xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision 8730046c)
1 /*
2  * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *	  copyright notice, this list of conditions and the following
16  *	  disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *	  copyright notice, this list of conditions and the following
20  *	  disclaimer in the documentation and/or other materials
21  *	  provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
42 
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
49 
50 #include <rdma/ib_addr.h>
51 
52 #include <libcxgb_cm.h>
53 #include "iw_cxgb4.h"
54 #include "clip_tbl.h"
55 
56 static char *states[] = {
57 	"idle",
58 	"listen",
59 	"connecting",
60 	"mpa_wait_req",
61 	"mpa_req_sent",
62 	"mpa_req_rcvd",
63 	"mpa_rep_sent",
64 	"fpdu_mode",
65 	"aborting",
66 	"closing",
67 	"moribund",
68 	"dead",
69 	NULL,
70 };
71 
72 static int nocong;
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75 
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79 
80 static int dack_mode = 1;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
83 
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 		 "Per-connection max ORD/IRD (default=32)");
88 
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92 
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96 
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 		 "Enable tcp window scaling (default=1)");
101 
102 int c4iw_debug;
103 module_param(c4iw_debug, int, 0644);
104 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
105 
106 static int peer2peer = 1;
107 module_param(peer2peer, int, 0644);
108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
109 
110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
111 module_param(p2p_type, int, 0644);
112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
113 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
114 
115 static int ep_timeout_secs = 60;
116 module_param(ep_timeout_secs, int, 0644);
117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
118 				   "in seconds (default=60)");
119 
120 static int mpa_rev = 2;
121 module_param(mpa_rev, int, 0644);
122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
123 		"1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
124 		" compliant (default=2)");
125 
126 static int markers_enabled;
127 module_param(markers_enabled, int, 0644);
128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
129 
130 static int crc_enabled = 1;
131 module_param(crc_enabled, int, 0644);
132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
133 
134 static int rcv_win = 256 * 1024;
135 module_param(rcv_win, int, 0644);
136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
137 
138 static int snd_win = 128 * 1024;
139 module_param(snd_win, int, 0644);
140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
141 
142 static struct workqueue_struct *workq;
143 
144 static struct sk_buff_head rxq;
145 
146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
147 static void ep_timeout(unsigned long arg);
148 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
150 
151 static LIST_HEAD(timeout_list);
152 static spinlock_t timeout_lock;
153 
154 static void deref_cm_id(struct c4iw_ep_common *epc)
155 {
156 	epc->cm_id->rem_ref(epc->cm_id);
157 	epc->cm_id = NULL;
158 	set_bit(CM_ID_DEREFED, &epc->history);
159 }
160 
161 static void ref_cm_id(struct c4iw_ep_common *epc)
162 {
163 	set_bit(CM_ID_REFED, &epc->history);
164 	epc->cm_id->add_ref(epc->cm_id);
165 }
166 
167 static void deref_qp(struct c4iw_ep *ep)
168 {
169 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
170 	clear_bit(QP_REFERENCED, &ep->com.flags);
171 	set_bit(QP_DEREFED, &ep->com.history);
172 }
173 
174 static void ref_qp(struct c4iw_ep *ep)
175 {
176 	set_bit(QP_REFERENCED, &ep->com.flags);
177 	set_bit(QP_REFED, &ep->com.history);
178 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
179 }
180 
181 static void start_ep_timer(struct c4iw_ep *ep)
182 {
183 	PDBG("%s ep %p\n", __func__, ep);
184 	if (timer_pending(&ep->timer)) {
185 		pr_err("%s timer already started! ep %p\n",
186 		       __func__, ep);
187 		return;
188 	}
189 	clear_bit(TIMEOUT, &ep->com.flags);
190 	c4iw_get_ep(&ep->com);
191 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
192 	ep->timer.data = (unsigned long)ep;
193 	ep->timer.function = ep_timeout;
194 	add_timer(&ep->timer);
195 }
196 
197 static int stop_ep_timer(struct c4iw_ep *ep)
198 {
199 	PDBG("%s ep %p stopping\n", __func__, ep);
200 	del_timer_sync(&ep->timer);
201 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
202 		c4iw_put_ep(&ep->com);
203 		return 0;
204 	}
205 	return 1;
206 }
207 
208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
209 		  struct l2t_entry *l2e)
210 {
211 	int	error = 0;
212 
213 	if (c4iw_fatal_error(rdev)) {
214 		kfree_skb(skb);
215 		PDBG("%s - device in error state - dropping\n", __func__);
216 		return -EIO;
217 	}
218 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
219 	if (error < 0)
220 		kfree_skb(skb);
221 	else if (error == NET_XMIT_DROP)
222 		return -ENOMEM;
223 	return error < 0 ? error : 0;
224 }
225 
226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
227 {
228 	int	error = 0;
229 
230 	if (c4iw_fatal_error(rdev)) {
231 		kfree_skb(skb);
232 		PDBG("%s - device in error state - dropping\n", __func__);
233 		return -EIO;
234 	}
235 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
236 	if (error < 0)
237 		kfree_skb(skb);
238 	return error < 0 ? error : 0;
239 }
240 
241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
242 {
243 	u32 len = roundup(sizeof(struct cpl_tid_release), 16);
244 
245 	skb = get_skb(skb, len, GFP_KERNEL);
246 	if (!skb)
247 		return;
248 
249 	cxgb_mk_tid_release(skb, len, hwtid, 0);
250 	c4iw_ofld_send(rdev, skb);
251 	return;
252 }
253 
254 static void set_emss(struct c4iw_ep *ep, u16 opt)
255 {
256 	ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
257 		   ((AF_INET == ep->com.remote_addr.ss_family) ?
258 		    sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
259 		   sizeof(struct tcphdr);
260 	ep->mss = ep->emss;
261 	if (TCPOPT_TSTAMP_G(opt))
262 		ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
263 	if (ep->emss < 128)
264 		ep->emss = 128;
265 	if (ep->emss & 7)
266 		PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
267 		     TCPOPT_MSS_G(opt), ep->mss, ep->emss);
268 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
269 	     ep->mss, ep->emss);
270 }
271 
272 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
273 {
274 	enum c4iw_ep_state state;
275 
276 	mutex_lock(&epc->mutex);
277 	state = epc->state;
278 	mutex_unlock(&epc->mutex);
279 	return state;
280 }
281 
282 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
283 {
284 	epc->state = new;
285 }
286 
287 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
288 {
289 	mutex_lock(&epc->mutex);
290 	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
291 	__state_set(epc, new);
292 	mutex_unlock(&epc->mutex);
293 	return;
294 }
295 
296 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
297 {
298 	struct sk_buff *skb;
299 	unsigned int i;
300 	size_t len;
301 
302 	len = roundup(sizeof(union cpl_wr_size), 16);
303 	for (i = 0; i < size; i++) {
304 		skb = alloc_skb(len, GFP_KERNEL);
305 		if (!skb)
306 			goto fail;
307 		skb_queue_tail(ep_skb_list, skb);
308 	}
309 	return 0;
310 fail:
311 	skb_queue_purge(ep_skb_list);
312 	return -ENOMEM;
313 }
314 
315 static void *alloc_ep(int size, gfp_t gfp)
316 {
317 	struct c4iw_ep_common *epc;
318 
319 	epc = kzalloc(size, gfp);
320 	if (epc) {
321 		kref_init(&epc->kref);
322 		mutex_init(&epc->mutex);
323 		c4iw_init_wr_wait(&epc->wr_wait);
324 	}
325 	PDBG("%s alloc ep %p\n", __func__, epc);
326 	return epc;
327 }
328 
329 static void remove_ep_tid(struct c4iw_ep *ep)
330 {
331 	unsigned long flags;
332 
333 	spin_lock_irqsave(&ep->com.dev->lock, flags);
334 	_remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
335 	if (idr_is_empty(&ep->com.dev->hwtid_idr))
336 		wake_up(&ep->com.dev->wait);
337 	spin_unlock_irqrestore(&ep->com.dev->lock, flags);
338 }
339 
340 static void insert_ep_tid(struct c4iw_ep *ep)
341 {
342 	unsigned long flags;
343 
344 	spin_lock_irqsave(&ep->com.dev->lock, flags);
345 	_insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
346 	spin_unlock_irqrestore(&ep->com.dev->lock, flags);
347 }
348 
349 /*
350  * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
351  */
352 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
353 {
354 	struct c4iw_ep *ep;
355 	unsigned long flags;
356 
357 	spin_lock_irqsave(&dev->lock, flags);
358 	ep = idr_find(&dev->hwtid_idr, tid);
359 	if (ep)
360 		c4iw_get_ep(&ep->com);
361 	spin_unlock_irqrestore(&dev->lock, flags);
362 	return ep;
363 }
364 
365 /*
366  * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
367  */
368 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
369 					       unsigned int stid)
370 {
371 	struct c4iw_listen_ep *ep;
372 	unsigned long flags;
373 
374 	spin_lock_irqsave(&dev->lock, flags);
375 	ep = idr_find(&dev->stid_idr, stid);
376 	if (ep)
377 		c4iw_get_ep(&ep->com);
378 	spin_unlock_irqrestore(&dev->lock, flags);
379 	return ep;
380 }
381 
382 void _c4iw_free_ep(struct kref *kref)
383 {
384 	struct c4iw_ep *ep;
385 
386 	ep = container_of(kref, struct c4iw_ep, com.kref);
387 	PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
388 	if (test_bit(QP_REFERENCED, &ep->com.flags))
389 		deref_qp(ep);
390 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
391 		if (ep->com.remote_addr.ss_family == AF_INET6) {
392 			struct sockaddr_in6 *sin6 =
393 					(struct sockaddr_in6 *)
394 					&ep->com.local_addr;
395 
396 			cxgb4_clip_release(
397 					ep->com.dev->rdev.lldi.ports[0],
398 					(const u32 *)&sin6->sin6_addr.s6_addr,
399 					1);
400 		}
401 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
402 		dst_release(ep->dst);
403 		cxgb4_l2t_release(ep->l2t);
404 		if (ep->mpa_skb)
405 			kfree_skb(ep->mpa_skb);
406 	}
407 	if (!skb_queue_empty(&ep->com.ep_skb_list))
408 		skb_queue_purge(&ep->com.ep_skb_list);
409 	kfree(ep);
410 }
411 
412 static void release_ep_resources(struct c4iw_ep *ep)
413 {
414 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
415 
416 	/*
417 	 * If we have a hwtid, then remove it from the idr table
418 	 * so lookups will no longer find this endpoint.  Otherwise
419 	 * we have a race where one thread finds the ep ptr just
420 	 * before the other thread is freeing the ep memory.
421 	 */
422 	if (ep->hwtid != -1)
423 		remove_ep_tid(ep);
424 	c4iw_put_ep(&ep->com);
425 }
426 
427 static int status2errno(int status)
428 {
429 	switch (status) {
430 	case CPL_ERR_NONE:
431 		return 0;
432 	case CPL_ERR_CONN_RESET:
433 		return -ECONNRESET;
434 	case CPL_ERR_ARP_MISS:
435 		return -EHOSTUNREACH;
436 	case CPL_ERR_CONN_TIMEDOUT:
437 		return -ETIMEDOUT;
438 	case CPL_ERR_TCAM_FULL:
439 		return -ENOMEM;
440 	case CPL_ERR_CONN_EXIST:
441 		return -EADDRINUSE;
442 	default:
443 		return -EIO;
444 	}
445 }
446 
447 /*
448  * Try and reuse skbs already allocated...
449  */
450 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
451 {
452 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
453 		skb_trim(skb, 0);
454 		skb_get(skb);
455 		skb_reset_transport_header(skb);
456 	} else {
457 		skb = alloc_skb(len, gfp);
458 	}
459 	t4_set_arp_err_handler(skb, NULL, NULL);
460 	return skb;
461 }
462 
463 static struct net_device *get_real_dev(struct net_device *egress_dev)
464 {
465 	return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
466 }
467 
468 static void arp_failure_discard(void *handle, struct sk_buff *skb)
469 {
470 	pr_err(MOD "ARP failure\n");
471 	kfree_skb(skb);
472 }
473 
474 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
475 {
476 	pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
477 }
478 
479 enum {
480 	NUM_FAKE_CPLS = 2,
481 	FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
482 	FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
483 };
484 
485 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
486 {
487 	struct c4iw_ep *ep;
488 
489 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
490 	release_ep_resources(ep);
491 	return 0;
492 }
493 
494 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
495 {
496 	struct c4iw_ep *ep;
497 
498 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
499 	c4iw_put_ep(&ep->parent_ep->com);
500 	release_ep_resources(ep);
501 	return 0;
502 }
503 
504 /*
505  * Fake up a special CPL opcode and call sched() so process_work() will call
506  * _put_ep_safe() in a safe context to free the ep resources.  This is needed
507  * because ARP error handlers are called in an ATOMIC context, and
508  * _c4iw_free_ep() needs to block.
509  */
510 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
511 				  int cpl)
512 {
513 	struct cpl_act_establish *rpl = cplhdr(skb);
514 
515 	/* Set our special ARP_FAILURE opcode */
516 	rpl->ot.opcode = cpl;
517 
518 	/*
519 	 * Save ep in the skb->cb area, after where sched() will save the dev
520 	 * ptr.
521 	 */
522 	*((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
523 	sched(ep->com.dev, skb);
524 }
525 
526 /* Handle an ARP failure for an accept */
527 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
528 {
529 	struct c4iw_ep *ep = handle;
530 
531 	pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
532 	       ep->hwtid);
533 
534 	__state_set(&ep->com, DEAD);
535 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
536 }
537 
538 /*
539  * Handle an ARP failure for an active open.
540  */
541 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
542 {
543 	struct c4iw_ep *ep = handle;
544 
545 	printk(KERN_ERR MOD "ARP failure during connect\n");
546 	connect_reply_upcall(ep, -EHOSTUNREACH);
547 	__state_set(&ep->com, DEAD);
548 	if (ep->com.remote_addr.ss_family == AF_INET6) {
549 		struct sockaddr_in6 *sin6 =
550 			(struct sockaddr_in6 *)&ep->com.local_addr;
551 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
552 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
553 	}
554 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
555 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
556 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
557 }
558 
559 /*
560  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
561  * and send it along.
