projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net: dsa: mv88e6xxx: better IEEE Prio Mapping Table description
[linux-2.6-block.git] / drivers / infiniband / hw / cxgb4 / cm.c
CommitLineData
cfdda9d7 1/*
9eccfe10 2 * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
cfdda9d7
SW		 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>
1cab775c 41#include <linux/if_vlan.h>
cfdda9d7
SW		 42
43#include <net/neighbour.h>
44#include <net/netevent.h>
45#include <net/route.h>
1cab775c 46#include <net/tcp.h>
830662f6
VP		 47#include <net/ip6_route.h>
48#include <net/addrconf.h>
cfdda9d7 49
11b8e22d
SW		 50#include <rdma/ib_addr.h>
51
85e42b04 52#include <libcxgb_cm.h>
cfdda9d7 53#include "iw_cxgb4.h"
84cc6ac6 54#include "clip_tbl.h"
cfdda9d7
SW		 55
56static 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
5be78ee9
VP		 72static int nocong;
73module_param(nocong, int, 0644);
74MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75
76static int enable_ecn;
77module_param(enable_ecn, int, 0644);
78MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79
b52fe09e 80static int dack_mode = 1;
ba6d3925 81module_param(dack_mode, int, 0644);
b52fe09e 82MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
ba6d3925 83
4c2c5763 84uint c4iw_max_read_depth = 32;
be4c9bad 85module_param(c4iw_max_read_depth, int, 0644);
4c2c5763
HS		 86MODULE_PARM_DESC(c4iw_max_read_depth,
87 "Per-connection max ORD/IRD (default=32)");
be4c9bad 88
cfdda9d7
SW		 89static int enable_tcp_timestamps;
90module_param(enable_tcp_timestamps, int, 0644);
91MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92
93static int enable_tcp_sack;
94module_param(enable_tcp_sack, int, 0644);
95MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96
97static int enable_tcp_window_scaling = 1;
98module_param(enable_tcp_window_scaling, int, 0644);
99MODULE_PARM_DESC(enable_tcp_window_scaling,
100 "Enable tcp window scaling (default=1)");
101
102int c4iw_debug;
103module_param(c4iw_debug, int, 0644);
a9a42886 104MODULE_PARM_DESC(c4iw_debug, "obsolete");
cfdda9d7 105
df2d5130 106static int peer2peer = 1;
cfdda9d7 107module_param(peer2peer, int, 0644);
df2d5130 108MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
cfdda9d7
SW		 109
110static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
111module_param(p2p_type, int, 0644);
112MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
113 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
114
115static int ep_timeout_secs = 60;
116module_param(ep_timeout_secs, int, 0644);
117MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
118 "in seconds (default=60)");
119
b8ac3112 120static int mpa_rev = 2;
cfdda9d7
SW		 121module_param(mpa_rev, int, 0644);
122MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
ccd2c30b 123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
b8ac3112 124 " compliant (default=2)");
cfdda9d7
SW		 125
126static int markers_enabled;
127module_param(markers_enabled, int, 0644);
128MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
129
130static int crc_enabled = 1;
131module_param(crc_enabled, int, 0644);
132MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
133
134static int rcv_win = 256 * 1024;
135module_param(rcv_win, int, 0644);
136MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
137
98ae68b7 138static int snd_win = 128 * 1024;
cfdda9d7 139module_param(snd_win, int, 0644);
98ae68b7 140MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
cfdda9d7 141
cfdda9d7 142static struct workqueue_struct *workq;
cfdda9d7
SW		 143
144static struct sk_buff_head rxq;
cfdda9d7
SW		 145
146static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
147static void ep_timeout(unsigned long arg);
148static void connect_reply_upcall(struct c4iw_ep *ep, int status);
9dec900c 149static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
cfdda9d7 150
be4c9bad
RD		 151static LIST_HEAD(timeout_list);
152static spinlock_t timeout_lock;
153
9ca6f7cf
H		 154static void deref_cm_id(struct c4iw_ep_common *epc)
155{
156 epc->cm_id->rem_ref(epc->cm_id);
157 epc->cm_id = NULL;
158 set_bit(CM_ID_DEREFED, &epc->history);
159}
160
161static void ref_cm_id(struct c4iw_ep_common *epc)
162{
163 set_bit(CM_ID_REFED, &epc->history);
164 epc->cm_id->add_ref(epc->cm_id);
165}
166
325abead
VP		 167static void deref_qp(struct c4iw_ep *ep)
168{
169 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
170 clear_bit(QP_REFERENCED, &ep->com.flags);
9ca6f7cf 171 set_bit(QP_DEREFED, &ep->com.history);
325abead
VP		 172}
173
174static void ref_qp(struct c4iw_ep *ep)
175{
176 set_bit(QP_REFERENCED, &ep->com.flags);
9ca6f7cf 177 set_bit(QP_REFED, &ep->com.history);
325abead
VP		 178 c4iw_qp_add_ref(&ep->com.qp->ibqp);
179}
180
cfdda9d7
SW		 181static void start_ep_timer(struct c4iw_ep *ep)
182{
a9a42886 183 pr_debug("%s ep %p\n", __func__, ep);
cfdda9d7 184 if (timer_pending(&ep->timer)) {
1ec779cc
VP		 185 pr_err("%s timer already started! ep %p\n",
186 __func__, ep);
187 return;
188 }
189 clear_bit(TIMEOUT, &ep->com.flags);
190 c4iw_get_ep(&ep->com);
cfdda9d7
SW		 191 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
192 ep->timer.data = (unsigned long)ep;
193 ep->timer.function = ep_timeout;
194 add_timer(&ep->timer);
195}
196
b33bd0cb 197static int stop_ep_timer(struct c4iw_ep *ep)
cfdda9d7 198{
a9a42886 199 pr_debug("%s ep %p stopping\n", __func__, ep);
cfdda9d7 200 del_timer_sync(&ep->timer);
b33bd0cb 201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
1ec779cc 202 c4iw_put_ep(&ep->com);
b33bd0cb
SW		 203 return 0;
204 }
205 return 1;
cfdda9d7
SW		 206}
207
208static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
209 struct l2t_entry *l2e)
210{
211 int error = 0;
212
213 if (c4iw_fatal_error(rdev)) {
214 kfree_skb(skb);
a9a42886 215 pr_debug("%s - device in error state - dropping\n", __func__);
cfdda9d7
SW		 216 return -EIO;
217 }
218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
219 if (error < 0)
220 kfree_skb(skb);
caa6c9f2
H		 221 else if (error == NET_XMIT_DROP)
222 return -ENOMEM;
74594861 223 return error < 0 ? error : 0;
cfdda9d7
SW		 224}
225
226int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
227{
228 int error = 0;
229
230 if (c4iw_fatal_error(rdev)) {
231 kfree_skb(skb);
a9a42886 232 pr_debug("%s - device in error state - dropping\n", __func__);
cfdda9d7
SW		 233 return -EIO;
234 }
235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
236 if (error < 0)
237 kfree_skb(skb);
74594861 238 return error < 0 ? error : 0;
cfdda9d7
SW		 239}
240
241static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
242{
a1a23454 243 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
cfdda9d7 244
a1a23454 245 skb = get_skb(skb, len, GFP_KERNEL);
cfdda9d7
SW		 246 if (!skb)
247 return;
a1a23454
VP		 248
249 cxgb_mk_tid_release(skb, len, hwtid, 0);
cfdda9d7
SW		 250 c4iw_ofld_send(rdev, skb);
251 return;
252}
253
254static void set_emss(struct c4iw_ep *ep, u16 opt)
255{
6c53e938 256 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
04524a47
H		 257 ((AF_INET == ep->com.remote_addr.ss_family) ?
258 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
259 sizeof(struct tcphdr);
cfdda9d7 260 ep->mss = ep->emss;
6c53e938 261 if (TCPOPT_TSTAMP_G(opt))
04524a47 262 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
cfdda9d7
SW		 263 if (ep->emss < 128)
264 ep->emss = 128;
92e7ae71 265 if (ep->emss & 7)
a9a42886
JP		 266 pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
267 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
268 pr_debug("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
269 ep->mss, ep->emss);
cfdda9d7
SW		 270}
271
272static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
273{
cfdda9d7
SW		 274 enum c4iw_ep_state state;
275
2f5b48c3 276 mutex_lock(&epc->mutex);
cfdda9d7 277 state = epc->state;
2f5b48c3 278 mutex_unlock(&epc->mutex);
cfdda9d7
SW		 279 return state;
280}
281
282static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
283{
284 epc->state = new;
285}
286
287static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
288{
2f5b48c3 289 mutex_lock(&epc->mutex);
a9a42886 290 pr_debug("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
cfdda9d7 291 __state_set(epc, new);
2f5b48c3 292 mutex_unlock(&epc->mutex);
cfdda9d7
SW		 293 return;
294}
295
4a740838
H		 296static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
297{
298 struct sk_buff *skb;
299 unsigned int i;
300 size_t len;
301
302 len = roundup(sizeof(union cpl_wr_size), 16);
303 for (i = 0; i < size; i++) {
304 skb = alloc_skb(len, GFP_KERNEL);
305 if (!skb)
306 goto fail;
307 skb_queue_tail(ep_skb_list, skb);
308 }
309 return 0;
310fail:
311 skb_queue_purge(ep_skb_list);
312 return -ENOMEM;
313}
314
cfdda9d7
SW		 315static void *alloc_ep(int size, gfp_t gfp)
316{
317 struct c4iw_ep_common *epc;
318
319 epc = kzalloc(size, gfp);
320 if (epc) {
321 kref_init(&epc->kref);
2f5b48c3 322 mutex_init(&epc->mutex);
aadc4df3 323 c4iw_init_wr_wait(&epc->wr_wait);
cfdda9d7 324 }
a9a42886 325 pr_debug("%s alloc ep %p\n", __func__, epc);
cfdda9d7
SW		 326 return epc;
327}
328
944661dd
H		 329static void remove_ep_tid(struct c4iw_ep *ep)
330{
331 unsigned long flags;
332
333 spin_lock_irqsave(&ep->com.dev->lock, flags);
334 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
37eb816c
SW		 335 if (idr_is_empty(&ep->com.dev->hwtid_idr))
336 wake_up(&ep->com.dev->wait);
944661dd
H		 337 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
338}
339
340static void insert_ep_tid(struct c4iw_ep *ep)
341{
342 unsigned long flags;
343
344 spin_lock_irqsave(&ep->com.dev->lock, flags);
345 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
346 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
347}
348
349/*
350 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
351 */
352static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
353{
354 struct c4iw_ep *ep;
355 unsigned long flags;
356
357 spin_lock_irqsave(&dev->lock, flags);
358 ep = idr_find(&dev->hwtid_idr, tid);
359 if (ep)
360 c4iw_get_ep(&ep->com);
361 spin_unlock_irqrestore(&dev->lock, flags);
362 return ep;
363}
364
f86fac79
H		 365/*
366 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
367 */
368static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
369 unsigned int stid)
370{
371 struct c4iw_listen_ep *ep;
372 unsigned long flags;
373
374 spin_lock_irqsave(&dev->lock, flags);
375 ep = idr_find(&dev->stid_idr, stid);
376 if (ep)
377 c4iw_get_ep(&ep->com);
378 spin_unlock_irqrestore(&dev->lock, flags);
379 return ep;
380}
381
cfdda9d7
SW		 382void _c4iw_free_ep(struct kref *kref)
383{
384 struct c4iw_ep *ep;
385
386 ep = container_of(kref, struct c4iw_ep, com.kref);
a9a42886 387 pr_debug("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
325abead
VP		 388 if (test_bit(QP_REFERENCED, &ep->com.flags))
389 deref_qp(ep);
cfdda9d7 390 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
84cc6ac6
H		 391 if (ep->com.remote_addr.ss_family == AF_INET6) {
392 struct sockaddr_in6 *sin6 =
393 (struct sockaddr_in6 *)
170003c8 394 &ep->com.local_addr;
84cc6ac6
H		 395
396 cxgb4_clip_release(
397 ep->com.dev->rdev.lldi.ports[0],
398 (const u32 *)&sin6->sin6_addr.s6_addr,
399 1);
400 }
1dec4cec
GG		 401 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
402 ep->com.local_addr.ss_family);
cfdda9d7
SW		 403 dst_release(ep->dst);
404 cxgb4_l2t_release(ep->l2t);
c878b706
H		 405 if (ep->mpa_skb)
406 kfree_skb(ep->mpa_skb);
cfdda9d7 407 }
4a740838
H		 408 if (!skb_queue_empty(&ep->com.ep_skb_list))
409 skb_queue_purge(&ep->com.ep_skb_list);
cfdda9d7
SW		 410 kfree(ep);
411}
412
413static void release_ep_resources(struct c4iw_ep *ep)
414{
415 set_bit(RELEASE_RESOURCES, &ep->com.flags);
944661dd
H		 416
417 /*
418 * If we have a hwtid, then remove it from the idr table
419 * so lookups will no longer find this endpoint. Otherwise
420 * we have a race where one thread finds the ep ptr just
421 * before the other thread is freeing the ep memory.
422 */
423 if (ep->hwtid != -1)
424 remove_ep_tid(ep);
cfdda9d7
SW		 425 c4iw_put_ep(&ep->com);
426}
427
cfdda9d7
SW		 428static int status2errno(int status)
429{
430 switch (status) {
431 case CPL_ERR_NONE:
432 return 0;
433 case CPL_ERR_CONN_RESET:
434 return -ECONNRESET;
435 case CPL_ERR_ARP_MISS:
436 return -EHOSTUNREACH;
437 case CPL_ERR_CONN_TIMEDOUT:
438 return -ETIMEDOUT;
439 case CPL_ERR_TCAM_FULL:
440 return -ENOMEM;
441 case CPL_ERR_CONN_EXIST:
442 return -EADDRINUSE;
443 default:
444 return -EIO;
445 }
446}
447
448/*
449 * Try and reuse skbs already allocated...
450 */
451static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
452{
453 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
454 skb_trim(skb, 0);
455 skb_get(skb);
456 skb_reset_transport_header(skb);
457 } else {
458 skb = alloc_skb(len, gfp);
459 }
b38a0ad8 460 t4_set_arp_err_handler(skb, NULL, NULL);
cfdda9d7
SW		 461 return skb;
462}
463
830662f6
VP		 464static struct net_device *get_real_dev(struct net_device *egress_dev)
465{
11b8e22d 466 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
830662f6
VP		 467}
468
cfdda9d7
SW		 469static void arp_failure_discard(void *handle, struct sk_buff *skb)
470{
700456bd 471 pr_err("ARP failure\n");
cfdda9d7
SW		 472 kfree_skb(skb);
473}
474
64bec74a
H		 475static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
476{
477 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
478}
479
9dec900c 480enum {
8d1f1a6b 481 NUM_FAKE_CPLS = 2,
9dec900c 482 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
8d1f1a6b 483 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
9dec900c
H		 484};
485
486static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
487{
488 struct c4iw_ep *ep;
489
490 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
491 release_ep_resources(ep);
1dad0ebe 492 kfree_skb(skb);
9dec900c
H		 493 return 0;
494}
495
8d1f1a6b
H		 496static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
497{
498 struct c4iw_ep *ep;
499
500 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
501 c4iw_put_ep(&ep->parent_ep->com);
502 release_ep_resources(ep);
1dad0ebe 503 kfree_skb(skb);
8d1f1a6b
H		 504 return 0;
505}
506
9dec900c
H		 507/*
508 * Fake up a special CPL opcode and call sched() so process_work() will call
509 * _put_ep_safe() in a safe context to free the ep resources. This is needed
510 * because ARP error handlers are called in an ATOMIC context, and
511 * _c4iw_free_ep() needs to block.
512 */
8d1f1a6b
H		 513static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
514 int cpl)
9dec900c
H		 515{
516 struct cpl_act_establish *rpl = cplhdr(skb);
517
518 /* Set our special ARP_FAILURE opcode */
8d1f1a6b 519 rpl->ot.opcode = cpl;
9dec900c
H		 520
521 /*
522 * Save ep in the skb->cb area, after where sched() will save the dev
523 * ptr.
524 */
525 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
526 sched(ep->com.dev, skb);
527}
528
529/* Handle an ARP failure for an accept */
530static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
531{
532 struct c4iw_ep *ep = handle;
533
700456bd 534 pr_err("ARP failure during accept - tid %u - dropping connection\n",
9dec900c
H		 535 ep->hwtid);
536
537 __state_set(&ep->com, DEAD);
8d1f1a6b 538 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
9dec900c
H		 539}
540
cfdda9d7
SW		 541/*
542 * Handle an ARP failure for an active open.
543 */
544static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
545{
5dab6d3a
H		 546 struct c4iw_ep *ep = handle;
547
700456bd 548 pr_err("ARP failure during connect\n");
5dab6d3a 549 connect_reply_upcall(ep, -EHOSTUNREACH);
9dec900c 550 __state_set(&ep->com, DEAD);
84cc6ac6
H		 551 if (ep->com.remote_addr.ss_family == AF_INET6) {
552 struct sockaddr_in6 *sin6 =
170003c8 553 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 554 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
555 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
556 }
5dab6d3a
H		 557 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
558 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
8d1f1a6b 559 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
cfdda9d7
SW		 560}
561
562/*
563 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
564 * and send it along.
