xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision 75f25bd3)
1 /*
2  * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *	  copyright notice, this list of conditions and the following
16  *	  disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *	  copyright notice, this list of conditions and the following
20  *	  disclaimer in the documentation and/or other materials
21  *	  provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41 
42 #include <net/neighbour.h>
43 #include <net/netevent.h>
44 #include <net/route.h>
45 
46 #include "iw_cxgb4.h"
47 
48 static char *states[] = {
49 	"idle",
50 	"listen",
51 	"connecting",
52 	"mpa_wait_req",
53 	"mpa_req_sent",
54 	"mpa_req_rcvd",
55 	"mpa_rep_sent",
56 	"fpdu_mode",
57 	"aborting",
58 	"closing",
59 	"moribund",
60 	"dead",
61 	NULL,
62 };
63 
64 static int dack_mode = 1;
65 module_param(dack_mode, int, 0644);
66 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
67 
68 int c4iw_max_read_depth = 8;
69 module_param(c4iw_max_read_depth, int, 0644);
70 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
71 
72 static int enable_tcp_timestamps;
73 module_param(enable_tcp_timestamps, int, 0644);
74 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
75 
76 static int enable_tcp_sack;
77 module_param(enable_tcp_sack, int, 0644);
78 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
79 
80 static int enable_tcp_window_scaling = 1;
81 module_param(enable_tcp_window_scaling, int, 0644);
82 MODULE_PARM_DESC(enable_tcp_window_scaling,
83 		 "Enable tcp window scaling (default=1)");
84 
85 int c4iw_debug;
86 module_param(c4iw_debug, int, 0644);
87 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
88 
89 static int peer2peer;
90 module_param(peer2peer, int, 0644);
91 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
92 
93 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
94 module_param(p2p_type, int, 0644);
95 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
96 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
97 
98 static int ep_timeout_secs = 60;
99 module_param(ep_timeout_secs, int, 0644);
100 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
101 				   "in seconds (default=60)");
102 
103 static int mpa_rev = 1;
104 module_param(mpa_rev, int, 0644);
105 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
106 		 "1 is spec compliant. (default=1)");
107 
108 static int markers_enabled;
109 module_param(markers_enabled, int, 0644);
110 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
111 
112 static int crc_enabled = 1;
113 module_param(crc_enabled, int, 0644);
114 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
115 
116 static int rcv_win = 256 * 1024;
117 module_param(rcv_win, int, 0644);
118 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
119 
120 static int snd_win = 128 * 1024;
121 module_param(snd_win, int, 0644);
122 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
123 
124 static struct workqueue_struct *workq;
125 
126 static struct sk_buff_head rxq;
127 
128 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
129 static void ep_timeout(unsigned long arg);
130 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
131 
132 static LIST_HEAD(timeout_list);
133 static spinlock_t timeout_lock;
134 
135 static void start_ep_timer(struct c4iw_ep *ep)
136 {
137 	PDBG("%s ep %p\n", __func__, ep);
138 	if (timer_pending(&ep->timer)) {
139 		PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
140 		del_timer_sync(&ep->timer);
141 	} else
142 		c4iw_get_ep(&ep->com);
143 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
144 	ep->timer.data = (unsigned long)ep;
145 	ep->timer.function = ep_timeout;
146 	add_timer(&ep->timer);
147 }
148 
149 static void stop_ep_timer(struct c4iw_ep *ep)
150 {
151 	PDBG("%s ep %p\n", __func__, ep);
152 	if (!timer_pending(&ep->timer)) {
153 		printk(KERN_ERR "%s timer stopped when its not running! "
154 		       "ep %p state %u\n", __func__, ep, ep->com.state);
155 		WARN_ON(1);
156 		return;
157 	}
158 	del_timer_sync(&ep->timer);
159 	c4iw_put_ep(&ep->com);
160 }
161 
162 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
163 		  struct l2t_entry *l2e)
164 {
165 	int	error = 0;
166 
167 	if (c4iw_fatal_error(rdev)) {
168 		kfree_skb(skb);
169 		PDBG("%s - device in error state - dropping\n", __func__);
170 		return -EIO;
171 	}
172 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
173 	if (error < 0)
174 		kfree_skb(skb);
175 	return error < 0 ? error : 0;
176 }
177 
178 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
179 {
180 	int	error = 0;
181 
182 	if (c4iw_fatal_error(rdev)) {
183 		kfree_skb(skb);
184 		PDBG("%s - device in error state - dropping\n", __func__);
185 		return -EIO;
186 	}
187 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
188 	if (error < 0)
189 		kfree_skb(skb);
190 	return error < 0 ? error : 0;
191 }
192 
193 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
194 {
195 	struct cpl_tid_release *req;
196 
197 	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
198 	if (!skb)
199 		return;
200 	req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
201 	INIT_TP_WR(req, hwtid);
202 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
203 	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
204 	c4iw_ofld_send(rdev, skb);
205 	return;
206 }
207 
208 static void set_emss(struct c4iw_ep *ep, u16 opt)
209 {
210 	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
211 	ep->mss = ep->emss;
212 	if (GET_TCPOPT_TSTAMP(opt))
213 		ep->emss -= 12;
214 	if (ep->emss < 128)
215 		ep->emss = 128;
216 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
217 	     ep->mss, ep->emss);
218 }
219 
220 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
221 {
222 	enum c4iw_ep_state state;
223 
224 	mutex_lock(&epc->mutex);
225 	state = epc->state;
226 	mutex_unlock(&epc->mutex);
227 	return state;
228 }
229 
230 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
231 {
232 	epc->state = new;
233 }
234 
235 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
236 {
237 	mutex_lock(&epc->mutex);
238 	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
239 	__state_set(epc, new);
240 	mutex_unlock(&epc->mutex);
241 	return;
242 }
243 
244 static void *alloc_ep(int size, gfp_t gfp)
245 {
246 	struct c4iw_ep_common *epc;
247 
248 	epc = kzalloc(size, gfp);
249 	if (epc) {
250 		kref_init(&epc->kref);
251 		mutex_init(&epc->mutex);
252 		c4iw_init_wr_wait(&epc->wr_wait);
253 	}
254 	PDBG("%s alloc ep %p\n", __func__, epc);
255 	return epc;
256 }
257 
258 void _c4iw_free_ep(struct kref *kref)
259 {
260 	struct c4iw_ep *ep;
261 
262 	ep = container_of(kref, struct c4iw_ep, com.kref);
263 	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
264 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
265 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
266 		dst_release(ep->dst);
267 		cxgb4_l2t_release(ep->l2t);
268 	}
269 	kfree(ep);
270 }
271 
272 static void release_ep_resources(struct c4iw_ep *ep)
273 {
274 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
275 	c4iw_put_ep(&ep->com);
276 }
277 
278 static int status2errno(int status)
279 {
280 	switch (status) {
281 	case CPL_ERR_NONE:
282 		return 0;
283 	case CPL_ERR_CONN_RESET:
284 		return -ECONNRESET;
285 	case CPL_ERR_ARP_MISS:
286 		return -EHOSTUNREACH;
287 	case CPL_ERR_CONN_TIMEDOUT:
288 		return -ETIMEDOUT;
289 	case CPL_ERR_TCAM_FULL:
290 		return -ENOMEM;
291 	case CPL_ERR_CONN_EXIST:
292 		return -EADDRINUSE;
293 	default:
294 		return -EIO;
295 	}
296 }
297 
298 /*
299  * Try and reuse skbs already allocated...
300  */
301 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
302 {
303 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
304 		skb_trim(skb, 0);
305 		skb_get(skb);
306 		skb_reset_transport_header(skb);
307 	} else {
308 		skb = alloc_skb(len, gfp);
309 	}
310 	return skb;
311 }
312 
313 static struct rtable *find_route(struct c4iw_dev *dev, __be32 local_ip,
314 				 __be32 peer_ip, __be16 local_port,
315 				 __be16 peer_port, u8 tos)
316 {
317 	struct rtable *rt;
318 	struct flowi4 fl4;
319 
320 	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
321 				   peer_port, local_port, IPPROTO_TCP,
322 				   tos, 0);
323 	if (IS_ERR(rt))
324 		return NULL;
325 	return rt;
326 }
327 
328 static void arp_failure_discard(void *handle, struct sk_buff *skb)
329 {
330 	PDBG("%s c4iw_dev %p\n", __func__, handle);
331 	kfree_skb(skb);
332 }
333 
334 /*
335  * Handle an ARP failure for an active open.
336  */
337 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
338 {
339 	printk(KERN_ERR MOD "ARP failure duing connect\n");
340 	kfree_skb(skb);
341 }
342 
343 /*
344  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
345  * and send it along.
