1	/*
2	* Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
3	*
4	* This software is available to you under a choice of one of two
5	* licenses. You may choose to be licensed under the terms of the GNU
6	* General Public License (GPL) Version 2, available from the file
7	* COPYING in the main directory of this source tree, or the
8	* OpenIB.org BSD license below:
9	*
10	* Redistribution and use in source and binary forms, with or
11	* without modification, are permitted provided that the following
12	* conditions are met:
13	*
14	* - Redistributions of source code must retain the above
15	* copyright notice, this list of conditions and the following
16	* disclaimer.
17	*
18	* - Redistributions in binary form must reproduce the above
19	* copyright notice, this list of conditions and the following
20	* disclaimer in the documentation and/or other materials
21	* provided with the distribution.
22	*
23	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30	* SOFTWARE.
31	*/
32	#include <linux/module.h>
33	#include <linux/list.h>
34	#include <linux/workqueue.h>
35	#include <linux/skbuff.h>
36	#include <linux/timer.h>
37	#include <linux/notifier.h>
38	#include <linux/inetdevice.h>
39	#include <linux/ip.h>
40	#include <linux/tcp.h>
41	#include <linux/if_vlan.h>
42
43	#include <net/neighbour.h>
44	#include <net/netevent.h>
45	#include <net/route.h>
46	#include <net/tcp.h>
47	#include <net/ip6_route.h>
48	#include <net/addrconf.h>
49
50	#include <rdma/ib_addr.h>
51
52	#include <libcxgb_cm.h>
53	#include "iw_cxgb4.h"
54	#include "clip_tbl.h"
55
56	static char *states[] = {
57	"idle",
58	"listen",
59	"connecting",
60	"mpa_wait_req",
61	"mpa_req_sent",
62	"mpa_req_rcvd",
63	"mpa_rep_sent",
64	"fpdu_mode",
65	"aborting",
66	"closing",
67	"moribund",
68	"dead",
69	NULL,
70	};
71
72	static int nocong;
73	module_param(nocong, int, 0644);
74	MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75
76	static int enable_ecn;
77	module_param(enable_ecn, int, 0644);
78	MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79
80	static int dack_mode;
81	module_param(dack_mode, int, 0644);
82	MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=0)");
83
84	uint c4iw_max_read_depth = 32;
85	module_param(c4iw_max_read_depth, int, 0644);
86	MODULE_PARM_DESC(c4iw_max_read_depth,
87	"Per-connection max ORD/IRD (default=32)");
88
89	static int enable_tcp_timestamps;
90	module_param(enable_tcp_timestamps, int, 0644);
91	MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92
93	static int enable_tcp_sack;
94	module_param(enable_tcp_sack, int, 0644);
95	MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96
97	static int enable_tcp_window_scaling = 1;
98	module_param(enable_tcp_window_scaling, int, 0644);
99	MODULE_PARM_DESC(enable_tcp_window_scaling,
100	"Enable tcp window scaling (default=1)");
101
102	static int peer2peer = 1;
103	module_param(peer2peer, int, 0644);
104	MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
105
106	static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
107	module_param(p2p_type, int, 0644);
108	MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
109	"1=RDMA_READ 0=RDMA_WRITE (default 1)");
110
111	static int ep_timeout_secs = 60;
112	module_param(ep_timeout_secs, int, 0644);
113	MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
114	"in seconds (default=60)");
115
116	static int mpa_rev = 2;
117	module_param(mpa_rev, int, 0644);
118	MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
119	"1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
120	" compliant (default=2)");
121
122	static int markers_enabled;
123	module_param(markers_enabled, int, 0644);
124	MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
125
126	static int crc_enabled = 1;
127	module_param(crc_enabled, int, 0644);
128	MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
129
130	static int rcv_win = 256 * 1024;
131	module_param(rcv_win, int, 0644);
132	MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
133
134	static int snd_win = 128 * 1024;
135	module_param(snd_win, int, 0644);
136	MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
137
138	static struct workqueue_struct *workq;
139
140	static struct sk_buff_head rxq;
141
142	static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
143	static void ep_timeout(struct timer_list *t);
144	static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145	static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
146
147	static LIST_HEAD(timeout_list);
148	static DEFINE_SPINLOCK(timeout_lock);
149
150	static void deref_cm_id(struct c4iw_ep_common *epc)
151	{
152	epc->cm_id->rem_ref(epc->cm_id);
153	epc->cm_id = NULL;
154	set_bit(nr: CM_ID_DEREFED, addr: &epc->history);
155	}
156
157	static void ref_cm_id(struct c4iw_ep_common *epc)
158	{
159	set_bit(nr: CM_ID_REFED, addr: &epc->history);
160	epc->cm_id->add_ref(epc->cm_id);
161	}
162
163	static void deref_qp(struct c4iw_ep *ep)
164	{
165	c4iw_qp_rem_ref(qp: &ep->com.qp->ibqp);
166	clear_bit(nr: QP_REFERENCED, addr: &ep->com.flags);
167	set_bit(nr: QP_DEREFED, addr: &ep->com.history);
168	}
169
170	static void ref_qp(struct c4iw_ep *ep)
171	{
172	set_bit(nr: QP_REFERENCED, addr: &ep->com.flags);
173	set_bit(nr: QP_REFED, addr: &ep->com.history);
174	c4iw_qp_add_ref(qp: &ep->com.qp->ibqp);
175	}
176
177	static void start_ep_timer(struct c4iw_ep *ep)
178	{
179	pr_debug("ep %p\n", ep);
180	if (timer_pending(timer: &ep->timer)) {
181	pr_err("%s timer already started! ep %p\n",
182	__func__, ep);
183	return;
184	}
185	clear_bit(nr: TIMEOUT, addr: &ep->com.flags);
186	c4iw_get_ep(&ep->com);
187	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
188	add_timer(timer: &ep->timer);
189	}
190
191	static int stop_ep_timer(struct c4iw_ep *ep)
192	{
193	pr_debug("ep %p stopping\n", ep);
194	del_timer_sync(timer: &ep->timer);
195	if (!test_and_set_bit(nr: TIMEOUT, addr: &ep->com.flags)) {
196	c4iw_put_ep(&ep->com);
197	return 0;
198	}
199	return 1;
200	}
201
202	static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
203	struct l2t_entry *l2e)
204	{
205	int error = 0;
206
207	if (c4iw_fatal_error(rdev)) {
208	kfree_skb(skb);
209	pr_err("%s - device in error state - dropping\n", __func__);
210	return -EIO;
211	}
212	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
213	if (error < 0)
214	kfree_skb(skb);
215	else if (error == NET_XMIT_DROP)
216	return -ENOMEM;
217	return error < 0 ? error : 0;
218	}
219
220	int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
221	{
222	int error = 0;
223
224	if (c4iw_fatal_error(rdev)) {
225	kfree_skb(skb);
226	pr_err("%s - device in error state - dropping\n", __func__);
227	return -EIO;
228	}
229	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
230	if (error < 0)
231	kfree_skb(skb);
232	return error < 0 ? error : 0;
233	}
234
235	static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
236	{
237	u32 len = roundup(sizeof(struct cpl_tid_release), 16);
238
239	skb = get_skb(skb, len, GFP_KERNEL);
240	if (!skb)
241	return;
242
243	cxgb_mk_tid_release(skb, len, hwtid, 0);
244	c4iw_ofld_send(rdev, skb);
245	return;
246	}
247
248	static void set_emss(struct c4iw_ep *ep, u16 opt)
249	{
250	ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
251	((AF_INET == ep->com.remote_addr.ss_family) ?
252	sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
253	sizeof(struct tcphdr);
254	ep->mss = ep->emss;
255	if (TCPOPT_TSTAMP_G(opt))
256	ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
257	if (ep->emss < 128)
258	ep->emss = 128;
259	if (ep->emss & 7)
260	pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
261	TCPOPT_MSS_G(opt), ep->mss, ep->emss);
262	pr_debug("mss_idx %u mss %u emss=%u\n", TCPOPT_MSS_G(opt), ep->mss,
263	ep->emss);
264	}
265
266	static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
267	{
268	enum c4iw_ep_state state;
269
270	mutex_lock(&epc->mutex);
271	state = epc->state;
272	mutex_unlock(lock: &epc->mutex);
273	return state;
274	}
275
276	static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
277	{
278	epc->state = new;
279	}
280
281	static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
282	{
283	mutex_lock(&epc->mutex);
284	pr_debug("%s -> %s\n", states[epc->state], states[new]);
285	__state_set(epc, new);
286	mutex_unlock(lock: &epc->mutex);
287	return;
288	}
289
290	static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
291	{
292	struct sk_buff *skb;
293	unsigned int i;
294	size_t len;
295
296	len = roundup(sizeof(union cpl_wr_size), 16);
297	for (i = 0; i < size; i++) {
298	skb = alloc_skb(size: len, GFP_KERNEL);
299	if (!skb)
300	goto fail;
301	skb_queue_tail(list: ep_skb_list, newsk: skb);
302	}
303	return 0;
304	fail:
305	skb_queue_purge(list: ep_skb_list);
306	return -ENOMEM;
307	}
308
309	static void *alloc_ep(int size, gfp_t gfp)
310	{
311	struct c4iw_ep_common *epc;
312
313	epc = kzalloc(size, flags: gfp);
314	if (epc) {
315	epc->wr_waitp = c4iw_alloc_wr_wait(gfp);
316	if (!epc->wr_waitp) {
317	kfree(objp: epc);
318	epc = NULL;
319	goto out;
320	}
321	kref_init(kref: &epc->kref);
322	mutex_init(&epc->mutex);
323	c4iw_init_wr_wait(wr_waitp: epc->wr_waitp);
324	}
325	pr_debug("alloc ep %p\n", epc);
326	out:
327	return epc;
328	}
329
330	static void remove_ep_tid(struct c4iw_ep *ep)
331	{
332	unsigned long flags;
333
334	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
335	__xa_erase(&ep->com.dev->hwtids, index: ep->hwtid);
336	if (xa_empty(xa: &ep->com.dev->hwtids))
337	wake_up(&ep->com.dev->wait);
338	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
339	}
340
341	static int insert_ep_tid(struct c4iw_ep *ep)
342	{
343	unsigned long flags;
344	int err;
345
346	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
347	err = __xa_insert(&ep->com.dev->hwtids, index: ep->hwtid, entry: ep, GFP_KERNEL);
348	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
349
350	return err;
351	}
352
353	/*
354	* Atomically lookup the ep ptr given the tid and grab a reference on the ep.
355	*/
356	static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
357	{
358	struct c4iw_ep *ep;
359	unsigned long flags;
360
361	xa_lock_irqsave(&dev->hwtids, flags);
362	ep = xa_load(&dev->hwtids, index: tid);
363	if (ep)
364	c4iw_get_ep(&ep->com);
365	xa_unlock_irqrestore(&dev->hwtids, flags);
366	return ep;
367	}
368
369	/*
370	* Atomically lookup the ep ptr given the stid and grab a reference on the ep.
371	*/
372	static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
373	unsigned int stid)
374	{
375	struct c4iw_listen_ep *ep;
376	unsigned long flags;
377
378	xa_lock_irqsave(&dev->stids, flags);
379	ep = xa_load(&dev->stids, index: stid);
380	if (ep)
381	c4iw_get_ep(&ep->com);
382	xa_unlock_irqrestore(&dev->stids, flags);
383	return ep;
384	}
385
386	void _c4iw_free_ep(struct kref *kref)
387	{
388	struct c4iw_ep *ep;
389
390	ep = container_of(kref, struct c4iw_ep, com.kref);
391	pr_debug("ep %p state %s\n", ep, states[ep->com.state]);
392	if (test_bit(QP_REFERENCED, &ep->com.flags))
393	deref_qp(ep);
394	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
395	if (ep->com.remote_addr.ss_family == AF_INET6) {
396	struct sockaddr_in6 *sin6 =
397	(struct sockaddr_in6 *)
398	&ep->com.local_addr;
399
400	cxgb4_clip_release(
401	ep->com.dev->rdev.lldi.ports[0],
402	(const u32 *)&sin6->sin6_addr.s6_addr,
403	1);
404	}
405	cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
406	ep->com.local_addr.ss_family);
407	dst_release(dst: ep->dst);
408	cxgb4_l2t_release(ep->l2t);
409	kfree_skb(skb: ep->mpa_skb);
410	}
411	if (!skb_queue_empty(list: &ep->com.ep_skb_list))
412	skb_queue_purge(list: &ep->com.ep_skb_list);
413	c4iw_put_wr_wait(wr_waitp: ep->com.wr_waitp);
414	kfree(objp: ep);
415	}
416
417	static void release_ep_resources(struct c4iw_ep *ep)
418	{
419	set_bit(nr: RELEASE_RESOURCES, addr: &ep->com.flags);
420
421	/*
422	* If we have a hwtid, then remove it from the idr table
423	* so lookups will no longer find this endpoint. Otherwise
424	* we have a race where one thread finds the ep ptr just
425	* before the other thread is freeing the ep memory.
426	*/
427	if (ep->hwtid != -1)
428	remove_ep_tid(ep);
429	c4iw_put_ep(&ep->com);
430	}
431
432	static int status2errno(int status)
433	{
434	switch (status) {
435	case CPL_ERR_NONE:
436	return 0;
437	case CPL_ERR_CONN_RESET:
438	return -ECONNRESET;
439	case CPL_ERR_ARP_MISS:
440	return -EHOSTUNREACH;
441	case CPL_ERR_CONN_TIMEDOUT:
442	return -ETIMEDOUT;
443	case CPL_ERR_TCAM_FULL:
444	return -ENOMEM;
445	case CPL_ERR_CONN_EXIST:
446	return -EADDRINUSE;
447	default:
448	return -EIO;
449	}
450	}
451
452	/*
453	* Try and reuse skbs already allocated...
454	*/
455	static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
456	{
457	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
458	skb_trim(skb, len: 0);
459	skb_get(skb);
460	skb_reset_transport_header(skb);
461	} else {
462	skb = alloc_skb(size: len, priority: gfp);
463	if (!skb)
464	return NULL;
465	}
466	t4_set_arp_err_handler(skb, NULL, NULL);
467	return skb;
468	}
469
470	static struct net_device *get_real_dev(struct net_device *egress_dev)
471	{
472	return rdma_vlan_dev_real_dev(dev: egress_dev) ? : egress_dev;
473	}
474
475	static void arp_failure_discard(void *handle, struct sk_buff *skb)
476	{
477	pr_err("ARP failure\n");
478	kfree_skb(skb);
479	}
480
481	static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
482	{
483	pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
484	}
485
486	enum {
487	NUM_FAKE_CPLS = 2,
488	FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
489	FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
490	};
491
492	static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
493	{
494	struct c4iw_ep *ep;
495
496	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497	release_ep_resources(ep);
498	return 0;
499	}
500
501	static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
502	{
503	struct c4iw_ep *ep;
504
505	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
506	c4iw_put_ep(&ep->parent_ep->com);
507	release_ep_resources(ep);
508	return 0;
509	}
510
511	/*
512	* Fake up a special CPL opcode and call sched() so process_work() will call
513	* _put_ep_safe() in a safe context to free the ep resources. This is needed
514	* because ARP error handlers are called in an ATOMIC context, and
515	* _c4iw_free_ep() needs to block.
516	*/
517	static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
518	int cpl)
519	{
520	struct cpl_act_establish *rpl = cplhdr(skb);
521
522	/* Set our special ARP_FAILURE opcode */
523	rpl->ot.opcode = cpl;
524
525	/*
526	* Save ep in the skb->cb area, after where sched() will save the dev
527	* ptr.
528	*/
529	*((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
530	sched(dev: ep->com.dev, skb);
531	}
532
533	/* Handle an ARP failure for an accept */
534	static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
535	{
536	struct c4iw_ep *ep = handle;
537
538	pr_err("ARP failure during accept - tid %u - dropping connection\n",
539	ep->hwtid);
540
541	__state_set(epc: &ep->com, new: DEAD);
542	queue_arp_failure_cpl(ep, skb, cpl: FAKE_CPL_PASS_PUT_EP_SAFE);
543	}
544
545	/*
546	* Handle an ARP failure for an active open.
547	*/
548	static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
549	{
550	struct c4iw_ep *ep = handle;
551
552	pr_err("ARP failure during connect\n");
553	connect_reply_upcall(ep, status: -EHOSTUNREACH);
554	__state_set(epc: &ep->com, new: DEAD);
555	if (ep->com.remote_addr.ss_family == AF_INET6) {
556	struct sockaddr_in6 *sin6 =
557	(struct sockaddr_in6 *)&ep->com.local_addr;
558	cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
559	(const u32 *)&sin6->sin6_addr.s6_addr, 1);
560	}
561	xa_erase_irq(xa: &ep->com.dev->atids, index: ep->atid);
562	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
563	queue_arp_failure_cpl(ep, skb, cpl: FAKE_CPL_PUT_EP_SAFE);
564	}
565
566	/*
567	* Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
568	* and send it along.
569	*/
570	static void abort_arp_failure(void *handle, struct sk_buff *skb)
571	{
572	int ret;
573	struct c4iw_ep *ep = handle;
574	struct c4iw_rdev *rdev = &ep->com.dev->rdev;
575	struct cpl_abort_req *req = cplhdr(skb);
576
577	pr_debug("rdev %p\n", rdev);
578	req->cmd = CPL_ABORT_NO_RST;
579	skb_get(skb);
580	ret = c4iw_ofld_send(rdev, skb);
581	if (ret) {
582	__state_set(epc: &ep->com, new: DEAD);
583	queue_arp_failure_cpl(ep, skb, cpl: FAKE_CPL_PUT_EP_SAFE);
584	} else
585	kfree_skb(skb);
586	}
587
588	static int send_flowc(struct c4iw_ep *ep)
589	{
590	struct fw_flowc_wr *flowc;
591	struct sk_buff *skb = skb_dequeue(list: &ep->com.ep_skb_list);
592	u16 vlan = ep->l2t->vlan;
593	int nparams;
594	int flowclen, flowclen16;
595
596	if (WARN_ON(!skb))
597	return -ENOMEM;
598
599	if (vlan == CPL_L2T_VLAN_NONE)
600	nparams = 9;
601	else
602	nparams = 10;
603
604	flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
605	flowclen16 = DIV_ROUND_UP(flowclen, 16);
606	flowclen = flowclen16 * 16;
607
608	flowc = __skb_put(skb, len: flowclen);
609	memset(flowc, 0, flowclen);
610
611	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
612	FW_FLOWC_WR_NPARAMS_V(nparams));
613	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
614	FW_WR_FLOWID_V(ep->hwtid));
615
616	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
617	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
618	(ep->com.dev->rdev.lldi.pf));
619	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
620	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
621	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
622	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
623	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
624	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
625	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
626	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
627	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
628	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
629	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
630	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
631	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
632	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
633	flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
634	flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale);
635	if (nparams == 10) {
636	u16 pri;
637	pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
638	flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
639	flowc->mnemval[9].val = cpu_to_be32(pri);
640	}
641
642	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
643	return c4iw_ofld_send(rdev: &ep->com.dev->rdev, skb);
644	}
645
646	static int send_halfclose(struct c4iw_ep *ep)
647	{
648	struct sk_buff *skb = skb_dequeue(list: &ep->com.ep_skb_list);
649	u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
650
651	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
652	if (WARN_ON(!skb))
653	return -ENOMEM;
654
655	cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
656	NULL, arp_failure_discard);
657
658	return c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb, l2e: ep->l2t);
659	}
660
661	static void read_tcb(struct c4iw_ep *ep)
662	{
663	struct sk_buff *skb;
664	struct cpl_get_tcb *req;
665	int wrlen = roundup(sizeof(*req), 16);
666
667	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
668	if (WARN_ON(!skb))
669	return;
670
671	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
672	req = (struct cpl_get_tcb *) skb_put(skb, len: wrlen);
673	memset(req, 0, wrlen);
674	INIT_TP_WR(req, ep->hwtid);
675	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
676	req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
677
678	/*
679	* keep a ref on the ep so the tcb is not unlocked before this
680	* cpl completes. The ref is released in read_tcb_rpl().