565 */
566static void abort_arp_failure(void *handle, struct sk_buff *skb)
567{
761e19a5
H		 568 int ret;
569 struct c4iw_ep *ep = handle;
570 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
cfdda9d7
SW		 571 struct cpl_abort_req *req = cplhdr(skb);
572
a9a42886 573 pr_debug("%s rdev %p\n", __func__, rdev);
cfdda9d7 574 req->cmd = CPL_ABORT_NO_RST;
1dad0ebe 575 skb_get(skb);
761e19a5
H		 576 ret = c4iw_ofld_send(rdev, skb);
577 if (ret) {
578 __state_set(&ep->com, DEAD);
579 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
1dad0ebe
RR		 580 } else
581 kfree_skb(skb);
cfdda9d7
SW		 582}
583
4a740838 584static int send_flowc(struct c4iw_ep *ep)
cfdda9d7 585{
cfdda9d7 586 struct fw_flowc_wr *flowc;
4a740838 587 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
cfdda9d7 588 int i;
ac8e4c69
H		 589 u16 vlan = ep->l2t->vlan;
590 int nparams;
591
4a740838
H		 592 if (WARN_ON(!skb))
593 return -ENOMEM;
594
ac8e4c69
H		 595 if (vlan == CPL_L2T_VLAN_NONE)
596 nparams = 8;
597 else
598 nparams = 9;
cfdda9d7 599
4df864c1 600 flowc = __skb_put(skb, FLOWC_LEN);
cfdda9d7 601
e2ac9628 602 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
ac8e4c69 603 FW_FLOWC_WR_NPARAMS_V(nparams));
4a740838 604 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
e2ac9628 605 16)) | FW_WR_FLOWID_V(ep->hwtid));
cfdda9d7
SW		 606
607 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
5167865a 608 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
35b1de55 609 (ep->com.dev->rdev.lldi.pf));
cfdda9d7
SW		 610 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
611 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
612 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
613 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
614 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
615 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
616 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
617 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
618 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
619 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
620 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
b408ff28 621 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
cfdda9d7
SW		 622 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
623 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
ac8e4c69
H		 624 if (nparams == 9) {
625 u16 pri;
626
627 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
628 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
629 flowc->mnemval[8].val = cpu_to_be32(pri);
630 } else {
631 /* Pad WR to 16 byte boundary */
632 flowc->mnemval[8].mnemonic = 0;
633 flowc->mnemval[8].val = 0;
634 }
cfdda9d7
SW		 635 for (i = 0; i < 9; i++) {
636 flowc->mnemval[i].r4[0] = 0;
637 flowc->mnemval[i].r4[1] = 0;
638 flowc->mnemval[i].r4[2] = 0;
639 }
640
641 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
fef4422d 642 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
cfdda9d7
SW		 643}
644
4a740838 645static int send_halfclose(struct c4iw_ep *ep)
cfdda9d7 646{
4a740838 647 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
29fb6f42 648 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
cfdda9d7 649
a9a42886 650 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
4a740838 651 if (WARN_ON(!skb))
cfdda9d7 652 return -ENOMEM;
4a740838 653
29fb6f42
VP		 654 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
655 NULL, arp_failure_discard);
656
cfdda9d7
SW		 657 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
658}
659
4a740838 660static int send_abort(struct c4iw_ep *ep)
cfdda9d7 661{
a7e1a97f 662 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
4a740838 663 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
cfdda9d7 664
a9a42886 665 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
4a740838 666 if (WARN_ON(!req_skb))
cfdda9d7 667 return -ENOMEM;
4a740838 668
a7e1a97f
VP		 669 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
670 ep, abort_arp_failure);
671
4a740838 672 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
cfdda9d7
SW		 673}
674
675static int send_connect(struct c4iw_ep *ep)
676{
963cab50
H		 677 struct cpl_act_open_req *req = NULL;
678 struct cpl_t5_act_open_req *t5req = NULL;
679 struct cpl_t6_act_open_req *t6req = NULL;
680 struct cpl_act_open_req6 *req6 = NULL;
681 struct cpl_t5_act_open_req6 *t5req6 = NULL;
682 struct cpl_t6_act_open_req6 *t6req6 = NULL;
cfdda9d7
SW		 683 struct sk_buff *skb;
684 u64 opt0;
685 u32 opt2;
686 unsigned int mtu_idx;
cc516700 687 u32 wscale;
963cab50 688 int win, sizev4, sizev6, wrlen;
9eccfe10 689 struct sockaddr_in *la = (struct sockaddr_in *)
170003c8 690 &ep->com.local_addr;
9eccfe10 691 struct sockaddr_in *ra = (struct sockaddr_in *)
170003c8 692 &ep->com.remote_addr;
9eccfe10 693 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
170003c8 694 &ep->com.local_addr;
9eccfe10 695 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
170003c8 696 &ep->com.remote_addr;
84cc6ac6 697 int ret;
963cab50
H		 698 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
699 u32 isn = (prandom_u32() & ~7UL) - 1;
192539f4
GG		 700 struct net_device *netdev;
701 u64 params;
702
703 netdev = ep->com.dev->rdev.lldi.ports[0];
963cab50
H		 704
705 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
706 case CHELSIO_T4:
707 sizev4 = sizeof(struct cpl_act_open_req);
708 sizev6 = sizeof(struct cpl_act_open_req6);
709 break;
710 case CHELSIO_T5:
711 sizev4 = sizeof(struct cpl_t5_act_open_req);
712 sizev6 = sizeof(struct cpl_t5_act_open_req6);
713 break;
714 case CHELSIO_T6:
715 sizev4 = sizeof(struct cpl_t6_act_open_req);
716 sizev6 = sizeof(struct cpl_t6_act_open_req6);
717 break;
718 default:
719 pr_err("T%d Chip is not supported\n",
720 CHELSIO_CHIP_VERSION(adapter_type));
721 return -EINVAL;
722 }
830662f6
VP		 723
724 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
725 roundup(sizev4, 16) :
726 roundup(sizev6, 16);
cfdda9d7 727
a9a42886 728 pr_debug("%s ep %p atid %u\n", __func__, ep, ep->atid);
cfdda9d7
SW		 729
730 skb = get_skb(NULL, wrlen, GFP_KERNEL);
731 if (!skb) {
700456bd 732 pr_err("%s - failed to alloc skb\n", __func__);
cfdda9d7
SW		 733 return -ENOMEM;
734 }
d4f1a5c6 735 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
cfdda9d7 736
44c6d069
VP		 737 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
738 enable_tcp_timestamps,
739 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
cc516700 740 wscale = cxgb_compute_wscale(rcv_win);
b408ff28
HS		 741
742 /*
743 * Specify the largest window that will fit in opt0. The
744 * remainder will be specified in the rx_data_ack.
745 */
746 win = ep->rcv_win >> 10;
d7990b0c
AB		 747 if (win > RCV_BUFSIZ_M)
748 win = RCV_BUFSIZ_M;
b408ff28 749
6c53e938 750 opt0 = (nocong ? NO_CONG_F : 0) |
d7990b0c 751 KEEP_ALIVE_F |
6c53e938 752 DELACK_F |
d7990b0c
AB		 753 WND_SCALE_V(wscale) |
754 MSS_IDX_V(mtu_idx) |
755 L2T_IDX_V(ep->l2t->idx) |
756 TX_CHAN_V(ep->tx_chan) |
757 SMAC_SEL_V(ep->smac_idx) |
ac8e4c69 758 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 759 ULP_MODE_V(ULP_MODE_TCPDDP) |
760 RCV_BUFSIZ_V(win);
761 opt2 = RX_CHANNEL_V(0) |
6c53e938 762 CCTRL_ECN_V(enable_ecn) |
d7990b0c 763 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
cfdda9d7 764 if (enable_tcp_timestamps)
6c53e938 765 opt2 |= TSTAMPS_EN_F;
cfdda9d7 766 if (enable_tcp_sack)
6c53e938 767 opt2 |= SACK_EN_F;
cfdda9d7 768 if (wscale && enable_tcp_window_scaling)
d7990b0c 769 opt2 |= WND_SCALE_EN_F;
963cab50
H		 770 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
771 if (peer2peer)
772 isn += 4;
773
d7990b0c 774 opt2 |= T5_OPT_2_VALID_F;
cf7fe64a 775 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
0b741047 776 opt2 |= T5_ISS_F;
92e5011a 777 }
84cc6ac6 778
192539f4
GG		 779 params = cxgb4_select_ntuple(netdev, ep->l2t);
780
84cc6ac6
H		 781 if (ep->com.remote_addr.ss_family == AF_INET6)
782 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
783 (const u32 *)&la6->sin6_addr.s6_addr, 1);
784
5dab6d3a 785 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
cfdda9d7 786
963cab50
H		 787 if (ep->com.remote_addr.ss_family == AF_INET) {
788 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
789 case CHELSIO_T4:
4df864c1 790 req = skb_put(skb, wrlen);
830662f6 791 INIT_TP_WR(req, 0);
963cab50
H		 792 break;
793 case CHELSIO_T5:
4df864c1 794 t5req = skb_put(skb, wrlen);
963cab50
H		 795 INIT_TP_WR(t5req, 0);
796 req = (struct cpl_act_open_req *)t5req;
797 break;
798 case CHELSIO_T6:
4df864c1 799 t6req = skb_put(skb, wrlen);
963cab50
H		 800 INIT_TP_WR(t6req, 0);
801 req = (struct cpl_act_open_req *)t6req;
802 t5req = (struct cpl_t5_act_open_req *)t6req;
803 break;
804 default:
805 pr_err("T%d Chip is not supported\n",
806 CHELSIO_CHIP_VERSION(adapter_type));
807 ret = -EINVAL;
808 goto clip_release;
809 }
810
811 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
812 ((ep->rss_qid<<14) | ep->atid)));
813 req->local_port = la->sin_port;
814 req->peer_port = ra->sin_port;
815 req->local_ip = la->sin_addr.s_addr;
816 req->peer_ip = ra->sin_addr.s_addr;
817 req->opt0 = cpu_to_be64(opt0);
818
819 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
192539f4 820 req->params = cpu_to_be32(params);
830662f6
VP		 821 req->opt2 = cpu_to_be32(opt2);
822 } else {
192539f4
GG		 823 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
824 t5req->params =
825 cpu_to_be64(FILTER_TUPLE_V(params));
826 t5req->rsvd = cpu_to_be32(isn);
a9a42886 827 pr_debug("%s snd_isn %u\n", __func__, t5req->rsvd);
192539f4
GG		 828 t5req->opt2 = cpu_to_be32(opt2);
829 } else {
830 t6req->params =
831 cpu_to_be64(FILTER_TUPLE_V(params));
832 t6req->rsvd = cpu_to_be32(isn);
339e7575 833 pr_debug("%s snd_isn %u\n", __func__, t6req->rsvd);
192539f4
GG		 834 t6req->opt2 = cpu_to_be32(opt2);
835 }
963cab50
H		 836 }
837 } else {
838 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
839 case CHELSIO_T4:
4df864c1 840 req6 = skb_put(skb, wrlen);
830662f6 841 INIT_TP_WR(req6, 0);
963cab50
H		 842 break;
843 case CHELSIO_T5:
4df864c1 844 t5req6 = skb_put(skb, wrlen);
963cab50
H		 845 INIT_TP_WR(t5req6, 0);
846 req6 = (struct cpl_act_open_req6 *)t5req6;
847 break;
848 case CHELSIO_T6:
4df864c1 849 t6req6 = skb_put(skb, wrlen);
963cab50
H		 850 INIT_TP_WR(t6req6, 0);
851 req6 = (struct cpl_act_open_req6 *)t6req6;
852 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
853 break;
854 default:
855 pr_err("T%d Chip is not supported\n",
856 CHELSIO_CHIP_VERSION(adapter_type));
857 ret = -EINVAL;
858 goto clip_release;
859 }
860
861 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
862 ((ep->rss_qid<<14)|ep->atid)));
863 req6->local_port = la6->sin6_port;
864 req6->peer_port = ra6->sin6_port;
865 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
866 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
867 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
868 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
869 req6->opt0 = cpu_to_be64(opt0);
870
871 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
192539f4
GG		 872 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
873 ep->l2t));
830662f6 874 req6->opt2 = cpu_to_be32(opt2);
830662f6 875 } else {
192539f4
GG		 876 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
877 t5req6->params =
878 cpu_to_be64(FILTER_TUPLE_V(params));
879 t5req6->rsvd = cpu_to_be32(isn);
a9a42886 880 pr_debug("%s snd_isn %u\n", __func__, t5req6->rsvd);
192539f4
GG		 881 t5req6->opt2 = cpu_to_be32(opt2);
882 } else {
883 t6req6->params =
884 cpu_to_be64(FILTER_TUPLE_V(params));
885 t6req6->rsvd = cpu_to_be32(isn);
339e7575 886 pr_debug("%s snd_isn %u\n", __func__, t6req6->rsvd);
192539f4
GG		 887 t6req6->opt2 = cpu_to_be32(opt2);
888 }
889
830662f6 890 }
f079af7a
VP		 891 }
892
793dad94 893 set_bit(ACT_OPEN_REQ, &ep->com.history);
84cc6ac6 894 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
963cab50 895clip_release:
84cc6ac6
H		 896 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
897 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
898 (const u32 *)&la6->sin6_addr.s6_addr, 1);
899 return ret;
cfdda9d7
SW		 900}
901
caa6c9f2
H		 902static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
903 u8 mpa_rev_to_use)
cfdda9d7 904{
caa6c9f2 905 int mpalen, wrlen, ret;
cfdda9d7
SW		 906 struct fw_ofld_tx_data_wr *req;
907 struct mpa_message *mpa;
d2fe99e8 908 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7 909
a9a42886
JP		 910 pr_debug("%s ep %p tid %u pd_len %d\n",
911 __func__, ep, ep->hwtid, ep->plen);
cfdda9d7
SW		 912
913 BUG_ON(skb_cloned(skb));
914
915 mpalen = sizeof(*mpa) + ep->plen;
d2fe99e8
KS		 916 if (mpa_rev_to_use == 2)
917 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 918 wrlen = roundup(mpalen + sizeof *req, 16);
919 skb = get_skb(skb, wrlen, GFP_KERNEL);
920 if (!skb) {
921 connect_reply_upcall(ep, -ENOMEM);
caa6c9f2 922 return -ENOMEM;
cfdda9d7
SW		 923 }
924 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
925
b080db58 926 req = skb_put_zero(skb, wrlen);
cfdda9d7 927 req->op_to_immdlen = cpu_to_be32(
e2ac9628
HS		 928 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
929 FW_WR_COMPL_F |
930 FW_WR_IMMDLEN_V(mpalen));
cfdda9d7 931 req->flowid_len16 = cpu_to_be32(
e2ac9628
HS		 932 FW_WR_FLOWID_V(ep->hwtid) |
933 FW_WR_LEN16_V(wrlen >> 4));
cfdda9d7
SW		 934 req->plen = cpu_to_be32(mpalen);
935 req->tunnel_to_proxy = cpu_to_be32(
e2ac9628
HS		 936 FW_OFLD_TX_DATA_WR_FLUSH_F |
937 FW_OFLD_TX_DATA_WR_SHOVE_F);
cfdda9d7
SW		 938
939 mpa = (struct mpa_message *)(req + 1);
940 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
3d4e7994
H		 941
942 mpa->flags = 0;
943 if (crc_enabled)
944 mpa->flags |= MPA_CRC;
945 if (markers_enabled) {
946 mpa->flags |= MPA_MARKERS;
947 ep->mpa_attr.recv_marker_enabled = 1;
948 } else {
949 ep->mpa_attr.recv_marker_enabled = 0;
950 }
951 if (mpa_rev_to_use == 2)
952 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
953
cfdda9d7 954 mpa->private_data_size = htons(ep->plen);
d2fe99e8 955 mpa->revision = mpa_rev_to_use;
01b225e1 956 if (mpa_rev_to_use == 1) {
d2fe99e8 957 ep->tried_with_mpa_v1 = 1;
01b225e1
KS		 958 ep->retry_with_mpa_v1 = 0;
959 }
d2fe99e8
KS		 960
961 if (mpa_rev_to_use == 2) {
f747c34a
RD		 962 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
963 sizeof (struct mpa_v2_conn_params));
a9a42886
JP		 964 pr_debug("%s initiator ird %u ord %u\n", __func__, ep->ird,
965 ep->ord);
d2fe99e8
KS		 966 mpa_v2_params.ird = htons((u16)ep->ird);
967 mpa_v2_params.ord = htons((u16)ep->ord);
968
969 if (peer2peer) {
970 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
971 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
972 mpa_v2_params.ord |=
973 htons(MPA_V2_RDMA_WRITE_RTR);
974 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
975 mpa_v2_params.ord |=
976 htons(MPA_V2_RDMA_READ_RTR);
977 }
978 memcpy(mpa->private_data, &mpa_v2_params,
979 sizeof(struct mpa_v2_conn_params));
cfdda9d7 980
d2fe99e8
KS		 981 if (ep->plen)
982 memcpy(mpa->private_data +
983 sizeof(struct mpa_v2_conn_params),
984 ep->mpa_pkt + sizeof(*mpa), ep->plen);
985 } else
986 if (ep->plen)
987 memcpy(mpa->private_data,
988 ep->mpa_pkt + sizeof(*mpa), ep->plen);
cfdda9d7
SW		 989
990 /*
991 * Reference the mpa skb. This ensures the data area
992 * will remain in memory until the hw acks the tx.
993 * Function fw4_ack() will deref it.
994 */
995 skb_get(skb);
996 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
997 BUG_ON(ep->mpa_skb);
998 ep->mpa_skb = skb;
caa6c9f2
H		 999 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1000 if (ret)
1001 return ret;
cfdda9d7 1002 start_ep_timer(ep);
a7db89eb 1003 __state_set(&ep->com, MPA_REQ_SENT);
cfdda9d7 1004 ep->mpa_attr.initiator = 1;
9c88aa00 1005 ep->snd_seq += mpalen;
caa6c9f2 1006 return ret;
cfdda9d7
SW		 1007}
1008
1009static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1010{
1011 int mpalen, wrlen;
1012 struct fw_ofld_tx_data_wr *req;
1013 struct mpa_message *mpa;
1014 struct sk_buff *skb;
d2fe99e8 1015 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7 1016
a9a42886
JP		 1017 pr_debug("%s ep %p tid %u pd_len %d\n",
1018 __func__, ep, ep->hwtid, ep->plen);
cfdda9d7
SW		 1019
1020 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS		 1021 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1022 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 1023 wrlen = roundup(mpalen + sizeof *req, 16);
1024
1025 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1026 if (!skb) {
700456bd 1027 pr_err("%s - cannot alloc skb!\n", __func__);
cfdda9d7
SW		 1028 return -ENOMEM;
1029 }
1030 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1031
b080db58 1032 req = skb_put_zero(skb, wrlen);
cfdda9d7 1033 req->op_to_immdlen = cpu_to_be32(
e2ac9628
HS		 1034 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1035 FW_WR_COMPL_F |
1036 FW_WR_IMMDLEN_V(mpalen));
cfdda9d7 1037 req->flowid_len16 = cpu_to_be32(
e2ac9628
HS		 1038 FW_WR_FLOWID_V(ep->hwtid) |
1039 FW_WR_LEN16_V(wrlen >> 4));
cfdda9d7
SW		 1040 req->plen = cpu_to_be32(mpalen);
1041 req->tunnel_to_proxy = cpu_to_be32(
e2ac9628
HS		 1042 FW_OFLD_TX_DATA_WR_FLUSH_F |
1043 FW_OFLD_TX_DATA_WR_SHOVE_F);
cfdda9d7
SW		 1044
1045 mpa = (struct mpa_message *)(req + 1);
1046 memset(mpa, 0, sizeof(*mpa));
1047 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1048 mpa->flags = MPA_REJECT;
fe7e0a4d 1049 mpa->revision = ep->mpa_attr.version;
cfdda9d7 1050 mpa->private_data_size = htons(plen);
d2fe99e8
KS		 1051
1052 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1053 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD		 1054 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1055 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS		 1056 mpa_v2_params.ird = htons(((u16)ep->ird) |
1057 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1058 0));
1059 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1060 (p2p_type ==
1061 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1062 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1063 FW_RI_INIT_P2PTYPE_READ_REQ ?