346  */
347 static void abort_arp_failure(void *handle, struct sk_buff *skb)
348 {
349 	struct c4iw_rdev *rdev = handle;
350 	struct cpl_abort_req *req = cplhdr(skb);
351 
352 	PDBG("%s rdev %p\n", __func__, rdev);
353 	req->cmd = CPL_ABORT_NO_RST;
354 	c4iw_ofld_send(rdev, skb);
355 }
356 
357 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
358 {
359 	unsigned int flowclen = 80;
360 	struct fw_flowc_wr *flowc;
361 	int i;
362 
363 	skb = get_skb(skb, flowclen, GFP_KERNEL);
364 	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
365 
366 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
367 					   FW_FLOWC_WR_NPARAMS(8));
368 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
369 					  16)) | FW_WR_FLOWID(ep->hwtid));
370 
371 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
372 	flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
373 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
374 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
375 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
376 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
377 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
378 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
379 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
380 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
381 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
382 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
383 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
384 	flowc->mnemval[6].val = cpu_to_be32(snd_win);
385 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
386 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
387 	/* Pad WR to 16 byte boundary */
388 	flowc->mnemval[8].mnemonic = 0;
389 	flowc->mnemval[8].val = 0;
390 	for (i = 0; i < 9; i++) {
391 		flowc->mnemval[i].r4[0] = 0;
392 		flowc->mnemval[i].r4[1] = 0;
393 		flowc->mnemval[i].r4[2] = 0;
394 	}
395 
396 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
397 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
398 }
399 
400 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
401 {
402 	struct cpl_close_con_req *req;
403 	struct sk_buff *skb;
404 	int wrlen = roundup(sizeof *req, 16);
405 
406 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
407 	skb = get_skb(NULL, wrlen, gfp);
408 	if (!skb) {
409 		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
410 		return -ENOMEM;
411 	}
412 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
413 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
414 	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
415 	memset(req, 0, wrlen);
416 	INIT_TP_WR(req, ep->hwtid);
417 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
418 						    ep->hwtid));
419 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
420 }
421 
422 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
423 {
424 	struct cpl_abort_req *req;
425 	int wrlen = roundup(sizeof *req, 16);
426 
427 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
428 	skb = get_skb(skb, wrlen, gfp);
429 	if (!skb) {
430 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
431 		       __func__);
432 		return -ENOMEM;
433 	}
434 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
435 	t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
436 	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
437 	memset(req, 0, wrlen);
438 	INIT_TP_WR(req, ep->hwtid);
439 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
440 	req->cmd = CPL_ABORT_SEND_RST;
441 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
442 }
443 
444 static int send_connect(struct c4iw_ep *ep)
445 {
446 	struct cpl_act_open_req *req;
447 	struct sk_buff *skb;
448 	u64 opt0;
449 	u32 opt2;
450 	unsigned int mtu_idx;
451 	int wscale;
452 	int wrlen = roundup(sizeof *req, 16);
453 
454 	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
455 
456 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
457 	if (!skb) {
458 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
459 		       __func__);
460 		return -ENOMEM;
461 	}
462 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
463 
464 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
465 	wscale = compute_wscale(rcv_win);
466 	opt0 = KEEP_ALIVE(1) |
467 	       DELACK(1) |
468 	       WND_SCALE(wscale) |
469 	       MSS_IDX(mtu_idx) |
470 	       L2T_IDX(ep->l2t->idx) |
471 	       TX_CHAN(ep->tx_chan) |
472 	       SMAC_SEL(ep->smac_idx) |
473 	       DSCP(ep->tos) |
474 	       ULP_MODE(ULP_MODE_TCPDDP) |
475 	       RCV_BUFSIZ(rcv_win>>10);
476 	opt2 = RX_CHANNEL(0) |
477 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
478 	if (enable_tcp_timestamps)
479 		opt2 |= TSTAMPS_EN(1);
480 	if (enable_tcp_sack)
481 		opt2 |= SACK_EN(1);
482 	if (wscale && enable_tcp_window_scaling)
483 		opt2 |= WND_SCALE_EN(1);
484 	t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
485 
486 	req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
487 	INIT_TP_WR(req, 0);
488 	OPCODE_TID(req) = cpu_to_be32(
489 		MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ((ep->rss_qid<<14)|ep->atid)));
490 	req->local_port = ep->com.local_addr.sin_port;
491 	req->peer_port = ep->com.remote_addr.sin_port;
492 	req->local_ip = ep->com.local_addr.sin_addr.s_addr;
493 	req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
494 	req->opt0 = cpu_to_be64(opt0);
495 	req->params = 0;
496 	req->opt2 = cpu_to_be32(opt2);
497 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
498 }
499 
500 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb)
501 {
502 	int mpalen, wrlen;
503 	struct fw_ofld_tx_data_wr *req;
504 	struct mpa_message *mpa;
505 
506 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
507 
508 	BUG_ON(skb_cloned(skb));
509 
510 	mpalen = sizeof(*mpa) + ep->plen;
511 	wrlen = roundup(mpalen + sizeof *req, 16);
512 	skb = get_skb(skb, wrlen, GFP_KERNEL);
513 	if (!skb) {
514 		connect_reply_upcall(ep, -ENOMEM);
515 		return;
516 	}
517 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
518 
519 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
520 	memset(req, 0, wrlen);
521 	req->op_to_immdlen = cpu_to_be32(
522 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
523 		FW_WR_COMPL(1) |
524 		FW_WR_IMMDLEN(mpalen));
525 	req->flowid_len16 = cpu_to_be32(
526 		FW_WR_FLOWID(ep->hwtid) |
527 		FW_WR_LEN16(wrlen >> 4));
528 	req->plen = cpu_to_be32(mpalen);
529 	req->tunnel_to_proxy = cpu_to_be32(
530 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
531 		FW_OFLD_TX_DATA_WR_SHOVE(1));
532 
533 	mpa = (struct mpa_message *)(req + 1);
534 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
535 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
536 		     (markers_enabled ? MPA_MARKERS : 0);
537 	mpa->private_data_size = htons(ep->plen);
538 	mpa->revision = mpa_rev;
539 
540 	if (ep->plen)
541 		memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
542 
543 	/*
544 	 * Reference the mpa skb.  This ensures the data area
545 	 * will remain in memory until the hw acks the tx.
546 	 * Function fw4_ack() will deref it.
547 	 */
548 	skb_get(skb);
549 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
550 	BUG_ON(ep->mpa_skb);
551 	ep->mpa_skb = skb;
552 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
553 	start_ep_timer(ep);
554 	state_set(&ep->com, MPA_REQ_SENT);
555 	ep->mpa_attr.initiator = 1;
556 	return;
557 }
558 
559 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
560 {
561 	int mpalen, wrlen;
562 	struct fw_ofld_tx_data_wr *req;
563 	struct mpa_message *mpa;
564 	struct sk_buff *skb;
565 
566 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
567 
568 	mpalen = sizeof(*mpa) + plen;
569 	wrlen = roundup(mpalen + sizeof *req, 16);
570 
571 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
572 	if (!skb) {
573 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
574 		return -ENOMEM;
575 	}
576 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
577 
578 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
579 	memset(req, 0, wrlen);
580 	req->op_to_immdlen = cpu_to_be32(
581 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
582 		FW_WR_COMPL(1) |
583 		FW_WR_IMMDLEN(mpalen));
584 	req->flowid_len16 = cpu_to_be32(
585 		FW_WR_FLOWID(ep->hwtid) |
586 		FW_WR_LEN16(wrlen >> 4));
587 	req->plen = cpu_to_be32(mpalen);
588 	req->tunnel_to_proxy = cpu_to_be32(
589 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
590 		FW_OFLD_TX_DATA_WR_SHOVE(1));
591 
592 	mpa = (struct mpa_message *)(req + 1);
593 	memset(mpa, 0, sizeof(*mpa));
594 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
595 	mpa->flags = MPA_REJECT;
596 	mpa->revision = mpa_rev;
597 	mpa->private_data_size = htons(plen);
598 	if (plen)
599 		memcpy(mpa->private_data, pdata, plen);
600 
601 	/*
602 	 * Reference the mpa skb again.  This ensures the data area
603 	 * will remain in memory until the hw acks the tx.
604 	 * Function fw4_ack() will deref it.
605 	 */
606 	skb_get(skb);
607 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
608 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
609 	BUG_ON(ep->mpa_skb);
610 	ep->mpa_skb = skb;
611 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
612 }
613 
614 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
615 {
616 	int mpalen, wrlen;
617 	struct fw_ofld_tx_data_wr *req;
618 	struct mpa_message *mpa;
619 	struct sk_buff *skb;
620 
621 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
622 
623 	mpalen = sizeof(*mpa) + plen;
624 	wrlen = roundup(mpalen + sizeof *req, 16);
625 
626 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
627 	if (!skb) {
628 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
629 		return -ENOMEM;
630 	}
631 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
632 
633 	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
634 	memset(req, 0, wrlen);
635 	req->op_to_immdlen = cpu_to_be32(
636 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
637 		FW_WR_COMPL(1) |
638 		FW_WR_IMMDLEN(mpalen));
639 	req->flowid_len16 = cpu_to_be32(
640 		FW_WR_FLOWID(ep->hwtid) |
641 		FW_WR_LEN16(wrlen >> 4));
642 	req->plen = cpu_to_be32(mpalen);
643 	req->tunnel_to_proxy = cpu_to_be32(
644 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
645 		FW_OFLD_TX_DATA_WR_SHOVE(1));
646 
647 	mpa = (struct mpa_message *)(req + 1);
648 	memset(mpa, 0, sizeof(*mpa));
649 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
650 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
651 		     (markers_enabled ? MPA_MARKERS : 0);
652 	mpa->revision = mpa_rev;
653 	mpa->private_data_size = htons(plen);
654 	if (plen)
655 		memcpy(mpa->private_data, pdata, plen);
656 
657 	/*
658 	 * Reference the mpa skb.  This ensures the data area
659 	 * will remain in memory until the hw acks the tx.
660 	 * Function fw4_ack() will deref it.
