xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision f4356947)
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;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=0)");
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 static int peer2peer = 1;
103 module_param(peer2peer, int, 0644);
104 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
105 
106 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
107 module_param(p2p_type, int, 0644);
108 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
109 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
110 
111 static int ep_timeout_secs = 60;
112 module_param(ep_timeout_secs, int, 0644);
113 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
114 				   "in seconds (default=60)");
115 
116 static int mpa_rev = 2;
117 module_param(mpa_rev, int, 0644);
118 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
119 		"1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
120 		" compliant (default=2)");
121 
122 static int markers_enabled;
123 module_param(markers_enabled, int, 0644);
124 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
125 
126 static int crc_enabled = 1;
127 module_param(crc_enabled, int, 0644);
128 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
129 
130 static int rcv_win = 256 * 1024;
131 module_param(rcv_win, int, 0644);
132 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
133 
134 static int snd_win = 128 * 1024;
135 module_param(snd_win, int, 0644);
136 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
137 
138 static struct workqueue_struct *workq;
139 
140 static struct sk_buff_head rxq;
141 
142 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
143 static void ep_timeout(struct timer_list *t);
144 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
146 
147 static LIST_HEAD(timeout_list);
148 static DEFINE_SPINLOCK(timeout_lock);
149 
150 static void deref_cm_id(struct c4iw_ep_common *epc)
151 {
152 	epc->cm_id->rem_ref(epc->cm_id);
153 	epc->cm_id = NULL;
154 	set_bit(CM_ID_DEREFED, &epc->history);
155 }
156 
157 static void ref_cm_id(struct c4iw_ep_common *epc)
158 {
159 	set_bit(CM_ID_REFED, &epc->history);
160 	epc->cm_id->add_ref(epc->cm_id);
161 }
162 
163 static void deref_qp(struct c4iw_ep *ep)
164 {
165 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
166 	clear_bit(QP_REFERENCED, &ep->com.flags);
167 	set_bit(QP_DEREFED, &ep->com.history);
168 }
169 
170 static void ref_qp(struct c4iw_ep *ep)
171 {
172 	set_bit(QP_REFERENCED, &ep->com.flags);
173 	set_bit(QP_REFED, &ep->com.history);
174 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
175 }
176 
177 static void start_ep_timer(struct c4iw_ep *ep)
178 {
179 	pr_debug("ep %p\n", ep);
180 	if (timer_pending(&ep->timer)) {
181 		pr_err("%s timer already started! ep %p\n",
182 		       __func__, ep);
183 		return;
184 	}
185 	clear_bit(TIMEOUT, &ep->com.flags);
186 	c4iw_get_ep(&ep->com);
187 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
188 	add_timer(&ep->timer);
189 }
190 
191 static int stop_ep_timer(struct c4iw_ep *ep)
192 {
193 	pr_debug("ep %p stopping\n", ep);
194 	del_timer_sync(&ep->timer);
195 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
196 		c4iw_put_ep(&ep->com);
197 		return 0;
198 	}
199 	return 1;
200 }
201 
202 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
203 		  struct l2t_entry *l2e)
204 {
205 	int	error = 0;
206 
207 	if (c4iw_fatal_error(rdev)) {
208 		kfree_skb(skb);
209 		pr_err("%s - device in error state - dropping\n", __func__);
210 		return -EIO;
211 	}
212 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
213 	if (error < 0)
214 		kfree_skb(skb);
215 	else if (error == NET_XMIT_DROP)
216 		return -ENOMEM;
217 	return error < 0 ? error : 0;
218 }
219 
220 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
221 {
222 	int	error = 0;
223 
224 	if (c4iw_fatal_error(rdev)) {
225 		kfree_skb(skb);
226 		pr_err("%s - device in error state - dropping\n", __func__);
227 		return -EIO;
228 	}
229 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
230 	if (error < 0)
231 		kfree_skb(skb);
232 	return error < 0 ? error : 0;
233 }
234 
235 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
236 {
237 	u32 len = roundup(sizeof(struct cpl_tid_release), 16);
238 
239 	skb = get_skb(skb, len, GFP_KERNEL);
240 	if (!skb)
241 		return;
242 
243 	cxgb_mk_tid_release(skb, len, hwtid, 0);
244 	c4iw_ofld_send(rdev, skb);
245 	return;
246 }
247 
248 static void set_emss(struct c4iw_ep *ep, u16 opt)
249 {
250 	ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
251 		   ((AF_INET == ep->com.remote_addr.ss_family) ?
252 		    sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
253 		   sizeof(struct tcphdr);
254 	ep->mss = ep->emss;
255 	if (TCPOPT_TSTAMP_G(opt))
256 		ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
257 	if (ep->emss < 128)
258 		ep->emss = 128;
259 	if (ep->emss & 7)
260 		pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
261 			 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
262 	pr_debug("mss_idx %u mss %u emss=%u\n", TCPOPT_MSS_G(opt), ep->mss,
263 		 ep->emss);
264 }
265 
266 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
267 {
268 	enum c4iw_ep_state state;
269 
270 	mutex_lock(&epc->mutex);
271 	state = epc->state;
272 	mutex_unlock(&epc->mutex);
273 	return state;
274 }
275 
276 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
277 {
278 	epc->state = new;
279 }
280 
281 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
282 {
283 	mutex_lock(&epc->mutex);
284 	pr_debug("%s -> %s\n", states[epc->state], states[new]);
285 	__state_set(epc, new);
286 	mutex_unlock(&epc->mutex);
287 	return;
288 }
289 
290 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
291 {
292 	struct sk_buff *skb;
293 	unsigned int i;
294 	size_t len;
295 
296 	len = roundup(sizeof(union cpl_wr_size), 16);
297 	for (i = 0; i < size; i++) {
298 		skb = alloc_skb(len, GFP_KERNEL);
299 		if (!skb)
300 			goto fail;
301 		skb_queue_tail(ep_skb_list, skb);
302 	}
303 	return 0;
304 fail:
305 	skb_queue_purge(ep_skb_list);
306 	return -ENOMEM;
307 }
308 
309 static void *alloc_ep(int size, gfp_t gfp)
310 {
311 	struct c4iw_ep_common *epc;
312 
313 	epc = kzalloc(size, gfp);
314 	if (epc) {
315 		epc->wr_waitp = c4iw_alloc_wr_wait(gfp);
316 		if (!epc->wr_waitp) {
317 			kfree(epc);
318 			epc = NULL;
319 			goto out;
320 		}
321 		kref_init(&epc->kref);
322 		mutex_init(&epc->mutex);
323 		c4iw_init_wr_wait(epc->wr_waitp);
324 	}
325 	pr_debug("alloc ep %p\n", epc);
326 out:
327 	return epc;
328 }
329 
330 static void remove_ep_tid(struct c4iw_ep *ep)
331 {
332 	unsigned long flags;
333 
334 	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
335 	__xa_erase(&ep->com.dev->hwtids, ep->hwtid);
336 	if (xa_empty(&ep->com.dev->hwtids))
337 		wake_up(&ep->com.dev->wait);
338 	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
339 }
340 
341 static int insert_ep_tid(struct c4iw_ep *ep)
342 {
343 	unsigned long flags;
344 	int err;
345 
346 	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
347 	err = __xa_insert(&ep->com.dev->hwtids, ep->hwtid, ep, GFP_KERNEL);
348 	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
349 
350 	return err;
351 }
352 
353 /*
354  * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
355  */
356 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
357 {
358 	struct c4iw_ep *ep;
359 	unsigned long flags;
360 
361 	xa_lock_irqsave(&dev->hwtids, flags);
362 	ep = xa_load(&dev->hwtids, tid);
363 	if (ep)
364 		c4iw_get_ep(&ep->com);
365 	xa_unlock_irqrestore(&dev->hwtids, flags);
366 	return ep;
367 }
368 
369 /*
370  * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
371  */
372 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
373 					       unsigned int stid)
374 {
375 	struct c4iw_listen_ep *ep;
376 	unsigned long flags;
377 
378 	xa_lock_irqsave(&dev->stids, flags);
379 	ep = xa_load(&dev->stids, stid);
380 	if (ep)
381 		c4iw_get_ep(&ep->com);
382 	xa_unlock_irqrestore(&dev->stids, flags);
383 	return ep;
384 }
385 
386 void _c4iw_free_ep(struct kref *kref)
387 {
388 	struct c4iw_ep *ep;
389 
390 	ep = container_of(kref, struct c4iw_ep, com.kref);
391 	pr_debug("ep %p state %s\n", ep, states[ep->com.state]);
392 	if (test_bit(QP_REFERENCED, &ep->com.flags))
393 		deref_qp(ep);
394 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
395 		if (ep->com.remote_addr.ss_family == AF_INET6) {
396 			struct sockaddr_in6 *sin6 =
397 					(struct sockaddr_in6 *)
398 					&ep->com.local_addr;
399 
400 			cxgb4_clip_release(
401 					ep->com.dev->rdev.lldi.ports[0],
402 					(const u32 *)&sin6->sin6_addr.s6_addr,
403 					1);
404 		}
405 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
406 				 ep->com.local_addr.ss_family);
407 		dst_release(ep->dst);
408 		cxgb4_l2t_release(ep->l2t);
409 		kfree_skb(ep->mpa_skb);
410 	}
411 	if (!skb_queue_empty(&ep->com.ep_skb_list))
412 		skb_queue_purge(&ep->com.ep_skb_list);
413 	c4iw_put_wr_wait(ep->com.wr_waitp);
414 	kfree(ep);
415 }
416 
417 static void release_ep_resources(struct c4iw_ep *ep)
418 {
419 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
420 
421 	/*
422 	 * If we have a hwtid, then remove it from the idr table
423 	 * so lookups will no longer find this endpoint.  Otherwise
424 	 * we have a race where one thread finds the ep ptr just
425 	 * before the other thread is freeing the ep memory.
426 	 */
427 	if (ep->hwtid != -1)
428 		remove_ep_tid(ep);
429 	c4iw_put_ep(&ep->com);
430 }
431 
432 static int status2errno(int status)
433 {
434 	switch (status) {
435 	case CPL_ERR_NONE:
436 		return 0;
437 	case CPL_ERR_CONN_RESET:
438 		return -ECONNRESET;
439 	case CPL_ERR_ARP_MISS:
440 		return -EHOSTUNREACH;
441 	case CPL_ERR_CONN_TIMEDOUT:
442 		return -ETIMEDOUT;
443 	case CPL_ERR_TCAM_FULL:
444 		return -ENOMEM;
445 	case CPL_ERR_CONN_EXIST:
446 		return -EADDRINUSE;
447 	default:
448 		return -EIO;
449 	}
450 }
451 
452 /*
453  * Try and reuse skbs already allocated...
454  */
455 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
456 {
457 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
458 		skb_trim(skb, 0);
459 		skb_get(skb);
460 		skb_reset_transport_header(skb);
461 	} else {
462 		skb = alloc_skb(len, gfp);
463 		if (!skb)
464 			return NULL;
465 	}
466 	t4_set_arp_err_handler(skb, NULL, NULL);
467 	return skb;
468 }
469 
470 static struct net_device *get_real_dev(struct net_device *egress_dev)
471 {
472 	return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
473 }
474 
475 static void arp_failure_discard(void *handle, struct sk_buff *skb)
476 {
477 	pr_err("ARP failure\n");
478 	kfree_skb(skb);
479 }
480 
481 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
482 {
483 	pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
484 }
485 
486 enum {
487 	NUM_FAKE_CPLS = 2,
488 	FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
489 	FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
490 };
491 
492 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
493 {
494 	struct c4iw_ep *ep;
495 
496 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497 	release_ep_resources(ep);
498 	return 0;
499 }
500 
501 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
502 {
503 	struct c4iw_ep *ep;
504 
505 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
506 	c4iw_put_ep(&ep->parent_ep->com);
507 	release_ep_resources(ep);
508 	return 0;
509 }
510 
511 /*
512  * Fake up a special CPL opcode and call sched() so process_work() will call
513  * _put_ep_safe() in a safe context to free the ep resources.  This is needed
514  * because ARP error handlers are called in an ATOMIC context, and
515  * _c4iw_free_ep() needs to block.
516  */
517 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
518 				  int cpl)
519 {
520 	struct cpl_act_establish *rpl = cplhdr(skb);
521 
522 	/* Set our special ARP_FAILURE opcode */
523 	rpl->ot.opcode = cpl;
524 
525 	/*
526 	 * Save ep in the skb->cb area, after where sched() will save the dev
527 	 * ptr.
528 	 */
529 	*((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
530 	sched(ep->com.dev, skb);
531 }
532 
533 /* Handle an ARP failure for an accept */
534 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
535 {
536 	struct c4iw_ep *ep = handle;
537 
538 	pr_err("ARP failure during accept - tid %u - dropping connection\n",
539 	       ep->hwtid);
540 
541 	__state_set(&ep->com, DEAD);
542 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
543 }
544 
545 /*
546  * Handle an ARP failure for an active open.
547  */
548 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
549 {
550 	struct c4iw_ep *ep = handle;
551 
552 	pr_err("ARP failure during connect\n");
553 	connect_reply_upcall(ep, -EHOSTUNREACH);
554 	__state_set(&ep->com, DEAD);
555 	if (ep->com.remote_addr.ss_family == AF_INET6) {
556 		struct sockaddr_in6 *sin6 =
557 			(struct sockaddr_in6 *)&ep->com.local_addr;
558 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
559 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
560 	}
561 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
562 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
563 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
564 }
565 
566 /*
567  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
568  * and send it along.
569  */
570 static void abort_arp_failure(void *handle, struct sk_buff *skb)
571 {
572 	int ret;
573 	struct c4iw_ep *ep = handle;
574 	struct c4iw_rdev *rdev = &ep->com.dev->rdev;
575 	struct cpl_abort_req *req = cplhdr(skb);
576 
577 	pr_debug("rdev %p\n", rdev);
578 	req->cmd = CPL_ABORT_NO_RST;
579 	skb_get(skb);
580 	ret = c4iw_ofld_send(rdev, skb);
581 	if (ret) {
582 		__state_set(&ep->com, DEAD);
583 		queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
584 	} else
585 		kfree_skb(skb);
586 }
587 
588 static int send_flowc(struct c4iw_ep *ep)
589 {
590 	struct fw_flowc_wr *flowc;
591 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
592 	u16 vlan = ep->l2t->vlan;
593 	int nparams;
594 	int flowclen, flowclen16;
595 
596 	if (WARN_ON(!skb))
597 		return -ENOMEM;
598 
599 	if (vlan == CPL_L2T_VLAN_NONE)
600 		nparams = 9;
601 	else
602 		nparams = 10;
603 
604 	flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
605 	flowclen16 = DIV_ROUND_UP(flowclen, 16);
606 	flowclen = flowclen16 * 16;
607 
608 	flowc = __skb_put(skb, flowclen);
609 	memset(flowc, 0, flowclen);
610 
611 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
612 					   FW_FLOWC_WR_NPARAMS_V(nparams));
613 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
614 					  FW_WR_FLOWID_V(ep->hwtid));
615 
616 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
617 	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
618 					    (ep->com.dev->rdev.lldi.pf));
619 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
620 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
621 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
622 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
623 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
624 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
625 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
626 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
627 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
628 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
629 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
630 	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
631 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
632 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
633 	flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
634 	flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale);
635 	if (nparams == 10) {
636 		u16 pri;
637 		pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
638 		flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
639 		flowc->mnemval[9].val = cpu_to_be32(pri);
640 	}
641 
642 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
643 	return c4iw_ofld_send(&ep->com.dev->rdev, skb);
644 }
645 
646 static int send_halfclose(struct c4iw_ep *ep)
647 {
648 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
649 	u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
650 
651 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
652 	if (WARN_ON(!skb))
653 		return -ENOMEM;
654 
655 	cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
656 			      NULL, arp_failure_discard);
657 
658 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
659 }
660 
661 static void read_tcb(struct c4iw_ep *ep)
662 {
663 	struct sk_buff *skb;
664 	struct cpl_get_tcb *req;
665 	int wrlen = roundup(sizeof(*req), 16);
666 
667 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
668 	if (WARN_ON(!skb))
669 		return;
670 
671 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
672 	req = (struct cpl_get_tcb *) skb_put(skb, wrlen);
673 	memset(req, 0, wrlen);
674 	INIT_TP_WR(req, ep->hwtid);
675 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
676 	req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
677 
678 	/*
679 	 * keep a ref on the ep so the tcb is not unlocked before this
680 	 * cpl completes. The ref is released in read_tcb_rpl().
