xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision 089a49b6)
1 /*
2  * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *	  copyright notice, this list of conditions and the following
16  *	  disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *	  copyright notice, this list of conditions and the following
20  *	  disclaimer in the documentation and/or other materials
21  *	  provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
42 
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
49 
50 #include "iw_cxgb4.h"
51 
52 static char *states[] = {
53 	"idle",
54 	"listen",
55 	"connecting",
56 	"mpa_wait_req",
57 	"mpa_req_sent",
58 	"mpa_req_rcvd",
59 	"mpa_rep_sent",
60 	"fpdu_mode",
61 	"aborting",
62 	"closing",
63 	"moribund",
64 	"dead",
65 	NULL,
66 };
67 
68 static int nocong;
69 module_param(nocong, int, 0644);
70 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
71 
72 static int enable_ecn;
73 module_param(enable_ecn, int, 0644);
74 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
75 
76 static int dack_mode = 1;
77 module_param(dack_mode, int, 0644);
78 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
79 
80 int c4iw_max_read_depth = 8;
81 module_param(c4iw_max_read_depth, int, 0644);
82 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
83 
84 static int enable_tcp_timestamps;
85 module_param(enable_tcp_timestamps, int, 0644);
86 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
87 
88 static int enable_tcp_sack;
89 module_param(enable_tcp_sack, int, 0644);
90 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
91 
92 static int enable_tcp_window_scaling = 1;
93 module_param(enable_tcp_window_scaling, int, 0644);
94 MODULE_PARM_DESC(enable_tcp_window_scaling,
95 		 "Enable tcp window scaling (default=1)");
96 
97 int c4iw_debug;
98 module_param(c4iw_debug, int, 0644);
99 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
100 
101 static int peer2peer;
102 module_param(peer2peer, int, 0644);
103 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
104 
105 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
106 module_param(p2p_type, int, 0644);
107 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
108 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
109 
110 static int ep_timeout_secs = 60;
111 module_param(ep_timeout_secs, int, 0644);
112 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
113 				   "in seconds (default=60)");
114 
115 static int mpa_rev = 1;
116 module_param(mpa_rev, int, 0644);
117 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
118 		"1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
119 		" compliant (default=1)");
120 
121 static int markers_enabled;
122 module_param(markers_enabled, int, 0644);
123 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
124 
125 static int crc_enabled = 1;
126 module_param(crc_enabled, int, 0644);
127 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
128 
129 static int rcv_win = 256 * 1024;
130 module_param(rcv_win, int, 0644);
131 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
132 
133 static int snd_win = 128 * 1024;
134 module_param(snd_win, int, 0644);
135 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
136 
137 static struct workqueue_struct *workq;
138 
139 static struct sk_buff_head rxq;
140 
141 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
142 static void ep_timeout(unsigned long arg);
143 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
144 
145 static LIST_HEAD(timeout_list);
146 static spinlock_t timeout_lock;
147 
148 static void deref_qp(struct c4iw_ep *ep)
149 {
150 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
151 	clear_bit(QP_REFERENCED, &ep->com.flags);
152 }
153 
154 static void ref_qp(struct c4iw_ep *ep)
155 {
156 	set_bit(QP_REFERENCED, &ep->com.flags);
157 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
158 }
159 
160 static void start_ep_timer(struct c4iw_ep *ep)
161 {
162 	PDBG("%s ep %p\n", __func__, ep);
163 	if (timer_pending(&ep->timer)) {
164 		pr_err("%s timer already started! ep %p\n",
165 		       __func__, ep);
166 		return;
167 	}
168 	clear_bit(TIMEOUT, &ep->com.flags);
169 	c4iw_get_ep(&ep->com);
170 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
171 	ep->timer.data = (unsigned long)ep;
172 	ep->timer.function = ep_timeout;
173 	add_timer(&ep->timer);
174 }
175 
176 static void stop_ep_timer(struct c4iw_ep *ep)
177 {
178 	PDBG("%s ep %p stopping\n", __func__, ep);
179 	del_timer_sync(&ep->timer);
180 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
181 		c4iw_put_ep(&ep->com);
182 }
183 
184 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
185 		  struct l2t_entry *l2e)
186 {
187 	int	error = 0;
188 
189 	if (c4iw_fatal_error(rdev)) {
190 		kfree_skb(skb);
191 		PDBG("%s - device in error state - dropping\n", __func__);
192 		return -EIO;
193 	}
194 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
195 	if (error < 0)
196 		kfree_skb(skb);
197 	return error < 0 ? error : 0;
198 }
199 
200 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
201 {
202 	int	error = 0;
203 
204 	if (c4iw_fatal_error(rdev)) {
205 		kfree_skb(skb);
206 		PDBG("%s - device in error state - dropping\n", __func__);
207 		return -EIO;
208 	}
209 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
210 	if (error < 0)
211 		kfree_skb(skb);
212 	return error < 0 ? error : 0;
213 }
214 
215 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
216 {
217 	struct cpl_tid_release *req;
218 
219 	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
220 	if (!skb)
221 		return;
222 	req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
223 	INIT_TP_WR(req, hwtid);
224 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
225 	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
226 	c4iw_ofld_send(rdev, skb);
227 	return;
228 }
229 
230 static void set_emss(struct c4iw_ep *ep, u16 opt)
231 {
232 	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
233 	ep->mss = ep->emss;
234 	if (GET_TCPOPT_TSTAMP(opt))
235 		ep->emss -= 12;
236 	if (ep->emss < 128)
237 		ep->emss = 128;
238 	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
239 	     ep->mss, ep->emss);
240 }
241 
242 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
243 {
244 	enum c4iw_ep_state state;
245 
246 	mutex_lock(&epc->mutex);
247 	state = epc->state;
248 	mutex_unlock(&epc->mutex);
249 	return state;
250 }
251 
252 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
253 {
254 	epc->state = new;
255 }
256 
257 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
258 {
259 	mutex_lock(&epc->mutex);
260 	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
261 	__state_set(epc, new);
262 	mutex_unlock(&epc->mutex);
263 	return;
264 }
265 
266 static void *alloc_ep(int size, gfp_t gfp)
267 {
268 	struct c4iw_ep_common *epc;
269 
270 	epc = kzalloc(size, gfp);
271 	if (epc) {
272 		kref_init(&epc->kref);
273 		mutex_init(&epc->mutex);
274 		c4iw_init_wr_wait(&epc->wr_wait);
275 	}
276 	PDBG("%s alloc ep %p\n", __func__, epc);
277 	return epc;
278 }
279 
280 void _c4iw_free_ep(struct kref *kref)
281 {
282 	struct c4iw_ep *ep;
283 
284 	ep = container_of(kref, struct c4iw_ep, com.kref);
285 	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
286 	if (test_bit(QP_REFERENCED, &ep->com.flags))
287 		deref_qp(ep);
288 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
289 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
290 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
291 		dst_release(ep->dst);
292 		cxgb4_l2t_release(ep->l2t);
293 	}
294 	kfree(ep);
295 }
296 
297 static void release_ep_resources(struct c4iw_ep *ep)
298 {
299 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
300 	c4iw_put_ep(&ep->com);
301 }
302 
303 static int status2errno(int status)
304 {
305 	switch (status) {
306 	case CPL_ERR_NONE:
307 		return 0;
308 	case CPL_ERR_CONN_RESET:
309 		return -ECONNRESET;
310 	case CPL_ERR_ARP_MISS:
311 		return -EHOSTUNREACH;
312 	case CPL_ERR_CONN_TIMEDOUT:
313 		return -ETIMEDOUT;
314 	case CPL_ERR_TCAM_FULL:
315 		return -ENOMEM;
316 	case CPL_ERR_CONN_EXIST:
317 		return -EADDRINUSE;
318 	default:
319 		return -EIO;
320 	}
321 }
322 
323 /*
324  * Try and reuse skbs already allocated...
325  */
326 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
327 {
328 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
329 		skb_trim(skb, 0);
330 		skb_get(skb);
331 		skb_reset_transport_header(skb);
332 	} else {
333 		skb = alloc_skb(len, gfp);
334 	}
335 	t4_set_arp_err_handler(skb, NULL, NULL);
336 	return skb;
337 }
338 
339 static struct net_device *get_real_dev(struct net_device *egress_dev)
340 {
341 	struct net_device *phys_dev = egress_dev;
342 	if (egress_dev->priv_flags & IFF_802_1Q_VLAN)
343 		phys_dev = vlan_dev_real_dev(egress_dev);
344 	return phys_dev;
345 }
346 
347 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
348 {
349 	int i;
350 
351 	egress_dev = get_real_dev(egress_dev);
352 	for (i = 0; i < dev->rdev.lldi.nports; i++)
353 		if (dev->rdev.lldi.ports[i] == egress_dev)
354 			return 1;
355 	return 0;
356 }
357 
358 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
359 				     __u8 *peer_ip, __be16 local_port,
360 				     __be16 peer_port, u8 tos,
361 				     __u32 sin6_scope_id)
362 {
363 	struct dst_entry *dst = NULL;
364 
365 	if (IS_ENABLED(CONFIG_IPV6)) {
366 		struct flowi6 fl6;
367 
368 		memset(&fl6, 0, sizeof(fl6));
369 		memcpy(&fl6.daddr, peer_ip, 16);
370 		memcpy(&fl6.saddr, local_ip, 16);
371 		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
372 			fl6.flowi6_oif = sin6_scope_id;
373 		dst = ip6_route_output(&init_net, NULL, &fl6);
374 		if (!dst)
375 			goto out;
376 		if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
377 		    !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
378 			dst_release(dst);
379 			dst = NULL;
380 		}
381 	}
382 
383 out:
384 	return dst;
385 }
386 
387 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
388 				 __be32 peer_ip, __be16 local_port,
389 				 __be16 peer_port, u8 tos)
390 {
391 	struct rtable *rt;
392 	struct flowi4 fl4;
393 	struct neighbour *n;
394 
395 	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
396 				   peer_port, local_port, IPPROTO_TCP,
397 				   tos, 0);
398 	if (IS_ERR(rt))
399 		return NULL;
400 	n = dst_neigh_lookup(&rt->dst, &peer_ip);
401 	if (!n)
402 		return NULL;
403 	if (!our_interface(dev, n->dev)) {
404 		dst_release(&rt->dst);
405 		return NULL;
406 	}
407 	neigh_release(n);
408 	return &rt->dst;
409 }
410 
411 static void arp_failure_discard(void *handle, struct sk_buff *skb)
412 {
413 	PDBG("%s c4iw_dev %p\n", __func__, handle);
414 	kfree_skb(skb);
415 }
416 
417 /*
418  * Handle an ARP failure for an active open.
419  */
420 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
421 {
422 	printk(KERN_ERR MOD "ARP failure duing connect\n");
423 	kfree_skb(skb);
424 }
425 
426 /*
427  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
428  * and send it along.