661 	 */
662 	skb_get(skb);
663 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
664 	ep->mpa_skb = skb;
665 	state_set(&ep->com, MPA_REP_SENT);
666 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
667 }
668 
669 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
670 {
671 	struct c4iw_ep *ep;
672 	struct cpl_act_establish *req = cplhdr(skb);
673 	unsigned int tid = GET_TID(req);
674 	unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
675 	struct tid_info *t = dev->rdev.lldi.tids;
676 
677 	ep = lookup_atid(t, atid);
678 
679 	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
680 	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
681 
682 	dst_confirm(ep->dst);
683 
684 	/* setup the hwtid for this connection */
685 	ep->hwtid = tid;
686 	cxgb4_insert_tid(t, ep, tid);
687 
688 	ep->snd_seq = be32_to_cpu(req->snd_isn);
689 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
690 
691 	set_emss(ep, ntohs(req->tcp_opt));
692 
693 	/* dealloc the atid */
694 	cxgb4_free_atid(t, atid);
695 
696 	/* start MPA negotiation */
697 	send_flowc(ep, NULL);
698 	send_mpa_req(ep, skb);
699 
700 	return 0;
701 }
702 
703 static void close_complete_upcall(struct c4iw_ep *ep)
704 {
705 	struct iw_cm_event event;
706 
707 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
708 	memset(&event, 0, sizeof(event));
709 	event.event = IW_CM_EVENT_CLOSE;
710 	if (ep->com.cm_id) {
711 		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
712 		     ep, ep->com.cm_id, ep->hwtid);
713 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
714 		ep->com.cm_id->rem_ref(ep->com.cm_id);
715 		ep->com.cm_id = NULL;
716 		ep->com.qp = NULL;
717 	}
718 }
719 
720 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
721 {
722 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
723 	close_complete_upcall(ep);
724 	state_set(&ep->com, ABORTING);
725 	return send_abort(ep, skb, gfp);
726 }
727 
728 static void peer_close_upcall(struct c4iw_ep *ep)
729 {
730 	struct iw_cm_event event;
731 
732 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
733 	memset(&event, 0, sizeof(event));
734 	event.event = IW_CM_EVENT_DISCONNECT;
735 	if (ep->com.cm_id) {
736 		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
737 		     ep, ep->com.cm_id, ep->hwtid);
738 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
739 	}
740 }
741 
742 static void peer_abort_upcall(struct c4iw_ep *ep)
743 {
744 	struct iw_cm_event event;
745 
746 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
747 	memset(&event, 0, sizeof(event));
748 	event.event = IW_CM_EVENT_CLOSE;
749 	event.status = -ECONNRESET;
750 	if (ep->com.cm_id) {
751 		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
752 		     ep->com.cm_id, ep->hwtid);
753 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
754 		ep->com.cm_id->rem_ref(ep->com.cm_id);
755 		ep->com.cm_id = NULL;
756 		ep->com.qp = NULL;
757 	}
758 }
759 
760 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
761 {
762 	struct iw_cm_event event;
763 
764 	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
765 	memset(&event, 0, sizeof(event));
766 	event.event = IW_CM_EVENT_CONNECT_REPLY;
767 	event.status = status;
768 	event.local_addr = ep->com.local_addr;
769 	event.remote_addr = ep->com.remote_addr;
770 
771 	if ((status == 0) || (status == -ECONNREFUSED)) {
772 		event.private_data_len = ep->plen;
773 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
774 	}
775 
776 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
777 	     ep->hwtid, status);
778 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
779 
780 	if (status < 0) {
781 		ep->com.cm_id->rem_ref(ep->com.cm_id);
782 		ep->com.cm_id = NULL;
783 		ep->com.qp = NULL;
784 	}
785 }
786 
787 static void connect_request_upcall(struct c4iw_ep *ep)
788 {
789 	struct iw_cm_event event;
790 
791 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
792 	memset(&event, 0, sizeof(event));
793 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
794 	event.local_addr = ep->com.local_addr;
795 	event.remote_addr = ep->com.remote_addr;
796 	event.private_data_len = ep->plen;
797 	event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
798 	event.provider_data = ep;
799 	if (state_read(&ep->parent_ep->com) != DEAD) {
800 		c4iw_get_ep(&ep->com);
801 		ep->parent_ep->com.cm_id->event_handler(
802 						ep->parent_ep->com.cm_id,
803 						&event);
804 	}
805 	c4iw_put_ep(&ep->parent_ep->com);
806 	ep->parent_ep = NULL;
807 }
808 
809 static void established_upcall(struct c4iw_ep *ep)
810 {
811 	struct iw_cm_event event;
812 
813 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
814 	memset(&event, 0, sizeof(event));
815 	event.event = IW_CM_EVENT_ESTABLISHED;
816 	if (ep->com.cm_id) {
817 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
818 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
819 	}
820 }
821 
822 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
823 {
824 	struct cpl_rx_data_ack *req;
825 	struct sk_buff *skb;
826 	int wrlen = roundup(sizeof *req, 16);
827 
828 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
829 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
830 	if (!skb) {
831 		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
832 		return 0;
833 	}
834 
835 	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
836 	memset(req, 0, wrlen);
837 	INIT_TP_WR(req, ep->hwtid);
838 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
839 						    ep->hwtid));
840 	req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
841 				       F_RX_DACK_CHANGE |
842 				       V_RX_DACK_MODE(dack_mode));
843 	set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
844 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
845 	return credits;
846 }
847 
848 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
849 {
850 	struct mpa_message *mpa;
851 	u16 plen;
852 	struct c4iw_qp_attributes attrs;
853 	enum c4iw_qp_attr_mask mask;
854 	int err;
855 
856 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
857 
858 	/*
859 	 * Stop mpa timer.  If it expired, then the state has
860 	 * changed and we bail since ep_timeout already aborted
861 	 * the connection.
862 	 */
863 	stop_ep_timer(ep);
864 	if (state_read(&ep->com) != MPA_REQ_SENT)
865 		return;
866 
867 	/*
868 	 * If we get more than the supported amount of private data
869 	 * then we must fail this connection.
870 	 */
871 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
872 		err = -EINVAL;
873 		goto err;
874 	}
875 
876 	/*
877 	 * copy the new data into our accumulation buffer.
878 	 */
879 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
880 				  skb->len);
881 	ep->mpa_pkt_len += skb->len;
882 
883 	/*
884 	 * if we don't even have the mpa message, then bail.
885 	 */
886 	if (ep->mpa_pkt_len < sizeof(*mpa))
887 		return;
888 	mpa = (struct mpa_message *) ep->mpa_pkt;
889 
890 	/* Validate MPA header. */
891 	if (mpa->revision != mpa_rev) {
892 		err = -EPROTO;
893 		goto err;
894 	}
895 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
896 		err = -EPROTO;
897 		goto err;
898 	}
899 
900 	plen = ntohs(mpa->private_data_size);
901 
902 	/*
903 	 * Fail if there's too much private data.
904 	 */
905 	if (plen > MPA_MAX_PRIVATE_DATA) {
906 		err = -EPROTO;
907 		goto err;
908 	}
909 
910 	/*
911 	 * If plen does not account for pkt size
912 	 */
913 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
914 		err = -EPROTO;
915 		goto err;
916 	}
917 
918 	ep->plen = (u8) plen;
919 
920 	/*
921 	 * If we don't have all the pdata yet, then bail.
922 	 * We'll continue process when more data arrives.
923 	 */
924 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
925 		return;
926 
927 	if (mpa->flags & MPA_REJECT) {
928 		err = -ECONNREFUSED;
929 		goto err;
930 	}
931 
932 	/*
933 	 * If we get here we have accumulated the entire mpa
934 	 * start reply message including private data. And
935 	 * the MPA header is valid.
936 	 */
937 	state_set(&ep->com, FPDU_MODE);
938 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
939 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
940 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
941 	ep->mpa_attr.version = mpa_rev;
942 	ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
943 					    FW_RI_INIT_P2PTYPE_DISABLED;
944 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
945 	     "xmit_marker_enabled=%d, version=%d\n", __func__,
946 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
947 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
948 
949 	attrs.mpa_attr = ep->mpa_attr;
950 	attrs.max_ird = ep->ird;
951 	attrs.max_ord = ep->ord;
952 	attrs.llp_stream_handle = ep;
953 	attrs.next_state = C4IW_QP_STATE_RTS;
954 
955 	mask = C4IW_QP_ATTR_NEXT_STATE |
956 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
957 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
958 
959 	/* bind QP and TID with INIT_WR */
960 	err = c4iw_modify_qp(ep->com.qp->rhp,
961 			     ep->com.qp, mask, &attrs, 1);
962 	if (err)
963 		goto err;
964 	goto out;
965 err:
966 	state_set(&ep->com, ABORTING);
967 	send_abort(ep, skb, GFP_KERNEL);
968 out:
969 	connect_reply_upcall(ep, err);
970 	return;
971 }
972 
973 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
974 {
975 	struct mpa_message *mpa;
976 	u16 plen;
977 
978 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
979 
980 	if (state_read(&ep->com) != MPA_REQ_WAIT)
981 		return;
982 
983 	/*
984 	 * If we get more than the supported amount of private data
985 	 * then we must fail this connection.
986 	 */
987 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
988 		stop_ep_timer(ep);
989 		abort_connection(ep, skb, GFP_KERNEL);
990 		return;
991 	}
992 
993 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
994 
995 	/*
996 	 * Copy the new data into our accumulation buffer.