562  */
563 static void abort_arp_failure(void *handle, struct sk_buff *skb)
564 {
565 	int ret;
566 	struct c4iw_ep *ep = handle;
567 	struct c4iw_rdev *rdev = &ep->com.dev->rdev;
568 	struct cpl_abort_req *req = cplhdr(skb);
569 
570 	PDBG("%s rdev %p\n", __func__, rdev);
571 	req->cmd = CPL_ABORT_NO_RST;
572 	ret = c4iw_ofld_send(rdev, skb);
573 	if (ret) {
574 		__state_set(&ep->com, DEAD);
575 		queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
576 	}
577 }
578 
579 static int send_flowc(struct c4iw_ep *ep)
580 {
581 	struct fw_flowc_wr *flowc;
582 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
583 	int i;
584 	u16 vlan = ep->l2t->vlan;
585 	int nparams;
586 
587 	if (WARN_ON(!skb))
588 		return -ENOMEM;
589 
590 	if (vlan == CPL_L2T_VLAN_NONE)
591 		nparams = 8;
592 	else
593 		nparams = 9;
594 
595 	flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
596 
597 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
598 					   FW_FLOWC_WR_NPARAMS_V(nparams));
599 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
600 					  16)) | FW_WR_FLOWID_V(ep->hwtid));
601 
602 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
603 	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
604 					    (ep->com.dev->rdev.lldi.pf));
605 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
606 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
607 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
608 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
609 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
610 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
611 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
612 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
613 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
614 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
615 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
616 	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
617 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
618 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
619 	if (nparams == 9) {
620 		u16 pri;
621 
622 		pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
623 		flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
624 		flowc->mnemval[8].val = cpu_to_be32(pri);
625 	} else {
626 		/* Pad WR to 16 byte boundary */
627 		flowc->mnemval[8].mnemonic = 0;
628 		flowc->mnemval[8].val = 0;
629 	}
630 	for (i = 0; i < 9; i++) {
631 		flowc->mnemval[i].r4[0] = 0;
632 		flowc->mnemval[i].r4[1] = 0;
633 		flowc->mnemval[i].r4[2] = 0;
634 	}
635 
636 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
637 	return c4iw_ofld_send(&ep->com.dev->rdev, skb);
638 }
639 
640 static int send_halfclose(struct c4iw_ep *ep)
641 {
642 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
643 	u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
644 
645 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
646 	if (WARN_ON(!skb))
647 		return -ENOMEM;
648 
649 	cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
650 			      NULL, arp_failure_discard);
651 
652 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
653 }
654 
655 static int send_abort(struct c4iw_ep *ep)
656 {
657 	u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
658 	struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
659 
660 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
661 	if (WARN_ON(!req_skb))
662 		return -ENOMEM;
663 
664 	cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
665 			  ep, abort_arp_failure);
666 
667 	return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
668 }
669 
670 static int send_connect(struct c4iw_ep *ep)
671 {
672 	struct cpl_act_open_req *req = NULL;
673 	struct cpl_t5_act_open_req *t5req = NULL;
674 	struct cpl_t6_act_open_req *t6req = NULL;
675 	struct cpl_act_open_req6 *req6 = NULL;
676 	struct cpl_t5_act_open_req6 *t5req6 = NULL;
677 	struct cpl_t6_act_open_req6 *t6req6 = NULL;
678 	struct sk_buff *skb;
679 	u64 opt0;
680 	u32 opt2;
681 	unsigned int mtu_idx;
682 	u32 wscale;
683 	int win, sizev4, sizev6, wrlen;
684 	struct sockaddr_in *la = (struct sockaddr_in *)
685 				 &ep->com.local_addr;
686 	struct sockaddr_in *ra = (struct sockaddr_in *)
687 				 &ep->com.remote_addr;
688 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
689 				   &ep->com.local_addr;
690 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
691 				   &ep->com.remote_addr;
692 	int ret;
693 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
694 	u32 isn = (prandom_u32() & ~7UL) - 1;
695 
696 	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
697 	case CHELSIO_T4:
698 		sizev4 = sizeof(struct cpl_act_open_req);
699 		sizev6 = sizeof(struct cpl_act_open_req6);
700 		break;
701 	case CHELSIO_T5:
702 		sizev4 = sizeof(struct cpl_t5_act_open_req);
703 		sizev6 = sizeof(struct cpl_t5_act_open_req6);
704 		break;
705 	case CHELSIO_T6:
706 		sizev4 = sizeof(struct cpl_t6_act_open_req);
707 		sizev6 = sizeof(struct cpl_t6_act_open_req6);
708 		break;
709 	default:
710 		pr_err("T%d Chip is not supported\n",
711 		       CHELSIO_CHIP_VERSION(adapter_type));
712 		return -EINVAL;
713 	}
714 
715 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
716 			roundup(sizev4, 16) :
717 			roundup(sizev6, 16);
718 
719 	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
720 
721 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
722 	if (!skb) {
723 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
724 		       __func__);
725 		return -ENOMEM;
726 	}
727 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
728 
729 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
730 		      enable_tcp_timestamps,
731 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
732 	wscale = cxgb_compute_wscale(rcv_win);
733 
734 	/*
735 	 * Specify the largest window that will fit in opt0. The
736 	 * remainder will be specified in the rx_data_ack.
737 	 */
738 	win = ep->rcv_win >> 10;
739 	if (win > RCV_BUFSIZ_M)
740 		win = RCV_BUFSIZ_M;
741 
742 	opt0 = (nocong ? NO_CONG_F : 0) |
743 	       KEEP_ALIVE_F |
744 	       DELACK_F |
745 	       WND_SCALE_V(wscale) |
746 	       MSS_IDX_V(mtu_idx) |
747 	       L2T_IDX_V(ep->l2t->idx) |
748 	       TX_CHAN_V(ep->tx_chan) |
749 	       SMAC_SEL_V(ep->smac_idx) |
750 	       DSCP_V(ep->tos >> 2) |
751 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
752 	       RCV_BUFSIZ_V(win);
753 	opt2 = RX_CHANNEL_V(0) |
754 	       CCTRL_ECN_V(enable_ecn) |
755 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
756 	if (enable_tcp_timestamps)
757 		opt2 |= TSTAMPS_EN_F;
758 	if (enable_tcp_sack)
759 		opt2 |= SACK_EN_F;
760 	if (wscale && enable_tcp_window_scaling)
761 		opt2 |= WND_SCALE_EN_F;
762 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
763 		if (peer2peer)
764 			isn += 4;
765 
766 		opt2 |= T5_OPT_2_VALID_F;
767 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
768 		opt2 |= T5_ISS_F;
769 	}
770 
771 	if (ep->com.remote_addr.ss_family == AF_INET6)
772 		cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
773 			       (const u32 *)&la6->sin6_addr.s6_addr, 1);
774 
775 	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
776 
777 	if (ep->com.remote_addr.ss_family == AF_INET) {
778 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
779 		case CHELSIO_T4:
780 			req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
781 			INIT_TP_WR(req, 0);
782 			break;
783 		case CHELSIO_T5:
784 			t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
785 					wrlen);
786 			INIT_TP_WR(t5req, 0);
787 			req = (struct cpl_act_open_req *)t5req;
788 			break;
789 		case CHELSIO_T6:
790 			t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
791 					wrlen);
792 			INIT_TP_WR(t6req, 0);
793 			req = (struct cpl_act_open_req *)t6req;
794 			t5req = (struct cpl_t5_act_open_req *)t6req;
795 			break;
796 		default:
797 			pr_err("T%d Chip is not supported\n",
798 			       CHELSIO_CHIP_VERSION(adapter_type));
799 			ret = -EINVAL;
800 			goto clip_release;
801 		}
802 
803 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
804 					((ep->rss_qid<<14) | ep->atid)));
805 		req->local_port = la->sin_port;
806 		req->peer_port = ra->sin_port;
807 		req->local_ip = la->sin_addr.s_addr;
808 		req->peer_ip = ra->sin_addr.s_addr;
809 		req->opt0 = cpu_to_be64(opt0);
810 
811 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
812 			req->params = cpu_to_be32(cxgb4_select_ntuple(
813 						ep->com.dev->rdev.lldi.ports[0],
814 						ep->l2t));
815 			req->opt2 = cpu_to_be32(opt2);
816 		} else {
817 			t5req->params = cpu_to_be64(FILTER_TUPLE_V(
818 						cxgb4_select_ntuple(
819 						ep->com.dev->rdev.lldi.ports[0],
820 						ep->l2t)));
821 			t5req->rsvd = cpu_to_be32(isn);
822 			PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
823 			t5req->opt2 = cpu_to_be32(opt2);
824 		}
825 	} else {
826 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
827 		case CHELSIO_T4:
828 			req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
829 			INIT_TP_WR(req6, 0);
830 			break;
831 		case CHELSIO_T5:
832 			t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
833 					wrlen);
834 			INIT_TP_WR(t5req6, 0);
835 			req6 = (struct cpl_act_open_req6 *)t5req6;
836 			break;
837 		case CHELSIO_T6:
838 			t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
839 					wrlen);
840 			INIT_TP_WR(t6req6, 0);
841 			req6 = (struct cpl_act_open_req6 *)t6req6;
842 			t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
843 			break;
844 		default:
845 			pr_err("T%d Chip is not supported\n",
846 			       CHELSIO_CHIP_VERSION(adapter_type));
847 			ret = -EINVAL;
848 			goto clip_release;
849 		}
850 
851 		OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
852 					((ep->rss_qid<<14)|ep->atid)));
853 		req6->local_port = la6->sin6_port;
854 		req6->peer_port = ra6->sin6_port;
855 		req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
856 		req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
857 		req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
858 		req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
859 		req6->opt0 = cpu_to_be64(opt0);
860 
861 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
862 			req6->params = cpu_to_be32(cxgb4_select_ntuple(
863 						ep->com.dev->rdev.lldi.ports[0],
864 						ep->l2t));
865 			req6->opt2 = cpu_to_be32(opt2);
866 		} else {
867 			t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
868 						cxgb4_select_ntuple(
869 						ep->com.dev->rdev.lldi.ports[0],
870 						ep->l2t)));
871 			t5req6->rsvd = cpu_to_be32(isn);
872 			PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
873 			t5req6->opt2 = cpu_to_be32(opt2);
874 		}
875 	}
876 
877 	set_bit(ACT_OPEN_REQ, &ep->com.history);
878 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
879 clip_release:
880 	if (ret && ep->com.remote_addr.ss_family == AF_INET6)
881 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
882 				   (const u32 *)&la6->sin6_addr.s6_addr, 1);
883 	return ret;
884 }
885 
886 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
887 			u8 mpa_rev_to_use)
888 {
889 	int mpalen, wrlen, ret;
890 	struct fw_ofld_tx_data_wr *req;
891 	struct mpa_message *mpa;
892 	struct mpa_v2_conn_params mpa_v2_params;
893 
894 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
895 
896 	BUG_ON(skb_cloned(skb));
897 
898 	mpalen = sizeof(*mpa) + ep->plen;
899 	if (mpa_rev_to_use == 2)
900 		mpalen += sizeof(struct mpa_v2_conn_params);
901 	wrlen = roundup(mpalen + sizeof *req, 16);
902 	skb = get_skb(skb, wrlen, GFP_KERNEL);
903 	if (!skb) {
904 		connect_reply_upcall(ep, -ENOMEM);
905 		return -ENOMEM;
906 	}
907 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
908 
909 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
910 	memset(req, 0, wrlen);
911 	req->op_to_immdlen = cpu_to_be32(
912 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
913 		FW_WR_COMPL_F |
914 		FW_WR_IMMDLEN_V(mpalen));
915 	req->flowid_len16 = cpu_to_be32(
916 		FW_WR_FLOWID_V(ep->hwtid) |
917 		FW_WR_LEN16_V(wrlen >> 4));
918 	req->plen = cpu_to_be32(mpalen);
919 	req->tunnel_to_proxy = cpu_to_be32(
920 		FW_OFLD_TX_DATA_WR_FLUSH_F |
921 		FW_OFLD_TX_DATA_WR_SHOVE_F);
922 
923 	mpa = (struct mpa_message *)(req + 1);
924 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
925 
926 	mpa->flags = 0;
927 	if (crc_enabled)
928 		mpa->flags |= MPA_CRC;
929 	if (markers_enabled) {
930 		mpa->flags |= MPA_MARKERS;
931 		ep->mpa_attr.recv_marker_enabled = 1;
932 	} else {
933 		ep->mpa_attr.recv_marker_enabled = 0;
934 	}
935 	if (mpa_rev_to_use == 2)
936 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
937 
938 	mpa->private_data_size = htons(ep->plen);
939 	mpa->revision = mpa_rev_to_use;
940 	if (mpa_rev_to_use == 1) {
941 		ep->tried_with_mpa_v1 = 1;
942 		ep->retry_with_mpa_v1 = 0;
943 	}
944 
945 	if (mpa_rev_to_use == 2) {
946 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
947 					       sizeof (struct mpa_v2_conn_params));
948 		PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
949 		     ep->ord);
950 		mpa_v2_params.ird = htons((u16)ep->ird);
951 		mpa_v2_params.ord = htons((u16)ep->ord);
952 
953 		if (peer2peer) {
954 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
955 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
956 				mpa_v2_params.ord |=
957 					htons(MPA_V2_RDMA_WRITE_RTR);
958 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
959 				mpa_v2_params.ord |=
960 					htons(MPA_V2_RDMA_READ_RTR);
961 		}
962 		memcpy(mpa->private_data, &mpa_v2_params,
963 		       sizeof(struct mpa_v2_conn_params));
964 
965 		if (ep->plen)
966 			memcpy(mpa->private_data +
967 			       sizeof(struct mpa_v2_conn_params),
968 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
969 	} else
970 		if (ep->plen)
971 			memcpy(mpa->private_data,
972 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
973 
974 	/*
975 	 * Reference the mpa skb.  This ensures the data area
976 	 * will remain in memory until the hw acks the tx.
977 	 * Function fw4_ack() will deref it.
978 	 */
979 	skb_get(skb);
980 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
981 	BUG_ON(ep->mpa_skb);
982 	ep->mpa_skb = skb;
983 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
984 	if (ret)
985 		return ret;
986 	start_ep_timer(ep);
987 	__state_set(&ep->com, MPA_REQ_SENT);
988 	ep->mpa_attr.initiator = 1;
989 	ep->snd_seq += mpalen;
990 	return ret;
991 }
992 
993 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
994 {
995 	int mpalen, wrlen;
996 	struct fw_ofld_tx_data_wr *req;
997 	struct mpa_message *mpa;
998 	struct sk_buff *skb;
999 	struct mpa_v2_conn_params mpa_v2_params;
1000 
1001 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1002 
1003 	mpalen = sizeof(*mpa) + plen;
1004 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1005 		mpalen += sizeof(struct mpa_v2_conn_params);
1006 	wrlen = roundup(mpalen + sizeof *req, 16);
1007 
1008 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1009 	if (!skb) {
1010 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1011 		return -ENOMEM;
1012 	}
1013 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1014 
1015 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1016 	memset(req, 0, wrlen);
1017 	req->op_to_immdlen = cpu_to_be32(
1018 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1019 		FW_WR_COMPL_F |
1020 		FW_WR_IMMDLEN_V(mpalen));
1021 	req->flowid_len16 = cpu_to_be32(
1022 		FW_WR_FLOWID_V(ep->hwtid) |
1023 		FW_WR_LEN16_V(wrlen >> 4));
1024 	req->plen = cpu_to_be32(mpalen);
1025 	req->tunnel_to_proxy = cpu_to_be32(
1026 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1027 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1028 
1029 	mpa = (struct mpa_message *)(req + 1);
1030 	memset(mpa, 0, sizeof(*mpa));
1031 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1032 	mpa->flags = MPA_REJECT;
1033 	mpa->revision = ep->mpa_attr.version;
1034 	mpa->private_data_size = htons(plen);
1035 
1036 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1037 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1038 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1039 					       sizeof (struct mpa_v2_conn_params));
1040 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1041 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1042 					   0));
1043 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1044 					  (p2p_type ==
1045 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1046 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1047 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
1048 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
1049 		memcpy(mpa->private_data, &mpa_v2_params,
1050 		       sizeof(struct mpa_v2_conn_params));
1051 
1052 		if (ep->plen)
1053 			memcpy(mpa->private_data +
1054 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1055 	} else
1056 		if (plen)
1057 			memcpy(mpa->private_data, pdata, plen);
1058 
1059 	/*
1060 	 * Reference the mpa skb again.  This ensures the data area
1061 	 * will remain in memory until the hw acks the tx.
1062 	 * Function fw4_ack() will deref it.
1063 	 */
1064 	skb_get(skb);
1065 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1066 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1067 	BUG_ON(ep->mpa_skb);
1068 	ep->mpa_skb = skb;
1069 	ep->snd_seq += mpalen;
1070 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1071 }
1072 
1073 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1074 {
1075 	int mpalen, wrlen;
1076 	struct fw_ofld_tx_data_wr *req;
1077 	struct mpa_message *mpa;
1078 	struct sk_buff *skb;
1079 	struct mpa_v2_conn_params mpa_v2_params;
1080 
1081 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1082 
1083 	mpalen = sizeof(*mpa) + plen;
1084 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1085 		mpalen += sizeof(struct mpa_v2_conn_params);
1086 	wrlen = roundup(mpalen + sizeof *req, 16);
1087 
1088 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1089 	if (!skb) {
1090 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1091 		return -ENOMEM;
1092 	}
1093 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1094 
1095 	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1096 	memset(req, 0, wrlen);
1097 	req->op_to_immdlen = cpu_to_be32(
1098 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1099 		FW_WR_COMPL_F |
1100 		FW_WR_IMMDLEN_V(mpalen));
1101 	req->flowid_len16 = cpu_to_be32(
1102 		FW_WR_FLOWID_V(ep->hwtid) |
1103 		FW_WR_LEN16_V(wrlen >> 4));
1104 	req->plen = cpu_to_be32(mpalen);
1105 	req->tunnel_to_proxy = cpu_to_be32(
1106 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1107 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1108 
1109 	mpa = (struct mpa_message *)(req + 1);
1110 	memset(mpa, 0, sizeof(*mpa));
1111 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1112 	mpa->flags = 0;
1113 	if (ep->mpa_attr.crc_enabled)
1114 		mpa->flags |= MPA_CRC;
1115 	if (ep->mpa_attr.recv_marker_enabled)
1116 		mpa->flags |= MPA_MARKERS;
1117 	mpa->revision = ep->mpa_attr.version;
1118 	mpa->private_data_size = htons(plen);
1119 
1120 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1121 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1122 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1123 					       sizeof (struct mpa_v2_conn_params));
1124 		mpa_v2_params.ird = htons((u16)ep->ird);
1125 		mpa_v2_params.ord = htons((u16)ep->ord);
1126 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1127 					FW_RI_INIT_P2PTYPE_DISABLED)) {
1128 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1129 
1130 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1131 				mpa_v2_params.ord |=
1132 					htons(MPA_V2_RDMA_WRITE_RTR);
1133 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1134 				mpa_v2_params.ord |=
1135 					htons(MPA_V2_RDMA_READ_RTR);
1136 		}
1137 
1138 		memcpy(mpa->private_data, &mpa_v2_params,
1139 		       sizeof(struct mpa_v2_conn_params));
1140 
1141 		if (ep->plen)
1142 			memcpy(mpa->private_data +
1143 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1144 	} else
1145 		if (plen)
1146 			memcpy(mpa->private_data, pdata, plen);
1147 
1148 	/*
1149 	 * Reference the mpa skb.  This ensures the data area
1150 	 * will remain in memory until the hw acks the tx.