1064 MPA_V2_RDMA_READ_RTR : 0) : 0));
1065 memcpy(mpa->private_data, &mpa_v2_params,
1066 sizeof(struct mpa_v2_conn_params));
1067
1068 if (ep->plen)
1069 memcpy(mpa->private_data +
1070 sizeof(struct mpa_v2_conn_params), pdata, plen);
1071 } else
1072 if (plen)
1073 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW		 1074
1075 /*
1076 * Reference the mpa skb again. This ensures the data area
1077 * will remain in memory until the hw acks the tx.
1078 * Function fw4_ack() will deref it.
1079 */
1080 skb_get(skb);
1081 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
64bec74a 1082 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
cfdda9d7
SW		 1083 BUG_ON(ep->mpa_skb);
1084 ep->mpa_skb = skb;
9c88aa00 1085 ep->snd_seq += mpalen;
cfdda9d7
SW		 1086 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1087}
1088
1089static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1090{
1091 int mpalen, wrlen;
1092 struct fw_ofld_tx_data_wr *req;
1093 struct mpa_message *mpa;
1094 struct sk_buff *skb;
d2fe99e8 1095 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7 1096
a9a42886
JP		 1097 pr_debug("%s ep %p tid %u pd_len %d\n",
1098 __func__, ep, ep->hwtid, ep->plen);
cfdda9d7
SW		 1099
1100 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS		 1101 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1102 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 1103 wrlen = roundup(mpalen + sizeof *req, 16);
1104
1105 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1106 if (!skb) {
700456bd 1107 pr_err("%s - cannot alloc skb!\n", __func__);
cfdda9d7
SW		 1108 return -ENOMEM;
1109 }
1110 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1111
b080db58 1112 req = skb_put_zero(skb, wrlen);
cfdda9d7 1113 req->op_to_immdlen = cpu_to_be32(
e2ac9628
HS		 1114 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1115 FW_WR_COMPL_F |
1116 FW_WR_IMMDLEN_V(mpalen));
cfdda9d7 1117 req->flowid_len16 = cpu_to_be32(
e2ac9628
HS		 1118 FW_WR_FLOWID_V(ep->hwtid) |
1119 FW_WR_LEN16_V(wrlen >> 4));
cfdda9d7
SW		 1120 req->plen = cpu_to_be32(mpalen);
1121 req->tunnel_to_proxy = cpu_to_be32(
e2ac9628
HS		 1122 FW_OFLD_TX_DATA_WR_FLUSH_F |
1123 FW_OFLD_TX_DATA_WR_SHOVE_F);
cfdda9d7
SW		 1124
1125 mpa = (struct mpa_message *)(req + 1);
1126 memset(mpa, 0, sizeof(*mpa));
1127 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
3d4e7994
H		 1128 mpa->flags = 0;
1129 if (ep->mpa_attr.crc_enabled)
1130 mpa->flags |= MPA_CRC;
1131 if (ep->mpa_attr.recv_marker_enabled)
1132 mpa->flags |= MPA_MARKERS;
d2fe99e8 1133 mpa->revision = ep->mpa_attr.version;
cfdda9d7 1134 mpa->private_data_size = htons(plen);
d2fe99e8
KS		 1135
1136 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1137 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD		 1138 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1139 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS		 1140 mpa_v2_params.ird = htons((u16)ep->ird);
1141 mpa_v2_params.ord = htons((u16)ep->ord);
1142 if (peer2peer && (ep->mpa_attr.p2p_type !=
1143 FW_RI_INIT_P2PTYPE_DISABLED)) {
1144 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1145
1146 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1147 mpa_v2_params.ord |=
1148 htons(MPA_V2_RDMA_WRITE_RTR);
1149 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1150 mpa_v2_params.ord |=
1151 htons(MPA_V2_RDMA_READ_RTR);
1152 }
1153
1154 memcpy(mpa->private_data, &mpa_v2_params,
1155 sizeof(struct mpa_v2_conn_params));
1156
1157 if (ep->plen)
1158 memcpy(mpa->private_data +
1159 sizeof(struct mpa_v2_conn_params), pdata, plen);
1160 } else
1161 if (plen)
1162 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW		 1163
1164 /*
1165 * Reference the mpa skb. This ensures the data area
1166 * will remain in memory until the hw acks the tx.
1167 * Function fw4_ack() will deref it.
1168 */
1169 skb_get(skb);
64bec74a 1170 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
cfdda9d7 1171 ep->mpa_skb = skb;
a7db89eb 1172 __state_set(&ep->com, MPA_REP_SENT);
9c88aa00 1173 ep->snd_seq += mpalen;
cfdda9d7
SW		 1174 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1175}
1176
1177static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1178{
1179 struct c4iw_ep *ep;
1180 struct cpl_act_establish *req = cplhdr(skb);
1181 unsigned int tid = GET_TID(req);
6c53e938 1182 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
cfdda9d7 1183 struct tid_info *t = dev->rdev.lldi.tids;
fef4422d 1184 int ret;
cfdda9d7
SW		 1185
1186 ep = lookup_atid(t, atid);
1187
a9a42886
JP		 1188 pr_debug("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1189 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
cfdda9d7 1190
a7db89eb 1191 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 1192 dst_confirm(ep->dst);
1193
1194 /* setup the hwtid for this connection */
1195 ep->hwtid = tid;
1dec4cec 1196 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
944661dd 1197 insert_ep_tid(ep);
cfdda9d7
SW		 1198
1199 ep->snd_seq = be32_to_cpu(req->snd_isn);
1200 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1201
1202 set_emss(ep, ntohs(req->tcp_opt));
1203
1204 /* dealloc the atid */
793dad94 1205 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7 1206 cxgb4_free_atid(t, atid);
793dad94 1207 set_bit(ACT_ESTAB, &ep->com.history);
cfdda9d7
SW		 1208
1209 /* start MPA negotiation */
4a740838 1210 ret = send_flowc(ep);
fef4422d
H		 1211 if (ret)
1212 goto err;
d2fe99e8 1213 if (ep->retry_with_mpa_v1)
caa6c9f2 1214 ret = send_mpa_req(ep, skb, 1);
d2fe99e8 1215 else
caa6c9f2
H		 1216 ret = send_mpa_req(ep, skb, mpa_rev);
1217 if (ret)
1218 goto err;
a7db89eb 1219 mutex_unlock(&ep->com.mutex);
cfdda9d7 1220 return 0;
fef4422d
H		 1221err:
1222 mutex_unlock(&ep->com.mutex);
1223 connect_reply_upcall(ep, -ENOMEM);
1224 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1225 return 0;
cfdda9d7
SW		 1226}
1227
be13b2df 1228static void close_complete_upcall(struct c4iw_ep *ep, int status)
cfdda9d7
SW		 1229{
1230 struct iw_cm_event event;
1231
a9a42886 1232 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 1233 memset(&event, 0, sizeof(event));
1234 event.event = IW_CM_EVENT_CLOSE;
be13b2df 1235 event.status = status;
cfdda9d7 1236 if (ep->com.cm_id) {
a9a42886
JP		 1237 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1238 ep, ep->com.cm_id, ep->hwtid);
cfdda9d7 1239 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
9ca6f7cf 1240 deref_cm_id(&ep->com);
793dad94 1241 set_bit(CLOSE_UPCALL, &ep->com.history);
cfdda9d7
SW		 1242 }
1243}
1244
cfdda9d7
SW		 1245static void peer_close_upcall(struct c4iw_ep *ep)
1246{
1247 struct iw_cm_event event;
1248
a9a42886 1249 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 1250 memset(&event, 0, sizeof(event));
1251 event.event = IW_CM_EVENT_DISCONNECT;
1252 if (ep->com.cm_id) {
a9a42886
JP		 1253 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1254 ep, ep->com.cm_id, ep->hwtid);
cfdda9d7 1255 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 1256 set_bit(DISCONN_UPCALL, &ep->com.history);
cfdda9d7
SW		 1257 }
1258}
1259
1260static void peer_abort_upcall(struct c4iw_ep *ep)
1261{
1262 struct iw_cm_event event;
1263
a9a42886 1264 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 1265 memset(&event, 0, sizeof(event));
1266 event.event = IW_CM_EVENT_CLOSE;
1267 event.status = -ECONNRESET;
1268 if (ep->com.cm_id) {
a9a42886
JP		 1269 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1270 ep->com.cm_id, ep->hwtid);
cfdda9d7 1271 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
9ca6f7cf 1272 deref_cm_id(&ep->com);
793dad94 1273 set_bit(ABORT_UPCALL, &ep->com.history);
cfdda9d7
SW		 1274 }
1275}
1276
1277static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1278{
1279 struct iw_cm_event event;
1280
a9a42886
JP		 1281 pr_debug("%s ep %p tid %u status %d\n",
1282 __func__, ep, ep->hwtid, status);
cfdda9d7
SW		 1283 memset(&event, 0, sizeof(event));
1284 event.event = IW_CM_EVENT_CONNECT_REPLY;
1285 event.status = status;
24d44a39
SW		 1286 memcpy(&event.local_addr, &ep->com.local_addr,
1287 sizeof(ep->com.local_addr));
1288 memcpy(&event.remote_addr, &ep->com.remote_addr,
1289 sizeof(ep->com.remote_addr));
cfdda9d7
SW		 1290
1291 if ((status == 0) || (status == -ECONNREFUSED)) {
d2fe99e8
KS		 1292 if (!ep->tried_with_mpa_v1) {
1293 /* this means MPA_v2 is used */
158c776d
H		 1294 event.ord = ep->ird;
1295 event.ird = ep->ord;
d2fe99e8
KS		 1296 event.private_data_len = ep->plen -
1297 sizeof(struct mpa_v2_conn_params);
1298 event.private_data = ep->mpa_pkt +
1299 sizeof(struct mpa_message) +
1300 sizeof(struct mpa_v2_conn_params);
1301 } else {
1302 /* this means MPA_v1 is used */
158c776d
H		 1303 event.ord = cur_max_read_depth(ep->com.dev);
1304 event.ird = cur_max_read_depth(ep->com.dev);
d2fe99e8
KS		 1305 event.private_data_len = ep->plen;
1306 event.private_data = ep->mpa_pkt +
1307 sizeof(struct mpa_message);
1308 }
cfdda9d7 1309 }
85963e4c 1310
a9a42886
JP		 1311 pr_debug("%s ep %p tid %u status %d\n", __func__, ep,
1312 ep->hwtid, status);
793dad94 1313 set_bit(CONN_RPL_UPCALL, &ep->com.history);
85963e4c
RD		 1314 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1315
9ca6f7cf
H		 1316 if (status < 0)
1317 deref_cm_id(&ep->com);
cfdda9d7
SW		 1318}
1319
be13b2df 1320static int connect_request_upcall(struct c4iw_ep *ep)
cfdda9d7
SW		 1321{
1322 struct iw_cm_event event;
be13b2df 1323 int ret;
cfdda9d7 1324
a9a42886 1325 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 1326 memset(&event, 0, sizeof(event));
1327 event.event = IW_CM_EVENT_CONNECT_REQUEST;
24d44a39
SW		 1328 memcpy(&event.local_addr, &ep->com.local_addr,
1329 sizeof(ep->com.local_addr));
1330 memcpy(&event.remote_addr, &ep->com.remote_addr,
1331 sizeof(ep->com.remote_addr));
cfdda9d7 1332 event.provider_data = ep;
d2fe99e8
KS		 1333 if (!ep->tried_with_mpa_v1) {
1334 /* this means MPA_v2 is used */
1335 event.ord = ep->ord;
1336 event.ird = ep->ird;
1337 event.private_data_len = ep->plen -
1338 sizeof(struct mpa_v2_conn_params);
1339 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1340 sizeof(struct mpa_v2_conn_params);
1341 } else {
1342 /* this means MPA_v1 is used. Send max supported */
4c2c5763
HS		 1343 event.ord = cur_max_read_depth(ep->com.dev);
1344 event.ird = cur_max_read_depth(ep->com.dev);
d2fe99e8
KS		 1345 event.private_data_len = ep->plen;
1346 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1347 }
be13b2df
SW		 1348 c4iw_get_ep(&ep->com);
1349 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1350 &event);
1351 if (ret)
1352 c4iw_put_ep(&ep->com);
793dad94 1353 set_bit(CONNREQ_UPCALL, &ep->com.history);
cfdda9d7 1354 c4iw_put_ep(&ep->parent_ep->com);
be13b2df 1355 return ret;
cfdda9d7
SW		 1356}
1357
1358static void established_upcall(struct c4iw_ep *ep)
1359{
1360 struct iw_cm_event event;
1361
a9a42886 1362 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 1363 memset(&event, 0, sizeof(event));
1364 event.event = IW_CM_EVENT_ESTABLISHED;
3dd9a5dc
H		 1365 event.ird = ep->ord;
1366 event.ord = ep->ird;
cfdda9d7 1367 if (ep->com.cm_id) {
a9a42886 1368 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7 1369 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 1370 set_bit(ESTAB_UPCALL, &ep->com.history);
cfdda9d7
SW		 1371 }
1372}
1373
1374static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1375{
cfdda9d7 1376 struct sk_buff *skb;
6e3b6fc2
VP		 1377 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1378 u32 credit_dack;
cfdda9d7 1379
a9a42886
JP		 1380 pr_debug("%s ep %p tid %u credits %u\n",
1381 __func__, ep, ep->hwtid, credits);
cfdda9d7
SW		 1382 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1383 if (!skb) {
700456bd 1384 pr_err("update_rx_credits - cannot alloc skb!\n");
cfdda9d7
SW		 1385 return 0;
1386 }
1387
b408ff28
HS		 1388 /*
1389 * If we couldn't specify the entire rcv window at connection setup
1390 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1391 * then add the overage in to the credits returned.
1392 */
d7990b0c
AB		 1393 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1394 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
b408ff28 1395
6e3b6fc2
VP		 1396 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1397 RX_DACK_MODE_V(dack_mode);
1398
1399 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1400 credit_dack);
1401
cfdda9d7
SW		 1402 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1403 return credits;
1404}
1405
4c2c5763
HS		 1406#define RELAXED_IRD_NEGOTIATION 1
1407
f8e1e1d1
H		 1408/*
1409 * process_mpa_reply - process streaming mode MPA reply
1410 *
1411 * Returns:
1412 *
1413 * 0 upon success indicating a connect request was delivered to the ULP
1414 * or the mpa request is incomplete but valid so far.
1415 *
1416 * 1 if a failure requires the caller to close the connection.
1417 *
1418 * 2 if a failure requires the caller to abort the connection.
1419 */
cc18b939 1420static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
cfdda9d7
SW		 1421{
1422 struct mpa_message *mpa;
d2fe99e8 1423 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7 1424 u16 plen;
d2fe99e8
KS		 1425 u16 resp_ird, resp_ord;
1426 u8 rtr_mismatch = 0, insuff_ird = 0;
cfdda9d7
SW		 1427 struct c4iw_qp_attributes attrs;
1428 enum c4iw_qp_attr_mask mask;
1429 int err;
cc18b939 1430 int disconnect = 0;
cfdda9d7 1431
a9a42886 1432 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7 1433
cfdda9d7
SW		 1434 /*
1435 * If we get more than the supported amount of private data
1436 * then we must fail this connection.
1437 */
1438 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1439 err = -EINVAL;
da1cecdf 1440 goto err_stop_timer;
cfdda9d7
SW		 1441 }
1442
1443 /*
1444 * copy the new data into our accumulation buffer.
1445 */
1446 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1447 skb->len);
1448 ep->mpa_pkt_len += skb->len;
1449
1450 /*
1451 * if we don't even have the mpa message, then bail.
1452 */
1453 if (ep->mpa_pkt_len < sizeof(*mpa))
cc18b939 1454 return 0;
cfdda9d7
SW		 1455 mpa = (struct mpa_message *) ep->mpa_pkt;
1456
1457 /* Validate MPA header. */
d2fe99e8 1458 if (mpa->revision > mpa_rev) {
700456bd
JP		 1459 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1460 __func__, mpa_rev, mpa->revision);
cfdda9d7 1461 err = -EPROTO;
da1cecdf 1462 goto err_stop_timer;
cfdda9d7
SW		 1463 }
1464 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1465 err = -EPROTO;
da1cecdf 1466 goto err_stop_timer;
cfdda9d7
SW		 1467 }
1468
1469 plen = ntohs(mpa->private_data_size);
1470
1471 /*
1472 * Fail if there's too much private data.
1473 */
1474 if (plen > MPA_MAX_PRIVATE_DATA) {
1475 err = -EPROTO;
da1cecdf 1476 goto err_stop_timer;
cfdda9d7
SW		 1477 }
1478
1479 /*
1480 * If plen does not account for pkt size
1481 */
1482 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1483 err = -EPROTO;
da1cecdf 1484 goto err_stop_timer;
cfdda9d7
SW		 1485 }
1486
1487 ep->plen = (u8) plen;
1488
1489 /*
1490 * If we don't have all the pdata yet, then bail.
1491 * We'll continue process when more data arrives.
1492 */
1493 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
cc18b939 1494 return 0;
cfdda9d7
SW		 1495
1496 if (mpa->flags & MPA_REJECT) {
1497 err = -ECONNREFUSED;
da1cecdf 1498 goto err_stop_timer;
cfdda9d7
SW		 1499 }
1500
da1cecdf
H		 1501 /*
1502 * Stop mpa timer. If it expired, then
1503 * we ignore the MPA reply. process_timeout()
1504 * will abort the connection.