997 	 */
998 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
999 				  skb->len);
1000 	ep->mpa_pkt_len += skb->len;
1001 
1002 	/*
1003 	 * If we don't even have the mpa message, then bail.
1004 	 * We'll continue process when more data arrives.
1005 	 */
1006 	if (ep->mpa_pkt_len < sizeof(*mpa))
1007 		return;
1008 
1009 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1010 	stop_ep_timer(ep);
1011 	mpa = (struct mpa_message *) ep->mpa_pkt;
1012 
1013 	/*
1014 	 * Validate MPA Header.
1015 	 */
1016 	if (mpa->revision != mpa_rev) {
1017 		abort_connection(ep, skb, GFP_KERNEL);
1018 		return;
1019 	}
1020 
1021 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1022 		abort_connection(ep, skb, GFP_KERNEL);
1023 		return;
1024 	}
1025 
1026 	plen = ntohs(mpa->private_data_size);
1027 
1028 	/*
1029 	 * Fail if there's too much private data.
1030 	 */
1031 	if (plen > MPA_MAX_PRIVATE_DATA) {
1032 		abort_connection(ep, skb, GFP_KERNEL);
1033 		return;
1034 	}
1035 
1036 	/*
1037 	 * If plen does not account for pkt size
1038 	 */
1039 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1040 		abort_connection(ep, skb, GFP_KERNEL);
1041 		return;
1042 	}
1043 	ep->plen = (u8) plen;
1044 
1045 	/*
1046 	 * If we don't have all the pdata yet, then bail.
1047 	 */
1048 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1049 		return;
1050 
1051 	/*
1052 	 * If we get here we have accumulated the entire mpa
1053 	 * start reply message including private data.
1054 	 */
1055 	ep->mpa_attr.initiator = 0;
1056 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1057 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1058 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1059 	ep->mpa_attr.version = mpa_rev;
1060 	ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
1061 					    FW_RI_INIT_P2PTYPE_DISABLED;
1062 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1063 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1064 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1065 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1066 	     ep->mpa_attr.p2p_type);
1067 
1068 	state_set(&ep->com, MPA_REQ_RCVD);
1069 
1070 	/* drive upcall */
1071 	connect_request_upcall(ep);
1072 	return;
1073 }
1074 
1075 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1076 {
1077 	struct c4iw_ep *ep;
1078 	struct cpl_rx_data *hdr = cplhdr(skb);
1079 	unsigned int dlen = ntohs(hdr->len);
1080 	unsigned int tid = GET_TID(hdr);
1081 	struct tid_info *t = dev->rdev.lldi.tids;
1082 
1083 	ep = lookup_tid(t, tid);
1084 	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1085 	skb_pull(skb, sizeof(*hdr));
1086 	skb_trim(skb, dlen);
1087 
1088 	ep->rcv_seq += dlen;
1089 	BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1090 
1091 	/* update RX credits */
1092 	update_rx_credits(ep, dlen);
1093 
1094 	switch (state_read(&ep->com)) {
1095 	case MPA_REQ_SENT:
1096 		process_mpa_reply(ep, skb);
1097 		break;
1098 	case MPA_REQ_WAIT:
1099 		process_mpa_request(ep, skb);
1100 		break;
1101 	case MPA_REP_SENT:
1102 		break;
1103 	default:
1104 		printk(KERN_ERR MOD "%s Unexpected streaming data."
1105 		       " ep %p state %d tid %u\n",
1106 		       __func__, ep, state_read(&ep->com), ep->hwtid);
1107 
1108 		/*
1109 		 * The ep will timeout and inform the ULP of the failure.
1110 		 * See ep_timeout().
1111 		 */
1112 		break;
1113 	}
1114 	return 0;
1115 }
1116 
1117 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1118 {
1119 	struct c4iw_ep *ep;
1120 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1121 	int release = 0;
1122 	unsigned int tid = GET_TID(rpl);
1123 	struct tid_info *t = dev->rdev.lldi.tids;
1124 
1125 	ep = lookup_tid(t, tid);
1126 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1127 	BUG_ON(!ep);
1128 	mutex_lock(&ep->com.mutex);
1129 	switch (ep->com.state) {
1130 	case ABORTING:
1131 		__state_set(&ep->com, DEAD);
1132 		release = 1;
1133 		break;
1134 	default:
1135 		printk(KERN_ERR "%s ep %p state %d\n",
1136 		     __func__, ep, ep->com.state);
1137 		break;
1138 	}
1139 	mutex_unlock(&ep->com.mutex);
1140 
1141 	if (release)
1142 		release_ep_resources(ep);
1143 	return 0;
1144 }
1145 
1146 /*
1147  * Return whether a failed active open has allocated a TID
1148  */
1149 static inline int act_open_has_tid(int status)
1150 {
1151 	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1152 	       status != CPL_ERR_ARP_MISS;
1153 }
1154 
1155 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1156 {
1157 	struct c4iw_ep *ep;
1158 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
1159 	unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1160 					ntohl(rpl->atid_status)));
1161 	struct tid_info *t = dev->rdev.lldi.tids;
1162 	int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1163 
1164 	ep = lookup_atid(t, atid);
1165 
1166 	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1167 	     status, status2errno(status));
1168 
1169 	if (status == CPL_ERR_RTX_NEG_ADVICE) {
1170 		printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1171 			atid);
1172 		return 0;
1173 	}
1174 
1175 	connect_reply_upcall(ep, status2errno(status));
1176 	state_set(&ep->com, DEAD);
1177 
1178 	if (status && act_open_has_tid(status))
1179 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1180 
1181 	cxgb4_free_atid(t, atid);
1182 	dst_release(ep->dst);
1183 	cxgb4_l2t_release(ep->l2t);
1184 	c4iw_put_ep(&ep->com);
1185 
1186 	return 0;
1187 }
1188 
1189 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1190 {
1191 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1192 	struct tid_info *t = dev->rdev.lldi.tids;
1193 	unsigned int stid = GET_TID(rpl);
1194 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1195 
1196 	if (!ep) {
1197 		printk(KERN_ERR MOD "stid %d lookup failure!\n", stid);
1198 		return 0;
1199 	}
1200 	PDBG("%s ep %p status %d error %d\n", __func__, ep,
1201 	     rpl->status, status2errno(rpl->status));
1202 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1203 
1204 	return 0;
1205 }
1206 
1207 static int listen_stop(struct c4iw_listen_ep *ep)
1208 {
1209 	struct sk_buff *skb;
1210 	struct cpl_close_listsvr_req *req;
1211 
1212 	PDBG("%s ep %p\n", __func__, ep);
1213 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1214 	if (!skb) {
1215 		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1216 		return -ENOMEM;
1217 	}
1218 	req = (struct cpl_close_listsvr_req *) skb_put(skb, sizeof(*req));
1219 	INIT_TP_WR(req, 0);
1220 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ,
1221 						    ep->stid));
1222 	req->reply_ctrl = cpu_to_be16(
1223 			  QUEUENO(ep->com.dev->rdev.lldi.rxq_ids[0]));
1224 	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
1225 	return c4iw_ofld_send(&ep->com.dev->rdev, skb);
1226 }
1227 
1228 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1229 {
1230 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1231 	struct tid_info *t = dev->rdev.lldi.tids;
1232 	unsigned int stid = GET_TID(rpl);
1233 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1234 
1235 	PDBG("%s ep %p\n", __func__, ep);
1236 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1237 	return 0;
1238 }
1239 
1240 static void accept_cr(struct c4iw_ep *ep, __be32 peer_ip, struct sk_buff *skb,
1241 		      struct cpl_pass_accept_req *req)
1242 {
1243 	struct cpl_pass_accept_rpl *rpl;
1244 	unsigned int mtu_idx;
1245 	u64 opt0;
1246 	u32 opt2;
1247 	int wscale;
1248 
1249 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1250 	BUG_ON(skb_cloned(skb));
1251 	skb_trim(skb, sizeof(*rpl));
1252 	skb_get(skb);
1253 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1254 	wscale = compute_wscale(rcv_win);
1255 	opt0 = KEEP_ALIVE(1) |
1256 	       DELACK(1) |
1257 	       WND_SCALE(wscale) |
1258 	       MSS_IDX(mtu_idx) |
1259 	       L2T_IDX(ep->l2t->idx) |
1260 	       TX_CHAN(ep->tx_chan) |
1261 	       SMAC_SEL(ep->smac_idx) |
1262 	       DSCP(ep->tos) |
1263 	       ULP_MODE(ULP_MODE_TCPDDP) |
1264 	       RCV_BUFSIZ(rcv_win>>10);
1265 	opt2 = RX_CHANNEL(0) |
1266 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1267 
1268 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
1269 		opt2 |= TSTAMPS_EN(1);
1270 	if (enable_tcp_sack && req->tcpopt.sack)
1271 		opt2 |= SACK_EN(1);
1272 	if (wscale && enable_tcp_window_scaling)
1273 		opt2 |= WND_SCALE_EN(1);
1274 
1275 	rpl = cplhdr(skb);
1276 	INIT_TP_WR(rpl, ep->hwtid);
1277 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1278 				      ep->hwtid));
1279 	rpl->opt0 = cpu_to_be64(opt0);
1280 	rpl->opt2 = cpu_to_be32(opt2);
1281 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