1151 	 * Function fw4_ack() will deref it.
1152 	 */
1153 	skb_get(skb);
1154 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1155 	ep->mpa_skb = skb;
1156 	__state_set(&ep->com, MPA_REP_SENT);
1157 	ep->snd_seq += mpalen;
1158 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1159 }
1160 
1161 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1162 {
1163 	struct c4iw_ep *ep;
1164 	struct cpl_act_establish *req = cplhdr(skb);
1165 	unsigned int tid = GET_TID(req);
1166 	unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1167 	struct tid_info *t = dev->rdev.lldi.tids;
1168 	int ret;
1169 
1170 	ep = lookup_atid(t, atid);
1171 
1172 	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1173 	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1174 
1175 	mutex_lock(&ep->com.mutex);
1176 	dst_confirm(ep->dst);
1177 
1178 	/* setup the hwtid for this connection */
1179 	ep->hwtid = tid;
1180 	cxgb4_insert_tid(t, ep, tid);
1181 	insert_ep_tid(ep);
1182 
1183 	ep->snd_seq = be32_to_cpu(req->snd_isn);
1184 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1185 
1186 	set_emss(ep, ntohs(req->tcp_opt));
1187 
1188 	/* dealloc the atid */
1189 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1190 	cxgb4_free_atid(t, atid);
1191 	set_bit(ACT_ESTAB, &ep->com.history);
1192 
1193 	/* start MPA negotiation */
1194 	ret = send_flowc(ep);
1195 	if (ret)
1196 		goto err;
1197 	if (ep->retry_with_mpa_v1)
1198 		ret = send_mpa_req(ep, skb, 1);
1199 	else
1200 		ret = send_mpa_req(ep, skb, mpa_rev);
1201 	if (ret)
1202 		goto err;
1203 	mutex_unlock(&ep->com.mutex);
1204 	return 0;
1205 err:
1206 	mutex_unlock(&ep->com.mutex);
1207 	connect_reply_upcall(ep, -ENOMEM);
1208 	c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1209 	return 0;
1210 }
1211 
1212 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1213 {
1214 	struct iw_cm_event event;
1215 
1216 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1217 	memset(&event, 0, sizeof(event));
1218 	event.event = IW_CM_EVENT_CLOSE;
1219 	event.status = status;
1220 	if (ep->com.cm_id) {
1221 		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1222 		     ep, ep->com.cm_id, ep->hwtid);
1223 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1224 		deref_cm_id(&ep->com);
1225 		set_bit(CLOSE_UPCALL, &ep->com.history);
1226 	}
1227 }
1228 
1229 static void peer_close_upcall(struct c4iw_ep *ep)
1230 {
1231 	struct iw_cm_event event;
1232 
1233 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1234 	memset(&event, 0, sizeof(event));
1235 	event.event = IW_CM_EVENT_DISCONNECT;
1236 	if (ep->com.cm_id) {
1237 		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1238 		     ep, ep->com.cm_id, ep->hwtid);
1239 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1240 		set_bit(DISCONN_UPCALL, &ep->com.history);
1241 	}
1242 }
1243 
1244 static void peer_abort_upcall(struct c4iw_ep *ep)
1245 {
1246 	struct iw_cm_event event;
1247 
1248 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1249 	memset(&event, 0, sizeof(event));
1250 	event.event = IW_CM_EVENT_CLOSE;
1251 	event.status = -ECONNRESET;
1252 	if (ep->com.cm_id) {
1253 		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1254 		     ep->com.cm_id, ep->hwtid);
1255 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1256 		deref_cm_id(&ep->com);
1257 		set_bit(ABORT_UPCALL, &ep->com.history);
1258 	}
1259 }
1260 
1261 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1262 {
1263 	struct iw_cm_event event;
1264 
1265 	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1266 	memset(&event, 0, sizeof(event));
1267 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1268 	event.status = status;
1269 	memcpy(&event.local_addr, &ep->com.local_addr,
1270 	       sizeof(ep->com.local_addr));
1271 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1272 	       sizeof(ep->com.remote_addr));
1273 
1274 	if ((status == 0) || (status == -ECONNREFUSED)) {
1275 		if (!ep->tried_with_mpa_v1) {
1276 			/* this means MPA_v2 is used */
1277 			event.ord = ep->ird;
1278 			event.ird = ep->ord;
1279 			event.private_data_len = ep->plen -
1280 				sizeof(struct mpa_v2_conn_params);
1281 			event.private_data = ep->mpa_pkt +
1282 				sizeof(struct mpa_message) +
1283 				sizeof(struct mpa_v2_conn_params);
1284 		} else {
1285 			/* this means MPA_v1 is used */
1286 			event.ord = cur_max_read_depth(ep->com.dev);
1287 			event.ird = cur_max_read_depth(ep->com.dev);
1288 			event.private_data_len = ep->plen;
1289 			event.private_data = ep->mpa_pkt +
1290 				sizeof(struct mpa_message);
1291 		}
1292 	}
1293 
1294 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1295 	     ep->hwtid, status);
1296 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1297 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1298 
1299 	if (status < 0)
1300 		deref_cm_id(&ep->com);
1301 }
1302 
1303 static int connect_request_upcall(struct c4iw_ep *ep)
1304 {
1305 	struct iw_cm_event event;
1306 	int ret;
1307 
1308 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1309 	memset(&event, 0, sizeof(event));
1310 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1311 	memcpy(&event.local_addr, &ep->com.local_addr,
1312 	       sizeof(ep->com.local_addr));
1313 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1314 	       sizeof(ep->com.remote_addr));
1315 	event.provider_data = ep;
1316 	if (!ep->tried_with_mpa_v1) {
1317 		/* this means MPA_v2 is used */
1318 		event.ord = ep->ord;
1319 		event.ird = ep->ird;
1320 		event.private_data_len = ep->plen -
1321 			sizeof(struct mpa_v2_conn_params);
1322 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1323 			sizeof(struct mpa_v2_conn_params);
1324 	} else {
1325 		/* this means MPA_v1 is used. Send max supported */
1326 		event.ord = cur_max_read_depth(ep->com.dev);
1327 		event.ird = cur_max_read_depth(ep->com.dev);
1328 		event.private_data_len = ep->plen;
1329 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1330 	}
1331 	c4iw_get_ep(&ep->com);
1332 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1333 						      &event);
1334 	if (ret)
1335 		c4iw_put_ep(&ep->com);
1336 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1337 	c4iw_put_ep(&ep->parent_ep->com);
1338 	return ret;
1339 }
1340 
1341 static void established_upcall(struct c4iw_ep *ep)
1342 {
1343 	struct iw_cm_event event;
1344 
1345 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1346 	memset(&event, 0, sizeof(event));
1347 	event.event = IW_CM_EVENT_ESTABLISHED;
1348 	event.ird = ep->ord;
1349 	event.ord = ep->ird;
1350 	if (ep->com.cm_id) {
1351 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1352 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1353 		set_bit(ESTAB_UPCALL, &ep->com.history);
1354 	}
1355 }
1356 
1357 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1358 {
1359 	struct sk_buff *skb;
1360 	u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1361 	u32 credit_dack;
1362 
1363 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1364 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1365 	if (!skb) {
1366 		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1367 		return 0;
1368 	}
1369 
1370 	/*
1371 	 * If we couldn't specify the entire rcv window at connection setup
1372 	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1373 	 * then add the overage in to the credits returned.
1374 	 */
1375 	if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1376 		credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1377 
1378 	credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1379 		      RX_DACK_MODE_V(dack_mode);
1380 
1381 	cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1382 			    credit_dack);
1383 
1384 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1385 	return credits;
1386 }
1387 
1388 #define RELAXED_IRD_NEGOTIATION 1
1389 
1390 /*
1391  * process_mpa_reply - process streaming mode MPA reply
1392  *
1393  * Returns:
1394  *
1395  * 0 upon success indicating a connect request was delivered to the ULP
1396  * or the mpa request is incomplete but valid so far.
1397  *
1398  * 1 if a failure requires the caller to close the connection.
1399  *
1400  * 2 if a failure requires the caller to abort the connection.
1401  */
1402 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1403 {
1404 	struct mpa_message *mpa;
1405 	struct mpa_v2_conn_params *mpa_v2_params;
1406 	u16 plen;
1407 	u16 resp_ird, resp_ord;
1408 	u8 rtr_mismatch = 0, insuff_ird = 0;
1409 	struct c4iw_qp_attributes attrs;
1410 	enum c4iw_qp_attr_mask mask;
1411 	int err;
1412 	int disconnect = 0;
1413 
1414 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1415 
1416 	/*
1417 	 * If we get more than the supported amount of private data
1418 	 * then we must fail this connection.
1419 	 */
1420 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1421 		err = -EINVAL;
1422 		goto err_stop_timer;
1423 	}
1424 
1425 	/*
1426 	 * copy the new data into our accumulation buffer.
1427 	 */
1428 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1429 				  skb->len);
1430 	ep->mpa_pkt_len += skb->len;
1431 
1432 	/*
1433 	 * if we don't even have the mpa message, then bail.
1434 	 */
1435 	if (ep->mpa_pkt_len < sizeof(*mpa))
1436 		return 0;
1437 	mpa = (struct mpa_message *) ep->mpa_pkt;
1438 
1439 	/* Validate MPA header. */
1440 	if (mpa->revision > mpa_rev) {
1441 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1442 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1443 		err = -EPROTO;
1444 		goto err_stop_timer;
1445 	}
1446 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1447 		err = -EPROTO;
1448 		goto err_stop_timer;
1449 	}
1450 
1451 	plen = ntohs(mpa->private_data_size);
1452 
1453 	/*
1454 	 * Fail if there's too much private data.
1455 	 */
1456 	if (plen > MPA_MAX_PRIVATE_DATA) {
1457 		err = -EPROTO;
1458 		goto err_stop_timer;
1459 	}
1460 
1461 	/*
1462 	 * If plen does not account for pkt size
1463 	 */
1464 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1465 		err = -EPROTO;
1466 		goto err_stop_timer;
1467 	}
1468 
1469 	ep->plen = (u8) plen;
1470 
1471 	/*
1472 	 * If we don't have all the pdata yet, then bail.
1473 	 * We'll continue process when more data arrives.
1474 	 */
1475 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1476 		return 0;
1477 
1478 	if (mpa->flags & MPA_REJECT) {
1479 		err = -ECONNREFUSED;
1480 		goto err_stop_timer;
1481 	}
1482 
1483 	/*
1484 	 * Stop mpa timer.  If it expired, then
1485 	 * we ignore the MPA reply.  process_timeout()
1486 	 * will abort the connection.
1487 	 */
1488 	if (stop_ep_timer(ep))
1489 		return 0;
1490 
1491 	/*
1492 	 * If we get here we have accumulated the entire mpa
1493 	 * start reply message including private data. And
1494 	 * the MPA header is valid.
1495 	 */
1496 	__state_set(&ep->com, FPDU_MODE);
1497 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1498 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1499 	ep->mpa_attr.version = mpa->revision;
1500 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1501 
1502 	if (mpa->revision == 2) {
1503 		ep->mpa_attr.enhanced_rdma_conn =
1504 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1505 		if (ep->mpa_attr.enhanced_rdma_conn) {
1506 			mpa_v2_params = (struct mpa_v2_conn_params *)
1507 				(ep->mpa_pkt + sizeof(*mpa));
1508 			resp_ird = ntohs(mpa_v2_params->ird) &
1509 				MPA_V2_IRD_ORD_MASK;
1510 			resp_ord = ntohs(mpa_v2_params->ord) &
1511 				MPA_V2_IRD_ORD_MASK;
1512 			PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1513 			     __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1514 
1515 			/*
1516 			 * This is a double-check. Ideally, below checks are
1517 			 * not required since ird/ord stuff has been taken
1518 			 * care of in c4iw_accept_cr
1519 			 */
1520 			if (ep->ird < resp_ord) {
1521 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1522 				    ep->com.dev->rdev.lldi.max_ordird_qp)
1523 					ep->ird = resp_ord;
1524 				else
1525 					insuff_ird = 1;
1526 			} else if (ep->ird > resp_ord) {
1527 				ep->ird = resp_ord;
1528 			}
1529 			if (ep->ord > resp_ird) {
1530 				if (RELAXED_IRD_NEGOTIATION)
1531 					ep->ord = resp_ird;
1532 				else
1533 					insuff_ird = 1;
1534 			}
1535 			if (insuff_ird) {
1536 				err = -ENOMEM;
1537 				ep->ird = resp_ord;
1538 				ep->ord = resp_ird;
1539 			}
1540 
1541 			if (ntohs(mpa_v2_params->ird) &
1542 					MPA_V2_PEER2PEER_MODEL) {
1543 				if (ntohs(mpa_v2_params->ord) &
1544 						MPA_V2_RDMA_WRITE_RTR)
1545 					ep->mpa_attr.p2p_type =
1546 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1547 				else if (ntohs(mpa_v2_params->ord) &
1548 						MPA_V2_RDMA_READ_RTR)
1549 					ep->mpa_attr.p2p_type =
1550 						FW_RI_INIT_P2PTYPE_READ_REQ;
1551 			}
1552 		}
1553 	} else if (mpa->revision == 1)
1554 		if (peer2peer)
1555 			ep->mpa_attr.p2p_type = p2p_type;
1556 
1557 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1558 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1559 	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
1560 	     ep->mpa_attr.recv_marker_enabled,
1561 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1562 	     ep->mpa_attr.p2p_type, p2p_type);
1563 
1564 	/*
1565 	 * If responder's RTR does not match with that of initiator, assign
1566 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1567 	 * generated when moving QP to RTS state.
1568 	 * A TERM message will be sent after QP has moved to RTS state
1569 	 */
1570 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1571 			(ep->mpa_attr.p2p_type != p2p_type)) {
1572 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1573 		rtr_mismatch = 1;
1574 	}
1575 
1576 	attrs.mpa_attr = ep->mpa_attr;
1577 	attrs.max_ird = ep->ird;
1578 	attrs.max_ord = ep->ord;
1579 	attrs.llp_stream_handle = ep;
1580 	attrs.next_state = C4IW_QP_STATE_RTS;
1581 
1582 	mask = C4IW_QP_ATTR_NEXT_STATE |
1583 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1584 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1585 
1586 	/* bind QP and TID with INIT_WR */
1587 	err = c4iw_modify_qp(ep->com.qp->rhp,
1588 			     ep->com.qp, mask, &attrs, 1);
1589 	if (err)
1590 		goto err;
1591 
1592 	/*
1593 	 * If responder's RTR requirement did not match with what initiator
1594 	 * supports, generate TERM message
1595 	 */
1596 	if (rtr_mismatch) {
1597 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1598 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1599 		attrs.ecode = MPA_NOMATCH_RTR;
1600 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1601 		attrs.send_term = 1;
1602 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1603 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1604 		err = -ENOMEM;
1605 		disconnect = 1;
1606 		goto out;
1607 	}
1608 
1609 	/*
1610 	 * Generate TERM if initiator IRD is not sufficient for responder
1611 	 * provided ORD. Currently, we do the same behaviour even when
1612 	 * responder provided IRD is also not sufficient as regards to
1613 	 * initiator ORD.
1614 	 */
1615 	if (insuff_ird) {
1616 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1617 				__func__);
1618 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1619 		attrs.ecode = MPA_INSUFF_IRD;
1620 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1621 		attrs.send_term = 1;
1622 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1623 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1624 		err = -ENOMEM;
1625 		disconnect = 1;
1626 		goto out;
1627 	}
1628 	goto out;
1629 err_stop_timer:
1630 	stop_ep_timer(ep);
1631 err:
1632 	disconnect = 2;
1633 out:
1634 	connect_reply_upcall(ep, err);
1635 	return disconnect;
1636 }
1637 
1638 /*
1639  * process_mpa_request - process streaming mode MPA request
1640  *
1641  * Returns:
1642  *
1643  * 0 upon success indicating a connect request was delivered to the ULP
1644  * or the mpa request is incomplete but valid so far.
1645  *
1646  * 1 if a failure requires the caller to close the connection.
1647  *
1648  * 2 if a failure requires the caller to abort the connection.