1505 */
1506 if (stop_ep_timer(ep))
1507 return 0;
1508
cfdda9d7
SW		 1509 /*
1510 * If we get here we have accumulated the entire mpa
1511 * start reply message including private data. And
1512 * the MPA header is valid.
1513 */
c529fb50 1514 __state_set(&ep->com, FPDU_MODE);
cfdda9d7 1515 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
cfdda9d7 1516 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS		 1517 ep->mpa_attr.version = mpa->revision;
1518 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1519
1520 if (mpa->revision == 2) {
1521 ep->mpa_attr.enhanced_rdma_conn =
1522 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1523 if (ep->mpa_attr.enhanced_rdma_conn) {
1524 mpa_v2_params = (struct mpa_v2_conn_params *)
1525 (ep->mpa_pkt + sizeof(*mpa));
1526 resp_ird = ntohs(mpa_v2_params->ird) &
1527 MPA_V2_IRD_ORD_MASK;
1528 resp_ord = ntohs(mpa_v2_params->ord) &
1529 MPA_V2_IRD_ORD_MASK;
a9a42886
JP		 1530 pr_debug("%s responder ird %u ord %u ep ird %u ord %u\n",
1531 __func__,
1532 resp_ird, resp_ord, ep->ird, ep->ord);
d2fe99e8
KS		 1533
1534 /*
1535 * This is a double-check. Ideally, below checks are
1536 * not required since ird/ord stuff has been taken
1537 * care of in c4iw_accept_cr
1538 */
4c2c5763
HS		 1539 if (ep->ird < resp_ord) {
1540 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1541 ep->com.dev->rdev.lldi.max_ordird_qp)
1542 ep->ird = resp_ord;
1543 else
1544 insuff_ird = 1;
1545 } else if (ep->ird > resp_ord) {
1546 ep->ird = resp_ord;
1547 }
1548 if (ep->ord > resp_ird) {
1549 if (RELAXED_IRD_NEGOTIATION)
1550 ep->ord = resp_ird;
1551 else
1552 insuff_ird = 1;
1553 }
1554 if (insuff_ird) {
d2fe99e8
KS		 1555 err = -ENOMEM;
1556 ep->ird = resp_ord;
1557 ep->ord = resp_ird;
d2fe99e8
KS		 1558 }
1559
1560 if (ntohs(mpa_v2_params->ird) &
1561 MPA_V2_PEER2PEER_MODEL) {
1562 if (ntohs(mpa_v2_params->ord) &
1563 MPA_V2_RDMA_WRITE_RTR)
1564 ep->mpa_attr.p2p_type =
1565 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1566 else if (ntohs(mpa_v2_params->ord) &
1567 MPA_V2_RDMA_READ_RTR)
1568 ep->mpa_attr.p2p_type =
1569 FW_RI_INIT_P2PTYPE_READ_REQ;
1570 }
1571 }
1572 } else if (mpa->revision == 1)
1573 if (peer2peer)
1574 ep->mpa_attr.p2p_type = p2p_type;
1575
a9a42886
JP		 1576 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1577 __func__, ep->mpa_attr.crc_enabled,
1578 ep->mpa_attr.recv_marker_enabled,
1579 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1580 ep->mpa_attr.p2p_type, p2p_type);
d2fe99e8
KS		 1581
1582 /*
1583 * If responder's RTR does not match with that of initiator, assign
1584 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1585 * generated when moving QP to RTS state.
1586 * A TERM message will be sent after QP has moved to RTS state
1587 */
91018f86 1588 if ((ep->mpa_attr.version == 2) && peer2peer &&
d2fe99e8
KS		 1589 (ep->mpa_attr.p2p_type != p2p_type)) {
1590 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1591 rtr_mismatch = 1;
1592 }
cfdda9d7
SW		 1593
1594 attrs.mpa_attr = ep->mpa_attr;
1595 attrs.max_ird = ep->ird;
1596 attrs.max_ord = ep->ord;
1597 attrs.llp_stream_handle = ep;
1598 attrs.next_state = C4IW_QP_STATE_RTS;
1599
1600 mask = C4IW_QP_ATTR_NEXT_STATE |
1601 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1602 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1603
1604 /* bind QP and TID with INIT_WR */
1605 err = c4iw_modify_qp(ep->com.qp->rhp,
1606 ep->com.qp, mask, &attrs, 1);
1607 if (err)
1608 goto err;
d2fe99e8
KS		 1609
1610 /*
1611 * If responder's RTR requirement did not match with what initiator
1612 * supports, generate TERM message
1613 */
1614 if (rtr_mismatch) {
700456bd 1615 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
d2fe99e8
KS		 1616 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1617 attrs.ecode = MPA_NOMATCH_RTR;
1618 attrs.next_state = C4IW_QP_STATE_TERMINATE;
cc18b939 1619 attrs.send_term = 1;
d2fe99e8 1620 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939 1621 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
d2fe99e8 1622 err = -ENOMEM;
cc18b939 1623 disconnect = 1;
d2fe99e8
KS		 1624 goto out;
1625 }
1626
1627 /*
1628 * Generate TERM if initiator IRD is not sufficient for responder
1629 * provided ORD. Currently, we do the same behaviour even when
1630 * responder provided IRD is also not sufficient as regards to
1631 * initiator ORD.
1632 */
1633 if (insuff_ird) {
700456bd 1634 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
d2fe99e8
KS		 1635 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1636 attrs.ecode = MPA_INSUFF_IRD;
1637 attrs.next_state = C4IW_QP_STATE_TERMINATE;
cc18b939 1638 attrs.send_term = 1;
d2fe99e8 1639 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939 1640 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
d2fe99e8 1641 err = -ENOMEM;
cc18b939 1642 disconnect = 1;
d2fe99e8
KS		 1643 goto out;
1644 }
cfdda9d7 1645 goto out;
da1cecdf
H		 1646err_stop_timer:
1647 stop_ep_timer(ep);
cfdda9d7 1648err:
f8e1e1d1 1649 disconnect = 2;
cfdda9d7
SW		 1650out:
1651 connect_reply_upcall(ep, err);
cc18b939 1652 return disconnect;
cfdda9d7
SW		 1653}
1654
fd6aabe4
H		 1655/*
1656 * process_mpa_request - process streaming mode MPA request
1657 *
1658 * Returns:
1659 *
1660 * 0 upon success indicating a connect request was delivered to the ULP
1661 * or the mpa request is incomplete but valid so far.
1662 *
1663 * 1 if a failure requires the caller to close the connection.
1664 *
1665 * 2 if a failure requires the caller to abort the connection.
1666 */
1667static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
cfdda9d7
SW		 1668{
1669 struct mpa_message *mpa;
d2fe99e8 1670 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7
SW		 1671 u16 plen;
1672
a9a42886 1673 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7 1674
cfdda9d7
SW		 1675 /*
1676 * If we get more than the supported amount of private data
1677 * then we must fail this connection.
1678 */
fd6aabe4
H		 1679 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1680 goto err_stop_timer;
cfdda9d7 1681
a9a42886 1682 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
cfdda9d7
SW		 1683
1684 /*
1685 * Copy the new data into our accumulation buffer.
1686 */
1687 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1688 skb->len);
1689 ep->mpa_pkt_len += skb->len;
1690
1691 /*
1692 * If we don't even have the mpa message, then bail.
1693 * We'll continue process when more data arrives.
1694 */
1695 if (ep->mpa_pkt_len < sizeof(*mpa))
fd6aabe4 1696 return 0;
cfdda9d7 1697
a9a42886 1698 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
cfdda9d7
SW		 1699 mpa = (struct mpa_message *) ep->mpa_pkt;
1700
1701 /*
1702 * Validate MPA Header.
1703 */
d2fe99e8 1704 if (mpa->revision > mpa_rev) {
700456bd
JP		 1705 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1706 __func__, mpa_rev, mpa->revision);
fd6aabe4 1707 goto err_stop_timer;
cfdda9d7
SW		 1708 }
1709
fd6aabe4
H		 1710 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1711 goto err_stop_timer;
cfdda9d7
SW		 1712
1713 plen = ntohs(mpa->private_data_size);
1714
1715 /*
1716 * Fail if there's too much private data.
1717 */
fd6aabe4
H		 1718 if (plen > MPA_MAX_PRIVATE_DATA)
1719 goto err_stop_timer;
cfdda9d7
SW		 1720
1721 /*
1722 * If plen does not account for pkt size
1723 */
fd6aabe4
H		 1724 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1725 goto err_stop_timer;
cfdda9d7
SW		 1726 ep->plen = (u8) plen;
1727
1728 /*
1729 * If we don't have all the pdata yet, then bail.
1730 */
1731 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
fd6aabe4 1732 return 0;
cfdda9d7
SW		 1733
1734 /*
1735 * If we get here we have accumulated the entire mpa
1736 * start reply message including private data.
1737 */
1738 ep->mpa_attr.initiator = 0;
1739 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1740 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1741 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS		 1742 ep->mpa_attr.version = mpa->revision;
1743 if (mpa->revision == 1)
1744 ep->tried_with_mpa_v1 = 1;
1745 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1746
1747 if (mpa->revision == 2) {
1748 ep->mpa_attr.enhanced_rdma_conn =
1749 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1750 if (ep->mpa_attr.enhanced_rdma_conn) {
1751 mpa_v2_params = (struct mpa_v2_conn_params *)
1752 (ep->mpa_pkt + sizeof(*mpa));
1753 ep->ird = ntohs(mpa_v2_params->ird) &
1754 MPA_V2_IRD_ORD_MASK;
7f446abf
SW		 1755 ep->ird = min_t(u32, ep->ird,
1756 cur_max_read_depth(ep->com.dev));
d2fe99e8
KS		 1757 ep->ord = ntohs(mpa_v2_params->ord) &
1758 MPA_V2_IRD_ORD_MASK;
7f446abf
SW		 1759 ep->ord = min_t(u32, ep->ord,
1760 cur_max_read_depth(ep->com.dev));
a9a42886
JP		 1761 pr_debug("%s initiator ird %u ord %u\n",
1762 __func__, ep->ird, ep->ord);
d2fe99e8
KS		 1763 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1764 if (peer2peer) {
1765 if (ntohs(mpa_v2_params->ord) &
1766 MPA_V2_RDMA_WRITE_RTR)
1767 ep->mpa_attr.p2p_type =
1768 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1769 else if (ntohs(mpa_v2_params->ord) &
1770 MPA_V2_RDMA_READ_RTR)
1771 ep->mpa_attr.p2p_type =
1772 FW_RI_INIT_P2PTYPE_READ_REQ;
1773 }
1774 }
1775 } else if (mpa->revision == 1)
1776 if (peer2peer)
1777 ep->mpa_attr.p2p_type = p2p_type;
1778
a9a42886
JP		 1779 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1780 __func__,
1781 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1782 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1783 ep->mpa_attr.p2p_type);
cfdda9d7 1784
e4b76a2a
H		 1785 __state_set(&ep->com, MPA_REQ_RCVD);
1786
1787 /* drive upcall */
1788 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1789 if (ep->parent_ep->com.state != DEAD) {
1790 if (connect_request_upcall(ep))
fd6aabe4 1791 goto err_unlock_parent;
e4b76a2a
H		 1792 } else {
1793 goto err_unlock_parent;
be13b2df 1794 }
e4b76a2a 1795 mutex_unlock(&ep->parent_ep->com.mutex);
fd6aabe4
H		 1796 return 0;
1797
1798err_unlock_parent:
1799 mutex_unlock(&ep->parent_ep->com.mutex);
1800 goto err_out;
1801err_stop_timer:
1802 (void)stop_ep_timer(ep);
1803err_out:
1804 return 2;
cfdda9d7
SW		 1805}
1806
1807static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1808{
1809 struct c4iw_ep *ep;
1810 struct cpl_rx_data *hdr = cplhdr(skb);
1811 unsigned int dlen = ntohs(hdr->len);
1812 unsigned int tid = GET_TID(hdr);
793dad94 1813 __u8 status = hdr->status;
cc18b939 1814 int disconnect = 0;
cfdda9d7 1815
944661dd 1816 ep = get_ep_from_tid(dev, tid);
977116c6
SW		 1817 if (!ep)
1818 return 0;
a9a42886 1819 pr_debug("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
cfdda9d7
SW		 1820 skb_pull(skb, sizeof(*hdr));
1821 skb_trim(skb, dlen);
c529fb50 1822 mutex_lock(&ep->com.mutex);
cfdda9d7 1823
c529fb50 1824 switch (ep->com.state) {
cfdda9d7 1825 case MPA_REQ_SENT:
3bcf96e0 1826 update_rx_credits(ep, dlen);
55abf8df 1827 ep->rcv_seq += dlen;
cc18b939 1828 disconnect = process_mpa_reply(ep, skb);
cfdda9d7
SW		 1829 break;
1830 case MPA_REQ_WAIT:
3bcf96e0 1831 update_rx_credits(ep, dlen);
55abf8df 1832 ep->rcv_seq += dlen;
4a4dd8db 1833 disconnect = process_mpa_request(ep, skb);
cfdda9d7 1834 break;
1557967b
VP		 1835 case FPDU_MODE: {
1836 struct c4iw_qp_attributes attrs;
3bcf96e0
SW		 1837
1838 update_rx_credits(ep, dlen);
1557967b 1839 BUG_ON(!ep->com.qp);
e8e5b927 1840 if (status)
1557967b 1841 pr_err("%s Unexpected streaming data." \
04236df2
VP		 1842 " qpid %u ep %p state %d tid %u status %d\n",
1843 __func__, ep->com.qp->wq.sq.qid, ep,
c529fb50 1844 ep->com.state, ep->hwtid, status);
97d7ec0c 1845 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1557967b 1846 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939
SW		 1847 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1848 disconnect = 1;
cfdda9d7
SW		 1849 break;
1850 }
1557967b
VP		 1851 default:
1852 break;
1853 }
c529fb50 1854 mutex_unlock(&ep->com.mutex);
cc18b939 1855 if (disconnect)
4a4dd8db 1856 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
944661dd 1857 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 1858 return 0;
1859}
1860
1861static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1862{
1863 struct c4iw_ep *ep;
1864 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
cfdda9d7
SW		 1865 int release = 0;
1866 unsigned int tid = GET_TID(rpl);
cfdda9d7 1867
944661dd 1868 ep = get_ep_from_tid(dev, tid);
4984037b 1869 if (!ep) {
700456bd 1870 pr_warn("Abort rpl to freed endpoint\n");
4984037b
VP		 1871 return 0;
1872 }
a9a42886 1873 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2f5b48c3 1874 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 1875 switch (ep->com.state) {
1876 case ABORTING:
91e9c071 1877 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 1878 __state_set(&ep->com, DEAD);
1879 release = 1;
1880 break;
1881 default:
700456bd 1882 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
cfdda9d7
SW		 1883 break;
1884 }
2f5b48c3 1885 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 1886
1887 if (release)
1888 release_ep_resources(ep);
944661dd 1889 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 1890 return 0;
1891}
1892
caa6c9f2 1893static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
5be78ee9
VP		 1894{
1895 struct sk_buff *skb;
1896 struct fw_ofld_connection_wr *req;
1897 unsigned int mtu_idx;
cc516700 1898 u32 wscale;
830662f6 1899 struct sockaddr_in *sin;
b408ff28 1900 int win;
5be78ee9
VP		 1901
1902 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
4df864c1 1903 req = __skb_put(skb, sizeof(*req));
5be78ee9 1904 memset(req, 0, sizeof(*req));
6c53e938 1905 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
e2ac9628 1906 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
41b4f86c
KS		 1907 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1908 ep->com.dev->rdev.lldi.ports[0],
5be78ee9 1909 ep->l2t));
170003c8 1910 sin = (struct sockaddr_in *)&ep->com.local_addr;
830662f6
VP		 1911 req->le.lport = sin->sin_port;
1912 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
170003c8 1913 sin = (struct sockaddr_in *)&ep->com.remote_addr;
830662f6
VP		 1914 req->le.pport = sin->sin_port;
1915 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
5be78ee9 1916 req->tcb.t_state_to_astid =
77a80e23
HS		 1917 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1918 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
5be78ee9 1919 req->tcb.cplrxdataack_cplpassacceptrpl =
77a80e23 1920 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
ef5d6355 1921 req->tcb.tx_max = (__force __be32) jiffies;
793dad94 1922 req->tcb.rcv_adv = htons(1);
44c6d069
VP		 1923 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1924 enable_tcp_timestamps,
1925 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
cc516700 1926 wscale = cxgb_compute_wscale(rcv_win);
b408ff28
HS		 1927
1928 /*
1929 * Specify the largest window that will fit in opt0. The
1930 * remainder will be specified in the rx_data_ack.