681 	 */
682 	c4iw_get_ep(&ep->com);
683 	if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
684 		c4iw_put_ep(&ep->com);
685 }
686 
687 static int send_abort_req(struct c4iw_ep *ep)
688 {
689 	u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
690 	struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
691 
692 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
693 	if (WARN_ON(!req_skb))
694 		return -ENOMEM;
695 
696 	cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
697 			  ep, abort_arp_failure);
698 
699 	return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
700 }
701 
702 static int send_abort(struct c4iw_ep *ep)
703 {
704 	if (!ep->com.qp || !ep->com.qp->srq) {
705 		send_abort_req(ep);
706 		return 0;
707 	}
708 	set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags);
709 	read_tcb(ep);
710 	return 0;
711 }
712 
713 static int send_connect(struct c4iw_ep *ep)
714 {
715 	struct cpl_act_open_req *req = NULL;
716 	struct cpl_t5_act_open_req *t5req = NULL;
717 	struct cpl_t6_act_open_req *t6req = NULL;
718 	struct cpl_act_open_req6 *req6 = NULL;
719 	struct cpl_t5_act_open_req6 *t5req6 = NULL;
720 	struct cpl_t6_act_open_req6 *t6req6 = NULL;
721 	struct sk_buff *skb;
722 	u64 opt0;
723 	u32 opt2;
724 	unsigned int mtu_idx;
725 	u32 wscale;
726 	int win, sizev4, sizev6, wrlen;
727 	struct sockaddr_in *la = (struct sockaddr_in *)
728 				 &ep->com.local_addr;
729 	struct sockaddr_in *ra = (struct sockaddr_in *)
730 				 &ep->com.remote_addr;
731 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
732 				   &ep->com.local_addr;
733 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
734 				   &ep->com.remote_addr;
735 	int ret;
736 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
737 	u32 isn = (get_random_u32() & ~7UL) - 1;
738 	struct net_device *netdev;
739 	u64 params;
740 
741 	netdev = ep->com.dev->rdev.lldi.ports[0];
742 
743 	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
744 	case CHELSIO_T4:
745 		sizev4 = sizeof(struct cpl_act_open_req);
746 		sizev6 = sizeof(struct cpl_act_open_req6);
747 		break;
748 	case CHELSIO_T5:
749 		sizev4 = sizeof(struct cpl_t5_act_open_req);
750 		sizev6 = sizeof(struct cpl_t5_act_open_req6);
751 		break;
752 	case CHELSIO_T6:
753 		sizev4 = sizeof(struct cpl_t6_act_open_req);
754 		sizev6 = sizeof(struct cpl_t6_act_open_req6);
755 		break;
756 	default:
757 		pr_err("T%d Chip is not supported\n",
758 		       CHELSIO_CHIP_VERSION(adapter_type));
759 		return -EINVAL;
760 	}
761 
762 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
763 			roundup(sizev4, 16) :
764 			roundup(sizev6, 16);
765 
766 	pr_debug("ep %p atid %u\n", ep, ep->atid);
767 
768 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
769 	if (!skb) {
770 		pr_err("%s - failed to alloc skb\n", __func__);
771 		return -ENOMEM;
772 	}
773 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
774 
775 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
776 		      enable_tcp_timestamps,
777 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
778 	wscale = cxgb_compute_wscale(rcv_win);
779 
780 	/*
781 	 * Specify the largest window that will fit in opt0. The
782 	 * remainder will be specified in the rx_data_ack.
783 	 */
784 	win = ep->rcv_win >> 10;
785 	if (win > RCV_BUFSIZ_M)
786 		win = RCV_BUFSIZ_M;
787 
788 	opt0 = (nocong ? NO_CONG_F : 0) |
789 	       KEEP_ALIVE_F |
790 	       DELACK_F |
791 	       WND_SCALE_V(wscale) |
792 	       MSS_IDX_V(mtu_idx) |
793 	       L2T_IDX_V(ep->l2t->idx) |
794 	       TX_CHAN_V(ep->tx_chan) |
795 	       SMAC_SEL_V(ep->smac_idx) |
796 	       DSCP_V(ep->tos >> 2) |
797 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
798 	       RCV_BUFSIZ_V(win);
799 	opt2 = RX_CHANNEL_V(0) |
800 	       CCTRL_ECN_V(enable_ecn) |
801 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
802 	if (enable_tcp_timestamps)
803 		opt2 |= TSTAMPS_EN_F;
804 	if (enable_tcp_sack)
805 		opt2 |= SACK_EN_F;
806 	if (wscale && enable_tcp_window_scaling)
807 		opt2 |= WND_SCALE_EN_F;
808 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
809 		if (peer2peer)
810 			isn += 4;
811 
812 		opt2 |= T5_OPT_2_VALID_F;
813 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
814 		opt2 |= T5_ISS_F;
815 	}
816 
817 	params = cxgb4_select_ntuple(netdev, ep->l2t);
818 
819 	if (ep->com.remote_addr.ss_family == AF_INET6)
820 		cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
821 			       (const u32 *)&la6->sin6_addr.s6_addr, 1);
822 
823 	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
824 
825 	if (ep->com.remote_addr.ss_family == AF_INET) {
826 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
827 		case CHELSIO_T4:
828 			req = skb_put(skb, wrlen);
829 			INIT_TP_WR(req, 0);
830 			break;
831 		case CHELSIO_T5:
832 			t5req = skb_put(skb, wrlen);
833 			INIT_TP_WR(t5req, 0);
834 			req = (struct cpl_act_open_req *)t5req;
835 			break;
836 		case CHELSIO_T6:
837 			t6req = skb_put(skb, wrlen);
838 			INIT_TP_WR(t6req, 0);
839 			req = (struct cpl_act_open_req *)t6req;
840 			t5req = (struct cpl_t5_act_open_req *)t6req;
841 			break;
842 		default:
843 			pr_err("T%d Chip is not supported\n",
844 			       CHELSIO_CHIP_VERSION(adapter_type));
845 			ret = -EINVAL;
846 			goto clip_release;
847 		}
848 
849 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
850 					((ep->rss_qid<<14) | ep->atid)));
851 		req->local_port = la->sin_port;
852 		req->peer_port = ra->sin_port;
853 		req->local_ip = la->sin_addr.s_addr;
854 		req->peer_ip = ra->sin_addr.s_addr;
855 		req->opt0 = cpu_to_be64(opt0);
856 
857 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
858 			req->params = cpu_to_be32(params);
859 			req->opt2 = cpu_to_be32(opt2);
860 		} else {
861 			if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
862 				t5req->params =
863 					  cpu_to_be64(FILTER_TUPLE_V(params));
864 				t5req->rsvd = cpu_to_be32(isn);
865 				pr_debug("snd_isn %u\n", t5req->rsvd);
866 				t5req->opt2 = cpu_to_be32(opt2);
867 			} else {
868 				t6req->params =
869 					  cpu_to_be64(FILTER_TUPLE_V(params));
870 				t6req->rsvd = cpu_to_be32(isn);
871 				pr_debug("snd_isn %u\n", t6req->rsvd);
872 				t6req->opt2 = cpu_to_be32(opt2);
873 			}
874 		}
875 	} else {
876 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
877 		case CHELSIO_T4:
878 			req6 = skb_put(skb, wrlen);
879 			INIT_TP_WR(req6, 0);
880 			break;
881 		case CHELSIO_T5:
882 			t5req6 = skb_put(skb, wrlen);
883 			INIT_TP_WR(t5req6, 0);
884 			req6 = (struct cpl_act_open_req6 *)t5req6;
885 			break;
886 		case CHELSIO_T6:
887 			t6req6 = skb_put(skb, wrlen);
888 			INIT_TP_WR(t6req6, 0);
889 			req6 = (struct cpl_act_open_req6 *)t6req6;
890 			t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
891 			break;
892 		default:
893 			pr_err("T%d Chip is not supported\n",
894 			       CHELSIO_CHIP_VERSION(adapter_type));
895 			ret = -EINVAL;
896 			goto clip_release;
897 		}
898 
899 		OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
900 					((ep->rss_qid<<14)|ep->atid)));
901 		req6->local_port = la6->sin6_port;
902 		req6->peer_port = ra6->sin6_port;
903 		req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
904 		req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
905 		req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
906 		req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
907 		req6->opt0 = cpu_to_be64(opt0);
908 
909 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
910 			req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
911 								      ep->l2t));
912 			req6->opt2 = cpu_to_be32(opt2);
913 		} else {
914 			if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
915 				t5req6->params =
916 					    cpu_to_be64(FILTER_TUPLE_V(params));
917 				t5req6->rsvd = cpu_to_be32(isn);
918 				pr_debug("snd_isn %u\n", t5req6->rsvd);
919 				t5req6->opt2 = cpu_to_be32(opt2);
920 			} else {
921 				t6req6->params =
922 					    cpu_to_be64(FILTER_TUPLE_V(params));
923 				t6req6->rsvd = cpu_to_be32(isn);
924 				pr_debug("snd_isn %u\n", t6req6->rsvd);
925 				t6req6->opt2 = cpu_to_be32(opt2);
926 			}
927 
928 		}
929 	}
930 
931 	set_bit(ACT_OPEN_REQ, &ep->com.history);
932 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
933 clip_release:
934 	if (ret && ep->com.remote_addr.ss_family == AF_INET6)
935 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
936 				   (const u32 *)&la6->sin6_addr.s6_addr, 1);
937 	return ret;
938 }
939 
940 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
941 			u8 mpa_rev_to_use)
942 {
943 	int mpalen, wrlen, ret;
944 	struct fw_ofld_tx_data_wr *req;
945 	struct mpa_message *mpa;
946 	struct mpa_v2_conn_params mpa_v2_params;
947 
948 	pr_debug("ep %p tid %u pd_len %d\n",
949 		 ep, ep->hwtid, ep->plen);
950 
951 	mpalen = sizeof(*mpa) + ep->plen;
952 	if (mpa_rev_to_use == 2)
953 		mpalen += sizeof(struct mpa_v2_conn_params);
954 	wrlen = roundup(mpalen + sizeof(*req), 16);
955 	skb = get_skb(skb, wrlen, GFP_KERNEL);
956 	if (!skb) {
957 		connect_reply_upcall(ep, -ENOMEM);
958 		return -ENOMEM;
959 	}
960 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
961 
962 	req = skb_put_zero(skb, wrlen);
963 	req->op_to_immdlen = cpu_to_be32(
964 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
965 		FW_WR_COMPL_F |
966 		FW_WR_IMMDLEN_V(mpalen));
967 	req->flowid_len16 = cpu_to_be32(
968 		FW_WR_FLOWID_V(ep->hwtid) |
969 		FW_WR_LEN16_V(wrlen >> 4));
970 	req->plen = cpu_to_be32(mpalen);
971 	req->tunnel_to_proxy = cpu_to_be32(
972 		FW_OFLD_TX_DATA_WR_FLUSH_F |
973 		FW_OFLD_TX_DATA_WR_SHOVE_F);
974 
975 	mpa = (struct mpa_message *)(req + 1);
976 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
977 
978 	mpa->flags = 0;
979 	if (crc_enabled)
980 		mpa->flags |= MPA_CRC;
981 	if (markers_enabled) {
982 		mpa->flags |= MPA_MARKERS;
983 		ep->mpa_attr.recv_marker_enabled = 1;
984 	} else {
985 		ep->mpa_attr.recv_marker_enabled = 0;
986 	}
987 	if (mpa_rev_to_use == 2)
988 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
989 
990 	mpa->private_data_size = htons(ep->plen);
991 	mpa->revision = mpa_rev_to_use;
992 	if (mpa_rev_to_use == 1) {
993 		ep->tried_with_mpa_v1 = 1;
994 		ep->retry_with_mpa_v1 = 0;
995 	}
996 
997 	if (mpa_rev_to_use == 2) {
998 		mpa->private_data_size =
999 			htons(ntohs(mpa->private_data_size) +
1000 			      sizeof(struct mpa_v2_conn_params));
1001 		pr_debug("initiator ird %u ord %u\n", ep->ird,
1002 			 ep->ord);
1003 		mpa_v2_params.ird = htons((u16)ep->ird);
1004 		mpa_v2_params.ord = htons((u16)ep->ord);
1005 
1006 		if (peer2peer) {
1007 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1008 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1009 				mpa_v2_params.ord |=
1010 					htons(MPA_V2_RDMA_WRITE_RTR);
1011 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1012 				mpa_v2_params.ord |=
1013 					htons(MPA_V2_RDMA_READ_RTR);
1014 		}
1015 		memcpy(mpa->private_data, &mpa_v2_params,
1016 		       sizeof(struct mpa_v2_conn_params));
1017 
1018 		if (ep->plen)
1019 			memcpy(mpa->private_data +
1020 			       sizeof(struct mpa_v2_conn_params),
1021 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
1022 	} else
1023 		if (ep->plen)
1024 			memcpy(mpa->private_data,
1025 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1026 
1027 	/*
1028 	 * Reference the mpa skb.  This ensures the data area
1029 	 * will remain in memory until the hw acks the tx.
1030 	 * Function fw4_ack() will deref it.
1031 	 */
1032 	skb_get(skb);
1033 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1034 	ep->mpa_skb = skb;
1035 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1036 	if (ret)
1037 		return ret;
1038 	start_ep_timer(ep);
1039 	__state_set(&ep->com, MPA_REQ_SENT);
1040 	ep->mpa_attr.initiator = 1;
1041 	ep->snd_seq += mpalen;
1042 	return ret;
1043 }
1044 
1045 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1046 {
1047 	int mpalen, wrlen;
1048 	struct fw_ofld_tx_data_wr *req;
1049 	struct mpa_message *mpa;
1050 	struct sk_buff *skb;
1051 	struct mpa_v2_conn_params mpa_v2_params;
1052 
1053 	pr_debug("ep %p tid %u pd_len %d\n",
1054 		 ep, ep->hwtid, ep->plen);
1055 
1056 	mpalen = sizeof(*mpa) + plen;
1057 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1058 		mpalen += sizeof(struct mpa_v2_conn_params);
1059 	wrlen = roundup(mpalen + sizeof(*req), 16);
1060 
1061 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1062 	if (!skb) {
1063 		pr_err("%s - cannot alloc skb!\n", __func__);
1064 		return -ENOMEM;
1065 	}
1066 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1067 
1068 	req = skb_put_zero(skb, wrlen);
1069 	req->op_to_immdlen = cpu_to_be32(
1070 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1071 		FW_WR_COMPL_F |
1072 		FW_WR_IMMDLEN_V(mpalen));
1073 	req->flowid_len16 = cpu_to_be32(
1074 		FW_WR_FLOWID_V(ep->hwtid) |
1075 		FW_WR_LEN16_V(wrlen >> 4));
1076 	req->plen = cpu_to_be32(mpalen);
1077 	req->tunnel_to_proxy = cpu_to_be32(
1078 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1079 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1080 
1081 	mpa = (struct mpa_message *)(req + 1);
1082 	memset(mpa, 0, sizeof(*mpa));
1083 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1084 	mpa->flags = MPA_REJECT;
1085 	mpa->revision = ep->mpa_attr.version;
1086 	mpa->private_data_size = htons(plen);
1087 
1088 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1089 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1090 		mpa->private_data_size =
1091 			htons(ntohs(mpa->private_data_size) +
1092 			      sizeof(struct mpa_v2_conn_params));
1093 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1094 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1095 					   0));
1096 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1097 					  (p2p_type ==
1098 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1099 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1100 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
1101 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
1102 		memcpy(mpa->private_data, &mpa_v2_params,
1103 		       sizeof(struct mpa_v2_conn_params));
1104 
1105 		if (ep->plen)
1106 			memcpy(mpa->private_data +
1107 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1108 	} else
1109 		if (plen)
1110 			memcpy(mpa->private_data, pdata, plen);
1111 
1112 	/*
1113 	 * Reference the mpa skb again.  This ensures the data area
1114 	 * will remain in memory until the hw acks the tx.
1115 	 * Function fw4_ack() will deref it.
1116 	 */
1117 	skb_get(skb);
1118 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1119 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1120 	ep->mpa_skb = skb;
1121 	ep->snd_seq += mpalen;
1122 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1123 }
1124 
1125 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1126 {
1127 	int mpalen, wrlen;
1128 	struct fw_ofld_tx_data_wr *req;
1129 	struct mpa_message *mpa;
1130 	struct sk_buff *skb;
1131 	struct mpa_v2_conn_params mpa_v2_params;
1132 
1133 	pr_debug("ep %p tid %u pd_len %d\n",
1134 		 ep, ep->hwtid, ep->plen);
1135 
1136 	mpalen = sizeof(*mpa) + plen;
1137 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1138 		mpalen += sizeof(struct mpa_v2_conn_params);
1139 	wrlen = roundup(mpalen + sizeof(*req), 16);
1140 
1141 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1142 	if (!skb) {
1143 		pr_err("%s - cannot alloc skb!\n", __func__);
1144 		return -ENOMEM;
1145 	}
1146 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1147 
1148 	req = skb_put_zero(skb, wrlen);
1149 	req->op_to_immdlen = cpu_to_be32(
1150 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1151 		FW_WR_COMPL_F |
1152 		FW_WR_IMMDLEN_V(mpalen));
1153 	req->flowid_len16 = cpu_to_be32(
1154 		FW_WR_FLOWID_V(ep->hwtid) |
1155 		FW_WR_LEN16_V(wrlen >> 4));
1156 	req->plen = cpu_to_be32(mpalen);
1157 	req->tunnel_to_proxy = cpu_to_be32(
1158 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1159 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1160 
1161 	mpa = (struct mpa_message *)(req + 1);
1162 	memset(mpa, 0, sizeof(*mpa));
1163 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1164 	mpa->flags = 0;
1165 	if (ep->mpa_attr.crc_enabled)
1166 		mpa->flags |= MPA_CRC;
1167 	if (ep->mpa_attr.recv_marker_enabled)
1168 		mpa->flags |= MPA_MARKERS;
1169 	mpa->revision = ep->mpa_attr.version;
1170 	mpa->private_data_size = htons(plen);
1171 
1172 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1173 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1174 		mpa->private_data_size =
1175 			htons(ntohs(mpa->private_data_size) +
1176 			      sizeof(struct mpa_v2_conn_params));
1177 		mpa_v2_params.ird = htons((u16)ep->ird);
1178 		mpa_v2_params.ord = htons((u16)ep->ord);
1179 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1180 					FW_RI_INIT_P2PTYPE_DISABLED)) {
1181 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1182 
1183 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1184 				mpa_v2_params.ord |=
1185 					htons(MPA_V2_RDMA_WRITE_RTR);
1186 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1187 				mpa_v2_params.ord |=
1188 					htons(MPA_V2_RDMA_READ_RTR);
1189 		}
1190 
1191 		memcpy(mpa->private_data, &mpa_v2_params,
1192 		       sizeof(struct mpa_v2_conn_params));
1193 
1194 		if (ep->plen)
1195 			memcpy(mpa->private_data +
1196 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1197 	} else
1198 		if (plen)
1199 			memcpy(mpa->private_data, pdata, plen);
1200 
1201 	/*
1202 	 * Reference the mpa skb.  This ensures the data area
1203 	 * will remain in memory until the hw acks the tx.