1649  */
1650 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1651 {
1652 	struct mpa_message *mpa;
1653 	struct mpa_v2_conn_params *mpa_v2_params;
1654 	u16 plen;
1655 
1656 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1657 
1658 	/*
1659 	 * If we get more than the supported amount of private data
1660 	 * then we must fail this connection.
1661 	 */
1662 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1663 		goto err_stop_timer;
1664 
1665 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1666 
1667 	/*
1668 	 * Copy the new data into our accumulation buffer.
1669 	 */
1670 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1671 				  skb->len);
1672 	ep->mpa_pkt_len += skb->len;
1673 
1674 	/*
1675 	 * If we don't even have the mpa message, then bail.
1676 	 * We'll continue process when more data arrives.
1677 	 */
1678 	if (ep->mpa_pkt_len < sizeof(*mpa))
1679 		return 0;
1680 
1681 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1682 	mpa = (struct mpa_message *) ep->mpa_pkt;
1683 
1684 	/*
1685 	 * Validate MPA Header.
1686 	 */
1687 	if (mpa->revision > mpa_rev) {
1688 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1689 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1690 		goto err_stop_timer;
1691 	}
1692 
1693 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1694 		goto err_stop_timer;
1695 
1696 	plen = ntohs(mpa->private_data_size);
1697 
1698 	/*
1699 	 * Fail if there's too much private data.
1700 	 */
1701 	if (plen > MPA_MAX_PRIVATE_DATA)
1702 		goto err_stop_timer;
1703 
1704 	/*
1705 	 * If plen does not account for pkt size
1706 	 */
1707 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1708 		goto err_stop_timer;
1709 	ep->plen = (u8) plen;
1710 
1711 	/*
1712 	 * If we don't have all the pdata yet, then bail.
1713 	 */
1714 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1715 		return 0;
1716 
1717 	/*
1718 	 * If we get here we have accumulated the entire mpa
1719 	 * start reply message including private data.
1720 	 */
1721 	ep->mpa_attr.initiator = 0;
1722 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1723 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1724 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1725 	ep->mpa_attr.version = mpa->revision;
1726 	if (mpa->revision == 1)
1727 		ep->tried_with_mpa_v1 = 1;
1728 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1729 
1730 	if (mpa->revision == 2) {
1731 		ep->mpa_attr.enhanced_rdma_conn =
1732 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1733 		if (ep->mpa_attr.enhanced_rdma_conn) {
1734 			mpa_v2_params = (struct mpa_v2_conn_params *)
1735 				(ep->mpa_pkt + sizeof(*mpa));
1736 			ep->ird = ntohs(mpa_v2_params->ird) &
1737 				MPA_V2_IRD_ORD_MASK;
1738 			ep->ird = min_t(u32, ep->ird,
1739 					cur_max_read_depth(ep->com.dev));
1740 			ep->ord = ntohs(mpa_v2_params->ord) &
1741 				MPA_V2_IRD_ORD_MASK;
1742 			ep->ord = min_t(u32, ep->ord,
1743 					cur_max_read_depth(ep->com.dev));
1744 			PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1745 			     ep->ord);
1746 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1747 				if (peer2peer) {
1748 					if (ntohs(mpa_v2_params->ord) &
1749 							MPA_V2_RDMA_WRITE_RTR)
1750 						ep->mpa_attr.p2p_type =
1751 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1752 					else if (ntohs(mpa_v2_params->ord) &
1753 							MPA_V2_RDMA_READ_RTR)
1754 						ep->mpa_attr.p2p_type =
1755 						FW_RI_INIT_P2PTYPE_READ_REQ;
1756 				}
1757 		}
1758 	} else if (mpa->revision == 1)
1759 		if (peer2peer)
1760 			ep->mpa_attr.p2p_type = p2p_type;
1761 
1762 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1763 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1764 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1765 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1766 	     ep->mpa_attr.p2p_type);
1767 
1768 	__state_set(&ep->com, MPA_REQ_RCVD);
1769 
1770 	/* drive upcall */
1771 	mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1772 	if (ep->parent_ep->com.state != DEAD) {
1773 		if (connect_request_upcall(ep))
1774 			goto err_unlock_parent;
1775 	} else {
1776 		goto err_unlock_parent;
1777 	}
1778 	mutex_unlock(&ep->parent_ep->com.mutex);
1779 	return 0;
1780 
1781 err_unlock_parent:
1782 	mutex_unlock(&ep->parent_ep->com.mutex);
1783 	goto err_out;
1784 err_stop_timer:
1785 	(void)stop_ep_timer(ep);
1786 err_out:
1787 	return 2;
1788 }
1789 
1790 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1791 {
1792 	struct c4iw_ep *ep;
1793 	struct cpl_rx_data *hdr = cplhdr(skb);
1794 	unsigned int dlen = ntohs(hdr->len);
1795 	unsigned int tid = GET_TID(hdr);
1796 	__u8 status = hdr->status;
1797 	int disconnect = 0;
1798 
1799 	ep = get_ep_from_tid(dev, tid);
1800 	if (!ep)
1801 		return 0;
1802 	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1803 	skb_pull(skb, sizeof(*hdr));
1804 	skb_trim(skb, dlen);
1805 	mutex_lock(&ep->com.mutex);
1806 
1807 	/* update RX credits */
1808 	update_rx_credits(ep, dlen);
1809 
1810 	switch (ep->com.state) {
1811 	case MPA_REQ_SENT:
1812 		ep->rcv_seq += dlen;
1813 		disconnect = process_mpa_reply(ep, skb);
1814 		break;
1815 	case MPA_REQ_WAIT:
1816 		ep->rcv_seq += dlen;
1817 		disconnect = process_mpa_request(ep, skb);
1818 		break;
1819 	case FPDU_MODE: {
1820 		struct c4iw_qp_attributes attrs;
1821 		BUG_ON(!ep->com.qp);
1822 		if (status)
1823 			pr_err("%s Unexpected streaming data." \
1824 			       " qpid %u ep %p state %d tid %u status %d\n",
1825 			       __func__, ep->com.qp->wq.sq.qid, ep,
1826 			       ep->com.state, ep->hwtid, status);
1827 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1828 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1829 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1830 		disconnect = 1;
1831 		break;
1832 	}
1833 	default:
1834 		break;
1835 	}
1836 	mutex_unlock(&ep->com.mutex);
1837 	if (disconnect)
1838 		c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1839 	c4iw_put_ep(&ep->com);
1840 	return 0;
1841 }
1842 
1843 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1844 {
1845 	struct c4iw_ep *ep;
1846 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1847 	int release = 0;
1848 	unsigned int tid = GET_TID(rpl);
1849 
1850 	ep = get_ep_from_tid(dev, tid);
1851 	if (!ep) {
1852 		printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1853 		return 0;
1854 	}
1855 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1856 	mutex_lock(&ep->com.mutex);
1857 	switch (ep->com.state) {
1858 	case ABORTING:
1859 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1860 		__state_set(&ep->com, DEAD);
1861 		release = 1;
1862 		break;
1863 	default:
1864 		printk(KERN_ERR "%s ep %p state %d\n",
1865 		     __func__, ep, ep->com.state);
1866 		break;
1867 	}
1868 	mutex_unlock(&ep->com.mutex);
1869 
1870 	if (release)
1871 		release_ep_resources(ep);
1872 	c4iw_put_ep(&ep->com);
1873 	return 0;
1874 }
1875 
1876 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1877 {
1878 	struct sk_buff *skb;
1879 	struct fw_ofld_connection_wr *req;
1880 	unsigned int mtu_idx;
1881 	u32 wscale;
1882 	struct sockaddr_in *sin;
1883 	int win;
1884 
1885 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1886 	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1887 	memset(req, 0, sizeof(*req));
1888 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1889 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1890 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1891 				     ep->com.dev->rdev.lldi.ports[0],
1892 				     ep->l2t));
1893 	sin = (struct sockaddr_in *)&ep->com.local_addr;
1894 	req->le.lport = sin->sin_port;
1895 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1896 	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1897 	req->le.pport = sin->sin_port;
1898 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1899 	req->tcb.t_state_to_astid =
1900 			htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1901 			FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1902 	req->tcb.cplrxdataack_cplpassacceptrpl =
1903 			htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1904 	req->tcb.tx_max = (__force __be32) jiffies;
1905 	req->tcb.rcv_adv = htons(1);
1906 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1907 		      enable_tcp_timestamps,
1908 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1909 	wscale = cxgb_compute_wscale(rcv_win);
1910 
1911 	/*
1912 	 * Specify the largest window that will fit in opt0. The
1913 	 * remainder will be specified in the rx_data_ack.
1914 	 */
1915 	win = ep->rcv_win >> 10;
1916 	if (win > RCV_BUFSIZ_M)
1917 		win = RCV_BUFSIZ_M;
1918 
1919 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1920 		(nocong ? NO_CONG_F : 0) |
1921 		KEEP_ALIVE_F |
1922 		DELACK_F |
1923 		WND_SCALE_V(wscale) |
1924 		MSS_IDX_V(mtu_idx) |
1925 		L2T_IDX_V(ep->l2t->idx) |
1926 		TX_CHAN_V(ep->tx_chan) |
1927 		SMAC_SEL_V(ep->smac_idx) |
1928 		DSCP_V(ep->tos >> 2) |
1929 		ULP_MODE_V(ULP_MODE_TCPDDP) |
1930 		RCV_BUFSIZ_V(win));
1931 	req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1932 		TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1933 		RX_CHANNEL_V(0) |
1934 		CCTRL_ECN_V(enable_ecn) |
1935 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1936 	if (enable_tcp_timestamps)
1937 		req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1938 	if (enable_tcp_sack)
1939 		req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1940 	if (wscale && enable_tcp_window_scaling)
1941 		req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1942 	req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1943 	req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1944 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1945 	set_bit(ACT_OFLD_CONN, &ep->com.history);
1946 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1947 }
1948 
1949 /*
1950  * Some of the error codes above implicitly indicate that there is no TID
1951  * allocated with the result of an ACT_OPEN.  We use this predicate to make
1952  * that explicit.
1953  */
1954 static inline int act_open_has_tid(int status)
1955 {
1956 	return (status != CPL_ERR_TCAM_PARITY &&
1957 		status != CPL_ERR_TCAM_MISS &&
1958 		status != CPL_ERR_TCAM_FULL &&
1959 		status != CPL_ERR_CONN_EXIST_SYNRECV &&
1960 		status != CPL_ERR_CONN_EXIST);
1961 }
1962 
1963 static char *neg_adv_str(unsigned int status)
1964 {
1965 	switch (status) {
1966 	case CPL_ERR_RTX_NEG_ADVICE:
1967 		return "Retransmit timeout";
1968 	case CPL_ERR_PERSIST_NEG_ADVICE:
1969 		return "Persist timeout";
1970 	case CPL_ERR_KEEPALV_NEG_ADVICE:
1971 		return "Keepalive timeout";
1972 	default:
1973 		return "Unknown";
1974 	}
1975 }
1976 
1977 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1978 {
1979 	ep->snd_win = snd_win;
1980 	ep->rcv_win = rcv_win;
1981 	PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1982 }
1983 
1984 #define ACT_OPEN_RETRY_COUNT 2
1985 
1986 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1987 		     struct dst_entry *dst, struct c4iw_dev *cdev,
1988 		     bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
1989 {
1990 	struct neighbour *n;
1991 	int err, step;
1992 	struct net_device *pdev;
1993 
1994 	n = dst_neigh_lookup(dst, peer_ip);
1995 	if (!n)
1996 		return -ENODEV;
1997 
1998 	rcu_read_lock();
1999 	err = -ENOMEM;
2000 	if (n->dev->flags & IFF_LOOPBACK) {
2001 		if (iptype == 4)
2002 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2003 		else if (IS_ENABLED(CONFIG_IPV6))
2004 			for_each_netdev(&init_net, pdev) {
2005 				if (ipv6_chk_addr(&init_net,
2006 						  (struct in6_addr *)peer_ip,
2007 						  pdev, 1))
2008 					break;
2009 			}
2010 		else
2011 			pdev = NULL;
2012 
2013 		if (!pdev) {
2014 			err = -ENODEV;
2015 			goto out;
2016 		}
2017 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2018 					n, pdev, rt_tos2priority(tos));
2019 		if (!ep->l2t) {
2020 			dev_put(pdev);
2021 			goto out;
2022 		}
2023 		ep->mtu = pdev->mtu;
2024 		ep->tx_chan = cxgb4_port_chan(pdev);
2025 		ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2026 						cxgb4_port_viid(pdev));
2027 		step = cdev->rdev.lldi.ntxq /
2028 			cdev->rdev.lldi.nchan;
2029 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2030 		step = cdev->rdev.lldi.nrxq /
2031 			cdev->rdev.lldi.nchan;
2032 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2033 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2034 			cxgb4_port_idx(pdev) * step];
2035 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2036 		dev_put(pdev);
2037 	} else {
2038 		pdev = get_real_dev(n->dev);
2039 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2040 					n, pdev, 0);
2041 		if (!ep->l2t)
2042 			goto out;
2043 		ep->mtu = dst_mtu(dst);
2044 		ep->tx_chan = cxgb4_port_chan(pdev);
2045 		ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2046 						cxgb4_port_viid(pdev));
2047 		step = cdev->rdev.lldi.ntxq /
2048 			cdev->rdev.lldi.nchan;
2049 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2050 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2051 		step = cdev->rdev.lldi.nrxq /
2052 			cdev->rdev.lldi.nchan;
2053 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2054 			cxgb4_port_idx(pdev) * step];
2055 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2056 
2057 		if (clear_mpa_v1) {
2058 			ep->retry_with_mpa_v1 = 0;
2059 			ep->tried_with_mpa_v1 = 0;
2060 		}
2061 	}
2062 	err = 0;
2063 out:
2064 	rcu_read_unlock();
2065 
2066 	neigh_release(n);
2067 
2068 	return err;
2069 }
2070 
2071 static int c4iw_reconnect(struct c4iw_ep *ep)
2072 {
2073 	int err = 0;
2074 	int size = 0;
2075 	struct sockaddr_in *laddr = (struct sockaddr_in *)
2076 				    &ep->com.cm_id->m_local_addr;
2077 	struct sockaddr_in *raddr = (struct sockaddr_in *)
2078 				    &ep->com.cm_id->m_remote_addr;
2079 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2080 				      &ep->com.cm_id->m_local_addr;
2081 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2082 				      &ep->com.cm_id->m_remote_addr;
2083 	int iptype;
2084 	__u8 *ra;
2085 
2086 	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2087 	init_timer(&ep->timer);
2088 	c4iw_init_wr_wait(&ep->com.wr_wait);
2089 
2090 	/* When MPA revision is different on nodes, the node with MPA_rev=2
2091 	 * tries to reconnect with MPA_rev 1 for the same EP through
2092 	 * c4iw_reconnect(), where the same EP is assigned with new tid for
2093 	 * further connection establishment. As we are using the same EP pointer
2094 	 * for reconnect, few skbs are used during the previous c4iw_connect(),
2095 	 * which leaves the EP with inadequate skbs for further
2096 	 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2097 	 * skb_list() during peer_abort(). Allocate skbs which is already used.
2098 	 */
2099 	size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2100 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2101 		err = -ENOMEM;
2102 		goto fail1;
2103 	}
2104 
2105 	/*
2106 	 * Allocate an active TID to initiate a TCP connection.
2107 	 */
2108 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2109 	if (ep->atid == -1) {
2110 		pr_err("%s - cannot alloc atid.\n", __func__);
2111 		err = -ENOMEM;
2112 		goto fail2;
2113 	}
2114 	insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2115 
2116 	/* find a route */
2117 	if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2118 		ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2119 					  laddr->sin_addr.s_addr,
2120 					  raddr->sin_addr.s_addr,
2121 					  laddr->sin_port,
2122 					  raddr->sin_port, ep->com.cm_id->tos);
2123 		iptype = 4;
2124 		ra = (__u8 *)&raddr->sin_addr;
2125 	} else {
2126 		ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2127 					   get_real_dev,
2128 					   laddr6->sin6_addr.s6_addr,
2129 					   raddr6->sin6_addr.s6_addr,
2130 					   laddr6->sin6_port,
2131 					   raddr6->sin6_port, 0,
2132 					   raddr6->sin6_scope_id);
2133 		iptype = 6;
2134 		ra = (__u8 *)&raddr6->sin6_addr;
2135 	}
2136 	if (!ep->dst) {
2137 		pr_err("%s - cannot find route.\n", __func__);
2138 		err = -EHOSTUNREACH;
2139 		goto fail3;
2140 	}
2141 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2142 			ep->com.dev->rdev.lldi.adapter_type,
2143 			ep->com.cm_id->tos);
2144 	if (err) {
2145 		pr_err("%s - cannot alloc l2e.\n", __func__);
2146 		goto fail4;
2147 	}
2148 
2149 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2150 	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2151 	     ep->l2t->idx);
2152 
2153 	state_set(&ep->com, CONNECTING);
2154 	ep->tos = ep->com.cm_id->tos;
2155 
2156 	/* send connect request to rnic */
2157 	err = send_connect(ep);
2158 	if (!err)
2159 		goto out;
2160 
2161 	cxgb4_l2t_release(ep->l2t);
2162 fail4:
2163 	dst_release(ep->dst);
2164 fail3:
2165 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2166 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2167 fail2:
2168 	/*
2169 	 * remember to send notification to upper layer.