429  */
430 static void abort_arp_failure(void *handle, struct sk_buff *skb)
431 {
432 	struct c4iw_rdev *rdev = handle;
433 	struct cpl_abort_req *req = cplhdr(skb);
434 
435 	PDBG("%s rdev %p\n", __func__, rdev);
436 	req->cmd = CPL_ABORT_NO_RST;
437 	c4iw_ofld_send(rdev, skb);
438 }
439 
440 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
441 {
442 	unsigned int flowclen = 80;
443 	struct fw_flowc_wr *flowc;
444 	int i;
445 
446 	skb = get_skb(skb, flowclen, GFP_KERNEL);
447 	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
448 
449 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
450 					   FW_FLOWC_WR_NPARAMS(8));
451 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
452 					  16)) | FW_WR_FLOWID(ep->hwtid));
453 
454 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
455 	flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
456 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
457 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
458 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
459 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
460 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
461 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
462 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
463 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
464 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
465 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
466 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
467 	flowc->mnemval[6].val = cpu_to_be32(snd_win);
468 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
469 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
470 	/* Pad WR to 16 byte boundary */
471 	flowc->mnemval[8].mnemonic = 0;
472 	flowc->mnemval[8].val = 0;
473 	for (i = 0; i < 9; i++) {
474 		flowc->mnemval[i].r4[0] = 0;
475 		flowc->mnemval[i].r4[1] = 0;
476 		flowc->mnemval[i].r4[2] = 0;
477 	}
478 
479 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
480 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
481 }
482 
483 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
484 {
485 	struct cpl_close_con_req *req;
486 	struct sk_buff *skb;
487 	int wrlen = roundup(sizeof *req, 16);
488 
489 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
490 	skb = get_skb(NULL, wrlen, gfp);
491 	if (!skb) {
492 		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
493 		return -ENOMEM;
494 	}
495 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
496 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
497 	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
498 	memset(req, 0, wrlen);
499 	INIT_TP_WR(req, ep->hwtid);
500 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
501 						    ep->hwtid));
502 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
503 }
504 
505 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
506 {
507 	struct cpl_abort_req *req;
508 	int wrlen = roundup(sizeof *req, 16);
509 
510 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
511 	skb = get_skb(skb, wrlen, gfp);
512 	if (!skb) {
513 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
514 		       __func__);
515 		return -ENOMEM;
516 	}
517 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
518 	t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
519 	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
520 	memset(req, 0, wrlen);
521 	INIT_TP_WR(req, ep->hwtid);
522 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
523 	req->cmd = CPL_ABORT_SEND_RST;
524 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
525 }
526 
527 #define VLAN_NONE 0xfff
528 #define FILTER_SEL_VLAN_NONE 0xffff
529 #define FILTER_SEL_WIDTH_P_FC (3+1) /* port uses 3 bits, FCoE one bit */
530 #define FILTER_SEL_WIDTH_VIN_P_FC \
531 	(6 + 7 + FILTER_SEL_WIDTH_P_FC) /* 6 bits are unused, VF uses 7 bits*/
532 #define FILTER_SEL_WIDTH_TAG_P_FC \
533 	(3 + FILTER_SEL_WIDTH_VIN_P_FC) /* PF uses 3 bits */
534 #define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC)
535 
536 static unsigned int select_ntuple(struct c4iw_dev *dev, struct dst_entry *dst,
537 				  struct l2t_entry *l2t)
538 {
539 	unsigned int ntuple = 0;
540 	u32 viid;
541 
542 	switch (dev->rdev.lldi.filt_mode) {
543 
544 	/* default filter mode */
545 	case HW_TPL_FR_MT_PR_IV_P_FC:
546 		if (l2t->vlan == VLAN_NONE)
547 			ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC;
548 		else {
549 			ntuple |= l2t->vlan << FILTER_SEL_WIDTH_P_FC;
550 			ntuple |= 1 << FILTER_SEL_WIDTH_TAG_P_FC;
551 		}
552 		ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
553 			  FILTER_SEL_WIDTH_VLD_TAG_P_FC;
554 		break;
555 	case HW_TPL_FR_MT_PR_OV_P_FC: {
556 		viid = cxgb4_port_viid(l2t->neigh->dev);
557 
558 		ntuple |= FW_VIID_VIN_GET(viid) << FILTER_SEL_WIDTH_P_FC;
559 		ntuple |= FW_VIID_PFN_GET(viid) << FILTER_SEL_WIDTH_VIN_P_FC;
560 		ntuple |= FW_VIID_VIVLD_GET(viid) << FILTER_SEL_WIDTH_TAG_P_FC;
561 		ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
562 			  FILTER_SEL_WIDTH_VLD_TAG_P_FC;
563 		break;
564 	}
565 	default:
566 		break;
567 	}
568 	return ntuple;
569 }
570 
571 static int send_connect(struct c4iw_ep *ep)
572 {
573 	struct cpl_act_open_req *req;
574 	struct cpl_t5_act_open_req *t5_req;
575 	struct cpl_act_open_req6 *req6;
576 	struct cpl_t5_act_open_req6 *t5_req6;
577 	struct sk_buff *skb;
578 	u64 opt0;
579 	u32 opt2;
580 	unsigned int mtu_idx;
581 	int wscale;
582 	int wrlen;
583 	int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
584 				sizeof(struct cpl_act_open_req) :
585 				sizeof(struct cpl_t5_act_open_req);
586 	int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
587 				sizeof(struct cpl_act_open_req6) :
588 				sizeof(struct cpl_t5_act_open_req6);
589 	struct sockaddr_in *la = (struct sockaddr_in *)&ep->com.local_addr;
590 	struct sockaddr_in *ra = (struct sockaddr_in *)&ep->com.remote_addr;
591 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
592 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
593 
594 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
595 			roundup(sizev4, 16) :
596 			roundup(sizev6, 16);
597 
598 	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
599 
600 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
601 	if (!skb) {
602 		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
603 		       __func__);
604 		return -ENOMEM;
605 	}
606 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
607 
608 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
609 	wscale = compute_wscale(rcv_win);
610 	opt0 = (nocong ? NO_CONG(1) : 0) |
611 	       KEEP_ALIVE(1) |
612 	       DELACK(1) |
613 	       WND_SCALE(wscale) |
614 	       MSS_IDX(mtu_idx) |
615 	       L2T_IDX(ep->l2t->idx) |
616 	       TX_CHAN(ep->tx_chan) |
617 	       SMAC_SEL(ep->smac_idx) |
618 	       DSCP(ep->tos) |
619 	       ULP_MODE(ULP_MODE_TCPDDP) |
620 	       RCV_BUFSIZ(rcv_win>>10);
621 	opt2 = RX_CHANNEL(0) |
622 	       CCTRL_ECN(enable_ecn) |
623 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
624 	if (enable_tcp_timestamps)
625 		opt2 |= TSTAMPS_EN(1);
626 	if (enable_tcp_sack)
627 		opt2 |= SACK_EN(1);
628 	if (wscale && enable_tcp_window_scaling)
629 		opt2 |= WND_SCALE_EN(1);
630 	t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
631 
632 	if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
633 		if (ep->com.remote_addr.ss_family == AF_INET) {
634 			req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
635 			INIT_TP_WR(req, 0);
636 			OPCODE_TID(req) = cpu_to_be32(
637 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
638 					((ep->rss_qid << 14) | ep->atid)));
639 			req->local_port = la->sin_port;
640 			req->peer_port = ra->sin_port;
641 			req->local_ip = la->sin_addr.s_addr;
642 			req->peer_ip = ra->sin_addr.s_addr;
643 			req->opt0 = cpu_to_be64(opt0);
644 			req->params = cpu_to_be32(select_ntuple(ep->com.dev,
645 						ep->dst, ep->l2t));
646 			req->opt2 = cpu_to_be32(opt2);
647 		} else {
648 			req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
649 
650 			INIT_TP_WR(req6, 0);
651 			OPCODE_TID(req6) = cpu_to_be32(
652 					   MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
653 					   ((ep->rss_qid<<14)|ep->atid)));
654 			req6->local_port = la6->sin6_port;
655 			req6->peer_port = ra6->sin6_port;
656 			req6->local_ip_hi = *((__be64 *)
657 						(la6->sin6_addr.s6_addr));
658 			req6->local_ip_lo = *((__be64 *)
659 						(la6->sin6_addr.s6_addr + 8));
660 			req6->peer_ip_hi = *((__be64 *)
661 						(ra6->sin6_addr.s6_addr));
662 			req6->peer_ip_lo = *((__be64 *)
663 						(ra6->sin6_addr.s6_addr + 8));
664 			req6->opt0 = cpu_to_be64(opt0);
665 			req6->params = cpu_to_be32(
666 					select_ntuple(ep->com.dev, ep->dst,
667 						      ep->l2t));
668 			req6->opt2 = cpu_to_be32(opt2);
669 		}
670 	} else {
671 		if (ep->com.remote_addr.ss_family == AF_INET) {
672 			t5_req = (struct cpl_t5_act_open_req *)
673 				 skb_put(skb, wrlen);
674 			INIT_TP_WR(t5_req, 0);
675 			OPCODE_TID(t5_req) = cpu_to_be32(
676 					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
677 					((ep->rss_qid << 14) | ep->atid)));
678 			t5_req->local_port = la->sin_port;
679 			t5_req->peer_port = ra->sin_port;
680 			t5_req->local_ip = la->sin_addr.s_addr;
681 			t5_req->peer_ip = ra->sin_addr.s_addr;
682 			t5_req->opt0 = cpu_to_be64(opt0);
683 			t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
684 						select_ntuple(ep->com.dev,
685 						ep->dst, ep->l2t)));
686 			t5_req->opt2 = cpu_to_be32(opt2);
687 		} else {
688 			t5_req6 = (struct cpl_t5_act_open_req6 *)
689 				  skb_put(skb, wrlen);
690 			INIT_TP_WR(t5_req6, 0);
691 			OPCODE_TID(t5_req6) = cpu_to_be32(
692 					      MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
693 					      ((ep->rss_qid<<14)|ep->atid)));
694 			t5_req6->local_port = la6->sin6_port;
695 			t5_req6->peer_port = ra6->sin6_port;
696 			t5_req6->local_ip_hi = *((__be64 *)
697 						(la6->sin6_addr.s6_addr));
698 			t5_req6->local_ip_lo = *((__be64 *)
699 						(la6->sin6_addr.s6_addr + 8));
700 			t5_req6->peer_ip_hi = *((__be64 *)
701 						(ra6->sin6_addr.s6_addr));
702 			t5_req6->peer_ip_lo = *((__be64 *)
703 						(ra6->sin6_addr.s6_addr + 8));
704 			t5_req6->opt0 = cpu_to_be64(opt0);
705 			t5_req6->params = (__force __be64)cpu_to_be32(
706 				select_ntuple(ep->com.dev, ep->dst, ep->l2t));
707 			t5_req6->opt2 = cpu_to_be32(opt2);
708 		}
709 	}
710 
711 	set_bit(ACT_OPEN_REQ, &ep->com.history);
712 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
713 }
714 
715 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
716 		u8 mpa_rev_to_use)
717 {
718 	int mpalen, wrlen;
719 	struct fw_ofld_tx_data_wr *req;
720 	struct mpa_message *mpa;
721 	struct mpa_v2_conn_params mpa_v2_params;
722 
723 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
724 
725 	BUG_ON(skb_cloned(skb));
726 
727 	mpalen = sizeof(*mpa) + ep->plen;
728 	if (mpa_rev_to_use == 2)
729 		mpalen += sizeof(struct mpa_v2_conn_params);
730 	wrlen = roundup(mpalen + sizeof *req, 16);
731 	skb = get_skb(skb, wrlen, GFP_KERNEL);
732 	if (!skb) {
733 		connect_reply_upcall(ep, -ENOMEM);
734 		return;
735 	}
736 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
737 
738 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
739 	memset(req, 0, wrlen);
740 	req->op_to_immdlen = cpu_to_be32(
741 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
742 		FW_WR_COMPL(1) |
743 		FW_WR_IMMDLEN(mpalen));
744 	req->flowid_len16 = cpu_to_be32(
745 		FW_WR_FLOWID(ep->hwtid) |
746 		FW_WR_LEN16(wrlen >> 4));
747 	req->plen = cpu_to_be32(mpalen);
748 	req->tunnel_to_proxy = cpu_to_be32(
749 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
750 		FW_OFLD_TX_DATA_WR_SHOVE(1));
751 
752 	mpa = (struct mpa_message *)(req + 1);
753 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
754 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
755 		     (markers_enabled ? MPA_MARKERS : 0) |
756 		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
757 	mpa->private_data_size = htons(ep->plen);
758 	mpa->revision = mpa_rev_to_use;
759 	if (mpa_rev_to_use == 1) {
760 		ep->tried_with_mpa_v1 = 1;
761 		ep->retry_with_mpa_v1 = 0;
762 	}
763 
764 	if (mpa_rev_to_use == 2) {
765 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
766 					       sizeof (struct mpa_v2_conn_params));
767 		mpa_v2_params.ird = htons((u16)ep->ird);
768 		mpa_v2_params.ord = htons((u16)ep->ord);
769 
770 		if (peer2peer) {
771 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
772 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
773 				mpa_v2_params.ord |=
774 					htons(MPA_V2_RDMA_WRITE_RTR);
775 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
776 				mpa_v2_params.ord |=
777 					htons(MPA_V2_RDMA_READ_RTR);
778 		}
779 		memcpy(mpa->private_data, &mpa_v2_params,
780 		       sizeof(struct mpa_v2_conn_params));
781 
782 		if (ep->plen)
783 			memcpy(mpa->private_data +
784 			       sizeof(struct mpa_v2_conn_params),
785 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
786 	} else
787 		if (ep->plen)
788 			memcpy(mpa->private_data,
789 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
790 
791 	/*
792 	 * Reference the mpa skb.  This ensures the data area
793 	 * will remain in memory until the hw acks the tx.
794 	 * Function fw4_ack() will deref it.
795 	 */
796 	skb_get(skb);
797 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
798 	BUG_ON(ep->mpa_skb);
799 	ep->mpa_skb = skb;
800 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
801 	start_ep_timer(ep);
802 	state_set(&ep->com, MPA_REQ_SENT);
803 	ep->mpa_attr.initiator = 1;
804 	return;
805 }
806 
807 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
808 {
809 	int mpalen, wrlen;
810 	struct fw_ofld_tx_data_wr *req;
811 	struct mpa_message *mpa;
812 	struct sk_buff *skb;
813 	struct mpa_v2_conn_params mpa_v2_params;
814 
815 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
816 
817 	mpalen = sizeof(*mpa) + plen;
818 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
819 		mpalen += sizeof(struct mpa_v2_conn_params);
820 	wrlen = roundup(mpalen + sizeof *req, 16);
821 
822 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
823 	if (!skb) {
824 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
825 		return -ENOMEM;
826 	}
827 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
828 
829 	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
830 	memset(req, 0, wrlen);
831 	req->op_to_immdlen = cpu_to_be32(
832 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
833 		FW_WR_COMPL(1) |
834 		FW_WR_IMMDLEN(mpalen));
835 	req->flowid_len16 = cpu_to_be32(
836 		FW_WR_FLOWID(ep->hwtid) |
837 		FW_WR_LEN16(wrlen >> 4));
838 	req->plen = cpu_to_be32(mpalen);
839 	req->tunnel_to_proxy = cpu_to_be32(
840 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
841 		FW_OFLD_TX_DATA_WR_SHOVE(1));
842 
843 	mpa = (struct mpa_message *)(req + 1);
844 	memset(mpa, 0, sizeof(*mpa));
845 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
846 	mpa->flags = MPA_REJECT;
847 	mpa->revision = ep->mpa_attr.version;
848 	mpa->private_data_size = htons(plen);
849 
850 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
851 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
852 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
853 					       sizeof (struct mpa_v2_conn_params));
854 		mpa_v2_params.ird = htons(((u16)ep->ird) |
855 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
856 					   0));
857 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
858 					  (p2p_type ==
859 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
860 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
861 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
862 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
863 		memcpy(mpa->private_data, &mpa_v2_params,
864 		       sizeof(struct mpa_v2_conn_params));
865 
866 		if (ep->plen)
867 			memcpy(mpa->private_data +
868 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
869 	} else
870 		if (plen)
871 			memcpy(mpa->private_data, pdata, plen);
872 
873 	/*
874 	 * Reference the mpa skb again.  This ensures the data area
875 	 * will remain in memory until the hw acks the tx.
876 	 * Function fw4_ack() will deref it.
877 	 */
878 	skb_get(skb);
879 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
880 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
881 	BUG_ON(ep->mpa_skb);
882 	ep->mpa_skb = skb;
883 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
884 }
885 
886 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
887 {
888 	int mpalen, wrlen;
889 	struct fw_ofld_tx_data_wr *req;
890 	struct mpa_message *mpa;
891 	struct sk_buff *skb;
892 	struct mpa_v2_conn_params mpa_v2_params;
893 
894 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
895 
896 	mpalen = sizeof(*mpa) + plen;
897 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
898 		mpalen += sizeof(struct mpa_v2_conn_params);
899 	wrlen = roundup(mpalen + sizeof *req, 16);
900 
901 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
902 	if (!skb) {
903 		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
904 		return -ENOMEM;
905 	}
906 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
907 
908 	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
909 	memset(req, 0, wrlen);
910 	req->op_to_immdlen = cpu_to_be32(
911 		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
912 		FW_WR_COMPL(1) |
913 		FW_WR_IMMDLEN(mpalen));
914 	req->flowid_len16 = cpu_to_be32(
915 		FW_WR_FLOWID(ep->hwtid) |
916 		FW_WR_LEN16(wrlen >> 4));
917 	req->plen = cpu_to_be32(mpalen);
918 	req->tunnel_to_proxy = cpu_to_be32(
919 		FW_OFLD_TX_DATA_WR_FLUSH(1) |
920 		FW_OFLD_TX_DATA_WR_SHOVE(1));
921 
922 	mpa = (struct mpa_message *)(req + 1);
923 	memset(mpa, 0, sizeof(*mpa));
924 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
925 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
926 		     (markers_enabled ? MPA_MARKERS : 0);
927 	mpa->revision = ep->mpa_attr.version;
928 	mpa->private_data_size = htons(plen);
929 
930 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
931 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
932 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
933 					       sizeof (struct mpa_v2_conn_params));
934 		mpa_v2_params.ird = htons((u16)ep->ird);
935 		mpa_v2_params.ord = htons((u16)ep->ord);
936 		if (peer2peer && (ep->mpa_attr.p2p_type !=
937 					FW_RI_INIT_P2PTYPE_DISABLED)) {
938 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
939 
940 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
941 				mpa_v2_params.ord |=
942 					htons(MPA_V2_RDMA_WRITE_RTR);
943 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
944 				mpa_v2_params.ord |=
945 					htons(MPA_V2_RDMA_READ_RTR);
946 		}
947 
948 		memcpy(mpa->private_data, &mpa_v2_params,
949 		       sizeof(struct mpa_v2_conn_params));
950 
951 		if (ep->plen)
952 			memcpy(mpa->private_data +
953 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
954 	} else
955 		if (plen)
956 			memcpy(mpa->private_data, pdata, plen);
957 
958 	/*
959 	 * Reference the mpa skb.  This ensures the data area
960 	 * will remain in memory until the hw acks the tx.
