pi = (struct port_info *)netdev_priv(pdev);
3395 		tx_chan = cxgb4_port_chan(pdev);
3396 		dev_put(pdev);
3397 	} else {
3398 		pdev = get_real_dev(neigh->dev);
3399 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3400 					pdev, 0);
3401 		pi = (struct port_info *)netdev_priv(pdev);
3402 		tx_chan = cxgb4_port_chan(pdev);
3403 	}
3404 	neigh_release(neigh);
3405 	if (!e) {
3406 		pr_err("%s - failed to allocate l2t entry!\n",
3407 		       __func__);
3408 		goto free_dst;
3409 	}
3410 
3411 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3412 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3413 	window = (__force u16) htons((__force u16)tcph->window);
3414 
3415 	/* Calcuate filter portion for LE region. */
3416 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3417 						    dev->rdev.lldi.ports[0],
3418 						    e));
3419 
3420 	/*
3421 	 * Synthesize the cpl_pass_accept_req. We have everything except the
3422 	 * TID. Once firmware sends a reply with TID we update the TID field
3423 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
3424 	 */
3425 	build_cpl_pass_accept_req(skb, stid, iph->tos);
3426 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3427 			      tcph->source, ntohl(tcph->seq), filter, window,
3428 			      rss_qid, pi->port_id);
3429 	cxgb4_l2t_release(e);
3430 free_dst:
3431 	dst_release(dst);
3432 reject:
3433 	return 0;
3434 }
3435 
3436 /*
3437  * These are the real handlers that are called from a
3438  * work queue.
3439  */
3440 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3441 	[CPL_ACT_ESTABLISH] = act_establish,
3442 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
3443 	[CPL_RX_DATA] = rx_data,
3444 	[CPL_ABORT_RPL_RSS] = abort_rpl,
3445 	[CPL_ABORT_RPL] = abort_rpl,
3446 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
3447 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3448 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3449 	[CPL_PASS_ESTABLISH] = pass_establish,
3450 	[CPL_PEER_CLOSE] = peer_close,
3451 	[CPL_ABORT_REQ_RSS] = peer_abort,
3452 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
3453 	[CPL_RDMA_TERMINATE] = terminate,
3454 	[CPL_FW4_ACK] = fw4_ack,
3455 	[CPL_FW6_MSG] = deferred_fw6_msg,
3456 	[CPL_RX_PKT] = rx_pkt
3457 };
3458 
3459 static void process_timeout(struct c4iw_ep *ep)
3460 {
3461 	struct c4iw_qp_attributes attrs;
3462 	int abort = 1;
3463 
3464 	mutex_lock(&ep->com.mutex);
3465 	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3466 	     ep->com.state);
3467 	set_bit(TIMEDOUT, &ep->com.history);
3468 	switch (ep->com.state) {
3469 	case MPA_REQ_SENT:
3470 		__state_set(&ep->com, ABORTING);
3471 		connect_reply_upcall(ep, -ETIMEDOUT);
3472 		break;
3473 	case MPA_REQ_WAIT:
3474 		__state_set(&ep->com, ABORTING);
3475 		break;
3476 	case CLOSING:
3477 	case MORIBUND:
3478 		if (ep->com.cm_id && ep->com.qp) {
3479 			attrs.next_state = C4IW_QP_STATE_ERROR;
3480 			c4iw_modify_qp(ep->com.qp->rhp,
3481 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3482 				     &attrs, 1);
3483 		}
3484 		__state_set(&ep->com, ABORTING);
3485 		close_complete_upcall(ep, -ETIMEDOUT);
3486 		break;
3487 	case ABORTING:
3488 	case DEAD:
3489 
3490 		/*
3491 		 * These states are expected if the ep timed out at the same
3492 		 * time as another thread was calling stop_ep_timer().
3493 		 * So we silently do nothing for these states.
3494 		 */
3495 		abort = 0;
3496 		break;
3497 	default:
3498 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3499 			__func__, ep, ep->hwtid, ep->com.state);
3500 		abort = 0;
3501 	}
3502 	if (abort)
3503 		abort_connection(ep, NULL, GFP_KERNEL);
3504 	mutex_unlock(&ep->com.mutex);
3505 	c4iw_put_ep(&ep->com);
3506 }
3507 
3508 static void process_timedout_eps(void)
3509 {
3510 	struct c4iw_ep *ep;
3511 
3512 	spin_lock_irq(&timeout_lock);
3513 	while (!list_empty(&timeout_list)) {
3514 		struct list_head *tmp;
3515 
3516 		tmp = timeout_list.next;
3517 		list_del(tmp);
3518 		tmp->next = NULL;
3519 		tmp->prev = NULL;
3520 		spin_unlock_irq(&timeout_lock);
3521 		ep = list_entry(tmp, struct c4iw_ep, entry);
3522 		process_timeout(ep);
3523 		spin_lock_irq(&timeout_lock);
3524 	}
3525 	spin_unlock_irq(&timeout_lock);
3526 }
3527 
3528 static void process_work(struct work_struct *work)
3529 {
3530 	struct sk_buff *skb = NULL;
3531 	struct c4iw_dev *dev;
3532 	struct cpl_act_establish *rpl;
3533 	unsigned int opcode;
3534 	int ret;
3535 
3536 	process_timedout_eps();
3537 	while ((skb = skb_dequeue(&rxq))) {
3538 		rpl = cplhdr(skb);
3539 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3540 		opcode = rpl->ot.opcode;
3541 
3542 		BUG_ON(!work_handlers[opcode]);
3543 		ret = work_handlers[opcode](dev, skb);
3544 		if (!ret)
3545 			kfree_skb(skb);
3546 		process_timedout_eps();
3547 	}
3548 }
3549 
3550 static DECLARE_WORK(skb_work, process_work);
3551 
3552 static void ep_timeout(unsigned long arg)
3553 {
3554 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3555 	int kickit = 0;
3556 
3557 	spin_lock(&timeout_lock);
3558 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3559 		/*
3560 		 * Only insert if it is not already on the list.