1282 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1283 
1284 	return;
1285 }
1286 
1287 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, __be32 peer_ip,
1288 		      struct sk_buff *skb)
1289 {
1290 	PDBG("%s c4iw_dev %p tid %u peer_ip %x\n", __func__, dev, hwtid,
1291 	     peer_ip);
1292 	BUG_ON(skb_cloned(skb));
1293 	skb_trim(skb, sizeof(struct cpl_tid_release));
1294 	skb_get(skb);
1295 	release_tid(&dev->rdev, hwtid, skb);
1296 	return;
1297 }
1298 
1299 static void get_4tuple(struct cpl_pass_accept_req *req,
1300 		       __be32 *local_ip, __be32 *peer_ip,
1301 		       __be16 *local_port, __be16 *peer_port)
1302 {
1303 	int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
1304 	int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
1305 	struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
1306 	struct tcphdr *tcp = (struct tcphdr *)
1307 			     ((u8 *)(req + 1) + eth_len + ip_len);
1308 
1309 	PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
1310 	     ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
1311 	     ntohs(tcp->dest));
1312 
1313 	*peer_ip = ip->saddr;
1314 	*local_ip = ip->daddr;
1315 	*peer_port = tcp->source;
1316 	*local_port = tcp->dest;
1317 
1318 	return;
1319 }
1320 
1321 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1322 {
1323 	struct c4iw_ep *child_ep, *parent_ep;
1324 	struct cpl_pass_accept_req *req = cplhdr(skb);
1325 	unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
1326 	struct tid_info *t = dev->rdev.lldi.tids;
1327 	unsigned int hwtid = GET_TID(req);
1328 	struct neighbour *neigh;
1329 	struct dst_entry *dst;
1330 	struct l2t_entry *l2t;
1331 	struct rtable *rt;
1332 	__be32 local_ip, peer_ip;
1333 	__be16 local_port, peer_port;
1334 	struct net_device *pdev;
1335 	u32 tx_chan, smac_idx;
1336 	u16 rss_qid;
1337 	u32 mtu;
1338 	int step;
1339 	int txq_idx, ctrlq_idx;
1340 
1341 	parent_ep = lookup_stid(t, stid);
1342 	PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1343 
1344 	get_4tuple(req, &local_ip, &peer_ip, &local_port, &peer_port);
1345 
1346 	if (state_read(&parent_ep->com) != LISTEN) {
1347 		printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1348 		       __func__);
1349 		goto reject;
1350 	}
1351 
1352 	/* Find output route */
1353 	rt = find_route(dev, local_ip, peer_ip, local_port, peer_port,
1354 			GET_POPEN_TOS(ntohl(req->tos_stid)));
1355 	if (!rt) {
1356 		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1357 		       __func__);
1358 		goto reject;
1359 	}
1360 	dst = &rt->dst;
1361 	neigh = dst_get_neighbour(dst);
1362 	if (neigh->dev->flags & IFF_LOOPBACK) {
1363 		pdev = ip_dev_find(&init_net, peer_ip);
1364 		BUG_ON(!pdev);
1365 		l2t = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, pdev, 0);
1366 		mtu = pdev->mtu;
1367 		tx_chan = cxgb4_port_chan(pdev);
1368 		smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1369 		step = dev->rdev.lldi.ntxq / dev->rdev.lldi.nchan;
1370 		txq_idx = cxgb4_port_idx(pdev) * step;
1371 		ctrlq_idx = cxgb4_port_idx(pdev);
1372 		step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
1373 		rss_qid = dev->rdev.lldi.rxq_ids[cxgb4_port_idx(pdev) * step];
1374 		dev_put(pdev);
1375 	} else {
1376 		l2t = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, neigh->dev, 0);
1377 		mtu = dst_mtu(dst);
1378 		tx_chan = cxgb4_port_chan(neigh->dev);
1379 		smac_idx = (cxgb4_port_viid(neigh->dev) & 0x7F) << 1;
1380 		step = dev->rdev.lldi.ntxq / dev->rdev.lldi.nchan;
1381 		txq_idx = cxgb4_port_idx(neigh->dev) * step;
1382 		ctrlq_idx = cxgb4_port_idx(neigh->dev);
1383 		step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
1384 		rss_qid = dev->rdev.lldi.rxq_ids[
1385 			  cxgb4_port_idx(neigh->dev) * step];
1386 	}
1387 	if (!l2t) {
1388 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1389 		       __func__);
1390 		dst_release(dst);
1391 		goto reject;
1392 	}
1393 
1394 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1395 	if (!child_ep) {
1396 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1397 		       __func__);
1398 		cxgb4_l2t_release(l2t);
1399 		dst_release(dst);
1400 		goto reject;
1401 	}
1402 	state_set(&child_ep->com, CONNECTING);
1403 	child_ep->com.dev = dev;
1404 	child_ep->com.cm_id = NULL;
1405 	child_ep->com.local_addr.sin_family = PF_INET;
1406 	child_ep->com.local_addr.sin_port = local_port;
1407 	child_ep->com.local_addr.sin_addr.s_addr = local_ip;
1408 	child_ep->com.remote_addr.sin_family = PF_INET;
1409 	child_ep->com.remote_addr.sin_port = peer_port;
1410 	child_ep->com.remote_addr.sin_addr.s_addr = peer_ip;
1411 	c4iw_get_ep(&parent_ep->com);
1412 	child_ep->parent_ep = parent_ep;
1413 	child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
1414 	child_ep->l2t = l2t;
1415 	child_ep->dst = dst;
1416 	child_ep->hwtid = hwtid;
1417 	child_ep->tx_chan = tx_chan;
1418 	child_ep->smac_idx = smac_idx;
1419 	child_ep->rss_qid = rss_qid;
1420 	child_ep->mtu = mtu;
1421 	child_ep->txq_idx = txq_idx;
1422 	child_ep->ctrlq_idx = ctrlq_idx;
1423 
1424 	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
1425 	     tx_chan, smac_idx, rss_qid);
1426 
1427 	init_timer(&child_ep->timer);
1428 	cxgb4_insert_tid(t, child_ep, hwtid);
1429 	accept_cr(child_ep, peer_ip, skb, req);
1430 	goto out;
1431 reject:
1432 	reject_cr(dev, hwtid, peer_ip, skb);
1433 out:
1434 	return 0;
1435 }
1436 
1437 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1438 {
1439 	struct c4iw_ep *ep;
1440 	struct cpl_pass_establish *req = cplhdr(skb);
1441 	struct tid_info *t = dev->rdev.lldi.tids;
1442 	unsigned int tid = GET_TID(req);
1443 
1444 	ep = lookup_tid(t, tid);
1445 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1446 	ep->snd_seq = be32_to_cpu(req->snd_isn);
1447 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1448 
1449 	set_emss(ep, ntohs(req->tcp_opt));
1450 
1451 	dst_confirm(ep->dst);
1452 	state_set(&ep->com, MPA_REQ_WAIT);
1453 	start_ep_timer(ep);
1454 	send_flowc(ep, skb);
1455 
1456 	return 0;
1457 }
1458 
1459 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
1460 {
1461 	struct cpl_peer_close *hdr = cplhdr(skb);
1462 	struct c4iw_ep *ep;
1463 	struct c4iw_qp_attributes attrs;
1464 	int disconnect = 1;
1465 	int release = 0;
1466 	struct tid_info *t = dev->rdev.lldi.tids;
1467 	unsigned int tid = GET_TID(hdr);
1468 	int ret;
1469 
1470 	ep = lookup_tid(t, tid);
1471 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1472 	dst_confirm(ep->dst);
1473 
1474 	mutex_lock(&ep->com.mutex);
1475 	switch (ep->com.state) {
1476 	case MPA_REQ_WAIT:
1477 		__state_set(&ep->com, CLOSING);
1478 		break;
1479 	case MPA_REQ_SENT:
1480 		__state_set(&ep->com, CLOSING);
1481 		connect_reply_upcall(ep, -ECONNRESET);
1482 		break;
1483 	case MPA_REQ_RCVD:
1484 
1485 		/*
1486 		 * We're gonna mark this puppy DEAD, but keep
1487 		 * the reference on it until the ULP accepts or
1488 		 * rejects the CR. Also wake up anyone waiting
1489 		 * in rdma connection migration (see c4iw_accept_cr()).
1490 		 */
1491 		__state_set(&ep->com, CLOSING);
1492 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1493 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1494 		break;
1495 	case MPA_REP_SENT:
1496 		__state_set(&ep->com, CLOSING);
1497 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1498 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1499 		break;
1500 	case FPDU_MODE:
1501 		start_ep_timer(ep);
1502 		__state_set(&ep->com, CLOSING);
1503 		attrs.next_state = C4IW_QP_STATE_CLOSING;
1504 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1505 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1506 		if (ret != -ECONNRESET) {
1507 			peer_close_upcall(ep);
1508 			disconnect = 1;
1509 		}
1510 		break;
1511 	case ABORTING:
1512 		disconnect = 0;
1513 		break;
1514 	case CLOSING:
1515 		__state_set(&ep->com, MORIBUND);
1516 		disconnect = 0;
1517 		break;
1518 	case MORIBUND:
1519 		stop_ep_timer(ep);
1520 		if (ep->com.cm_id && ep->com.qp) {
1521 			attrs.next_state = C4IW_QP_STATE_IDLE;
1522 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1523 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1524 		}
1525 		close_complete_upcall(ep);
1526 		__state_set(&ep->com, DEAD);
1527 		release = 1;
1528 		disconnect = 0;
1529 		break;
1530 	case DEAD:
1531 		disconnect = 0;
1532 		break;
1533 	default:
1534 		BUG_ON(1);
1535 	}
1536 	mutex_unlock(&ep->com.mutex);
1537 	if (disconnect)
1538 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1539 	if (release)
1540 		release_ep_resources(ep);
1541 	return 0;
1542 }
1543 
1544 /*
1545  * Returns whether an ABORT_REQ_RSS message is a negative advice.