1204 	 * Function fw4_ack() will deref it.
1205 	 */
1206 	skb_get(skb);
1207 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1208 	ep->mpa_skb = skb;
1209 	__state_set(&ep->com, MPA_REP_SENT);
1210 	ep->snd_seq += mpalen;
1211 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1212 }
1213 
1214 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1215 {
1216 	struct c4iw_ep *ep;
1217 	struct cpl_act_establish *req = cplhdr(skb);
1218 	unsigned short tcp_opt = ntohs(req->tcp_opt);
1219 	unsigned int tid = GET_TID(req);
1220 	unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1221 	struct tid_info *t = dev->rdev.lldi.tids;
1222 	int ret;
1223 
1224 	ep = lookup_atid(t, atid);
1225 
1226 	pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1227 		 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1228 
1229 	mutex_lock(&ep->com.mutex);
1230 	dst_confirm(ep->dst);
1231 
1232 	/* setup the hwtid for this connection */
1233 	ep->hwtid = tid;
1234 	cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1235 	insert_ep_tid(ep);
1236 
1237 	ep->snd_seq = be32_to_cpu(req->snd_isn);
1238 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1239 	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1240 
1241 	set_emss(ep, tcp_opt);
1242 
1243 	/* dealloc the atid */
1244 	xa_erase_irq(&ep->com.dev->atids, atid);
1245 	cxgb4_free_atid(t, atid);
1246 	set_bit(ACT_ESTAB, &ep->com.history);
1247 
1248 	/* start MPA negotiation */
1249 	ret = send_flowc(ep);
1250 	if (ret)
1251 		goto err;
1252 	if (ep->retry_with_mpa_v1)
1253 		ret = send_mpa_req(ep, skb, 1);
1254 	else
1255 		ret = send_mpa_req(ep, skb, mpa_rev);
1256 	if (ret)
1257 		goto err;
1258 	mutex_unlock(&ep->com.mutex);
1259 	return 0;
1260 err:
1261 	mutex_unlock(&ep->com.mutex);
1262 	connect_reply_upcall(ep, -ENOMEM);
1263 	c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1264 	return 0;
1265 }
1266 
1267 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1268 {
1269 	struct iw_cm_event event;
1270 
1271 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1272 	memset(&event, 0, sizeof(event));
1273 	event.event = IW_CM_EVENT_CLOSE;
1274 	event.status = status;
1275 	if (ep->com.cm_id) {
1276 		pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1277 			 ep, ep->com.cm_id, ep->hwtid);
1278 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1279 		deref_cm_id(&ep->com);
1280 		set_bit(CLOSE_UPCALL, &ep->com.history);
1281 	}
1282 }
1283 
1284 static void peer_close_upcall(struct c4iw_ep *ep)
1285 {
1286 	struct iw_cm_event event;
1287 
1288 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1289 	memset(&event, 0, sizeof(event));
1290 	event.event = IW_CM_EVENT_DISCONNECT;
1291 	if (ep->com.cm_id) {
1292 		pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1293 			 ep, ep->com.cm_id, ep->hwtid);
1294 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1295 		set_bit(DISCONN_UPCALL, &ep->com.history);
1296 	}
1297 }
1298 
1299 static void peer_abort_upcall(struct c4iw_ep *ep)
1300 {
1301 	struct iw_cm_event event;
1302 
1303 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1304 	memset(&event, 0, sizeof(event));
1305 	event.event = IW_CM_EVENT_CLOSE;
1306 	event.status = -ECONNRESET;
1307 	if (ep->com.cm_id) {
1308 		pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1309 			 ep->com.cm_id, ep->hwtid);
1310 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1311 		deref_cm_id(&ep->com);
1312 		set_bit(ABORT_UPCALL, &ep->com.history);
1313 	}
1314 }
1315 
1316 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1317 {
1318 	struct iw_cm_event event;
1319 
1320 	pr_debug("ep %p tid %u status %d\n",
1321 		 ep, ep->hwtid, status);
1322 	memset(&event, 0, sizeof(event));
1323 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1324 	event.status = status;
1325 	memcpy(&event.local_addr, &ep->com.local_addr,
1326 	       sizeof(ep->com.local_addr));
1327 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1328 	       sizeof(ep->com.remote_addr));
1329 
1330 	if ((status == 0) || (status == -ECONNREFUSED)) {
1331 		if (!ep->tried_with_mpa_v1) {
1332 			/* this means MPA_v2 is used */
1333 			event.ord = ep->ird;
1334 			event.ird = ep->ord;
1335 			event.private_data_len = ep->plen -
1336 				sizeof(struct mpa_v2_conn_params);
1337 			event.private_data = ep->mpa_pkt +
1338 				sizeof(struct mpa_message) +
1339 				sizeof(struct mpa_v2_conn_params);
1340 		} else {
1341 			/* this means MPA_v1 is used */
1342 			event.ord = cur_max_read_depth(ep->com.dev);
1343 			event.ird = cur_max_read_depth(ep->com.dev);
1344 			event.private_data_len = ep->plen;
1345 			event.private_data = ep->mpa_pkt +
1346 				sizeof(struct mpa_message);
1347 		}
1348 	}
1349 
1350 	pr_debug("ep %p tid %u status %d\n", ep,
1351 		 ep->hwtid, status);
1352 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1353 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1354 
1355 	if (status < 0)
1356 		deref_cm_id(&ep->com);
1357 }
1358 
1359 static int connect_request_upcall(struct c4iw_ep *ep)
1360 {
1361 	struct iw_cm_event event;
1362 	int ret;
1363 
1364 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1365 	memset(&event, 0, sizeof(event));
1366 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1367 	memcpy(&event.local_addr, &ep->com.local_addr,
1368 	       sizeof(ep->com.local_addr));
1369 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1370 	       sizeof(ep->com.remote_addr));
1371 	event.provider_data = ep;
1372 	if (!ep->tried_with_mpa_v1) {
1373 		/* this means MPA_v2 is used */
1374 		event.ord = ep->ord;
1375 		event.ird = ep->ird;
1376 		event.private_data_len = ep->plen -
1377 			sizeof(struct mpa_v2_conn_params);
1378 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1379 			sizeof(struct mpa_v2_conn_params);
1380 	} else {
1381 		/* this means MPA_v1 is used. Send max supported */
1382 		event.ord = cur_max_read_depth(ep->com.dev);
1383 		event.ird = cur_max_read_depth(ep->com.dev);
1384 		event.private_data_len = ep->plen;
1385 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1386 	}
1387 	c4iw_get_ep(&ep->com);
1388 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1389 						      &event);
1390 	if (ret)
1391 		c4iw_put_ep(&ep->com);
1392 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1393 	c4iw_put_ep(&ep->parent_ep->com);
1394 	return ret;
1395 }
1396 
1397 static void established_upcall(struct c4iw_ep *ep)
1398 {
1399 	struct iw_cm_event event;
1400 
1401 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1402 	memset(&event, 0, sizeof(event));
1403 	event.event = IW_CM_EVENT_ESTABLISHED;
1404 	event.ird = ep->ord;
1405 	event.ord = ep->ird;
1406 	if (ep->com.cm_id) {
1407 		pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1408 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1409 		set_bit(ESTAB_UPCALL, &ep->com.history);
1410 	}
1411 }
1412 
1413 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1414 {
1415 	struct sk_buff *skb;
1416 	u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1417 	u32 credit_dack;
1418 
1419 	pr_debug("ep %p tid %u credits %u\n",
1420 		 ep, ep->hwtid, credits);
1421 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1422 	if (!skb) {
1423 		pr_err("update_rx_credits - cannot alloc skb!\n");
1424 		return 0;
1425 	}
1426 
1427 	/*
1428 	 * If we couldn't specify the entire rcv window at connection setup
1429 	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1430 	 * then add the overage in to the credits returned.
1431 	 */
1432 	if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1433 		credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1434 
1435 	credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1436 		      RX_DACK_MODE_V(dack_mode);
1437 
1438 	cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1439 			    credit_dack);
1440 
1441 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1442 	return credits;
1443 }
1444 
1445 #define RELAXED_IRD_NEGOTIATION 1
1446 
1447 /*
1448  * process_mpa_reply - process streaming mode MPA reply
1449  *
1450  * Returns:
1451  *
1452  * 0 upon success indicating a connect request was delivered to the ULP
1453  * or the mpa request is incomplete but valid so far.
1454  *
1455  * 1 if a failure requires the caller to close the connection.
1456  *
1457  * 2 if a failure requires the caller to abort the connection.
1458  */
1459 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1460 {
1461 	struct mpa_message *mpa;
1462 	struct mpa_v2_conn_params *mpa_v2_params;
1463 	u16 plen;
1464 	u16 resp_ird, resp_ord;
1465 	u8 rtr_mismatch = 0, insuff_ird = 0;
1466 	struct c4iw_qp_attributes attrs;
1467 	enum c4iw_qp_attr_mask mask;
1468 	int err;
1469 	int disconnect = 0;
1470 
1471 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1472 
1473 	/*
1474 	 * If we get more than the supported amount of private data
1475 	 * then we must fail this connection.
1476 	 */
1477 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1478 		err = -EINVAL;
1479 		goto err_stop_timer;
1480 	}
1481 
1482 	/*
1483 	 * copy the new data into our accumulation buffer.
1484 	 */
1485 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1486 				  skb->len);
1487 	ep->mpa_pkt_len += skb->len;
1488 
1489 	/*
1490 	 * if we don't even have the mpa message, then bail.
1491 	 */
1492 	if (ep->mpa_pkt_len < sizeof(*mpa))
1493 		return 0;
1494 	mpa = (struct mpa_message *) ep->mpa_pkt;
1495 
1496 	/* Validate MPA header. */
1497 	if (mpa->revision > mpa_rev) {
1498 		pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1499 		       __func__, mpa_rev, mpa->revision);
1500 		err = -EPROTO;
1501 		goto err_stop_timer;
1502 	}
1503 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1504 		err = -EPROTO;
1505 		goto err_stop_timer;
1506 	}
1507 
1508 	plen = ntohs(mpa->private_data_size);
1509 
1510 	/*
1511 	 * Fail if there's too much private data.
1512 	 */
1513 	if (plen > MPA_MAX_PRIVATE_DATA) {
1514 		err = -EPROTO;
1515 		goto err_stop_timer;
1516 	}
1517 
1518 	/*
1519 	 * If plen does not account for pkt size
1520 	 */
1521 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1522 		err = -EPROTO;
1523 		goto err_stop_timer;
1524 	}
1525 
1526 	ep->plen = (u8) plen;
1527 
1528 	/*
1529 	 * If we don't have all the pdata yet, then bail.
1530 	 * We'll continue process when more data arrives.
1531 	 */
1532 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1533 		return 0;
1534 
1535 	if (mpa->flags & MPA_REJECT) {
1536 		err = -ECONNREFUSED;
1537 		goto err_stop_timer;
1538 	}
1539 
1540 	/*
1541 	 * Stop mpa timer.  If it expired, then
1542 	 * we ignore the MPA reply.  process_timeout()
1543 	 * will abort the connection.
1544 	 */
1545 	if (stop_ep_timer(ep))
1546 		return 0;
1547 
1548 	/*
1549 	 * If we get here we have accumulated the entire mpa
1550 	 * start reply message including private data. And
1551 	 * the MPA header is valid.
1552 	 */
1553 	__state_set(&ep->com, FPDU_MODE);
1554 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1555 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1556 	ep->mpa_attr.version = mpa->revision;
1557 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1558 
1559 	if (mpa->revision == 2) {
1560 		ep->mpa_attr.enhanced_rdma_conn =
1561 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1562 		if (ep->mpa_attr.enhanced_rdma_conn) {
1563 			mpa_v2_params = (struct mpa_v2_conn_params *)
1564 				(ep->mpa_pkt + sizeof(*mpa));
1565 			resp_ird = ntohs(mpa_v2_params->ird) &
1566 				MPA_V2_IRD_ORD_MASK;
1567 			resp_ord = ntohs(mpa_v2_params->ord) &
1568 				MPA_V2_IRD_ORD_MASK;
1569 			pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1570 				 resp_ird, resp_ord, ep->ird, ep->ord);
1571 
1572 			/*
1573 			 * This is a double-check. Ideally, below checks are
1574 			 * not required since ird/ord stuff has been taken
1575 			 * care of in c4iw_accept_cr
1576 			 */
1577 			if (ep->ird < resp_ord) {
1578 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1579 				    ep->com.dev->rdev.lldi.max_ordird_qp)
1580 					ep->ird = resp_ord;
1581 				else
1582 					insuff_ird = 1;
1583 			} else if (ep->ird > resp_ord) {
1584 				ep->ird = resp_ord;
1585 			}
1586 			if (ep->ord > resp_ird) {
1587 				if (RELAXED_IRD_NEGOTIATION)
1588 					ep->ord = resp_ird;
1589 				else
1590 					insuff_ird = 1;
1591 			}
1592 			if (insuff_ird) {
1593 				err = -ENOMEM;
1594 				ep->ird = resp_ord;
1595 				ep->ord = resp_ird;
1596 			}
1597 
1598 			if (ntohs(mpa_v2_params->ird) &
1599 					MPA_V2_PEER2PEER_MODEL) {
1600 				if (ntohs(mpa_v2_params->ord) &
1601 						MPA_V2_RDMA_WRITE_RTR)
1602 					ep->mpa_attr.p2p_type =
1603 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1604 				else if (ntohs(mpa_v2_params->ord) &
1605 						MPA_V2_RDMA_READ_RTR)
1606 					ep->mpa_attr.p2p_type =
1607 						FW_RI_INIT_P2PTYPE_READ_REQ;
1608 			}
1609 		}
1610 	} else if (mpa->revision == 1)
1611 		if (peer2peer)
1612 			ep->mpa_attr.p2p_type = p2p_type;
1613 
1614 	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1615 		 ep->mpa_attr.crc_enabled,
1616 		 ep->mpa_attr.recv_marker_enabled,
1617 		 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1618 		 ep->mpa_attr.p2p_type, p2p_type);
1619 
1620 	/*
1621 	 * If responder's RTR does not match with that of initiator, assign
1622 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1623 	 * generated when moving QP to RTS state.
1624 	 * A TERM message will be sent after QP has moved to RTS state
1625 	 */
1626 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1627 			(ep->mpa_attr.p2p_type != p2p_type)) {
1628 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1629 		rtr_mismatch = 1;
1630 	}
1631 
1632 	attrs.mpa_attr = ep->mpa_attr;
1633 	attrs.max_ird = ep->ird;
1634 	attrs.max_ord = ep->ord;
1635 	attrs.llp_stream_handle = ep;
1636 	attrs.next_state = C4IW_QP_STATE_RTS;
1637 
1638 	mask = C4IW_QP_ATTR_NEXT_STATE |
1639 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1640 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1641 
1642 	/* bind QP and TID with INIT_WR */
1643 	err = c4iw_modify_qp(ep->com.qp->rhp,
1644 			     ep->com.qp, mask, &attrs, 1);
1645 	if (err)
1646 		goto err;
1647 
1648 	/*
1649 	 * If responder's RTR requirement did not match with what initiator
1650 	 * supports, generate TERM message
1651 	 */
1652 	if (rtr_mismatch) {
1653 		pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1654 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1655 		attrs.ecode = MPA_NOMATCH_RTR;
1656 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1657 		attrs.send_term = 1;
1658 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1659 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1660 		err = -ENOMEM;
1661 		disconnect = 1;
1662 		goto out;
1663 	}
1664 
1665 	/*
1666 	 * Generate TERM if initiator IRD is not sufficient for responder
1667 	 * provided ORD. Currently, we do the same behaviour even when
1668 	 * responder provided IRD is also not sufficient as regards to
1669 	 * initiator ORD.
1670 	 */
1671 	if (insuff_ird) {
1672 		pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1673 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1674 		attrs.ecode = MPA_INSUFF_IRD;
1675 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1676 		attrs.send_term = 1;
1677 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1678 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1679 		err = -ENOMEM;
1680 		disconnect = 1;
1681 		goto out;
1682 	}
1683 	goto out;
1684 err_stop_timer:
1685 	stop_ep_timer(ep);
1686 err:
1687 	disconnect = 2;
1688 out:
1689 	connect_reply_upcall(ep, err);
1690 	return disconnect;
1691 }
1692 
1693 /*
1694  * process_mpa_request - process streaming mode MPA request
1695  *
1696  * Returns:
1697  *
1698  * 0 upon success indicating a connect request was delivered to the ULP
1699  * or the mpa request is incomplete but valid so far.
1700  *
1701  * 1 if a failure requires the caller to close the connection.
1702  *
1703  * 2 if a failure requires the caller to abort the connection.
1704  */
1705 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1706 {
1707 	struct mpa_message *mpa;
1708 	struct mpa_v2_conn_params *mpa_v2_params;
1709 	u16 plen;
1710 
1711 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1712 
1713 	/*
1714 	 * If we get more than the supported amount of private data
1715 	 * then we must fail this connection.
1716 	 */
1717 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1718 		goto err_stop_timer;
1719 
1720 	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1721 
1722 	/*
1723 	 * Copy the new data into our accumulation buffer.
1724 	 */
1725 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1726 				  skb->len);
1727 	ep->mpa_pkt_len += skb->len;
1728 
1729 	/*
1730 	 * If we don't even have the mpa message, then bail.
1731 	 * We'll continue process when more data arrives.
1732 	 */
1733 	if (ep->mpa_pkt_len < sizeof(*mpa))
1734 		return 0;
1735 
1736 	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1737 	mpa = (struct mpa_message *) ep->mpa_pkt;
1738 
1739 	/*
1740 	 * Validate MPA Header.