961 	 * Function fw4_ack() will deref it.
962 	 */
963 	skb_get(skb);
964 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
965 	ep->mpa_skb = skb;
966 	state_set(&ep->com, MPA_REP_SENT);
967 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
968 }
969 
970 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
971 {
972 	struct c4iw_ep *ep;
973 	struct cpl_act_establish *req = cplhdr(skb);
974 	unsigned int tid = GET_TID(req);
975 	unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
976 	struct tid_info *t = dev->rdev.lldi.tids;
977 
978 	ep = lookup_atid(t, atid);
979 
980 	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
981 	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
982 
983 	dst_confirm(ep->dst);
984 
985 	/* setup the hwtid for this connection */
986 	ep->hwtid = tid;
987 	cxgb4_insert_tid(t, ep, tid);
988 	insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
989 
990 	ep->snd_seq = be32_to_cpu(req->snd_isn);
991 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
992 
993 	set_emss(ep, ntohs(req->tcp_opt));
994 
995 	/* dealloc the atid */
996 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
997 	cxgb4_free_atid(t, atid);
998 	set_bit(ACT_ESTAB, &ep->com.history);
999 
1000 	/* start MPA negotiation */
1001 	send_flowc(ep, NULL);
1002 	if (ep->retry_with_mpa_v1)
1003 		send_mpa_req(ep, skb, 1);
1004 	else
1005 		send_mpa_req(ep, skb, mpa_rev);
1006 
1007 	return 0;
1008 }
1009 
1010 static void close_complete_upcall(struct c4iw_ep *ep)
1011 {
1012 	struct iw_cm_event event;
1013 
1014 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1015 	memset(&event, 0, sizeof(event));
1016 	event.event = IW_CM_EVENT_CLOSE;
1017 	if (ep->com.cm_id) {
1018 		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1019 		     ep, ep->com.cm_id, ep->hwtid);
1020 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1021 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1022 		ep->com.cm_id = NULL;
1023 		set_bit(CLOSE_UPCALL, &ep->com.history);
1024 	}
1025 }
1026 
1027 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1028 {
1029 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1030 	close_complete_upcall(ep);
1031 	state_set(&ep->com, ABORTING);
1032 	set_bit(ABORT_CONN, &ep->com.history);
1033 	return send_abort(ep, skb, gfp);
1034 }
1035 
1036 static void peer_close_upcall(struct c4iw_ep *ep)
1037 {
1038 	struct iw_cm_event event;
1039 
1040 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1041 	memset(&event, 0, sizeof(event));
1042 	event.event = IW_CM_EVENT_DISCONNECT;
1043 	if (ep->com.cm_id) {
1044 		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1045 		     ep, ep->com.cm_id, ep->hwtid);
1046 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1047 		set_bit(DISCONN_UPCALL, &ep->com.history);
1048 	}
1049 }
1050 
1051 static void peer_abort_upcall(struct c4iw_ep *ep)
1052 {
1053 	struct iw_cm_event event;
1054 
1055 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1056 	memset(&event, 0, sizeof(event));
1057 	event.event = IW_CM_EVENT_CLOSE;
1058 	event.status = -ECONNRESET;
1059 	if (ep->com.cm_id) {
1060 		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1061 		     ep->com.cm_id, ep->hwtid);
1062 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1063 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1064 		ep->com.cm_id = NULL;
1065 		set_bit(ABORT_UPCALL, &ep->com.history);
1066 	}
1067 }
1068 
1069 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1070 {
1071 	struct iw_cm_event event;
1072 
1073 	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1074 	memset(&event, 0, sizeof(event));
1075 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1076 	event.status = status;
1077 	memcpy(&event.local_addr, &ep->com.local_addr,
1078 	       sizeof(ep->com.local_addr));
1079 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1080 	       sizeof(ep->com.remote_addr));
1081 
1082 	if ((status == 0) || (status == -ECONNREFUSED)) {
1083 		if (!ep->tried_with_mpa_v1) {
1084 			/* this means MPA_v2 is used */
1085 			event.private_data_len = ep->plen -
1086 				sizeof(struct mpa_v2_conn_params);
1087 			event.private_data = ep->mpa_pkt +
1088 				sizeof(struct mpa_message) +
1089 				sizeof(struct mpa_v2_conn_params);
1090 		} else {
1091 			/* this means MPA_v1 is used */
1092 			event.private_data_len = ep->plen;
1093 			event.private_data = ep->mpa_pkt +
1094 				sizeof(struct mpa_message);
1095 		}
1096 	}
1097 
1098 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1099 	     ep->hwtid, status);
1100 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1101 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1102 
1103 	if (status < 0) {
1104 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1105 		ep->com.cm_id = NULL;
1106 	}
1107 }
1108 
1109 static void connect_request_upcall(struct c4iw_ep *ep)
1110 {
1111 	struct iw_cm_event event;
1112 
1113 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1114 	memset(&event, 0, sizeof(event));
1115 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1116 	memcpy(&event.local_addr, &ep->com.local_addr,
1117 	       sizeof(ep->com.local_addr));
1118 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1119 	       sizeof(ep->com.remote_addr));
1120 	event.provider_data = ep;
1121 	if (!ep->tried_with_mpa_v1) {
1122 		/* this means MPA_v2 is used */
1123 		event.ord = ep->ord;
1124 		event.ird = ep->ird;
1125 		event.private_data_len = ep->plen -
1126 			sizeof(struct mpa_v2_conn_params);
1127 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1128 			sizeof(struct mpa_v2_conn_params);
1129 	} else {
1130 		/* this means MPA_v1 is used. Send max supported */
1131 		event.ord = c4iw_max_read_depth;
1132 		event.ird = c4iw_max_read_depth;
1133 		event.private_data_len = ep->plen;
1134 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1135 	}
1136 	if (state_read(&ep->parent_ep->com) != DEAD) {
1137 		c4iw_get_ep(&ep->com);
1138 		ep->parent_ep->com.cm_id->event_handler(
1139 						ep->parent_ep->com.cm_id,
1140 						&event);
1141 	}
1142 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1143 	c4iw_put_ep(&ep->parent_ep->com);
1144 	ep->parent_ep = NULL;
1145 }
1146 
1147 static void established_upcall(struct c4iw_ep *ep)
1148 {
1149 	struct iw_cm_event event;
1150 
1151 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1152 	memset(&event, 0, sizeof(event));
1153 	event.event = IW_CM_EVENT_ESTABLISHED;
1154 	event.ird = ep->ird;
1155 	event.ord = ep->ord;
1156 	if (ep->com.cm_id) {
1157 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1158 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1159 		set_bit(ESTAB_UPCALL, &ep->com.history);
1160 	}
1161 }
1162 
1163 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1164 {
1165 	struct cpl_rx_data_ack *req;
1166 	struct sk_buff *skb;
1167 	int wrlen = roundup(sizeof *req, 16);
1168 
1169 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1170 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1171 	if (!skb) {
1172 		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1173 		return 0;
1174 	}
1175 
1176 	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1177 	memset(req, 0, wrlen);
1178 	INIT_TP_WR(req, ep->hwtid);
1179 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1180 						    ep->hwtid));
1181 	req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1182 				       F_RX_DACK_CHANGE |
1183 				       V_RX_DACK_MODE(dack_mode));
1184 	set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1185 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1186 	return credits;
1187 }
1188 
1189 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1190 {
1191 	struct mpa_message *mpa;
1192 	struct mpa_v2_conn_params *mpa_v2_params;
1193 	u16 plen;
1194 	u16 resp_ird, resp_ord;
1195 	u8 rtr_mismatch = 0, insuff_ird = 0;
1196 	struct c4iw_qp_attributes attrs;
1197 	enum c4iw_qp_attr_mask mask;
1198 	int err;
1199 
1200 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1201 
1202 	/*
1203 	 * Stop mpa timer.  If it expired, then the state has
1204 	 * changed and we bail since ep_timeout already aborted
1205 	 * the connection.
1206 	 */
1207 	stop_ep_timer(ep);
1208 	if (state_read(&ep->com) != MPA_REQ_SENT)
1209 		return;
1210 
1211 	/*
1212 	 * If we get more than the supported amount of private data
1213 	 * then we must fail this connection.
1214 	 */
1215 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1216 		err = -EINVAL;
1217 		goto err;
1218 	}
1219 
1220 	/*
1221 	 * copy the new data into our accumulation buffer.
1222 	 */
1223 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1224 				  skb->len);
1225 	ep->mpa_pkt_len += skb->len;
1226 
1227 	/*
1228 	 * if we don't even have the mpa message, then bail.
1229 	 */
1230 	if (ep->mpa_pkt_len < sizeof(*mpa))
1231 		return;
1232 	mpa = (struct mpa_message *) ep->mpa_pkt;
1233 
1234 	/* Validate MPA header. */
1235 	if (mpa->revision > mpa_rev) {
1236 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1237 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1238 		err = -EPROTO;
1239 		goto err;
1240 	}
1241 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1242 		err = -EPROTO;
1243 		goto err;
1244 	}
1245 
1246 	plen = ntohs(mpa->private_data_size);
1247 
1248 	/*
1249 	 * Fail if there's too much private data.
1250 	 */
1251 	if (plen > MPA_MAX_PRIVATE_DATA) {
1252 		err = -EPROTO;
1253 		goto err;
1254 	}
1255 
1256 	/*
1257 	 * If plen does not account for pkt size
1258 	 */
1259 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1260 		err = -EPROTO;
1261 		goto err;
1262 	}
1263 
1264 	ep->plen = (u8) plen;
1265 
1266 	/*
1267 	 * If we don't have all the pdata yet, then bail.
1268 	 * We'll continue process when more data arrives.
1269 	 */
1270 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1271 		return;
1272 
1273 	if (mpa->flags & MPA_REJECT) {
1274 		err = -ECONNREFUSED;
1275 		goto err;
1276 	}
1277 
1278 	/*
1279 	 * If we get here we have accumulated the entire mpa
1280 	 * start reply message including private data. And
1281 	 * the MPA header is valid.
1282 	 */
1283 	state_set(&ep->com, FPDU_MODE);
1284 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1285 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1286 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1287 	ep->mpa_attr.version = mpa->revision;
1288 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1289 
1290 	if (mpa->revision == 2) {
1291 		ep->mpa_attr.enhanced_rdma_conn =
1292 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1293 		if (ep->mpa_attr.enhanced_rdma_conn) {
1294 			mpa_v2_params = (struct mpa_v2_conn_params *)
1295 				(ep->mpa_pkt + sizeof(*mpa));
1296 			resp_ird = ntohs(mpa_v2_params->ird) &
1297 				MPA_V2_IRD_ORD_MASK;
1298 			resp_ord = ntohs(mpa_v2_params->ord) &
1299 				MPA_V2_IRD_ORD_MASK;
1300 
1301 			/*
1302 			 * This is a double-check. Ideally, below checks are
1303 			 * not required since ird/ord stuff has been taken
1304 			 * care of in c4iw_accept_cr
1305 			 */
1306 			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1307 				err = -ENOMEM;
1308 				ep->ird = resp_ord;
1309 				ep->ord = resp_ird;
1310 				insuff_ird = 1;
1311 			}
1312 
1313 			if (ntohs(mpa_v2_params->ird) &
1314 					MPA_V2_PEER2PEER_MODEL) {
1315 				if (ntohs(mpa_v2_params->ord) &
1316 						MPA_V2_RDMA_WRITE_RTR)
1317 					ep->mpa_attr.p2p_type =
1318 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1319 				else if (ntohs(mpa_v2_params->ord) &
1320 						MPA_V2_RDMA_READ_RTR)
1321 					ep->mpa_attr.p2p_type =
1322 						FW_RI_INIT_P2PTYPE_READ_REQ;
1323 			}
1324 		}
1325 	} else if (mpa->revision == 1)
1326 		if (peer2peer)
1327 			ep->mpa_attr.p2p_type = p2p_type;
1328 
1329 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1330 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1331 	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
1332 	     ep->mpa_attr.recv_marker_enabled,
1333 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1334 	     ep->mpa_attr.p2p_type, p2p_type);
1335 
1336 	/*
1337 	 * If responder's RTR does not match with that of initiator, assign
1338 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1339 	 * generated when moving QP to RTS state.
1340 	 * A TERM message will be sent after QP has moved to RTS state
1341 	 */
1342 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1343 			(ep->mpa_attr.p2p_type != p2p_type)) {
1344 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1345 		rtr_mismatch = 1;
1346 	}
1347 
1348 	attrs.mpa_attr = ep->mpa_attr;
1349 	attrs.max_ird = ep->ird;
1350 	attrs.max_ord = ep->ord;
1351 	attrs.llp_stream_handle = ep;
1352 	attrs.next_state = C4IW_QP_STATE_RTS;
1353 
1354 	mask = C4IW_QP_ATTR_NEXT_STATE |
1355 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1356 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1357 
1358 	/* bind QP and TID with INIT_WR */
1359 	err = c4iw_modify_qp(ep->com.qp->rhp,
1360 			     ep->com.qp, mask, &attrs, 1);
1361 	if (err)
1362 		goto err;
1363 
1364 	/*
1365 	 * If responder's RTR requirement did not match with what initiator
1366 	 * supports, generate TERM message
1367 	 */
1368 	if (rtr_mismatch) {
1369 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1370 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1371 		attrs.ecode = MPA_NOMATCH_RTR;
1372 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1373 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1374 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1375 		err = -ENOMEM;
1376 		goto out;
1377 	}
1378 
1379 	/*
1380 	 * Generate TERM if initiator IRD is not sufficient for responder
1381 	 * provided ORD. Currently, we do the same behaviour even when
1382 	 * responder provided IRD is also not sufficient as regards to
1383 	 * initiator ORD.