2170 	 * We are in here so the upper layer is not aware that this is
2171 	 * re-connect attempt and so, upper layer is still waiting for
2172 	 * response of 1st connect request.
2173 	 */
2174 	connect_reply_upcall(ep, -ECONNRESET);
2175 fail1:
2176 	c4iw_put_ep(&ep->com);
2177 out:
2178 	return err;
2179 }
2180 
2181 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2182 {
2183 	struct c4iw_ep *ep;
2184 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
2185 	unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2186 				      ntohl(rpl->atid_status)));
2187 	struct tid_info *t = dev->rdev.lldi.tids;
2188 	int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2189 	struct sockaddr_in *la;
2190 	struct sockaddr_in *ra;
2191 	struct sockaddr_in6 *la6;
2192 	struct sockaddr_in6 *ra6;
2193 	int ret = 0;
2194 
2195 	ep = lookup_atid(t, atid);
2196 	la = (struct sockaddr_in *)&ep->com.local_addr;
2197 	ra = (struct sockaddr_in *)&ep->com.remote_addr;
2198 	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2199 	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2200 
2201 	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2202 	     status, status2errno(status));
2203 
2204 	if (cxgb_is_neg_adv(status)) {
2205 		PDBG("%s Connection problems for atid %u status %u (%s)\n",
2206 		     __func__, atid, status, neg_adv_str(status));
2207 		ep->stats.connect_neg_adv++;
2208 		mutex_lock(&dev->rdev.stats.lock);
2209 		dev->rdev.stats.neg_adv++;
2210 		mutex_unlock(&dev->rdev.stats.lock);
2211 		return 0;
2212 	}
2213 
2214 	set_bit(ACT_OPEN_RPL, &ep->com.history);
2215 
2216 	/*
2217 	 * Log interesting failures.
2218 	 */
2219 	switch (status) {
2220 	case CPL_ERR_CONN_RESET:
2221 	case CPL_ERR_CONN_TIMEDOUT:
2222 		break;
2223 	case CPL_ERR_TCAM_FULL:
2224 		mutex_lock(&dev->rdev.stats.lock);
2225 		dev->rdev.stats.tcam_full++;
2226 		mutex_unlock(&dev->rdev.stats.lock);
2227 		if (ep->com.local_addr.ss_family == AF_INET &&
2228 		    dev->rdev.lldi.enable_fw_ofld_conn) {
2229 			ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2230 						   ntohl(rpl->atid_status))));
2231 			if (ret)
2232 				goto fail;
2233 			return 0;
2234 		}
2235 		break;
2236 	case CPL_ERR_CONN_EXIST:
2237 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2238 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
2239 			if (ep->com.remote_addr.ss_family == AF_INET6) {
2240 				struct sockaddr_in6 *sin6 =
2241 						(struct sockaddr_in6 *)
2242 						&ep->com.local_addr;
2243 				cxgb4_clip_release(
2244 						ep->com.dev->rdev.lldi.ports[0],
2245 						(const u32 *)
2246 						&sin6->sin6_addr.s6_addr, 1);
2247 			}
2248 			remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2249 					atid);
2250 			cxgb4_free_atid(t, atid);
2251 			dst_release(ep->dst);
2252 			cxgb4_l2t_release(ep->l2t);
2253 			c4iw_reconnect(ep);
2254 			return 0;
2255 		}
2256 		break;
2257 	default:
2258 		if (ep->com.local_addr.ss_family == AF_INET) {
2259 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2260 				atid, status, status2errno(status),
2261 				&la->sin_addr.s_addr, ntohs(la->sin_port),
2262 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
2263 		} else {
2264 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2265 				atid, status, status2errno(status),
2266 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2267 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2268 		}
2269 		break;
2270 	}
2271 
2272 fail:
2273 	connect_reply_upcall(ep, status2errno(status));
2274 	state_set(&ep->com, DEAD);
2275 
2276 	if (ep->com.remote_addr.ss_family == AF_INET6) {
2277 		struct sockaddr_in6 *sin6 =
2278 			(struct sockaddr_in6 *)&ep->com.local_addr;
2279 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2280 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2281 	}
2282 	if (status && act_open_has_tid(status))
2283 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2284 
2285 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2286 	cxgb4_free_atid(t, atid);
2287 	dst_release(ep->dst);
2288 	cxgb4_l2t_release(ep->l2t);
2289 	c4iw_put_ep(&ep->com);
2290 
2291 	return 0;
2292 }
2293 
2294 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2295 {
2296 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2297 	unsigned int stid = GET_TID(rpl);
2298 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2299 
2300 	if (!ep) {
2301 		PDBG("%s stid %d lookup failure!\n", __func__, stid);
2302 		goto out;
2303 	}
2304 	PDBG("%s ep %p status %d error %d\n", __func__, ep,
2305 	     rpl->status, status2errno(rpl->status));
2306 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2307 	c4iw_put_ep(&ep->com);
2308 out:
2309 	return 0;
2310 }
2311 
2312 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2313 {
2314 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2315 	unsigned int stid = GET_TID(rpl);
2316 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2317 
2318 	PDBG("%s ep %p\n", __func__, ep);
2319 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2320 	c4iw_put_ep(&ep->com);
2321 	return 0;
2322 }
2323 
2324 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2325 		     struct cpl_pass_accept_req *req)
2326 {
2327 	struct cpl_pass_accept_rpl *rpl;
2328 	unsigned int mtu_idx;
2329 	u64 opt0;
2330 	u32 opt2;
2331 	u32 wscale;
2332 	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2333 	int win;
2334 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2335 
2336 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2337 	BUG_ON(skb_cloned(skb));
2338 
2339 	skb_get(skb);
2340 	rpl = cplhdr(skb);
2341 	if (!is_t4(adapter_type)) {
2342 		skb_trim(skb, roundup(sizeof(*rpl5), 16));
2343 		rpl5 = (void *)rpl;
2344 		INIT_TP_WR(rpl5, ep->hwtid);
2345 	} else {
2346 		skb_trim(skb, sizeof(*rpl));
2347 		INIT_TP_WR(rpl, ep->hwtid);
2348 	}
2349 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2350 						    ep->hwtid));
2351 
2352 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2353 		      enable_tcp_timestamps && req->tcpopt.tstamp,
2354 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2355 	wscale = cxgb_compute_wscale(rcv_win);
2356 
2357 	/*
2358 	 * Specify the largest window that will fit in opt0. The
2359 	 * remainder will be specified in the rx_data_ack.
2360 	 */
2361 	win = ep->rcv_win >> 10;
2362 	if (win > RCV_BUFSIZ_M)
2363 		win = RCV_BUFSIZ_M;
2364 	opt0 = (nocong ? NO_CONG_F : 0) |
2365 	       KEEP_ALIVE_F |
2366 	       DELACK_F |
2367 	       WND_SCALE_V(wscale) |
2368 	       MSS_IDX_V(mtu_idx) |
2369 	       L2T_IDX_V(ep->l2t->idx) |
2370 	       TX_CHAN_V(ep->tx_chan) |
2371 	       SMAC_SEL_V(ep->smac_idx) |
2372 	       DSCP_V(ep->tos >> 2) |
2373 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
2374 	       RCV_BUFSIZ_V(win);
2375 	opt2 = RX_CHANNEL_V(0) |
2376 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2377 
2378 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2379 		opt2 |= TSTAMPS_EN_F;
2380 	if (enable_tcp_sack && req->tcpopt.sack)
2381 		opt2 |= SACK_EN_F;
2382 	if (wscale && enable_tcp_window_scaling)
2383 		opt2 |= WND_SCALE_EN_F;
2384 	if (enable_ecn) {
2385 		const struct tcphdr *tcph;
2386 		u32 hlen = ntohl(req->hdr_len);
2387 
2388 		if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2389 			tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2390 				IP_HDR_LEN_G(hlen);
2391 		else
2392 			tcph = (const void *)(req + 1) +
2393 				T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2394 		if (tcph->ece && tcph->cwr)
2395 			opt2 |= CCTRL_ECN_V(1);
2396 	}
2397 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2398 		u32 isn = (prandom_u32() & ~7UL) - 1;
2399 		opt2 |= T5_OPT_2_VALID_F;
2400 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2401 		opt2 |= T5_ISS_F;
2402 		rpl5 = (void *)rpl;
2403 		memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2404 		if (peer2peer)
2405 			isn += 4;
2406 		rpl5->iss = cpu_to_be32(isn);
2407 		PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2408 	}
2409 
2410 	rpl->opt0 = cpu_to_be64(opt0);
2411 	rpl->opt2 = cpu_to_be32(opt2);
2412 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2413 	t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2414 
2415 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2416 }
2417 
2418 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2419 {
2420 	PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2421 	BUG_ON(skb_cloned(skb));
2422 	skb_trim(skb, sizeof(struct cpl_tid_release));
2423 	release_tid(&dev->rdev, hwtid, skb);
2424 	return;
2425 }
2426 
2427 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2428 {
2429 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2430 	struct cpl_pass_accept_req *req = cplhdr(skb);
2431 	unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2432 	struct tid_info *t = dev->rdev.lldi.tids;
2433 	unsigned int hwtid = GET_TID(req);
2434 	struct dst_entry *dst;
2435 	__u8 local_ip[16], peer_ip[16];
2436 	__be16 local_port, peer_port;
2437 	struct sockaddr_in6 *sin6;
2438 	int err;
2439 	u16 peer_mss = ntohs(req->tcpopt.mss);
2440 	int iptype;
2441 	unsigned short hdrs;
2442 	u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2443 
2444 	parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2445 	if (!parent_ep) {
2446 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2447 		goto reject;
2448 	}
2449 
2450 	if (state_read(&parent_ep->com) != LISTEN) {
2451 		PDBG("%s - listening ep not in LISTEN\n", __func__);
2452 		goto reject;
2453 	}
2454 
2455 	cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2456 			&iptype, local_ip, peer_ip, &local_port, &peer_port);
2457 
2458 	/* Find output route */
2459 	if (iptype == 4)  {
2460 		PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2461 		     , __func__, parent_ep, hwtid,
2462 		     local_ip, peer_ip, ntohs(local_port),
2463 		     ntohs(peer_port), peer_mss);
2464 		dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2465 				      *(__be32 *)local_ip, *(__be32 *)peer_ip,
2466 				      local_port, peer_port, tos);
2467 	} else {
2468 		PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2469 		     , __func__, parent_ep, hwtid,
2470 		     local_ip, peer_ip, ntohs(local_port),
2471 		     ntohs(peer_port), peer_mss);
2472 		dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2473 				local_ip, peer_ip, local_port, peer_port,
2474 				PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2475 				((struct sockaddr_in6 *)
2476 				 &parent_ep->com.local_addr)->sin6_scope_id);
2477 	}
2478 	if (!dst) {
2479 		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2480 		       __func__);
2481 		goto reject;
2482 	}
2483 
2484 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2485 	if (!child_ep) {
2486 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2487 		       __func__);
2488 		dst_release(dst);
2489 		goto reject;
2490 	}
2491 
2492 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2493 			parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2494 	if (err) {
2495 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2496 		       __func__);
2497 		dst_release(dst);
2498 		kfree(child_ep);
2499 		goto reject;
2500 	}
2501 
2502 	hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2503 	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2504 	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2505 		child_ep->mtu = peer_mss + hdrs;
2506 
2507 	skb_queue_head_init(&child_ep->com.ep_skb_list);
2508 	if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2509 		goto fail;
2510 
2511 	state_set(&child_ep->com, CONNECTING);
2512 	child_ep->com.dev = dev;
2513 	child_ep->com.cm_id = NULL;
2514 
2515 	if (iptype == 4) {
2516 		struct sockaddr_in *sin = (struct sockaddr_in *)
2517 			&child_ep->com.local_addr;
2518 
2519 		sin->sin_family = PF_INET;
2520 		sin->sin_port = local_port;
2521 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2522 
2523 		sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2524 		sin->sin_family = PF_INET;
2525 		sin->sin_port = ((struct sockaddr_in *)
2526 				 &parent_ep->com.local_addr)->sin_port;
2527 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2528 
2529 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2530 		sin->sin_family = PF_INET;
2531 		sin->sin_port = peer_port;
2532 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2533 	} else {
2534 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2535 		sin6->sin6_family = PF_INET6;
2536 		sin6->sin6_port = local_port;
2537 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2538 
2539 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2540 		sin6->sin6_family = PF_INET6;
2541 		sin6->sin6_port = ((struct sockaddr_in6 *)
2542 				   &parent_ep->com.local_addr)->sin6_port;
2543 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2544 
2545 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2546 		sin6->sin6_family = PF_INET6;
2547 		sin6->sin6_port = peer_port;
2548 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2549 	}
2550 
2551 	c4iw_get_ep(&parent_ep->com);
2552 	child_ep->parent_ep = parent_ep;
2553 	child_ep->tos = tos;
2554 	child_ep->dst = dst;
2555 	child_ep->hwtid = hwtid;
2556 
2557 	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2558 	     child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2559 
2560 	init_timer(&child_ep->timer);
2561 	cxgb4_insert_tid(t, child_ep, hwtid);
2562 	insert_ep_tid(child_ep);
2563 	if (accept_cr(child_ep, skb, req)) {
2564 		c4iw_put_ep(&parent_ep->com);
2565 		release_ep_resources(child_ep);
2566 	} else {
2567 		set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2568 	}
2569 	if (iptype == 6) {
2570 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2571 		cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2572 			       (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2573 	}
2574 	goto out;
2575 fail:
2576 	c4iw_put_ep(&child_ep->com);
2577 reject:
2578 	reject_cr(dev, hwtid, skb);
2579 	if (parent_ep)
2580 		c4iw_put_ep(&parent_ep->com);
2581 out:
2582 	return 0;
2583 }
2584 
2585 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2586 {
2587 	struct c4iw_ep *ep;
2588 	struct cpl_pass_establish *req = cplhdr(skb);
2589 	unsigned int tid = GET_TID(req);
2590 	int ret;
2591 
2592 	ep = get_ep_from_tid(dev, tid);
2593 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2594 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2595 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2596 
2597 	PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2598 	     ntohs(req->tcp_opt));
2599 
2600 	set_emss(ep, ntohs(req->tcp_opt));
2601 
2602 	dst_confirm(ep->dst);
2603 	mutex_lock(&ep->com.mutex);
2604 	ep->com.state = MPA_REQ_WAIT;
2605 	start_ep_timer(ep);
2606 	set_bit(PASS_ESTAB, &ep->com.history);
2607 	ret = send_flowc(ep);
2608 	mutex_unlock(&ep->com.mutex);
2609 	if (ret)
2610 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2611 	c4iw_put_ep(&ep->com);
2612 
2613 	return 0;
2614 }
2615 
2616 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2617 {
2618 	struct cpl_peer_close *hdr = cplhdr(skb);
2619 	struct c4iw_ep *ep;
2620 	struct c4iw_qp_attributes attrs;
2621 	int disconnect = 1;
2622 	int release = 0;
2623 	unsigned int tid = GET_TID(hdr);
2624 	int ret;
2625 
2626 	ep = get_ep_from_tid(dev, tid);
2627 	if (!ep)
2628 		return 0;
2629 
2630 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2631 	dst_confirm(ep->dst);
2632 
2633 	set_bit(PEER_CLOSE, &ep->com.history);
2634 	mutex_lock(&ep->com.mutex);
2635 	switch (ep->com.state) {
2636 	case MPA_REQ_WAIT:
2637 		__state_set(&ep->com, CLOSING);
2638 		break;
2639 	case MPA_REQ_SENT:
2640 		__state_set(&ep->com, CLOSING);
2641 		connect_reply_upcall(ep, -ECONNRESET);
2642 		break;
2643 	case MPA_REQ_RCVD:
2644 
2645 		/*
2646 		 * We're gonna mark this puppy DEAD, but keep
2647 		 * the reference on it until the ULP accepts or
2648 		 * rejects the CR. Also wake up anyone waiting
2649 		 * in rdma connection migration (see c4iw_accept_cr()).