3561 		 */
3562 		if (!ep->entry.next) {
3563 			list_add_tail(&ep->entry, &timeout_list);
3564 			kickit = 1;
3565 		}
3566 	}
3567 	spin_unlock(&timeout_lock);
3568 	if (kickit)
3569 		queue_work(workq, &skb_work);
3570 }
3571 
3572 /*
3573  * All the CM events are handled on a work queue to have a safe context.
3574  */
3575 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3576 {
3577 
3578 	/*
3579 	 * Save dev in the skb->cb area.
3580 	 */
3581 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3582 
3583 	/*
3584 	 * Queue the skb and schedule the worker thread.
3585 	 */
3586 	skb_queue_tail(&rxq, skb);
3587 	queue_work(workq, &skb_work);
3588 	return 0;
3589 }
3590 
3591 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3592 {
3593 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3594 
3595 	if (rpl->status != CPL_ERR_NONE) {
3596 		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3597 		       "for tid %u\n", rpl->status, GET_TID(rpl));
3598 	}
3599 	kfree_skb(skb);
3600 	return 0;
3601 }
3602 
3603 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3604 {
3605 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3606 	struct c4iw_wr_wait *wr_waitp;
3607 	int ret;
3608 
3609 	PDBG("%s type %u\n", __func__, rpl->type);
3610 
3611 	switch (rpl->type) {
3612 	case FW6_TYPE_WR_RPL:
3613 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3614 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3615 		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3616 		if (wr_waitp)
3617 			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3618 		kfree_skb(skb);
3619 		break;
3620 	case FW6_TYPE_CQE:
3621 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3622 		sched(dev, skb);
3623 		break;
3624 	default:
3625 		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3626 		       rpl->type);
3627 		kfree_skb(skb);
3628 		break;
3629 	}
3630 	return 0;
3631 }
3632 
3633 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3634 {
3635 	struct cpl_abort_req_rss *req = cplhdr(skb);
3636 	struct c4iw_ep *ep;
3637 	struct tid_info *t = dev->rdev.lldi.tids;
3638 	unsigned int tid = GET_TID(req);
3639 
3640 	ep = lookup_tid(t, tid);
3641 	if (!ep) {
3642 		printk(KERN_WARNING MOD
3643 		       "Abort on non-existent endpoint, tid %d\n", tid);
3644 		kfree_skb(skb);
3645 		return 0;
3646 	}
3647 	if (is_neg_adv(req->status)) {
3648 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3649 		     ep->hwtid);
3650 		kfree_skb(skb);
3651 		return 0;
3652 	}
3653 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3654 	     ep->com.state);
3655 
3656 	/*
3657 	 * Wake up any threads in rdma_init() or rdma_fini().
3658 	 * However, if we are on MPAv2 and want to retry with MPAv1
3659 	 * then, don't wake up yet.
3660 	 */
3661 	if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3662 		if (ep->com.state != MPA_REQ_SENT)
3663 			c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3664 	} else
3665 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3666 	sched(dev, skb);
3667 	return 0;
3668 }
3669 
3670 /*
3671  * Most upcalls from the T4 Core go to sched() to
3672  * schedule the processing on a work queue.
3673  */
3674 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3675 	[CPL_ACT_ESTABLISH] = sched,
3676 	[CPL_ACT_OPEN_RPL] = sched,
3677 	[CPL_RX_DATA] = sched,
3678 	[CPL_ABORT_RPL_RSS] = sched,
3679 	[CPL_ABORT_RPL] = sched,
3680 	[CPL_PASS_OPEN_RPL] = sched,
3681 	[CPL_CLOSE_LISTSRV_RPL] = sched,
3682 	[CPL_PASS_ACCEPT_REQ] = sched,
3683 	[CPL_PASS_ESTABLISH] = sched,
3684 	[CPL_PEER_CLOSE] = sched,
3685 	[CPL_CLOSE_CON_RPL] = sched,
3686 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
3687 	[CPL_RDMA_TERMINATE] = sched,
3688 	[CPL_FW4_ACK] = sched,
3689 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
3690 	[CPL_FW6_MSG] = fw6_msg,
3691 	[CPL_RX_PKT] = sched
3692 };
3693 
3694 int __init c4iw_cm_init(void)
3695 {
3696 	spin_lock_init(&timeout_lock);
3697 	skb_queue_head_init(&rxq);
3698 
3699 	workq = create_singlethread_workqueue("iw_cxgb4");
3700 	if (!workq)
3701 		return -ENOMEM;
3702 
3703 	return 0;
3704 }
3705 
3706 void __exit c4iw_cm_term(void)
3707 {
3708 	WARN_ON(!list_empty(&timeout_list));
3709 	flush_workqueue(workq);
3710 	destroy_workqueue(workq);
3711 }
3712