1384 	 */
1385 	if (insuff_ird) {
1386 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1387 				__func__);
1388 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1389 		attrs.ecode = MPA_INSUFF_IRD;
1390 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1391 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1392 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1393 		err = -ENOMEM;
1394 		goto out;
1395 	}
1396 	goto out;
1397 err:
1398 	state_set(&ep->com, ABORTING);
1399 	send_abort(ep, skb, GFP_KERNEL);
1400 out:
1401 	connect_reply_upcall(ep, err);
1402 	return;
1403 }
1404 
1405 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1406 {
1407 	struct mpa_message *mpa;
1408 	struct mpa_v2_conn_params *mpa_v2_params;
1409 	u16 plen;
1410 
1411 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1412 
1413 	if (state_read(&ep->com) != MPA_REQ_WAIT)
1414 		return;
1415 
1416 	/*
1417 	 * If we get more than the supported amount of private data
1418 	 * then we must fail this connection.
1419 	 */
1420 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1421 		stop_ep_timer(ep);
1422 		abort_connection(ep, skb, GFP_KERNEL);
1423 		return;
1424 	}
1425 
1426 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1427 
1428 	/*
1429 	 * Copy the new data into our accumulation buffer.
1430 	 */
1431 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1432 				  skb->len);
1433 	ep->mpa_pkt_len += skb->len;
1434 
1435 	/*
1436 	 * If we don't even have the mpa message, then bail.
1437 	 * We'll continue process when more data arrives.
1438 	 */
1439 	if (ep->mpa_pkt_len < sizeof(*mpa))
1440 		return;
1441 
1442 	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1443 	stop_ep_timer(ep);
1444 	mpa = (struct mpa_message *) ep->mpa_pkt;
1445 
1446 	/*
1447 	 * Validate MPA Header.
1448 	 */
1449 	if (mpa->revision > mpa_rev) {
1450 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1451 		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
1452 		stop_ep_timer(ep);
1453 		abort_connection(ep, skb, GFP_KERNEL);
1454 		return;
1455 	}
1456 
1457 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1458 		stop_ep_timer(ep);
1459 		abort_connection(ep, skb, GFP_KERNEL);
1460 		return;
1461 	}
1462 
1463 	plen = ntohs(mpa->private_data_size);
1464 
1465 	/*
1466 	 * Fail if there's too much private data.
1467 	 */
1468 	if (plen > MPA_MAX_PRIVATE_DATA) {
1469 		stop_ep_timer(ep);
1470 		abort_connection(ep, skb, GFP_KERNEL);
1471 		return;
1472 	}
1473 
1474 	/*
1475 	 * If plen does not account for pkt size
1476 	 */
1477 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1478 		stop_ep_timer(ep);
1479 		abort_connection(ep, skb, GFP_KERNEL);
1480 		return;
1481 	}
1482 	ep->plen = (u8) plen;
1483 
1484 	/*
1485 	 * If we don't have all the pdata yet, then bail.
1486 	 */
1487 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1488 		return;
1489 
1490 	/*
1491 	 * If we get here we have accumulated the entire mpa
1492 	 * start reply message including private data.
1493 	 */
1494 	ep->mpa_attr.initiator = 0;
1495 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1496 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1497 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1498 	ep->mpa_attr.version = mpa->revision;
1499 	if (mpa->revision == 1)
1500 		ep->tried_with_mpa_v1 = 1;
1501 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1502 
1503 	if (mpa->revision == 2) {
1504 		ep->mpa_attr.enhanced_rdma_conn =
1505 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1506 		if (ep->mpa_attr.enhanced_rdma_conn) {
1507 			mpa_v2_params = (struct mpa_v2_conn_params *)
1508 				(ep->mpa_pkt + sizeof(*mpa));
1509 			ep->ird = ntohs(mpa_v2_params->ird) &
1510 				MPA_V2_IRD_ORD_MASK;
1511 			ep->ord = ntohs(mpa_v2_params->ord) &
1512 				MPA_V2_IRD_ORD_MASK;
1513 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1514 				if (peer2peer) {
1515 					if (ntohs(mpa_v2_params->ord) &
1516 							MPA_V2_RDMA_WRITE_RTR)
1517 						ep->mpa_attr.p2p_type =
1518 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1519 					else if (ntohs(mpa_v2_params->ord) &
1520 							MPA_V2_RDMA_READ_RTR)
1521 						ep->mpa_attr.p2p_type =
1522 						FW_RI_INIT_P2PTYPE_READ_REQ;
1523 				}
1524 		}
1525 	} else if (mpa->revision == 1)
1526 		if (peer2peer)
1527 			ep->mpa_attr.p2p_type = p2p_type;
1528 
1529 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1530 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1531 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1532 	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1533 	     ep->mpa_attr.p2p_type);
1534 
1535 	state_set(&ep->com, MPA_REQ_RCVD);
1536 
1537 	/* drive upcall */
1538 	connect_request_upcall(ep);
1539 	return;
1540 }
1541 
1542 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1543 {
1544 	struct c4iw_ep *ep;
1545 	struct cpl_rx_data *hdr = cplhdr(skb);
1546 	unsigned int dlen = ntohs(hdr->len);
1547 	unsigned int tid = GET_TID(hdr);
1548 	struct tid_info *t = dev->rdev.lldi.tids;
1549 	__u8 status = hdr->status;
1550 
1551 	ep = lookup_tid(t, tid);
1552 	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1553 	skb_pull(skb, sizeof(*hdr));
1554 	skb_trim(skb, dlen);
1555 
1556 	/* update RX credits */
1557 	update_rx_credits(ep, dlen);
1558 
1559 	switch (state_read(&ep->com)) {
1560 	case MPA_REQ_SENT:
1561 		ep->rcv_seq += dlen;
1562 		process_mpa_reply(ep, skb);
1563 		break;
1564 	case MPA_REQ_WAIT:
1565 		ep->rcv_seq += dlen;
1566 		process_mpa_request(ep, skb);
1567 		break;
1568 	case FPDU_MODE: {
1569 		struct c4iw_qp_attributes attrs;
1570 		BUG_ON(!ep->com.qp);
1571 		if (status)
1572 			pr_err("%s Unexpected streaming data." \
1573 			       " qpid %u ep %p state %d tid %u status %d\n",
1574 			       __func__, ep->com.qp->wq.sq.qid, ep,
1575 			       state_read(&ep->com), ep->hwtid, status);
1576 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1577 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1578 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1579 		break;
1580 	}
1581 	default:
1582 		break;
1583 	}
1584 	return 0;
1585 }
1586 
1587 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1588 {
1589 	struct c4iw_ep *ep;
1590 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1591 	int release = 0;
1592 	unsigned int tid = GET_TID(rpl);
1593 	struct tid_info *t = dev->rdev.lldi.tids;
1594 
1595 	ep = lookup_tid(t, tid);
1596 	if (!ep) {
1597 		printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1598 		return 0;
1599 	}
1600 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1601 	mutex_lock(&ep->com.mutex);
1602 	switch (ep->com.state) {
1603 	case ABORTING:
1604 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1605 		__state_set(&ep->com, DEAD);
1606 		release = 1;
1607 		break;
1608 	default:
1609 		printk(KERN_ERR "%s ep %p state %d\n",
1610 		     __func__, ep, ep->com.state);
1611 		break;
1612 	}
1613 	mutex_unlock(&ep->com.mutex);
1614 
1615 	if (release)
1616 		release_ep_resources(ep);
1617 	return 0;
1618 }
1619 
1620 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1621 {
1622 	struct sk_buff *skb;
1623 	struct fw_ofld_connection_wr *req;
1624 	unsigned int mtu_idx;
1625 	int wscale;
1626 	struct sockaddr_in *sin;
1627 
1628 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1629 	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1630 	memset(req, 0, sizeof(*req));
1631 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1632 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1633 	req->le.filter = cpu_to_be32(select_ntuple(ep->com.dev, ep->dst,
1634 				     ep->l2t));
1635 	sin = (struct sockaddr_in *)&ep->com.local_addr;
1636 	req->le.lport = sin->sin_port;
1637 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1638 	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1639 	req->le.pport = sin->sin_port;
1640 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1641 	req->tcb.t_state_to_astid =
1642 			htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1643 			V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1644 	req->tcb.cplrxdataack_cplpassacceptrpl =
1645 			htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1646 	req->tcb.tx_max = (__force __be32) jiffies;
1647 	req->tcb.rcv_adv = htons(1);
1648 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1649 	wscale = compute_wscale(rcv_win);
1650 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1651 		(nocong ? NO_CONG(1) : 0) |
1652 		KEEP_ALIVE(1) |
1653 		DELACK(1) |
1654 		WND_SCALE(wscale) |
1655 		MSS_IDX(mtu_idx) |
1656 		L2T_IDX(ep->l2t->idx) |
1657 		TX_CHAN(ep->tx_chan) |
1658 		SMAC_SEL(ep->smac_idx) |
1659 		DSCP(ep->tos) |
1660 		ULP_MODE(ULP_MODE_TCPDDP) |
1661 		RCV_BUFSIZ(rcv_win >> 10));
1662 	req->tcb.opt2 = (__force __be32) (PACE(1) |
1663 		TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1664 		RX_CHANNEL(0) |
1665 		CCTRL_ECN(enable_ecn) |
1666 		RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1667 	if (enable_tcp_timestamps)
1668 		req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1669 	if (enable_tcp_sack)
1670 		req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1671 	if (wscale && enable_tcp_window_scaling)
1672 		req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1673 	req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1674 	req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1675 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1676 	set_bit(ACT_OFLD_CONN, &ep->com.history);
1677 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1678 }
1679 
1680 /*
1681  * Return whether a failed active open has allocated a TID
1682  */
1683 static inline int act_open_has_tid(int status)
1684 {
1685 	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1686 	       status != CPL_ERR_ARP_MISS;
1687 }
1688 
1689 #define ACT_OPEN_RETRY_COUNT 2
1690 
1691 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1692 		     struct dst_entry *dst, struct c4iw_dev *cdev,
1693 		     bool clear_mpa_v1)
1694 {
1695 	struct neighbour *n;
1696 	int err, step;
1697 	struct net_device *pdev;
1698 
1699 	n = dst_neigh_lookup(dst, peer_ip);
1700 	if (!n)
1701 		return -ENODEV;
1702 
1703 	rcu_read_lock();
1704 	err = -ENOMEM;
1705 	if (n->dev->flags & IFF_LOOPBACK) {
1706 		if (iptype == 4)
1707 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1708 		else if (IS_ENABLED(CONFIG_IPV6))
1709 			for_each_netdev(&init_net, pdev) {
1710 				if (ipv6_chk_addr(&init_net,
1711 						  (struct in6_addr *)peer_ip,
1712 						  pdev, 1))
1713 					break;
1714 			}
1715 		else
1716 			pdev = NULL;
1717 
1718 		if (!pdev) {
1719 			err = -ENODEV;
1720 			goto out;
1721 		}
1722 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1723 					n, pdev, 0);
1724 		if (!ep->l2t)
1725 			goto out;
1726 		ep->mtu = pdev->mtu;
1727 		ep->tx_chan = cxgb4_port_chan(pdev);
1728 		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1729 		step = cdev->rdev.lldi.ntxq /
1730 			cdev->rdev.lldi.nchan;
1731 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
1732 		step = cdev->rdev.lldi.nrxq /
1733 			cdev->rdev.lldi.nchan;
1734 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
1735 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1736 			cxgb4_port_idx(pdev) * step];
1737 		dev_put(pdev);
1738 	} else {
1739 		pdev = get_real_dev(n->dev);
1740 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1741 					n, pdev, 0);
1742 		if (!ep->l2t)
1743 			goto out;
1744 		ep->mtu = dst_mtu(dst);
1745 		ep->tx_chan = cxgb4_port_chan(n->dev);
1746 		ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1747 		step = cdev->rdev.lldi.ntxq /
1748 			cdev->rdev.lldi.nchan;
1749 		ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1750 		ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1751 		step = cdev->rdev.lldi.nrxq /
1752 			cdev->rdev.lldi.nchan;
1753 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1754 			cxgb4_port_idx(n->dev) * step];
1755 
1756 		if (clear_mpa_v1) {
1757 			ep->retry_with_mpa_v1 = 0;
1758 			ep->tried_with_mpa_v1 = 0;
1759 		}
1760 	}
1761 	err = 0;
1762 out:
1763 	rcu_read_unlock();
1764 
1765 	neigh_release(n);
1766 
1767 	return err;
1768 }
1769 
1770 static int c4iw_reconnect(struct c4iw_ep *ep)
1771 {
1772 	int err = 0;
1773 	struct sockaddr_in *laddr = (struct sockaddr_in *)
1774 				    &ep->com.cm_id->local_addr;
1775 	struct sockaddr_in *raddr = (struct sockaddr_in *)
1776 				    &ep->com.cm_id->remote_addr;
1777 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1778 				      &ep->com.cm_id->local_addr;
1779 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1780 				      &ep->com.cm_id->remote_addr;
1781 	int iptype;
1782 	__u8 *ra;
1783 
1784 	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1785 	init_timer(&ep->timer);
1786 
1787 	/*
1788 	 * Allocate an active TID to initiate a TCP connection.
1789 	 */
1790 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1791 	if (ep->atid == -1) {
1792 		pr_err("%s - cannot alloc atid.\n", __func__);
1793 		err = -ENOMEM;
1794 		goto fail2;
1795 	}
1796 	insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1797 
1798 	/* find a route */
1799 	if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1800 		ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1801 				     raddr->sin_addr.s_addr, laddr->sin_port,
1802 				     raddr->sin_port, 0);
1803 		iptype = 4;
1804 		ra = (__u8 *)&raddr->sin_addr;
1805 	} else {
1806 		ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1807 				      raddr6->sin6_addr.s6_addr,
1808 				      laddr6->sin6_port, raddr6->sin6_port, 0,
1809 				      raddr6->sin6_scope_id);
1810 		iptype = 6;
1811 		ra = (__u8 *)&raddr6->sin6_addr;
1812 	}
1813 	if (!ep->dst) {
1814 		pr_err("%s - cannot find route.\n", __func__);
1815 		err = -EHOSTUNREACH;
1816 		goto fail3;
1817 	}
1818 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1819 	if (err) {
1820 		pr_err("%s - cannot alloc l2e.\n", __func__);
1821 		goto fail4;
1822 	}
1823 
1824 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1825 	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1826 	     ep->l2t->idx);
1827 
1828 	state_set(&ep->com, CONNECTING);
1829 	ep->tos = 0;
1830 
1831 	/* send connect request to rnic */
1832 	err = send_connect(ep);
1833 	if (!err)
1834 		goto out;
1835 
1836 	cxgb4_l2t_release(ep->l2t);
1837 fail4:
1838 	dst_release(ep->dst);
1839 fail3:
1840 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1841 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1842 fail2:
1843 	/*
1844 	 * remember to send notification to upper layer.