1741 	 */
1742 	if (mpa->revision > mpa_rev) {
1743 		pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1744 		       __func__, mpa_rev, mpa->revision);
1745 		goto err_stop_timer;
1746 	}
1747 
1748 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1749 		goto err_stop_timer;
1750 
1751 	plen = ntohs(mpa->private_data_size);
1752 
1753 	/*
1754 	 * Fail if there's too much private data.
1755 	 */
1756 	if (plen > MPA_MAX_PRIVATE_DATA)
1757 		goto err_stop_timer;
1758 
1759 	/*
1760 	 * If plen does not account for pkt size
1761 	 */
1762 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1763 		goto err_stop_timer;
1764 	ep->plen = (u8) plen;
1765 
1766 	/*
1767 	 * If we don't have all the pdata yet, then bail.
1768 	 */
1769 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1770 		return 0;
1771 
1772 	/*
1773 	 * If we get here we have accumulated the entire mpa
1774 	 * start reply message including private data.
1775 	 */
1776 	ep->mpa_attr.initiator = 0;
1777 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1778 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1779 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1780 	ep->mpa_attr.version = mpa->revision;
1781 	if (mpa->revision == 1)
1782 		ep->tried_with_mpa_v1 = 1;
1783 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1784 
1785 	if (mpa->revision == 2) {
1786 		ep->mpa_attr.enhanced_rdma_conn =
1787 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1788 		if (ep->mpa_attr.enhanced_rdma_conn) {
1789 			mpa_v2_params = (struct mpa_v2_conn_params *)
1790 				(ep->mpa_pkt + sizeof(*mpa));
1791 			ep->ird = ntohs(mpa_v2_params->ird) &
1792 				MPA_V2_IRD_ORD_MASK;
1793 			ep->ird = min_t(u32, ep->ird,
1794 					cur_max_read_depth(ep->com.dev));
1795 			ep->ord = ntohs(mpa_v2_params->ord) &
1796 				MPA_V2_IRD_ORD_MASK;
1797 			ep->ord = min_t(u32, ep->ord,
1798 					cur_max_read_depth(ep->com.dev));
1799 			pr_debug("initiator ird %u ord %u\n",
1800 				 ep->ird, ep->ord);
1801 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1802 				if (peer2peer) {
1803 					if (ntohs(mpa_v2_params->ord) &
1804 							MPA_V2_RDMA_WRITE_RTR)
1805 						ep->mpa_attr.p2p_type =
1806 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1807 					else if (ntohs(mpa_v2_params->ord) &
1808 							MPA_V2_RDMA_READ_RTR)
1809 						ep->mpa_attr.p2p_type =
1810 						FW_RI_INIT_P2PTYPE_READ_REQ;
1811 				}
1812 		}
1813 	} else if (mpa->revision == 1)
1814 		if (peer2peer)
1815 			ep->mpa_attr.p2p_type = p2p_type;
1816 
1817 	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1818 		 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1819 		 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1820 		 ep->mpa_attr.p2p_type);
1821 
1822 	__state_set(&ep->com, MPA_REQ_RCVD);
1823 
1824 	/* drive upcall */
1825 	mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1826 	if (ep->parent_ep->com.state != DEAD) {
1827 		if (connect_request_upcall(ep))
1828 			goto err_unlock_parent;
1829 	} else {
1830 		goto err_unlock_parent;
1831 	}
1832 	mutex_unlock(&ep->parent_ep->com.mutex);
1833 	return 0;
1834 
1835 err_unlock_parent:
1836 	mutex_unlock(&ep->parent_ep->com.mutex);
1837 	goto err_out;
1838 err_stop_timer:
1839 	(void)stop_ep_timer(ep);
1840 err_out:
1841 	return 2;
1842 }
1843 
1844 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1845 {
1846 	struct c4iw_ep *ep;
1847 	struct cpl_rx_data *hdr = cplhdr(skb);
1848 	unsigned int dlen = ntohs(hdr->len);
1849 	unsigned int tid = GET_TID(hdr);
1850 	__u8 status = hdr->status;
1851 	int disconnect = 0;
1852 
1853 	ep = get_ep_from_tid(dev, tid);
1854 	if (!ep)
1855 		return 0;
1856 	pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1857 	skb_pull(skb, sizeof(*hdr));
1858 	skb_trim(skb, dlen);
1859 	mutex_lock(&ep->com.mutex);
1860 
1861 	switch (ep->com.state) {
1862 	case MPA_REQ_SENT:
1863 		update_rx_credits(ep, dlen);
1864 		ep->rcv_seq += dlen;
1865 		disconnect = process_mpa_reply(ep, skb);
1866 		break;
1867 	case MPA_REQ_WAIT:
1868 		update_rx_credits(ep, dlen);
1869 		ep->rcv_seq += dlen;
1870 		disconnect = process_mpa_request(ep, skb);
1871 		break;
1872 	case FPDU_MODE: {
1873 		struct c4iw_qp_attributes attrs;
1874 
1875 		update_rx_credits(ep, dlen);
1876 		if (status)
1877 			pr_err("%s Unexpected streaming data." \
1878 			       " qpid %u ep %p state %d tid %u status %d\n",
1879 			       __func__, ep->com.qp->wq.sq.qid, ep,
1880 			       ep->com.state, ep->hwtid, status);
1881 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1882 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1883 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1884 		disconnect = 1;
1885 		break;
1886 	}
1887 	default:
1888 		break;
1889 	}
1890 	mutex_unlock(&ep->com.mutex);
1891 	if (disconnect)
1892 		c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1893 	c4iw_put_ep(&ep->com);
1894 	return 0;
1895 }
1896 
1897 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1898 {
1899 	enum chip_type adapter_type;
1900 
1901 	adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1902 
1903 	/*
1904 	 * If this TCB had a srq buffer cached, then we must complete
1905 	 * it. For user mode, that means saving the srqidx in the
1906 	 * user/kernel status page for this qp.  For kernel mode, just
1907 	 * synthesize the CQE now.
1908 	 */
1909 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1910 		if (ep->com.qp->ibqp.uobject)
1911 			t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1912 		else
1913 			c4iw_flush_srqidx(ep->com.qp, srqidx);
1914 	}
1915 }
1916 
1917 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1918 {
1919 	u32 srqidx;
1920 	struct c4iw_ep *ep;
1921 	struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1922 	int release = 0;
1923 	unsigned int tid = GET_TID(rpl);
1924 
1925 	ep = get_ep_from_tid(dev, tid);
1926 	if (!ep) {
1927 		pr_warn("Abort rpl to freed endpoint\n");
1928 		return 0;
1929 	}
1930 
1931 	if (ep->com.qp && ep->com.qp->srq) {
1932 		srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1933 		complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
1934 	}
1935 
1936 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1937 	mutex_lock(&ep->com.mutex);
1938 	switch (ep->com.state) {
1939 	case ABORTING:
1940 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1941 		__state_set(&ep->com, DEAD);
1942 		release = 1;
1943 		break;
1944 	default:
1945 		pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1946 		break;
1947 	}
1948 	mutex_unlock(&ep->com.mutex);
1949 
1950 	if (release) {
1951 		close_complete_upcall(ep, -ECONNRESET);
1952 		release_ep_resources(ep);
1953 	}
1954 	c4iw_put_ep(&ep->com);
1955 	return 0;
1956 }
1957 
1958 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1959 {
1960 	struct sk_buff *skb;
1961 	struct fw_ofld_connection_wr *req;
1962 	unsigned int mtu_idx;
1963 	u32 wscale;
1964 	struct sockaddr_in *sin;
1965 	int win;
1966 
1967 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1968 	req = __skb_put_zero(skb, sizeof(*req));
1969 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1970 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1971 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1972 				     ep->com.dev->rdev.lldi.ports[0],
1973 				     ep->l2t));
1974 	sin = (struct sockaddr_in *)&ep->com.local_addr;
1975 	req->le.lport = sin->sin_port;
1976 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1977 	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1978 	req->le.pport = sin->sin_port;
1979 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1980 	req->tcb.t_state_to_astid =
1981 			htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1982 			FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1983 	req->tcb.cplrxdataack_cplpassacceptrpl =
1984 			htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1985 	req->tcb.tx_max = (__force __be32) jiffies;
1986 	req->tcb.rcv_adv = htons(1);
1987 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1988 		      enable_tcp_timestamps,
1989 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1990 	wscale = cxgb_compute_wscale(rcv_win);
1991 
1992 	/*
1993 	 * Specify the largest window that will fit in opt0. The
1994 	 * remainder will be specified in the rx_data_ack.
1995 	 */
1996 	win = ep->rcv_win >> 10;
1997 	if (win > RCV_BUFSIZ_M)
1998 		win = RCV_BUFSIZ_M;
1999 
2000 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2001 		(nocong ? NO_CONG_F : 0) |
2002 		KEEP_ALIVE_F |
2003 		DELACK_F |
2004 		WND_SCALE_V(wscale) |
2005 		MSS_IDX_V(mtu_idx) |
2006 		L2T_IDX_V(ep->l2t->idx) |
2007 		TX_CHAN_V(ep->tx_chan) |
2008 		SMAC_SEL_V(ep->smac_idx) |
2009 		DSCP_V(ep->tos >> 2) |
2010 		ULP_MODE_V(ULP_MODE_TCPDDP) |
2011 		RCV_BUFSIZ_V(win));
2012 	req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2013 		TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2014 		RX_CHANNEL_V(0) |
2015 		CCTRL_ECN_V(enable_ecn) |
2016 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2017 	if (enable_tcp_timestamps)
2018 		req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2019 	if (enable_tcp_sack)
2020 		req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2021 	if (wscale && enable_tcp_window_scaling)
2022 		req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2023 	req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2024 	req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2025 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2026 	set_bit(ACT_OFLD_CONN, &ep->com.history);
2027 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2028 }
2029 
2030 /*
2031  * Some of the error codes above implicitly indicate that there is no TID
2032  * allocated with the result of an ACT_OPEN.  We use this predicate to make
2033  * that explicit.
2034  */
2035 static inline int act_open_has_tid(int status)
2036 {
2037 	return (status != CPL_ERR_TCAM_PARITY &&
2038 		status != CPL_ERR_TCAM_MISS &&
2039 		status != CPL_ERR_TCAM_FULL &&
2040 		status != CPL_ERR_CONN_EXIST_SYNRECV &&
2041 		status != CPL_ERR_CONN_EXIST);
2042 }
2043 
2044 static char *neg_adv_str(unsigned int status)
2045 {
2046 	switch (status) {
2047 	case CPL_ERR_RTX_NEG_ADVICE:
2048 		return "Retransmit timeout";
2049 	case CPL_ERR_PERSIST_NEG_ADVICE:
2050 		return "Persist timeout";
2051 	case CPL_ERR_KEEPALV_NEG_ADVICE:
2052 		return "Keepalive timeout";
2053 	default:
2054 		return "Unknown";
2055 	}
2056 }
2057 
2058 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2059 {
2060 	ep->snd_win = snd_win;
2061 	ep->rcv_win = rcv_win;
2062 	pr_debug("snd_win %d rcv_win %d\n",
2063 		 ep->snd_win, ep->rcv_win);
2064 }
2065 
2066 #define ACT_OPEN_RETRY_COUNT 2
2067 
2068 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2069 		     struct dst_entry *dst, struct c4iw_dev *cdev,
2070 		     bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2071 {
2072 	struct neighbour *n;
2073 	int err, step;
2074 	struct net_device *pdev;
2075 
2076 	n = dst_neigh_lookup(dst, peer_ip);
2077 	if (!n)
2078 		return -ENODEV;
2079 
2080 	rcu_read_lock();
2081 	err = -ENOMEM;
2082 	if (n->dev->flags & IFF_LOOPBACK) {
2083 		if (iptype == 4)
2084 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2085 		else if (IS_ENABLED(CONFIG_IPV6))
2086 			for_each_netdev(&init_net, pdev) {
2087 				if (ipv6_chk_addr(&init_net,
2088 						  (struct in6_addr *)peer_ip,
2089 						  pdev, 1))
2090 					break;
2091 			}
2092 		else
2093 			pdev = NULL;
2094 
2095 		if (!pdev) {
2096 			err = -ENODEV;
2097 			goto out;
2098 		}
2099 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2100 					n, pdev, rt_tos2priority(tos));
2101 		if (!ep->l2t) {
2102 			dev_put(pdev);
2103 			goto out;
2104 		}
2105 		ep->mtu = pdev->mtu;
2106 		ep->tx_chan = cxgb4_port_chan(pdev);
2107 		ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2108 		step = cdev->rdev.lldi.ntxq /
2109 			cdev->rdev.lldi.nchan;
2110 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2111 		step = cdev->rdev.lldi.nrxq /
2112 			cdev->rdev.lldi.nchan;
2113 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2114 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2115 			cxgb4_port_idx(pdev) * step];
2116 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2117 		dev_put(pdev);
2118 	} else {
2119 		pdev = get_real_dev(n->dev);
2120 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2121 					n, pdev, rt_tos2priority(tos));
2122 		if (!ep->l2t)
2123 			goto out;
2124 		ep->mtu = dst_mtu(dst);
2125 		ep->tx_chan = cxgb4_port_chan(pdev);
2126 		ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2127 		step = cdev->rdev.lldi.ntxq /
2128 			cdev->rdev.lldi.nchan;
2129 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2130 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2131 		step = cdev->rdev.lldi.nrxq /
2132 			cdev->rdev.lldi.nchan;
2133 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2134 			cxgb4_port_idx(pdev) * step];
2135 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2136 
2137 		if (clear_mpa_v1) {
2138 			ep->retry_with_mpa_v1 = 0;
2139 			ep->tried_with_mpa_v1 = 0;
2140 		}
2141 	}
2142 	err = 0;
2143 out:
2144 	rcu_read_unlock();
2145 
2146 	neigh_release(n);
2147 
2148 	return err;
2149 }
2150 
2151 static int c4iw_reconnect(struct c4iw_ep *ep)
2152 {
2153 	int err = 0;
2154 	int size = 0;
2155 	struct sockaddr_in *laddr = (struct sockaddr_in *)
2156 				    &ep->com.cm_id->m_local_addr;
2157 	struct sockaddr_in *raddr = (struct sockaddr_in *)
2158 				    &ep->com.cm_id->m_remote_addr;
2159 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2160 				      &ep->com.cm_id->m_local_addr;
2161 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2162 				      &ep->com.cm_id->m_remote_addr;
2163 	int iptype;
2164 	__u8 *ra;
2165 
2166 	pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2167 	c4iw_init_wr_wait(ep->com.wr_waitp);
2168 
2169 	/* When MPA revision is different on nodes, the node with MPA_rev=2
2170 	 * tries to reconnect with MPA_rev 1 for the same EP through
2171 	 * c4iw_reconnect(), where the same EP is assigned with new tid for
2172 	 * further connection establishment. As we are using the same EP pointer
2173 	 * for reconnect, few skbs are used during the previous c4iw_connect(),
2174 	 * which leaves the EP with inadequate skbs for further
2175 	 * c4iw_reconnect(), Further causing a crash due to an empty
2176 	 * skb_list() during peer_abort(). Allocate skbs which is already used.
2177 	 */
2178 	size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2179 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2180 		err = -ENOMEM;
2181 		goto fail1;
2182 	}
2183 
2184 	/*
2185 	 * Allocate an active TID to initiate a TCP connection.
2186 	 */
2187 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2188 	if (ep->atid == -1) {
2189 		pr_err("%s - cannot alloc atid\n", __func__);
2190 		err = -ENOMEM;
2191 		goto fail2;
2192 	}
2193 	err = xa_insert_irq(&ep->com.dev->atids, ep->atid, ep, GFP_KERNEL);
2194 	if (err)
2195 		goto fail2a;
2196 
2197 	/* find a route */
2198 	if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2199 		ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2200 					  laddr->sin_addr.s_addr,
2201 					  raddr->sin_addr.s_addr,
2202 					  laddr->sin_port,
2203 					  raddr->sin_port, ep->com.cm_id->tos);
2204 		iptype = 4;
2205 		ra = (__u8 *)&raddr->sin_addr;
2206 	} else {
2207 		ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2208 					   get_real_dev,
2209 					   laddr6->sin6_addr.s6_addr,
2210 					   raddr6->sin6_addr.s6_addr,
2211 					   laddr6->sin6_port,
2212 					   raddr6->sin6_port,
2213 					   ep->com.cm_id->tos,
2214 					   raddr6->sin6_scope_id);
2215 		iptype = 6;
2216 		ra = (__u8 *)&raddr6->sin6_addr;
2217 	}
2218 	if (!ep->dst) {
2219 		pr_err("%s - cannot find route\n", __func__);
2220 		err = -EHOSTUNREACH;
2221 		goto fail3;
2222 	}
2223 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2224 			ep->com.dev->rdev.lldi.adapter_type,
2225 			ep->com.cm_id->tos);
2226 	if (err) {
2227 		pr_err("%s - cannot alloc l2e\n", __func__);
2228 		goto fail4;
2229 	}
2230 
2231 	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2232 		 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2233 		 ep->l2t->idx);
2234 
2235 	state_set(&ep->com, CONNECTING);
2236 	ep->tos = ep->com.cm_id->tos;
2237 
2238 	/* send connect request to rnic */
2239 	err = send_connect(ep);
2240 	if (!err)
2241 		goto out;
2242 
2243 	cxgb4_l2t_release(ep->l2t);
2244 fail4:
2245 	dst_release(ep->dst);
2246 fail3:
2247 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
2248 fail2a:
2249 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2250 fail2:
2251 	/*
2252 	 * remember to send notification to upper layer.
2253 	 * We are in here so the upper layer is not aware that this is
2254 	 * re-connect attempt and so, upper layer is still waiting for
2255 	 * response of 1st connect request.