681	*/
682	c4iw_get_ep(&ep->com);
683	if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
684	c4iw_put_ep(&ep->com);
685	}
686
687	static int send_abort_req(struct c4iw_ep *ep)
688	{
689	u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
690	struct sk_buff *req_skb = skb_dequeue(list: &ep->com.ep_skb_list);
691
692	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
693	if (WARN_ON(!req_skb))
694	return -ENOMEM;
695
696	cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
697	ep, abort_arp_failure);
698
699	return c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb: req_skb, l2e: ep->l2t);
700	}
701
702	static int send_abort(struct c4iw_ep *ep)
703	{
704	if (!ep->com.qp || !ep->com.qp->srq) {
705	send_abort_req(ep);
706	return 0;
707	}
708	set_bit(nr: ABORT_REQ_IN_PROGRESS, addr: &ep->com.flags);
709	read_tcb(ep);
710	return 0;
711	}
712
713	static int send_connect(struct c4iw_ep *ep)
714	{
715	struct cpl_act_open_req *req = NULL;
716	struct cpl_t5_act_open_req *t5req = NULL;
717	struct cpl_t6_act_open_req *t6req = NULL;
718	struct cpl_act_open_req6 *req6 = NULL;
719	struct cpl_t5_act_open_req6 *t5req6 = NULL;
720	struct cpl_t6_act_open_req6 *t6req6 = NULL;
721	struct sk_buff *skb;
722	u64 opt0;
723	u32 opt2;
724	unsigned int mtu_idx;
725	u32 wscale;
726	int win, sizev4, sizev6, wrlen;
727	struct sockaddr_in *la = (struct sockaddr_in *)
728	&ep->com.local_addr;
729	struct sockaddr_in *ra = (struct sockaddr_in *)
730	&ep->com.remote_addr;
731	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
732	&ep->com.local_addr;
733	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
734	&ep->com.remote_addr;
735	int ret;
736	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
737	u32 isn = (get_random_u32() & ~7UL) - 1;
738	struct net_device *netdev;
739	u64 params;
740
741	netdev = ep->com.dev->rdev.lldi.ports[0];
742
743	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
744	case CHELSIO_T4:
745	sizev4 = sizeof(struct cpl_act_open_req);
746	sizev6 = sizeof(struct cpl_act_open_req6);
747	break;
748	case CHELSIO_T5:
749	sizev4 = sizeof(struct cpl_t5_act_open_req);
750	sizev6 = sizeof(struct cpl_t5_act_open_req6);
751	break;
752	case CHELSIO_T6:
753	sizev4 = sizeof(struct cpl_t6_act_open_req);
754	sizev6 = sizeof(struct cpl_t6_act_open_req6);
755	break;
756	default:
757	pr_err("T%d Chip is not supported\n",
758	CHELSIO_CHIP_VERSION(adapter_type));
759	return -EINVAL;
760	}
761
762	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
763	roundup(sizev4, 16) :
764	roundup(sizev6, 16);
765
766	pr_debug("ep %p atid %u\n", ep, ep->atid);
767
768	skb = get_skb(NULL, len: wrlen, GFP_KERNEL);
769	if (!skb) {
770	pr_err("%s - failed to alloc skb\n", __func__);
771	return -ENOMEM;
772	}
773	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
774
775	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
776	enable_tcp_timestamps,
777	(ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
778	wscale = cxgb_compute_wscale(rcv_win);
779
780	/*
781	* Specify the largest window that will fit in opt0. The
782	* remainder will be specified in the rx_data_ack.
783	*/
784	win = ep->rcv_win >> 10;
785	if (win > RCV_BUFSIZ_M)
786	win = RCV_BUFSIZ_M;
787
788	opt0 = (nocong ? NO_CONG_F : 0) |
789	KEEP_ALIVE_F |
790	DELACK_F |
791	WND_SCALE_V(wscale) |
792	MSS_IDX_V(mtu_idx) |
793	L2T_IDX_V(ep->l2t->idx) |
794	TX_CHAN_V(ep->tx_chan) |
795	SMAC_SEL_V(ep->smac_idx) |
796	DSCP_V(ep->tos >> 2) |
797	ULP_MODE_V(ULP_MODE_TCPDDP) |
798	RCV_BUFSIZ_V(win);
799	opt2 = RX_CHANNEL_V(0) |
800	CCTRL_ECN_V(enable_ecn) |
801	RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
802	if (enable_tcp_timestamps)
803	opt2 |= TSTAMPS_EN_F;
804	if (enable_tcp_sack)
805	opt2 |= SACK_EN_F;
806	if (wscale && enable_tcp_window_scaling)
807	opt2 |= WND_SCALE_EN_F;
808	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
809	if (peer2peer)
810	isn += 4;
811
812	opt2 |= T5_OPT_2_VALID_F;
813	opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
814	opt2 |= T5_ISS_F;
815	}
816
817	params = cxgb4_select_ntuple(netdev, ep->l2t);
818
819	if (ep->com.remote_addr.ss_family == AF_INET6)
820	cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
821	(const u32 *)&la6->sin6_addr.s6_addr, 1);
822
823	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
824
825	if (ep->com.remote_addr.ss_family == AF_INET) {
826	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
827	case CHELSIO_T4:
828	req = skb_put(skb, len: wrlen);
829	INIT_TP_WR(req, 0);
830	break;
831	case CHELSIO_T5:
832	t5req = skb_put(skb, len: wrlen);
833	INIT_TP_WR(t5req, 0);
834	req = (struct cpl_act_open_req *)t5req;
835	break;
836	case CHELSIO_T6:
837	t6req = skb_put(skb, len: wrlen);
838	INIT_TP_WR(t6req, 0);
839	req = (struct cpl_act_open_req *)t6req;
840	t5req = (struct cpl_t5_act_open_req *)t6req;
841	break;
842	default:
843	pr_err("T%d Chip is not supported\n",
844	CHELSIO_CHIP_VERSION(adapter_type));
845	ret = -EINVAL;
846	goto clip_release;
847	}
848
849	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
850	((ep->rss_qid<<14) | ep->atid)));
851	req->local_port = la->sin_port;
852	req->peer_port = ra->sin_port;
853	req->local_ip = la->sin_addr.s_addr;
854	req->peer_ip = ra->sin_addr.s_addr;
855	req->opt0 = cpu_to_be64(opt0);
856
857	if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
858	req->params = cpu_to_be32(params);
859	req->opt2 = cpu_to_be32(opt2);
860	} else {
861	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
862	t5req->params =
863	cpu_to_be64(FILTER_TUPLE_V(params));
864	t5req->rsvd = cpu_to_be32(isn);
865	pr_debug("snd_isn %u\n", t5req->rsvd);
866	t5req->opt2 = cpu_to_be32(opt2);
867	} else {
868	t6req->params =
869	cpu_to_be64(FILTER_TUPLE_V(params));
870	t6req->rsvd = cpu_to_be32(isn);
871	pr_debug("snd_isn %u\n", t6req->rsvd);
872	t6req->opt2 = cpu_to_be32(opt2);
873	}
874	}
875	} else {
876	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
877	case CHELSIO_T4:
878	req6 = skb_put(skb, len: wrlen);
879	INIT_TP_WR(req6, 0);
880	break;
881	case CHELSIO_T5:
882	t5req6 = skb_put(skb, len: wrlen);
883	INIT_TP_WR(t5req6, 0);
884	req6 = (struct cpl_act_open_req6 *)t5req6;
885	break;
886	case CHELSIO_T6:
887	t6req6 = skb_put(skb, len: wrlen);
888	INIT_TP_WR(t6req6, 0);
889	req6 = (struct cpl_act_open_req6 *)t6req6;
890	t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
891	break;
892	default:
893	pr_err("T%d Chip is not supported\n",
894	CHELSIO_CHIP_VERSION(adapter_type));
895	ret = -EINVAL;
896	goto clip_release;
897	}
898
899	OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
900	((ep->rss_qid<<14)|ep->atid)));
901	req6->local_port = la6->sin6_port;
902	req6->peer_port = ra6->sin6_port;
903	req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
904	req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
905	req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
906	req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
907	req6->opt0 = cpu_to_be64(opt0);
908
909	if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
910	req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
911	ep->l2t));
912	req6->opt2 = cpu_to_be32(opt2);
913	} else {
914	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
915	t5req6->params =
916	cpu_to_be64(FILTER_TUPLE_V(params));
917	t5req6->rsvd = cpu_to_be32(isn);
918	pr_debug("snd_isn %u\n", t5req6->rsvd);
919	t5req6->opt2 = cpu_to_be32(opt2);
920	} else {
921	t6req6->params =
922	cpu_to_be64(FILTER_TUPLE_V(params));
923	t6req6->rsvd = cpu_to_be32(isn);
924	pr_debug("snd_isn %u\n", t6req6->rsvd);
925	t6req6->opt2 = cpu_to_be32(opt2);
926	}
927
928	}
929	}
930
931	set_bit(nr: ACT_OPEN_REQ, addr: &ep->com.history);
932	ret = c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb, l2e: ep->l2t);
933	clip_release:
934	if (ret && ep->com.remote_addr.ss_family == AF_INET6)
935	cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
936	(const u32 *)&la6->sin6_addr.s6_addr, 1);
937	return ret;
938	}
939
940	static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
941	u8 mpa_rev_to_use)
942	{
943	int mpalen, wrlen, ret;
944	struct fw_ofld_tx_data_wr *req;
945	struct mpa_message *mpa;
946	struct mpa_v2_conn_params mpa_v2_params;
947
948	pr_debug("ep %p tid %u pd_len %d\n",
949	ep, ep->hwtid, ep->plen);
950
951	mpalen = sizeof(*mpa) + ep->plen;
952	if (mpa_rev_to_use == 2)
953	mpalen += sizeof(struct mpa_v2_conn_params);
954	wrlen = roundup(mpalen + sizeof(*req), 16);
955	skb = get_skb(skb, len: wrlen, GFP_KERNEL);
956	if (!skb) {
957	connect_reply_upcall(ep, status: -ENOMEM);
958	return -ENOMEM;
959	}
960	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
961
962	req = skb_put_zero(skb, len: wrlen);
963	req->op_to_immdlen = cpu_to_be32(
964	FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
965	FW_WR_COMPL_F |
966	FW_WR_IMMDLEN_V(mpalen));
967	req->flowid_len16 = cpu_to_be32(
968	FW_WR_FLOWID_V(ep->hwtid) |
969	FW_WR_LEN16_V(wrlen >> 4));
970	req->plen = cpu_to_be32(mpalen);
971	req->tunnel_to_proxy = cpu_to_be32(
972	FW_OFLD_TX_DATA_WR_FLUSH_F |
973	FW_OFLD_TX_DATA_WR_SHOVE_F);
974
975	mpa = (struct mpa_message *)(req + 1);
976	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
977
978	mpa->flags = 0;
979	if (crc_enabled)
980	mpa->flags |= MPA_CRC;
981	if (markers_enabled) {
982	mpa->flags |= MPA_MARKERS;
983	ep->mpa_attr.recv_marker_enabled = 1;
984	} else {
985	ep->mpa_attr.recv_marker_enabled = 0;
986	}
987	if (mpa_rev_to_use == 2)
988	mpa->flags |= MPA_ENHANCED_RDMA_CONN;
989
990	mpa->private_data_size = htons(ep->plen);
991	mpa->revision = mpa_rev_to_use;
992	if (mpa_rev_to_use == 1) {
993	ep->tried_with_mpa_v1 = 1;
994	ep->retry_with_mpa_v1 = 0;
995	}
996
997	if (mpa_rev_to_use == 2) {
998	mpa->private_data_size =
999	htons(ntohs(mpa->private_data_size) +
1000	sizeof(struct mpa_v2_conn_params));
1001	pr_debug("initiator ird %u ord %u\n", ep->ird,
1002	ep->ord);
1003	mpa_v2_params.ird = htons((u16)ep->ird);
1004	mpa_v2_params.ord = htons((u16)ep->ord);
1005
1006	if (peer2peer) {
1007	mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1008	if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1009	mpa_v2_params.ord |=
1010	htons(MPA_V2_RDMA_WRITE_RTR);
1011	else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1012	mpa_v2_params.ord |=
1013	htons(MPA_V2_RDMA_READ_RTR);
1014	}
1015	memcpy(mpa->private_data, &mpa_v2_params,
1016	sizeof(struct mpa_v2_conn_params));
1017
1018	if (ep->plen)
1019	memcpy(mpa->private_data +
1020	sizeof(struct mpa_v2_conn_params),
1021	ep->mpa_pkt + sizeof(*mpa), ep->plen);
1022	} else
1023	if (ep->plen)
1024	memcpy(mpa->private_data,
1025	ep->mpa_pkt + sizeof(*mpa), ep->plen);
1026
1027	/*
1028	* Reference the mpa skb. This ensures the data area
1029	* will remain in memory until the hw acks the tx.
1030	* Function fw4_ack() will deref it.
1031	*/
1032	skb_get(skb);
1033	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1034	ep->mpa_skb = skb;
1035	ret = c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb, l2e: ep->l2t);
1036	if (ret)
1037	return ret;
1038	start_ep_timer(ep);
1039	__state_set(epc: &ep->com, new: MPA_REQ_SENT);
1040	ep->mpa_attr.initiator = 1;
1041	ep->snd_seq += mpalen;
1042	return ret;
1043	}
1044
1045	static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1046	{
1047	int mpalen, wrlen;
1048	struct fw_ofld_tx_data_wr *req;
1049	struct mpa_message *mpa;
1050	struct sk_buff *skb;
1051	struct mpa_v2_conn_params mpa_v2_params;
1052
1053	pr_debug("ep %p tid %u pd_len %d\n",
1054	ep, ep->hwtid, ep->plen);
1055
1056	mpalen = sizeof(*mpa) + plen;
1057	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1058	mpalen += sizeof(struct mpa_v2_conn_params);
1059	wrlen = roundup(mpalen + sizeof(*req), 16);
1060
1061	skb = get_skb(NULL, len: wrlen, GFP_KERNEL);
1062	if (!skb) {
1063	pr_err("%s - cannot alloc skb!\n", __func__);
1064	return -ENOMEM;
1065	}
1066	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1067
1068	req = skb_put_zero(skb, len: wrlen);
1069	req->op_to_immdlen = cpu_to_be32(
1070	FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1071	FW_WR_COMPL_F |
1072	FW_WR_IMMDLEN_V(mpalen));
1073	req->flowid_len16 = cpu_to_be32(
1074	FW_WR_FLOWID_V(ep->hwtid) |
1075	FW_WR_LEN16_V(wrlen >> 4));
1076	req->plen = cpu_to_be32(mpalen);
1077	req->tunnel_to_proxy = cpu_to_be32(
1078	FW_OFLD_TX_DATA_WR_FLUSH_F |
1079	FW_OFLD_TX_DATA_WR_SHOVE_F);
1080
1081	mpa = (struct mpa_message *)(req + 1);
1082	memset(mpa, 0, sizeof(*mpa));
1083	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1084	mpa->flags = MPA_REJECT;
1085	mpa->revision = ep->mpa_attr.version;
1086	mpa->private_data_size = htons(plen);
1087
1088	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1089	mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1090	mpa->private_data_size =
1091	htons(ntohs(mpa->private_data_size) +
1092	sizeof(struct mpa_v2_conn_params));
1093	mpa_v2_params.ird = htons(((u16)ep->ird) |
1094	(peer2peer ? MPA_V2_PEER2PEER_MODEL :
1095	0));
1096	mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1097	(p2p_type ==
1098	FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1099	MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1100	FW_RI_INIT_P2PTYPE_READ_REQ ?
1101	MPA_V2_RDMA_READ_RTR : 0) : 0));
1102	memcpy(mpa->private_data, &mpa_v2_params,
1103	sizeof(struct mpa_v2_conn_params));
1104
1105	if (ep->plen)
1106	memcpy(mpa->private_data +
1107	sizeof(struct mpa_v2_conn_params), pdata, plen);
1108	} else
1109	if (plen)
1110	memcpy(mpa->private_data, pdata, plen);
1111
1112	/*
1113	* Reference the mpa skb again. This ensures the data area
1114	* will remain in memory until the hw acks the tx.
1115	* Function fw4_ack() will deref it.
1116	*/
1117	skb_get(skb);
1118	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1119	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1120	ep->mpa_skb = skb;
1121	ep->snd_seq += mpalen;
1122	return c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb, l2e: ep->l2t);
1123	}
1124
1125	static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1126	{
1127	int mpalen, wrlen;
1128	struct fw_ofld_tx_data_wr *req;
1129	struct mpa_message *mpa;
1130	struct sk_buff *skb;
1131	struct mpa_v2_conn_params mpa_v2_params;
1132
1133	pr_debug("ep %p tid %u pd_len %d\n",
1134	ep, ep->hwtid, ep->plen);
1135
1136	mpalen = sizeof(*mpa) + plen;
1137	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1138	mpalen += sizeof(struct mpa_v2_conn_params);
1139	wrlen = roundup(mpalen + sizeof(*req), 16);
1140
1141	skb = get_skb(NULL, len: wrlen, GFP_KERNEL);
1142	if (!skb) {
1143	pr_err("%s - cannot alloc skb!\n", __func__);
1144	return -ENOMEM;
1145	}
1146	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1147
1148	req = skb_put_zero(skb, len: wrlen);
1149	req->op_to_immdlen = cpu_to_be32(
1150	FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1151	FW_WR_COMPL_F |
1152	FW_WR_IMMDLEN_V(mpalen));
1153	req->flowid_len16 = cpu_to_be32(
1154	FW_WR_FLOWID_V(ep->hwtid) |
1155	FW_WR_LEN16_V(wrlen >> 4));
1156	req->plen = cpu_to_be32(mpalen);
1157	req->tunnel_to_proxy = cpu_to_be32(
1158	FW_OFLD_TX_DATA_WR_FLUSH_F |
1159	FW_OFLD_TX_DATA_WR_SHOVE_F);
1160
1161	mpa = (struct mpa_message *)(req + 1);
1162	memset(mpa, 0, sizeof(*mpa));
1163	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1164	mpa->flags = 0;
1165	if (ep->mpa_attr.crc_enabled)
1166	mpa->flags |= MPA_CRC;
1167	if (ep->mpa_attr.recv_marker_enabled)
1168	mpa->flags |= MPA_MARKERS;
1169	mpa->revision = ep->mpa_attr.version;
1170	mpa->private_data_size = htons(plen);
1171
1172	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1173	mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1174	mpa->private_data_size =
1175	htons(ntohs(mpa->private_data_size) +
1176	sizeof(struct mpa_v2_conn_params));
1177	mpa_v2_params.ird = htons((u16)ep->ird);
1178	mpa_v2_params.ord = htons((u16)ep->ord);
1179	if (peer2peer && (ep->mpa_attr.p2p_type !=
1180	FW_RI_INIT_P2PTYPE_DISABLED)) {
1181	mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1182
1183	if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1184	mpa_v2_params.ord |=
1185	htons(MPA_V2_RDMA_WRITE_RTR);
1186	else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1187	mpa_v2_params.ord |=
1188	htons(MPA_V2_RDMA_READ_RTR);
1189	}
1190
1191	memcpy(mpa->private_data, &mpa_v2_params,
1192	sizeof(struct mpa_v2_conn_params));
1193
1194	if (ep->plen)
1195	memcpy(mpa->private_data +
1196	sizeof(struct mpa_v2_conn_params), pdata, plen);
1197	} else
1198	if (plen)
1199	memcpy(mpa->private_data, pdata, plen);
1200
1201	/*
1202	* Reference the mpa skb. This ensures the data area
1203	* will remain in memory until the hw acks the tx.