1845 	 * We are in here so the upper layer is not aware that this is
1846 	 * re-connect attempt and so, upper layer is still waiting for
1847 	 * response of 1st connect request.
1848 	 */
1849 	connect_reply_upcall(ep, -ECONNRESET);
1850 	c4iw_put_ep(&ep->com);
1851 out:
1852 	return err;
1853 }
1854 
1855 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1856 {
1857 	struct c4iw_ep *ep;
1858 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
1859 	unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1860 					ntohl(rpl->atid_status)));
1861 	struct tid_info *t = dev->rdev.lldi.tids;
1862 	int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1863 	struct sockaddr_in *la;
1864 	struct sockaddr_in *ra;
1865 	struct sockaddr_in6 *la6;
1866 	struct sockaddr_in6 *ra6;
1867 
1868 	ep = lookup_atid(t, atid);
1869 	la = (struct sockaddr_in *)&ep->com.local_addr;
1870 	ra = (struct sockaddr_in *)&ep->com.remote_addr;
1871 	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1872 	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1873 
1874 	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1875 	     status, status2errno(status));
1876 
1877 	if (status == CPL_ERR_RTX_NEG_ADVICE) {
1878 		printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1879 			atid);
1880 		return 0;
1881 	}
1882 
1883 	set_bit(ACT_OPEN_RPL, &ep->com.history);
1884 
1885 	/*
1886 	 * Log interesting failures.
1887 	 */
1888 	switch (status) {
1889 	case CPL_ERR_CONN_RESET:
1890 	case CPL_ERR_CONN_TIMEDOUT:
1891 		break;
1892 	case CPL_ERR_TCAM_FULL:
1893 		mutex_lock(&dev->rdev.stats.lock);
1894 		dev->rdev.stats.tcam_full++;
1895 		mutex_unlock(&dev->rdev.stats.lock);
1896 		if (ep->com.local_addr.ss_family == AF_INET &&
1897 		    dev->rdev.lldi.enable_fw_ofld_conn) {
1898 			send_fw_act_open_req(ep,
1899 					     GET_TID_TID(GET_AOPEN_ATID(
1900 					     ntohl(rpl->atid_status))));
1901 			return 0;
1902 		}
1903 		break;
1904 	case CPL_ERR_CONN_EXIST:
1905 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1906 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
1907 			remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1908 					atid);
1909 			cxgb4_free_atid(t, atid);
1910 			dst_release(ep->dst);
1911 			cxgb4_l2t_release(ep->l2t);
1912 			c4iw_reconnect(ep);
1913 			return 0;
1914 		}
1915 		break;
1916 	default:
1917 		if (ep->com.local_addr.ss_family == AF_INET) {
1918 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1919 				atid, status, status2errno(status),
1920 				&la->sin_addr.s_addr, ntohs(la->sin_port),
1921 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
1922 		} else {
1923 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1924 				atid, status, status2errno(status),
1925 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1926 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1927 		}
1928 		break;
1929 	}
1930 
1931 	connect_reply_upcall(ep, status2errno(status));
1932 	state_set(&ep->com, DEAD);
1933 
1934 	if (status && act_open_has_tid(status))
1935 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1936 
1937 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1938 	cxgb4_free_atid(t, atid);
1939 	dst_release(ep->dst);
1940 	cxgb4_l2t_release(ep->l2t);
1941 	c4iw_put_ep(&ep->com);
1942 
1943 	return 0;
1944 }
1945 
1946 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1947 {
1948 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1949 	struct tid_info *t = dev->rdev.lldi.tids;
1950 	unsigned int stid = GET_TID(rpl);
1951 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1952 
1953 	if (!ep) {
1954 		PDBG("%s stid %d lookup failure!\n", __func__, stid);
1955 		goto out;
1956 	}
1957 	PDBG("%s ep %p status %d error %d\n", __func__, ep,
1958 	     rpl->status, status2errno(rpl->status));
1959 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1960 
1961 out:
1962 	return 0;
1963 }
1964 
1965 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1966 {
1967 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1968 	struct tid_info *t = dev->rdev.lldi.tids;
1969 	unsigned int stid = GET_TID(rpl);
1970 	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1971 
1972 	PDBG("%s ep %p\n", __func__, ep);
1973 	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1974 	return 0;
1975 }
1976 
1977 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1978 		      struct cpl_pass_accept_req *req)
1979 {
1980 	struct cpl_pass_accept_rpl *rpl;
1981 	unsigned int mtu_idx;
1982 	u64 opt0;
1983 	u32 opt2;
1984 	int wscale;
1985 
1986 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1987 	BUG_ON(skb_cloned(skb));
1988 	skb_trim(skb, sizeof(*rpl));
1989 	skb_get(skb);
1990 	cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1991 	wscale = compute_wscale(rcv_win);
1992 	opt0 = (nocong ? NO_CONG(1) : 0) |
1993 	       KEEP_ALIVE(1) |
1994 	       DELACK(1) |
1995 	       WND_SCALE(wscale) |
1996 	       MSS_IDX(mtu_idx) |
1997 	       L2T_IDX(ep->l2t->idx) |
1998 	       TX_CHAN(ep->tx_chan) |
1999 	       SMAC_SEL(ep->smac_idx) |
2000 	       DSCP(ep->tos >> 2) |
2001 	       ULP_MODE(ULP_MODE_TCPDDP) |
2002 	       RCV_BUFSIZ(rcv_win>>10);
2003 	opt2 = RX_CHANNEL(0) |
2004 	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
2005 
2006 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2007 		opt2 |= TSTAMPS_EN(1);
2008 	if (enable_tcp_sack && req->tcpopt.sack)
2009 		opt2 |= SACK_EN(1);
2010 	if (wscale && enable_tcp_window_scaling)
2011 		opt2 |= WND_SCALE_EN(1);
2012 	if (enable_ecn) {
2013 		const struct tcphdr *tcph;
2014 		u32 hlen = ntohl(req->hdr_len);
2015 
2016 		tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2017 			G_IP_HDR_LEN(hlen);
2018 		if (tcph->ece && tcph->cwr)
2019 			opt2 |= CCTRL_ECN(1);
2020 	}
2021 
2022 	rpl = cplhdr(skb);
2023 	INIT_TP_WR(rpl, ep->hwtid);
2024 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2025 				      ep->hwtid));
2026 	rpl->opt0 = cpu_to_be64(opt0);
2027 	rpl->opt2 = cpu_to_be32(opt2);
2028 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2029 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
2030 	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2031 
2032 	return;
2033 }
2034 
2035 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2036 {
2037 	PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2038 	BUG_ON(skb_cloned(skb));
2039 	skb_trim(skb, sizeof(struct cpl_tid_release));
2040 	skb_get(skb);
2041 	release_tid(&dev->rdev, hwtid, skb);
2042 	return;
2043 }
2044 
2045 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2046 		       __u8 *local_ip, __u8 *peer_ip,
2047 		       __be16 *local_port, __be16 *peer_port)
2048 {
2049 	int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2050 	int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2051 	struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2052 	struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2053 	struct tcphdr *tcp = (struct tcphdr *)
2054 			     ((u8 *)(req + 1) + eth_len + ip_len);
2055 
2056 	if (ip->version == 4) {
2057 		PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2058 		     ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2059 		     ntohs(tcp->dest));
2060 		*iptype = 4;
2061 		memcpy(peer_ip, &ip->saddr, 4);
2062 		memcpy(local_ip, &ip->daddr, 4);
2063 	} else {
2064 		PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2065 		     ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2066 		     ntohs(tcp->dest));
2067 		*iptype = 6;
2068 		memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2069 		memcpy(local_ip, ip6->daddr.s6_addr, 16);
2070 	}
2071 	*peer_port = tcp->source;
2072 	*local_port = tcp->dest;
2073 
2074 	return;
2075 }
2076 
2077 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2078 {
2079 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2080 	struct cpl_pass_accept_req *req = cplhdr(skb);
2081 	unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2082 	struct tid_info *t = dev->rdev.lldi.tids;
2083 	unsigned int hwtid = GET_TID(req);
2084 	struct dst_entry *dst;
2085 	__u8 local_ip[16], peer_ip[16];
2086 	__be16 local_port, peer_port;
2087 	int err;
2088 	u16 peer_mss = ntohs(req->tcpopt.mss);
2089 	int iptype;
2090 
2091 	parent_ep = lookup_stid(t, stid);
2092 	if (!parent_ep) {
2093 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2094 		goto reject;
2095 	}
2096 
2097 	if (state_read(&parent_ep->com) != LISTEN) {
2098 		printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2099 		       __func__);
2100 		goto reject;
2101 	}
2102 
2103 	get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2104 
2105 	/* Find output route */
2106 	if (iptype == 4)  {
2107 		PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2108 		     , __func__, parent_ep, hwtid,
2109 		     local_ip, peer_ip, ntohs(local_port),
2110 		     ntohs(peer_port), peer_mss);
2111 		dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2112 				 local_port, peer_port,
2113 				 GET_POPEN_TOS(ntohl(req->tos_stid)));
2114 	} else {
2115 		PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2116 		     , __func__, parent_ep, hwtid,
2117 		     local_ip, peer_ip, ntohs(local_port),
2118 		     ntohs(peer_port), peer_mss);
2119 		dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2120 				  PASS_OPEN_TOS(ntohl(req->tos_stid)),
2121 				  ((struct sockaddr_in6 *)
2122 				  &parent_ep->com.local_addr)->sin6_scope_id);
2123 	}
2124 	if (!dst) {
2125 		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2126 		       __func__);
2127 		goto reject;
2128 	}
2129 
2130 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2131 	if (!child_ep) {
2132 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2133 		       __func__);
2134 		dst_release(dst);
2135 		goto reject;
2136 	}
2137 
2138 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2139 	if (err) {
2140 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2141 		       __func__);
2142 		dst_release(dst);
2143 		kfree(child_ep);
2144 		goto reject;
2145 	}
2146 
2147 	if (peer_mss && child_ep->mtu > (peer_mss + 40))
2148 		child_ep->mtu = peer_mss + 40;
2149 
2150 	state_set(&child_ep->com, CONNECTING);
2151 	child_ep->com.dev = dev;
2152 	child_ep->com.cm_id = NULL;
2153 	if (iptype == 4) {
2154 		struct sockaddr_in *sin = (struct sockaddr_in *)
2155 			&child_ep->com.local_addr;
2156 		sin->sin_family = PF_INET;
2157 		sin->sin_port = local_port;
2158 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2159 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2160 		sin->sin_family = PF_INET;
2161 		sin->sin_port = peer_port;
2162 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2163 	} else {
2164 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2165 			&child_ep->com.local_addr;
2166 		sin6->sin6_family = PF_INET6;
2167 		sin6->sin6_port = local_port;
2168 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2169 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2170 		sin6->sin6_family = PF_INET6;
2171 		sin6->sin6_port = peer_port;
2172 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2173 	}
2174 	c4iw_get_ep(&parent_ep->com);
2175 	child_ep->parent_ep = parent_ep;
2176 	child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2177 	child_ep->dst = dst;
2178 	child_ep->hwtid = hwtid;
2179 
2180 	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2181 	     child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2182 
2183 	init_timer(&child_ep->timer);
2184 	cxgb4_insert_tid(t, child_ep, hwtid);
2185 	insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2186 	accept_cr(child_ep, skb, req);
2187 	set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2188 	goto out;
2189 reject:
2190 	reject_cr(dev, hwtid, skb);
2191 out:
2192 	return 0;
2193 }
2194 
2195 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2196 {
2197 	struct c4iw_ep *ep;
2198 	struct cpl_pass_establish *req = cplhdr(skb);
2199 	struct tid_info *t = dev->rdev.lldi.tids;
2200 	unsigned int tid = GET_TID(req);
2201 
2202 	ep = lookup_tid(t, tid);
2203 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2204 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2205 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2206 
2207 	PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2208 	     ntohs(req->tcp_opt));
2209 
2210 	set_emss(ep, ntohs(req->tcp_opt));
2211 
2212 	dst_confirm(ep->dst);
2213 	state_set(&ep->com, MPA_REQ_WAIT);
2214 	start_ep_timer(ep);
2215 	send_flowc(ep, skb);
2216 	set_bit(PASS_ESTAB, &ep->com.history);
2217 
2218 	return 0;
2219 }
2220 
2221 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2222 {
2223 	struct cpl_peer_close *hdr = cplhdr(skb);
2224 	struct c4iw_ep *ep;
2225 	struct c4iw_qp_attributes attrs;
2226 	int disconnect = 1;
2227 	int release = 0;
2228 	struct tid_info *t = dev->rdev.lldi.tids;
2229 	unsigned int tid = GET_TID(hdr);
2230 	int ret;
2231 
2232 	ep = lookup_tid(t, tid);
2233 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2234 	dst_confirm(ep->dst);
2235 
2236 	set_bit(PEER_CLOSE, &ep->com.history);
2237 	mutex_lock(&ep->com.mutex);
2238 	switch (ep->com.state) {
2239 	case MPA_REQ_WAIT:
2240 		__state_set(&ep->com, CLOSING);
2241 		break;
2242 	case MPA_REQ_SENT:
2243 		__state_set(&ep->com, CLOSING);
2244 		connect_reply_upcall(ep, -ECONNRESET);
2245 		break;
2246 	case MPA_REQ_RCVD:
2247 
2248 		/*
2249 		 * We're gonna mark this puppy DEAD, but keep
2250 		 * the reference on it until the ULP accepts or
2251 		 * rejects the CR. Also wake up anyone waiting
2252 		 * in rdma connection migration (see c4iw_accept_cr()).