2256 	 */
2257 	connect_reply_upcall(ep, -ECONNRESET);
2258 fail1:
2259 	c4iw_put_ep(&ep->com);
2260 out:
2261 	return err;
2262 }
2263 
2264 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2265 {
2266 	struct c4iw_ep *ep;
2267 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
2268 	unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2269 				      ntohl(rpl->atid_status)));
2270 	struct tid_info *t = dev->rdev.lldi.tids;
2271 	int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2272 	struct sockaddr_in *la;
2273 	struct sockaddr_in *ra;
2274 	struct sockaddr_in6 *la6;
2275 	struct sockaddr_in6 *ra6;
2276 	int ret = 0;
2277 
2278 	ep = lookup_atid(t, atid);
2279 	la = (struct sockaddr_in *)&ep->com.local_addr;
2280 	ra = (struct sockaddr_in *)&ep->com.remote_addr;
2281 	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2282 	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2283 
2284 	pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2285 		 status, status2errno(status));
2286 
2287 	if (cxgb_is_neg_adv(status)) {
2288 		pr_debug("Connection problems for atid %u status %u (%s)\n",
2289 			 atid, status, neg_adv_str(status));
2290 		ep->stats.connect_neg_adv++;
2291 		mutex_lock(&dev->rdev.stats.lock);
2292 		dev->rdev.stats.neg_adv++;
2293 		mutex_unlock(&dev->rdev.stats.lock);
2294 		return 0;
2295 	}
2296 
2297 	set_bit(ACT_OPEN_RPL, &ep->com.history);
2298 
2299 	/*
2300 	 * Log interesting failures.
2301 	 */
2302 	switch (status) {
2303 	case CPL_ERR_CONN_RESET:
2304 	case CPL_ERR_CONN_TIMEDOUT:
2305 		break;
2306 	case CPL_ERR_TCAM_FULL:
2307 		mutex_lock(&dev->rdev.stats.lock);
2308 		dev->rdev.stats.tcam_full++;
2309 		mutex_unlock(&dev->rdev.stats.lock);
2310 		if (ep->com.local_addr.ss_family == AF_INET &&
2311 		    dev->rdev.lldi.enable_fw_ofld_conn) {
2312 			ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2313 						   ntohl(rpl->atid_status))));
2314 			if (ret)
2315 				goto fail;
2316 			return 0;
2317 		}
2318 		break;
2319 	case CPL_ERR_CONN_EXIST:
2320 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2321 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
2322 			if (ep->com.remote_addr.ss_family == AF_INET6) {
2323 				struct sockaddr_in6 *sin6 =
2324 						(struct sockaddr_in6 *)
2325 						&ep->com.local_addr;
2326 				cxgb4_clip_release(
2327 						ep->com.dev->rdev.lldi.ports[0],
2328 						(const u32 *)
2329 						&sin6->sin6_addr.s6_addr, 1);
2330 			}
2331 			xa_erase_irq(&ep->com.dev->atids, atid);
2332 			cxgb4_free_atid(t, atid);
2333 			dst_release(ep->dst);
2334 			cxgb4_l2t_release(ep->l2t);
2335 			c4iw_reconnect(ep);
2336 			return 0;
2337 		}
2338 		break;
2339 	default:
2340 		if (ep->com.local_addr.ss_family == AF_INET) {
2341 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2342 				atid, status, status2errno(status),
2343 				&la->sin_addr.s_addr, ntohs(la->sin_port),
2344 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
2345 		} else {
2346 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2347 				atid, status, status2errno(status),
2348 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2349 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2350 		}
2351 		break;
2352 	}
2353 
2354 fail:
2355 	connect_reply_upcall(ep, status2errno(status));
2356 	state_set(&ep->com, DEAD);
2357 
2358 	if (ep->com.remote_addr.ss_family == AF_INET6) {
2359 		struct sockaddr_in6 *sin6 =
2360 			(struct sockaddr_in6 *)&ep->com.local_addr;
2361 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2362 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2363 	}
2364 	if (status && act_open_has_tid(status))
2365 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2366 				 ep->com.local_addr.ss_family);
2367 
2368 	xa_erase_irq(&ep->com.dev->atids, atid);
2369 	cxgb4_free_atid(t, atid);
2370 	dst_release(ep->dst);
2371 	cxgb4_l2t_release(ep->l2t);
2372 	c4iw_put_ep(&ep->com);
2373 
2374 	return 0;
2375 }
2376 
2377 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2378 {
2379 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2380 	unsigned int stid = GET_TID(rpl);
2381 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2382 
2383 	if (!ep) {
2384 		pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2385 		goto out;
2386 	}
2387 	pr_debug("ep %p status %d error %d\n", ep,
2388 		 rpl->status, status2errno(rpl->status));
2389 	c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2390 	c4iw_put_ep(&ep->com);
2391 out:
2392 	return 0;
2393 }
2394 
2395 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2396 {
2397 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2398 	unsigned int stid = GET_TID(rpl);
2399 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2400 
2401 	if (!ep) {
2402 		pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2403 		goto out;
2404 	}
2405 	pr_debug("ep %p\n", ep);
2406 	c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2407 	c4iw_put_ep(&ep->com);
2408 out:
2409 	return 0;
2410 }
2411 
2412 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2413 		     struct cpl_pass_accept_req *req)
2414 {
2415 	struct cpl_pass_accept_rpl *rpl;
2416 	unsigned int mtu_idx;
2417 	u64 opt0;
2418 	u32 opt2;
2419 	u32 wscale;
2420 	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2421 	int win;
2422 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2423 
2424 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2425 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2426 		      enable_tcp_timestamps && req->tcpopt.tstamp,
2427 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2428 	wscale = cxgb_compute_wscale(rcv_win);
2429 
2430 	/*
2431 	 * Specify the largest window that will fit in opt0. The
2432 	 * remainder will be specified in the rx_data_ack.
2433 	 */
2434 	win = ep->rcv_win >> 10;
2435 	if (win > RCV_BUFSIZ_M)
2436 		win = RCV_BUFSIZ_M;
2437 	opt0 = (nocong ? NO_CONG_F : 0) |
2438 	       KEEP_ALIVE_F |
2439 	       DELACK_F |
2440 	       WND_SCALE_V(wscale) |
2441 	       MSS_IDX_V(mtu_idx) |
2442 	       L2T_IDX_V(ep->l2t->idx) |
2443 	       TX_CHAN_V(ep->tx_chan) |
2444 	       SMAC_SEL_V(ep->smac_idx) |
2445 	       DSCP_V(ep->tos >> 2) |
2446 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
2447 	       RCV_BUFSIZ_V(win);
2448 	opt2 = RX_CHANNEL_V(0) |
2449 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2450 
2451 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2452 		opt2 |= TSTAMPS_EN_F;
2453 	if (enable_tcp_sack && req->tcpopt.sack)
2454 		opt2 |= SACK_EN_F;
2455 	if (wscale && enable_tcp_window_scaling)
2456 		opt2 |= WND_SCALE_EN_F;
2457 	if (enable_ecn) {
2458 		const struct tcphdr *tcph;
2459 		u32 hlen = ntohl(req->hdr_len);
2460 
2461 		if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2462 			tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2463 				IP_HDR_LEN_G(hlen);
2464 		else
2465 			tcph = (const void *)(req + 1) +
2466 				T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2467 		if (tcph->ece && tcph->cwr)
2468 			opt2 |= CCTRL_ECN_V(1);
2469 	}
2470 
2471 	if (!is_t4(adapter_type)) {
2472 		u32 isn = (get_random_u32() & ~7UL) - 1;
2473 
2474 		skb = get_skb(skb, roundup(sizeof(*rpl5), 16), GFP_KERNEL);
2475 		rpl5 = __skb_put_zero(skb, roundup(sizeof(*rpl5), 16));
2476 		rpl = (void *)rpl5;
2477 		INIT_TP_WR_CPL(rpl5, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2478 		opt2 |= T5_OPT_2_VALID_F;
2479 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2480 		opt2 |= T5_ISS_F;
2481 		if (peer2peer)
2482 			isn += 4;
2483 		rpl5->iss = cpu_to_be32(isn);
2484 		pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2485 	} else {
2486 		skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2487 		rpl = __skb_put_zero(skb, sizeof(*rpl));
2488 		INIT_TP_WR_CPL(rpl, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2489 	}
2490 
2491 	rpl->opt0 = cpu_to_be64(opt0);
2492 	rpl->opt2 = cpu_to_be32(opt2);
2493 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2494 	t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2495 
2496 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2497 }
2498 
2499 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2500 {
2501 	pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2502 	skb_trim(skb, sizeof(struct cpl_tid_release));
2503 	release_tid(&dev->rdev, hwtid, skb);
2504 	return;
2505 }
2506 
2507 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2508 {
2509 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2510 	struct cpl_pass_accept_req *req = cplhdr(skb);
2511 	unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2512 	struct tid_info *t = dev->rdev.lldi.tids;
2513 	unsigned int hwtid = GET_TID(req);
2514 	struct dst_entry *dst;
2515 	__u8 local_ip[16], peer_ip[16];
2516 	__be16 local_port, peer_port;
2517 	struct sockaddr_in6 *sin6;
2518 	int err;
2519 	u16 peer_mss = ntohs(req->tcpopt.mss);
2520 	int iptype;
2521 	unsigned short hdrs;
2522 	u8 tos;
2523 
2524 	parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2525 	if (!parent_ep) {
2526 		pr_err("%s connect request on invalid stid %d\n",
2527 		       __func__, stid);
2528 		goto reject;
2529 	}
2530 
2531 	if (state_read(&parent_ep->com) != LISTEN) {
2532 		pr_err("%s - listening ep not in LISTEN\n", __func__);
2533 		goto reject;
2534 	}
2535 
2536 	if (parent_ep->com.cm_id->tos_set)
2537 		tos = parent_ep->com.cm_id->tos;
2538 	else
2539 		tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2540 
2541 	cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2542 			&iptype, local_ip, peer_ip, &local_port, &peer_port);
2543 
2544 	/* Find output route */
2545 	if (iptype == 4)  {
2546 		pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2547 			 , parent_ep, hwtid,
2548 			 local_ip, peer_ip, ntohs(local_port),
2549 			 ntohs(peer_port), peer_mss);
2550 		dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2551 				      *(__be32 *)local_ip, *(__be32 *)peer_ip,
2552 				      local_port, peer_port, tos);
2553 	} else {
2554 		pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2555 			 , parent_ep, hwtid,
2556 			 local_ip, peer_ip, ntohs(local_port),
2557 			 ntohs(peer_port), peer_mss);
2558 		dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2559 				local_ip, peer_ip, local_port, peer_port,
2560 				tos,
2561 				((struct sockaddr_in6 *)
2562 				 &parent_ep->com.local_addr)->sin6_scope_id);
2563 	}
2564 	if (!dst) {
2565 		pr_err("%s - failed to find dst entry!\n", __func__);
2566 		goto reject;
2567 	}
2568 
2569 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2570 	if (!child_ep) {
2571 		pr_err("%s - failed to allocate ep entry!\n", __func__);
2572 		dst_release(dst);
2573 		goto reject;
2574 	}
2575 
2576 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2577 			parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2578 	if (err) {
2579 		pr_err("%s - failed to allocate l2t entry!\n", __func__);
2580 		dst_release(dst);
2581 		kfree(child_ep);
2582 		goto reject;
2583 	}
2584 
2585 	hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2586 	       sizeof(struct tcphdr) +
2587 	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2588 	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2589 		child_ep->mtu = peer_mss + hdrs;
2590 
2591 	skb_queue_head_init(&child_ep->com.ep_skb_list);
2592 	if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2593 		goto fail;
2594 
2595 	state_set(&child_ep->com, CONNECTING);
2596 	child_ep->com.dev = dev;
2597 	child_ep->com.cm_id = NULL;
2598 
2599 	if (iptype == 4) {
2600 		struct sockaddr_in *sin = (struct sockaddr_in *)
2601 			&child_ep->com.local_addr;
2602 
2603 		sin->sin_family = AF_INET;
2604 		sin->sin_port = local_port;
2605 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2606 
2607 		sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2608 		sin->sin_family = AF_INET;
2609 		sin->sin_port = ((struct sockaddr_in *)
2610 				 &parent_ep->com.local_addr)->sin_port;
2611 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2612 
2613 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2614 		sin->sin_family = AF_INET;
2615 		sin->sin_port = peer_port;
2616 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2617 	} else {
2618 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2619 		sin6->sin6_family = PF_INET6;
2620 		sin6->sin6_port = local_port;
2621 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2622 
2623 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2624 		sin6->sin6_family = PF_INET6;
2625 		sin6->sin6_port = ((struct sockaddr_in6 *)
2626 				   &parent_ep->com.local_addr)->sin6_port;
2627 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2628 
2629 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2630 		sin6->sin6_family = PF_INET6;
2631 		sin6->sin6_port = peer_port;
2632 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2633 	}
2634 
2635 	c4iw_get_ep(&parent_ep->com);
2636 	child_ep->parent_ep = parent_ep;
2637 	child_ep->tos = tos;
2638 	child_ep->dst = dst;
2639 	child_ep->hwtid = hwtid;
2640 
2641 	pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2642 		 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2643 
2644 	timer_setup(&child_ep->timer, ep_timeout, 0);
2645 	cxgb4_insert_tid(t, child_ep, hwtid,
2646 			 child_ep->com.local_addr.ss_family);
2647 	insert_ep_tid(child_ep);
2648 	if (accept_cr(child_ep, skb, req)) {
2649 		c4iw_put_ep(&parent_ep->com);
2650 		release_ep_resources(child_ep);
2651 	} else {
2652 		set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2653 	}
2654 	if (iptype == 6) {
2655 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2656 		cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2657 			       (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2658 	}
2659 	goto out;
2660 fail:
2661 	c4iw_put_ep(&child_ep->com);
2662 reject:
2663 	reject_cr(dev, hwtid, skb);
2664 out:
2665 	if (parent_ep)
2666 		c4iw_put_ep(&parent_ep->com);
2667 	return 0;
2668 }
2669 
2670 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2671 {
2672 	struct c4iw_ep *ep;
2673 	struct cpl_pass_establish *req = cplhdr(skb);
2674 	unsigned int tid = GET_TID(req);
2675 	int ret;
2676 	u16 tcp_opt = ntohs(req->tcp_opt);
2677 
2678 	ep = get_ep_from_tid(dev, tid);
2679 	if (!ep)
2680 		return 0;
2681 
2682 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2683 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2684 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2685 	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2686 
2687 	pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2688 
2689 	set_emss(ep, tcp_opt);
2690 
2691 	dst_confirm(ep->dst);
2692 	mutex_lock(&ep->com.mutex);
2693 	ep->com.state = MPA_REQ_WAIT;
2694 	start_ep_timer(ep);
2695 	set_bit(PASS_ESTAB, &ep->com.history);
2696 	ret = send_flowc(ep);
2697 	mutex_unlock(&ep->com.mutex);
2698 	if (ret)
2699 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2700 	c4iw_put_ep(&ep->com);
2701 
2702 	return 0;
2703 }
2704 
2705 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2706 {
2707 	struct cpl_peer_close *hdr = cplhdr(skb);
2708 	struct c4iw_ep *ep;
2709 	struct c4iw_qp_attributes attrs;
2710 	int disconnect = 1;
2711 	int release = 0;
2712 	unsigned int tid = GET_TID(hdr);
2713 	int ret;
2714 
2715 	ep = get_ep_from_tid(dev, tid);
2716 	if (!ep)
2717 		return 0;
2718 
2719 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2720 	dst_confirm(ep->dst);
2721 
2722 	set_bit(PEER_CLOSE, &ep->com.history);
2723 	mutex_lock(&ep->com.mutex);
2724 	switch (ep->com.state) {
2725 	case MPA_REQ_WAIT:
2726 		__state_set(&ep->com, CLOSING);
2727 		break;
2728 	case MPA_REQ_SENT:
2729 		__state_set(&ep->com, CLOSING);
2730 		connect_reply_upcall(ep, -ECONNRESET);
2731 		break;
2732 	case MPA_REQ_RCVD:
2733 
2734 		/*
2735 		 * We're gonna mark this puppy DEAD, but keep
2736 		 * the reference on it until the ULP accepts or
2737 		 * rejects the CR. Also wake up anyone waiting
2738 		 * in rdma connection migration (see c4iw_accept_cr()).
2739 		 */
2740 		__state_set(&ep->com, CLOSING);
2741 		pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2742 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2743 		break;
2744 	case MPA_REP_SENT:
2745 		__state_set(&ep->com, CLOSING);
2746 		pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2747 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2748 		break;
2749 	case FPDU_MODE:
2750 		start_ep_timer(ep);
2751 		__state_set(&ep->com, CLOSING);
2752 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2753 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2754 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2755 		if (ret != -ECONNRESET) {
2756 			peer_close_upcall(ep);
2757 			disconnect = 1;
2758 		}
2759 		break;
2760 	case ABORTING:
2761 		disconnect = 0;
2762 		break;
2763 	case CLOSING:
2764 		__state_set(&ep->com, MORIBUND);
2765 		disconnect = 0;
2766 		break;
2767 	case MORIBUND:
2768 		(void)stop_ep_timer(ep);
2769 		if (ep->com.cm_id && ep->com.qp) {
2770 			attrs.next_state = C4IW_QP_STATE_IDLE;
2771 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2772 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2773 		}
2774 		close_complete_upcall(ep, 0);
2775 		__state_set(&ep->com, DEAD);
2776 		release = 1;
2777 		disconnect = 0;
2778 		break;
2779 	case DEAD:
2780 		disconnect = 0;
2781 		break;
2782 	default:
2783 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2784 	}
2785 	mutex_unlock(&ep->com.mutex);
2786 	if (disconnect)
2787 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2788 	if (release)
2789 		release_ep_resources(ep);
2790 	c4iw_put_ep(&ep->com);
2791 	return 0;
2792 }
2793 
2794 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2795 {
2796 	complete_cached_srq_buffers(ep, ep->srqe_idx);
2797 	if (ep->com.cm_id && ep->com.qp) {
2798 		struct c4iw_qp_attributes attrs;
2799 
2800 		attrs.next_state = C4IW_QP_STATE_ERROR;
2801 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2802 			       C4IW_QP_ATTR_NEXT_STATE,	&attrs, 1);
2803 	}
2804 	peer_abort_upcall(ep);
2805 	release_ep_resources(ep);
2806 	c4iw_put_ep(&ep->com);
2807 }
2808 
2809 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2810 {
2811 	struct cpl_abort_req_rss6 *req = cplhdr(skb);
2812 	struct c4iw_ep *ep;
2813 	struct sk_buff *rpl_skb;
2814 	struct c4iw_qp_attributes attrs;
2815 	int ret;
2816 	int release = 0;
2817 	unsigned int tid = GET_TID(req);
2818 	u8 status;
2819 	u32 srqidx;
2820 
2821 	u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2822 
2823 	ep = get_ep_from_tid(dev, tid);
2824 	if (!ep)
2825 		return 0;
2826 
2827 	status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2828 
2829 	if (cxgb_is_neg_adv(status)) {
2830 		pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2831 			 ep->hwtid, status, neg_adv_str(status));
2832 		ep->stats.abort_neg_adv++;
2833 		mutex_lock(&dev->rdev.stats.lock);
2834 		dev->rdev.stats.neg_adv++;
2835 		mutex_unlock(&dev->rdev.stats.lock);
2836 		goto deref_ep;
2837 	}
2838 
2839 	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2840 		 ep->com.state);
2841 	set_bit(PEER_ABORT, &ep->com.history);
2842 
2843 	/*
2844 	 * Wake up any threads in rdma_init() or rdma_fini().