1204	* Function fw4_ack() will deref it.
1205	*/
1206	skb_get(skb);
1207	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1208	ep->mpa_skb = skb;
1209	__state_set(epc: &ep->com, new: MPA_REP_SENT);
1210	ep->snd_seq += mpalen;
1211	return c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb, l2e: ep->l2t);
1212	}
1213
1214	static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1215	{
1216	struct c4iw_ep *ep;
1217	struct cpl_act_establish *req = cplhdr(skb);
1218	unsigned short tcp_opt = ntohs(req->tcp_opt);
1219	unsigned int tid = GET_TID(req);
1220	unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1221	struct tid_info *t = dev->rdev.lldi.tids;
1222	int ret;
1223
1224	ep = lookup_atid(t, atid);
1225
1226	pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1227	be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1228
1229	mutex_lock(&ep->com.mutex);
1230	dst_confirm(dst: ep->dst);
1231
1232	/* setup the hwtid for this connection */
1233	ep->hwtid = tid;
1234	cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1235	insert_ep_tid(ep);
1236
1237	ep->snd_seq = be32_to_cpu(req->snd_isn);
1238	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1239	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1240
1241	set_emss(ep, opt: tcp_opt);
1242
1243	/* dealloc the atid */
1244	xa_erase_irq(xa: &ep->com.dev->atids, index: atid);
1245	cxgb4_free_atid(t, atid);
1246	set_bit(nr: ACT_ESTAB, addr: &ep->com.history);
1247
1248	/* start MPA negotiation */
1249	ret = send_flowc(ep);
1250	if (ret)
1251	goto err;
1252	if (ep->retry_with_mpa_v1)
1253	ret = send_mpa_req(ep, skb, mpa_rev_to_use: 1);
1254	else
1255	ret = send_mpa_req(ep, skb, mpa_rev_to_use: mpa_rev);
1256	if (ret)
1257	goto err;
1258	mutex_unlock(lock: &ep->com.mutex);
1259	return 0;
1260	err:
1261	mutex_unlock(lock: &ep->com.mutex);
1262	connect_reply_upcall(ep, status: -ENOMEM);
1263	c4iw_ep_disconnect(ep, abrupt: 0, GFP_KERNEL);
1264	return 0;
1265	}
1266
1267	static void close_complete_upcall(struct c4iw_ep *ep, int status)
1268	{
1269	struct iw_cm_event event;
1270
1271	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1272	memset(&event, 0, sizeof(event));
1273	event.event = IW_CM_EVENT_CLOSE;
1274	event.status = status;
1275	if (ep->com.cm_id) {
1276	pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1277	ep, ep->com.cm_id, ep->hwtid);
1278	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1279	deref_cm_id(epc: &ep->com);
1280	set_bit(nr: CLOSE_UPCALL, addr: &ep->com.history);
1281	}
1282	}
1283
1284	static void peer_close_upcall(struct c4iw_ep *ep)
1285	{
1286	struct iw_cm_event event;
1287
1288	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1289	memset(&event, 0, sizeof(event));
1290	event.event = IW_CM_EVENT_DISCONNECT;
1291	if (ep->com.cm_id) {
1292	pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1293	ep, ep->com.cm_id, ep->hwtid);
1294	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1295	set_bit(nr: DISCONN_UPCALL, addr: &ep->com.history);
1296	}
1297	}
1298
1299	static void peer_abort_upcall(struct c4iw_ep *ep)
1300	{
1301	struct iw_cm_event event;
1302
1303	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1304	memset(&event, 0, sizeof(event));
1305	event.event = IW_CM_EVENT_CLOSE;
1306	event.status = -ECONNRESET;
1307	if (ep->com.cm_id) {
1308	pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1309	ep->com.cm_id, ep->hwtid);
1310	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1311	deref_cm_id(epc: &ep->com);
1312	set_bit(nr: ABORT_UPCALL, addr: &ep->com.history);
1313	}
1314	}
1315
1316	static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1317	{
1318	struct iw_cm_event event;
1319
1320	pr_debug("ep %p tid %u status %d\n",
1321	ep, ep->hwtid, status);
1322	memset(&event, 0, sizeof(event));
1323	event.event = IW_CM_EVENT_CONNECT_REPLY;
1324	event.status = status;
1325	memcpy(&event.local_addr, &ep->com.local_addr,
1326	sizeof(ep->com.local_addr));
1327	memcpy(&event.remote_addr, &ep->com.remote_addr,
1328	sizeof(ep->com.remote_addr));
1329
1330	if ((status == 0) || (status == -ECONNREFUSED)) {
1331	if (!ep->tried_with_mpa_v1) {
1332	/* this means MPA_v2 is used */
1333	event.ord = ep->ird;
1334	event.ird = ep->ord;
1335	event.private_data_len = ep->plen -
1336	sizeof(struct mpa_v2_conn_params);
1337	event.private_data = ep->mpa_pkt +
1338	sizeof(struct mpa_message) +
1339	sizeof(struct mpa_v2_conn_params);
1340	} else {
1341	/* this means MPA_v1 is used */
1342	event.ord = cur_max_read_depth(dev: ep->com.dev);
1343	event.ird = cur_max_read_depth(dev: ep->com.dev);
1344	event.private_data_len = ep->plen;
1345	event.private_data = ep->mpa_pkt +
1346	sizeof(struct mpa_message);
1347	}
1348	}
1349
1350	pr_debug("ep %p tid %u status %d\n", ep,
1351	ep->hwtid, status);
1352	set_bit(nr: CONN_RPL_UPCALL, addr: &ep->com.history);
1353	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1354
1355	if (status < 0)
1356	deref_cm_id(epc: &ep->com);
1357	}
1358
1359	static int connect_request_upcall(struct c4iw_ep *ep)
1360	{
1361	struct iw_cm_event event;
1362	int ret;
1363
1364	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1365	memset(&event, 0, sizeof(event));
1366	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1367	memcpy(&event.local_addr, &ep->com.local_addr,
1368	sizeof(ep->com.local_addr));
1369	memcpy(&event.remote_addr, &ep->com.remote_addr,
1370	sizeof(ep->com.remote_addr));
1371	event.provider_data = ep;
1372	if (!ep->tried_with_mpa_v1) {
1373	/* this means MPA_v2 is used */
1374	event.ord = ep->ord;
1375	event.ird = ep->ird;
1376	event.private_data_len = ep->plen -
1377	sizeof(struct mpa_v2_conn_params);
1378	event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1379	sizeof(struct mpa_v2_conn_params);
1380	} else {
1381	/* this means MPA_v1 is used. Send max supported */
1382	event.ord = cur_max_read_depth(dev: ep->com.dev);
1383	event.ird = cur_max_read_depth(dev: ep->com.dev);
1384	event.private_data_len = ep->plen;
1385	event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1386	}
1387	c4iw_get_ep(&ep->com);
1388	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1389	&event);
1390	if (ret)
1391	c4iw_put_ep(&ep->com);
1392	set_bit(nr: CONNREQ_UPCALL, addr: &ep->com.history);
1393	c4iw_put_ep(&ep->parent_ep->com);
1394	return ret;
1395	}
1396
1397	static void established_upcall(struct c4iw_ep *ep)
1398	{
1399	struct iw_cm_event event;
1400
1401	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1402	memset(&event, 0, sizeof(event));
1403	event.event = IW_CM_EVENT_ESTABLISHED;
1404	event.ird = ep->ord;
1405	event.ord = ep->ird;
1406	if (ep->com.cm_id) {
1407	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1408	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1409	set_bit(nr: ESTAB_UPCALL, addr: &ep->com.history);
1410	}
1411	}
1412
1413	static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1414	{
1415	struct sk_buff *skb;
1416	u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1417	u32 credit_dack;
1418
1419	pr_debug("ep %p tid %u credits %u\n",
1420	ep, ep->hwtid, credits);
1421	skb = get_skb(NULL, len: wrlen, GFP_KERNEL);
1422	if (!skb) {
1423	pr_err("update_rx_credits - cannot alloc skb!\n");
1424	return 0;
1425	}
1426
1427	/*
1428	* If we couldn't specify the entire rcv window at connection setup
1429	* due to the limit in the number of bits in the RCV_BUFSIZ field,
1430	* then add the overage in to the credits returned.
1431	*/
1432	if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1433	credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1434
1435	credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1436	RX_DACK_MODE_V(dack_mode);
1437
1438	cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1439	credit_dack);
1440
1441	c4iw_ofld_send(rdev: &ep->com.dev->rdev, skb);
1442	return credits;
1443	}
1444
1445	#define RELAXED_IRD_NEGOTIATION 1
1446
1447	/*
1448	* process_mpa_reply - process streaming mode MPA reply
1449	*
1450	* Returns:
1451	*
1452	* 0 upon success indicating a connect request was delivered to the ULP
1453	* or the mpa request is incomplete but valid so far.
1454	*
1455	* 1 if a failure requires the caller to close the connection.
1456	*
1457	* 2 if a failure requires the caller to abort the connection.
1458	*/
1459	static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1460	{
1461	struct mpa_message *mpa;
1462	struct mpa_v2_conn_params *mpa_v2_params;
1463	u16 plen;
1464	u16 resp_ird, resp_ord;
1465	u8 rtr_mismatch = 0, insuff_ird = 0;
1466	struct c4iw_qp_attributes attrs;
1467	enum c4iw_qp_attr_mask mask;
1468	int err;
1469	int disconnect = 0;
1470
1471	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1472
1473	/*
1474	* If we get more than the supported amount of private data
1475	* then we must fail this connection.
1476	*/
1477	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1478	err = -EINVAL;
1479	goto err_stop_timer;
1480	}
1481
1482	/*
1483	* copy the new data into our accumulation buffer.
1484	*/
1485	skb_copy_from_linear_data(skb, to: &(ep->mpa_pkt[ep->mpa_pkt_len]),
1486	len: skb->len);
1487	ep->mpa_pkt_len += skb->len;
1488
1489	/*
1490	* if we don't even have the mpa message, then bail.
1491	*/
1492	if (ep->mpa_pkt_len < sizeof(*mpa))
1493	return 0;
1494	mpa = (struct mpa_message *) ep->mpa_pkt;
1495
1496	/* Validate MPA header. */
1497	if (mpa->revision > mpa_rev) {
1498	pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1499	__func__, mpa_rev, mpa->revision);
1500	err = -EPROTO;
1501	goto err_stop_timer;
1502	}
1503	if (memcmp(p: mpa->key, MPA_KEY_REP, size: sizeof(mpa->key))) {
1504	err = -EPROTO;
1505	goto err_stop_timer;
1506	}
1507
1508	plen = ntohs(mpa->private_data_size);
1509
1510	/*
1511	* Fail if there's too much private data.
1512	*/
1513	if (plen > MPA_MAX_PRIVATE_DATA) {
1514	err = -EPROTO;
1515	goto err_stop_timer;
1516	}
1517
1518	/*
1519	* If plen does not account for pkt size
1520	*/
1521	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1522	err = -EPROTO;
1523	goto err_stop_timer;
1524	}
1525
1526	ep->plen = (u8) plen;
1527
1528	/*
1529	* If we don't have all the pdata yet, then bail.
1530	* We'll continue process when more data arrives.
1531	*/
1532	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1533	return 0;
1534
1535	if (mpa->flags & MPA_REJECT) {
1536	err = -ECONNREFUSED;
1537	goto err_stop_timer;
1538	}
1539
1540	/*
1541	* Stop mpa timer. If it expired, then
1542	* we ignore the MPA reply. process_timeout()
1543	* will abort the connection.
1544	*/
1545	if (stop_ep_timer(ep))
1546	return 0;
1547
1548	/*
1549	* If we get here we have accumulated the entire mpa
1550	* start reply message including private data. And
1551	* the MPA header is valid.
1552	*/
1553	__state_set(epc: &ep->com, new: FPDU_MODE);
1554	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1555	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1556	ep->mpa_attr.version = mpa->revision;
1557	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1558
1559	if (mpa->revision == 2) {
1560	ep->mpa_attr.enhanced_rdma_conn =
1561	mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1562	if (ep->mpa_attr.enhanced_rdma_conn) {
1563	mpa_v2_params = (struct mpa_v2_conn_params *)
1564	(ep->mpa_pkt + sizeof(*mpa));
1565	resp_ird = ntohs(mpa_v2_params->ird) &
1566	MPA_V2_IRD_ORD_MASK;
1567	resp_ord = ntohs(mpa_v2_params->ord) &
1568	MPA_V2_IRD_ORD_MASK;
1569	pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1570	resp_ird, resp_ord, ep->ird, ep->ord);
1571
1572	/*
1573	* This is a double-check. Ideally, below checks are
1574	* not required since ird/ord stuff has been taken
1575	* care of in c4iw_accept_cr
1576	*/
1577	if (ep->ird < resp_ord) {
1578	if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1579	ep->com.dev->rdev.lldi.max_ordird_qp)
1580	ep->ird = resp_ord;
1581	else
1582	insuff_ird = 1;
1583	} else if (ep->ird > resp_ord) {
1584	ep->ird = resp_ord;
1585	}
1586	if (ep->ord > resp_ird) {
1587	if (RELAXED_IRD_NEGOTIATION)
1588	ep->ord = resp_ird;
1589	else
1590	insuff_ird = 1;
1591	}
1592	if (insuff_ird) {
1593	err = -ENOMEM;
1594	ep->ird = resp_ord;
1595	ep->ord = resp_ird;
1596	}
1597
1598	if (ntohs(mpa_v2_params->ird) &
1599	MPA_V2_PEER2PEER_MODEL) {
1600	if (ntohs(mpa_v2_params->ord) &
1601	MPA_V2_RDMA_WRITE_RTR)
1602	ep->mpa_attr.p2p_type =
1603	FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1604	else if (ntohs(mpa_v2_params->ord) &
1605	MPA_V2_RDMA_READ_RTR)
1606	ep->mpa_attr.p2p_type =
1607	FW_RI_INIT_P2PTYPE_READ_REQ;
1608	}
1609	}
1610	} else if (mpa->revision == 1)
1611	if (peer2peer)
1612	ep->mpa_attr.p2p_type = p2p_type;
1613
1614	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1615	ep->mpa_attr.crc_enabled,
1616	ep->mpa_attr.recv_marker_enabled,
1617	ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1618	ep->mpa_attr.p2p_type, p2p_type);
1619
1620	/*
1621	* If responder's RTR does not match with that of initiator, assign
1622	* FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1623	* generated when moving QP to RTS state.
1624	* A TERM message will be sent after QP has moved to RTS state
1625	*/
1626	if ((ep->mpa_attr.version == 2) && peer2peer &&
1627	(ep->mpa_attr.p2p_type != p2p_type)) {
1628	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1629	rtr_mismatch = 1;
1630	}
1631
1632	attrs.mpa_attr = ep->mpa_attr;
1633	attrs.max_ird = ep->ird;
1634	attrs.max_ord = ep->ord;
1635	attrs.llp_stream_handle = ep;
1636	attrs.next_state = C4IW_QP_STATE_RTS;
1637
1638	mask = C4IW_QP_ATTR_NEXT_STATE |
1639	C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1640	C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1641
1642	/* bind QP and TID with INIT_WR */
1643	err = c4iw_modify_qp(rhp: ep->com.qp->rhp,
1644	qhp: ep->com.qp, mask, attrs: &attrs, internal: 1);
1645	if (err)
1646	goto err;
1647
1648	/*
1649	* If responder's RTR requirement did not match with what initiator
1650	* supports, generate TERM message
1651	*/
1652	if (rtr_mismatch) {
1653	pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1654	attrs.layer_etype = LAYER_MPA | DDP_LLP;
1655	attrs.ecode = MPA_NOMATCH_RTR;
1656	attrs.next_state = C4IW_QP_STATE_TERMINATE;
1657	attrs.send_term = 1;
1658	err = c4iw_modify_qp(rhp: ep->com.qp->rhp, qhp: ep->com.qp,
1659	mask: C4IW_QP_ATTR_NEXT_STATE, attrs: &attrs, internal: 1);
1660	err = -ENOMEM;
1661	disconnect = 1;
1662	goto out;
1663	}
1664
1665	/*
1666	* Generate TERM if initiator IRD is not sufficient for responder
1667	* provided ORD. Currently, we do the same behaviour even when
1668	* responder provided IRD is also not sufficient as regards to
1669	* initiator ORD.
1670	*/
1671	if (insuff_ird) {
1672	pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1673	attrs.layer_etype = LAYER_MPA | DDP_LLP;
1674	attrs.ecode = MPA_INSUFF_IRD;
1675	attrs.next_state = C4IW_QP_STATE_TERMINATE;
1676	attrs.send_term = 1;
1677	err = c4iw_modify_qp(rhp: ep->com.qp->rhp, qhp: ep->com.qp,
1678	mask: C4IW_QP_ATTR_NEXT_STATE, attrs: &attrs, internal: 1);
1679	err = -ENOMEM;
1680	disconnect = 1;
1681	goto out;
1682	}
1683	goto out;
1684	err_stop_timer:
1685	stop_ep_timer(ep);
1686	err:
1687	disconnect = 2;
1688	out:
1689	connect_reply_upcall(ep, status: err);
1690	return disconnect;
1691	}
1692
1693	/*
1694	* process_mpa_request - process streaming mode MPA request
1695	*
1696	* Returns:
1697	*
1698	* 0 upon success indicating a connect request was delivered to the ULP
1699	* or the mpa request is incomplete but valid so far.
1700	*
1701	* 1 if a failure requires the caller to close the connection.
1702	*
1703	* 2 if a failure requires the caller to abort the connection.
1704	*/
1705	static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1706	{
1707	struct mpa_message *mpa;
1708	struct mpa_v2_conn_params *mpa_v2_params;
1709	u16 plen;
1710
1711	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1712
1713	/*
1714	* If we get more than the supported amount of private data
1715	* then we must fail this connection.
1716	*/
1717	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1718	goto err_stop_timer;
1719
1720	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1721
1722	/*
1723	* Copy the new data into our accumulation buffer.