2253 		 */
2254 		__state_set(&ep->com, CLOSING);
2255 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2256 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2257 		break;
2258 	case MPA_REP_SENT:
2259 		__state_set(&ep->com, CLOSING);
2260 		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2261 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2262 		break;
2263 	case FPDU_MODE:
2264 		start_ep_timer(ep);
2265 		__state_set(&ep->com, CLOSING);
2266 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2267 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2268 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2269 		if (ret != -ECONNRESET) {
2270 			peer_close_upcall(ep);
2271 			disconnect = 1;
2272 		}
2273 		break;
2274 	case ABORTING:
2275 		disconnect = 0;
2276 		break;
2277 	case CLOSING:
2278 		__state_set(&ep->com, MORIBUND);
2279 		disconnect = 0;
2280 		break;
2281 	case MORIBUND:
2282 		stop_ep_timer(ep);
2283 		if (ep->com.cm_id && ep->com.qp) {
2284 			attrs.next_state = C4IW_QP_STATE_IDLE;
2285 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2286 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2287 		}
2288 		close_complete_upcall(ep);
2289 		__state_set(&ep->com, DEAD);
2290 		release = 1;
2291 		disconnect = 0;
2292 		break;
2293 	case DEAD:
2294 		disconnect = 0;
2295 		break;
2296 	default:
2297 		BUG_ON(1);
2298 	}
2299 	mutex_unlock(&ep->com.mutex);
2300 	if (disconnect)
2301 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2302 	if (release)
2303 		release_ep_resources(ep);
2304 	return 0;
2305 }
2306 
2307 /*
2308  * Returns whether an ABORT_REQ_RSS message is a negative advice.
2309  */
2310 static int is_neg_adv_abort(unsigned int status)
2311 {
2312 	return status == CPL_ERR_RTX_NEG_ADVICE ||
2313 	       status == CPL_ERR_PERSIST_NEG_ADVICE;
2314 }
2315 
2316 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2317 {
2318 	struct cpl_abort_req_rss *req = cplhdr(skb);
2319 	struct c4iw_ep *ep;
2320 	struct cpl_abort_rpl *rpl;
2321 	struct sk_buff *rpl_skb;
2322 	struct c4iw_qp_attributes attrs;
2323 	int ret;
2324 	int release = 0;
2325 	struct tid_info *t = dev->rdev.lldi.tids;
2326 	unsigned int tid = GET_TID(req);
2327 
2328 	ep = lookup_tid(t, tid);
2329 	if (is_neg_adv_abort(req->status)) {
2330 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2331 		     ep->hwtid);
2332 		return 0;
2333 	}
2334 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2335 	     ep->com.state);
2336 	set_bit(PEER_ABORT, &ep->com.history);
2337 
2338 	/*
2339 	 * Wake up any threads in rdma_init() or rdma_fini().
2340 	 * However, this is not needed if com state is just
2341 	 * MPA_REQ_SENT
2342 	 */
2343 	if (ep->com.state != MPA_REQ_SENT)
2344 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2345 
2346 	mutex_lock(&ep->com.mutex);
2347 	switch (ep->com.state) {
2348 	case CONNECTING:
2349 		break;
2350 	case MPA_REQ_WAIT:
2351 		stop_ep_timer(ep);
2352 		break;
2353 	case MPA_REQ_SENT:
2354 		stop_ep_timer(ep);
2355 		if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2356 			connect_reply_upcall(ep, -ECONNRESET);
2357 		else {
2358 			/*
2359 			 * we just don't send notification upwards because we
2360 			 * want to retry with mpa_v1 without upper layers even
2361 			 * knowing it.
2362 			 *
2363 			 * do some housekeeping so as to re-initiate the
2364 			 * connection
2365 			 */
2366 			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2367 			     mpa_rev);
2368 			ep->retry_with_mpa_v1 = 1;
2369 		}
2370 		break;
2371 	case MPA_REP_SENT:
2372 		break;
2373 	case MPA_REQ_RCVD:
2374 		break;
2375 	case MORIBUND:
2376 	case CLOSING:
2377 		stop_ep_timer(ep);
2378 		/*FALLTHROUGH*/
2379 	case FPDU_MODE:
2380 		if (ep->com.cm_id && ep->com.qp) {
2381 			attrs.next_state = C4IW_QP_STATE_ERROR;
2382 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2383 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2384 				     &attrs, 1);
2385 			if (ret)
2386 				printk(KERN_ERR MOD
2387 				       "%s - qp <- error failed!\n",
2388 				       __func__);
2389 		}
2390 		peer_abort_upcall(ep);
2391 		break;
2392 	case ABORTING:
2393 		break;
2394 	case DEAD:
2395 		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2396 		mutex_unlock(&ep->com.mutex);
2397 		return 0;
2398 	default:
2399 		BUG_ON(1);
2400 		break;
2401 	}
2402 	dst_confirm(ep->dst);
2403 	if (ep->com.state != ABORTING) {
2404 		__state_set(&ep->com, DEAD);
2405 		/* we don't release if we want to retry with mpa_v1 */
2406 		if (!ep->retry_with_mpa_v1)
2407 			release = 1;
2408 	}
2409 	mutex_unlock(&ep->com.mutex);
2410 
2411 	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2412 	if (!rpl_skb) {
2413 		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2414 		       __func__);
2415 		release = 1;
2416 		goto out;
2417 	}
2418 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2419 	rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2420 	INIT_TP_WR(rpl, ep->hwtid);
2421 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2422 	rpl->cmd = CPL_ABORT_NO_RST;
2423 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2424 out:
2425 	if (release)
2426 		release_ep_resources(ep);
2427 	else if (ep->retry_with_mpa_v1) {
2428 		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2429 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2430 		dst_release(ep->dst);
2431 		cxgb4_l2t_release(ep->l2t);
2432 		c4iw_reconnect(ep);
2433 	}
2434 
2435 	return 0;
2436 }
2437 
2438 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2439 {
2440 	struct c4iw_ep *ep;
2441 	struct c4iw_qp_attributes attrs;
2442 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2443 	int release = 0;
2444 	struct tid_info *t = dev->rdev.lldi.tids;
2445 	unsigned int tid = GET_TID(rpl);
2446 
2447 	ep = lookup_tid(t, tid);
2448 
2449 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2450 	BUG_ON(!ep);
2451 
2452 	/* The cm_id may be null if we failed to connect */
2453 	mutex_lock(&ep->com.mutex);
2454 	switch (ep->com.state) {
2455 	case CLOSING:
2456 		__state_set(&ep->com, MORIBUND);
2457 		break;
2458 	case MORIBUND:
2459 		stop_ep_timer(ep);
2460 		if ((ep->com.cm_id) && (ep->com.qp)) {
2461 			attrs.next_state = C4IW_QP_STATE_IDLE;
2462 			c4iw_modify_qp(ep->com.qp->rhp,
2463 					     ep->com.qp,
2464 					     C4IW_QP_ATTR_NEXT_STATE,
2465 					     &attrs, 1);
2466 		}
2467 		close_complete_upcall(ep);
2468 		__state_set(&ep->com, DEAD);
2469 		release = 1;
2470 		break;
2471 	case ABORTING:
2472 	case DEAD:
2473 		break;
2474 	default:
2475 		BUG_ON(1);
2476 		break;
2477 	}
2478 	mutex_unlock(&ep->com.mutex);
2479 	if (release)
2480 		release_ep_resources(ep);
2481 	return 0;
2482 }
2483 
2484 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2485 {
2486 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
2487 	struct tid_info *t = dev->rdev.lldi.tids;
2488 	unsigned int tid = GET_TID(rpl);
2489 	struct c4iw_ep *ep;
2490 	struct c4iw_qp_attributes attrs;
2491 
2492 	ep = lookup_tid(t, tid);
2493 	BUG_ON(!ep);
2494 
2495 	if (ep && ep->com.qp) {
2496 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2497 		       ep->com.qp->wq.sq.qid);
2498 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2499 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2500 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2501 	} else
2502 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2503 
2504 	return 0;
2505 }
2506 
2507 /*
2508  * Upcall from the adapter indicating data has been transmitted.
2509  * For us its just the single MPA request or reply.  We can now free
2510  * the skb holding the mpa message.
2511  */
2512 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2513 {
2514 	struct c4iw_ep *ep;
2515 	struct cpl_fw4_ack *hdr = cplhdr(skb);
2516 	u8 credits = hdr->credits;
2517 	unsigned int tid = GET_TID(hdr);
2518 	struct tid_info *t = dev->rdev.lldi.tids;
2519 
2520 
2521 	ep = lookup_tid(t, tid);
2522 	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2523 	if (credits == 0) {
2524 		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2525 		     __func__, ep, ep->hwtid, state_read(&ep->com));
2526 		return 0;
2527 	}
2528 
2529 	dst_confirm(ep->dst);
2530 	if (ep->mpa_skb) {
2531 		PDBG("%s last streaming msg ack ep %p tid %u state %u "
2532 		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2533 		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2534 		kfree_skb(ep->mpa_skb);
2535 		ep->mpa_skb = NULL;
2536 	}
2537 	return 0;
2538 }
2539 
2540 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2541 {
2542 	int err;
2543 	struct c4iw_ep *ep = to_ep(cm_id);
2544 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2545 
2546 	if (state_read(&ep->com) == DEAD) {
2547 		c4iw_put_ep(&ep->com);
2548 		return -ECONNRESET;
2549 	}
2550 	set_bit(ULP_REJECT, &ep->com.history);
2551 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2552 	if (mpa_rev == 0)
2553 		abort_connection(ep, NULL, GFP_KERNEL);
2554 	else {
2555 		err = send_mpa_reject(ep, pdata, pdata_len);
2556 		err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2557 	}
2558 	c4iw_put_ep(&ep->com);
2559 	return 0;
2560 }
2561 
2562 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2563 {
2564 	int err;
2565 	struct c4iw_qp_attributes attrs;
2566 	enum c4iw_qp_attr_mask mask;
2567 	struct c4iw_ep *ep = to_ep(cm_id);
2568 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2569 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2570 
2571 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2572 	if (state_read(&ep->com) == DEAD) {
2573 		err = -ECONNRESET;
2574 		goto err;
2575 	}
2576 
2577 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2578 	BUG_ON(!qp);
2579 
2580 	set_bit(ULP_ACCEPT, &ep->com.history);
2581 	if ((conn_param->ord > c4iw_max_read_depth) ||
2582 	    (conn_param->ird > c4iw_max_read_depth)) {
2583 		abort_connection(ep, NULL, GFP_KERNEL);
2584 		err = -EINVAL;
2585 		goto err;
2586 	}
2587 
2588 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2589 		if (conn_param->ord > ep->ird) {
2590 			ep->ird = conn_param->ird;
2591 			ep->ord = conn_param->ord;
2592 			send_mpa_reject(ep, conn_param->private_data,
2593 					conn_param->private_data_len);
2594 			abort_connection(ep, NULL, GFP_KERNEL);
2595 			err = -ENOMEM;
2596 			goto err;
2597 		}
2598 		if (conn_param->ird > ep->ord) {
2599 			if (!ep->ord)
2600 				conn_param->ird = 1;
2601 			else {
2602 				abort_connection(ep, NULL, GFP_KERNEL);
2603 				err = -ENOMEM;
2604 				goto err;
2605 			}
2606 		}
2607 
2608 	}
2609 	ep->ird = conn_param->ird;
2610 	ep->ord = conn_param->ord;
2611 
2612 	if (ep->mpa_attr.version != 2)
2613 		if (peer2peer && ep->ird == 0)
2614 			ep->ird = 1;
2615 
2616 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2617 
2618 	cm_id->add_ref(cm_id);
2619 	ep->com.cm_id = cm_id;
2620 	ep->com.qp = qp;
2621 	ref_qp(ep);
2622 
2623 	/* bind QP to EP and move to RTS */
2624 	attrs.mpa_attr = ep->mpa_attr;
2625 	attrs.max_ird = ep->ird;
2626 	attrs.max_ord = ep->ord;
2627 	attrs.llp_stream_handle = ep;
2628 	attrs.next_state = C4IW_QP_STATE_RTS;
2629 
2630 	/* bind QP and TID with INIT_WR */
2631 	mask = C4IW_QP_ATTR_NEXT_STATE |
2632 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2633 			     C4IW_QP_ATTR_MPA_ATTR |
2634 			     C4IW_QP_ATTR_MAX_IRD |
2635 			     C4IW_QP_ATTR_MAX_ORD;
2636 
2637 	err = c4iw_modify_qp(ep->com.qp->rhp,
2638 			     ep->com.qp, mask, &attrs, 1);
2639 	if (err)
2640 		goto err1;
2641 	err = send_mpa_reply(ep, conn_param->private_data,
2642 			     conn_param->private_data_len);
2643 	if (err)
2644 		goto err1;
2645 
2646 	state_set(&ep->com, FPDU_MODE);
2647 	established_upcall(ep);
2648 	c4iw_put_ep(&ep->com);
2649 	return 0;
2650 err1:
2651 	ep->com.cm_id = NULL;
2652 	cm_id->rem_ref(cm_id);
2653 err:
2654 	c4iw_put_ep(&ep->com);
2655 	return err;
2656 }
2657 
2658 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2659 {
2660 	struct in_device *ind;
2661 	int found = 0;
2662 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2663 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2664 
2665 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
2666 	if (!ind)
2667 		return -EADDRNOTAVAIL;
2668 	for_primary_ifa(ind) {
2669 		laddr->sin_addr.s_addr = ifa->ifa_address;
2670 		raddr->sin_addr.s_addr = ifa->ifa_address;
2671 		found = 1;
2672 		break;
2673 	}
2674 	endfor_ifa(ind);
2675 	in_dev_put(ind);
2676 	return found ? 0 : -EADDRNOTAVAIL;
2677 }
2678 
2679 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2680 		      unsigned char banned_flags)
2681 {
2682 	struct inet6_dev *idev;
2683 	int err = -EADDRNOTAVAIL;
2684 
2685 	rcu_read_lock();
2686 	idev = __in6_dev_get(dev);
2687 	if (idev != NULL) {
2688 		struct inet6_ifaddr *ifp;
2689 
2690 		read_lock_bh(&idev->lock);
2691 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
2692 			if (ifp->scope == IFA_LINK &&
2693 			    !(ifp->flags & banned_flags)) {
2694 				memcpy(addr, &ifp->addr, 16);
2695 				err = 0;
2696 				break;
2697 			}
2698 		}
2699 		read_unlock_bh(&idev->lock);
2700 	}
2701 	rcu_read_unlock();
2702 	return err;
2703 }
2704 
2705 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2706 {
2707 	struct in6_addr uninitialized_var(addr);
2708 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2709 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2710 
2711 	if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2712 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2713 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2714 		return 0;
2715 	}
2716 	return -EADDRNOTAVAIL;
2717 }
2718 
2719 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2720 {
2721 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2722 	struct c4iw_ep *ep;
2723 	int err = 0;
2724 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2725 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2726 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2727 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2728 				      &cm_id->remote_addr;
2729 	__u8 *ra;
2730 	int iptype;
2731 
2732 	if ((conn_param->ord > c4iw_max_read_depth) ||
2733 	    (conn_param->ird > c4iw_max_read_depth)) {
2734 		err = -EINVAL;
2735 		goto out;
2736 	}
2737 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2738 	if (!ep) {
2739 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2740 		err = -ENOMEM;
2741 		goto out;
2742 	}
2743 	init_timer(&ep->timer);
2744 	ep->plen = conn_param->private_data_len;
2745 	if (ep->plen)
2746 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2747 		       conn_param->private_data, ep->plen);
2748 	ep->ird = conn_param->ird;
2749 	ep->ord = conn_param->ord;
2750 
2751 	if (peer2peer && ep->ord == 0)
2752 		ep->ord = 1;
2753 
2754 	cm_id->add_ref(cm_id);
2755 	ep->com.dev = dev;
2756 	ep->com.cm_id = cm_id;
2757 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2758 	if (!ep->com.qp) {
2759 		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2760 		err = -EINVAL;
2761 		goto fail2;
2762 	}
2763 	ref_qp(ep);
2764 	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2765 	     ep->com.qp, cm_id);
2766 
2767 	/*
2768 	 * Allocate an active TID to initiate a TCP connection.