2845 	 * However, this is not needed if com state is just
2846 	 * MPA_REQ_SENT
2847 	 */
2848 	if (ep->com.state != MPA_REQ_SENT)
2849 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2850 
2851 	mutex_lock(&ep->com.mutex);
2852 	switch (ep->com.state) {
2853 	case CONNECTING:
2854 		c4iw_put_ep(&ep->parent_ep->com);
2855 		break;
2856 	case MPA_REQ_WAIT:
2857 		(void)stop_ep_timer(ep);
2858 		break;
2859 	case MPA_REQ_SENT:
2860 		(void)stop_ep_timer(ep);
2861 		if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2862 		    (mpa_rev == 2 && ep->tried_with_mpa_v1))
2863 			connect_reply_upcall(ep, -ECONNRESET);
2864 		else {
2865 			/*
2866 			 * we just don't send notification upwards because we
2867 			 * want to retry with mpa_v1 without upper layers even
2868 			 * knowing it.
2869 			 *
2870 			 * do some housekeeping so as to re-initiate the
2871 			 * connection
2872 			 */
2873 			pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2874 				__func__, mpa_rev);
2875 			ep->retry_with_mpa_v1 = 1;
2876 		}
2877 		break;
2878 	case MPA_REP_SENT:
2879 		break;
2880 	case MPA_REQ_RCVD:
2881 		break;
2882 	case MORIBUND:
2883 	case CLOSING:
2884 		stop_ep_timer(ep);
2885 		fallthrough;
2886 	case FPDU_MODE:
2887 		if (ep->com.qp && ep->com.qp->srq) {
2888 			srqidx = ABORT_RSS_SRQIDX_G(
2889 					be32_to_cpu(req->srqidx_status));
2890 			if (srqidx) {
2891 				complete_cached_srq_buffers(ep, srqidx);
2892 			} else {
2893 				/* Hold ep ref until finish_peer_abort() */
2894 				c4iw_get_ep(&ep->com);
2895 				__state_set(&ep->com, ABORTING);
2896 				set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
2897 				read_tcb(ep);
2898 				break;
2899 
2900 			}
2901 		}
2902 
2903 		if (ep->com.cm_id && ep->com.qp) {
2904 			attrs.next_state = C4IW_QP_STATE_ERROR;
2905 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2906 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2907 				     &attrs, 1);
2908 			if (ret)
2909 				pr_err("%s - qp <- error failed!\n", __func__);
2910 		}
2911 		peer_abort_upcall(ep);
2912 		break;
2913 	case ABORTING:
2914 		break;
2915 	case DEAD:
2916 		pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2917 		mutex_unlock(&ep->com.mutex);
2918 		goto deref_ep;
2919 	default:
2920 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2921 		break;
2922 	}
2923 	dst_confirm(ep->dst);
2924 	if (ep->com.state != ABORTING) {
2925 		__state_set(&ep->com, DEAD);
2926 		/* we don't release if we want to retry with mpa_v1 */
2927 		if (!ep->retry_with_mpa_v1)
2928 			release = 1;
2929 	}
2930 	mutex_unlock(&ep->com.mutex);
2931 
2932 	rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2933 	if (WARN_ON(!rpl_skb)) {
2934 		release = 1;
2935 		goto out;
2936 	}
2937 
2938 	cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2939 
2940 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2941 out:
2942 	if (release)
2943 		release_ep_resources(ep);
2944 	else if (ep->retry_with_mpa_v1) {
2945 		if (ep->com.remote_addr.ss_family == AF_INET6) {
2946 			struct sockaddr_in6 *sin6 =
2947 					(struct sockaddr_in6 *)
2948 					&ep->com.local_addr;
2949 			cxgb4_clip_release(
2950 					ep->com.dev->rdev.lldi.ports[0],
2951 					(const u32 *)&sin6->sin6_addr.s6_addr,
2952 					1);
2953 		}
2954 		xa_erase_irq(&ep->com.dev->hwtids, ep->hwtid);
2955 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2956 				 ep->com.local_addr.ss_family);
2957 		dst_release(ep->dst);
2958 		cxgb4_l2t_release(ep->l2t);
2959 		c4iw_reconnect(ep);
2960 	}
2961 
2962 deref_ep:
2963 	c4iw_put_ep(&ep->com);
2964 	/* Dereferencing ep, referenced in peer_abort_intr() */
2965 	c4iw_put_ep(&ep->com);
2966 	return 0;
2967 }
2968 
2969 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2970 {
2971 	struct c4iw_ep *ep;
2972 	struct c4iw_qp_attributes attrs;
2973 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2974 	int release = 0;
2975 	unsigned int tid = GET_TID(rpl);
2976 
2977 	ep = get_ep_from_tid(dev, tid);
2978 	if (!ep)
2979 		return 0;
2980 
2981 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2982 
2983 	/* The cm_id may be null if we failed to connect */
2984 	mutex_lock(&ep->com.mutex);
2985 	set_bit(CLOSE_CON_RPL, &ep->com.history);
2986 	switch (ep->com.state) {
2987 	case CLOSING:
2988 		__state_set(&ep->com, MORIBUND);
2989 		break;
2990 	case MORIBUND:
2991 		(void)stop_ep_timer(ep);
2992 		if ((ep->com.cm_id) && (ep->com.qp)) {
2993 			attrs.next_state = C4IW_QP_STATE_IDLE;
2994 			c4iw_modify_qp(ep->com.qp->rhp,
2995 					     ep->com.qp,
2996 					     C4IW_QP_ATTR_NEXT_STATE,
2997 					     &attrs, 1);
2998 		}
2999 		close_complete_upcall(ep, 0);
3000 		__state_set(&ep->com, DEAD);
3001 		release = 1;
3002 		break;
3003 	case ABORTING:
3004 	case DEAD:
3005 		break;
3006 	default:
3007 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3008 		break;
3009 	}
3010 	mutex_unlock(&ep->com.mutex);
3011 	if (release)
3012 		release_ep_resources(ep);
3013 	c4iw_put_ep(&ep->com);
3014 	return 0;
3015 }
3016 
3017 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3018 {
3019 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
3020 	unsigned int tid = GET_TID(rpl);
3021 	struct c4iw_ep *ep;
3022 	struct c4iw_qp_attributes attrs;
3023 
3024 	ep = get_ep_from_tid(dev, tid);
3025 
3026 	if (ep) {
3027 		if (ep->com.qp) {
3028 			pr_warn("TERM received tid %u qpid %u\n", tid,
3029 				ep->com.qp->wq.sq.qid);
3030 			attrs.next_state = C4IW_QP_STATE_TERMINATE;
3031 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3032 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3033 		}
3034 
3035 		/* As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3,
3036 		 * when entering the TERM state the RNIC MUST initiate a CLOSE.
3037 		 */
3038 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3039 		c4iw_put_ep(&ep->com);
3040 	} else
3041 		pr_warn("TERM received tid %u no ep/qp\n", tid);
3042 
3043 	return 0;
3044 }
3045 
3046 /*
3047  * Upcall from the adapter indicating data has been transmitted.
3048  * For us its just the single MPA request or reply.  We can now free
3049  * the skb holding the mpa message.
3050  */
3051 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3052 {
3053 	struct c4iw_ep *ep;
3054 	struct cpl_fw4_ack *hdr = cplhdr(skb);
3055 	u8 credits = hdr->credits;
3056 	unsigned int tid = GET_TID(hdr);
3057 
3058 
3059 	ep = get_ep_from_tid(dev, tid);
3060 	if (!ep)
3061 		return 0;
3062 	pr_debug("ep %p tid %u credits %u\n",
3063 		 ep, ep->hwtid, credits);
3064 	if (credits == 0) {
3065 		pr_debug("0 credit ack ep %p tid %u state %u\n",
3066 			 ep, ep->hwtid, state_read(&ep->com));
3067 		goto out;
3068 	}
3069 
3070 	dst_confirm(ep->dst);
3071 	if (ep->mpa_skb) {
3072 		pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3073 			 ep, ep->hwtid, state_read(&ep->com),
3074 			 ep->mpa_attr.initiator ? 1 : 0);
3075 		mutex_lock(&ep->com.mutex);
3076 		kfree_skb(ep->mpa_skb);
3077 		ep->mpa_skb = NULL;
3078 		if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3079 			stop_ep_timer(ep);
3080 		mutex_unlock(&ep->com.mutex);
3081 	}
3082 out:
3083 	c4iw_put_ep(&ep->com);
3084 	return 0;
3085 }
3086 
3087 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3088 {
3089 	int abort;
3090 	struct c4iw_ep *ep = to_ep(cm_id);
3091 
3092 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3093 
3094 	mutex_lock(&ep->com.mutex);
3095 	if (ep->com.state != MPA_REQ_RCVD) {
3096 		mutex_unlock(&ep->com.mutex);
3097 		c4iw_put_ep(&ep->com);
3098 		return -ECONNRESET;
3099 	}
3100 	set_bit(ULP_REJECT, &ep->com.history);
3101 	if (mpa_rev == 0)
3102 		abort = 1;
3103 	else
3104 		abort = send_mpa_reject(ep, pdata, pdata_len);
3105 	mutex_unlock(&ep->com.mutex);
3106 
3107 	stop_ep_timer(ep);
3108 	c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3109 	c4iw_put_ep(&ep->com);
3110 	return 0;
3111 }
3112 
3113 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3114 {
3115 	int err;
3116 	struct c4iw_qp_attributes attrs;
3117 	enum c4iw_qp_attr_mask mask;
3118 	struct c4iw_ep *ep = to_ep(cm_id);
3119 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3120 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3121 	int abort = 0;
3122 
3123 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3124 
3125 	mutex_lock(&ep->com.mutex);
3126 	if (ep->com.state != MPA_REQ_RCVD) {
3127 		err = -ECONNRESET;
3128 		goto err_out;
3129 	}
3130 
3131 	if (!qp) {
3132 		err = -EINVAL;
3133 		goto err_out;
3134 	}
3135 
3136 	set_bit(ULP_ACCEPT, &ep->com.history);
3137 	if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3138 	    (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3139 		err = -EINVAL;
3140 		goto err_abort;
3141 	}
3142 
3143 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3144 		if (conn_param->ord > ep->ird) {
3145 			if (RELAXED_IRD_NEGOTIATION) {
3146 				conn_param->ord = ep->ird;
3147 			} else {
3148 				ep->ird = conn_param->ird;
3149 				ep->ord = conn_param->ord;
3150 				send_mpa_reject(ep, conn_param->private_data,
3151 						conn_param->private_data_len);
3152 				err = -ENOMEM;
3153 				goto err_abort;
3154 			}
3155 		}
3156 		if (conn_param->ird < ep->ord) {
3157 			if (RELAXED_IRD_NEGOTIATION &&
3158 			    ep->ord <= h->rdev.lldi.max_ordird_qp) {
3159 				conn_param->ird = ep->ord;
3160 			} else {
3161 				err = -ENOMEM;
3162 				goto err_abort;
3163 			}
3164 		}
3165 	}
3166 	ep->ird = conn_param->ird;
3167 	ep->ord = conn_param->ord;
3168 
3169 	if (ep->mpa_attr.version == 1) {
3170 		if (peer2peer && ep->ird == 0)
3171 			ep->ird = 1;
3172 	} else {
3173 		if (peer2peer &&
3174 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3175 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3176 			ep->ird = 1;
3177 	}
3178 
3179 	pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3180 
3181 	ep->com.cm_id = cm_id;
3182 	ref_cm_id(&ep->com);
3183 	ep->com.qp = qp;
3184 	ref_qp(ep);
3185 
3186 	/* bind QP to EP and move to RTS */
3187 	attrs.mpa_attr = ep->mpa_attr;
3188 	attrs.max_ird = ep->ird;
3189 	attrs.max_ord = ep->ord;
3190 	attrs.llp_stream_handle = ep;
3191 	attrs.next_state = C4IW_QP_STATE_RTS;
3192 
3193 	/* bind QP and TID with INIT_WR */
3194 	mask = C4IW_QP_ATTR_NEXT_STATE |
3195 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3196 			     C4IW_QP_ATTR_MPA_ATTR |
3197 			     C4IW_QP_ATTR_MAX_IRD |
3198 			     C4IW_QP_ATTR_MAX_ORD;
3199 
3200 	err = c4iw_modify_qp(ep->com.qp->rhp,
3201 			     ep->com.qp, mask, &attrs, 1);
3202 	if (err)
3203 		goto err_deref_cm_id;
3204 
3205 	set_bit(STOP_MPA_TIMER, &ep->com.flags);
3206 	err = send_mpa_reply(ep, conn_param->private_data,
3207 			     conn_param->private_data_len);
3208 	if (err)
3209 		goto err_deref_cm_id;
3210 
3211 	__state_set(&ep->com, FPDU_MODE);
3212 	established_upcall(ep);
3213 	mutex_unlock(&ep->com.mutex);
3214 	c4iw_put_ep(&ep->com);
3215 	return 0;
3216 err_deref_cm_id:
3217 	deref_cm_id(&ep->com);
3218 err_abort:
3219 	abort = 1;
3220 err_out:
3221 	mutex_unlock(&ep->com.mutex);
3222 	if (abort)
3223 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3224 	c4iw_put_ep(&ep->com);
3225 	return err;
3226 }
3227 
3228 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3229 {
3230 	struct in_device *ind;
3231 	int found = 0;
3232 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3233 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3234 	const struct in_ifaddr *ifa;
3235 
3236 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
3237 	if (!ind)
3238 		return -EADDRNOTAVAIL;
3239 	rcu_read_lock();
3240 	in_dev_for_each_ifa_rcu(ifa, ind) {
3241 		if (ifa->ifa_flags & IFA_F_SECONDARY)
3242 			continue;
3243 		laddr->sin_addr.s_addr = ifa->ifa_address;
3244 		raddr->sin_addr.s_addr = ifa->ifa_address;
3245 		found = 1;
3246 		break;
3247 	}
3248 	rcu_read_unlock();
3249 
3250 	in_dev_put(ind);
3251 	return found ? 0 : -EADDRNOTAVAIL;
3252 }
3253 
3254 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3255 		      unsigned char banned_flags)
3256 {
3257 	struct inet6_dev *idev;
3258 	int err = -EADDRNOTAVAIL;
3259 
3260 	rcu_read_lock();
3261 	idev = __in6_dev_get(dev);
3262 	if (idev != NULL) {
3263 		struct inet6_ifaddr *ifp;
3264 
3265 		read_lock_bh(&idev->lock);
3266 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
3267 			if (ifp->scope == IFA_LINK &&
3268 			    !(ifp->flags & banned_flags)) {
3269 				memcpy(addr, &ifp->addr, 16);
3270 				err = 0;
3271 				break;
3272 			}
3273 		}
3274 		read_unlock_bh(&idev->lock);
3275 	}
3276 	rcu_read_unlock();
3277 	return err;
3278 }
3279 
3280 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3281 {
3282 	struct in6_addr addr;
3283 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3284 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3285 
3286 	if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3287 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3288 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3289 		return 0;
3290 	}
3291 	return -EADDRNOTAVAIL;
3292 }
3293 
3294 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3295 {
3296 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3297 	struct c4iw_ep *ep;
3298 	int err = 0;
3299 	struct sockaddr_in *laddr;
3300 	struct sockaddr_in *raddr;
3301 	struct sockaddr_in6 *laddr6;
3302 	struct sockaddr_in6 *raddr6;
3303 	__u8 *ra;
3304 	int iptype;
3305 
3306 	if ((conn_param->ord > cur_max_read_depth(dev)) ||
3307 	    (conn_param->ird > cur_max_read_depth(dev))) {
3308 		err = -EINVAL;
3309 		goto out;
3310 	}
3311 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3312 	if (!ep) {
3313 		pr_err("%s - cannot alloc ep\n", __func__);
3314 		err = -ENOMEM;
3315 		goto out;
3316 	}
3317 
3318 	skb_queue_head_init(&ep->com.ep_skb_list);
3319 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3320 		err = -ENOMEM;
3321 		goto fail1;
3322 	}
3323 
3324 	timer_setup(&ep->timer, ep_timeout, 0);
3325 	ep->plen = conn_param->private_data_len;
3326 	if (ep->plen)
3327 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3328 		       conn_param->private_data, ep->plen);
3329 	ep->ird = conn_param->ird;
3330 	ep->ord = conn_param->ord;
3331 
3332 	if (peer2peer && ep->ord == 0)
3333 		ep->ord = 1;
3334 
3335 	ep->com.cm_id = cm_id;
3336 	ref_cm_id(&ep->com);
3337 	cm_id->provider_data = ep;
3338 	ep->com.dev = dev;
3339 	ep->com.qp = get_qhp(dev, conn_param->qpn);
3340 	if (!ep->com.qp) {
3341 		pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3342 		err = -EINVAL;
3343 		goto fail2;
3344 	}
3345 	ref_qp(ep);
3346 	pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3347 		 ep->com.qp, cm_id);
3348 
3349 	/*
3350 	 * Allocate an active TID to initiate a TCP connection.