2650 		 */
2651 		__state_set(&ep->com, CLOSING);
2652 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2653 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2654 		break;
2655 	case MPA_REP_SENT:
2656 		__state_set(&ep->com, CLOSING);
2657 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2658 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2659 		break;
2660 	case FPDU_MODE:
2661 		start_ep_timer(ep);
2662 		__state_set(&ep->com, CLOSING);
2663 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2664 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2665 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2666 		if (ret != -ECONNRESET) {
2667 			peer_close_upcall(ep);
2668 			disconnect = 1;
2669 		}
2670 		break;
2671 	case ABORTING:
2672 		disconnect = 0;
2673 		break;
2674 	case CLOSING:
2675 		__state_set(&ep->com, MORIBUND);
2676 		disconnect = 0;
2677 		break;
2678 	case MORIBUND:
2679 		(void)stop_ep_timer(ep);
2680 		if (ep->com.cm_id && ep->com.qp) {
2681 			attrs.next_state = C4IW_QP_STATE_IDLE;
2682 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2683 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2684 		}
2685 		close_complete_upcall(ep, 0);
2686 		__state_set(&ep->com, DEAD);
2687 		release = 1;
2688 		disconnect = 0;
2689 		break;
2690 	case DEAD:
2691 		disconnect = 0;
2692 		break;
2693 	default:
2694 		BUG_ON(1);
2695 	}
2696 	mutex_unlock(&ep->com.mutex);
2697 	if (disconnect)
2698 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2699 	if (release)
2700 		release_ep_resources(ep);
2701 	c4iw_put_ep(&ep->com);
2702 	return 0;
2703 }
2704 
2705 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2706 {
2707 	struct cpl_abort_req_rss *req = cplhdr(skb);
2708 	struct c4iw_ep *ep;
2709 	struct sk_buff *rpl_skb;
2710 	struct c4iw_qp_attributes attrs;
2711 	int ret;
2712 	int release = 0;
2713 	unsigned int tid = GET_TID(req);
2714 	u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2715 
2716 	ep = get_ep_from_tid(dev, tid);
2717 	if (!ep)
2718 		return 0;
2719 
2720 	if (cxgb_is_neg_adv(req->status)) {
2721 		PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2722 		     __func__, ep->hwtid, req->status,
2723 		     neg_adv_str(req->status));
2724 		ep->stats.abort_neg_adv++;
2725 		mutex_lock(&dev->rdev.stats.lock);
2726 		dev->rdev.stats.neg_adv++;
2727 		mutex_unlock(&dev->rdev.stats.lock);
2728 		goto deref_ep;
2729 	}
2730 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2731 	     ep->com.state);
2732 	set_bit(PEER_ABORT, &ep->com.history);
2733 
2734 	/*
2735 	 * Wake up any threads in rdma_init() or rdma_fini().
2736 	 * However, this is not needed if com state is just
2737 	 * MPA_REQ_SENT
2738 	 */
2739 	if (ep->com.state != MPA_REQ_SENT)
2740 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2741 
2742 	mutex_lock(&ep->com.mutex);
2743 	switch (ep->com.state) {
2744 	case CONNECTING:
2745 		c4iw_put_ep(&ep->parent_ep->com);
2746 		break;
2747 	case MPA_REQ_WAIT:
2748 		(void)stop_ep_timer(ep);
2749 		break;
2750 	case MPA_REQ_SENT:
2751 		(void)stop_ep_timer(ep);
2752 		if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2753 			connect_reply_upcall(ep, -ECONNRESET);
2754 		else {
2755 			/*
2756 			 * we just don't send notification upwards because we
2757 			 * want to retry with mpa_v1 without upper layers even
2758 			 * knowing it.
2759 			 *
2760 			 * do some housekeeping so as to re-initiate the
2761 			 * connection
2762 			 */
2763 			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2764 			     mpa_rev);
2765 			ep->retry_with_mpa_v1 = 1;
2766 		}
2767 		break;
2768 	case MPA_REP_SENT:
2769 		break;
2770 	case MPA_REQ_RCVD:
2771 		break;
2772 	case MORIBUND:
2773 	case CLOSING:
2774 		stop_ep_timer(ep);
2775 		/*FALLTHROUGH*/
2776 	case FPDU_MODE:
2777 		if (ep->com.cm_id && ep->com.qp) {
2778 			attrs.next_state = C4IW_QP_STATE_ERROR;
2779 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2780 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2781 				     &attrs, 1);
2782 			if (ret)
2783 				printk(KERN_ERR MOD
2784 				       "%s - qp <- error failed!\n",
2785 				       __func__);
2786 		}
2787 		peer_abort_upcall(ep);
2788 		break;
2789 	case ABORTING:
2790 		break;
2791 	case DEAD:
2792 		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2793 		mutex_unlock(&ep->com.mutex);
2794 		goto deref_ep;
2795 	default:
2796 		BUG_ON(1);
2797 		break;
2798 	}
2799 	dst_confirm(ep->dst);
2800 	if (ep->com.state != ABORTING) {
2801 		__state_set(&ep->com, DEAD);
2802 		/* we don't release if we want to retry with mpa_v1 */
2803 		if (!ep->retry_with_mpa_v1)
2804 			release = 1;
2805 	}
2806 	mutex_unlock(&ep->com.mutex);
2807 
2808 	rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2809 	if (WARN_ON(!rpl_skb)) {
2810 		release = 1;
2811 		goto out;
2812 	}
2813 
2814 	cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2815 
2816 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2817 out:
2818 	if (release)
2819 		release_ep_resources(ep);
2820 	else if (ep->retry_with_mpa_v1) {
2821 		if (ep->com.remote_addr.ss_family == AF_INET6) {
2822 			struct sockaddr_in6 *sin6 =
2823 					(struct sockaddr_in6 *)
2824 					&ep->com.local_addr;
2825 			cxgb4_clip_release(
2826 					ep->com.dev->rdev.lldi.ports[0],
2827 					(const u32 *)&sin6->sin6_addr.s6_addr,
2828 					1);
2829 		}
2830 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2831 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2832 		dst_release(ep->dst);
2833 		cxgb4_l2t_release(ep->l2t);
2834 		c4iw_reconnect(ep);
2835 	}
2836 
2837 deref_ep:
2838 	c4iw_put_ep(&ep->com);
2839 	/* Dereferencing ep, referenced in peer_abort_intr() */
2840 	c4iw_put_ep(&ep->com);
2841 	return 0;
2842 }
2843 
2844 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2845 {
2846 	struct c4iw_ep *ep;
2847 	struct c4iw_qp_attributes attrs;
2848 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2849 	int release = 0;
2850 	unsigned int tid = GET_TID(rpl);
2851 
2852 	ep = get_ep_from_tid(dev, tid);
2853 	if (!ep)
2854 		return 0;
2855 
2856 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2857 	BUG_ON(!ep);
2858 
2859 	/* The cm_id may be null if we failed to connect */
2860 	mutex_lock(&ep->com.mutex);
2861 	set_bit(CLOSE_CON_RPL, &ep->com.history);
2862 	switch (ep->com.state) {
2863 	case CLOSING:
2864 		__state_set(&ep->com, MORIBUND);
2865 		break;
2866 	case MORIBUND:
2867 		(void)stop_ep_timer(ep);
2868 		if ((ep->com.cm_id) && (ep->com.qp)) {
2869 			attrs.next_state = C4IW_QP_STATE_IDLE;
2870 			c4iw_modify_qp(ep->com.qp->rhp,
2871 					     ep->com.qp,
2872 					     C4IW_QP_ATTR_NEXT_STATE,
2873 					     &attrs, 1);
2874 		}
2875 		close_complete_upcall(ep, 0);
2876 		__state_set(&ep->com, DEAD);
2877 		release = 1;
2878 		break;
2879 	case ABORTING:
2880 	case DEAD:
2881 		break;
2882 	default:
2883 		BUG_ON(1);
2884 		break;
2885 	}
2886 	mutex_unlock(&ep->com.mutex);
2887 	if (release)
2888 		release_ep_resources(ep);
2889 	c4iw_put_ep(&ep->com);
2890 	return 0;
2891 }
2892 
2893 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2894 {
2895 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
2896 	unsigned int tid = GET_TID(rpl);
2897 	struct c4iw_ep *ep;
2898 	struct c4iw_qp_attributes attrs;
2899 
2900 	ep = get_ep_from_tid(dev, tid);
2901 	BUG_ON(!ep);
2902 
2903 	if (ep && ep->com.qp) {
2904 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2905 		       ep->com.qp->wq.sq.qid);
2906 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2907 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2908 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2909 	} else
2910 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2911 	c4iw_put_ep(&ep->com);
2912 
2913 	return 0;
2914 }
2915 
2916 /*
2917  * Upcall from the adapter indicating data has been transmitted.
2918  * For us its just the single MPA request or reply.  We can now free
2919  * the skb holding the mpa message.
2920  */
2921 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2922 {
2923 	struct c4iw_ep *ep;
2924 	struct cpl_fw4_ack *hdr = cplhdr(skb);
2925 	u8 credits = hdr->credits;
2926 	unsigned int tid = GET_TID(hdr);
2927 
2928 
2929 	ep = get_ep_from_tid(dev, tid);
2930 	if (!ep)
2931 		return 0;
2932 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2933 	if (credits == 0) {
2934 		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2935 		     __func__, ep, ep->hwtid, state_read(&ep->com));
2936 		goto out;
2937 	}
2938 
2939 	dst_confirm(ep->dst);
2940 	if (ep->mpa_skb) {
2941 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
2942 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2943 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2944 		mutex_lock(&ep->com.mutex);
2945 		kfree_skb(ep->mpa_skb);
2946 		ep->mpa_skb = NULL;
2947 		if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2948 			stop_ep_timer(ep);
2949 		mutex_unlock(&ep->com.mutex);
2950 	}
2951 out:
2952 	c4iw_put_ep(&ep->com);
2953 	return 0;
2954 }
2955 
2956 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2957 {
2958 	int abort;
2959 	struct c4iw_ep *ep = to_ep(cm_id);
2960 
2961 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2962 
2963 	mutex_lock(&ep->com.mutex);
2964 	if (ep->com.state != MPA_REQ_RCVD) {
2965 		mutex_unlock(&ep->com.mutex);
2966 		c4iw_put_ep(&ep->com);
2967 		return -ECONNRESET;
2968 	}
2969 	set_bit(ULP_REJECT, &ep->com.history);
2970 	if (mpa_rev == 0)
2971 		abort = 1;
2972 	else
2973 		abort = send_mpa_reject(ep, pdata, pdata_len);
2974 	mutex_unlock(&ep->com.mutex);
2975 
2976 	stop_ep_timer(ep);
2977 	c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2978 	c4iw_put_ep(&ep->com);
2979 	return 0;
2980 }
2981 
2982 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2983 {
2984 	int err;
2985 	struct c4iw_qp_attributes attrs;
2986 	enum c4iw_qp_attr_mask mask;
2987 	struct c4iw_ep *ep = to_ep(cm_id);
2988 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2989 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2990 	int abort = 0;
2991 
2992 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2993 
2994 	mutex_lock(&ep->com.mutex);
2995 	if (ep->com.state != MPA_REQ_RCVD) {
2996 		err = -ECONNRESET;
2997 		goto err_out;
2998 	}
2999 
3000 	BUG_ON(!qp);
3001 
3002 	set_bit(ULP_ACCEPT, &ep->com.history);
3003 	if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3004 	    (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3005 		err = -EINVAL;
3006 		goto err_abort;
3007 	}
3008 
3009 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3010 		if (conn_param->ord > ep->ird) {
3011 			if (RELAXED_IRD_NEGOTIATION) {
3012 				conn_param->ord = ep->ird;
3013 			} else {
3014 				ep->ird = conn_param->ird;
3015 				ep->ord = conn_param->ord;
3016 				send_mpa_reject(ep, conn_param->private_data,
3017 						conn_param->private_data_len);
3018 				err = -ENOMEM;
3019 				goto err_abort;
3020 			}
3021 		}
3022 		if (conn_param->ird < ep->ord) {
3023 			if (RELAXED_IRD_NEGOTIATION &&
3024 			    ep->ord <= h->rdev.lldi.max_ordird_qp) {
3025 				conn_param->ird = ep->ord;
3026 			} else {
3027 				err = -ENOMEM;
3028 				goto err_abort;
3029 			}
3030 		}
3031 	}
3032 	ep->ird = conn_param->ird;
3033 	ep->ord = conn_param->ord;
3034 
3035 	if (ep->mpa_attr.version == 1) {
3036 		if (peer2peer && ep->ird == 0)
3037 			ep->ird = 1;
3038 	} else {
3039 		if (peer2peer &&
3040 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3041 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3042 			ep->ird = 1;
3043 	}
3044 
3045 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3046 
3047 	ep->com.cm_id = cm_id;
3048 	ref_cm_id(&ep->com);
3049 	ep->com.qp = qp;
3050 	ref_qp(ep);
3051 
3052 	/* bind QP to EP and move to RTS */
3053 	attrs.mpa_attr = ep->mpa_attr;
3054 	attrs.max_ird = ep->ird;
3055 	attrs.max_ord = ep->ord;
3056 	attrs.llp_stream_handle = ep;
3057 	attrs.next_state = C4IW_QP_STATE_RTS;
3058 
3059 	/* bind QP and TID with INIT_WR */
3060 	mask = C4IW_QP_ATTR_NEXT_STATE |
3061 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3062 			     C4IW_QP_ATTR_MPA_ATTR |
3063 			     C4IW_QP_ATTR_MAX_IRD |
3064 			     C4IW_QP_ATTR_MAX_ORD;
3065 
3066 	err = c4iw_modify_qp(ep->com.qp->rhp,
3067 			     ep->com.qp, mask, &attrs, 1);
3068 	if (err)
3069 		goto err_deref_cm_id;
3070 
3071 	set_bit(STOP_MPA_TIMER, &ep->com.flags);
3072 	err = send_mpa_reply(ep, conn_param->private_data,
3073 			     conn_param->private_data_len);
3074 	if (err)
3075 		goto err_deref_cm_id;
3076 
3077 	__state_set(&ep->com, FPDU_MODE);
3078 	established_upcall(ep);
3079 	mutex_unlock(&ep->com.mutex);
3080 	c4iw_put_ep(&ep->com);
3081 	return 0;
3082 err_deref_cm_id:
3083 	deref_cm_id(&ep->com);
3084 err_abort:
3085 	abort = 1;
3086 err_out:
3087 	mutex_unlock(&ep->com.mutex);
3088 	if (abort)
3089 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3090 	c4iw_put_ep(&ep->com);
3091 	return err;
3092 }
3093 
3094 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3095 {
3096 	struct in_device *ind;
3097 	int found = 0;
3098 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3099 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3100 
3101 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
3102 	if (!ind)
3103 		return -EADDRNOTAVAIL;
3104 	for_primary_ifa(ind) {
3105 		laddr->sin_addr.s_addr = ifa->ifa_address;
3106 		raddr->sin_addr.s_addr = ifa->ifa_address;
3107 		found = 1;
3108 		break;
3109 	}
3110 	endfor_ifa(ind);
3111 	in_dev_put(ind);
3112 	return found ? 0 : -EADDRNOTAVAIL;
3113 }
3114 
3115 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3116 		      unsigned char banned_flags)
3117 {
3118 	struct inet6_dev *idev;
3119 	int err = -EADDRNOTAVAIL;
3120 
3121 	rcu_read_lock();
3122 	idev = __in6_dev_get(dev);
3123 	if (idev != NULL) {
3124 		struct inet6_ifaddr *ifp;
3125 
3126 		read_lock_bh(&idev->lock);
3127 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
3128 			if (ifp->scope == IFA_LINK &&
3129 			    !(ifp->flags & banned_flags)) {
3130 				memcpy(addr, &ifp->addr, 16);
3131 				err = 0;
3132 				break;
3133 			}
3134 		}
3135 		read_unlock_bh(&idev->lock);
3136 	}
3137 	rcu_read_unlock();
3138 	return err;
3139 }
3140 
3141 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3142 {
3143 	struct in6_addr uninitialized_var(addr);
3144 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3145 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3146 
3147 	if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3148 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3149 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3150 		return 0;
3151 	}
3152 	return -EADDRNOTAVAIL;
3153 }
3154 
3155 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3156 {
3157 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3158 	struct c4iw_ep *ep;
3159 	int err = 0;
3160 	struct sockaddr_in *laddr;
3161 	struct sockaddr_in *raddr;
3162 	struct sockaddr_in6 *laddr6;
3163 	struct sockaddr_in6 *raddr6;
3164 	__u8 *ra;
3165 	int iptype;
3166 
3167 	if ((conn_param->ord > cur_max_read_depth(dev)) ||
3168 	    (conn_param->ird > cur_max_read_depth(dev))) {
3169 		err = -EINVAL;
3170 		goto out;
3171 	}
3172 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3173 	if (!ep) {
3174 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3175 		err = -ENOMEM;
3176 		goto out;
3177 	}
3178 
3179 	skb_queue_head_init(&ep->com.ep_skb_list);
3180 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3181 		err = -ENOMEM;
3182 		goto fail1;
3183 	}
3184 
3185 	init_timer(&ep->timer);
3186 	ep->plen = conn_param->private_data_len;
3187 	if (ep->plen)
3188 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3189 		       conn_param->private_data, ep->plen);
3190 	ep->ird = conn_param->ird;
3191 	ep->ord = conn_param->ord;
3192 
3193 	if (peer2peer && ep->ord == 0)
3194 		ep->ord = 1;
3195 
3196 	ep->com.cm_id = cm_id;
3197 	ref_cm_id(&ep->com);
3198 	ep->com.dev = dev;
3199 	ep->com.qp = get_qhp(dev, conn_param->qpn);
3200 	if (!ep->com.qp) {
3201 		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3202 		err = -EINVAL;
3203 		goto fail2;
3204 	}
3205 	ref_qp(ep);
3206 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3207 	     ep->com.qp, cm_id);
3208 
3209 	/*
3210 	 * Allocate an active TID to initiate a TCP connection.