2769 	 */
2770 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2771 	if (ep->atid == -1) {
2772 		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2773 		err = -ENOMEM;
2774 		goto fail2;
2775 	}
2776 	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2777 
2778 	if (cm_id->remote_addr.ss_family == AF_INET) {
2779 		iptype = 4;
2780 		ra = (__u8 *)&raddr->sin_addr;
2781 
2782 		/*
2783 		 * Handle loopback requests to INADDR_ANY.
2784 		 */
2785 		if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2786 			err = pick_local_ipaddrs(dev, cm_id);
2787 			if (err)
2788 				goto fail2;
2789 		}
2790 
2791 		/* find a route */
2792 		PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2793 		     __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2794 		     ra, ntohs(raddr->sin_port));
2795 		ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2796 				     raddr->sin_addr.s_addr, laddr->sin_port,
2797 				     raddr->sin_port, 0);
2798 	} else {
2799 		iptype = 6;
2800 		ra = (__u8 *)&raddr6->sin6_addr;
2801 
2802 		/*
2803 		 * Handle loopback requests to INADDR_ANY.
2804 		 */
2805 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2806 			err = pick_local_ip6addrs(dev, cm_id);
2807 			if (err)
2808 				goto fail2;
2809 		}
2810 
2811 		/* find a route */
2812 		PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2813 		     __func__, laddr6->sin6_addr.s6_addr,
2814 		     ntohs(laddr6->sin6_port),
2815 		     raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2816 		ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2817 				      raddr6->sin6_addr.s6_addr,
2818 				      laddr6->sin6_port, raddr6->sin6_port, 0,
2819 				      raddr6->sin6_scope_id);
2820 	}
2821 	if (!ep->dst) {
2822 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2823 		err = -EHOSTUNREACH;
2824 		goto fail3;
2825 	}
2826 
2827 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2828 	if (err) {
2829 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2830 		goto fail4;
2831 	}
2832 
2833 	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2834 		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2835 		ep->l2t->idx);
2836 
2837 	state_set(&ep->com, CONNECTING);
2838 	ep->tos = 0;
2839 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
2840 	       sizeof(ep->com.local_addr));
2841 	memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2842 	       sizeof(ep->com.remote_addr));
2843 
2844 	/* send connect request to rnic */
2845 	err = send_connect(ep);
2846 	if (!err)
2847 		goto out;
2848 
2849 	cxgb4_l2t_release(ep->l2t);
2850 fail4:
2851 	dst_release(ep->dst);
2852 fail3:
2853 	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2854 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2855 fail2:
2856 	cm_id->rem_ref(cm_id);
2857 	c4iw_put_ep(&ep->com);
2858 out:
2859 	return err;
2860 }
2861 
2862 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2863 {
2864 	int err;
2865 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2866 
2867 	c4iw_init_wr_wait(&ep->com.wr_wait);
2868 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2869 				   ep->stid, &sin6->sin6_addr,
2870 				   sin6->sin6_port,
2871 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
2872 	if (!err)
2873 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2874 					  &ep->com.wr_wait,
2875 					  0, 0, __func__);
2876 	if (err)
2877 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2878 		       err, ep->stid,
2879 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2880 	return err;
2881 }
2882 
2883 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2884 {
2885 	int err;
2886 	struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2887 
2888 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
2889 		do {
2890 			err = cxgb4_create_server_filter(
2891 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
2892 				sin->sin_addr.s_addr, sin->sin_port, 0,
2893 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2894 			if (err == -EBUSY) {
2895 				set_current_state(TASK_UNINTERRUPTIBLE);
2896 				schedule_timeout(usecs_to_jiffies(100));
2897 			}
2898 		} while (err == -EBUSY);
2899 	} else {
2900 		c4iw_init_wr_wait(&ep->com.wr_wait);
2901 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2902 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2903 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2904 		if (!err)
2905 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2906 						  &ep->com.wr_wait,
2907 						  0, 0, __func__);
2908 	}
2909 	if (err)
2910 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2911 		       , err, ep->stid,
2912 		       &sin->sin_addr, ntohs(sin->sin_port));
2913 	return err;
2914 }
2915 
2916 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2917 {
2918 	int err = 0;
2919 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2920 	struct c4iw_listen_ep *ep;
2921 
2922 	might_sleep();
2923 
2924 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2925 	if (!ep) {
2926 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2927 		err = -ENOMEM;
2928 		goto fail1;
2929 	}
2930 	PDBG("%s ep %p\n", __func__, ep);
2931 	cm_id->add_ref(cm_id);
2932 	ep->com.cm_id = cm_id;
2933 	ep->com.dev = dev;
2934 	ep->backlog = backlog;
2935 	memcpy(&ep->com.local_addr, &cm_id->local_addr,
2936 	       sizeof(ep->com.local_addr));
2937 
2938 	/*
2939 	 * Allocate a server TID.
2940 	 */
2941 	if (dev->rdev.lldi.enable_fw_ofld_conn)
2942 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2943 					     cm_id->local_addr.ss_family, ep);
2944 	else
2945 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2946 					    cm_id->local_addr.ss_family, ep);
2947 
2948 	if (ep->stid == -1) {
2949 		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2950 		err = -ENOMEM;
2951 		goto fail2;
2952 	}
2953 	insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2954 	state_set(&ep->com, LISTEN);
2955 	if (ep->com.local_addr.ss_family == AF_INET)
2956 		err = create_server4(dev, ep);
2957 	else
2958 		err = create_server6(dev, ep);
2959 	if (!err) {
2960 		cm_id->provider_data = ep;
2961 		goto out;
2962 	}
2963 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2964 			ep->com.local_addr.ss_family);
2965 fail2:
2966 	cm_id->rem_ref(cm_id);
2967 	c4iw_put_ep(&ep->com);
2968 fail1:
2969 out:
2970 	return err;
2971 }
2972 
2973 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2974 {
2975 	int err;
2976 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2977 
2978 	PDBG("%s ep %p\n", __func__, ep);
2979 
2980 	might_sleep();
2981 	state_set(&ep->com, DEAD);
2982 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2983 	    ep->com.local_addr.ss_family == AF_INET) {
2984 		err = cxgb4_remove_server_filter(
2985 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
2986 			ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2987 	} else {
2988 		c4iw_init_wr_wait(&ep->com.wr_wait);
2989 		err = cxgb4_remove_server(
2990 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
2991 				ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2992 		if (err)
2993 			goto done;
2994 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2995 					  0, 0, __func__);
2996 	}
2997 	remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
2998 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2999 			ep->com.local_addr.ss_family);
3000 done:
3001 	cm_id->rem_ref(cm_id);
3002 	c4iw_put_ep(&ep->com);
3003 	return err;
3004 }
3005 
3006 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3007 {
3008 	int ret = 0;
3009 	int close = 0;
3010 	int fatal = 0;
3011 	struct c4iw_rdev *rdev;
3012 
3013 	mutex_lock(&ep->com.mutex);
3014 
3015 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3016 	     states[ep->com.state], abrupt);
3017 
3018 	rdev = &ep->com.dev->rdev;
3019 	if (c4iw_fatal_error(rdev)) {
3020 		fatal = 1;
3021 		close_complete_upcall(ep);
3022 		ep->com.state = DEAD;
3023 	}
3024 	switch (ep->com.state) {
3025 	case MPA_REQ_WAIT:
3026 	case MPA_REQ_SENT:
3027 	case MPA_REQ_RCVD:
3028 	case MPA_REP_SENT:
3029 	case FPDU_MODE:
3030 		close = 1;
3031 		if (abrupt)
3032 			ep->com.state = ABORTING;
3033 		else {
3034 			ep->com.state = CLOSING;
3035 			start_ep_timer(ep);
3036 		}
3037 		set_bit(CLOSE_SENT, &ep->com.flags);
3038 		break;
3039 	case CLOSING:
3040 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3041 			close = 1;
3042 			if (abrupt) {
3043 				stop_ep_timer(ep);
3044 				ep->com.state = ABORTING;
3045 			} else
3046 				ep->com.state = MORIBUND;
3047 		}
3048 		break;
3049 	case MORIBUND:
3050 	case ABORTING:
3051 	case DEAD:
3052 		PDBG("%s ignoring disconnect ep %p state %u\n",
3053 		     __func__, ep, ep->com.state);
3054 		break;
3055 	default:
3056 		BUG();
3057 		break;
3058 	}
3059 
3060 	if (close) {
3061 		if (abrupt) {
3062 			set_bit(EP_DISC_ABORT, &ep->com.history);
3063 			close_complete_upcall(ep);
3064 			ret = send_abort(ep, NULL, gfp);
3065 		} else {
3066 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3067 			ret = send_halfclose(ep, gfp);
3068 		}
3069 		if (ret)
3070 			fatal = 1;
3071 	}
3072 	mutex_unlock(&ep->com.mutex);
3073 	if (fatal)
3074 		release_ep_resources(ep);
3075 	return ret;
3076 }
3077 
3078 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3079 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3080 {
3081 	struct c4iw_ep *ep;
3082 	int atid = be32_to_cpu(req->tid);
3083 
3084 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3085 					   (__force u32) req->tid);
3086 	if (!ep)
3087 		return;
3088 
3089 	switch (req->retval) {
3090 	case FW_ENOMEM:
3091 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3092 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3093 			send_fw_act_open_req(ep, atid);
3094 			return;
3095 		}
3096 	case FW_EADDRINUSE:
3097 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3098 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3099 			send_fw_act_open_req(ep, atid);
3100 			return;
3101 		}
3102 		break;
3103 	default:
3104 		pr_info("%s unexpected ofld conn wr retval %d\n",
3105 		       __func__, req->retval);
3106 		break;
3107 	}
3108 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3109 	       req->retval, atid);
3110 	mutex_lock(&dev->rdev.stats.lock);
3111 	dev->rdev.stats.act_ofld_conn_fails++;
3112 	mutex_unlock(&dev->rdev.stats.lock);
3113 	connect_reply_upcall(ep, status2errno(req->retval));
3114 	state_set(&ep->com, DEAD);
3115 	remove_handle(dev, &dev->atid_idr, atid);
3116 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3117 	dst_release(ep->dst);
3118 	cxgb4_l2t_release(ep->l2t);
3119 	c4iw_put_ep(&ep->com);
3120 }
3121 
3122 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3123 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3124 {
3125 	struct sk_buff *rpl_skb;
3126 	struct cpl_pass_accept_req *cpl;
3127 	int ret;
3128 
3129 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3130 	BUG_ON(!rpl_skb);
3131 	if (req->retval) {
3132 		PDBG("%s passive open failure %d\n", __func__, req->retval);
3133 		mutex_lock(&dev->rdev.stats.lock);
3134 		dev->rdev.stats.pas_ofld_conn_fails++;
3135 		mutex_unlock(&dev->rdev.stats.lock);
3136 		kfree_skb(rpl_skb);
3137 	} else {
3138 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3139 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3140 					(__force u32) htonl(
3141 					(__force u32) req->tid)));
3142 		ret = pass_accept_req(dev, rpl_skb);
3143 		if (!ret)
3144 			kfree_skb(rpl_skb);
3145 	}
3146 	return;
3147 }
3148 
3149 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3150 {
3151 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3152 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3153 
3154 	switch (rpl->type) {
3155 	case FW6_TYPE_CQE:
3156 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3157 		break;
3158 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3159 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3160 		switch (req->t_state) {
3161 		case TCP_SYN_SENT:
3162 			active_ofld_conn_reply(dev, skb, req);
3163 			break;
3164 		case TCP_SYN_RECV:
3165 			passive_ofld_conn_reply(dev, skb, req);
3166 			break;
3167 		default:
3168 			pr_err("%s unexpected ofld conn wr state %d\n",
3169 			       __func__, req->t_state);
3170 			break;
3171 		}
3172 		break;
3173 	}
3174 	return 0;
3175 }
3176 
3177 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3178 {
3179 	u32 l2info;
3180 	u16 vlantag, len, hdr_len, eth_hdr_len;
3181 	u8 intf;
3182 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3183 	struct cpl_pass_accept_req *req;
3184 	struct tcp_options_received tmp_opt;
3185 	struct c4iw_dev *dev;
3186 
3187 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3188 	/* Store values from cpl_rx_pkt in temporary location. */
3189 	vlantag = (__force u16) cpl->vlan;
3190 	len = (__force u16) cpl->len;
3191 	l2info  = (__force u32) cpl->l2info;
3192 	hdr_len = (__force u16) cpl->hdr_len;
3193 	intf = cpl->iff;
3194 
3195 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3196 
3197 	/*
3198 	 * We need to parse the TCP options from SYN packet.
3199 	 * to generate cpl_pass_accept_req.