1546  */
1547 static int is_neg_adv_abort(unsigned int status)
1548 {
1549 	return status == CPL_ERR_RTX_NEG_ADVICE ||
1550 	       status == CPL_ERR_PERSIST_NEG_ADVICE;
1551 }
1552 
1553 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1554 {
1555 	struct cpl_abort_req_rss *req = cplhdr(skb);
1556 	struct c4iw_ep *ep;
1557 	struct cpl_abort_rpl *rpl;
1558 	struct sk_buff *rpl_skb;
1559 	struct c4iw_qp_attributes attrs;
1560 	int ret;
1561 	int release = 0;
1562 	struct tid_info *t = dev->rdev.lldi.tids;
1563 	unsigned int tid = GET_TID(req);
1564 
1565 	ep = lookup_tid(t, tid);
1566 	if (is_neg_adv_abort(req->status)) {
1567 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
1568 		     ep->hwtid);
1569 		return 0;
1570 	}
1571 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
1572 	     ep->com.state);
1573 
1574 	/*
1575 	 * Wake up any threads in rdma_init() or rdma_fini().
1576 	 */
1577 	c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1578 
1579 	mutex_lock(&ep->com.mutex);
1580 	switch (ep->com.state) {
1581 	case CONNECTING:
1582 		break;
1583 	case MPA_REQ_WAIT:
1584 		stop_ep_timer(ep);
1585 		break;
1586 	case MPA_REQ_SENT:
1587 		stop_ep_timer(ep);
1588 		connect_reply_upcall(ep, -ECONNRESET);
1589 		break;
1590 	case MPA_REP_SENT:
1591 		break;
1592 	case MPA_REQ_RCVD:
1593 		break;
1594 	case MORIBUND:
1595 	case CLOSING:
1596 		stop_ep_timer(ep);
1597 		/*FALLTHROUGH*/
1598 	case FPDU_MODE:
1599 		if (ep->com.cm_id && ep->com.qp) {
1600 			attrs.next_state = C4IW_QP_STATE_ERROR;
1601 			ret = c4iw_modify_qp(ep->com.qp->rhp,
1602 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
1603 				     &attrs, 1);
1604 			if (ret)
1605 				printk(KERN_ERR MOD
1606 				       "%s - qp <- error failed!\n",
1607 				       __func__);
1608 		}
1609 		peer_abort_upcall(ep);
1610 		break;
1611 	case ABORTING:
1612 		break;
1613 	case DEAD:
1614 		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1615 		mutex_unlock(&ep->com.mutex);
1616 		return 0;
1617 	default:
1618 		BUG_ON(1);
1619 		break;
1620 	}
1621 	dst_confirm(ep->dst);
1622 	if (ep->com.state != ABORTING) {
1623 		__state_set(&ep->com, DEAD);
1624 		release = 1;
1625 	}
1626 	mutex_unlock(&ep->com.mutex);
1627 
1628 	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1629 	if (!rpl_skb) {
1630 		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1631 		       __func__);
1632 		release = 1;
1633 		goto out;
1634 	}
1635 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1636 	rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1637 	INIT_TP_WR(rpl, ep->hwtid);
1638 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1639 	rpl->cmd = CPL_ABORT_NO_RST;
1640 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
1641 out:
1642 	if (release)
1643 		release_ep_resources(ep);
1644 	return 0;
1645 }
1646 
1647 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1648 {
1649 	struct c4iw_ep *ep;
1650 	struct c4iw_qp_attributes attrs;
1651 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
1652 	int release = 0;
1653 	struct tid_info *t = dev->rdev.lldi.tids;
1654 	unsigned int tid = GET_TID(rpl);
1655 
1656 	ep = lookup_tid(t, tid);
1657 
1658 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1659 	BUG_ON(!ep);
1660 
1661 	/* The cm_id may be null if we failed to connect */
1662 	mutex_lock(&ep->com.mutex);
1663 	switch (ep->com.state) {
1664 	case CLOSING:
1665 		__state_set(&ep->com, MORIBUND);
1666 		break;
1667 	case MORIBUND:
1668 		stop_ep_timer(ep);
1669 		if ((ep->com.cm_id) && (ep->com.qp)) {
1670 			attrs.next_state = C4IW_QP_STATE_IDLE;
1671 			c4iw_modify_qp(ep->com.qp->rhp,
1672 					     ep->com.qp,
1673 					     C4IW_QP_ATTR_NEXT_STATE,
1674 					     &attrs, 1);
1675 		}
1676 		close_complete_upcall(ep);
1677 		__state_set(&ep->com, DEAD);
1678 		release = 1;
1679 		break;
1680 	case ABORTING:
1681 	case DEAD:
1682 		break;
1683 	default:
1684 		BUG_ON(1);
1685 		break;
1686 	}
1687 	mutex_unlock(&ep->com.mutex);
1688 	if (release)
1689 		release_ep_resources(ep);
1690 	return 0;
1691 }
1692 
1693 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
1694 {
1695 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
1696 	struct tid_info *t = dev->rdev.lldi.tids;
1697 	unsigned int tid = GET_TID(rpl);
1698 	struct c4iw_ep *ep;
1699 	struct c4iw_qp_attributes attrs;
1700 
1701 	ep = lookup_tid(t, tid);
1702 	BUG_ON(!ep);
1703 
1704 	if (ep && ep->com.qp) {
1705 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
1706 		       ep->com.qp->wq.sq.qid);
1707 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1708 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1709 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1710 	} else
1711 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
1712 
1713 	return 0;
1714 }
1715 
1716 /*
1717  * Upcall from the adapter indicating data has been transmitted.
1718  * For us its just the single MPA request or reply.  We can now free
1719  * the skb holding the mpa message.
1720  */
1721 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
1722 {
1723 	struct c4iw_ep *ep;
1724 	struct cpl_fw4_ack *hdr = cplhdr(skb);
1725 	u8 credits = hdr->credits;
1726 	unsigned int tid = GET_TID(hdr);
1727 	struct tid_info *t = dev->rdev.lldi.tids;
1728 
1729 
1730 	ep = lookup_tid(t, tid);
1731 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1732 	if (credits == 0) {
1733 		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
1734 		     __func__, ep, ep->hwtid, state_read(&ep->com));
1735 		return 0;
1736 	}
1737 
1738 	dst_confirm(ep->dst);
1739 	if (ep->mpa_skb) {
1740 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
1741 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
1742 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
1743 		kfree_skb(ep->mpa_skb);
1744 		ep->mpa_skb = NULL;
1745 	}
1746 	return 0;
1747 }
1748 
1749 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1750 {
1751 	int err;
1752 	struct c4iw_ep *ep = to_ep(cm_id);
1753 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1754 
1755 	if (state_read(&ep->com) == DEAD) {
1756 		c4iw_put_ep(&ep->com);
1757 		return -ECONNRESET;
1758 	}
1759 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1760 	if (mpa_rev == 0)
1761 		abort_connection(ep, NULL, GFP_KERNEL);
1762 	else {
1763 		err = send_mpa_reject(ep, pdata, pdata_len);
1764 		err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1765 	}
1766 	c4iw_put_ep(&ep->com);
1767 	return 0;
1768 }
1769 
1770 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1771 {
1772 	int err;
1773 	struct c4iw_qp_attributes attrs;
1774 	enum c4iw_qp_attr_mask mask;
1775 	struct c4iw_ep *ep = to_ep(cm_id);
1776 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1777 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1778 
1779 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1780 	if (state_read(&ep->com) == DEAD) {
1781 		err = -ECONNRESET;
1782 		goto err;
1783 	}
1784 
1785 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1786 	BUG_ON(!qp);
1787 
1788 	if ((conn_param->ord > c4iw_max_read_depth) ||
1789 	    (conn_param->ird > c4iw_max_read_depth)) {
1790 		abort_connection(ep, NULL, GFP_KERNEL);
1791 		err = -EINVAL;
1792 		goto err;
1793 	}
1794 
1795 	cm_id->add_ref(cm_id);
1796 	ep->com.cm_id = cm_id;
1797 	ep->com.qp = qp;
1798 
1799 	ep->ird = conn_param->ird;
1800 	ep->ord = conn_param->ord;
1801 
1802 	if (peer2peer && ep->ird == 0)
1803 		ep->ird = 1;
1804 
1805 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1806 
1807 	/* bind QP to EP and move to RTS */
1808 	attrs.mpa_attr = ep->mpa_attr;
1809 	attrs.max_ird = ep->ird;
1810 	attrs.max_ord = ep->ord;
1811 	attrs.llp_stream_handle = ep;
1812 	attrs.next_state = C4IW_QP_STATE_RTS;
1813 
1814 	/* bind QP and TID with INIT_WR */
1815 	mask = C4IW_QP_ATTR_NEXT_STATE |
1816 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1817 			     C4IW_QP_ATTR_MPA_ATTR |
1818 			     C4IW_QP_ATTR_MAX_IRD |
1819 			     C4IW_QP_ATTR_MAX_ORD;
1820 
1821 	err = c4iw_modify_qp(ep->com.qp->rhp,
1822 			     ep->com.qp, mask, &attrs, 1);
1823 	if (err)
1824 		goto err1;
1825 	err = send_mpa_reply(ep, conn_param->private_data,
1826 			     conn_param->private_data_len);
1827 	if (err)
1828 		goto err1;
1829 
1830 	state_set(&ep->com, FPDU_MODE);
1831 	established_upcall(ep);
1832 	c4iw_put_ep(&ep->com);
1833 	return 0;
1834 err1:
1835 	ep->com.cm_id = NULL;
1836 	ep->com.qp = NULL;
1837 	cm_id->rem_ref(cm_id);
1838 err:
1839 	c4iw_put_ep(&ep->com);
1840 	return err;
1841 }
1842 
1843 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1844 {
1845 	int err = 0;
1846 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
1847 	struct c4iw_ep *ep;
1848 	struct rtable *rt;
1849 	struct net_device *pdev;
1850 	struct neighbour *neigh;
1851 	int step;
1852 
1853 	if ((conn_param->ord > c4iw_max_read_depth) ||
1854 	    (conn_param->ird > c4iw_max_read_depth)) {
1855 		err = -EINVAL;
1856 		goto out;
1857 	}
1858 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1859 	if (!ep) {
1860 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1861 		err = -ENOMEM;
1862 		goto out;
1863 	}
1864 	init_timer(&ep->timer);
1865 	ep->plen = conn_param->private_data_len;
1866 	if (ep->plen)
1867 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1868 		       conn_param->private_data, ep->plen);
1869 	ep->ird = conn_param->ird;
1870 	ep->ord = conn_param->ord;
1871 
1872 	if (peer2peer && ep->ord == 0)
1873 		ep->ord = 1;
1874 
1875 	cm_id->add_ref(cm_id);
1876 	ep->com.dev = dev;
1877 	ep->com.cm_id = cm_id;
1878 	ep->com.qp = get_qhp(dev, conn_param->qpn);
1879 	BUG_ON(!ep->com.qp);
1880 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1881 	     ep->com.qp, cm_id);
1882 
1883 	/*
1884 	 * Allocate an active TID to initiate a TCP connection.