1931 */
1932 win = ep->rcv_win >> 10;
d7990b0c
AB		 1933 if (win > RCV_BUFSIZ_M)
1934 win = RCV_BUFSIZ_M;
b408ff28 1935
6c53e938
HS		 1936 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1937 (nocong ? NO_CONG_F : 0) |
d7990b0c 1938 KEEP_ALIVE_F |
6c53e938 1939 DELACK_F |
d7990b0c
AB		 1940 WND_SCALE_V(wscale) |
1941 MSS_IDX_V(mtu_idx) |
1942 L2T_IDX_V(ep->l2t->idx) |
1943 TX_CHAN_V(ep->tx_chan) |
1944 SMAC_SEL_V(ep->smac_idx) |
ac8e4c69 1945 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 1946 ULP_MODE_V(ULP_MODE_TCPDDP) |
1947 RCV_BUFSIZ_V(win));
6c53e938
HS		 1948 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1949 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
d7990b0c 1950 RX_CHANNEL_V(0) |
6c53e938 1951 CCTRL_ECN_V(enable_ecn) |
d7990b0c 1952 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
5be78ee9 1953 if (enable_tcp_timestamps)
6c53e938 1954 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
5be78ee9 1955 if (enable_tcp_sack)
6c53e938 1956 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
5be78ee9 1957 if (wscale && enable_tcp_window_scaling)
d7990b0c
AB		 1958 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1959 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1960 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
793dad94
VP		 1961 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1962 set_bit(ACT_OFLD_CONN, &ep->com.history);
caa6c9f2 1963 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
5be78ee9
VP		 1964}
1965
cfdda9d7 1966/*
4c72efef
H		 1967 * Some of the error codes above implicitly indicate that there is no TID
1968 * allocated with the result of an ACT_OPEN. We use this predicate to make
1969 * that explicit.
cfdda9d7
SW		 1970 */
1971static inline int act_open_has_tid(int status)
1972{
4c72efef
H		 1973 return (status != CPL_ERR_TCAM_PARITY &&
1974 status != CPL_ERR_TCAM_MISS &&
1975 status != CPL_ERR_TCAM_FULL &&
1976 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1977 status != CPL_ERR_CONN_EXIST);
cfdda9d7
SW		 1978}
1979
dd92b124
HS		 1980static char *neg_adv_str(unsigned int status)
1981{
1982 switch (status) {
1983 case CPL_ERR_RTX_NEG_ADVICE:
1984 return "Retransmit timeout";
1985 case CPL_ERR_PERSIST_NEG_ADVICE:
1986 return "Persist timeout";
1987 case CPL_ERR_KEEPALV_NEG_ADVICE:
1988 return "Keepalive timeout";
1989 default:
1990 return "Unknown";
1991 }
1992}
1993
b408ff28
HS		 1994static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1995{
1996 ep->snd_win = snd_win;
1997 ep->rcv_win = rcv_win;
a9a42886
JP		 1998 pr_debug("%s snd_win %d rcv_win %d\n",
1999 __func__, ep->snd_win, ep->rcv_win);
b408ff28
HS		 2000}
2001
793dad94
VP		 2002#define ACT_OPEN_RETRY_COUNT 2
2003
830662f6
VP		 2004static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2005 struct dst_entry *dst, struct c4iw_dev *cdev,
ac8e4c69 2006 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
830662f6
VP		 2007{
2008 struct neighbour *n;
2009 int err, step;
2010 struct net_device *pdev;
2011
2012 n = dst_neigh_lookup(dst, peer_ip);
2013 if (!n)
2014 return -ENODEV;
2015
2016 rcu_read_lock();
2017 err = -ENOMEM;
2018 if (n->dev->flags & IFF_LOOPBACK) {
2019 if (iptype == 4)
2020 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2021 else if (IS_ENABLED(CONFIG_IPV6))
2022 for_each_netdev(&init_net, pdev) {
2023 if (ipv6_chk_addr(&init_net,
2024 (struct in6_addr *)peer_ip,
2025 pdev, 1))
2026 break;
2027 }
2028 else
2029 pdev = NULL;
2030
2031 if (!pdev) {
2032 err = -ENODEV;
2033 goto out;
2034 }
2035 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
ac8e4c69 2036 n, pdev, rt_tos2priority(tos));
609e941a
SW		 2037 if (!ep->l2t) {
2038 dev_put(pdev);
830662f6 2039 goto out;
609e941a 2040 }
830662f6
VP		 2041 ep->mtu = pdev->mtu;
2042 ep->tx_chan = cxgb4_port_chan(pdev);
963cab50
H		 2043 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2044 cxgb4_port_viid(pdev));
830662f6
VP		 2045 step = cdev->rdev.lldi.ntxq /
2046 cdev->rdev.lldi.nchan;
2047 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2048 step = cdev->rdev.lldi.nrxq /
2049 cdev->rdev.lldi.nchan;
2050 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2051 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2052 cxgb4_port_idx(pdev) * step];
b408ff28 2053 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
830662f6
VP		 2054 dev_put(pdev);
2055 } else {
2056 pdev = get_real_dev(n->dev);
2057 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2058 n, pdev, 0);
2059 if (!ep->l2t)
2060 goto out;
2061 ep->mtu = dst_mtu(dst);
11b8e22d 2062 ep->tx_chan = cxgb4_port_chan(pdev);
963cab50
H		 2063 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2064 cxgb4_port_viid(pdev));
830662f6
VP		 2065 step = cdev->rdev.lldi.ntxq /
2066 cdev->rdev.lldi.nchan;
11b8e22d
SW		 2067 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2068 ep->ctrlq_idx = cxgb4_port_idx(pdev);
830662f6
VP		 2069 step = cdev->rdev.lldi.nrxq /
2070 cdev->rdev.lldi.nchan;
2071 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
11b8e22d 2072 cxgb4_port_idx(pdev) * step];
b408ff28 2073 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
830662f6
VP		 2074
2075 if (clear_mpa_v1) {
2076 ep->retry_with_mpa_v1 = 0;
2077 ep->tried_with_mpa_v1 = 0;
2078 }
2079 }
2080 err = 0;
2081out:
2082 rcu_read_unlock();
2083
2084 neigh_release(n);
2085
2086 return err;
2087}
2088
793dad94
VP		 2089static int c4iw_reconnect(struct c4iw_ep *ep)
2090{
2091 int err = 0;
4a740838 2092 int size = 0;
24d44a39 2093 struct sockaddr_in *laddr = (struct sockaddr_in *)
170003c8 2094 &ep->com.cm_id->m_local_addr;
24d44a39 2095 struct sockaddr_in *raddr = (struct sockaddr_in *)
170003c8 2096 &ep->com.cm_id->m_remote_addr;
830662f6 2097 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
170003c8 2098 &ep->com.cm_id->m_local_addr;
830662f6 2099 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
170003c8 2100 &ep->com.cm_id->m_remote_addr;
830662f6
VP		 2101 int iptype;
2102 __u8 *ra;
793dad94 2103
a9a42886 2104 pr_debug("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
793dad94 2105 init_timer(&ep->timer);
093108cb 2106 c4iw_init_wr_wait(&ep->com.wr_wait);
793dad94 2107
4a740838
H		 2108 /* When MPA revision is different on nodes, the node with MPA_rev=2
2109 * tries to reconnect with MPA_rev 1 for the same EP through
2110 * c4iw_reconnect(), where the same EP is assigned with new tid for
2111 * further connection establishment. As we are using the same EP pointer
2112 * for reconnect, few skbs are used during the previous c4iw_connect(),
2113 * which leaves the EP with inadequate skbs for further
2114 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2115 * skb_list() during peer_abort(). Allocate skbs which is already used.
2116 */
2117 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2118 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2119 err = -ENOMEM;
2120 goto fail1;
2121 }
2122
793dad94
VP		 2123 /*
2124 * Allocate an active TID to initiate a TCP connection.
2125 */
2126 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2127 if (ep->atid == -1) {
700456bd 2128 pr_err("%s - cannot alloc atid\n", __func__);
793dad94
VP		 2129 err = -ENOMEM;
2130 goto fail2;
2131 }
2132 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2133
2134 /* find a route */
170003c8 2135 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
804c2f3e
VP		 2136 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2137 laddr->sin_addr.s_addr,
2138 raddr->sin_addr.s_addr,
2139 laddr->sin_port,
2140 raddr->sin_port, ep->com.cm_id->tos);
830662f6
VP		 2141 iptype = 4;
2142 ra = (__u8 *)&raddr->sin_addr;
2143 } else {
95554761
VP		 2144 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2145 get_real_dev,
2146 laddr6->sin6_addr.s6_addr,
2147 raddr6->sin6_addr.s6_addr,
2148 laddr6->sin6_port,
2149 raddr6->sin6_port, 0,
2150 raddr6->sin6_scope_id);
830662f6
VP		 2151 iptype = 6;
2152 ra = (__u8 *)&raddr6->sin6_addr;
2153 }
2154 if (!ep->dst) {
700456bd 2155 pr_err("%s - cannot find route\n", __func__);
793dad94
VP		 2156 err = -EHOSTUNREACH;
2157 goto fail3;
2158 }
963cab50 2159 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
ac8e4c69
H		 2160 ep->com.dev->rdev.lldi.adapter_type,
2161 ep->com.cm_id->tos);
830662f6 2162 if (err) {
700456bd 2163 pr_err("%s - cannot alloc l2e\n", __func__);
793dad94
VP		 2164 goto fail4;
2165 }
2166
a9a42886
JP		 2167 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2168 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2169 ep->l2t->idx);
793dad94
VP		 2170
2171 state_set(&ep->com, CONNECTING);
ac8e4c69 2172 ep->tos = ep->com.cm_id->tos;
793dad94
VP		 2173
2174 /* send connect request to rnic */
2175 err = send_connect(ep);
2176 if (!err)
2177 goto out;
2178
2179 cxgb4_l2t_release(ep->l2t);
2180fail4:
2181 dst_release(ep->dst);
2182fail3:
2183 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2184 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2185fail2:
2186 /*
2187 * remember to send notification to upper layer.
2188 * We are in here so the upper layer is not aware that this is
2189 * re-connect attempt and so, upper layer is still waiting for
2190 * response of 1st connect request.
2191 */
2192 connect_reply_upcall(ep, -ECONNRESET);
4a740838 2193fail1:
793dad94
VP		 2194 c4iw_put_ep(&ep->com);
2195out:
2196 return err;
2197}
2198
cfdda9d7
SW		 2199static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2200{
2201 struct c4iw_ep *ep;
2202 struct cpl_act_open_rpl *rpl = cplhdr(skb);
6c53e938
HS		 2203 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2204 ntohl(rpl->atid_status)));
cfdda9d7 2205 struct tid_info *t = dev->rdev.lldi.tids;
6c53e938 2206 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
830662f6
VP		 2207 struct sockaddr_in *la;
2208 struct sockaddr_in *ra;
2209 struct sockaddr_in6 *la6;
2210 struct sockaddr_in6 *ra6;
caa6c9f2 2211 int ret = 0;
cfdda9d7
SW		 2212
2213 ep = lookup_atid(t, atid);
170003c8
SW		 2214 la = (struct sockaddr_in *)&ep->com.local_addr;
2215 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2216 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2217 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
cfdda9d7 2218
a9a42886
JP		 2219 pr_debug("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2220 status, status2errno(status));
cfdda9d7 2221
b65eef0a 2222 if (cxgb_is_neg_adv(status)) {
a9a42886
JP		 2223 pr_debug("%s Connection problems for atid %u status %u (%s)\n",
2224 __func__, atid, status, neg_adv_str(status));
179d03bb
H		 2225 ep->stats.connect_neg_adv++;
2226 mutex_lock(&dev->rdev.stats.lock);
2227 dev->rdev.stats.neg_adv++;
2228 mutex_unlock(&dev->rdev.stats.lock);
cfdda9d7
SW		 2229 return 0;
2230 }
2231
793dad94
VP		 2232 set_bit(ACT_OPEN_RPL, &ep->com.history);
2233
d716a2a0
VP		 2234 /*
2235 * Log interesting failures.
2236 */
2237 switch (status) {
2238 case CPL_ERR_CONN_RESET:
2239 case CPL_ERR_CONN_TIMEDOUT:
2240 break;
5be78ee9 2241 case CPL_ERR_TCAM_FULL:
830662f6 2242 mutex_lock(&dev->rdev.stats.lock);
3b174d94 2243 dev->rdev.stats.tcam_full++;
830662f6
VP		 2244 mutex_unlock(&dev->rdev.stats.lock);
2245 if (ep->com.local_addr.ss_family == AF_INET &&
2246 dev->rdev.lldi.enable_fw_ofld_conn) {
caa6c9f2
H		 2247 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2248 ntohl(rpl->atid_status))));
2249 if (ret)
2250 goto fail;
793dad94
VP		 2251 return 0;
2252 }
2253 break;
2254 case CPL_ERR_CONN_EXIST:
2255 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2256 set_bit(ACT_RETRY_INUSE, &ep->com.history);
84cc6ac6
H		 2257 if (ep->com.remote_addr.ss_family == AF_INET6) {
2258 struct sockaddr_in6 *sin6 =
2259 (struct sockaddr_in6 *)
170003c8 2260 &ep->com.local_addr;
84cc6ac6
H		 2261 cxgb4_clip_release(
2262 ep->com.dev->rdev.lldi.ports[0],
2263 (const u32 *)
2264 &sin6->sin6_addr.s6_addr, 1);
2265 }
793dad94
VP		 2266 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2267 atid);
2268 cxgb4_free_atid(t, atid);
2269 dst_release(ep->dst);
2270 cxgb4_l2t_release(ep->l2t);
2271 c4iw_reconnect(ep);
2272 return 0;
2273 }
5be78ee9 2274 break;
d716a2a0 2275 default:
830662f6
VP		 2276 if (ep->com.local_addr.ss_family == AF_INET) {
2277 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2278 atid, status, status2errno(status),
2279 &la->sin_addr.s_addr, ntohs(la->sin_port),
2280 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2281 } else {
2282 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2283 atid, status, status2errno(status),
2284 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2285 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2286 }
d716a2a0
VP		 2287 break;
2288 }
2289
caa6c9f2 2290fail:
cfdda9d7
SW		 2291 connect_reply_upcall(ep, status2errno(status));
2292 state_set(&ep->com, DEAD);
2293
84cc6ac6
H		 2294 if (ep->com.remote_addr.ss_family == AF_INET6) {
2295 struct sockaddr_in6 *sin6 =
170003c8 2296 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 2297 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2298 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2299 }
cfdda9d7 2300 if (status && act_open_has_tid(status))
1dec4cec
GG		 2301 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2302 ep->com.local_addr.ss_family);
cfdda9d7 2303
793dad94 2304 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7
SW		 2305 cxgb4_free_atid(t, atid);
2306 dst_release(ep->dst);
2307 cxgb4_l2t_release(ep->l2t);
2308 c4iw_put_ep(&ep->com);
2309
2310 return 0;
2311}
2312
2313static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2314{
2315 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
cfdda9d7 2316 unsigned int stid = GET_TID(rpl);
f86fac79 2317 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
cfdda9d7
SW		 2318
2319 if (!ep) {
a9a42886 2320 pr_debug("%s stid %d lookup failure!\n", __func__, stid);
1cab775c 2321 goto out;
cfdda9d7 2322 }
a9a42886
JP		 2323 pr_debug("%s ep %p status %d error %d\n", __func__, ep,
2324 rpl->status, status2errno(rpl->status));
d9594d99 2325 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
f86fac79 2326 c4iw_put_ep(&ep->com);
1cab775c 2327out:
cfdda9d7
SW		 2328 return 0;
2329}
2330
cfdda9d7
SW		 2331static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2332{
2333 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
cfdda9d7 2334 unsigned int stid = GET_TID(rpl);
f86fac79 2335 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
cfdda9d7 2336
a9a42886 2337 pr_debug("%s ep %p\n", __func__, ep);
d9594d99 2338 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
f86fac79 2339 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2340 return 0;
2341}
2342
9dec900c
H		 2343static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2344 struct cpl_pass_accept_req *req)
cfdda9d7
SW		 2345{
2346 struct cpl_pass_accept_rpl *rpl;
2347 unsigned int mtu_idx;
2348 u64 opt0;
2349 u32 opt2;
cc516700 2350 u32 wscale;
92e7ae71 2351 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
b408ff28 2352 int win;
963cab50 2353 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
cfdda9d7 2354
a9a42886 2355 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7 2356 BUG_ON(skb_cloned(skb));
92e7ae71 2357
cfdda9d7 2358 skb_get(skb);
92e7ae71 2359 rpl = cplhdr(skb);
963cab50 2360 if (!is_t4(adapter_type)) {
92e7ae71
HS		 2361 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2362 rpl5 = (void *)rpl;
2363 INIT_TP_WR(rpl5, ep->hwtid);
2364 } else {
2365 skb_trim(skb, sizeof(*rpl));
2366 INIT_TP_WR(rpl, ep->hwtid);
2367 }
2368 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2369 ep->hwtid));
2370
44c6d069
VP		 2371 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2372 enable_tcp_timestamps && req->tcpopt.tstamp,
2373 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
cc516700 2374 wscale = cxgb_compute_wscale(rcv_win);
b408ff28
HS		 2375
2376 /*
2377 * Specify the largest window that will fit in opt0. The
2378 * remainder will be specified in the rx_data_ack.