3351 	 */
3352 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3353 	if (ep->atid == -1) {
3354 		pr_err("%s - cannot alloc atid\n", __func__);
3355 		err = -ENOMEM;
3356 		goto fail2;
3357 	}
3358 	err = xa_insert_irq(&dev->atids, ep->atid, ep, GFP_KERNEL);
3359 	if (err)
3360 		goto fail5;
3361 
3362 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3363 	       sizeof(ep->com.local_addr));
3364 	memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3365 	       sizeof(ep->com.remote_addr));
3366 
3367 	laddr = (struct sockaddr_in *)&ep->com.local_addr;
3368 	raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3369 	laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3370 	raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3371 
3372 	if (cm_id->m_remote_addr.ss_family == AF_INET) {
3373 		iptype = 4;
3374 		ra = (__u8 *)&raddr->sin_addr;
3375 
3376 		/*
3377 		 * Handle loopback requests to INADDR_ANY.
3378 		 */
3379 		if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3380 			err = pick_local_ipaddrs(dev, cm_id);
3381 			if (err)
3382 				goto fail3;
3383 		}
3384 
3385 		/* find a route */
3386 		pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3387 			 &laddr->sin_addr, ntohs(laddr->sin_port),
3388 			 ra, ntohs(raddr->sin_port));
3389 		ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3390 					  laddr->sin_addr.s_addr,
3391 					  raddr->sin_addr.s_addr,
3392 					  laddr->sin_port,
3393 					  raddr->sin_port, cm_id->tos);
3394 	} else {
3395 		iptype = 6;
3396 		ra = (__u8 *)&raddr6->sin6_addr;
3397 
3398 		/*
3399 		 * Handle loopback requests to INADDR_ANY.
3400 		 */
3401 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3402 			err = pick_local_ip6addrs(dev, cm_id);
3403 			if (err)
3404 				goto fail3;
3405 		}
3406 
3407 		/* find a route */
3408 		pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3409 			 laddr6->sin6_addr.s6_addr,
3410 			 ntohs(laddr6->sin6_port),
3411 			 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3412 		ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3413 					   laddr6->sin6_addr.s6_addr,
3414 					   raddr6->sin6_addr.s6_addr,
3415 					   laddr6->sin6_port,
3416 					   raddr6->sin6_port, cm_id->tos,
3417 					   raddr6->sin6_scope_id);
3418 	}
3419 	if (!ep->dst) {
3420 		pr_err("%s - cannot find route\n", __func__);
3421 		err = -EHOSTUNREACH;
3422 		goto fail3;
3423 	}
3424 
3425 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3426 			ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3427 	if (err) {
3428 		pr_err("%s - cannot alloc l2e\n", __func__);
3429 		goto fail4;
3430 	}
3431 
3432 	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3433 		 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3434 		 ep->l2t->idx);
3435 
3436 	state_set(&ep->com, CONNECTING);
3437 	ep->tos = cm_id->tos;
3438 
3439 	/* send connect request to rnic */
3440 	err = send_connect(ep);
3441 	if (!err)
3442 		goto out;
3443 
3444 	cxgb4_l2t_release(ep->l2t);
3445 fail4:
3446 	dst_release(ep->dst);
3447 fail3:
3448 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
3449 fail5:
3450 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3451 fail2:
3452 	skb_queue_purge(&ep->com.ep_skb_list);
3453 	deref_cm_id(&ep->com);
3454 fail1:
3455 	c4iw_put_ep(&ep->com);
3456 out:
3457 	return err;
3458 }
3459 
3460 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3461 {
3462 	int err;
3463 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3464 				    &ep->com.local_addr;
3465 
3466 	if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3467 		err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3468 				     (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3469 		if (err)
3470 			return err;
3471 	}
3472 	c4iw_init_wr_wait(ep->com.wr_waitp);
3473 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3474 				   ep->stid, &sin6->sin6_addr,
3475 				   sin6->sin6_port,
3476 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
3477 	if (!err)
3478 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3479 					  ep->com.wr_waitp,
3480 					  0, 0, __func__);
3481 	else if (err > 0)
3482 		err = net_xmit_errno(err);
3483 	if (err) {
3484 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3485 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3486 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3487 		       err, ep->stid,
3488 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3489 	}
3490 	return err;
3491 }
3492 
3493 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3494 {
3495 	int err;
3496 	struct sockaddr_in *sin = (struct sockaddr_in *)
3497 				  &ep->com.local_addr;
3498 
3499 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
3500 		do {
3501 			err = cxgb4_create_server_filter(
3502 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3503 				sin->sin_addr.s_addr, sin->sin_port, 0,
3504 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3505 			if (err == -EBUSY) {
3506 				if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3507 					err = -EIO;
3508 					break;
3509 				}
3510 				set_current_state(TASK_UNINTERRUPTIBLE);
3511 				schedule_timeout(usecs_to_jiffies(100));
3512 			}
3513 		} while (err == -EBUSY);
3514 	} else {
3515 		c4iw_init_wr_wait(ep->com.wr_waitp);
3516 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3517 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3518 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3519 		if (!err)
3520 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3521 						  ep->com.wr_waitp,
3522 						  0, 0, __func__);
3523 		else if (err > 0)
3524 			err = net_xmit_errno(err);
3525 	}
3526 	if (err)
3527 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3528 		       , err, ep->stid,
3529 		       &sin->sin_addr, ntohs(sin->sin_port));
3530 	return err;
3531 }
3532 
3533 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3534 {
3535 	int err = 0;
3536 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3537 	struct c4iw_listen_ep *ep;
3538 
3539 	might_sleep();
3540 
3541 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3542 	if (!ep) {
3543 		pr_err("%s - cannot alloc ep\n", __func__);
3544 		err = -ENOMEM;
3545 		goto fail1;
3546 	}
3547 	skb_queue_head_init(&ep->com.ep_skb_list);
3548 	pr_debug("ep %p\n", ep);
3549 	ep->com.cm_id = cm_id;
3550 	ref_cm_id(&ep->com);
3551 	ep->com.dev = dev;
3552 	ep->backlog = backlog;
3553 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3554 	       sizeof(ep->com.local_addr));
3555 
3556 	/*
3557 	 * Allocate a server TID.
3558 	 */
3559 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
3560 	    ep->com.local_addr.ss_family == AF_INET)
3561 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3562 					     cm_id->m_local_addr.ss_family, ep);
3563 	else
3564 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3565 					    cm_id->m_local_addr.ss_family, ep);
3566 
3567 	if (ep->stid == -1) {
3568 		pr_err("%s - cannot alloc stid\n", __func__);
3569 		err = -ENOMEM;
3570 		goto fail2;
3571 	}
3572 	err = xa_insert_irq(&dev->stids, ep->stid, ep, GFP_KERNEL);
3573 	if (err)
3574 		goto fail3;
3575 
3576 	state_set(&ep->com, LISTEN);
3577 	if (ep->com.local_addr.ss_family == AF_INET)
3578 		err = create_server4(dev, ep);
3579 	else
3580 		err = create_server6(dev, ep);
3581 	if (!err) {
3582 		cm_id->provider_data = ep;
3583 		goto out;
3584 	}
3585 	xa_erase_irq(&ep->com.dev->stids, ep->stid);
3586 fail3:
3587 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3588 			ep->com.local_addr.ss_family);
3589 fail2:
3590 	deref_cm_id(&ep->com);
3591 	c4iw_put_ep(&ep->com);
3592 fail1:
3593 out:
3594 	return err;
3595 }
3596 
3597 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3598 {
3599 	int err;
3600 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3601 
3602 	pr_debug("ep %p\n", ep);
3603 
3604 	might_sleep();
3605 	state_set(&ep->com, DEAD);
3606 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3607 	    ep->com.local_addr.ss_family == AF_INET) {
3608 		err = cxgb4_remove_server_filter(
3609 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
3610 			ep->com.dev->rdev.lldi.rxq_ids[0], false);
3611 	} else {
3612 		struct sockaddr_in6 *sin6;
3613 		c4iw_init_wr_wait(ep->com.wr_waitp);
3614 		err = cxgb4_remove_server(
3615 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3616 				ep->com.dev->rdev.lldi.rxq_ids[0],
3617 				ep->com.local_addr.ss_family == AF_INET6);
3618 		if (err)
3619 			goto done;
3620 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3621 					  0, 0, __func__);
3622 		sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3623 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3624 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3625 	}
3626 	xa_erase_irq(&ep->com.dev->stids, ep->stid);
3627 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3628 			ep->com.local_addr.ss_family);
3629 done:
3630 	deref_cm_id(&ep->com);
3631 	c4iw_put_ep(&ep->com);
3632 	return err;
3633 }
3634 
3635 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3636 {
3637 	int ret = 0;
3638 	int close = 0;
3639 	int fatal = 0;
3640 	struct c4iw_rdev *rdev;
3641 
3642 	mutex_lock(&ep->com.mutex);
3643 
3644 	pr_debug("ep %p state %s, abrupt %d\n", ep,
3645 		 states[ep->com.state], abrupt);
3646 
3647 	/*
3648 	 * Ref the ep here in case we have fatal errors causing the
3649 	 * ep to be released and freed.
3650 	 */
3651 	c4iw_get_ep(&ep->com);
3652 
3653 	rdev = &ep->com.dev->rdev;
3654 	if (c4iw_fatal_error(rdev)) {
3655 		fatal = 1;
3656 		close_complete_upcall(ep, -EIO);
3657 		ep->com.state = DEAD;
3658 	}
3659 	switch (ep->com.state) {
3660 	case MPA_REQ_WAIT:
3661 	case MPA_REQ_SENT:
3662 	case MPA_REQ_RCVD:
3663 	case MPA_REP_SENT:
3664 	case FPDU_MODE:
3665 	case CONNECTING:
3666 		close = 1;
3667 		if (abrupt)
3668 			ep->com.state = ABORTING;
3669 		else {
3670 			ep->com.state = CLOSING;
3671 
3672 			/*
3673 			 * if we close before we see the fw4_ack() then we fix
3674 			 * up the timer state since we're reusing it.
3675 			 */
3676 			if (ep->mpa_skb &&
3677 			    test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3678 				clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3679 				stop_ep_timer(ep);
3680 			}
3681 			start_ep_timer(ep);
3682 		}
3683 		set_bit(CLOSE_SENT, &ep->com.flags);
3684 		break;
3685 	case CLOSING:
3686 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3687 			close = 1;
3688 			if (abrupt) {
3689 				(void)stop_ep_timer(ep);
3690 				ep->com.state = ABORTING;
3691 			} else
3692 				ep->com.state = MORIBUND;
3693 		}
3694 		break;
3695 	case MORIBUND:
3696 	case ABORTING:
3697 	case DEAD:
3698 		pr_debug("ignoring disconnect ep %p state %u\n",
3699 			 ep, ep->com.state);
3700 		break;
3701 	default:
3702 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3703 		break;
3704 	}
3705 
3706 	if (close) {
3707 		if (abrupt) {
3708 			set_bit(EP_DISC_ABORT, &ep->com.history);
3709 			ret = send_abort(ep);
3710 		} else {
3711 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3712 			ret = send_halfclose(ep);
3713 		}
3714 		if (ret) {
3715 			set_bit(EP_DISC_FAIL, &ep->com.history);
3716 			if (!abrupt) {
3717 				stop_ep_timer(ep);
3718 				close_complete_upcall(ep, -EIO);
3719 			}
3720 			if (ep->com.qp) {
3721 				struct c4iw_qp_attributes attrs;
3722 
3723 				attrs.next_state = C4IW_QP_STATE_ERROR;
3724 				ret = c4iw_modify_qp(ep->com.qp->rhp,
3725 						     ep->com.qp,
3726 						     C4IW_QP_ATTR_NEXT_STATE,
3727 						     &attrs, 1);
3728 				if (ret)
3729 					pr_err("%s - qp <- error failed!\n",
3730 					       __func__);
3731 			}
3732 			fatal = 1;
3733 		}
3734 	}
3735 	mutex_unlock(&ep->com.mutex);
3736 	c4iw_put_ep(&ep->com);
3737 	if (fatal)
3738 		release_ep_resources(ep);
3739 	return ret;
3740 }
3741 
3742 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3743 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3744 {
3745 	struct c4iw_ep *ep;
3746 	int atid = be32_to_cpu(req->tid);
3747 
3748 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3749 					   (__force u32) req->tid);
3750 	if (!ep)
3751 		return;
3752 
3753 	switch (req->retval) {
3754 	case FW_ENOMEM:
3755 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3756 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3757 			send_fw_act_open_req(ep, atid);
3758 			return;
3759 		}
3760 		fallthrough;
3761 	case FW_EADDRINUSE:
3762 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3763 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3764 			send_fw_act_open_req(ep, atid);
3765 			return;
3766 		}
3767 		break;
3768 	default:
3769 		pr_info("%s unexpected ofld conn wr retval %d\n",
3770 		       __func__, req->retval);
3771 		break;
3772 	}
3773 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3774 	       req->retval, atid);
3775 	mutex_lock(&dev->rdev.stats.lock);
3776 	dev->rdev.stats.act_ofld_conn_fails++;
3777 	mutex_unlock(&dev->rdev.stats.lock);
3778 	connect_reply_upcall(ep, status2errno(req->retval));
3779 	state_set(&ep->com, DEAD);
3780 	if (ep->com.remote_addr.ss_family == AF_INET6) {
3781 		struct sockaddr_in6 *sin6 =
3782 			(struct sockaddr_in6 *)&ep->com.local_addr;
3783 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3784 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3785 	}
3786 	xa_erase_irq(&dev->atids, atid);
3787 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3788 	dst_release(ep->dst);
3789 	cxgb4_l2t_release(ep->l2t);
3790 	c4iw_put_ep(&ep->com);
3791 }
3792 
3793 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3794 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3795 {
3796 	struct sk_buff *rpl_skb;
3797 	struct cpl_pass_accept_req *cpl;
3798 	int ret;
3799 
3800 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3801 	if (req->retval) {
3802 		pr_err("%s passive open failure %d\n", __func__, req->retval);
3803 		mutex_lock(&dev->rdev.stats.lock);
3804 		dev->rdev.stats.pas_ofld_conn_fails++;
3805 		mutex_unlock(&dev->rdev.stats.lock);
3806 		kfree_skb(rpl_skb);
3807 	} else {
3808 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3809 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3810 					(__force u32) htonl(
3811 					(__force u32) req->tid)));
3812 		ret = pass_accept_req(dev, rpl_skb);
3813 		if (!ret)
3814 			kfree_skb(rpl_skb);
3815 	}
3816 	return;
3817 }
3818 
3819 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3820 {
3821 	u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3822 	u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3823 	u64 t;
3824 	u32 shift = 32;
3825 
3826 	t = (thi << shift) | (tlo >> shift);
3827 
3828 	return t;
3829 }
3830 
3831 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3832 {
3833 	u32 v;
3834 	u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3835 
3836 	if (word & 0x1)
3837 		shift += 32;
3838 	v = (t >> shift) & mask;
3839 	return v;
3840 }
3841 
3842 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3843 {
3844 	struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3845 	__be64 *tcb = (__be64 *)(rpl + 1);
3846 	unsigned int tid = GET_TID(rpl);
3847 	struct c4iw_ep *ep;
3848 	u64 t_flags_64;
3849 	u32 rx_pdu_out;
3850 
3851 	ep = get_ep_from_tid(dev, tid);
3852 	if (!ep)
3853 		return 0;
3854 	/* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3855 	 * determine if there's a rx PDU feedback event pending.
3856 	 *
3857 	 * If that bit is set, it means we'll need to re-read the TCB's
3858 	 * rq_start value. The final value is the one present in a TCB
3859 	 * with the TF_RX_PDU_OUT bit cleared.
3860 	 */
3861 
3862 	t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3863 	rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3864 
3865 	c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3866 	c4iw_put_ep(&ep->com); /* from read_tcb() */
3867 
3868 	/* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3869 	if (rx_pdu_out) {
3870 		if (++ep->rx_pdu_out_cnt >= 2) {
3871 			WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3872 			goto cleanup;
3873 		}
3874 		read_tcb(ep);
3875 		return 0;
3876 	}
3877 
3878 	ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_M,
3879 					  TCB_RQ_START_S);
3880 cleanup:
3881 	pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3882 
3883 	if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3884 		finish_peer_abort(dev, ep);
3885 	else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3886 		send_abort_req(ep);
3887 	else
3888 		WARN_ONCE(1, "unexpected state!");
3889 
3890 	return 0;
3891 }
3892 
3893 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3894 {
3895 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3896 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3897 
3898 	switch (rpl->type) {
3899 	case FW6_TYPE_CQE:
3900 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3901 		break;
3902 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3903 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3904 		switch (req->t_state) {
3905 		case TCP_SYN_SENT:
3906 			active_ofld_conn_reply(dev, skb, req);
3907 			break;
3908 		case TCP_SYN_RECV:
3909 			passive_ofld_conn_reply(dev, skb, req);
3910 			break;
3911 		default:
3912 			pr_err("%s unexpected ofld conn wr state %d\n",
3913 			       __func__, req->t_state);
3914 			break;
3915 		}
3916 		break;
3917 	}
3918 	return 0;
3919 }
3920 
3921 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3922 {
3923 	__be32 l2info;
3924 	__be16 hdr_len, vlantag, len;
3925 	u16 eth_hdr_len;
3926 	int tcp_hdr_len, ip_hdr_len;
3927 	u8 intf;
3928 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3929 	struct cpl_pass_accept_req *req;
3930 	struct tcp_options_received tmp_opt;
3931 	struct c4iw_dev *dev;
3932 	enum chip_type type;
3933 
3934 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3935 	/* Store values from cpl_rx_pkt in temporary location. */
3936 	vlantag = cpl->vlan;
3937 	len = cpl->len;
3938 	l2info  = cpl->l2info;
3939 	hdr_len = cpl->hdr_len;
3940 	intf = cpl->iff;
3941 
3942 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3943 
3944 	/*
3945 	 * We need to parse the TCP options from SYN packet.