1885 	 */
1886 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
1887 	if (ep->atid == -1) {
1888 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1889 		err = -ENOMEM;
1890 		goto fail2;
1891 	}
1892 
1893 	PDBG("%s saddr 0x%x sport 0x%x raddr 0x%x rport 0x%x\n", __func__,
1894 	     ntohl(cm_id->local_addr.sin_addr.s_addr),
1895 	     ntohs(cm_id->local_addr.sin_port),
1896 	     ntohl(cm_id->remote_addr.sin_addr.s_addr),
1897 	     ntohs(cm_id->remote_addr.sin_port));
1898 
1899 	/* find a route */
1900 	rt = find_route(dev,
1901 			cm_id->local_addr.sin_addr.s_addr,
1902 			cm_id->remote_addr.sin_addr.s_addr,
1903 			cm_id->local_addr.sin_port,
1904 			cm_id->remote_addr.sin_port, 0);
1905 	if (!rt) {
1906 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1907 		err = -EHOSTUNREACH;
1908 		goto fail3;
1909 	}
1910 	ep->dst = &rt->dst;
1911 
1912 	neigh = dst_get_neighbour(ep->dst);
1913 
1914 	/* get a l2t entry */
1915 	if (neigh->dev->flags & IFF_LOOPBACK) {
1916 		PDBG("%s LOOPBACK\n", __func__);
1917 		pdev = ip_dev_find(&init_net,
1918 				   cm_id->remote_addr.sin_addr.s_addr);
1919 		ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1920 					neigh, pdev, 0);
1921 		ep->mtu = pdev->mtu;
1922 		ep->tx_chan = cxgb4_port_chan(pdev);
1923 		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1924 		step = ep->com.dev->rdev.lldi.ntxq /
1925 		       ep->com.dev->rdev.lldi.nchan;
1926 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1927 		step = ep->com.dev->rdev.lldi.nrxq /
1928 		       ep->com.dev->rdev.lldi.nchan;
1929 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1930 		ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1931 			      cxgb4_port_idx(pdev) * step];
1932 		dev_put(pdev);
1933 	} else {
1934 		ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1935 					neigh, neigh->dev, 0);
1936 		ep->mtu = dst_mtu(ep->dst);
1937 		ep->tx_chan = cxgb4_port_chan(neigh->dev);
1938 		ep->smac_idx = (cxgb4_port_viid(neigh->dev) & 0x7F) << 1;
1939 		step = ep->com.dev->rdev.lldi.ntxq /
1940 		       ep->com.dev->rdev.lldi.nchan;
1941 		ep->txq_idx = cxgb4_port_idx(neigh->dev) * step;
1942 		ep->ctrlq_idx = cxgb4_port_idx(neigh->dev);
1943 		step = ep->com.dev->rdev.lldi.nrxq /
1944 		       ep->com.dev->rdev.lldi.nchan;
1945 		ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1946 			      cxgb4_port_idx(neigh->dev) * step];
1947 	}
1948 	if (!ep->l2t) {
1949 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1950 		err = -ENOMEM;
1951 		goto fail4;
1952 	}
1953 
1954 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1955 		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1956 		ep->l2t->idx);
1957 
1958 	state_set(&ep->com, CONNECTING);
1959 	ep->tos = 0;
1960 	ep->com.local_addr = cm_id->local_addr;
1961 	ep->com.remote_addr = cm_id->remote_addr;
1962 
1963 	/* send connect request to rnic */
1964 	err = send_connect(ep);
1965 	if (!err)
1966 		goto out;
1967 
1968 	cxgb4_l2t_release(ep->l2t);
1969 fail4:
1970 	dst_release(ep->dst);
1971 fail3:
1972 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1973 fail2:
1974 	cm_id->rem_ref(cm_id);
1975 	c4iw_put_ep(&ep->com);
1976 out:
1977 	return err;
1978 }
1979 
1980 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
1981 {
1982 	int err = 0;
1983 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
1984 	struct c4iw_listen_ep *ep;
1985 
1986 
1987 	might_sleep();
1988 
1989 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1990 	if (!ep) {
1991 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1992 		err = -ENOMEM;
1993 		goto fail1;
1994 	}
1995 	PDBG("%s ep %p\n", __func__, ep);
1996 	cm_id->add_ref(cm_id);
1997 	ep->com.cm_id = cm_id;
1998 	ep->com.dev = dev;
1999 	ep->backlog = backlog;
2000 	ep->com.local_addr = cm_id->local_addr;
2001 
2002 	/*
2003 	 * Allocate a server TID.
2004 	 */
2005 	ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, PF_INET, ep);
2006 	if (ep->stid == -1) {
2007 		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2008 		err = -ENOMEM;
2009 		goto fail2;
2010 	}
2011 
2012 	state_set(&ep->com, LISTEN);
2013 	c4iw_init_wr_wait(&ep->com.wr_wait);
2014 	err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], ep->stid,
2015 				  ep->com.local_addr.sin_addr.s_addr,
2016 				  ep->com.local_addr.sin_port,
2017 				  ep->com.dev->rdev.lldi.rxq_ids[0]);
2018 	if (err)
2019 		goto fail3;
2020 
2021 	/* wait for pass_open_rpl */
2022 	err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0,
2023 				  __func__);
2024 	if (!err) {
2025 		cm_id->provider_data = ep;
2026 		goto out;
2027 	}
2028 fail3:
2029 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2030 fail2:
2031 	cm_id->rem_ref(cm_id);
2032 	c4iw_put_ep(&ep->com);
2033 fail1:
2034 out:
2035 	return err;
2036 }
2037 
2038 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2039 {
2040 	int err;
2041 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2042 
2043 	PDBG("%s ep %p\n", __func__, ep);
2044 
2045 	might_sleep();
2046 	state_set(&ep->com, DEAD);
2047 	c4iw_init_wr_wait(&ep->com.wr_wait);
2048 	err = listen_stop(ep);
2049 	if (err)
2050 		goto done;
2051 	err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0,
2052 				  __func__);
2053 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2054 done:
2055 	cm_id->rem_ref(cm_id);
2056 	c4iw_put_ep(&ep->com);
2057 	return err;
2058 }
2059 
2060 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2061 {
2062 	int ret = 0;
2063 	int close = 0;
2064 	int fatal = 0;
2065 	struct c4iw_rdev *rdev;
2066 
2067 	mutex_lock(&ep->com.mutex);
2068 
2069 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2070 	     states[ep->com.state], abrupt);
2071 
2072 	rdev = &ep->com.dev->rdev;
2073 	if (c4iw_fatal_error(rdev)) {
2074 		fatal = 1;
2075 		close_complete_upcall(ep);
2076 		ep->com.state = DEAD;
2077 	}
2078 	switch (ep->com.state) {
2079 	case MPA_REQ_WAIT:
2080 	case MPA_REQ_SENT:
2081 	case MPA_REQ_RCVD:
2082 	case MPA_REP_SENT:
2083 	case FPDU_MODE:
2084 		close = 1;
2085 		if (abrupt)
2086 			ep->com.state = ABORTING;
2087 		else {
2088 			ep->com.state = CLOSING;
2089 			start_ep_timer(ep);
2090 		}
2091 		set_bit(CLOSE_SENT, &ep->com.flags);
2092 		break;
2093 	case CLOSING:
2094 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2095 			close = 1;
2096 			if (abrupt) {
2097 				stop_ep_timer(ep);
2098 				ep->com.state = ABORTING;
2099 			} else
2100 				ep->com.state = MORIBUND;
2101 		}
2102 		break;
2103 	case MORIBUND:
2104 	case ABORTING:
2105 	case DEAD:
2106 		PDBG("%s ignoring disconnect ep %p state %u\n",
2107 		     __func__, ep, ep->com.state);
2108 		break;
2109 	default:
2110 		BUG();
2111 		break;
2112 	}
2113 
2114 	if (close) {
2115 		if (abrupt) {
2116 			close_complete_upcall(ep);
2117 			ret = send_abort(ep, NULL, gfp);
2118 		} else
2119 			ret = send_halfclose(ep, gfp);
2120 		if (ret)
2121 			fatal = 1;
2122 	}
2123 	mutex_unlock(&ep->com.mutex);
2124 	if (fatal)
2125 		release_ep_resources(ep);
2126 	return ret;
2127 }
2128 
2129 static int async_event(struct c4iw_dev *dev, struct sk_buff *skb)
2130 {
2131 	struct cpl_fw6_msg *rpl = cplhdr(skb);
2132 	c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
2133 	return 0;
2134 }
2135 
2136 /*
2137  * These are the real handlers that are called from a
2138  * work queue.