1724	*/
1725	skb_copy_from_linear_data(skb, to: &(ep->mpa_pkt[ep->mpa_pkt_len]),
1726	len: skb->len);
1727	ep->mpa_pkt_len += skb->len;
1728
1729	/*
1730	* If we don't even have the mpa message, then bail.
1731	* We'll continue process when more data arrives.
1732	*/
1733	if (ep->mpa_pkt_len < sizeof(*mpa))
1734	return 0;
1735
1736	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1737	mpa = (struct mpa_message *) ep->mpa_pkt;
1738
1739	/*
1740	* Validate MPA Header.
1741	*/
1742	if (mpa->revision > mpa_rev) {
1743	pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1744	__func__, mpa_rev, mpa->revision);
1745	goto err_stop_timer;
1746	}
1747
1748	if (memcmp(p: mpa->key, MPA_KEY_REQ, size: sizeof(mpa->key)))
1749	goto err_stop_timer;
1750
1751	plen = ntohs(mpa->private_data_size);
1752
1753	/*
1754	* Fail if there's too much private data.
1755	*/
1756	if (plen > MPA_MAX_PRIVATE_DATA)
1757	goto err_stop_timer;
1758
1759	/*
1760	* If plen does not account for pkt size
1761	*/
1762	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1763	goto err_stop_timer;
1764	ep->plen = (u8) plen;
1765
1766	/*
1767	* If we don't have all the pdata yet, then bail.
1768	*/
1769	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1770	return 0;
1771
1772	/*
1773	* If we get here we have accumulated the entire mpa
1774	* start reply message including private data.
1775	*/
1776	ep->mpa_attr.initiator = 0;
1777	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1778	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1779	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1780	ep->mpa_attr.version = mpa->revision;
1781	if (mpa->revision == 1)
1782	ep->tried_with_mpa_v1 = 1;
1783	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1784
1785	if (mpa->revision == 2) {
1786	ep->mpa_attr.enhanced_rdma_conn =
1787	mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1788	if (ep->mpa_attr.enhanced_rdma_conn) {
1789	mpa_v2_params = (struct mpa_v2_conn_params *)
1790	(ep->mpa_pkt + sizeof(*mpa));
1791	ep->ird = ntohs(mpa_v2_params->ird) &
1792	MPA_V2_IRD_ORD_MASK;
1793	ep->ird = min_t(u32, ep->ird,
1794	cur_max_read_depth(ep->com.dev));
1795	ep->ord = ntohs(mpa_v2_params->ord) &
1796	MPA_V2_IRD_ORD_MASK;
1797	ep->ord = min_t(u32, ep->ord,
1798	cur_max_read_depth(ep->com.dev));
1799	pr_debug("initiator ird %u ord %u\n",
1800	ep->ird, ep->ord);
1801	if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1802	if (peer2peer) {
1803	if (ntohs(mpa_v2_params->ord) &
1804	MPA_V2_RDMA_WRITE_RTR)
1805	ep->mpa_attr.p2p_type =
1806	FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1807	else if (ntohs(mpa_v2_params->ord) &
1808	MPA_V2_RDMA_READ_RTR)
1809	ep->mpa_attr.p2p_type =
1810	FW_RI_INIT_P2PTYPE_READ_REQ;
1811	}
1812	}
1813	} else if (mpa->revision == 1)
1814	if (peer2peer)
1815	ep->mpa_attr.p2p_type = p2p_type;
1816
1817	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1818	ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1819	ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1820	ep->mpa_attr.p2p_type);
1821
1822	__state_set(epc: &ep->com, new: MPA_REQ_RCVD);
1823
1824	/* drive upcall */
1825	mutex_lock_nested(lock: &ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1826	if (ep->parent_ep->com.state != DEAD) {
1827	if (connect_request_upcall(ep))
1828	goto err_unlock_parent;
1829	} else {
1830	goto err_unlock_parent;
1831	}
1832	mutex_unlock(lock: &ep->parent_ep->com.mutex);
1833	return 0;
1834
1835	err_unlock_parent:
1836	mutex_unlock(lock: &ep->parent_ep->com.mutex);
1837	goto err_out;
1838	err_stop_timer:
1839	(void)stop_ep_timer(ep);
1840	err_out:
1841	return 2;
1842	}
1843
1844	static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1845	{
1846	struct c4iw_ep *ep;
1847	struct cpl_rx_data *hdr = cplhdr(skb);
1848	unsigned int dlen = ntohs(hdr->len);
1849	unsigned int tid = GET_TID(hdr);
1850	__u8 status = hdr->status;
1851	int disconnect = 0;
1852
1853	ep = get_ep_from_tid(dev, tid);
1854	if (!ep)
1855	return 0;
1856	pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1857	skb_pull(skb, sizeof(*hdr));
1858	skb_trim(skb, len: dlen);
1859	mutex_lock(&ep->com.mutex);
1860
1861	switch (ep->com.state) {
1862	case MPA_REQ_SENT:
1863	update_rx_credits(ep, credits: dlen);
1864	ep->rcv_seq += dlen;
1865	disconnect = process_mpa_reply(ep, skb);
1866	break;
1867	case MPA_REQ_WAIT:
1868	update_rx_credits(ep, credits: dlen);
1869	ep->rcv_seq += dlen;
1870	disconnect = process_mpa_request(ep, skb);
1871	break;
1872	case FPDU_MODE: {
1873	struct c4iw_qp_attributes attrs;
1874
1875	update_rx_credits(ep, credits: dlen);
1876	if (status)
1877	pr_err("%s Unexpected streaming data." \
1878	" qpid %u ep %p state %d tid %u status %d\n",
1879	__func__, ep->com.qp->wq.sq.qid, ep,
1880	ep->com.state, ep->hwtid, status);
1881	attrs.next_state = C4IW_QP_STATE_TERMINATE;
1882	c4iw_modify_qp(rhp: ep->com.qp->rhp, qhp: ep->com.qp,
1883	mask: C4IW_QP_ATTR_NEXT_STATE, attrs: &attrs, internal: 1);
1884	disconnect = 1;
1885	break;
1886	}
1887	default:
1888	break;
1889	}
1890	mutex_unlock(lock: &ep->com.mutex);
1891	if (disconnect)
1892	c4iw_ep_disconnect(ep, abrupt: disconnect == 2, GFP_KERNEL);
1893	c4iw_put_ep(&ep->com);
1894	return 0;
1895	}
1896
1897	static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1898	{
1899	enum chip_type adapter_type;
1900
1901	adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1902
1903	/*
1904	* If this TCB had a srq buffer cached, then we must complete
1905	* it. For user mode, that means saving the srqidx in the
1906	* user/kernel status page for this qp. For kernel mode, just
1907	* synthesize the CQE now.
1908	*/
1909	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1910	if (ep->com.qp->ibqp.uobject)
1911	t4_set_wq_in_error(wq: &ep->com.qp->wq, srqidx);
1912	else
1913	c4iw_flush_srqidx(qhp: ep->com.qp, srqidx);
1914	}
1915	}
1916
1917	static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1918	{
1919	u32 srqidx;
1920	struct c4iw_ep *ep;
1921	struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1922	int release = 0;
1923	unsigned int tid = GET_TID(rpl);
1924
1925	ep = get_ep_from_tid(dev, tid);
1926	if (!ep) {
1927	pr_warn("Abort rpl to freed endpoint\n");
1928	return 0;
1929	}
1930
1931	if (ep->com.qp && ep->com.qp->srq) {
1932	srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1933	complete_cached_srq_buffers(ep, srqidx: srqidx ? srqidx : ep->srqe_idx);
1934	}
1935
1936	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1937	mutex_lock(&ep->com.mutex);
1938	switch (ep->com.state) {
1939	case ABORTING:
1940	c4iw_wake_up_noref(wr_waitp: ep->com.wr_waitp, ret: -ECONNRESET);
1941	__state_set(epc: &ep->com, new: DEAD);
1942	release = 1;
1943	break;
1944	default:
1945	pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1946	break;
1947	}
1948	mutex_unlock(lock: &ep->com.mutex);
1949
1950	if (release) {
1951	close_complete_upcall(ep, status: -ECONNRESET);
1952	release_ep_resources(ep);
1953	}
1954	c4iw_put_ep(&ep->com);
1955	return 0;
1956	}
1957
1958	static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1959	{
1960	struct sk_buff *skb;
1961	struct fw_ofld_connection_wr *req;
1962	unsigned int mtu_idx;
1963	u32 wscale;
1964	struct sockaddr_in *sin;
1965	int win;
1966
1967	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1968	if (!skb)
1969	return -ENOMEM;
1970
1971	req = __skb_put_zero(skb, sizeof(*req));
1972	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1973	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1974	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1975	ep->com.dev->rdev.lldi.ports[0],
1976	ep->l2t));
1977	sin = (struct sockaddr_in *)&ep->com.local_addr;
1978	req->le.lport = sin->sin_port;
1979	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1980	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1981	req->le.pport = sin->sin_port;
1982	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1983	req->tcb.t_state_to_astid =
1984	htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1985	FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1986	req->tcb.cplrxdataack_cplpassacceptrpl =
1987	htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1988	req->tcb.tx_max = (__force __be32) jiffies;
1989	req->tcb.rcv_adv = htons(1);
1990	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1991	enable_tcp_timestamps,
1992	(ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1993	wscale = cxgb_compute_wscale(rcv_win);
1994
1995	/*
1996	* Specify the largest window that will fit in opt0. The
1997	* remainder will be specified in the rx_data_ack.
1998	*/
1999	win = ep->rcv_win >> 10;
2000	if (win > RCV_BUFSIZ_M)
2001	win = RCV_BUFSIZ_M;
2002
2003	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2004	(nocong ? NO_CONG_F : 0) |
2005	KEEP_ALIVE_F |
2006	DELACK_F |
2007	WND_SCALE_V(wscale) |
2008	MSS_IDX_V(mtu_idx) |
2009	L2T_IDX_V(ep->l2t->idx) |
2010	TX_CHAN_V(ep->tx_chan) |
2011	SMAC_SEL_V(ep->smac_idx) |
2012	DSCP_V(ep->tos >> 2) |
2013	ULP_MODE_V(ULP_MODE_TCPDDP) |
2014	RCV_BUFSIZ_V(win));
2015	req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2016	TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2017	RX_CHANNEL_V(0) |
2018	CCTRL_ECN_V(enable_ecn) |
2019	RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2020	if (enable_tcp_timestamps)
2021	req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2022	if (enable_tcp_sack)
2023	req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2024	if (wscale && enable_tcp_window_scaling)
2025	req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2026	req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2027	req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2028	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2029	set_bit(nr: ACT_OFLD_CONN, addr: &ep->com.history);
2030	return c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb, l2e: ep->l2t);
2031	}
2032
2033	/*
2034	* Some of the error codes above implicitly indicate that there is no TID
2035	* allocated with the result of an ACT_OPEN. We use this predicate to make
2036	* that explicit.
2037	*/
2038	static inline int act_open_has_tid(int status)
2039	{
2040	return (status != CPL_ERR_TCAM_PARITY &&
2041	status != CPL_ERR_TCAM_MISS &&
2042	status != CPL_ERR_TCAM_FULL &&
2043	status != CPL_ERR_CONN_EXIST_SYNRECV &&
2044	status != CPL_ERR_CONN_EXIST);
2045	}
2046
2047	static char *neg_adv_str(unsigned int status)
2048	{
2049	switch (status) {
2050	case CPL_ERR_RTX_NEG_ADVICE:
2051	return "Retransmit timeout";
2052	case CPL_ERR_PERSIST_NEG_ADVICE:
2053	return "Persist timeout";
2054	case CPL_ERR_KEEPALV_NEG_ADVICE:
2055	return "Keepalive timeout";
2056	default:
2057	return "Unknown";
2058	}
2059	}
2060
2061	static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2062	{
2063	ep->snd_win = snd_win;
2064	ep->rcv_win = rcv_win;
2065	pr_debug("snd_win %d rcv_win %d\n",
2066	ep->snd_win, ep->rcv_win);
2067	}
2068
2069	#define ACT_OPEN_RETRY_COUNT 2
2070
2071	static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2072	struct dst_entry *dst, struct c4iw_dev *cdev,
2073	bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2074	{
2075	struct neighbour *n;
2076	int err, step;
2077	struct net_device *pdev;
2078
2079	n = dst_neigh_lookup(dst, daddr: peer_ip);
2080	if (!n)
2081	return -ENODEV;
2082
2083	rcu_read_lock();
2084	err = -ENOMEM;
2085	if (n->dev->flags & IFF_LOOPBACK) {
2086	if (iptype == 4)
2087	pdev = ip_dev_find(net: &init_net, addr: *(__be32 *)peer_ip);
2088	else if (IS_ENABLED(CONFIG_IPV6))
2089	for_each_netdev(&init_net, pdev) {
2090	if (ipv6_chk_addr(net: &init_net,
2091	addr: (struct in6_addr *)peer_ip,
2092	dev: pdev, strict: 1))
2093	break;
2094	}
2095	else
2096	pdev = NULL;
2097
2098	if (!pdev) {
2099	err = -ENODEV;
2100	goto out;
2101	}
2102	ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2103	n, pdev, rt_tos2priority(tos));
2104	if (!ep->l2t) {
2105	dev_put(dev: pdev);
2106	goto out;
2107	}
2108	ep->mtu = pdev->mtu;
2109	ep->tx_chan = cxgb4_port_chan(pdev);
2110	ep->smac_idx = ((struct port_info *)netdev_priv(dev: pdev))->smt_idx;
2111	step = cdev->rdev.lldi.ntxq /
2112	cdev->rdev.lldi.nchan;
2113	ep->txq_idx = cxgb4_port_idx(pdev) * step;
2114	step = cdev->rdev.lldi.nrxq /
2115	cdev->rdev.lldi.nchan;
2116	ep->ctrlq_idx = cxgb4_port_idx(pdev);
2117	ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2118	cxgb4_port_idx(pdev) * step];
2119	set_tcp_window(ep, pi: (struct port_info *)netdev_priv(dev: pdev));
2120	dev_put(dev: pdev);
2121	} else {
2122	pdev = get_real_dev(egress_dev: n->dev);
2123	ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2124	n, pdev, rt_tos2priority(tos));
2125	if (!ep->l2t)
2126	goto out;
2127	ep->mtu = dst_mtu(dst);
2128	ep->tx_chan = cxgb4_port_chan(pdev);
2129	ep->smac_idx = ((struct port_info *)netdev_priv(dev: pdev))->smt_idx;
2130	step = cdev->rdev.lldi.ntxq /
2131	cdev->rdev.lldi.nchan;
2132	ep->txq_idx = cxgb4_port_idx(pdev) * step;
2133	ep->ctrlq_idx = cxgb4_port_idx(pdev);
2134	step = cdev->rdev.lldi.nrxq /
2135	cdev->rdev.lldi.nchan;
2136	ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2137	cxgb4_port_idx(pdev) * step];
2138	set_tcp_window(ep, pi: (struct port_info *)netdev_priv(dev: pdev));
2139
2140	if (clear_mpa_v1) {
2141	ep->retry_with_mpa_v1 = 0;
2142	ep->tried_with_mpa_v1 = 0;
2143	}
2144	}
2145	err = 0;
2146	out:
2147	rcu_read_unlock();
2148
2149	neigh_release(neigh: n);
2150
2151	return err;
2152	}
2153
2154	static int c4iw_reconnect(struct c4iw_ep *ep)
2155	{
2156	int err = 0;
2157	int size = 0;
2158	struct sockaddr_in *laddr = (struct sockaddr_in *)
2159	&ep->com.cm_id->m_local_addr;
2160	struct sockaddr_in *raddr = (struct sockaddr_in *)
2161	&ep->com.cm_id->m_remote_addr;
2162	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2163	&ep->com.cm_id->m_local_addr;
2164	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2165	&ep->com.cm_id->m_remote_addr;
2166	int iptype;
2167	__u8 *ra;
2168
2169	pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2170	c4iw_init_wr_wait(wr_waitp: ep->com.wr_waitp);
2171
2172	/* When MPA revision is different on nodes, the node with MPA_rev=2
2173	* tries to reconnect with MPA_rev 1 for the same EP through
2174	* c4iw_reconnect(), where the same EP is assigned with new tid for
2175	* further connection establishment. As we are using the same EP pointer
2176	* for reconnect, few skbs are used during the previous c4iw_connect(),
2177	* which leaves the EP with inadequate skbs for further
2178	* c4iw_reconnect(), Further causing a crash due to an empty
2179	* skb_list() during peer_abort(). Allocate skbs which is already used.
2180	*/
2181	size = (CN_MAX_CON_BUF - skb_queue_len(list_: &ep->com.ep_skb_list));
2182	if (alloc_ep_skb_list(ep_skb_list: &ep->com.ep_skb_list, size)) {
2183	err = -ENOMEM;
2184	goto fail1;
2185	}
2186
2187	/*
2188	* Allocate an active TID to initiate a TCP connection.
2189	*/
2190	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2191	if (ep->atid == -1) {
2192	pr_err("%s - cannot alloc atid\n", __func__);
2193	err = -ENOMEM;
2194	goto fail2;
2195	}
2196	err = xa_insert_irq(xa: &ep->com.dev->atids, index: ep->atid, entry: ep, GFP_KERNEL);
2197	if (err)
2198	goto fail2a;
2199
2200	/* find a route */
2201	if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2202	ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2203	laddr->sin_addr.s_addr,
2204	raddr->sin_addr.s_addr,
2205	laddr->sin_port,
2206	raddr->sin_port, ep->com.cm_id->tos);
2207	iptype = 4;
2208	ra = (__u8 *)&raddr->sin_addr;
2209	} else {
2210	ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2211	get_real_dev,
2212	laddr6->sin6_addr.s6_addr,
2213	raddr6->sin6_addr.s6_addr,
2214	laddr6->sin6_port,
2215	raddr6->sin6_port,
2216	ep->com.cm_id->tos,
2217	raddr6->sin6_scope_id);
2218	iptype = 6;
2219	ra = (__u8 *)&raddr6->sin6_addr;
2220	}
2221	if (!ep->dst) {
2222	pr_err("%s - cannot find route\n", __func__);
2223	err = -EHOSTUNREACH;
2224	goto fail3;
2225	}
2226	err = import_ep(ep, iptype, peer_ip: ra, dst: ep->dst, cdev: ep->com.dev, clear_mpa_v1: false,
2227	adapter_type: ep->com.dev->rdev.lldi.adapter_type,
2228	tos: ep->com.cm_id->tos);
2229	if (err) {
2230	pr_err("%s - cannot alloc l2e\n", __func__);
2231	goto fail4;
2232	}
2233
2234	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2235	ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2236	ep->l2t->idx);
2237
2238	state_set(epc: &ep->com, new: CONNECTING);
2239	ep->tos = ep->com.cm_id->tos;
2240
2241	/* send connect request to rnic */
2242	err = send_connect(ep);
2243	if (!err)
2244	goto out;
2245
2246	cxgb4_l2t_release(ep->l2t);
2247	fail4:
2248	dst_release(dst: ep->dst);
2249	fail3:
2250	xa_erase_irq(xa: &ep->com.dev->atids, index: ep->atid);
2251	fail2a:
2252	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2253	fail2:
2254	/*
2255	* remember to send notification to upper layer.