2379 */
2380 win = ep->rcv_win >> 10;
d7990b0c
AB		 2381 if (win > RCV_BUFSIZ_M)
2382 win = RCV_BUFSIZ_M;
6c53e938 2383 opt0 = (nocong ? NO_CONG_F : 0) |
d7990b0c 2384 KEEP_ALIVE_F |
6c53e938 2385 DELACK_F |
d7990b0c
AB		 2386 WND_SCALE_V(wscale) |
2387 MSS_IDX_V(mtu_idx) |
2388 L2T_IDX_V(ep->l2t->idx) |
2389 TX_CHAN_V(ep->tx_chan) |
2390 SMAC_SEL_V(ep->smac_idx) |
6c53e938 2391 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 2392 ULP_MODE_V(ULP_MODE_TCPDDP) |
2393 RCV_BUFSIZ_V(win);
2394 opt2 = RX_CHANNEL_V(0) |
2395 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
cfdda9d7
SW		 2396
2397 if (enable_tcp_timestamps && req->tcpopt.tstamp)
6c53e938 2398 opt2 |= TSTAMPS_EN_F;
cfdda9d7 2399 if (enable_tcp_sack && req->tcpopt.sack)
6c53e938 2400 opt2 |= SACK_EN_F;
cfdda9d7 2401 if (wscale && enable_tcp_window_scaling)
d7990b0c 2402 opt2 |= WND_SCALE_EN_F;
5be78ee9
VP		 2403 if (enable_ecn) {
2404 const struct tcphdr *tcph;
2405 u32 hlen = ntohl(req->hdr_len);
2406
963cab50
H		 2407 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2408 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2409 IP_HDR_LEN_G(hlen);
2410 else
2411 tcph = (const void *)(req + 1) +
2412 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
5be78ee9 2413 if (tcph->ece && tcph->cwr)
6c53e938 2414 opt2 |= CCTRL_ECN_V(1);
5be78ee9 2415 }
963cab50 2416 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
92e7ae71 2417 u32 isn = (prandom_u32() & ~7UL) - 1;
d7990b0c 2418 opt2 |= T5_OPT_2_VALID_F;
cf7fe64a 2419 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
0b741047 2420 opt2 |= T5_ISS_F;
92e7ae71
HS		 2421 rpl5 = (void *)rpl;
2422 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2423 if (peer2peer)
2424 isn += 4;
2425 rpl5->iss = cpu_to_be32(isn);
a9a42886 2426 pr_debug("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
92e5011a 2427 }
cfdda9d7 2428
cfdda9d7
SW		 2429 rpl->opt0 = cpu_to_be64(opt0);
2430 rpl->opt2 = cpu_to_be32(opt2);
d4f1a5c6 2431 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
9dec900c 2432 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
cfdda9d7 2433
9dec900c 2434 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
cfdda9d7
SW		 2435}
2436
830662f6 2437static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
cfdda9d7 2438{
a9a42886 2439 pr_debug("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
cfdda9d7
SW		 2440 BUG_ON(skb_cloned(skb));
2441 skb_trim(skb, sizeof(struct cpl_tid_release));
cfdda9d7
SW		 2442 release_tid(&dev->rdev, hwtid, skb);
2443 return;
2444}
2445
cfdda9d7
SW		 2446static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2447{
793dad94 2448 struct c4iw_ep *child_ep = NULL, *parent_ep;
cfdda9d7 2449 struct cpl_pass_accept_req *req = cplhdr(skb);
6c53e938 2450 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
cfdda9d7
SW		 2451 struct tid_info *t = dev->rdev.lldi.tids;
2452 unsigned int hwtid = GET_TID(req);
2453 struct dst_entry *dst;
830662f6 2454 __u8 local_ip[16], peer_ip[16];
cfdda9d7 2455 __be16 local_port, peer_port;
84cc6ac6 2456 struct sockaddr_in6 *sin6;
3786cf18 2457 int err;
1cab775c 2458 u16 peer_mss = ntohs(req->tcpopt.mss);
830662f6 2459 int iptype;
92e7ae71 2460 unsigned short hdrs;
ac8e4c69 2461 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
cfdda9d7 2462
f86fac79 2463 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
1cab775c 2464 if (!parent_ep) {
a9a42886
JP		 2465 pr_debug("%s connect request on invalid stid %d\n",
2466 __func__, stid);
1cab775c
VP		 2467 goto reject;
2468 }
1cab775c 2469
cfdda9d7 2470 if (state_read(&parent_ep->com) != LISTEN) {
a9a42886 2471 pr_debug("%s - listening ep not in LISTEN\n", __func__);
cfdda9d7
SW		 2472 goto reject;
2473 }
2474
85e42b04
VP		 2475 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2476 &iptype, local_ip, peer_ip, &local_port, &peer_port);
830662f6 2477
cfdda9d7 2478 /* Find output route */
830662f6 2479 if (iptype == 4) {
a9a42886
JP		 2480 pr_debug("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2481 , __func__, parent_ep, hwtid,
2482 local_ip, peer_ip, ntohs(local_port),
2483 ntohs(peer_port), peer_mss);
804c2f3e
VP		 2484 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2485 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2486 local_port, peer_port, tos);
830662f6 2487 } else {
a9a42886
JP		 2488 pr_debug("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2489 , __func__, parent_ep, hwtid,
2490 local_ip, peer_ip, ntohs(local_port),
2491 ntohs(peer_port), peer_mss);
95554761
VP		 2492 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2493 local_ip, peer_ip, local_port, peer_port,
2494 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2495 ((struct sockaddr_in6 *)
2496 &parent_ep->com.local_addr)->sin6_scope_id);
830662f6
VP		 2497 }
2498 if (!dst) {
700456bd 2499 pr_err("%s - failed to find dst entry!\n", __func__);
cfdda9d7
SW		 2500 goto reject;
2501 }
3786cf18
DM		 2502
2503 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2504 if (!child_ep) {
700456bd 2505 pr_err("%s - failed to allocate ep entry!\n", __func__);
cfdda9d7
SW		 2506 dst_release(dst);
2507 goto reject;
2508 }
2509
963cab50 2510 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
ac8e4c69 2511 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
3786cf18 2512 if (err) {
700456bd 2513 pr_err("%s - failed to allocate l2t entry!\n", __func__);
cfdda9d7 2514 dst_release(dst);
3786cf18 2515 kfree(child_ep);
cfdda9d7
SW		 2516 goto reject;
2517 }
3786cf18 2518
98b80a2a
RR		 2519 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2520 sizeof(struct tcphdr) +
92e7ae71
HS		 2521 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2522 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2523 child_ep->mtu = peer_mss + hdrs;
1cab775c 2524
4a740838
H		 2525 skb_queue_head_init(&child_ep->com.ep_skb_list);
2526 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2527 goto fail;
2528
cfdda9d7
SW		 2529 state_set(&child_ep->com, CONNECTING);
2530 child_ep->com.dev = dev;
2531 child_ep->com.cm_id = NULL;
5b6b8fe6 2532
830662f6
VP		 2533 if (iptype == 4) {
2534 struct sockaddr_in *sin = (struct sockaddr_in *)
170003c8 2535 &child_ep->com.local_addr;
5b6b8fe6 2536
b462b06e 2537 sin->sin_family = AF_INET;
830662f6
VP		 2538 sin->sin_port = local_port;
2539 sin->sin_addr.s_addr = *(__be32 *)local_ip;
5b6b8fe6
SW		 2540
2541 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
b462b06e 2542 sin->sin_family = AF_INET;
5b6b8fe6
SW		 2543 sin->sin_port = ((struct sockaddr_in *)
2544 &parent_ep->com.local_addr)->sin_port;
2545 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2546
170003c8 2547 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
b462b06e 2548 sin->sin_family = AF_INET;
830662f6
VP		 2549 sin->sin_port = peer_port;
2550 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2551 } else {
170003c8 2552 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
830662f6
VP		 2553 sin6->sin6_family = PF_INET6;
2554 sin6->sin6_port = local_port;
2555 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
5b6b8fe6
SW		 2556
2557 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2558 sin6->sin6_family = PF_INET6;
2559 sin6->sin6_port = ((struct sockaddr_in6 *)
2560 &parent_ep->com.local_addr)->sin6_port;
2561 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2562
170003c8 2563 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
830662f6
VP		 2564 sin6->sin6_family = PF_INET6;
2565 sin6->sin6_port = peer_port;
2566 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2567 }
5b6b8fe6 2568
cfdda9d7
SW		 2569 c4iw_get_ep(&parent_ep->com);
2570 child_ep->parent_ep = parent_ep;
ac8e4c69 2571 child_ep->tos = tos;
cfdda9d7
SW		 2572 child_ep->dst = dst;
2573 child_ep->hwtid = hwtid;
cfdda9d7 2574
a9a42886
JP		 2575 pr_debug("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2576 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
cfdda9d7
SW		 2577
2578 init_timer(&child_ep->timer);
1dec4cec
GG		 2579 cxgb4_insert_tid(t, child_ep, hwtid,
2580 child_ep->com.local_addr.ss_family);
944661dd 2581 insert_ep_tid(child_ep);
9dec900c
H		 2582 if (accept_cr(child_ep, skb, req)) {
2583 c4iw_put_ep(&parent_ep->com);
2584 release_ep_resources(child_ep);
2585 } else {
2586 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2587 }
84cc6ac6 2588 if (iptype == 6) {
170003c8 2589 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
84cc6ac6
H		 2590 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2591 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2592 }
cfdda9d7 2593 goto out;
4a740838
H		 2594fail:
2595 c4iw_put_ep(&child_ep->com);
cfdda9d7 2596reject:
830662f6 2597 reject_cr(dev, hwtid, skb);
f86fac79
H		 2598 if (parent_ep)
2599 c4iw_put_ep(&parent_ep->com);
cfdda9d7
SW		 2600out:
2601 return 0;
2602}
2603
2604static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2605{
2606 struct c4iw_ep *ep;
2607 struct cpl_pass_establish *req = cplhdr(skb);
cfdda9d7 2608 unsigned int tid = GET_TID(req);
fef4422d 2609 int ret;
cfdda9d7 2610
944661dd 2611 ep = get_ep_from_tid(dev, tid);
a9a42886 2612 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 2613 ep->snd_seq = be32_to_cpu(req->snd_isn);
2614 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2615
a9a42886
JP		 2616 pr_debug("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2617 ntohs(req->tcp_opt));
1cab775c 2618
cfdda9d7
SW		 2619 set_emss(ep, ntohs(req->tcp_opt));
2620
2621 dst_confirm(ep->dst);
fef4422d
H		 2622 mutex_lock(&ep->com.mutex);
2623 ep->com.state = MPA_REQ_WAIT;
cfdda9d7 2624 start_ep_timer(ep);
793dad94 2625 set_bit(PASS_ESTAB, &ep->com.history);
4a740838 2626 ret = send_flowc(ep);
fef4422d
H		 2627 mutex_unlock(&ep->com.mutex);
2628 if (ret)
2629 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
944661dd 2630 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2631
2632 return 0;
2633}
2634
2635static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2636{
2637 struct cpl_peer_close *hdr = cplhdr(skb);
2638 struct c4iw_ep *ep;
2639 struct c4iw_qp_attributes attrs;
cfdda9d7
SW		 2640 int disconnect = 1;
2641 int release = 0;
cfdda9d7 2642 unsigned int tid = GET_TID(hdr);
8da7e7a5 2643 int ret;
cfdda9d7 2644
944661dd
H		 2645 ep = get_ep_from_tid(dev, tid);
2646 if (!ep)
2647 return 0;
2648
a9a42886 2649 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 2650 dst_confirm(ep->dst);
2651
793dad94 2652 set_bit(PEER_CLOSE, &ep->com.history);
2f5b48c3 2653 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2654 switch (ep->com.state) {
2655 case MPA_REQ_WAIT:
2656 __state_set(&ep->com, CLOSING);
2657 break;
2658 case MPA_REQ_SENT:
2659 __state_set(&ep->com, CLOSING);
2660 connect_reply_upcall(ep, -ECONNRESET);
2661 break;
2662 case MPA_REQ_RCVD:
2663
2664 /*
2665 * We're gonna mark this puppy DEAD, but keep
2666 * the reference on it until the ULP accepts or
2667 * rejects the CR. Also wake up anyone waiting
2668 * in rdma connection migration (see c4iw_accept_cr()).
2669 */
2670 __state_set(&ep->com, CLOSING);
a9a42886 2671 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2672 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2673 break;
2674 case MPA_REP_SENT:
2675 __state_set(&ep->com, CLOSING);
a9a42886 2676 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2677 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2678 break;
2679 case FPDU_MODE:
ca5a2202 2680 start_ep_timer(ep);
cfdda9d7 2681 __state_set(&ep->com, CLOSING);
30c95c2d 2682 attrs.next_state = C4IW_QP_STATE_CLOSING;
8da7e7a5 2683 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
30c95c2d 2684 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
8da7e7a5
SW		 2685 if (ret != -ECONNRESET) {
2686 peer_close_upcall(ep);
2687 disconnect = 1;
2688 }
cfdda9d7
SW		 2689 break;
2690 case ABORTING:
2691 disconnect = 0;
2692 break;
2693 case CLOSING:
2694 __state_set(&ep->com, MORIBUND);
2695 disconnect = 0;
2696 break;
2697 case MORIBUND:
b33bd0cb 2698 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2699 if (ep->com.cm_id && ep->com.qp) {
2700 attrs.next_state = C4IW_QP_STATE_IDLE;
2701 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2702 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2703 }
be13b2df 2704 close_complete_upcall(ep, 0);
cfdda9d7
SW		 2705 __state_set(&ep->com, DEAD);
2706 release = 1;
2707 disconnect = 0;
2708 break;
2709 case DEAD:
2710 disconnect = 0;
2711 break;
2712 default:
2713 BUG_ON(1);
2714 }
2f5b48c3 2715 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2716 if (disconnect)
2717 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2718 if (release)
2719 release_ep_resources(ep);
944661dd 2720 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2721 return 0;
2722}
2723
cfdda9d7
SW		 2724static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2725{
2726 struct cpl_abort_req_rss *req = cplhdr(skb);
2727 struct c4iw_ep *ep;
cfdda9d7
SW		 2728 struct sk_buff *rpl_skb;
2729 struct c4iw_qp_attributes attrs;
2730 int ret;
2731 int release = 0;
cfdda9d7 2732 unsigned int tid = GET_TID(req);
052f4731 2733 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
cfdda9d7 2734
944661dd
H		 2735 ep = get_ep_from_tid(dev, tid);
2736 if (!ep)
2737 return 0;
2738
b65eef0a 2739 if (cxgb_is_neg_adv(req->status)) {
a9a42886
JP		 2740 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
2741 __func__, ep->hwtid, req->status,
2742 neg_adv_str(req->status));
179d03bb
H		 2743 ep->stats.abort_neg_adv++;
2744 mutex_lock(&dev->rdev.stats.lock);
2745 dev->rdev.stats.neg_adv++;
2746 mutex_unlock(&dev->rdev.stats.lock);
944661dd 2747 goto deref_ep;
cfdda9d7 2748 }
a9a42886
JP		 2749 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2750 ep->com.state);
793dad94 2751 set_bit(PEER_ABORT, &ep->com.history);
2f5b48c3
SW		 2752
2753 /*
2754 * Wake up any threads in rdma_init() or rdma_fini().
d2fe99e8
KS		 2755 * However, this is not needed if com state is just
2756 * MPA_REQ_SENT
2f5b48c3 2757 */
d2fe99e8
KS		 2758 if (ep->com.state != MPA_REQ_SENT)
2759 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2f5b48c3
SW		 2760
2761 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2762 switch (ep->com.state) {
2763 case CONNECTING:
9dec900c 2764 c4iw_put_ep(&ep->parent_ep->com);
cfdda9d7
SW		 2765 break;
2766 case MPA_REQ_WAIT:
b33bd0cb 2767 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2768 break;
2769 case MPA_REQ_SENT:
b33bd0cb 2770 (void)stop_ep_timer(ep);
fe7e0a4d 2771 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
d2fe99e8
KS		 2772 connect_reply_upcall(ep, -ECONNRESET);
2773 else {
2774 /*
2775 * we just don't send notification upwards because we
2776 * want to retry with mpa_v1 without upper layers even
2777 * knowing it.
2778 *
2779 * do some housekeeping so as to re-initiate the
2780 * connection
2781 */
a9a42886
JP		 2782 pr_debug("%s: mpa_rev=%d. Retrying with mpav1\n",
2783 __func__, mpa_rev);
d2fe99e8
KS		 2784 ep->retry_with_mpa_v1 = 1;
2785 }
cfdda9d7
SW		 2786 break;
2787 case MPA_REP_SENT:
cfdda9d7
SW		 2788 break;
2789 case MPA_REQ_RCVD:
cfdda9d7
SW		 2790 break;
2791 case MORIBUND:
2792 case CLOSING:
ca5a2202 2793 stop_ep_timer(ep);
cfdda9d7
SW		 2794 /*FALLTHROUGH*/
2795 case FPDU_MODE:
2796 if (ep->com.cm_id && ep->com.qp) {
2797 attrs.next_state = C4IW_QP_STATE_ERROR;
2798 ret = c4iw_modify_qp(ep->com.qp->rhp,
2799 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2800 &attrs, 1);
2801 if (ret)
700456bd 2802 pr_err("%s - qp <- error failed!\n", __func__);
cfdda9d7
SW		 2803 }
2804 peer_abort_upcall(ep);
2805 break;
2806 case ABORTING:
2807 break;
2808 case DEAD:
a9a42886 2809 pr_debug("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2f5b48c3 2810 mutex_unlock(&ep->com.mutex);
944661dd 2811 goto deref_ep;
cfdda9d7
SW		 2812 default:
2813 BUG_ON(1);
2814 break;
2815 }
2816 dst_confirm(ep->dst);
2817 if (ep->com.state != ABORTING) {
2818 __state_set(&ep->com, DEAD);
d2fe99e8
KS		 2819 /* we don't release if we want to retry with mpa_v1 */
2820 if (!ep->retry_with_mpa_v1)
2821 release = 1;
cfdda9d7 2822 }
2f5b48c3 2823 mutex_unlock(&ep->com.mutex);
cfdda9d7 2824
4a740838
H		 2825 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2826 if (WARN_ON(!rpl_skb)) {
cfdda9d7
SW		 2827 release = 1;
2828 goto out;
2829 }
052f4731
VP		 2830
2831 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2832
cfdda9d7
SW		 2833 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2834out:
cfdda9d7
SW		 2835 if (release)
2836 release_ep_resources(ep);
fe7e0a4d 2837 else if (ep->retry_with_mpa_v1) {
84cc6ac6
H		 2838 if (ep->com.remote_addr.ss_family == AF_INET6) {
2839 struct sockaddr_in6 *sin6 =
2840 (struct sockaddr_in6 *)
170003c8 2841 &ep->com.local_addr;
84cc6ac6
H		 2842 cxgb4_clip_release(
2843 ep->com.dev->rdev.lldi.ports[0],
2844 (const u32 *)&sin6->sin6_addr.s6_addr,
2845 1);
2846 }
fe7e0a4d 2847 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
1dec4cec
GG		 2848 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2849 ep->com.local_addr.ss_family);
d2fe99e8
KS		 2850 dst_release(ep->dst);
2851 cxgb4_l2t_release(ep->l2t);
2852 c4iw_reconnect(ep);
2853 }
2854
944661dd
H		 2855deref_ep:
2856 c4iw_put_ep(&ep->com);
2857 /* Dereferencing ep, referenced in peer_abort_intr() */
2858 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2859 return 0;
2860}
2861
2862static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2863{
2864 struct c4iw_ep *ep;
2865 struct c4iw_qp_attributes attrs;
2866 struct cpl_close_con_rpl *rpl = cplhdr(skb);
cfdda9d7 2867 int release = 0;
cfdda9d7 2868 unsigned int tid = GET_TID(rpl);
cfdda9d7 2869
944661dd
H		 2870 ep = get_ep_from_tid(dev, tid);
2871 if (!ep)
2872 return 0;
cfdda9d7 2873
a9a42886 2874 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 2875 BUG_ON(!ep);
2876
2877 /* The cm_id may be null if we failed to connect */
2f5b48c3 2878 mutex_lock(&ep->com.mutex);
9ca6f7cf 2879 set_bit(CLOSE_CON_RPL, &ep->com.history);
cfdda9d7
SW		 2880 switch (ep->com.state) {
2881 case CLOSING:
2882 __state_set(&ep->com, MORIBUND);
2883 break;
2884 case MORIBUND:
b33bd0cb 2885 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2886 if ((ep->com.cm_id) && (ep->com.qp)) {
2887 attrs.next_state = C4IW_QP_STATE_IDLE;
2888 c4iw_modify_qp(ep->com.qp->rhp,
2889 ep->com.qp,
2890 C4IW_QP_ATTR_NEXT_STATE,
2891 &attrs, 1);
2892 }
be13b2df 2893 close_complete_upcall(ep, 0);
cfdda9d7
SW		 2894 __state_set(&ep->com, DEAD);
2895 release = 1;
2896 break;
2897 case ABORTING:
2898 case DEAD:
2899 break;
2900 default:
2901 BUG_ON(1);
2902 break;
2903 }
2f5b48c3 2904 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2905 if (release)
2906 release_ep_resources(ep);
944661dd 2907 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2908 return 0;
2909}
2910
2911static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2912{
0e42c1f4 2913 struct cpl_rdma_terminate *rpl = cplhdr(skb);
0e42c1f4
SW		 2914 unsigned int tid = GET_TID(rpl);
2915 struct c4iw_ep *ep;
2916 struct c4iw_qp_attributes attrs;
cfdda9d7 2917
944661dd 2918 ep = get_ep_from_tid(dev, tid);
0e42c1f4 2919 BUG_ON(!ep);
cfdda9d7 2920
30c95c2d 2921 if (ep && ep->com.qp) {
700456bd
JP		 2922 pr_warn("TERM received tid %u qpid %u\n",
2923 tid, ep->com.qp->wq.sq.qid);
0e42c1f4
SW		 2924 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2925 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2926 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2927 } else
700456bd 2928 pr_warn("TERM received tid %u no ep/qp\n", tid);
944661dd 2929 c4iw_put_ep(&ep->com);
cfdda9d7 2930
cfdda9d7
SW		 2931 return 0;
2932}
2933
2934/*
2935 * Upcall from the adapter indicating data has been transmitted.