3211 	 */
3212 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3213 	if (ep->atid == -1) {
3214 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3215 		err = -ENOMEM;
3216 		goto fail2;
3217 	}
3218 	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3219 
3220 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3221 	       sizeof(ep->com.local_addr));
3222 	memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3223 	       sizeof(ep->com.remote_addr));
3224 
3225 	laddr = (struct sockaddr_in *)&ep->com.local_addr;
3226 	raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3227 	laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3228 	raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3229 
3230 	if (cm_id->m_remote_addr.ss_family == AF_INET) {
3231 		iptype = 4;
3232 		ra = (__u8 *)&raddr->sin_addr;
3233 
3234 		/*
3235 		 * Handle loopback requests to INADDR_ANY.
3236 		 */
3237 		if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3238 			err = pick_local_ipaddrs(dev, cm_id);
3239 			if (err)
3240 				goto fail2;
3241 		}
3242 
3243 		/* find a route */
3244 		PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3245 		     __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3246 		     ra, ntohs(raddr->sin_port));
3247 		ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3248 					  laddr->sin_addr.s_addr,
3249 					  raddr->sin_addr.s_addr,
3250 					  laddr->sin_port,
3251 					  raddr->sin_port, cm_id->tos);
3252 	} else {
3253 		iptype = 6;
3254 		ra = (__u8 *)&raddr6->sin6_addr;
3255 
3256 		/*
3257 		 * Handle loopback requests to INADDR_ANY.
3258 		 */
3259 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3260 			err = pick_local_ip6addrs(dev, cm_id);
3261 			if (err)
3262 				goto fail2;
3263 		}
3264 
3265 		/* find a route */
3266 		PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3267 		     __func__, laddr6->sin6_addr.s6_addr,
3268 		     ntohs(laddr6->sin6_port),
3269 		     raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3270 		ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3271 					   laddr6->sin6_addr.s6_addr,
3272 					   raddr6->sin6_addr.s6_addr,
3273 					   laddr6->sin6_port,
3274 					   raddr6->sin6_port, 0,
3275 					   raddr6->sin6_scope_id);
3276 	}
3277 	if (!ep->dst) {
3278 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3279 		err = -EHOSTUNREACH;
3280 		goto fail3;
3281 	}
3282 
3283 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3284 			ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3285 	if (err) {
3286 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3287 		goto fail4;
3288 	}
3289 
3290 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3291 		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3292 		ep->l2t->idx);
3293 
3294 	state_set(&ep->com, CONNECTING);
3295 	ep->tos = cm_id->tos;
3296 
3297 	/* send connect request to rnic */
3298 	err = send_connect(ep);
3299 	if (!err)
3300 		goto out;
3301 
3302 	cxgb4_l2t_release(ep->l2t);
3303 fail4:
3304 	dst_release(ep->dst);
3305 fail3:
3306 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3307 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3308 fail2:
3309 	skb_queue_purge(&ep->com.ep_skb_list);
3310 	deref_cm_id(&ep->com);
3311 fail1:
3312 	c4iw_put_ep(&ep->com);
3313 out:
3314 	return err;
3315 }
3316 
3317 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3318 {
3319 	int err;
3320 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3321 				    &ep->com.local_addr;
3322 
3323 	if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3324 		err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3325 				     (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3326 		if (err)
3327 			return err;
3328 	}
3329 	c4iw_init_wr_wait(&ep->com.wr_wait);
3330 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3331 				   ep->stid, &sin6->sin6_addr,
3332 				   sin6->sin6_port,
3333 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
3334 	if (!err)
3335 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3336 					  &ep->com.wr_wait,
3337 					  0, 0, __func__);
3338 	else if (err > 0)
3339 		err = net_xmit_errno(err);
3340 	if (err) {
3341 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3342 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3343 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3344 		       err, ep->stid,
3345 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3346 	}
3347 	return err;
3348 }
3349 
3350 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3351 {
3352 	int err;
3353 	struct sockaddr_in *sin = (struct sockaddr_in *)
3354 				  &ep->com.local_addr;
3355 
3356 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
3357 		do {
3358 			err = cxgb4_create_server_filter(
3359 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3360 				sin->sin_addr.s_addr, sin->sin_port, 0,
3361 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3362 			if (err == -EBUSY) {
3363 				if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3364 					err = -EIO;
3365 					break;
3366 				}
3367 				set_current_state(TASK_UNINTERRUPTIBLE);
3368 				schedule_timeout(usecs_to_jiffies(100));
3369 			}
3370 		} while (err == -EBUSY);
3371 	} else {
3372 		c4iw_init_wr_wait(&ep->com.wr_wait);
3373 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3374 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3375 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3376 		if (!err)
3377 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3378 						  &ep->com.wr_wait,
3379 						  0, 0, __func__);
3380 		else if (err > 0)
3381 			err = net_xmit_errno(err);
3382 	}
3383 	if (err)
3384 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3385 		       , err, ep->stid,
3386 		       &sin->sin_addr, ntohs(sin->sin_port));
3387 	return err;
3388 }
3389 
3390 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3391 {
3392 	int err = 0;
3393 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3394 	struct c4iw_listen_ep *ep;
3395 
3396 	might_sleep();
3397 
3398 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3399 	if (!ep) {
3400 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3401 		err = -ENOMEM;
3402 		goto fail1;
3403 	}
3404 	skb_queue_head_init(&ep->com.ep_skb_list);
3405 	PDBG("%s ep %p\n", __func__, ep);
3406 	ep->com.cm_id = cm_id;
3407 	ref_cm_id(&ep->com);
3408 	ep->com.dev = dev;
3409 	ep->backlog = backlog;
3410 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3411 	       sizeof(ep->com.local_addr));
3412 
3413 	/*
3414 	 * Allocate a server TID.
3415 	 */
3416 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
3417 	    ep->com.local_addr.ss_family == AF_INET)
3418 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3419 					     cm_id->m_local_addr.ss_family, ep);
3420 	else
3421 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3422 					    cm_id->m_local_addr.ss_family, ep);
3423 
3424 	if (ep->stid == -1) {
3425 		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3426 		err = -ENOMEM;
3427 		goto fail2;
3428 	}
3429 	insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3430 
3431 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3432 	       sizeof(ep->com.local_addr));
3433 
3434 	state_set(&ep->com, LISTEN);
3435 	if (ep->com.local_addr.ss_family == AF_INET)
3436 		err = create_server4(dev, ep);
3437 	else
3438 		err = create_server6(dev, ep);
3439 	if (!err) {
3440 		cm_id->provider_data = ep;
3441 		goto out;
3442 	}
3443 
3444 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3445 			ep->com.local_addr.ss_family);
3446 fail2:
3447 	deref_cm_id(&ep->com);
3448 	c4iw_put_ep(&ep->com);
3449 fail1:
3450 out:
3451 	return err;
3452 }
3453 
3454 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3455 {
3456 	int err;
3457 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3458 
3459 	PDBG("%s ep %p\n", __func__, ep);
3460 
3461 	might_sleep();
3462 	state_set(&ep->com, DEAD);
3463 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3464 	    ep->com.local_addr.ss_family == AF_INET) {
3465 		err = cxgb4_remove_server_filter(
3466 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
3467 			ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3468 	} else {
3469 		struct sockaddr_in6 *sin6;
3470 		c4iw_init_wr_wait(&ep->com.wr_wait);
3471 		err = cxgb4_remove_server(
3472 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3473 				ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3474 		if (err)
3475 			goto done;
3476 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3477 					  0, 0, __func__);
3478 		sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3479 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3480 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3481 	}
3482 	remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3483 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3484 			ep->com.local_addr.ss_family);
3485 done:
3486 	deref_cm_id(&ep->com);
3487 	c4iw_put_ep(&ep->com);
3488 	return err;
3489 }
3490 
3491 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3492 {
3493 	int ret = 0;
3494 	int close = 0;
3495 	int fatal = 0;
3496 	struct c4iw_rdev *rdev;
3497 
3498 	mutex_lock(&ep->com.mutex);
3499 
3500 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3501 	     states[ep->com.state], abrupt);
3502 
3503 	/*
3504 	 * Ref the ep here in case we have fatal errors causing the
3505 	 * ep to be released and freed.
3506 	 */
3507 	c4iw_get_ep(&ep->com);
3508 
3509 	rdev = &ep->com.dev->rdev;
3510 	if (c4iw_fatal_error(rdev)) {
3511 		fatal = 1;
3512 		close_complete_upcall(ep, -EIO);
3513 		ep->com.state = DEAD;
3514 	}
3515 	switch (ep->com.state) {
3516 	case MPA_REQ_WAIT:
3517 	case MPA_REQ_SENT:
3518 	case MPA_REQ_RCVD:
3519 	case MPA_REP_SENT:
3520 	case FPDU_MODE:
3521 	case CONNECTING:
3522 		close = 1;
3523 		if (abrupt)
3524 			ep->com.state = ABORTING;
3525 		else {
3526 			ep->com.state = CLOSING;
3527 
3528 			/*
3529 			 * if we close before we see the fw4_ack() then we fix
3530 			 * up the timer state since we're reusing it.
3531 			 */
3532 			if (ep->mpa_skb &&
3533 			    test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3534 				clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3535 				stop_ep_timer(ep);
3536 			}
3537 			start_ep_timer(ep);
3538 		}
3539 		set_bit(CLOSE_SENT, &ep->com.flags);
3540 		break;
3541 	case CLOSING:
3542 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3543 			close = 1;
3544 			if (abrupt) {
3545 				(void)stop_ep_timer(ep);
3546 				ep->com.state = ABORTING;
3547 			} else
3548 				ep->com.state = MORIBUND;
3549 		}
3550 		break;
3551 	case MORIBUND:
3552 	case ABORTING:
3553 	case DEAD:
3554 		PDBG("%s ignoring disconnect ep %p state %u\n",
3555 		     __func__, ep, ep->com.state);
3556 		break;
3557 	default:
3558 		BUG();
3559 		break;
3560 	}
3561 
3562 	if (close) {
3563 		if (abrupt) {
3564 			set_bit(EP_DISC_ABORT, &ep->com.history);
3565 			close_complete_upcall(ep, -ECONNRESET);
3566 			ret = send_abort(ep);
3567 		} else {
3568 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3569 			ret = send_halfclose(ep);
3570 		}
3571 		if (ret) {
3572 			set_bit(EP_DISC_FAIL, &ep->com.history);
3573 			if (!abrupt) {
3574 				stop_ep_timer(ep);
3575 				close_complete_upcall(ep, -EIO);
3576 			}
3577 			if (ep->com.qp) {
3578 				struct c4iw_qp_attributes attrs;
3579 
3580 				attrs.next_state = C4IW_QP_STATE_ERROR;
3581 				ret = c4iw_modify_qp(ep->com.qp->rhp,
3582 						     ep->com.qp,
3583 						     C4IW_QP_ATTR_NEXT_STATE,
3584 						     &attrs, 1);
3585 				if (ret)
3586 					pr_err(MOD
3587 					       "%s - qp <- error failed!\n",
3588 					       __func__);
3589 			}
3590 			fatal = 1;
3591 		}
3592 	}
3593 	mutex_unlock(&ep->com.mutex);
3594 	c4iw_put_ep(&ep->com);
3595 	if (fatal)
3596 		release_ep_resources(ep);
3597 	return ret;
3598 }
3599 
3600 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3601 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3602 {
3603 	struct c4iw_ep *ep;
3604 	int atid = be32_to_cpu(req->tid);
3605 
3606 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3607 					   (__force u32) req->tid);
3608 	if (!ep)
3609 		return;
3610 
3611 	switch (req->retval) {
3612 	case FW_ENOMEM:
3613 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3614 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3615 			send_fw_act_open_req(ep, atid);
3616 			return;
3617 		}
3618 	case FW_EADDRINUSE:
3619 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3620 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3621 			send_fw_act_open_req(ep, atid);
3622 			return;
3623 		}
3624 		break;
3625 	default:
3626 		pr_info("%s unexpected ofld conn wr retval %d\n",
3627 		       __func__, req->retval);
3628 		break;
3629 	}
3630 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3631 	       req->retval, atid);
3632 	mutex_lock(&dev->rdev.stats.lock);
3633 	dev->rdev.stats.act_ofld_conn_fails++;
3634 	mutex_unlock(&dev->rdev.stats.lock);
3635 	connect_reply_upcall(ep, status2errno(req->retval));
3636 	state_set(&ep->com, DEAD);
3637 	if (ep->com.remote_addr.ss_family == AF_INET6) {
3638 		struct sockaddr_in6 *sin6 =
3639 			(struct sockaddr_in6 *)&ep->com.local_addr;
3640 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3641 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3642 	}
3643 	remove_handle(dev, &dev->atid_idr, atid);
3644 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3645 	dst_release(ep->dst);
3646 	cxgb4_l2t_release(ep->l2t);
3647 	c4iw_put_ep(&ep->com);
3648 }
3649 
3650 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3651 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3652 {
3653 	struct sk_buff *rpl_skb;
3654 	struct cpl_pass_accept_req *cpl;
3655 	int ret;
3656 
3657 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3658 	BUG_ON(!rpl_skb);
3659 	if (req->retval) {
3660 		PDBG("%s passive open failure %d\n", __func__, req->retval);
3661 		mutex_lock(&dev->rdev.stats.lock);
3662 		dev->rdev.stats.pas_ofld_conn_fails++;
3663 		mutex_unlock(&dev->rdev.stats.lock);
3664 		kfree_skb(rpl_skb);
3665 	} else {
3666 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3667 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3668 					(__force u32) htonl(
3669 					(__force u32) req->tid)));
3670 		ret = pass_accept_req(dev, rpl_skb);
3671 		if (!ret)
3672 			kfree_skb(rpl_skb);
3673 	}
3674 	return;
3675 }
3676 
3677 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3678 {
3679 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3680 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3681 
3682 	switch (rpl->type) {
3683 	case FW6_TYPE_CQE:
3684 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3685 		break;
3686 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3687 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3688 		switch (req->t_state) {
3689 		case TCP_SYN_SENT:
3690 			active_ofld_conn_reply(dev, skb, req);
3691 			break;
3692 		case TCP_SYN_RECV:
3693 			passive_ofld_conn_reply(dev, skb, req);
3694 			break;
3695 		default:
3696 			pr_err("%s unexpected ofld conn wr state %d\n",
3697 			       __func__, req->t_state);
3698 			break;
3699 		}
3700 		break;
3701 	}
3702 	return 0;
3703 }
3704 
3705 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3706 {
3707 	__be32 l2info;
3708 	__be16 hdr_len, vlantag, len;
3709 	u16 eth_hdr_len;
3710 	int tcp_hdr_len, ip_hdr_len;
3711 	u8 intf;
3712 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3713 	struct cpl_pass_accept_req *req;
3714 	struct tcp_options_received tmp_opt;
3715 	struct c4iw_dev *dev;
3716 	enum chip_type type;
3717 
3718 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3719 	/* Store values from cpl_rx_pkt in temporary location. */
3720 	vlantag = cpl->vlan;
3721 	len = cpl->len;
3722 	l2info  = cpl->l2info;
3723 	hdr_len = cpl->hdr_len;
3724 	intf = cpl->iff;
3725 
3726 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3727 
3728 	/*
3729 	 * We need to parse the TCP options from SYN packet.
3730 	 * to generate cpl_pass_accept_req.