2256	* We are in here so the upper layer is not aware that this is
2257	* re-connect attempt and so, upper layer is still waiting for
2258	* response of 1st connect request.
2259	*/
2260	connect_reply_upcall(ep, status: -ECONNRESET);
2261	fail1:
2262	c4iw_put_ep(&ep->com);
2263	out:
2264	return err;
2265	}
2266
2267	static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2268	{
2269	struct c4iw_ep *ep;
2270	struct cpl_act_open_rpl *rpl = cplhdr(skb);
2271	unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2272	ntohl(rpl->atid_status)));
2273	struct tid_info *t = dev->rdev.lldi.tids;
2274	int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2275	struct sockaddr_in *la;
2276	struct sockaddr_in *ra;
2277	struct sockaddr_in6 *la6;
2278	struct sockaddr_in6 *ra6;
2279	int ret = 0;
2280
2281	ep = lookup_atid(t, atid);
2282	la = (struct sockaddr_in *)&ep->com.local_addr;
2283	ra = (struct sockaddr_in *)&ep->com.remote_addr;
2284	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2285	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2286
2287	pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2288	status, status2errno(status));
2289
2290	if (cxgb_is_neg_adv(status)) {
2291	pr_debug("Connection problems for atid %u status %u (%s)\n",
2292	atid, status, neg_adv_str(status));
2293	ep->stats.connect_neg_adv++;
2294	mutex_lock(&dev->rdev.stats.lock);
2295	dev->rdev.stats.neg_adv++;
2296	mutex_unlock(lock: &dev->rdev.stats.lock);
2297	return 0;
2298	}
2299
2300	set_bit(nr: ACT_OPEN_RPL, addr: &ep->com.history);
2301
2302	/*
2303	* Log interesting failures.
2304	*/
2305	switch (status) {
2306	case CPL_ERR_CONN_RESET:
2307	case CPL_ERR_CONN_TIMEDOUT:
2308	break;
2309	case CPL_ERR_TCAM_FULL:
2310	mutex_lock(&dev->rdev.stats.lock);
2311	dev->rdev.stats.tcam_full++;
2312	mutex_unlock(lock: &dev->rdev.stats.lock);
2313	if (ep->com.local_addr.ss_family == AF_INET &&
2314	dev->rdev.lldi.enable_fw_ofld_conn) {
2315	ret = send_fw_act_open_req(ep, atid: TID_TID_G(AOPEN_ATID_G(
2316	ntohl(rpl->atid_status))));
2317	if (ret)
2318	goto fail;
2319	return 0;
2320	}
2321	break;
2322	case CPL_ERR_CONN_EXIST:
2323	if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2324	set_bit(nr: ACT_RETRY_INUSE, addr: &ep->com.history);
2325	if (ep->com.remote_addr.ss_family == AF_INET6) {
2326	struct sockaddr_in6 *sin6 =
2327	(struct sockaddr_in6 *)
2328	&ep->com.local_addr;
2329	cxgb4_clip_release(
2330	ep->com.dev->rdev.lldi.ports[0],
2331	(const u32 *)
2332	&sin6->sin6_addr.s6_addr, 1);
2333	}
2334	xa_erase_irq(xa: &ep->com.dev->atids, index: atid);
2335	cxgb4_free_atid(t, atid);
2336	dst_release(dst: ep->dst);
2337	cxgb4_l2t_release(ep->l2t);
2338	c4iw_reconnect(ep);
2339	return 0;
2340	}
2341	break;
2342	default:
2343	if (ep->com.local_addr.ss_family == AF_INET) {
2344	pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2345	atid, status, status2errno(status),
2346	&la->sin_addr.s_addr, ntohs(la->sin_port),
2347	&ra->sin_addr.s_addr, ntohs(ra->sin_port));
2348	} else {
2349	pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2350	atid, status, status2errno(status),
2351	la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2352	ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2353	}
2354	break;
2355	}
2356
2357	fail:
2358	connect_reply_upcall(ep, status: status2errno(status));
2359	state_set(epc: &ep->com, new: DEAD);
2360
2361	if (ep->com.remote_addr.ss_family == AF_INET6) {
2362	struct sockaddr_in6 *sin6 =
2363	(struct sockaddr_in6 *)&ep->com.local_addr;
2364	cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2365	(const u32 *)&sin6->sin6_addr.s6_addr, 1);
2366	}
2367	if (status && act_open_has_tid(status))
2368	cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2369	ep->com.local_addr.ss_family);
2370
2371	xa_erase_irq(xa: &ep->com.dev->atids, index: atid);
2372	cxgb4_free_atid(t, atid);
2373	dst_release(dst: ep->dst);
2374	cxgb4_l2t_release(ep->l2t);
2375	c4iw_put_ep(&ep->com);
2376
2377	return 0;
2378	}
2379
2380	static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2381	{
2382	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2383	unsigned int stid = GET_TID(rpl);
2384	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2385
2386	if (!ep) {
2387	pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2388	goto out;
2389	}
2390	pr_debug("ep %p status %d error %d\n", ep,
2391	rpl->status, status2errno(rpl->status));
2392	c4iw_wake_up_noref(wr_waitp: ep->com.wr_waitp, ret: status2errno(status: rpl->status));
2393	c4iw_put_ep(&ep->com);
2394	out:
2395	return 0;
2396	}
2397
2398	static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2399	{
2400	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2401	unsigned int stid = GET_TID(rpl);
2402	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2403
2404	if (!ep) {
2405	pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2406	goto out;
2407	}
2408	pr_debug("ep %p\n", ep);
2409	c4iw_wake_up_noref(wr_waitp: ep->com.wr_waitp, ret: status2errno(status: rpl->status));
2410	c4iw_put_ep(&ep->com);
2411	out:
2412	return 0;
2413	}
2414
2415	static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2416	struct cpl_pass_accept_req *req)
2417	{
2418	struct cpl_pass_accept_rpl *rpl;
2419	unsigned int mtu_idx;
2420	u64 opt0;
2421	u32 opt2;
2422	u32 wscale;
2423	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2424	int win;
2425	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2426
2427	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2428	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2429	enable_tcp_timestamps && req->tcpopt.tstamp,
2430	(ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2431	wscale = cxgb_compute_wscale(rcv_win);
2432
2433	/*
2434	* Specify the largest window that will fit in opt0. The
2435	* remainder will be specified in the rx_data_ack.
2436	*/
2437	win = ep->rcv_win >> 10;
2438	if (win > RCV_BUFSIZ_M)
2439	win = RCV_BUFSIZ_M;
2440	opt0 = (nocong ? NO_CONG_F : 0) |
2441	KEEP_ALIVE_F |
2442	DELACK_F |
2443	WND_SCALE_V(wscale) |
2444	MSS_IDX_V(mtu_idx) |
2445	L2T_IDX_V(ep->l2t->idx) |
2446	TX_CHAN_V(ep->tx_chan) |
2447	SMAC_SEL_V(ep->smac_idx) |
2448	DSCP_V(ep->tos >> 2) |
2449	ULP_MODE_V(ULP_MODE_TCPDDP) |
2450	RCV_BUFSIZ_V(win);
2451	opt2 = RX_CHANNEL_V(0) |
2452	RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2453
2454	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2455	opt2 |= TSTAMPS_EN_F;
2456	if (enable_tcp_sack && req->tcpopt.sack)
2457	opt2 |= SACK_EN_F;
2458	if (wscale && enable_tcp_window_scaling)
2459	opt2 |= WND_SCALE_EN_F;
2460	if (enable_ecn) {
2461	const struct tcphdr *tcph;
2462	u32 hlen = ntohl(req->hdr_len);
2463
2464	if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2465	tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2466	IP_HDR_LEN_G(hlen);
2467	else
2468	tcph = (const void *)(req + 1) +
2469	T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2470	if (tcph->ece && tcph->cwr)
2471	opt2 |= CCTRL_ECN_V(1);
2472	}
2473
2474	if (!is_t4(adapter_type)) {
2475	u32 isn = (get_random_u32() & ~7UL) - 1;
2476
2477	skb = get_skb(skb, roundup(sizeof(*rpl5), 16), GFP_KERNEL);
2478	rpl5 = __skb_put_zero(skb, roundup(sizeof(*rpl5), 16));
2479	rpl = (void *)rpl5;
2480	INIT_TP_WR_CPL(rpl5, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2481	opt2 |= T5_OPT_2_VALID_F;
2482	opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2483	opt2 |= T5_ISS_F;
2484	if (peer2peer)
2485	isn += 4;
2486	rpl5->iss = cpu_to_be32(isn);
2487	pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2488	} else {
2489	skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2490	rpl = __skb_put_zero(skb, sizeof(*rpl));
2491	INIT_TP_WR_CPL(rpl, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2492	}
2493
2494	rpl->opt0 = cpu_to_be64(opt0);
2495	rpl->opt2 = cpu_to_be32(opt2);
2496	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2497	t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2498
2499	return c4iw_l2t_send(rdev: &ep->com.dev->rdev, skb, l2e: ep->l2t);
2500	}
2501
2502	static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2503	{
2504	pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2505	skb_trim(skb, sizeof(struct cpl_tid_release));
2506	release_tid(rdev: &dev->rdev, hwtid, skb);
2507	return;
2508	}
2509
2510	static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2511	{
2512	struct c4iw_ep *child_ep = NULL, *parent_ep;
2513	struct cpl_pass_accept_req *req = cplhdr(skb);
2514	unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2515	struct tid_info *t = dev->rdev.lldi.tids;
2516	unsigned int hwtid = GET_TID(req);
2517	struct dst_entry *dst;
2518	__u8 local_ip[16], peer_ip[16];
2519	__be16 local_port, peer_port;
2520	struct sockaddr_in6 *sin6;
2521	int err;
2522	u16 peer_mss = ntohs(req->tcpopt.mss);
2523	int iptype;
2524	unsigned short hdrs;
2525	u8 tos;
2526
2527	parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2528	if (!parent_ep) {
2529	pr_err("%s connect request on invalid stid %d\n",
2530	__func__, stid);
2531	goto reject;
2532	}
2533
2534	if (state_read(epc: &parent_ep->com) != LISTEN) {
2535	pr_err("%s - listening ep not in LISTEN\n", __func__);
2536	goto reject;
2537	}
2538
2539	if (parent_ep->com.cm_id->tos_set)
2540	tos = parent_ep->com.cm_id->tos;
2541	else
2542	tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2543
2544	cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2545	&iptype, local_ip, peer_ip, &local_port, &peer_port);
2546
2547	/* Find output route */
2548	if (iptype == 4) {
2549	pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2550	, parent_ep, hwtid,
2551	local_ip, peer_ip, ntohs(local_port),
2552	ntohs(peer_port), peer_mss);
2553	dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2554	*(__be32 *)local_ip, *(__be32 *)peer_ip,
2555	local_port, peer_port, tos);
2556	} else {
2557	pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2558	, parent_ep, hwtid,
2559	local_ip, peer_ip, ntohs(local_port),
2560	ntohs(peer_port), peer_mss);
2561	dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2562	local_ip, peer_ip, local_port, peer_port,
2563	tos,
2564	((struct sockaddr_in6 *)
2565	&parent_ep->com.local_addr)->sin6_scope_id);
2566	}
2567	if (!dst) {
2568	pr_err("%s - failed to find dst entry!\n", __func__);
2569	goto reject;
2570	}
2571
2572	child_ep = alloc_ep(size: sizeof(*child_ep), GFP_KERNEL);
2573	if (!child_ep) {
2574	pr_err("%s - failed to allocate ep entry!\n", __func__);
2575	dst_release(dst);
2576	goto reject;
2577	}
2578
2579	err = import_ep(ep: child_ep, iptype, peer_ip, dst, cdev: dev, clear_mpa_v1: false,
2580	adapter_type: parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2581	if (err) {
2582	pr_err("%s - failed to allocate l2t entry!\n", __func__);
2583	dst_release(dst);
2584	kfree(objp: child_ep);
2585	goto reject;
2586	}
2587
2588	hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2589	sizeof(struct tcphdr) +
2590	((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2591	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2592	child_ep->mtu = peer_mss + hdrs;
2593
2594	skb_queue_head_init(list: &child_ep->com.ep_skb_list);
2595	if (alloc_ep_skb_list(ep_skb_list: &child_ep->com.ep_skb_list, size: CN_MAX_CON_BUF))
2596	goto fail;
2597
2598	state_set(epc: &child_ep->com, new: CONNECTING);
2599	child_ep->com.dev = dev;
2600	child_ep->com.cm_id = NULL;
2601
2602	if (iptype == 4) {
2603	struct sockaddr_in *sin = (struct sockaddr_in *)
2604	&child_ep->com.local_addr;
2605
2606	sin->sin_family = AF_INET;
2607	sin->sin_port = local_port;
2608	sin->sin_addr.s_addr = *(__be32 *)local_ip;
2609
2610	sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2611	sin->sin_family = AF_INET;
2612	sin->sin_port = ((struct sockaddr_in *)
2613	&parent_ep->com.local_addr)->sin_port;
2614	sin->sin_addr.s_addr = *(__be32 *)local_ip;
2615
2616	sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2617	sin->sin_family = AF_INET;
2618	sin->sin_port = peer_port;
2619	sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2620	} else {
2621	sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2622	sin6->sin6_family = PF_INET6;
2623	sin6->sin6_port = local_port;
2624	memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2625
2626	sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2627	sin6->sin6_family = PF_INET6;
2628	sin6->sin6_port = ((struct sockaddr_in6 *)
2629	&parent_ep->com.local_addr)->sin6_port;
2630	memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2631
2632	sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2633	sin6->sin6_family = PF_INET6;
2634	sin6->sin6_port = peer_port;
2635	memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2636	}
2637
2638	c4iw_get_ep(&parent_ep->com);
2639	child_ep->parent_ep = parent_ep;
2640	child_ep->tos = tos;
2641	child_ep->dst = dst;
2642	child_ep->hwtid = hwtid;
2643
2644	pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2645	child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2646
2647	timer_setup(&child_ep->timer, ep_timeout, 0);
2648	cxgb4_insert_tid(t, child_ep, hwtid,
2649	child_ep->com.local_addr.ss_family);
2650	insert_ep_tid(ep: child_ep);
2651	if (accept_cr(ep: child_ep, skb, req)) {
2652	c4iw_put_ep(&parent_ep->com);
2653	release_ep_resources(ep: child_ep);
2654	} else {
2655	set_bit(nr: PASS_ACCEPT_REQ, addr: &child_ep->com.history);
2656	}
2657	if (iptype == 6) {
2658	sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2659	cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2660	(const u32 *)&sin6->sin6_addr.s6_addr, 1);
2661	}
2662	goto out;
2663	fail:
2664	c4iw_put_ep(&child_ep->com);
2665	reject:
2666	reject_cr(dev, hwtid, skb);
2667	out:
2668	if (parent_ep)
2669	c4iw_put_ep(&parent_ep->com);
2670	return 0;
2671	}
2672
2673	static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2674	{
2675	struct c4iw_ep *ep;
2676	struct cpl_pass_establish *req = cplhdr(skb);
2677	unsigned int tid = GET_TID(req);
2678	int ret;
2679	u16 tcp_opt = ntohs(req->tcp_opt);
2680
2681	ep = get_ep_from_tid(dev, tid);
2682	if (!ep)
2683	return 0;
2684
2685	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2686	ep->snd_seq = be32_to_cpu(req->snd_isn);
2687	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2688	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2689
2690	pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2691
2692	set_emss(ep, opt: tcp_opt);
2693
2694	dst_confirm(dst: ep->dst);
2695	mutex_lock(&ep->com.mutex);
2696	ep->com.state = MPA_REQ_WAIT;
2697	start_ep_timer(ep);
2698	set_bit(nr: PASS_ESTAB, addr: &ep->com.history);
2699	ret = send_flowc(ep);
2700	mutex_unlock(lock: &ep->com.mutex);
2701	if (ret)
2702	c4iw_ep_disconnect(ep, abrupt: 1, GFP_KERNEL);
2703	c4iw_put_ep(&ep->com);
2704
2705	return 0;
2706	}
2707
2708	static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2709	{
2710	struct cpl_peer_close *hdr = cplhdr(skb);
2711	struct c4iw_ep *ep;
2712	struct c4iw_qp_attributes attrs;
2713	int disconnect = 1;
2714	int release = 0;
2715	unsigned int tid = GET_TID(hdr);
2716	int ret;
2717
2718	ep = get_ep_from_tid(dev, tid);
2719	if (!ep)
2720	return 0;
2721
2722	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2723	dst_confirm(dst: ep->dst);
2724
2725	set_bit(nr: PEER_CLOSE, addr: &ep->com.history);
2726	mutex_lock(&ep->com.mutex);
2727	switch (ep->com.state) {
2728	case MPA_REQ_WAIT:
2729	__state_set(epc: &ep->com, new: CLOSING);
2730	break;
2731	case MPA_REQ_SENT:
2732	__state_set(epc: &ep->com, new: CLOSING);
2733	connect_reply_upcall(ep, status: -ECONNRESET);
2734	break;
2735	case MPA_REQ_RCVD:
2736
2737	/*
2738	* We're gonna mark this puppy DEAD, but keep
2739	* the reference on it until the ULP accepts or
2740	* rejects the CR. Also wake up anyone waiting
2741	* in rdma connection migration (see c4iw_accept_cr()).