2139  */
2140 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
2141 	[CPL_ACT_ESTABLISH] = act_establish,
2142 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
2143 	[CPL_RX_DATA] = rx_data,
2144 	[CPL_ABORT_RPL_RSS] = abort_rpl,
2145 	[CPL_ABORT_RPL] = abort_rpl,
2146 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
2147 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2148 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2149 	[CPL_PASS_ESTABLISH] = pass_establish,
2150 	[CPL_PEER_CLOSE] = peer_close,
2151 	[CPL_ABORT_REQ_RSS] = peer_abort,
2152 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
2153 	[CPL_RDMA_TERMINATE] = terminate,
2154 	[CPL_FW4_ACK] = fw4_ack,
2155 	[CPL_FW6_MSG] = async_event
2156 };
2157 
2158 static void process_timeout(struct c4iw_ep *ep)
2159 {
2160 	struct c4iw_qp_attributes attrs;
2161 	int abort = 1;
2162 
2163 	mutex_lock(&ep->com.mutex);
2164 	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
2165 	     ep->com.state);
2166 	switch (ep->com.state) {
2167 	case MPA_REQ_SENT:
2168 		__state_set(&ep->com, ABORTING);
2169 		connect_reply_upcall(ep, -ETIMEDOUT);
2170 		break;
2171 	case MPA_REQ_WAIT:
2172 		__state_set(&ep->com, ABORTING);
2173 		break;
2174 	case CLOSING:
2175 	case MORIBUND:
2176 		if (ep->com.cm_id && ep->com.qp) {
2177 			attrs.next_state = C4IW_QP_STATE_ERROR;
2178 			c4iw_modify_qp(ep->com.qp->rhp,
2179 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2180 				     &attrs, 1);
2181 		}
2182 		__state_set(&ep->com, ABORTING);
2183 		break;
2184 	default:
2185 		printk(KERN_ERR "%s unexpected state ep %p tid %u state %u\n",
2186 			__func__, ep, ep->hwtid, ep->com.state);
2187 		WARN_ON(1);
2188 		abort = 0;
2189 	}
2190 	mutex_unlock(&ep->com.mutex);
2191 	if (abort)
2192 		abort_connection(ep, NULL, GFP_KERNEL);
2193 	c4iw_put_ep(&ep->com);
2194 }
2195 
2196 static void process_timedout_eps(void)
2197 {
2198 	struct c4iw_ep *ep;
2199 
2200 	spin_lock_irq(&timeout_lock);
2201 	while (!list_empty(&timeout_list)) {
2202 		struct list_head *tmp;
2203 
2204 		tmp = timeout_list.next;
2205 		list_del(tmp);
2206 		spin_unlock_irq(&timeout_lock);
2207 		ep = list_entry(tmp, struct c4iw_ep, entry);
2208 		process_timeout(ep);
2209 		spin_lock_irq(&timeout_lock);
2210 	}
2211 	spin_unlock_irq(&timeout_lock);
2212 }
2213 
2214 static void process_work(struct work_struct *work)
2215 {
2216 	struct sk_buff *skb = NULL;
2217 	struct c4iw_dev *dev;
2218 	struct cpl_act_establish *rpl;
2219 	unsigned int opcode;
2220 	int ret;
2221 
2222 	while ((skb = skb_dequeue(&rxq))) {
2223 		rpl = cplhdr(skb);
2224 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
2225 		opcode = rpl->ot.opcode;
2226 
2227 		BUG_ON(!work_handlers[opcode]);
2228 		ret = work_handlers[opcode](dev, skb);
2229 		if (!ret)
2230 			kfree_skb(skb);
2231 	}
2232 	process_timedout_eps();
2233 }
2234 
2235 static DECLARE_WORK(skb_work, process_work);
2236 
2237 static void ep_timeout(unsigned long arg)
2238 {
2239 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2240 
2241 	spin_lock(&timeout_lock);
2242 	list_add_tail(&ep->entry, &timeout_list);
2243 	spin_unlock(&timeout_lock);
2244 	queue_work(workq, &skb_work);
2245 }
2246 
2247 /*
2248  * All the CM events are handled on a work queue to have a safe context.
2249  */
2250 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
2251 {
2252 
2253 	/*
2254 	 * Save dev in the skb->cb area.
2255 	 */
2256 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
2257 
2258 	/*
2259 	 * Queue the skb and schedule the worker thread.
2260 	 */
2261 	skb_queue_tail(&rxq, skb);
2262 	queue_work(workq, &skb_work);
2263 	return 0;
2264 }
2265 
2266 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2267 {
2268 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2269 
2270 	if (rpl->status != CPL_ERR_NONE) {
2271 		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2272 		       "for tid %u\n", rpl->status, GET_TID(rpl));
2273 	}
2274 	kfree_skb(skb);
2275 	return 0;
2276 }
2277 
2278 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2279 {
2280 	struct cpl_fw6_msg *rpl = cplhdr(skb);
2281 	struct c4iw_wr_wait *wr_waitp;
2282 	int ret;
2283 
2284 	PDBG("%s type %u\n", __func__, rpl->type);
2285 
2286 	switch (rpl->type) {
2287 	case 1:
2288 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
2289 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
2290 		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
2291 		if (wr_waitp)
2292 			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2293 		kfree_skb(skb);
2294 		break;
2295 	case 2:
2296 		sched(dev, skb);
2297 		break;
2298 	default:
2299 		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
2300 		       rpl->type);
2301 		kfree_skb(skb);
2302 		break;
2303 	}
2304 	return 0;
2305 }
2306 
2307 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
2308 {
2309 	struct cpl_abort_req_rss *req = cplhdr(skb);
2310 	struct c4iw_ep *ep;
2311 	struct tid_info *t = dev->rdev.lldi.tids;
2312 	unsigned int tid = GET_TID(req);
2313 
2314 	ep = lookup_tid(t, tid);
2315 	if (is_neg_adv_abort(req->status)) {
2316 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2317 		     ep->hwtid);
2318 		kfree_skb(skb);
2319 		return 0;
2320 	}
2321 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2322 	     ep->com.state);
2323 
2324 	/*
2325 	 * Wake up any threads in rdma_init() or rdma_fini().
2326 	 */
2327 	c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2328 	sched(dev, skb);
2329 	return 0;
2330 }
2331 
2332 /*
2333  * Most upcalls from the T4 Core go to sched() to
2334  * schedule the processing on a work queue.
2335  */
2336 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
2337 	[CPL_ACT_ESTABLISH] = sched,
2338 	[CPL_ACT_OPEN_RPL] = sched,
2339 	[CPL_RX_DATA] = sched,
2340 	[CPL_ABORT_RPL_RSS] = sched,
2341 	[CPL_ABORT_RPL] = sched,
2342 	[CPL_PASS_OPEN_RPL] = sched,
2343 	[CPL_CLOSE_LISTSRV_RPL] = sched,
2344 	[CPL_PASS_ACCEPT_REQ] = sched,
2345 	[CPL_PASS_ESTABLISH] = sched,
2346 	[CPL_PEER_CLOSE] = sched,
2347 	[CPL_CLOSE_CON_RPL] = sched,
2348 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
2349 	[CPL_RDMA_TERMINATE] = sched,
2350 	[CPL_FW4_ACK] = sched,
2351 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
2352 	[CPL_FW6_MSG] = fw6_msg
2353 };
2354 
2355 int __init c4iw_cm_init(void)
2356 {
2357 	spin_lock_init(&timeout_lock);
2358 	skb_queue_head_init(&rxq);
2359 
2360 	workq = create_singlethread_workqueue("iw_cxgb4");
2361 	if (!workq)
2362 		return -ENOMEM;
2363 
2364 	return 0;
2365 }
2366 
2367 void __exit c4iw_cm_term(void)
2368 {
2369 	WARN_ON(!list_empty(&timeout_list));
2370 	flush_workqueue(workq);
2371 	destroy_workqueue(workq);
2372 }
2373