3200 	 */
3201 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3202 	tcp_clear_options(&tmp_opt);
3203 	tcp_parse_options(skb, &tmp_opt, 0, NULL);
3204 
3205 	req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3206 	memset(req, 0, sizeof(*req));
3207 	req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3208 			 V_SYN_MAC_IDX(G_RX_MACIDX(
3209 			 (__force int) htonl(l2info))) |
3210 			 F_SYN_XACT_MATCH);
3211 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3212 			    G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3213 			    G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3214 	req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3215 					(__force int) htonl(l2info))) |
3216 				   V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3217 					(__force int) htons(hdr_len))) |
3218 				   V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3219 					(__force int) htons(hdr_len))) |
3220 				   V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3221 	req->vlan = (__force __be16) vlantag;
3222 	req->len = (__force __be16) len;
3223 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3224 				    PASS_OPEN_TOS(tos));
3225 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3226 	if (tmp_opt.wscale_ok)
3227 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3228 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3229 	if (tmp_opt.sack_ok)
3230 		req->tcpopt.sack = 1;
3231 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3232 	return;
3233 }
3234 
3235 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3236 				  __be32 laddr, __be16 lport,
3237 				  __be32 raddr, __be16 rport,
3238 				  u32 rcv_isn, u32 filter, u16 window,
3239 				  u32 rss_qid, u8 port_id)
3240 {
3241 	struct sk_buff *req_skb;
3242 	struct fw_ofld_connection_wr *req;
3243 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
3244 
3245 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3246 	req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3247 	memset(req, 0, sizeof(*req));
3248 	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3249 	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3250 	req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3251 	req->le.filter = (__force __be32) filter;
3252 	req->le.lport = lport;
3253 	req->le.pport = rport;
3254 	req->le.u.ipv4.lip = laddr;
3255 	req->le.u.ipv4.pip = raddr;
3256 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3257 	req->tcb.rcv_adv = htons(window);
3258 	req->tcb.t_state_to_astid =
3259 		 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3260 			V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3261 			V_FW_OFLD_CONNECTION_WR_ASTID(
3262 			GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3263 
3264 	/*
3265 	 * We store the qid in opt2 which will be used by the firmware
3266 	 * to send us the wr response.
3267 	 */
3268 	req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3269 
3270 	/*
3271 	 * We initialize the MSS index in TCB to 0xF.
3272 	 * So that when driver sends cpl_pass_accept_rpl
3273 	 * TCB picks up the correct value. If this was 0
3274 	 * TP will ignore any value > 0 for MSS index.
3275 	 */
3276 	req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3277 	req->cookie = (unsigned long)skb;
3278 
3279 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3280 	cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3281 }
3282 
3283 /*
3284  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3285  * messages when a filter is being used instead of server to
3286  * redirect a syn packet. When packets hit filter they are redirected
3287  * to the offload queue and driver tries to establish the connection
3288  * using firmware work request.
3289  */
3290 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3291 {
3292 	int stid;
3293 	unsigned int filter;
3294 	struct ethhdr *eh = NULL;
3295 	struct vlan_ethhdr *vlan_eh = NULL;
3296 	struct iphdr *iph;
3297 	struct tcphdr *tcph;
3298 	struct rss_header *rss = (void *)skb->data;
3299 	struct cpl_rx_pkt *cpl = (void *)skb->data;
3300 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
3301 	struct l2t_entry *e;
3302 	struct dst_entry *dst;
3303 	struct c4iw_ep *lep;
3304 	u16 window;
3305 	struct port_info *pi;
3306 	struct net_device *pdev;
3307 	u16 rss_qid, eth_hdr_len;
3308 	int step;
3309 	u32 tx_chan;
3310 	struct neighbour *neigh;
3311 
3312 	/* Drop all non-SYN packets */
3313 	if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3314 		goto reject;
3315 
3316 	/*
3317 	 * Drop all packets which did not hit the filter.
3318 	 * Unlikely to happen.
3319 	 */
3320 	if (!(rss->filter_hit && rss->filter_tid))
3321 		goto reject;
3322 
3323 	/*
3324 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3325 	 */
3326 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val)
3327 					  - dev->rdev.lldi.tids->sftid_base
3328 					  + dev->rdev.lldi.tids->nstids;
3329 
3330 	lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3331 	if (!lep) {
3332 		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3333 		goto reject;
3334 	}
3335 
3336 	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3337 			    G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3338 			    G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3339 	if (eth_hdr_len == ETH_HLEN) {
3340 		eh = (struct ethhdr *)(req + 1);
3341 		iph = (struct iphdr *)(eh + 1);
3342 	} else {
3343 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
3344 		iph = (struct iphdr *)(vlan_eh + 1);
3345 		skb->vlan_tci = ntohs(cpl->vlan);
3346 	}
3347 
3348 	if (iph->version != 0x4)
3349 		goto reject;
3350 
3351 	tcph = (struct tcphdr *)(iph + 1);
3352 	skb_set_network_header(skb, (void *)iph - (void *)rss);
3353 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3354 	skb_get(skb);
3355 
3356 	PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3357 	     ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3358 	     ntohs(tcph->source), iph->tos);
3359 
3360 	dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3361 			 iph->tos);
3362 	if (!dst) {
3363 		pr_err("%s - failed to find dst entry!\n",
3364 		       __func__);
3365 		goto reject;
3366 	}
3367 	neigh = dst_neigh_lookup_skb(dst, skb);
3368 
3369 	if (!neigh) {
3370 		pr_err("%s - failed to allocate neigh!\n",
3371 		       __func__);
3372 		goto free_dst;
3373 	}
3374 
3375 	if (neigh->dev->flags & IFF_LOOPBACK) {
3376 		pdev = ip_dev_find(&init_net, iph->daddr);
3377 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3378 				    pdev, 0);
3379 		pi = (struct port_info *)netdev_priv(pdev);
3380 		tx_chan = cxgb4_port_chan(pdev);
3381 		dev_put(pdev);
3382 	} else {
3383 		pdev = get_real_dev(neigh->dev);
3384 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3385 					pdev, 0);
3386 		pi = (struct port_info *)netdev_priv(pdev);
3387 		tx_chan = cxgb4_port_chan(pdev);
3388 	}
3389 	if (!e) {
3390 		pr_err("%s - failed to allocate l2t entry!\n",
3391 		       __func__);
3392 		goto free_dst;
3393 	}
3394 
3395 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3396 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3397 	window = (__force u16) htons((__force u16)tcph->window);
3398 
3399 	/* Calcuate filter portion for LE region. */
3400 	filter = (__force unsigned int) cpu_to_be32(select_ntuple(dev, dst, e));
3401 
3402 	/*
3403 	 * Synthesize the cpl_pass_accept_req. We have everything except the
3404 	 * TID. Once firmware sends a reply with TID we update the TID field
3405 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
3406 	 */
3407 	build_cpl_pass_accept_req(skb, stid, iph->tos);
3408 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3409 			      tcph->source, ntohl(tcph->seq), filter, window,
3410 			      rss_qid, pi->port_id);
3411 	cxgb4_l2t_release(e);
3412 free_dst:
3413 	dst_release(dst);
3414 reject:
3415 	return 0;
3416 }
3417 
3418 /*
3419  * These are the real handlers that are called from a
3420  * work queue.
3421  */
3422 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3423 	[CPL_ACT_ESTABLISH] = act_establish,
3424 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
3425 	[CPL_RX_DATA] = rx_data,
3426 	[CPL_ABORT_RPL_RSS] = abort_rpl,
3427 	[CPL_ABORT_RPL] = abort_rpl,
3428 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
3429 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3430 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3431 	[CPL_PASS_ESTABLISH] = pass_establish,
3432 	[CPL_PEER_CLOSE] = peer_close,
3433 	[CPL_ABORT_REQ_RSS] = peer_abort,
3434 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
3435 	[CPL_RDMA_TERMINATE] = terminate,
3436 	[CPL_FW4_ACK] = fw4_ack,
3437 	[CPL_FW6_MSG] = deferred_fw6_msg,
3438 	[CPL_RX_PKT] = rx_pkt
3439 };
3440 
3441 static void process_timeout(struct c4iw_ep *ep)
3442 {
3443 	struct c4iw_qp_attributes attrs;
3444 	int abort = 1;
3445 
3446 	mutex_lock(&ep->com.mutex);
3447 	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3448 	     ep->com.state);
3449 	set_bit(TIMEDOUT, &ep->com.history);
3450 	switch (ep->com.state) {
3451 	case MPA_REQ_SENT:
3452 		__state_set(&ep->com, ABORTING);
3453 		connect_reply_upcall(ep, -ETIMEDOUT);
3454 		break;
3455 	case MPA_REQ_WAIT:
3456 		__state_set(&ep->com, ABORTING);
3457 		break;
3458 	case CLOSING:
3459 	case MORIBUND:
3460 		if (ep->com.cm_id && ep->com.qp) {
3461 			attrs.next_state = C4IW_QP_STATE_ERROR;
3462 			c4iw_modify_qp(ep->com.qp->rhp,
3463 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3464 				     &attrs, 1);
3465 		}
3466 		__state_set(&ep->com, ABORTING);
3467 		break;
3468 	default:
3469 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3470 			__func__, ep, ep->hwtid, ep->com.state);
3471 		abort = 0;
3472 	}
3473 	mutex_unlock(&ep->com.mutex);
3474 	if (abort)
3475 		abort_connection(ep, NULL, GFP_KERNEL);
3476 	c4iw_put_ep(&ep->com);
3477 }
3478 
3479 static void process_timedout_eps(void)
3480 {
3481 	struct c4iw_ep *ep;
3482 
3483 	spin_lock_irq(&timeout_lock);
3484 	while (!list_empty(&timeout_list)) {
3485 		struct list_head *tmp;
3486 
3487 		tmp = timeout_list.next;
3488 		list_del(tmp);
3489 		spin_unlock_irq(&timeout_lock);
3490 		ep = list_entry(tmp, struct c4iw_ep, entry);
3491 		process_timeout(ep);
3492 		spin_lock_irq(&timeout_lock);
3493 	}
3494 	spin_unlock_irq(&timeout_lock);
3495 }
3496 
3497 static void process_work(struct work_struct *work)
3498 {
3499 	struct sk_buff *skb = NULL;
3500 	struct c4iw_dev *dev;
3501 	struct cpl_act_establish *rpl;
3502 	unsigned int opcode;
3503 	int ret;
3504 
3505 	while ((skb = skb_dequeue(&rxq))) {
3506 		rpl = cplhdr(skb);
3507 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3508 		opcode = rpl->ot.opcode;
3509 
3510 		BUG_ON(!work_handlers[opcode]);
3511 		ret = work_handlers[opcode](dev, skb);
3512 		if (!ret)
3513 			kfree_skb(skb);
3514 	}
3515 	process_timedout_eps();
3516 }
3517 
3518 static DECLARE_WORK(skb_work, process_work);
3519 
3520 static void ep_timeout(unsigned long arg)
3521 {
3522 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3523 	int kickit = 0;
3524 
3525 	spin_lock(&timeout_lock);
3526 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3527 		list_add_tail(&ep->entry, &timeout_list);
3528 		kickit = 1;
3529 	}
3530 	spin_unlock(&timeout_lock);
3531 	if (kickit)
3532 		queue_work(workq, &skb_work);
3533 }
3534 
3535 /*
3536  * All the CM events are handled on a work queue to have a safe context.
3537  */
3538 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3539 {
3540 
3541 	/*
3542 	 * Save dev in the skb->cb area.
3543 	 */
3544 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3545 
3546 	/*
3547 	 * Queue the skb and schedule the worker thread.
3548 	 */
3549 	skb_queue_tail(&rxq, skb);
3550 	queue_work(workq, &skb_work);
3551 	return 0;
3552 }
3553 
3554 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3555 {
3556 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3557 
3558 	if (rpl->status != CPL_ERR_NONE) {
3559 		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3560 		       "for tid %u\n", rpl->status, GET_TID(rpl));
3561 	}
3562 	kfree_skb(skb);
3563 	return 0;
3564 }
3565 
3566 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3567 {
3568 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3569 	struct c4iw_wr_wait *wr_waitp;
3570 	int ret;
3571 
3572 	PDBG("%s type %u\n", __func__, rpl->type);
3573 
3574 	switch (rpl->type) {
3575 	case FW6_TYPE_WR_RPL:
3576 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3577 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3578 		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3579 		if (wr_waitp)
3580 			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3581 		kfree_skb(skb);
3582 		break;
3583 	case FW6_TYPE_CQE:
3584 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3585 		sched(dev, skb);
3586 		break;
3587 	default:
3588 		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3589 		       rpl->type);
3590 		kfree_skb(skb);
3591 		break;
3592 	}
3593 	return 0;
3594 }
3595 
3596 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3597 {
3598 	struct cpl_abort_req_rss *req = cplhdr(skb);
3599 	struct c4iw_ep *ep;
3600 	struct tid_info *t = dev->rdev.lldi.tids;
3601 	unsigned int tid = GET_TID(req);
3602 
3603 	ep = lookup_tid(t, tid);
3604 	if (!ep) {
3605 		printk(KERN_WARNING MOD
3606 		       "Abort on non-existent endpoint, tid %d\n", tid);
3607 		kfree_skb(skb);
3608 		return 0;
3609 	}
3610 	if (is_neg_adv_abort(req->status)) {
3611 		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3612 		     ep->hwtid);
3613 		kfree_skb(skb);
3614 		return 0;
3615 	}
3616 	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3617 	     ep->com.state);
3618 
3619 	/*
3620 	 * Wake up any threads in rdma_init() or rdma_fini().
3621 	 * However, if we are on MPAv2 and want to retry with MPAv1
3622 	 * then, don't wake up yet.
3623 	 */
3624 	if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3625 		if (ep->com.state != MPA_REQ_SENT)
3626 			c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3627 	} else
3628 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3629 	sched(dev, skb);
3630 	return 0;
3631 }
3632 
3633 /*
3634  * Most upcalls from the T4 Core go to sched() to
3635  * schedule the processing on a work queue.
3636  */
3637 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3638 	[CPL_ACT_ESTABLISH] = sched,
3639 	[CPL_ACT_OPEN_RPL] = sched,
3640 	[CPL_RX_DATA] = sched,
3641 	[CPL_ABORT_RPL_RSS] = sched,
3642 	[CPL_ABORT_RPL] = sched,
3643 	[CPL_PASS_OPEN_RPL] = sched,
3644 	[CPL_CLOSE_LISTSRV_RPL] = sched,
3645 	[CPL_PASS_ACCEPT_REQ] = sched,
3646 	[CPL_PASS_ESTABLISH] = sched,
3647 	[CPL_PEER_CLOSE] = sched,
3648 	[CPL_CLOSE_CON_RPL] = sched,
3649 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
3650 	[CPL_RDMA_TERMINATE] = sched,
3651 	[CPL_FW4_ACK] = sched,
3652 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
3653 	[CPL_FW6_MSG] = fw6_msg,
3654 	[CPL_RX_PKT] = sched
3655 };
3656 
3657 int __init c4iw_cm_init(void)
3658 {
3659 	spin_lock_init(&timeout_lock);
3660 	skb_queue_head_init(&rxq);
3661 
3662 	workq = create_singlethread_workqueue("iw_cxgb4");
3663 	if (!workq)
3664 		return -ENOMEM;
3665 
3666 	return 0;
3667 }
3668 
3669 void __exit c4iw_cm_term(void)
3670 {
3671 	WARN_ON(!list_empty(&timeout_list));
3672 	flush_workqueue(workq);
3673 	destroy_workqueue(workq);
3674 }
3675