2742	*/
2743	__state_set(epc: &ep->com, new: CLOSING);
2744	pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2745	c4iw_wake_up_noref(wr_waitp: ep->com.wr_waitp, ret: -ECONNRESET);
2746	break;
2747	case MPA_REP_SENT:
2748	__state_set(epc: &ep->com, new: CLOSING);
2749	pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2750	c4iw_wake_up_noref(wr_waitp: ep->com.wr_waitp, ret: -ECONNRESET);
2751	break;
2752	case FPDU_MODE:
2753	start_ep_timer(ep);
2754	__state_set(epc: &ep->com, new: CLOSING);
2755	attrs.next_state = C4IW_QP_STATE_CLOSING;
2756	ret = c4iw_modify_qp(rhp: ep->com.qp->rhp, qhp: ep->com.qp,
2757	mask: C4IW_QP_ATTR_NEXT_STATE, attrs: &attrs, internal: 1);
2758	if (ret != -ECONNRESET) {
2759	peer_close_upcall(ep);
2760	disconnect = 1;
2761	}
2762	break;
2763	case ABORTING:
2764	disconnect = 0;
2765	break;
2766	case CLOSING:
2767	__state_set(epc: &ep->com, new: MORIBUND);
2768	disconnect = 0;
2769	break;
2770	case MORIBUND:
2771	(void)stop_ep_timer(ep);
2772	if (ep->com.cm_id && ep->com.qp) {
2773	attrs.next_state = C4IW_QP_STATE_IDLE;
2774	c4iw_modify_qp(rhp: ep->com.qp->rhp, qhp: ep->com.qp,
2775	mask: C4IW_QP_ATTR_NEXT_STATE, attrs: &attrs, internal: 1);
2776	}
2777	close_complete_upcall(ep, status: 0);
2778	__state_set(epc: &ep->com, new: DEAD);
2779	release = 1;
2780	disconnect = 0;
2781	break;
2782	case DEAD:
2783	disconnect = 0;
2784	break;
2785	default:
2786	WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2787	}
2788	mutex_unlock(lock: &ep->com.mutex);
2789	if (disconnect)
2790	c4iw_ep_disconnect(ep, abrupt: 0, GFP_KERNEL);
2791	if (release)
2792	release_ep_resources(ep);
2793	c4iw_put_ep(&ep->com);
2794	return 0;
2795	}
2796
2797	static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2798	{
2799	complete_cached_srq_buffers(ep, srqidx: ep->srqe_idx);
2800	if (ep->com.cm_id && ep->com.qp) {
2801	struct c4iw_qp_attributes attrs;
2802
2803	attrs.next_state = C4IW_QP_STATE_ERROR;
2804	c4iw_modify_qp(rhp: ep->com.qp->rhp, qhp: ep->com.qp,
2805	mask: C4IW_QP_ATTR_NEXT_STATE, attrs: &attrs, internal: 1);
2806	}
2807	peer_abort_upcall(ep);
2808	release_ep_resources(ep);
2809	c4iw_put_ep(&ep->com);
2810	}
2811
2812	static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2813	{
2814	struct cpl_abort_req_rss6 *req = cplhdr(skb);
2815	struct c4iw_ep *ep;
2816	struct sk_buff *rpl_skb;
2817	struct c4iw_qp_attributes attrs;
2818	int ret;
2819	int release = 0;
2820	unsigned int tid = GET_TID(req);
2821	u8 status;
2822	u32 srqidx;
2823
2824	u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2825
2826	ep = get_ep_from_tid(dev, tid);
2827	if (!ep)
2828	return 0;
2829
2830	status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2831
2832	if (cxgb_is_neg_adv(status)) {
2833	pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2834	ep->hwtid, status, neg_adv_str(status));
2835	ep->stats.abort_neg_adv++;
2836	mutex_lock(&dev->rdev.stats.lock);
2837	dev->rdev.stats.neg_adv++;
2838	mutex_unlock(lock: &dev->rdev.stats.lock);
2839	goto deref_ep;
2840	}
2841
2842	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2843	ep->com.state);
2844	set_bit(nr: PEER_ABORT, addr: &ep->com.history);
2845
2846	/*
2847	* Wake up any threads in rdma_init() or rdma_fini().
2848	* However, this is not needed if com state is just
2849	* MPA_REQ_SENT
2850	*/
2851	if (ep->com.state != MPA_REQ_SENT)
2852	c4iw_wake_up_noref(wr_waitp: ep->com.wr_waitp, ret: -ECONNRESET);
2853
2854	mutex_lock(&ep->com.mutex);
2855	switch (ep->com.state) {
2856	case CONNECTING:
2857	c4iw_put_ep(&ep->parent_ep->com);
2858	break;
2859	case MPA_REQ_WAIT:
2860	(void)stop_ep_timer(ep);
2861	break;
2862	case MPA_REQ_SENT:
2863	(void)stop_ep_timer(ep);
2864	if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2865	(mpa_rev == 2 && ep->tried_with_mpa_v1))
2866	connect_reply_upcall(ep, status: -ECONNRESET);
2867	else {
2868	/*
2869	* we just don't send notification upwards because we
2870	* want to retry with mpa_v1 without upper layers even
2871	* knowing it.
2872	*
2873	* do some housekeeping so as to re-initiate the
2874	* connection
2875	*/
2876	pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2877	__func__, mpa_rev);
2878	ep->retry_with_mpa_v1 = 1;
2879	}
2880	break;
2881	case MPA_REP_SENT:
2882	break;
2883	case MPA_REQ_RCVD:
2884	break;
2885	case MORIBUND:
2886	case CLOSING:
2887	stop_ep_timer(ep);
2888	fallthrough;
2889	case FPDU_MODE:
2890	if (ep->com.qp && ep->com.qp->srq) {
2891	srqidx = ABORT_RSS_SRQIDX_G(
2892	be32_to_cpu(req->srqidx_status));
2893	if (srqidx) {
2894	complete_cached_srq_buffers(ep, srqidx);
2895	} else {
2896	/* Hold ep ref until finish_peer_abort() */
2897	c4iw_get_ep(&ep->com);
2898	__state_set(epc: &ep->com, new: ABORTING);
2899	set_bit(nr: PEER_ABORT_IN_PROGRESS, addr: &ep->com.flags);
2900	read_tcb(ep);
2901	break;
2902
2903	}
2904	}
2905
2906	if (ep->com.cm_id && ep->com.qp) {
2907	attrs.next_state = C4IW_QP_STATE_ERROR;
2908	ret = c4iw_modify_qp(rhp: ep->com.qp->rhp,
2909	qhp: ep->com.qp, mask: C4IW_QP_ATTR_NEXT_STATE,
2910	attrs: &attrs, internal: 1);
2911	if (ret)
2912	pr_err("%s - qp <- error failed!\n", __func__);
2913	}
2914	peer_abort_upcall(ep);
2915	break;
2916	case ABORTING:
2917	break;
2918	case DEAD:
2919	pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2920	mutex_unlock(lock: &ep->com.mutex);
2921	goto deref_ep;
2922	default:
2923	WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2924	break;
2925	}
2926	dst_confirm(dst: ep->dst);
2927	if (ep->com.state != ABORTING) {
2928	__state_set(epc: &ep->com, new: DEAD);
2929	/* we don't release if we want to retry with mpa_v1 */
2930	if (!ep->retry_with_mpa_v1)
2931	release = 1;
2932	}
2933	mutex_unlock(lock: &ep->com.mutex);
2934
2935	rpl_skb = skb_dequeue(list: &ep->com.ep_skb_list);
2936	if (WARN_ON(!rpl_skb)) {
2937	release = 1;
2938	goto out;
2939	}
2940
2941	cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2942
2943	c4iw_ofld_send(rdev: &ep->com.dev->rdev, skb: rpl_skb);
2944	out:
2945	if (release)
2946	release_ep_resources(ep);
2947	else if (ep->retry_with_mpa_v1) {
2948	if (ep->com.remote_addr.ss_family == AF_INET6) {
2949	struct sockaddr_in6 *sin6 =
2950	(struct sockaddr_in6 *)
2951	&ep->com.local_addr;
2952	cxgb4_clip_release(
2953	ep->com.dev->rdev.lldi.ports[0],
2954	(const u32 *)&sin6->sin6_addr.s6_addr,
2955	1);
2956	}
2957	xa_erase_irq(xa: &ep->com.dev->hwtids, index: ep->hwtid);
2958	cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2959	ep->com.local_addr.ss_family);
2960	dst_release(dst: ep->dst);
2961	cxgb4_l2t_release(ep->l2t);
2962	c4iw_reconnect(ep);
2963	}
2964
2965	deref_ep:
2966	c4iw_put_ep(&ep->com);
2967	/* Dereferencing ep, referenced in peer_abort_intr() */
2968	c4iw_put_ep(&ep->com);
2969	return 0;
2970	}
2971
2972	static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2973	{
2974	struct c4iw_ep *ep;
2975	struct c4iw_qp_attributes attrs;
2976	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2977	int release = 0;
2978	unsigned int tid = GET_TID(rpl);
2979
2980	ep = get_ep_from_tid(dev, tid);
2981	if (!ep)
2982	return 0;
2983
2984	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2985
2986	/* The cm_id may be null if we failed to connect */
2987	mutex_lock(&ep->com.mutex);
2988	set_bit(nr: CLOSE_CON_RPL, addr: &ep->com.history);
2989	switch (ep->com.state) {
2990	case CLOSING:
2991	__state_set(epc: &ep->com, new: MORIBUND);
2992	break;
2993	case MORIBUND:
2994	(void)stop_ep_timer(ep);
2995	if ((ep->com.cm_id) && (ep->com.qp)) {
2996	attrs.next_state = C4IW_QP_STATE_IDLE;
2997	c4iw_modify_qp(rhp: ep->com.qp->rhp,
2998	qhp: ep->com.qp,
2999	mask: C4IW_QP_ATTR_NEXT_STATE,
3000	attrs: &attrs, internal: 1);
3001	}
3002	close_complete_upcall(ep, status: 0);
3003	__state_set(epc: &ep->com, new: DEAD);
3004	release = 1;
3005	break;
3006	case ABORTING:
3007	case DEAD:
3008	break;
3009	default:
3010	WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3011	break;
3012	}
3013	mutex_unlock(lock: &ep->com.mutex);
3014	if (release)
3015	release_ep_resources(ep);
3016	c4iw_put_ep(&ep->com);
3017	return 0;
3018	}
3019
3020	static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3021	{
3022	struct cpl_rdma_terminate *rpl = cplhdr(skb);
3023	unsigned int tid = GET_TID(rpl);
3024	struct c4iw_ep *ep;
3025	struct c4iw_qp_attributes attrs;
3026
3027	ep = get_ep_from_tid(dev, tid);
3028
3029	if (ep) {
3030	if (ep->com.qp) {
3031	pr_warn("TERM received tid %u qpid %u\n", tid,
3032	ep->com.qp->wq.sq.qid);
3033	attrs.next_state = C4IW_QP_STATE_TERMINATE;
3034	c4iw_modify_qp(rhp: ep->com.qp->rhp, qhp: ep->com.qp,
3035	mask: C4IW_QP_ATTR_NEXT_STATE, attrs: &attrs, internal: 1);
3036	}
3037
3038	/* As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3,
3039	* when entering the TERM state the RNIC MUST initiate a CLOSE.
3040	*/
3041	c4iw_ep_disconnect(ep, abrupt: 1, GFP_KERNEL);
3042	c4iw_put_ep(&ep->com);
3043	} else
3044	pr_warn("TERM received tid %u no ep/qp\n", tid);
3045
3046	return 0;
3047	}
3048
3049	/*
3050	* Upcall from the adapter indicating data has been transmitted.
3051	* For us its just the single MPA request or reply. We can now free
3052	* the skb holding the mpa message.
3053	*/
3054	static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3055	{
3056	struct c4iw_ep *ep;
3057	struct cpl_fw4_ack *hdr = cplhdr(skb);
3058	u8 credits = hdr->credits;
3059	unsigned int tid = GET_TID(hdr);
3060
3061
3062	ep = get_ep_from_tid(dev, tid);
3063	if (!ep)
3064	return 0;
3065	pr_debug("ep %p tid %u credits %u\n",
3066	ep, ep->hwtid, credits);
3067	if (credits == 0) {
3068	pr_debug("0 credit ack ep %p tid %u state %u\n",
3069	ep, ep->hwtid, state_read(&ep->com));
3070	goto out;
3071	}
3072
3073	dst_confirm(dst: ep->dst);
3074	if (ep->mpa_skb) {
3075	pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3076	ep, ep->hwtid, state_read(&ep->com),
3077	ep->mpa_attr.initiator ? 1 : 0);
3078	mutex_lock(&ep->com.mutex);
3079	kfree_skb(skb: ep->mpa_skb);
3080	ep->mpa_skb = NULL;
3081	if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3082	stop_ep_timer(ep);
3083	mutex_unlock(lock: &ep->com.mutex);
3084	}
3085	out:
3086	c4iw_put_ep(&ep->com);
3087	return 0;
3088	}
3089
3090	int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3091	{
3092	int abort;
3093	struct c4iw_ep *ep = to_ep(cm_id);
3094
3095	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3096
3097	mutex_lock(&ep->com.mutex);
3098	if (ep->com.state != MPA_REQ_RCVD) {
3099	mutex_unlock(lock: &ep->com.mutex);
3100	c4iw_put_ep(&ep->com);
3101	return -ECONNRESET;
3102	}
3103	set_bit(nr: ULP_REJECT, addr: &ep->com.history);
3104	if (mpa_rev == 0)
3105	abort = 1;
3106	else
3107	abort = send_mpa_reject(ep, pdata, plen: pdata_len);
3108	mutex_unlock(lock: &ep->com.mutex);
3109
3110	stop_ep_timer(ep);
3111	c4iw_ep_disconnect(ep, abrupt: abort != 0, GFP_KERNEL);
3112	c4iw_put_ep(&ep->com);
3113	return 0;
3114	}
3115
3116	int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3117	{
3118	int err;
3119	struct c4iw_qp_attributes attrs;
3120	enum c4iw_qp_attr_mask mask;
3121	struct c4iw_ep *ep = to_ep(cm_id);
3122	struct c4iw_dev *h = to_c4iw_dev(ibdev: cm_id->device);
3123	struct c4iw_qp *qp = get_qhp(rhp: h, qpid: conn_param->qpn);
3124	int abort = 0;
3125
3126	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3127
3128	mutex_lock(&ep->com.mutex);
3129	if (ep->com.state != MPA_REQ_RCVD) {
3130	err = -ECONNRESET;
3131	goto err_out;
3132	}
3133
3134	if (!qp) {
3135	err = -EINVAL;
3136	goto err_out;
3137	}
3138
3139	set_bit(nr: ULP_ACCEPT, addr: &ep->com.history);
3140	if ((conn_param->ord > cur_max_read_depth(dev: ep->com.dev)) ||
3141	(conn_param->ird > cur_max_read_depth(dev: ep->com.dev))) {
3142	err = -EINVAL;
3143	goto err_abort;
3144	}
3145
3146	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3147	if (conn_param->ord > ep->ird) {
3148	if (RELAXED_IRD_NEGOTIATION) {
3149	conn_param->ord = ep->ird;
3150	} else {
3151	ep->ird = conn_param->ird;
3152	ep->ord = conn_param->ord;
3153	send_mpa_reject(ep, pdata: conn_param->private_data,
3154	plen: conn_param->private_data_len);
3155	err = -ENOMEM;
3156	goto err_abort;
3157	}
3158	}
3159	if (conn_param->ird < ep->ord) {
3160	if (RELAXED_IRD_NEGOTIATION &&
3161	ep->ord <= h->rdev.lldi.max_ordird_qp) {
3162	conn_param->ird = ep->ord;
3163	} else {
3164	err = -ENOMEM;
3165	goto err_abort;
3166	}
3167	}
3168	}
3169	ep->ird = conn_param->ird;
3170	ep->ord = conn_param->ord;
3171
3172	if (ep->mpa_attr.version == 1) {
3173	if (peer2peer && ep->ird == 0)
3174	ep->ird = 1;
3175	} else {
3176	if (peer2peer &&
3177	(ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3178	(p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3179	ep->ird = 1;
3180	}
3181
3182	pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3183
3184	ep->com.cm_id = cm_id;
3185	ref_cm_id(epc: &ep->com);
3186	ep->com.qp = qp;
3187	ref_qp(ep);
3188
3189	/* bind QP to EP and move to RTS */
3190	attrs.mpa_attr = ep->mpa_attr;
3191	attrs.max_ird = ep->ird;
3192	attrs.max_ord = ep->ord;
3193	attrs.llp_stream_handle = ep;
3194	attrs.next_state = C4IW_QP_STATE_RTS;
3195
3196	/* bind QP and TID with INIT_WR */
3197	mask = C4IW_QP_ATTR_NEXT_STATE |
3198	C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3199	C4IW_QP_ATTR_MPA_ATTR |
3200	C4IW_QP_ATTR_MAX_IRD |
3201	C4IW_QP_ATTR_MAX_ORD;
3202
3203	err = c4iw_modify_qp(rhp: ep->com.qp->rhp,
3204	qhp: ep->com.qp, mask, attrs: &attrs, internal: 1);
3205	if (err)
3206	goto err_deref_cm_id;
3207
3208	set_bit(nr: STOP_MPA_TIMER, addr: &ep->com.flags);
3209	err = send_mpa_reply(ep, pdata: conn_param->private_data,
3210	plen: conn_param->private_data_len);
3211	if (err)
3212	goto err_deref_cm_id;
3213
3214	__state_set(epc: &ep->com, new: FPDU_MODE);
3215	established_upcall(ep);
3216	mutex_unlock(lock: &ep->com.mutex);
3217	c4iw_put_ep(&ep->com);
3218	return 0;
3219	err_deref_cm_id:
3220	deref_cm_id(epc: &ep->com);
3221	err_abort:
3222	abort = 1;
3223	err_out:
3224	mutex_unlock(lock: &ep->com.mutex);
3225	if (abort)
3226	c4iw_ep_disconnect(ep, abrupt: 1, GFP_KERNEL);
3227	c4iw_put_ep(&ep->com);
3228	return err;
3229	}
3230
3231	static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3232	{
3233	struct in_device *ind;
3234	int found = 0;
3235	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3236	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3237	const struct in_ifaddr *ifa;
3238
3239	ind = in_dev_get(dev: dev->rdev.lldi.ports[0]);
3240	if (!ind)
3241	return -EADDRNOTAVAIL;
3242	rcu_read_lock();
3243	in_dev_for_each_ifa_rcu(ifa, ind) {
3244	if (ifa->ifa_flags & IFA_F_SECONDARY)
3245	continue;
3246	laddr->sin_addr.s_addr = ifa->ifa_address;
3247	raddr->sin_addr.s_addr = ifa->ifa_address;
3248	found = 1;
3249	break;
3250	}
3251	rcu_read_unlock();
3252
3253	in_dev_put(idev: ind);
3254	return found ? 0 : -EADDRNOTAVAIL;
3255	}
3256
3257	static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3258	unsigned char banned_flags)
3259	{
3260	struct inet6_dev *idev;
3261	int err = -EADDRNOTAVAIL;
3262
3263	rcu_read_lock();
3264	idev = __in6_dev_get(dev);
3265	if (idev != NULL) {
3266	struct inet6_ifaddr *ifp;
3267
3268	read_lock_bh(&idev->lock);
3269	list_for_each_entry(ifp, &idev->addr_list, if_list) {
3270	if (ifp->scope == IFA_LINK &&
3271	!(ifp->flags & banned_flags)) {
3272	memcpy(addr, &ifp->addr, 16);
3273	err = 0;
3274	break;
3275	}
3276	}
3277	read_unlock_bh(&idev->lock);
3278	}
3279	rcu_read_unlock();
3280	return err;
3281	}
3282
3283	static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3284	{
3285	struct in6_addr addr;
3286	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3287	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3288
3289	if (!get_lladdr(dev: dev->rdev.lldi.ports[0], addr: &addr, IFA_F_TENTATIVE)) {
3290	memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3291	memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3292	return 0;
3293	}
3294	return -EADDRNOTAVAIL;
3295	}
3296
3297	int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3298	{
3299	struct c4iw_dev *dev = to_c4iw_dev(ibdev: cm_id->device);
3300	struct c4iw_ep *ep;
3301	int err = 0;
3302	struct sockaddr_in *laddr;
3303	struct sockaddr_in *raddr;
3304	struct sockaddr_in6 *laddr6;
3305	struct sockaddr_in6 *raddr6;
3306	__u8 *ra;
3307	int iptype;
3308
3309	if ((conn_param->ord > cur_max_read_depth(dev)) ||
3310	(conn_param->ird > cur_max_read_depth(dev))) {
3311	err = -EINVAL;
3312	goto out;
3313	}
3314	ep = alloc_ep(size: sizeof(*ep), GFP_KERNEL);
3315	if (!ep) {
3316	pr_err("%s - cannot alloc ep\n", __func__);
3317	err = -ENOMEM;
3318	goto out;
3319	}
3320
3321	skb_queue_head_init(list: &ep->com.ep_skb_list);
3322	if (alloc_ep_skb_list(ep_skb_list: &ep->com.ep_skb_list, size: CN_MAX_CON_BUF)) {
3323	err = -ENOMEM;
3324	goto fail1;
3325	}
3326
3327	timer_setup(&ep->timer, ep_timeout, 0);
3328	ep->plen = conn_param->private_data_len;
3329	if (ep->plen)
3330	memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3331	conn_param->private_data, ep->plen);
3332	ep->ird = conn_param->ird;
3333	ep->ord = conn_param->ord;
3334
3335	if (peer2peer && ep->ord == 0)
3336	ep->ord = 1;
3337
3338	ep->com.cm_id = cm_id;
3339	ref_cm_id(epc: &ep->com);
3340	cm_id->provider_data = ep;
3341	ep->com.dev = dev;
3342	ep->com.qp = get_qhp(rhp: dev, qpid: conn_param->qpn);
3343	if (!ep->com.qp) {
3344	pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3345	err = -EINVAL;
3346	goto fail2;
3347	}
3348	ref_qp(ep);
3349	pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3350	ep->com.qp, cm_id);
3351
3352	/*
3353	* Allocate an active TID to initiate a TCP connection.