2936 * For us its just the single MPA request or reply. We can now free
2937 * the skb holding the mpa message.
2938 */
2939static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2940{
2941 struct c4iw_ep *ep;
2942 struct cpl_fw4_ack *hdr = cplhdr(skb);
2943 u8 credits = hdr->credits;
2944 unsigned int tid = GET_TID(hdr);
cfdda9d7
SW		 2945
2946
944661dd
H		 2947 ep = get_ep_from_tid(dev, tid);
2948 if (!ep)
2949 return 0;
a9a42886
JP		 2950 pr_debug("%s ep %p tid %u credits %u\n",
2951 __func__, ep, ep->hwtid, credits);
cfdda9d7 2952 if (credits == 0) {
a9a42886
JP		 2953 pr_debug("%s 0 credit ack ep %p tid %u state %u\n",
2954 __func__, ep, ep->hwtid, state_read(&ep->com));
944661dd 2955 goto out;
cfdda9d7
SW		 2956 }
2957
2958 dst_confirm(ep->dst);
2959 if (ep->mpa_skb) {
a9a42886
JP		 2960 pr_debug("%s last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
2961 __func__, ep, ep->hwtid,
2962 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
12eb5137 2963 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2964 kfree_skb(ep->mpa_skb);
2965 ep->mpa_skb = NULL;
e4b76a2a
H		 2966 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2967 stop_ep_timer(ep);
2968 mutex_unlock(&ep->com.mutex);
cfdda9d7 2969 }
944661dd
H		 2970out:
2971 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2972 return 0;
2973}
2974
cfdda9d7
SW		 2975int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2976{
bce2841f 2977 int abort;
cfdda9d7 2978 struct c4iw_ep *ep = to_ep(cm_id);
bce2841f 2979
a9a42886 2980 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7 2981
a7db89eb 2982 mutex_lock(&ep->com.mutex);
e8667a9b 2983 if (ep->com.state != MPA_REQ_RCVD) {
a7db89eb 2984 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2985 c4iw_put_ep(&ep->com);
2986 return -ECONNRESET;
2987 }
793dad94 2988 set_bit(ULP_REJECT, &ep->com.history);
cfdda9d7 2989 if (mpa_rev == 0)
bce2841f
H		 2990 abort = 1;
2991 else
2992 abort = send_mpa_reject(ep, pdata, pdata_len);
a7db89eb 2993 mutex_unlock(&ep->com.mutex);
bce2841f
H		 2994
2995 stop_ep_timer(ep);
2996 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
cfdda9d7
SW		 2997 c4iw_put_ep(&ep->com);
2998 return 0;
2999}
3000
3001int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3002{
3003 int err;
3004 struct c4iw_qp_attributes attrs;
3005 enum c4iw_qp_attr_mask mask;
3006 struct c4iw_ep *ep = to_ep(cm_id);
3007 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3008 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
eaf4c6d4 3009 int abort = 0;
cfdda9d7 3010
a9a42886 3011 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
a7db89eb
SW		 3012
3013 mutex_lock(&ep->com.mutex);
e8667a9b 3014 if (ep->com.state != MPA_REQ_RCVD) {
cfdda9d7 3015 err = -ECONNRESET;
eaf4c6d4 3016 goto err_out;
cfdda9d7
SW		 3017 }
3018
cfdda9d7
SW		 3019 BUG_ON(!qp);
3020
793dad94 3021 set_bit(ULP_ACCEPT, &ep->com.history);
4c2c5763
HS		 3022 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3023 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
cfdda9d7 3024 err = -EINVAL;
eaf4c6d4 3025 goto err_abort;
cfdda9d7
SW		 3026 }
3027
d2fe99e8
KS		 3028 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3029 if (conn_param->ord > ep->ird) {
4c2c5763 3030 if (RELAXED_IRD_NEGOTIATION) {
30b03b15 3031 conn_param->ord = ep->ird;
4c2c5763
HS		 3032 } else {
3033 ep->ird = conn_param->ird;
3034 ep->ord = conn_param->ord;
3035 send_mpa_reject(ep, conn_param->private_data,
3036 conn_param->private_data_len);
4c2c5763 3037 err = -ENOMEM;
eaf4c6d4 3038 goto err_abort;
4c2c5763 3039 }
d2fe99e8 3040 }
4c2c5763
HS		 3041 if (conn_param->ird < ep->ord) {
3042 if (RELAXED_IRD_NEGOTIATION &&
3043 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3044 conn_param->ird = ep->ord;
3045 } else {
d2fe99e8 3046 err = -ENOMEM;
eaf4c6d4 3047 goto err_abort;
d2fe99e8
KS		 3048 }
3049 }
d2fe99e8 3050 }
cfdda9d7
SW		 3051 ep->ird = conn_param->ird;
3052 ep->ord = conn_param->ord;
3053
4c2c5763 3054 if (ep->mpa_attr.version == 1) {
d2fe99e8
KS		 3055 if (peer2peer && ep->ird == 0)
3056 ep->ird = 1;
4c2c5763
HS		 3057 } else {
3058 if (peer2peer &&
3059 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
f57b780c 3060 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
4c2c5763
HS		 3061 ep->ird = 1;
3062 }
cfdda9d7 3063
a9a42886 3064 pr_debug("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
cfdda9d7 3065
d2fe99e8 3066 ep->com.cm_id = cm_id;
9ca6f7cf 3067 ref_cm_id(&ep->com);
d2fe99e8 3068 ep->com.qp = qp;
325abead 3069 ref_qp(ep);
d2fe99e8 3070
cfdda9d7
SW		 3071 /* bind QP to EP and move to RTS */
3072 attrs.mpa_attr = ep->mpa_attr;
3073 attrs.max_ird = ep->ird;
3074 attrs.max_ord = ep->ord;
3075 attrs.llp_stream_handle = ep;
3076 attrs.next_state = C4IW_QP_STATE_RTS;
3077
3078 /* bind QP and TID with INIT_WR */
3079 mask = C4IW_QP_ATTR_NEXT_STATE |
3080 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3081 C4IW_QP_ATTR_MPA_ATTR |
3082 C4IW_QP_ATTR_MAX_IRD |
3083 C4IW_QP_ATTR_MAX_ORD;
3084
3085 err = c4iw_modify_qp(ep->com.qp->rhp,
3086 ep->com.qp, mask, &attrs, 1);
3087 if (err)
eaf4c6d4 3088 goto err_deref_cm_id;
e4b76a2a
H		 3089
3090 set_bit(STOP_MPA_TIMER, &ep->com.flags);
cfdda9d7
SW		 3091 err = send_mpa_reply(ep, conn_param->private_data,
3092 conn_param->private_data_len);
3093 if (err)
eaf4c6d4 3094 goto err_deref_cm_id;
cfdda9d7 3095
a7db89eb 3096 __state_set(&ep->com, FPDU_MODE);
cfdda9d7 3097 established_upcall(ep);
a7db89eb 3098 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 3099 c4iw_put_ep(&ep->com);
3100 return 0;
eaf4c6d4 3101err_deref_cm_id:
9ca6f7cf 3102 deref_cm_id(&ep->com);
eaf4c6d4
H		 3103err_abort:
3104 abort = 1;
3105err_out:
a7db89eb 3106 mutex_unlock(&ep->com.mutex);
eaf4c6d4
H		 3107 if (abort)
3108 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
cfdda9d7
SW		 3109 c4iw_put_ep(&ep->com);
3110 return err;
3111}
3112
830662f6
VP		 3113static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3114{
3115 struct in_device *ind;
3116 int found = 0;
170003c8
SW		 3117 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3118 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
830662f6
VP		 3119
3120 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3121 if (!ind)
3122 return -EADDRNOTAVAIL;
3123 for_primary_ifa(ind) {
3124 laddr->sin_addr.s_addr = ifa->ifa_address;
3125 raddr->sin_addr.s_addr = ifa->ifa_address;
3126 found = 1;
3127 break;
3128 }
3129 endfor_ifa(ind);
3130 in_dev_put(ind);
3131 return found ? 0 : -EADDRNOTAVAIL;
3132}
3133
3134static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3135 unsigned char banned_flags)
3136{
3137 struct inet6_dev *idev;
3138 int err = -EADDRNOTAVAIL;
3139
3140 rcu_read_lock();
3141 idev = __in6_dev_get(dev);
3142 if (idev != NULL) {
3143 struct inet6_ifaddr *ifp;
3144
3145 read_lock_bh(&idev->lock);
3146 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3147 if (ifp->scope == IFA_LINK &&
3148 !(ifp->flags & banned_flags)) {
3149 memcpy(addr, &ifp->addr, 16);
3150 err = 0;
3151 break;
3152 }
3153 }
3154 read_unlock_bh(&idev->lock);
3155 }
3156 rcu_read_unlock();
3157 return err;
3158}
3159
3160static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3161{
3162 struct in6_addr uninitialized_var(addr);
170003c8
SW		 3163 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3164 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
830662f6 3165
54b9a96f 3166 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
830662f6
VP		 3167 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3168 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3169 return 0;
3170 }
3171 return -EADDRNOTAVAIL;
3172}
3173
cfdda9d7
SW		 3174int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3175{
cfdda9d7
SW		 3176 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3177 struct c4iw_ep *ep;
3786cf18 3178 int err = 0;
9eccfe10
SW		 3179 struct sockaddr_in *laddr;
3180 struct sockaddr_in *raddr;
3181 struct sockaddr_in6 *laddr6;
3182 struct sockaddr_in6 *raddr6;
830662f6
VP		 3183 __u8 *ra;
3184 int iptype;
cfdda9d7 3185
4c2c5763
HS		 3186 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3187 (conn_param->ird > cur_max_read_depth(dev))) {
be4c9bad
RD		 3188 err = -EINVAL;
3189 goto out;
3190 }
cfdda9d7
SW		 3191 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3192 if (!ep) {
700456bd 3193 pr_err("%s - cannot alloc ep\n", __func__);
cfdda9d7
SW		 3194 err = -ENOMEM;
3195 goto out;
3196 }
4a740838
H		 3197
3198 skb_queue_head_init(&ep->com.ep_skb_list);
3199 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3200 err = -ENOMEM;
3201 goto fail1;
3202 }
3203
cfdda9d7
SW		 3204 init_timer(&ep->timer);
3205 ep->plen = conn_param->private_data_len;
3206 if (ep->plen)
3207 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3208 conn_param->private_data, ep->plen);
3209 ep->ird = conn_param->ird;
3210 ep->ord = conn_param->ord;
3211
3212 if (peer2peer && ep->ord == 0)
3213 ep->ord = 1;
3214
cfdda9d7 3215 ep->com.cm_id = cm_id;
9ca6f7cf
H		 3216 ref_cm_id(&ep->com);
3217 ep->com.dev = dev;
cfdda9d7 3218 ep->com.qp = get_qhp(dev, conn_param->qpn);
830662f6 3219 if (!ep->com.qp) {
a9a42886 3220 pr_debug("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
830662f6 3221 err = -EINVAL;
4a740838 3222 goto fail2;
830662f6 3223 }
325abead 3224 ref_qp(ep);
a9a42886
JP		 3225 pr_debug("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3226 ep->com.qp, cm_id);
cfdda9d7
SW		 3227
3228 /*
3229 * Allocate an active TID to initiate a TCP connection.
3230 */
3231 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3232 if (ep->atid == -1) {
700456bd 3233 pr_err("%s - cannot alloc atid\n", __func__);
cfdda9d7 3234 err = -ENOMEM;
4a740838 3235 goto fail2;
cfdda9d7 3236 }
793dad94 3237 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
cfdda9d7 3238
170003c8 3239 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
9eccfe10 3240 sizeof(ep->com.local_addr));
170003c8 3241 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
9eccfe10
SW		 3242 sizeof(ep->com.remote_addr));
3243
170003c8
SW		 3244 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3245 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3246 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3247 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
9eccfe10 3248
170003c8 3249 if (cm_id->m_remote_addr.ss_family == AF_INET) {
830662f6
VP		 3250 iptype = 4;
3251 ra = (__u8 *)&raddr->sin_addr;
cfdda9d7 3252
830662f6
VP		 3253 /*
3254 * Handle loopback requests to INADDR_ANY.
3255 */
ba987e51 3256 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
830662f6
VP		 3257 err = pick_local_ipaddrs(dev, cm_id);
3258 if (err)
4a740838 3259 goto fail2;
830662f6
VP		 3260 }
3261
3262 /* find a route */
a9a42886
JP		 3263 pr_debug("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3264 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3265 ra, ntohs(raddr->sin_port));
804c2f3e
VP		 3266 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3267 laddr->sin_addr.s_addr,
3268 raddr->sin_addr.s_addr,
3269 laddr->sin_port,
3270 raddr->sin_port, cm_id->tos);
830662f6
VP		 3271 } else {
3272 iptype = 6;
3273 ra = (__u8 *)&raddr6->sin6_addr;
3274
3275 /*
3276 * Handle loopback requests to INADDR_ANY.
3277 */
3278 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3279 err = pick_local_ip6addrs(dev, cm_id);
3280 if (err)
4a740838 3281 goto fail2;
830662f6
VP		 3282 }
3283
3284 /* find a route */
a9a42886
JP		 3285 pr_debug("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3286 __func__, laddr6->sin6_addr.s6_addr,
3287 ntohs(laddr6->sin6_port),
3288 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
95554761
VP		 3289 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3290 laddr6->sin6_addr.s6_addr,
3291 raddr6->sin6_addr.s6_addr,
3292 laddr6->sin6_port,
3293 raddr6->sin6_port, 0,
3294 raddr6->sin6_scope_id);
830662f6
VP		 3295 }
3296 if (!ep->dst) {
700456bd 3297 pr_err("%s - cannot find route\n", __func__);
cfdda9d7 3298 err = -EHOSTUNREACH;
4a740838 3299 goto fail3;
cfdda9d7 3300 }
cfdda9d7 3301
963cab50 3302 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
ac8e4c69 3303 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3786cf18 3304 if (err) {
700456bd 3305 pr_err("%s - cannot alloc l2e\n", __func__);
4a740838 3306 goto fail4;
cfdda9d7
SW		 3307 }
3308
a9a42886
JP		 3309 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3310 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3311 ep->l2t->idx);
cfdda9d7
SW		 3312
3313 state_set(&ep->com, CONNECTING);
ac8e4c69 3314 ep->tos = cm_id->tos;
cfdda9d7
SW		 3315
3316 /* send connect request to rnic */
3317 err = send_connect(ep);
3318 if (!err)
3319 goto out;
3320
3321 cxgb4_l2t_release(ep->l2t);
4a740838 3322fail4:
9eccfe10 3323 dst_release(ep->dst);
4a740838 3324fail3:
793dad94 3325 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
cfdda9d7 3326 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
4a740838
H		 3327fail2:
3328 skb_queue_purge(&ep->com.ep_skb_list);
9ca6f7cf 3329 deref_cm_id(&ep->com);
4a740838 3330fail1:
cfdda9d7
SW		 3331 c4iw_put_ep(&ep->com);
3332out:
3333 return err;
3334}
3335
830662f6
VP		 3336static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3337{
3338 int err;
9eccfe10 3339 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
170003c8 3340 &ep->com.local_addr;
830662f6 3341
28de1f74
H		 3342 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3343 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3344 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3345 if (err)
3346 return err;
3347 }
830662f6
VP		 3348 c4iw_init_wr_wait(&ep->com.wr_wait);
3349 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3350 ep->stid, &sin6->sin6_addr,
3351 sin6->sin6_port,
3352 ep->com.dev->rdev.lldi.rxq_ids[0]);
3353 if (!err)
3354 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3355 &ep->com.wr_wait,
3356 0, 0, __func__);
e6b11163
H		 3357 else if (err > 0)
3358 err = net_xmit_errno(err);
28de1f74
H		 3359 if (err) {
3360 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3361 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
830662f6
VP		 3362 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3363 err, ep->stid,
3364 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
28de1f74 3365 }
830662f6
VP		 3366 return err;
3367}
3368
3369static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3370{
3371 int err;
9eccfe10 3372 struct sockaddr_in *sin = (struct sockaddr_in *)
170003c8 3373 &ep->com.local_addr;
830662f6
VP		 3374
3375 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3376 do {
3377 err = cxgb4_create_server_filter(
3378 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3379 sin->sin_addr.s_addr, sin->sin_port, 0,
3380 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3381 if (err == -EBUSY) {
99718e59
H		 3382 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3383 err = -EIO;
3384 break;
3385 }
830662f6
VP		 3386 set_current_state(TASK_UNINTERRUPTIBLE);
3387 schedule_timeout(usecs_to_jiffies(100));
3388 }
3389 } while (err == -EBUSY);
3390 } else {
3391 c4iw_init_wr_wait(&ep->com.wr_wait);
3392 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3393 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3394 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3395 if (!err)
3396 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3397 &ep->com.wr_wait,
3398 0, 0, __func__);
e6b11163
H		 3399 else if (err > 0)
3400 err = net_xmit_errno(err);
830662f6
VP		 3401 }
3402 if (err)
3403 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3404 , err, ep->stid,
3405 &sin->sin_addr, ntohs(sin->sin_port));
3406 return err;
3407}
3408
cfdda9d7
SW		 3409int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3410{
3411 int err = 0;
3412 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3413 struct c4iw_listen_ep *ep;
3414
cfdda9d7
SW		 3415 might_sleep();
3416
3417 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3418 if (!ep) {
700456bd 3419 pr_err("%s - cannot alloc ep\n", __func__);
cfdda9d7
SW		 3420 err = -ENOMEM;
3421 goto fail1;
3422 }
4a740838 3423 skb_queue_head_init(&ep->com.ep_skb_list);
a9a42886 3424 pr_debug("%s ep %p\n", __func__, ep);
cfdda9d7 3425 ep->com.cm_id = cm_id;
9ca6f7cf 3426 ref_cm_id(&ep->com);
cfdda9d7
SW		 3427 ep->com.dev = dev;
3428 ep->backlog = backlog;
170003c8 3429 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
24d44a39 3430 sizeof(ep->com.local_addr));
cfdda9d7
SW		 3431
3432 /*
3433 * Allocate a server TID.