3946 	 * to generate cpl_pass_accept_req.
3947 	 */
3948 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3949 	tcp_clear_options(&tmp_opt);
3950 	tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3951 
3952 	req = __skb_push(skb, sizeof(*req));
3953 	memset(req, 0, sizeof(*req));
3954 	req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3955 			 SYN_MAC_IDX_V(RX_MACIDX_G(
3956 			 be32_to_cpu(l2info))) |
3957 			 SYN_XACT_MATCH_F);
3958 	type = dev->rdev.lldi.adapter_type;
3959 	tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3960 	ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3961 	req->hdr_len =
3962 		cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3963 	if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3964 		eth_hdr_len = is_t4(type) ?
3965 				RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3966 				RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3967 		req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3968 					    IP_HDR_LEN_V(ip_hdr_len) |
3969 					    ETH_HDR_LEN_V(eth_hdr_len));
3970 	} else { /* T6 and later */
3971 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3972 		req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3973 					    T6_IP_HDR_LEN_V(ip_hdr_len) |
3974 					    T6_ETH_HDR_LEN_V(eth_hdr_len));
3975 	}
3976 	req->vlan = vlantag;
3977 	req->len = len;
3978 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3979 				    PASS_OPEN_TOS_V(tos));
3980 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3981 	if (tmp_opt.wscale_ok)
3982 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3983 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3984 	if (tmp_opt.sack_ok)
3985 		req->tcpopt.sack = 1;
3986 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3987 	return;
3988 }
3989 
3990 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3991 				  __be32 laddr, __be16 lport,
3992 				  __be32 raddr, __be16 rport,
3993 				  u32 rcv_isn, u32 filter, u16 window,
3994 				  u32 rss_qid, u8 port_id)
3995 {
3996 	struct sk_buff *req_skb;
3997 	struct fw_ofld_connection_wr *req;
3998 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
3999 	int ret;
4000 
4001 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
4002 	if (!req_skb)
4003 		return;
4004 	req = __skb_put_zero(req_skb, sizeof(*req));
4005 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
4006 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4007 	req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4008 	req->le.filter = (__force __be32) filter;
4009 	req->le.lport = lport;
4010 	req->le.pport = rport;
4011 	req->le.u.ipv4.lip = laddr;
4012 	req->le.u.ipv4.pip = raddr;
4013 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4014 	req->tcb.rcv_adv = htons(window);
4015 	req->tcb.t_state_to_astid =
4016 		 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4017 			FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4018 			FW_OFLD_CONNECTION_WR_ASTID_V(
4019 			PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4020 
4021 	/*
4022 	 * We store the qid in opt2 which will be used by the firmware
4023 	 * to send us the wr response.
4024 	 */
4025 	req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4026 
4027 	/*
4028 	 * We initialize the MSS index in TCB to 0xF.
4029 	 * So that when driver sends cpl_pass_accept_rpl
4030 	 * TCB picks up the correct value. If this was 0
4031 	 * TP will ignore any value > 0 for MSS index.
4032 	 */
4033 	req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4034 	req->cookie = (uintptr_t)skb;
4035 
4036 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4037 	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4038 	if (ret < 0) {
4039 		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4040 		       ret);
4041 		kfree_skb(skb);
4042 		kfree_skb(req_skb);
4043 	}
4044 }
4045 
4046 /*
4047  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4048  * messages when a filter is being used instead of server to
4049  * redirect a syn packet. When packets hit filter they are redirected
4050  * to the offload queue and driver tries to establish the connection
4051  * using firmware work request.
4052  */
4053 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4054 {
4055 	int stid;
4056 	unsigned int filter;
4057 	struct ethhdr *eh = NULL;
4058 	struct vlan_ethhdr *vlan_eh = NULL;
4059 	struct iphdr *iph;
4060 	struct tcphdr *tcph;
4061 	struct rss_header *rss = (void *)skb->data;
4062 	struct cpl_rx_pkt *cpl = (void *)skb->data;
4063 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
4064 	struct l2t_entry *e;
4065 	struct dst_entry *dst;
4066 	struct c4iw_ep *lep = NULL;
4067 	u16 window;
4068 	struct port_info *pi;
4069 	struct net_device *pdev;
4070 	u16 rss_qid, eth_hdr_len;
4071 	int step;
4072 	struct neighbour *neigh;
4073 
4074 	/* Drop all non-SYN packets */
4075 	if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4076 		goto reject;
4077 
4078 	/*
4079 	 * Drop all packets which did not hit the filter.
4080 	 * Unlikely to happen.
4081 	 */
4082 	if (!(rss->filter_hit && rss->filter_tid))
4083 		goto reject;
4084 
4085 	/*
4086 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4087 	 */
4088 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4089 
4090 	lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4091 	if (!lep) {
4092 		pr_warn("%s connect request on invalid stid %d\n",
4093 			__func__, stid);
4094 		goto reject;
4095 	}
4096 
4097 	switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4098 	case CHELSIO_T4:
4099 		eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4100 		break;
4101 	case CHELSIO_T5:
4102 		eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4103 		break;
4104 	case CHELSIO_T6:
4105 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4106 		break;
4107 	default:
4108 		pr_err("T%d Chip is not supported\n",
4109 		       CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4110 		goto reject;
4111 	}
4112 
4113 	if (eth_hdr_len == ETH_HLEN) {
4114 		eh = (struct ethhdr *)(req + 1);
4115 		iph = (struct iphdr *)(eh + 1);
4116 	} else {
4117 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
4118 		iph = (struct iphdr *)(vlan_eh + 1);
4119 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4120 	}
4121 
4122 	if (iph->version != 0x4)
4123 		goto reject;
4124 
4125 	tcph = (struct tcphdr *)(iph + 1);
4126 	skb_set_network_header(skb, (void *)iph - (void *)rss);
4127 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4128 	skb_get(skb);
4129 
4130 	pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4131 		 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4132 		 ntohs(tcph->source), iph->tos);
4133 
4134 	dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4135 			      iph->daddr, iph->saddr, tcph->dest,
4136 			      tcph->source, iph->tos);
4137 	if (!dst) {
4138 		pr_err("%s - failed to find dst entry!\n", __func__);
4139 		goto reject;
4140 	}
4141 	neigh = dst_neigh_lookup_skb(dst, skb);
4142 
4143 	if (!neigh) {
4144 		pr_err("%s - failed to allocate neigh!\n", __func__);
4145 		goto free_dst;
4146 	}
4147 
4148 	if (neigh->dev->flags & IFF_LOOPBACK) {
4149 		pdev = ip_dev_find(&init_net, iph->daddr);
4150 		if (!pdev) {
4151 			pr_err("%s - failed to find device!\n", __func__);
4152 			goto free_dst;
4153 		}
4154 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4155 				    pdev, 0);
4156 		pi = (struct port_info *)netdev_priv(pdev);
4157 		dev_put(pdev);
4158 	} else {
4159 		pdev = get_real_dev(neigh->dev);
4160 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4161 					pdev, 0);
4162 		pi = (struct port_info *)netdev_priv(pdev);
4163 	}
4164 	neigh_release(neigh);
4165 	if (!e) {
4166 		pr_err("%s - failed to allocate l2t entry!\n",
4167 		       __func__);
4168 		goto free_dst;
4169 	}
4170 
4171 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4172 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4173 	window = (__force u16) htons((__force u16)tcph->window);
4174 
4175 	/* Calcuate filter portion for LE region. */
4176 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4177 						    dev->rdev.lldi.ports[0],
4178 						    e));
4179 
4180 	/*
4181 	 * Synthesize the cpl_pass_accept_req. We have everything except the
4182 	 * TID. Once firmware sends a reply with TID we update the TID field
4183 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
4184 	 */
4185 	build_cpl_pass_accept_req(skb, stid, iph->tos);
4186 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4187 			      tcph->source, ntohl(tcph->seq), filter, window,
4188 			      rss_qid, pi->port_id);
4189 	cxgb4_l2t_release(e);
4190 free_dst:
4191 	dst_release(dst);
4192 reject:
4193 	if (lep)
4194 		c4iw_put_ep(&lep->com);
4195 	return 0;
4196 }
4197 
4198 /*
4199  * These are the real handlers that are called from a
4200  * work queue.
4201  */
4202 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4203 	[CPL_ACT_ESTABLISH] = act_establish,
4204 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
4205 	[CPL_RX_DATA] = rx_data,
4206 	[CPL_ABORT_RPL_RSS] = abort_rpl,
4207 	[CPL_ABORT_RPL] = abort_rpl,
4208 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
4209 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4210 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4211 	[CPL_PASS_ESTABLISH] = pass_establish,
4212 	[CPL_PEER_CLOSE] = peer_close,
4213 	[CPL_ABORT_REQ_RSS] = peer_abort,
4214 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
4215 	[CPL_RDMA_TERMINATE] = terminate,
4216 	[CPL_FW4_ACK] = fw4_ack,
4217 	[CPL_GET_TCB_RPL] = read_tcb_rpl,
4218 	[CPL_FW6_MSG] = deferred_fw6_msg,
4219 	[CPL_RX_PKT] = rx_pkt,
4220 	[FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4221 	[FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4222 };
4223 
4224 static void process_timeout(struct c4iw_ep *ep)
4225 {
4226 	struct c4iw_qp_attributes attrs;
4227 	int abort = 1;
4228 
4229 	mutex_lock(&ep->com.mutex);
4230 	pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4231 	set_bit(TIMEDOUT, &ep->com.history);
4232 	switch (ep->com.state) {
4233 	case MPA_REQ_SENT:
4234 		connect_reply_upcall(ep, -ETIMEDOUT);
4235 		break;
4236 	case MPA_REQ_WAIT:
4237 	case MPA_REQ_RCVD:
4238 	case MPA_REP_SENT:
4239 	case FPDU_MODE:
4240 		break;
4241 	case CLOSING:
4242 	case MORIBUND:
4243 		if (ep->com.cm_id && ep->com.qp) {
4244 			attrs.next_state = C4IW_QP_STATE_ERROR;
4245 			c4iw_modify_qp(ep->com.qp->rhp,
4246 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4247 				     &attrs, 1);
4248 		}
4249 		close_complete_upcall(ep, -ETIMEDOUT);
4250 		break;
4251 	case ABORTING:
4252 	case DEAD:
4253 
4254 		/*
4255 		 * These states are expected if the ep timed out at the same
4256 		 * time as another thread was calling stop_ep_timer().
4257 		 * So we silently do nothing for these states.
4258 		 */
4259 		abort = 0;
4260 		break;
4261 	default:
4262 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4263 			__func__, ep, ep->hwtid, ep->com.state);
4264 		abort = 0;
4265 	}
4266 	mutex_unlock(&ep->com.mutex);
4267 	if (abort)
4268 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4269 	c4iw_put_ep(&ep->com);
4270 }
4271 
4272 static void process_timedout_eps(void)
4273 {
4274 	struct c4iw_ep *ep;
4275 
4276 	spin_lock_irq(&timeout_lock);
4277 	while (!list_empty(&timeout_list)) {
4278 		struct list_head *tmp;
4279 
4280 		tmp = timeout_list.next;
4281 		list_del(tmp);
4282 		tmp->next = NULL;
4283 		tmp->prev = NULL;
4284 		spin_unlock_irq(&timeout_lock);
4285 		ep = list_entry(tmp, struct c4iw_ep, entry);
4286 		process_timeout(ep);
4287 		spin_lock_irq(&timeout_lock);
4288 	}
4289 	spin_unlock_irq(&timeout_lock);
4290 }
4291 
4292 static void process_work(struct work_struct *work)
4293 {
4294 	struct sk_buff *skb = NULL;
4295 	struct c4iw_dev *dev;
4296 	struct cpl_act_establish *rpl;
4297 	unsigned int opcode;
4298 	int ret;
4299 
4300 	process_timedout_eps();
4301 	while ((skb = skb_dequeue(&rxq))) {
4302 		rpl = cplhdr(skb);
4303 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4304 		opcode = rpl->ot.opcode;
4305 
4306 		if (opcode >= ARRAY_SIZE(work_handlers) ||
4307 		    !work_handlers[opcode]) {
4308 			pr_err("No handler for opcode 0x%x.\n", opcode);
4309 			kfree_skb(skb);
4310 		} else {
4311 			ret = work_handlers[opcode](dev, skb);
4312 			if (!ret)
4313 				kfree_skb(skb);
4314 		}
4315 		process_timedout_eps();
4316 	}
4317 }
4318 
4319 static DECLARE_WORK(skb_work, process_work);
4320 
4321 static void ep_timeout(struct timer_list *t)
4322 {
4323 	struct c4iw_ep *ep = from_timer(ep, t, timer);
4324 	int kickit = 0;
4325 
4326 	spin_lock(&timeout_lock);
4327 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4328 		/*
4329 		 * Only insert if it is not already on the list.
4330 		 */
4331 		if (!ep->entry.next) {
4332 			list_add_tail(&ep->entry, &timeout_list);
4333 			kickit = 1;
4334 		}
4335 	}
4336 	spin_unlock(&timeout_lock);
4337 	if (kickit)
4338 		queue_work(workq, &skb_work);
4339 }
4340 
4341 /*
4342  * All the CM events are handled on a work queue to have a safe context.
4343  */
4344 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4345 {
4346 
4347 	/*
4348 	 * Save dev in the skb->cb area.
4349 	 */
4350 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4351 
4352 	/*
4353 	 * Queue the skb and schedule the worker thread.
4354 	 */
4355 	skb_queue_tail(&rxq, skb);
4356 	queue_work(workq, &skb_work);
4357 	return 0;
4358 }
4359 
4360 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4361 {
4362 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4363 
4364 	if (rpl->status != CPL_ERR_NONE) {
4365 		pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4366 		       rpl->status, GET_TID(rpl));
4367 	}
4368 	kfree_skb(skb);
4369 	return 0;
4370 }
4371 
4372 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4373 {
4374 	struct cpl_fw6_msg *rpl = cplhdr(skb);
4375 	struct c4iw_wr_wait *wr_waitp;
4376 	int ret;
4377 
4378 	pr_debug("type %u\n", rpl->type);
4379 
4380 	switch (rpl->type) {
4381 	case FW6_TYPE_WR_RPL:
4382 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4383 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4384 		pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4385 		if (wr_waitp)
4386 			c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4387 		kfree_skb(skb);
4388 		break;
4389 	case FW6_TYPE_CQE:
4390 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4391 		sched(dev, skb);
4392 		break;
4393 	default:
4394 		pr_err("%s unexpected fw6 msg type %u\n",
4395 		       __func__, rpl->type);
4396 		kfree_skb(skb);
4397 		break;
4398 	}
4399 	return 0;
4400 }
4401 
4402 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4403 {
4404 	struct cpl_abort_req_rss *req = cplhdr(skb);
4405 	struct c4iw_ep *ep;
4406 	unsigned int tid = GET_TID(req);
4407 
4408 	ep = get_ep_from_tid(dev, tid);
4409 	/* This EP will be dereferenced in peer_abort() */
4410 	if (!ep) {
4411 		pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4412 		kfree_skb(skb);
4413 		return 0;
4414 	}
4415 	if (cxgb_is_neg_adv(req->status)) {
4416 		pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4417 			 ep->hwtid, req->status,
4418 			 neg_adv_str(req->status));
4419 		goto out;
4420 	}
4421 	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4422 
4423 	c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4424 out:
4425 	sched(dev, skb);
4426 	return 0;
4427 }
4428 
4429 /*
4430  * Most upcalls from the T4 Core go to sched() to
4431  * schedule the processing on a work queue.
4432  */
4433 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4434 	[CPL_ACT_ESTABLISH] = sched,
4435 	[CPL_ACT_OPEN_RPL] = sched,
4436 	[CPL_RX_DATA] = sched,
4437 	[CPL_ABORT_RPL_RSS] = sched,
4438 	[CPL_ABORT_RPL] = sched,
4439 	[CPL_PASS_OPEN_RPL] = sched,
4440 	[CPL_CLOSE_LISTSRV_RPL] = sched,
4441 	[CPL_PASS_ACCEPT_REQ] = sched,
4442 	[CPL_PASS_ESTABLISH] = sched,
4443 	[CPL_PEER_CLOSE] = sched,
4444 	[CPL_CLOSE_CON_RPL] = sched,
4445 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
4446 	[CPL_RDMA_TERMINATE] = sched,
4447 	[CPL_FW4_ACK] = sched,
4448 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
4449 	[CPL_GET_TCB_RPL] = sched,
4450 	[CPL_FW6_MSG] = fw6_msg,
4451 	[CPL_RX_PKT] = sched
4452 };
4453 
4454 int __init c4iw_cm_init(void)
4455 {
4456 	skb_queue_head_init(&rxq);
4457 
4458 	workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4459 	if (!workq)
4460 		return -ENOMEM;
4461 
4462 	return 0;
4463 }
4464 
4465 void c4iw_cm_term(void)
4466 {
4467 	WARN_ON(!list_empty(&timeout_list));
4468 	destroy_workqueue(workq);
4469 }
4470