3354	*/
3355	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3356	if (ep->atid == -1) {
3357	pr_err("%s - cannot alloc atid\n", __func__);
3358	err = -ENOMEM;
3359	goto fail2;
3360	}
3361	err = xa_insert_irq(xa: &dev->atids, index: ep->atid, entry: ep, GFP_KERNEL);
3362	if (err)
3363	goto fail5;
3364
3365	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3366	sizeof(ep->com.local_addr));
3367	memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3368	sizeof(ep->com.remote_addr));
3369
3370	laddr = (struct sockaddr_in *)&ep->com.local_addr;
3371	raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3372	laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3373	raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3374
3375	if (cm_id->m_remote_addr.ss_family == AF_INET) {
3376	iptype = 4;
3377	ra = (__u8 *)&raddr->sin_addr;
3378
3379	/*
3380	* Handle loopback requests to INADDR_ANY.
3381	*/
3382	if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3383	err = pick_local_ipaddrs(dev, cm_id);
3384	if (err)
3385	goto fail3;
3386	}
3387
3388	/* find a route */
3389	pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3390	&laddr->sin_addr, ntohs(laddr->sin_port),
3391	ra, ntohs(raddr->sin_port));
3392	ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3393	laddr->sin_addr.s_addr,
3394	raddr->sin_addr.s_addr,
3395	laddr->sin_port,
3396	raddr->sin_port, cm_id->tos);
3397	} else {
3398	iptype = 6;
3399	ra = (__u8 *)&raddr6->sin6_addr;
3400
3401	/*
3402	* Handle loopback requests to INADDR_ANY.
3403	*/
3404	if (ipv6_addr_type(addr: &raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3405	err = pick_local_ip6addrs(dev, cm_id);
3406	if (err)
3407	goto fail3;
3408	}
3409
3410	/* find a route */
3411	pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3412	laddr6->sin6_addr.s6_addr,
3413	ntohs(laddr6->sin6_port),
3414	raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3415	ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3416	laddr6->sin6_addr.s6_addr,
3417	raddr6->sin6_addr.s6_addr,
3418	laddr6->sin6_port,
3419	raddr6->sin6_port, cm_id->tos,
3420	raddr6->sin6_scope_id);
3421	}
3422	if (!ep->dst) {
3423	pr_err("%s - cannot find route\n", __func__);
3424	err = -EHOSTUNREACH;
3425	goto fail3;
3426	}
3427
3428	err = import_ep(ep, iptype, peer_ip: ra, dst: ep->dst, cdev: ep->com.dev, clear_mpa_v1: true,
3429	adapter_type: ep->com.dev->rdev.lldi.adapter_type, tos: cm_id->tos);
3430	if (err) {
3431	pr_err("%s - cannot alloc l2e\n", __func__);
3432	goto fail4;
3433	}
3434
3435	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3436	ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3437	ep->l2t->idx);
3438
3439	state_set(epc: &ep->com, new: CONNECTING);
3440	ep->tos = cm_id->tos;
3441
3442	/* send connect request to rnic */
3443	err = send_connect(ep);
3444	if (!err)
3445	goto out;
3446
3447	cxgb4_l2t_release(ep->l2t);
3448	fail4:
3449	dst_release(dst: ep->dst);
3450	fail3:
3451	xa_erase_irq(xa: &ep->com.dev->atids, index: ep->atid);
3452	fail5:
3453	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3454	fail2:
3455	skb_queue_purge(list: &ep->com.ep_skb_list);
3456	deref_cm_id(epc: &ep->com);
3457	fail1:
3458	c4iw_put_ep(&ep->com);
3459	out:
3460	return err;
3461	}
3462
3463	static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3464	{
3465	int err;
3466	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3467	&ep->com.local_addr;
3468
3469	if (ipv6_addr_type(addr: &sin6->sin6_addr) != IPV6_ADDR_ANY) {
3470	err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3471	(const u32 *)&sin6->sin6_addr.s6_addr, 1);
3472	if (err)
3473	return err;
3474	}
3475	c4iw_init_wr_wait(wr_waitp: ep->com.wr_waitp);
3476	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3477	ep->stid, &sin6->sin6_addr,
3478	sin6->sin6_port,
3479	ep->com.dev->rdev.lldi.rxq_ids[0]);
3480	if (!err)
3481	err = c4iw_wait_for_reply(rdev: &ep->com.dev->rdev,
3482	wr_waitp: ep->com.wr_waitp,
3483	hwtid: 0, qpid: 0, func: __func__);
3484	else if (err > 0)
3485	err = net_xmit_errno(err);
3486	if (err) {
3487	cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3488	(const u32 *)&sin6->sin6_addr.s6_addr, 1);
3489	pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3490	err, ep->stid,
3491	sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3492	}
3493	return err;
3494	}
3495
3496	static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3497	{
3498	int err;
3499	struct sockaddr_in *sin = (struct sockaddr_in *)
3500	&ep->com.local_addr;
3501
3502	if (dev->rdev.lldi.enable_fw_ofld_conn) {
3503	do {
3504	err = cxgb4_create_server_filter(
3505	ep->com.dev->rdev.lldi.ports[0], ep->stid,
3506	sin->sin_addr.s_addr, sin->sin_port, 0,
3507	ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3508	if (err == -EBUSY) {
3509	if (c4iw_fatal_error(rdev: &ep->com.dev->rdev)) {
3510	err = -EIO;
3511	break;
3512	}
3513	set_current_state(TASK_UNINTERRUPTIBLE);
3514	schedule_timeout(timeout: usecs_to_jiffies(u: 100));
3515	}
3516	} while (err == -EBUSY);
3517	} else {
3518	c4iw_init_wr_wait(wr_waitp: ep->com.wr_waitp);
3519	err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3520	ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3521	0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3522	if (!err)
3523	err = c4iw_wait_for_reply(rdev: &ep->com.dev->rdev,
3524	wr_waitp: ep->com.wr_waitp,
3525	hwtid: 0, qpid: 0, func: __func__);
3526	else if (err > 0)
3527	err = net_xmit_errno(err);
3528	}
3529	if (err)
3530	pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3531	, err, ep->stid,
3532	&sin->sin_addr, ntohs(sin->sin_port));
3533	return err;
3534	}
3535
3536	int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3537	{
3538	int err = 0;
3539	struct c4iw_dev *dev = to_c4iw_dev(ibdev: cm_id->device);
3540	struct c4iw_listen_ep *ep;
3541
3542	might_sleep();
3543
3544	ep = alloc_ep(size: sizeof(*ep), GFP_KERNEL);
3545	if (!ep) {
3546	pr_err("%s - cannot alloc ep\n", __func__);
3547	err = -ENOMEM;
3548	goto fail1;
3549	}
3550	skb_queue_head_init(list: &ep->com.ep_skb_list);
3551	pr_debug("ep %p\n", ep);
3552	ep->com.cm_id = cm_id;
3553	ref_cm_id(epc: &ep->com);
3554	ep->com.dev = dev;
3555	ep->backlog = backlog;
3556	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3557	sizeof(ep->com.local_addr));
3558
3559	/*
3560	* Allocate a server TID.
3561	*/
3562	if (dev->rdev.lldi.enable_fw_ofld_conn &&
3563	ep->com.local_addr.ss_family == AF_INET)
3564	ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3565	cm_id->m_local_addr.ss_family, ep);
3566	else
3567	ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3568	cm_id->m_local_addr.ss_family, ep);
3569
3570	if (ep->stid == -1) {
3571	pr_err("%s - cannot alloc stid\n", __func__);
3572	err = -ENOMEM;
3573	goto fail2;
3574	}
3575	err = xa_insert_irq(xa: &dev->stids, index: ep->stid, entry: ep, GFP_KERNEL);
3576	if (err)
3577	goto fail3;
3578
3579	state_set(epc: &ep->com, new: LISTEN);
3580	if (ep->com.local_addr.ss_family == AF_INET)
3581	err = create_server4(dev, ep);
3582	else
3583	err = create_server6(dev, ep);
3584	if (!err) {
3585	cm_id->provider_data = ep;
3586	goto out;
3587	}
3588	xa_erase_irq(xa: &ep->com.dev->stids, index: ep->stid);
3589	fail3:
3590	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3591	ep->com.local_addr.ss_family);
3592	fail2:
3593	deref_cm_id(epc: &ep->com);
3594	c4iw_put_ep(&ep->com);
3595	fail1:
3596	out:
3597	return err;
3598	}
3599
3600	int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3601	{
3602	int err;
3603	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3604
3605	pr_debug("ep %p\n", ep);
3606
3607	might_sleep();
3608	state_set(epc: &ep->com, new: DEAD);
3609	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3610	ep->com.local_addr.ss_family == AF_INET) {
3611	err = cxgb4_remove_server_filter(
3612	ep->com.dev->rdev.lldi.ports[0], ep->stid,
3613	ep->com.dev->rdev.lldi.rxq_ids[0], false);
3614	} else {
3615	struct sockaddr_in6 *sin6;
3616	c4iw_init_wr_wait(wr_waitp: ep->com.wr_waitp);
3617	err = cxgb4_remove_server(
3618	ep->com.dev->rdev.lldi.ports[0], ep->stid,
3619	ep->com.dev->rdev.lldi.rxq_ids[0],
3620	ep->com.local_addr.ss_family == AF_INET6);
3621	if (err)
3622	goto done;
3623	err = c4iw_wait_for_reply(rdev: &ep->com.dev->rdev, wr_waitp: ep->com.wr_waitp,
3624	hwtid: 0, qpid: 0, func: __func__);
3625	sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3626	cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3627	(const u32 *)&sin6->sin6_addr.s6_addr, 1);
3628	}
3629	xa_erase_irq(xa: &ep->com.dev->stids, index: ep->stid);
3630	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3631	ep->com.local_addr.ss_family);
3632	done:
3633	deref_cm_id(epc: &ep->com);
3634	c4iw_put_ep(&ep->com);
3635	return err;
3636	}
3637
3638	int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3639	{
3640	int ret = 0;
3641	int close = 0;
3642	int fatal = 0;
3643	struct c4iw_rdev *rdev;
3644
3645	mutex_lock(&ep->com.mutex);
3646
3647	pr_debug("ep %p state %s, abrupt %d\n", ep,
3648	states[ep->com.state], abrupt);
3649
3650	/*
3651	* Ref the ep here in case we have fatal errors causing the
3652	* ep to be released and freed.
3653	*/
3654	c4iw_get_ep(&ep->com);
3655
3656	rdev = &ep->com.dev->rdev;
3657	if (c4iw_fatal_error(rdev)) {
3658	fatal = 1;
3659	close_complete_upcall(ep, status: -EIO);
3660	ep->com.state = DEAD;
3661	}
3662	switch (ep->com.state) {
3663	case MPA_REQ_WAIT:
3664	case MPA_REQ_SENT:
3665	case MPA_REQ_RCVD:
3666	case MPA_REP_SENT:
3667	case FPDU_MODE:
3668	case CONNECTING:
3669	close = 1;
3670	if (abrupt)
3671	ep->com.state = ABORTING;
3672	else {
3673	ep->com.state = CLOSING;
3674
3675	/*
3676	* if we close before we see the fw4_ack() then we fix
3677	* up the timer state since we're reusing it.
3678	*/
3679	if (ep->mpa_skb &&
3680	test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3681	clear_bit(nr: STOP_MPA_TIMER, addr: &ep->com.flags);
3682	stop_ep_timer(ep);
3683	}
3684	start_ep_timer(ep);
3685	}
3686	set_bit(nr: CLOSE_SENT, addr: &ep->com.flags);
3687	break;
3688	case CLOSING:
3689	if (!test_and_set_bit(nr: CLOSE_SENT, addr: &ep->com.flags)) {
3690	close = 1;
3691	if (abrupt) {
3692	(void)stop_ep_timer(ep);
3693	ep->com.state = ABORTING;
3694	} else
3695	ep->com.state = MORIBUND;
3696	}
3697	break;
3698	case MORIBUND:
3699	case ABORTING:
3700	case DEAD:
3701	pr_debug("ignoring disconnect ep %p state %u\n",
3702	ep, ep->com.state);
3703	break;
3704	default:
3705	WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3706	break;
3707	}
3708
3709	if (close) {
3710	if (abrupt) {
3711	set_bit(nr: EP_DISC_ABORT, addr: &ep->com.history);
3712	ret = send_abort(ep);
3713	} else {
3714	set_bit(nr: EP_DISC_CLOSE, addr: &ep->com.history);
3715	ret = send_halfclose(ep);
3716	}
3717	if (ret) {
3718	set_bit(nr: EP_DISC_FAIL, addr: &ep->com.history);
3719	if (!abrupt) {
3720	stop_ep_timer(ep);
3721	close_complete_upcall(ep, status: -EIO);
3722	}
3723	if (ep->com.qp) {
3724	struct c4iw_qp_attributes attrs;
3725
3726	attrs.next_state = C4IW_QP_STATE_ERROR;
3727	ret = c4iw_modify_qp(rhp: ep->com.qp->rhp,
3728	qhp: ep->com.qp,
3729	mask: C4IW_QP_ATTR_NEXT_STATE,
3730	attrs: &attrs, internal: 1);
3731	if (ret)
3732	pr_err("%s - qp <- error failed!\n",
3733	__func__);
3734	}
3735	fatal = 1;
3736	}
3737	}
3738	mutex_unlock(lock: &ep->com.mutex);
3739	c4iw_put_ep(&ep->com);
3740	if (fatal)
3741	release_ep_resources(ep);
3742	return ret;
3743	}
3744
3745	static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3746	struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3747	{
3748	struct c4iw_ep *ep;
3749	int atid = be32_to_cpu(req->tid);
3750
3751	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3752	(__force u32) req->tid);
3753	if (!ep)
3754	return;
3755
3756	switch (req->retval) {
3757	case FW_ENOMEM:
3758	set_bit(nr: ACT_RETRY_NOMEM, addr: &ep->com.history);
3759	if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3760	send_fw_act_open_req(ep, atid);
3761	return;
3762	}
3763	fallthrough;
3764	case FW_EADDRINUSE:
3765	set_bit(nr: ACT_RETRY_INUSE, addr: &ep->com.history);
3766	if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3767	send_fw_act_open_req(ep, atid);
3768	return;
3769	}
3770	break;
3771	default:
3772	pr_info("%s unexpected ofld conn wr retval %d\n",
3773	__func__, req->retval);
3774	break;
3775	}
3776	pr_err("active ofld_connect_wr failure %d atid %d\n",
3777	req->retval, atid);
3778	mutex_lock(&dev->rdev.stats.lock);
3779	dev->rdev.stats.act_ofld_conn_fails++;
3780	mutex_unlock(lock: &dev->rdev.stats.lock);
3781	connect_reply_upcall(ep, status: status2errno(status: req->retval));
3782	state_set(epc: &ep->com, new: DEAD);
3783	if (ep->com.remote_addr.ss_family == AF_INET6) {
3784	struct sockaddr_in6 *sin6 =
3785	(struct sockaddr_in6 *)&ep->com.local_addr;
3786	cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3787	(const u32 *)&sin6->sin6_addr.s6_addr, 1);
3788	}
3789	xa_erase_irq(xa: &dev->atids, index: atid);
3790	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3791	dst_release(dst: ep->dst);
3792	cxgb4_l2t_release(ep->l2t);
3793	c4iw_put_ep(&ep->com);
3794	}
3795
3796	static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3797	struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3798	{
3799	struct sk_buff *rpl_skb;
3800	struct cpl_pass_accept_req *cpl;
3801	int ret;
3802
3803	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3804	if (req->retval) {
3805	pr_err("%s passive open failure %d\n", __func__, req->retval);
3806	mutex_lock(&dev->rdev.stats.lock);
3807	dev->rdev.stats.pas_ofld_conn_fails++;
3808	mutex_unlock(lock: &dev->rdev.stats.lock);
3809	kfree_skb(skb: rpl_skb);
3810	} else {
3811	cpl = (struct cpl_pass_accept_req *)cplhdr(skb: rpl_skb);
3812	OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3813	(__force u32) htonl(
3814	(__force u32) req->tid)));
3815	ret = pass_accept_req(dev, skb: rpl_skb);
3816	if (!ret)
3817	kfree_skb(skb: rpl_skb);
3818	}
3819	return;
3820	}
3821
3822	static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3823	{
3824	u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3825	u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3826	u64 t;
3827	u32 shift = 32;
3828
3829	t = (thi << shift) | (tlo >> shift);
3830
3831	return t;
3832	}
3833
3834	static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3835	{
3836	u32 v;
3837	u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3838
3839	if (word & 0x1)
3840	shift += 32;
3841	v = (t >> shift) & mask;
3842	return v;
3843	}
3844
3845	static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3846	{
3847	struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3848	__be64 *tcb = (__be64 *)(rpl + 1);
3849	unsigned int tid = GET_TID(rpl);
3850	struct c4iw_ep *ep;
3851	u64 t_flags_64;
3852	u32 rx_pdu_out;
3853
3854	ep = get_ep_from_tid(dev, tid);
3855	if (!ep)
3856	return 0;
3857	/* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3858	* determine if there's a rx PDU feedback event pending.
3859	*
3860	* If that bit is set, it means we'll need to re-read the TCB's
3861	* rq_start value. The final value is the one present in a TCB
3862	* with the TF_RX_PDU_OUT bit cleared.