3731 	 */
3732 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3733 	tcp_clear_options(&tmp_opt);
3734 	tcp_parse_options(skb, &tmp_opt, 0, NULL);
3735 
3736 	req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3737 	memset(req, 0, sizeof(*req));
3738 	req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3739 			 SYN_MAC_IDX_V(RX_MACIDX_G(
3740 			 be32_to_cpu(l2info))) |
3741 			 SYN_XACT_MATCH_F);
3742 	type = dev->rdev.lldi.adapter_type;
3743 	tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3744 	ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3745 	req->hdr_len =
3746 		cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3747 	if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3748 		eth_hdr_len = is_t4(type) ?
3749 				RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3750 				RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3751 		req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3752 					    IP_HDR_LEN_V(ip_hdr_len) |
3753 					    ETH_HDR_LEN_V(eth_hdr_len));
3754 	} else { /* T6 and later */
3755 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3756 		req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3757 					    T6_IP_HDR_LEN_V(ip_hdr_len) |
3758 					    T6_ETH_HDR_LEN_V(eth_hdr_len));
3759 	}
3760 	req->vlan = vlantag;
3761 	req->len = len;
3762 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3763 				    PASS_OPEN_TOS_V(tos));
3764 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3765 	if (tmp_opt.wscale_ok)
3766 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3767 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3768 	if (tmp_opt.sack_ok)
3769 		req->tcpopt.sack = 1;
3770 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3771 	return;
3772 }
3773 
3774 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3775 				  __be32 laddr, __be16 lport,
3776 				  __be32 raddr, __be16 rport,
3777 				  u32 rcv_isn, u32 filter, u16 window,
3778 				  u32 rss_qid, u8 port_id)
3779 {
3780 	struct sk_buff *req_skb;
3781 	struct fw_ofld_connection_wr *req;
3782 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
3783 	int ret;
3784 
3785 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3786 	req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3787 	memset(req, 0, sizeof(*req));
3788 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3789 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3790 	req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3791 	req->le.filter = (__force __be32) filter;
3792 	req->le.lport = lport;
3793 	req->le.pport = rport;
3794 	req->le.u.ipv4.lip = laddr;
3795 	req->le.u.ipv4.pip = raddr;
3796 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3797 	req->tcb.rcv_adv = htons(window);
3798 	req->tcb.t_state_to_astid =
3799 		 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3800 			FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3801 			FW_OFLD_CONNECTION_WR_ASTID_V(
3802 			PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3803 
3804 	/*
3805 	 * We store the qid in opt2 which will be used by the firmware
3806 	 * to send us the wr response.
3807 	 */
3808 	req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3809 
3810 	/*
3811 	 * We initialize the MSS index in TCB to 0xF.
3812 	 * So that when driver sends cpl_pass_accept_rpl
3813 	 * TCB picks up the correct value. If this was 0
3814 	 * TP will ignore any value > 0 for MSS index.
3815 	 */
3816 	req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3817 	req->cookie = (uintptr_t)skb;
3818 
3819 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3820 	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3821 	if (ret < 0) {
3822 		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3823 		       ret);
3824 		kfree_skb(skb);
3825 		kfree_skb(req_skb);
3826 	}
3827 }
3828 
3829 /*
3830  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3831  * messages when a filter is being used instead of server to
3832  * redirect a syn packet. When packets hit filter they are redirected
3833  * to the offload queue and driver tries to establish the connection
3834  * using firmware work request.
3835  */
3836 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3837 {
3838 	int stid;
3839 	unsigned int filter;
3840 	struct ethhdr *eh = NULL;
3841 	struct vlan_ethhdr *vlan_eh = NULL;
3842 	struct iphdr *iph;
3843 	struct tcphdr *tcph;
3844 	struct rss_header *rss = (void *)skb->data;
3845 	struct cpl_rx_pkt *cpl = (void *)skb->data;
3846 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
3847 	struct l2t_entry *e;
3848 	struct dst_entry *dst;
3849 	struct c4iw_ep *lep = NULL;
3850 	u16 window;
3851 	struct port_info *pi;
3852 	struct net_device *pdev;
3853 	u16 rss_qid, eth_hdr_len;
3854 	int step;
3855 	u32 tx_chan;
3856 	struct neighbour *neigh;
3857 
3858 	/* Drop all non-SYN packets */
3859 	if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3860 		goto reject;
3861 
3862 	/*
3863 	 * Drop all packets which did not hit the filter.
3864 	 * Unlikely to happen.
3865 	 */
3866 	if (!(rss->filter_hit && rss->filter_tid))
3867 		goto reject;
3868 
3869 	/*
3870 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3871 	 */
3872 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3873 
3874 	lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3875 	if (!lep) {
3876 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3877 		goto reject;
3878 	}
3879 
3880 	switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3881 	case CHELSIO_T4:
3882 		eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3883 		break;
3884 	case CHELSIO_T5:
3885 		eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3886 		break;
3887 	case CHELSIO_T6:
3888 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3889 		break;
3890 	default:
3891 		pr_err("T%d Chip is not supported\n",
3892 		       CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3893 		goto reject;
3894 	}
3895 
3896 	if (eth_hdr_len == ETH_HLEN) {
3897 		eh = (struct ethhdr *)(req + 1);
3898 		iph = (struct iphdr *)(eh + 1);
3899 	} else {
3900 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
3901 		iph = (struct iphdr *)(vlan_eh + 1);
3902 		skb->vlan_tci = ntohs(cpl->vlan);
3903 	}
3904 
3905 	if (iph->version != 0x4)
3906 		goto reject;
3907 
3908 	tcph = (struct tcphdr *)(iph + 1);
3909 	skb_set_network_header(skb, (void *)iph - (void *)rss);
3910 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3911 	skb_get(skb);
3912 
3913 	PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3914 	     ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3915 	     ntohs(tcph->source), iph->tos);
3916 
3917 	dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3918 			      iph->daddr, iph->saddr, tcph->dest,
3919 			      tcph->source, iph->tos);
3920 	if (!dst) {
3921 		pr_err("%s - failed to find dst entry!\n",
3922 		       __func__);
3923 		goto reject;
3924 	}
3925 	neigh = dst_neigh_lookup_skb(dst, skb);
3926 
3927 	if (!neigh) {
3928 		pr_err("%s - failed to allocate neigh!\n",
3929 		       __func__);
3930 		goto free_dst;
3931 	}
3932 
3933 	if (neigh->dev->flags & IFF_LOOPBACK) {
3934 		pdev = ip_dev_find(&init_net, iph->daddr);
3935 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3936 				    pdev, 0);
3937 		pi = (struct port_info *)netdev_priv(pdev);
3938 		tx_chan = cxgb4_port_chan(pdev);
3939 		dev_put(pdev);
3940 	} else {
3941 		pdev = get_real_dev(neigh->dev);
3942 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3943 					pdev, 0);
3944 		pi = (struct port_info *)netdev_priv(pdev);
3945 		tx_chan = cxgb4_port_chan(pdev);
3946 	}
3947 	neigh_release(neigh);
3948 	if (!e) {
3949 		pr_err("%s - failed to allocate l2t entry!\n",
3950 		       __func__);
3951 		goto free_dst;
3952 	}
3953 
3954 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3955 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3956 	window = (__force u16) htons((__force u16)tcph->window);
3957 
3958 	/* Calcuate filter portion for LE region. */
3959 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3960 						    dev->rdev.lldi.ports[0],
3961 						    e));
3962 
3963 	/*
3964 	 * Synthesize the cpl_pass_accept_req. We have everything except the
3965 	 * TID. Once firmware sends a reply with TID we update the TID field
3966 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
3967 	 */
3968 	build_cpl_pass_accept_req(skb, stid, iph->tos);
3969 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3970 			      tcph->source, ntohl(tcph->seq), filter, window,
3971 			      rss_qid, pi->port_id);
3972 	cxgb4_l2t_release(e);
3973 free_dst:
3974 	dst_release(dst);
3975 reject:
3976 	if (lep)
3977 		c4iw_put_ep(&lep->com);
3978 	return 0;
3979 }
3980 
3981 /*
3982  * These are the real handlers that are called from a
3983  * work queue.
3984  */
3985 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
3986 	[CPL_ACT_ESTABLISH] = act_establish,
3987 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
3988 	[CPL_RX_DATA] = rx_data,
3989 	[CPL_ABORT_RPL_RSS] = abort_rpl,
3990 	[CPL_ABORT_RPL] = abort_rpl,
3991 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
3992 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3993 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3994 	[CPL_PASS_ESTABLISH] = pass_establish,
3995 	[CPL_PEER_CLOSE] = peer_close,
3996 	[CPL_ABORT_REQ_RSS] = peer_abort,
3997 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
3998 	[CPL_RDMA_TERMINATE] = terminate,
3999 	[CPL_FW4_ACK] = fw4_ack,
4000 	[CPL_FW6_MSG] = deferred_fw6_msg,
4001 	[CPL_RX_PKT] = rx_pkt,
4002 	[FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4003 	[FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4004 };
4005 
4006 static void process_timeout(struct c4iw_ep *ep)
4007 {
4008 	struct c4iw_qp_attributes attrs;
4009 	int abort = 1;
4010 
4011 	mutex_lock(&ep->com.mutex);
4012 	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4013 	     ep->com.state);
4014 	set_bit(TIMEDOUT, &ep->com.history);
4015 	switch (ep->com.state) {
4016 	case MPA_REQ_SENT:
4017 		connect_reply_upcall(ep, -ETIMEDOUT);
4018 		break;
4019 	case MPA_REQ_WAIT:
4020 	case MPA_REQ_RCVD:
4021 	case MPA_REP_SENT:
4022 	case FPDU_MODE:
4023 		break;
4024 	case CLOSING:
4025 	case MORIBUND:
4026 		if (ep->com.cm_id && ep->com.qp) {
4027 			attrs.next_state = C4IW_QP_STATE_ERROR;
4028 			c4iw_modify_qp(ep->com.qp->rhp,
4029 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4030 				     &attrs, 1);
4031 		}
4032 		close_complete_upcall(ep, -ETIMEDOUT);
4033 		break;
4034 	case ABORTING:
4035 	case DEAD:
4036 
4037 		/*
4038 		 * These states are expected if the ep timed out at the same
4039 		 * time as another thread was calling stop_ep_timer().
4040 		 * So we silently do nothing for these states.
4041 		 */
4042 		abort = 0;
4043 		break;
4044 	default:
4045 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4046 			__func__, ep, ep->hwtid, ep->com.state);
4047 		abort = 0;
4048 	}
4049 	mutex_unlock(&ep->com.mutex);
4050 	if (abort)
4051 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4052 	c4iw_put_ep(&ep->com);
4053 }
4054 
4055 static void process_timedout_eps(void)
4056 {
4057 	struct c4iw_ep *ep;
4058 
4059 	spin_lock_irq(&timeout_lock);
4060 	while (!list_empty(&timeout_list)) {
4061 		struct list_head *tmp;
4062 
4063 		tmp = timeout_list.next;
4064 		list_del(tmp);
4065 		tmp->next = NULL;
4066 		tmp->prev = NULL;
4067 		spin_unlock_irq(&timeout_lock);
4068 		ep = list_entry(tmp, struct c4iw_ep, entry);
4069 		process_timeout(ep);
4070 		spin_lock_irq(&timeout_lock);
4071 	}
4072 	spin_unlock_irq(&timeout_lock);
4073 }
4074 
4075 static void process_work(struct work_struct *work)
4076 {
4077 	struct sk_buff *skb = NULL;
4078 	struct c4iw_dev *dev;
4079 	struct cpl_act_establish *rpl;
4080 	unsigned int opcode;
4081 	int ret;
4082 
4083 	process_timedout_eps();
4084 	while ((skb = skb_dequeue(&rxq))) {
4085 		rpl = cplhdr(skb);
4086 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4087 		opcode = rpl->ot.opcode;
4088 
4089 		BUG_ON(!work_handlers[opcode]);
4090 		ret = work_handlers[opcode](dev, skb);
4091 		if (!ret)
4092 			kfree_skb(skb);
4093 		process_timedout_eps();
4094 	}
4095 }
4096 
4097 static DECLARE_WORK(skb_work, process_work);
4098 
4099 static void ep_timeout(unsigned long arg)
4100 {
4101 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4102 	int kickit = 0;
4103 
4104 	spin_lock(&timeout_lock);
4105 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4106 		/*
4107 		 * Only insert if it is not already on the list.
4108 		 */
4109 		if (!ep->entry.next) {
4110 			list_add_tail(&ep->entry, &timeout_list);
4111 			kickit = 1;
4112 		}
4113 	}
4114 	spin_unlock(&timeout_lock);
4115 	if (kickit)
4116 		queue_work(workq, &skb_work);
4117 }
4118 
4119 /*
4120  * All the CM events are handled on a work queue to have a safe context.
4121  */
4122 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4123 {
4124 
4125 	/*
4126 	 * Save dev in the skb->cb area.
4127 	 */
4128 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4129 
4130 	/*
4131 	 * Queue the skb and schedule the worker thread.
4132 	 */
4133 	skb_queue_tail(&rxq, skb);
4134 	queue_work(workq, &skb_work);
4135 	return 0;
4136 }
4137 
4138 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4139 {
4140 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4141 
4142 	if (rpl->status != CPL_ERR_NONE) {
4143 		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4144 		       "for tid %u\n", rpl->status, GET_TID(rpl));
4145 	}
4146 	kfree_skb(skb);
4147 	return 0;
4148 }
4149 
4150 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4151 {
4152 	struct cpl_fw6_msg *rpl = cplhdr(skb);
4153 	struct c4iw_wr_wait *wr_waitp;
4154 	int ret;
4155 
4156 	PDBG("%s type %u\n", __func__, rpl->type);
4157 
4158 	switch (rpl->type) {
4159 	case FW6_TYPE_WR_RPL:
4160 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4161 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4162 		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4163 		if (wr_waitp)
4164 			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4165 		kfree_skb(skb);
4166 		break;
4167 	case FW6_TYPE_CQE:
4168 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4169 		sched(dev, skb);
4170 		break;
4171 	default:
4172 		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4173 		       rpl->type);
4174 		kfree_skb(skb);
4175 		break;
4176 	}
4177 	return 0;
4178 }
4179 
4180 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4181 {
4182 	struct cpl_abort_req_rss *req = cplhdr(skb);
4183 	struct c4iw_ep *ep;
4184 	unsigned int tid = GET_TID(req);
4185 
4186 	ep = get_ep_from_tid(dev, tid);
4187 	/* This EP will be dereferenced in peer_abort() */
4188 	if (!ep) {
4189 		printk(KERN_WARNING MOD
4190 		       "Abort on non-existent endpoint, tid %d\n", tid);
4191 		kfree_skb(skb);
4192 		return 0;
4193 	}
4194 	if (cxgb_is_neg_adv(req->status)) {
4195 		PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4196 		     __func__, ep->hwtid, req->status,
4197 		     neg_adv_str(req->status));
4198 		goto out;
4199 	}
4200 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4201 	     ep->com.state);
4202 
4203 	c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4204 out:
4205 	sched(dev, skb);
4206 	return 0;
4207 }
4208 
4209 /*
4210  * Most upcalls from the T4 Core go to sched() to
4211  * schedule the processing on a work queue.
4212  */
4213 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4214 	[CPL_ACT_ESTABLISH] = sched,
4215 	[CPL_ACT_OPEN_RPL] = sched,
4216 	[CPL_RX_DATA] = sched,
4217 	[CPL_ABORT_RPL_RSS] = sched,
4218 	[CPL_ABORT_RPL] = sched,
4219 	[CPL_PASS_OPEN_RPL] = sched,
4220 	[CPL_CLOSE_LISTSRV_RPL] = sched,
4221 	[CPL_PASS_ACCEPT_REQ] = sched,
4222 	[CPL_PASS_ESTABLISH] = sched,
4223 	[CPL_PEER_CLOSE] = sched,
4224 	[CPL_CLOSE_CON_RPL] = sched,
4225 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
4226 	[CPL_RDMA_TERMINATE] = sched,
4227 	[CPL_FW4_ACK] = sched,
4228 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
4229 	[CPL_FW6_MSG] = fw6_msg,
4230 	[CPL_RX_PKT] = sched
4231 };
4232 
4233 int __init c4iw_cm_init(void)
4234 {
4235 	spin_lock_init(&timeout_lock);
4236 	skb_queue_head_init(&rxq);
4237 
4238 	workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4239 	if (!workq)
4240 		return -ENOMEM;
4241 
4242 	return 0;
4243 }
4244 
4245 void c4iw_cm_term(void)
4246 {
4247 	WARN_ON(!list_empty(&timeout_list));
4248 	flush_workqueue(workq);
4249 	destroy_workqueue(workq);
4250 }
4251