3863	*/
3864
3865	t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3866	rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3867
3868	c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3869	c4iw_put_ep(&ep->com); /* from read_tcb() */
3870
3871	/* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3872	if (rx_pdu_out) {
3873	if (++ep->rx_pdu_out_cnt >= 2) {
3874	WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3875	goto cleanup;
3876	}
3877	read_tcb(ep);
3878	return 0;
3879	}
3880
3881	ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_M,
3882	TCB_RQ_START_S);
3883	cleanup:
3884	pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3885
3886	if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3887	finish_peer_abort(dev, ep);
3888	else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3889	send_abort_req(ep);
3890	else
3891	WARN_ONCE(1, "unexpected state!");
3892
3893	return 0;
3894	}
3895
3896	static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3897	{
3898	struct cpl_fw6_msg *rpl = cplhdr(skb);
3899	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3900
3901	switch (rpl->type) {
3902	case FW6_TYPE_CQE:
3903	c4iw_ev_dispatch(dev, err_cqe: (struct t4_cqe *)&rpl->data[0]);
3904	break;
3905	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3906	req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3907	switch (req->t_state) {
3908	case TCP_SYN_SENT:
3909	active_ofld_conn_reply(dev, skb, req);
3910	break;
3911	case TCP_SYN_RECV:
3912	passive_ofld_conn_reply(dev, skb, req);
3913	break;
3914	default:
3915	pr_err("%s unexpected ofld conn wr state %d\n",
3916	__func__, req->t_state);
3917	break;
3918	}
3919	break;
3920	}
3921	return 0;
3922	}
3923
3924	static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3925	{
3926	__be32 l2info;
3927	__be16 hdr_len, vlantag, len;
3928	u16 eth_hdr_len;
3929	int tcp_hdr_len, ip_hdr_len;
3930	u8 intf;
3931	struct cpl_rx_pkt *cpl = cplhdr(skb);
3932	struct cpl_pass_accept_req *req;
3933	struct tcp_options_received tmp_opt;
3934	struct c4iw_dev *dev;
3935	enum chip_type type;
3936
3937	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3938	/* Store values from cpl_rx_pkt in temporary location. */
3939	vlantag = cpl->vlan;
3940	len = cpl->len;
3941	l2info = cpl->l2info;
3942	hdr_len = cpl->hdr_len;
3943	intf = cpl->iff;
3944
3945	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3946
3947	/*
3948	* We need to parse the TCP options from SYN packet.
3949	* to generate cpl_pass_accept_req.
3950	*/
3951	memset(&tmp_opt, 0, sizeof(tmp_opt));
3952	tcp_clear_options(rx_opt: &tmp_opt);
3953	tcp_parse_options(net: &init_net, skb, opt_rx: &tmp_opt, estab: 0, NULL);
3954
3955	req = __skb_push(skb, sizeof(*req));
3956	memset(req, 0, sizeof(*req));
3957	req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3958	SYN_MAC_IDX_V(RX_MACIDX_G(
3959	be32_to_cpu(l2info))) |
3960	SYN_XACT_MATCH_F);
3961	type = dev->rdev.lldi.adapter_type;
3962	tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3963	ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3964	req->hdr_len =
3965	cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3966	if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3967	eth_hdr_len = is_t4(type) ?
3968	RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3969	RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3970	req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3971	IP_HDR_LEN_V(ip_hdr_len) |
3972	ETH_HDR_LEN_V(eth_hdr_len));
3973	} else { /* T6 and later */
3974	eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3975	req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3976	T6_IP_HDR_LEN_V(ip_hdr_len) |
3977	T6_ETH_HDR_LEN_V(eth_hdr_len));
3978	}
3979	req->vlan = vlantag;
3980	req->len = len;
3981	req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3982	PASS_OPEN_TOS_V(tos));
3983	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3984	if (tmp_opt.wscale_ok)
3985	req->tcpopt.wsf = tmp_opt.snd_wscale;
3986	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3987	if (tmp_opt.sack_ok)
3988	req->tcpopt.sack = 1;
3989	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3990	return;
3991	}
3992
3993	static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3994	__be32 laddr, __be16 lport,
3995	__be32 raddr, __be16 rport,
3996	u32 rcv_isn, u32 filter, u16 window,
3997	u32 rss_qid, u8 port_id)
3998	{
3999	struct sk_buff *req_skb;
4000	struct fw_ofld_connection_wr *req;
4001	struct cpl_pass_accept_req *cpl = cplhdr(skb);
4002	int ret;
4003
4004	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
4005	if (!req_skb)
4006	return;
4007	req = __skb_put_zero(req_skb, sizeof(*req));
4008	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
4009	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4010	req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4011	req->le.filter = (__force __be32) filter;
4012	req->le.lport = lport;
4013	req->le.pport = rport;
4014	req->le.u.ipv4.lip = laddr;
4015	req->le.u.ipv4.pip = raddr;
4016	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4017	req->tcb.rcv_adv = htons(window);
4018	req->tcb.t_state_to_astid =
4019	htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4020	FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4021	FW_OFLD_CONNECTION_WR_ASTID_V(
4022	PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4023
4024	/*
4025	* We store the qid in opt2 which will be used by the firmware
4026	* to send us the wr response.
4027	*/
4028	req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4029
4030	/*
4031	* We initialize the MSS index in TCB to 0xF.
4032	* So that when driver sends cpl_pass_accept_rpl
4033	* TCB picks up the correct value. If this was 0
4034	* TP will ignore any value > 0 for MSS index.
4035	*/
4036	req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4037	req->cookie = (uintptr_t)skb;
4038
4039	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4040	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4041	if (ret < 0) {
4042	pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4043	ret);
4044	kfree_skb(skb);
4045	kfree_skb(skb: req_skb);
4046	}
4047	}
4048
4049	/*
4050	* Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4051	* messages when a filter is being used instead of server to
4052	* redirect a syn packet. When packets hit filter they are redirected
4053	* to the offload queue and driver tries to establish the connection
4054	* using firmware work request.
4055	*/
4056	static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4057	{
4058	int stid;
4059	unsigned int filter;
4060	struct ethhdr *eh = NULL;
4061	struct vlan_ethhdr *vlan_eh = NULL;
4062	struct iphdr *iph;
4063	struct tcphdr *tcph;
4064	struct rss_header *rss = (void *)skb->data;
4065	struct cpl_rx_pkt *cpl = (void *)skb->data;
4066	struct cpl_pass_accept_req *req = (void *)(rss + 1);
4067	struct l2t_entry *e;
4068	struct dst_entry *dst;
4069	struct c4iw_ep *lep = NULL;
4070	u16 window;
4071	struct port_info *pi;
4072	struct net_device *pdev;
4073	u16 rss_qid, eth_hdr_len;
4074	int step;
4075	struct neighbour *neigh;
4076
4077	/* Drop all non-SYN packets */
4078	if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4079	goto reject;
4080
4081	/*
4082	* Drop all packets which did not hit the filter.
4083	* Unlikely to happen.
4084	*/
4085	if (!(rss->filter_hit && rss->filter_tid))
4086	goto reject;
4087
4088	/*
4089	* Calculate the server tid from filter hit index from cpl_rx_pkt.
4090	*/
4091	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4092
4093	lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4094	if (!lep) {
4095	pr_warn("%s connect request on invalid stid %d\n",
4096	__func__, stid);
4097	goto reject;
4098	}
4099
4100	switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4101	case CHELSIO_T4:
4102	eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4103	break;
4104	case CHELSIO_T5:
4105	eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4106	break;
4107	case CHELSIO_T6:
4108	eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4109	break;
4110	default:
4111	pr_err("T%d Chip is not supported\n",
4112	CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4113	goto reject;
4114	}
4115
4116	if (eth_hdr_len == ETH_HLEN) {
4117	eh = (struct ethhdr *)(req + 1);
4118	iph = (struct iphdr *)(eh + 1);
4119	} else {
4120	vlan_eh = (struct vlan_ethhdr *)(req + 1);
4121	iph = (struct iphdr *)(vlan_eh + 1);
4122	__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4123	}
4124
4125	if (iph->version != 0x4)
4126	goto reject;
4127
4128	tcph = (struct tcphdr *)(iph + 1);
4129	skb_set_network_header(skb, offset: (void *)iph - (void *)rss);
4130	skb_set_transport_header(skb, offset: (void *)tcph - (void *)rss);
4131	skb_get(skb);
4132
4133	pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4134	ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4135	ntohs(tcph->source), iph->tos);
4136
4137	dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4138	iph->daddr, iph->saddr, tcph->dest,
4139	tcph->source, iph->tos);
4140	if (!dst) {
4141	pr_err("%s - failed to find dst entry!\n", __func__);
4142	goto reject;
4143	}
4144	neigh = dst_neigh_lookup_skb(dst, skb);
4145
4146	if (!neigh) {
4147	pr_err("%s - failed to allocate neigh!\n", __func__);
4148	goto free_dst;
4149	}
4150
4151	if (neigh->dev->flags & IFF_LOOPBACK) {
4152	pdev = ip_dev_find(net: &init_net, addr: iph->daddr);
4153	if (!pdev) {
4154	pr_err("%s - failed to find device!\n", __func__);
4155	goto free_dst;
4156	}
4157	e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4158	pdev, 0);
4159	pi = (struct port_info *)netdev_priv(dev: pdev);
4160	dev_put(dev: pdev);
4161	} else {
4162	pdev = get_real_dev(egress_dev: neigh->dev);
4163	e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4164	pdev, 0);
4165	pi = (struct port_info *)netdev_priv(dev: pdev);
4166	}
4167	neigh_release(neigh);
4168	if (!e) {
4169	pr_err("%s - failed to allocate l2t entry!\n",
4170	__func__);
4171	goto free_dst;
4172	}
4173
4174	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4175	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4176	window = (__force u16) htons((__force u16)tcph->window);
4177
4178	/* Calcuate filter portion for LE region. */
4179	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4180	dev->rdev.lldi.ports[0],
4181	e));
4182
4183	/*
4184	* Synthesize the cpl_pass_accept_req. We have everything except the
4185	* TID. Once firmware sends a reply with TID we update the TID field
4186	* in cpl and pass it through the regular cpl_pass_accept_req path.
4187	*/
4188	build_cpl_pass_accept_req(skb, stid, tos: iph->tos);
4189	send_fw_pass_open_req(dev, skb, laddr: iph->daddr, lport: tcph->dest, raddr: iph->saddr,
4190	rport: tcph->source, ntohl(tcph->seq), filter, window,
4191	rss_qid, port_id: pi->port_id);
4192	cxgb4_l2t_release(e);
4193	free_dst:
4194	dst_release(dst);
4195	reject:
4196	if (lep)
4197	c4iw_put_ep(&lep->com);
4198	return 0;
4199	}
4200
4201	/*
4202	* These are the real handlers that are called from a
4203	* work queue.
4204	*/
4205	static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4206	[CPL_ACT_ESTABLISH] = act_establish,
4207	[CPL_ACT_OPEN_RPL] = act_open_rpl,
4208	[CPL_RX_DATA] = rx_data,
4209	[CPL_ABORT_RPL_RSS] = abort_rpl,
4210	[CPL_ABORT_RPL] = abort_rpl,
4211	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
4212	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4213	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4214	[CPL_PASS_ESTABLISH] = pass_establish,
4215	[CPL_PEER_CLOSE] = peer_close,
4216	[CPL_ABORT_REQ_RSS] = peer_abort,
4217	[CPL_CLOSE_CON_RPL] = close_con_rpl,
4218	[CPL_RDMA_TERMINATE] = terminate,
4219	[CPL_FW4_ACK] = fw4_ack,
4220	[CPL_GET_TCB_RPL] = read_tcb_rpl,
4221	[CPL_FW6_MSG] = deferred_fw6_msg,
4222	[CPL_RX_PKT] = rx_pkt,
4223	[FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4224	[FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4225	};
4226
4227	static void process_timeout(struct c4iw_ep *ep)
4228	{
4229	struct c4iw_qp_attributes attrs;
4230	int abort = 1;
4231
4232	mutex_lock(&ep->com.mutex);
4233	pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4234	set_bit(nr: TIMEDOUT, addr: &ep->com.history);
4235	switch (ep->com.state) {
4236	case MPA_REQ_SENT:
4237	connect_reply_upcall(ep, status: -ETIMEDOUT);
4238	break;
4239	case MPA_REQ_WAIT:
4240	case MPA_REQ_RCVD:
4241	case MPA_REP_SENT:
4242	case FPDU_MODE:
4243	break;
4244	case CLOSING:
4245	case MORIBUND:
4246	if (ep->com.cm_id && ep->com.qp) {
4247	attrs.next_state = C4IW_QP_STATE_ERROR;
4248	c4iw_modify_qp(rhp: ep->com.qp->rhp,
4249	qhp: ep->com.qp, mask: C4IW_QP_ATTR_NEXT_STATE,
4250	attrs: &attrs, internal: 1);
4251	}
4252	close_complete_upcall(ep, status: -ETIMEDOUT);
4253	break;
4254	case ABORTING:
4255	case DEAD:
4256
4257	/*
4258	* These states are expected if the ep timed out at the same
4259	* time as another thread was calling stop_ep_timer().
4260	* So we silently do nothing for these states.
4261	*/
4262	abort = 0;
4263	break;
4264	default:
4265	WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4266	__func__, ep, ep->hwtid, ep->com.state);
4267	abort = 0;
4268	}
4269	mutex_unlock(lock: &ep->com.mutex);
4270	if (abort)
4271	c4iw_ep_disconnect(ep, abrupt: 1, GFP_KERNEL);
4272	c4iw_put_ep(&ep->com);
4273	}
4274
4275	static void process_timedout_eps(void)
4276	{
4277	struct c4iw_ep *ep;
4278
4279	spin_lock_irq(lock: &timeout_lock);
4280	while (!list_empty(head: &timeout_list)) {
4281	struct list_head *tmp;
4282
4283	tmp = timeout_list.next;
4284	list_del(entry: tmp);
4285	tmp->next = NULL;
4286	tmp->prev = NULL;
4287	spin_unlock_irq(lock: &timeout_lock);
4288	ep = list_entry(tmp, struct c4iw_ep, entry);
4289	process_timeout(ep);
4290	spin_lock_irq(lock: &timeout_lock);
4291	}
4292	spin_unlock_irq(lock: &timeout_lock);
4293	}
4294
4295	static void process_work(struct work_struct *work)
4296	{
4297	struct sk_buff *skb = NULL;
4298	struct c4iw_dev *dev;
4299	struct cpl_act_establish *rpl;
4300	unsigned int opcode;
4301	int ret;
4302
4303	process_timedout_eps();
4304	while ((skb = skb_dequeue(list: &rxq))) {
4305	rpl = cplhdr(skb);
4306	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4307	opcode = rpl->ot.opcode;
4308
4309	if (opcode >= ARRAY_SIZE(work_handlers) ||
4310	!work_handlers[opcode]) {
4311	pr_err("No handler for opcode 0x%x.\n", opcode);
4312	kfree_skb(skb);
4313	} else {
4314	ret = work_handlers[opcode](dev, skb);
4315	if (!ret)
4316	kfree_skb(skb);
4317	}
4318	process_timedout_eps();
4319	}
4320	}
4321
4322	static DECLARE_WORK(skb_work, process_work);
4323
4324	static void ep_timeout(struct timer_list *t)
4325	{
4326	struct c4iw_ep *ep = from_timer(ep, t, timer);
4327	int kickit = 0;
4328
4329	spin_lock(lock: &timeout_lock);
4330	if (!test_and_set_bit(nr: TIMEOUT, addr: &ep->com.flags)) {
4331	/*
4332	* Only insert if it is not already on the list.
4333	*/
4334	if (!ep->entry.next) {
4335	list_add_tail(new: &ep->entry, head: &timeout_list);
4336	kickit = 1;
4337	}
4338	}
4339	spin_unlock(lock: &timeout_lock);
4340	if (kickit)
4341	queue_work(wq: workq, work: &skb_work);
4342	}
4343
4344	/*
4345	* All the CM events are handled on a work queue to have a safe context.
4346	*/
4347	static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4348	{
4349
4350	/*
4351	* Save dev in the skb->cb area.
4352	*/
4353	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4354
4355	/*
4356	* Queue the skb and schedule the worker thread.
4357	*/
4358	skb_queue_tail(list: &rxq, newsk: skb);
4359	queue_work(wq: workq, work: &skb_work);
4360	return 0;
4361	}
4362
4363	static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4364	{
4365	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4366
4367	if (rpl->status != CPL_ERR_NONE) {
4368	pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4369	rpl->status, GET_TID(rpl));
4370	}
4371	kfree_skb(skb);
4372	return 0;
4373	}
4374
4375	static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4376	{
4377	struct cpl_fw6_msg *rpl = cplhdr(skb);
4378	struct c4iw_wr_wait *wr_waitp;
4379	int ret;
4380
4381	pr_debug("type %u\n", rpl->type);
4382
4383	switch (rpl->type) {
4384	case FW6_TYPE_WR_RPL:
4385	ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4386	wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4387	pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4388	if (wr_waitp)
4389	c4iw_wake_up_deref(wr_waitp, ret: ret ? -ret : 0);
4390	kfree_skb(skb);
4391	break;
4392	case FW6_TYPE_CQE:
4393	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4394	sched(dev, skb);
4395	break;
4396	default:
4397	pr_err("%s unexpected fw6 msg type %u\n",
4398	__func__, rpl->type);
4399	kfree_skb(skb);
4400	break;
4401	}
4402	return 0;
4403	}
4404
4405	static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4406	{
4407	struct cpl_abort_req_rss *req = cplhdr(skb);
4408	struct c4iw_ep *ep;
4409	unsigned int tid = GET_TID(req);
4410
4411	ep = get_ep_from_tid(dev, tid);
4412	/* This EP will be dereferenced in peer_abort() */
4413	if (!ep) {
4414	pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4415	kfree_skb(skb);
4416	return 0;
4417	}
4418	if (cxgb_is_neg_adv(req->status)) {
4419	pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4420	ep->hwtid, req->status,
4421	neg_adv_str(req->status));
4422	goto out;
4423	}
4424	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4425
4426	c4iw_wake_up_noref(wr_waitp: ep->com.wr_waitp, ret: -ECONNRESET);
4427	out:
4428	sched(dev, skb);
4429	return 0;
4430	}
4431
4432	/*
4433	* Most upcalls from the T4 Core go to sched() to
4434	* schedule the processing on a work queue.
4435	*/
4436	c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4437	[CPL_ACT_ESTABLISH] = sched,
4438	[CPL_ACT_OPEN_RPL] = sched,
4439	[CPL_RX_DATA] = sched,
4440	[CPL_ABORT_RPL_RSS] = sched,
4441	[CPL_ABORT_RPL] = sched,
4442	[CPL_PASS_OPEN_RPL] = sched,
4443	[CPL_CLOSE_LISTSRV_RPL] = sched,
4444	[CPL_PASS_ACCEPT_REQ] = sched,
4445	[CPL_PASS_ESTABLISH] = sched,
4446	[CPL_PEER_CLOSE] = sched,
4447	[CPL_CLOSE_CON_RPL] = sched,
4448	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
4449	[CPL_RDMA_TERMINATE] = sched,
4450	[CPL_FW4_ACK] = sched,
4451	[CPL_SET_TCB_RPL] = set_tcb_rpl,
4452	[CPL_GET_TCB_RPL] = sched,
4453	[CPL_FW6_MSG] = fw6_msg,
4454	[CPL_RX_PKT] = sched
4455	};
4456
4457	int __init c4iw_cm_init(void)
4458	{
4459	skb_queue_head_init(list: &rxq);
4460
4461	workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4462	if (!workq)
4463	return -ENOMEM;
4464
4465	return 0;
4466	}
4467
4468	void c4iw_cm_term(void)
4469	{
4470	WARN_ON(!list_empty(&timeout_list));
4471	destroy_workqueue(wq: workq);
4472	}
4473