3434 */
8c044690
KS		 3435 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3436 ep->com.local_addr.ss_family == AF_INET)
830662f6 3437 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
170003c8 3438 cm_id->m_local_addr.ss_family, ep);
1cab775c 3439 else
830662f6 3440 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
170003c8 3441 cm_id->m_local_addr.ss_family, ep);
1cab775c 3442
cfdda9d7 3443 if (ep->stid == -1) {
700456bd 3444 pr_err("%s - cannot alloc stid\n", __func__);
cfdda9d7
SW		 3445 err = -ENOMEM;
3446 goto fail2;
3447 }
793dad94 3448 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
9eccfe10 3449
170003c8
SW		 3450 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3451 sizeof(ep->com.local_addr));
9eccfe10 3452
cfdda9d7 3453 state_set(&ep->com, LISTEN);
830662f6
VP		 3454 if (ep->com.local_addr.ss_family == AF_INET)
3455 err = create_server4(dev, ep);
3456 else
3457 err = create_server6(dev, ep);
cfdda9d7
SW		 3458 if (!err) {
3459 cm_id->provider_data = ep;
3460 goto out;
3461 }
9eccfe10 3462
830662f6
VP		 3463 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3464 ep->com.local_addr.ss_family);
cfdda9d7 3465fail2:
9ca6f7cf 3466 deref_cm_id(&ep->com);
cfdda9d7
SW		 3467 c4iw_put_ep(&ep->com);
3468fail1:
3469out:
3470 return err;
3471}
3472
3473int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3474{
3475 int err;
3476 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3477
a9a42886 3478 pr_debug("%s ep %p\n", __func__, ep);
cfdda9d7
SW		 3479
3480 might_sleep();
3481 state_set(&ep->com, DEAD);
830662f6
VP		 3482 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3483 ep->com.local_addr.ss_family == AF_INET) {
1cab775c
VP		 3484 err = cxgb4_remove_server_filter(
3485 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3486 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3487 } else {
84cc6ac6 3488 struct sockaddr_in6 *sin6;
1cab775c 3489 c4iw_init_wr_wait(&ep->com.wr_wait);
830662f6
VP		 3490 err = cxgb4_remove_server(
3491 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3492 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
1cab775c
VP		 3493 if (err)
3494 goto done;
3495 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3496 0, 0, __func__);
170003c8 3497 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 3498 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3499 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
1cab775c 3500 }
793dad94 3501 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
830662f6
VP		 3502 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3503 ep->com.local_addr.ss_family);
cfdda9d7 3504done:
9ca6f7cf 3505 deref_cm_id(&ep->com);
cfdda9d7
SW		 3506 c4iw_put_ep(&ep->com);
3507 return err;
3508}
3509
3510int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3511{
3512 int ret = 0;
cfdda9d7
SW		 3513 int close = 0;
3514 int fatal = 0;
3515 struct c4iw_rdev *rdev;
cfdda9d7 3516
2f5b48c3 3517 mutex_lock(&ep->com.mutex);
cfdda9d7 3518
a9a42886
JP		 3519 pr_debug("%s ep %p state %s, abrupt %d\n", __func__, ep,
3520 states[ep->com.state], abrupt);
cfdda9d7 3521
6e410d8f
H		 3522 /*
3523 * Ref the ep here in case we have fatal errors causing the
3524 * ep to be released and freed.
3525 */
3526 c4iw_get_ep(&ep->com);
3527
cfdda9d7
SW		 3528 rdev = &ep->com.dev->rdev;
3529 if (c4iw_fatal_error(rdev)) {
3530 fatal = 1;
be13b2df 3531 close_complete_upcall(ep, -EIO);
cfdda9d7
SW		 3532 ep->com.state = DEAD;
3533 }
3534 switch (ep->com.state) {
3535 case MPA_REQ_WAIT:
3536 case MPA_REQ_SENT:
3537 case MPA_REQ_RCVD:
3538 case MPA_REP_SENT:
3539 case FPDU_MODE:
4a740838 3540 case CONNECTING:
cfdda9d7
SW		 3541 close = 1;
3542 if (abrupt)
3543 ep->com.state = ABORTING;
3544 else {
3545 ep->com.state = CLOSING;
12eb5137
SW		 3546
3547 /*
3548 * if we close before we see the fw4_ack() then we fix
3549 * up the timer state since we're reusing it.
3550 */
3551 if (ep->mpa_skb &&
3552 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3553 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3554 stop_ep_timer(ep);
3555 }
ca5a2202 3556 start_ep_timer(ep);
cfdda9d7
SW		 3557 }
3558 set_bit(CLOSE_SENT, &ep->com.flags);
3559 break;
3560 case CLOSING:
3561 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3562 close = 1;
3563 if (abrupt) {
b33bd0cb 3564 (void)stop_ep_timer(ep);
cfdda9d7
SW		 3565 ep->com.state = ABORTING;
3566 } else
3567 ep->com.state = MORIBUND;
3568 }
3569 break;
3570 case MORIBUND:
3571 case ABORTING:
3572 case DEAD:
a9a42886
JP		 3573 pr_debug("%s ignoring disconnect ep %p state %u\n",
3574 __func__, ep, ep->com.state);
cfdda9d7
SW		 3575 break;
3576 default:
3577 BUG();
3578 break;
3579 }
3580
cfdda9d7 3581 if (close) {
8da7e7a5 3582 if (abrupt) {
793dad94 3583 set_bit(EP_DISC_ABORT, &ep->com.history);
be13b2df 3584 close_complete_upcall(ep, -ECONNRESET);
4a740838 3585 ret = send_abort(ep);
793dad94
VP		 3586 } else {
3587 set_bit(EP_DISC_CLOSE, &ep->com.history);
4a740838 3588 ret = send_halfclose(ep);
793dad94 3589 }
88bc230d 3590 if (ret) {
9ca6f7cf 3591 set_bit(EP_DISC_FAIL, &ep->com.history);
88bc230d
H		 3592 if (!abrupt) {
3593 stop_ep_timer(ep);
3594 close_complete_upcall(ep, -EIO);
3595 }
c00dcbaf
H		 3596 if (ep->com.qp) {
3597 struct c4iw_qp_attributes attrs;
3598
3599 attrs.next_state = C4IW_QP_STATE_ERROR;
3600 ret = c4iw_modify_qp(ep->com.qp->rhp,
3601 ep->com.qp,
3602 C4IW_QP_ATTR_NEXT_STATE,
3603 &attrs, 1);
3604 if (ret)
700456bd 3605 pr_err("%s - qp <- error failed!\n",
c00dcbaf
H		 3606 __func__);
3607 }
cfdda9d7 3608 fatal = 1;
88bc230d 3609 }
cfdda9d7 3610 }
8da7e7a5 3611 mutex_unlock(&ep->com.mutex);
6e410d8f 3612 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 3613 if (fatal)
3614 release_ep_resources(ep);
3615 return ret;
3616}
3617
1cab775c
VP		 3618static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3619 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3620{
3621 struct c4iw_ep *ep;
793dad94 3622 int atid = be32_to_cpu(req->tid);
1cab775c 3623
ef5d6355
VP		 3624 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3625 (__force u32) req->tid);
1cab775c
VP		 3626 if (!ep)
3627 return;
3628
3629 switch (req->retval) {
3630 case FW_ENOMEM:
793dad94
VP		 3631 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3632 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3633 send_fw_act_open_req(ep, atid);
3634 return;
3635 }
1cab775c 3636 case FW_EADDRINUSE:
793dad94
VP		 3637 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3638 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3639 send_fw_act_open_req(ep, atid);
3640 return;
3641 }
1cab775c
VP		 3642 break;
3643 default:
3644 pr_info("%s unexpected ofld conn wr retval %d\n",
3645 __func__, req->retval);
3646 break;
3647 }
793dad94
VP		 3648 pr_err("active ofld_connect_wr failure %d atid %d\n",
3649 req->retval, atid);
3650 mutex_lock(&dev->rdev.stats.lock);
3651 dev->rdev.stats.act_ofld_conn_fails++;
3652 mutex_unlock(&dev->rdev.stats.lock);
1cab775c 3653 connect_reply_upcall(ep, status2errno(req->retval));
793dad94 3654 state_set(&ep->com, DEAD);
84cc6ac6
H		 3655 if (ep->com.remote_addr.ss_family == AF_INET6) {
3656 struct sockaddr_in6 *sin6 =
170003c8 3657 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 3658 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3659 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3660 }
793dad94
VP		 3661 remove_handle(dev, &dev->atid_idr, atid);
3662 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3663 dst_release(ep->dst);
3664 cxgb4_l2t_release(ep->l2t);
3665 c4iw_put_ep(&ep->com);
1cab775c
VP		 3666}
3667
3668static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3669 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3670{
3671 struct sk_buff *rpl_skb;
3672 struct cpl_pass_accept_req *cpl;
3673 int ret;
3674
710a3110 3675 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
1cab775c
VP		 3676 BUG_ON(!rpl_skb);
3677 if (req->retval) {
a9a42886 3678 pr_debug("%s passive open failure %d\n", __func__, req->retval);
793dad94
VP		 3679 mutex_lock(&dev->rdev.stats.lock);
3680 dev->rdev.stats.pas_ofld_conn_fails++;
3681 mutex_unlock(&dev->rdev.stats.lock);
1cab775c
VP		 3682 kfree_skb(rpl_skb);
3683 } else {
3684 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3685 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
ef5d6355
VP		 3686 (__force u32) htonl(
3687 (__force u32) req->tid)));
1cab775c
VP		 3688 ret = pass_accept_req(dev, rpl_skb);
3689 if (!ret)
3690 kfree_skb(rpl_skb);
3691 }
3692 return;
3693}
3694
3695static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2f5b48c3
SW		 3696{
3697 struct cpl_fw6_msg *rpl = cplhdr(skb);
1cab775c
VP		 3698 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3699
3700 switch (rpl->type) {
3701 case FW6_TYPE_CQE:
3702 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3703 break;
3704 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3705 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3706 switch (req->t_state) {
3707 case TCP_SYN_SENT:
3708 active_ofld_conn_reply(dev, skb, req);
3709 break;
3710 case TCP_SYN_RECV:
3711 passive_ofld_conn_reply(dev, skb, req);
3712 break;
3713 default:
3714 pr_err("%s unexpected ofld conn wr state %d\n",
3715 __func__, req->t_state);
3716 break;
3717 }
3718 break;
3719 }
3720 return 0;
3721}
3722
3723static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3724{
963cab50
H		 3725 __be32 l2info;
3726 __be16 hdr_len, vlantag, len;
3727 u16 eth_hdr_len;
3728 int tcp_hdr_len, ip_hdr_len;
1cab775c
VP		 3729 u8 intf;
3730 struct cpl_rx_pkt *cpl = cplhdr(skb);
3731 struct cpl_pass_accept_req *req;
3732 struct tcp_options_received tmp_opt;
f079af7a 3733 struct c4iw_dev *dev;
963cab50 3734 enum chip_type type;
1cab775c 3735
f079af7a 3736 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
1cab775c 3737 /* Store values from cpl_rx_pkt in temporary location. */
963cab50
H		 3738 vlantag = cpl->vlan;
3739 len = cpl->len;
3740 l2info = cpl->l2info;
3741 hdr_len = cpl->hdr_len;
1cab775c
VP		 3742 intf = cpl->iff;
3743
3744 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3745
3746 /*
3747 * We need to parse the TCP options from SYN packet.
3748 * to generate cpl_pass_accept_req.
3749 */
3750 memset(&tmp_opt, 0, sizeof(tmp_opt));
3751 tcp_clear_options(&tmp_opt);
eed29f17 3752 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
1cab775c 3753
d58ff351 3754 req = __skb_push(skb, sizeof(*req));
1cab775c 3755 memset(req, 0, sizeof(*req));
cf7fe64a
HS		 3756 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3757 SYN_MAC_IDX_V(RX_MACIDX_G(
963cab50 3758 be32_to_cpu(l2info))) |
cf7fe64a 3759 SYN_XACT_MATCH_F);
963cab50
H		 3760 type = dev->rdev.lldi.adapter_type;
3761 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3762 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3763 req->hdr_len =
3764 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3765 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3766 eth_hdr_len = is_t4(type) ?
3767 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3768 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3769 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3770 IP_HDR_LEN_V(ip_hdr_len) |
3771 ETH_HDR_LEN_V(eth_hdr_len));
3772 } else { /* T6 and later */
3773 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3774 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3775 T6_IP_HDR_LEN_V(ip_hdr_len) |
3776 T6_ETH_HDR_LEN_V(eth_hdr_len));
3777 }
3778 req->vlan = vlantag;
3779 req->len = len;
6c53e938
HS		 3780 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3781 PASS_OPEN_TOS_V(tos));
1cab775c
VP		 3782 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3783 if (tmp_opt.wscale_ok)
3784 req->tcpopt.wsf = tmp_opt.snd_wscale;
3785 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3786 if (tmp_opt.sack_ok)
3787 req->tcpopt.sack = 1;
3788 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3789 return;
3790}
3791
3792static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3793 __be32 laddr, __be16 lport,
3794 __be32 raddr, __be16 rport,
3795 u32 rcv_isn, u32 filter, u16 window,
3796 u32 rss_qid, u8 port_id)
3797{
3798 struct sk_buff *req_skb;
3799 struct fw_ofld_connection_wr *req;
3800 struct cpl_pass_accept_req *cpl = cplhdr(skb);
1ce1d471 3801 int ret;
1cab775c
VP		 3802
3803 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
9ef63f31
PB		 3804 if (!req_skb)
3805 return;
4df864c1 3806 req = __skb_put(req_skb, sizeof(*req));
1cab775c 3807 memset(req, 0, sizeof(*req));
6c53e938 3808 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
e2ac9628 3809 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
77a80e23 3810 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
ef5d6355 3811 req->le.filter = (__force __be32) filter;
1cab775c
VP		 3812 req->le.lport = lport;
3813 req->le.pport = rport;
3814 req->le.u.ipv4.lip = laddr;
3815 req->le.u.ipv4.pip = raddr;
3816 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3817 req->tcb.rcv_adv = htons(window);
3818 req->tcb.t_state_to_astid =
77a80e23
HS		 3819 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3820 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3821 FW_OFLD_CONNECTION_WR_ASTID_V(
6c53e938 3822 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
1cab775c
VP		 3823
3824 /*
3825 * We store the qid in opt2 which will be used by the firmware
3826 * to send us the wr response.
3827 */
d7990b0c 3828 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
1cab775c
VP		 3829
3830 /*
3831 * We initialize the MSS index in TCB to 0xF.
3832 * So that when driver sends cpl_pass_accept_rpl
3833 * TCB picks up the correct value. If this was 0
3834 * TP will ignore any value > 0 for MSS index.
3835 */
d7990b0c 3836 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
6198dd8d 3837 req->cookie = (uintptr_t)skb;
1cab775c
VP		 3838
3839 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
1ce1d471
SW		 3840 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3841 if (ret < 0) {
3842 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3843 ret);
3844 kfree_skb(skb);
3845 kfree_skb(req_skb);
3846 }
1cab775c
VP		 3847}
3848
3849/*
3850 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3851 * messages when a filter is being used instead of server to
3852 * redirect a syn packet. When packets hit filter they are redirected
3853 * to the offload queue and driver tries to establish the connection
3854 * using firmware work request.
3855 */
3856static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3857{
3858 int stid;
3859 unsigned int filter;
3860 struct ethhdr *eh = NULL;
3861 struct vlan_ethhdr *vlan_eh = NULL;
3862 struct iphdr *iph;
3863 struct tcphdr *tcph;
3864 struct rss_header *rss = (void *)skb->data;
3865 struct cpl_rx_pkt *cpl = (void *)skb->data;
3866 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3867 struct l2t_entry *e;
3868 struct dst_entry *dst;
f86fac79 3869 struct c4iw_ep *lep = NULL;
1cab775c
VP		 3870 u16 window;
3871 struct port_info *pi;
3872 struct net_device *pdev;
f079af7a 3873 u16 rss_qid, eth_hdr_len;
1cab775c
VP		 3874 int step;
3875 u32 tx_chan;
3876 struct neighbour *neigh;
3877
3878 /* Drop all non-SYN packets */
bdc590b9 3879 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
1cab775c
VP		 3880 goto reject;
3881
3882 /*
3883 * Drop all packets which did not hit the filter.
3884 * Unlikely to happen.
3885 */
3886 if (!(rss->filter_hit && rss->filter_tid))
3887 goto reject;
3888
3889 /*
3890 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3891 */
a4ea025f 3892 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
1cab775c 3893
f86fac79 3894 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
1cab775c 3895 if (!lep) {
a9a42886
JP		 3896 pr_debug("%s connect request on invalid stid %d\n",
3897 __func__, stid);
1cab775c
VP		 3898 goto reject;
3899 }
3900
963cab50
H		 3901 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3902 case CHELSIO_T4:
3903 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3904 break;
3905 case CHELSIO_T5:
3906 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3907 break;
3908 case CHELSIO_T6:
3909 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3910 break;
3911 default:
3912 pr_err("T%d Chip is not supported\n",
3913 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3914 goto reject;
3915 }
3916
f079af7a 3917 if (eth_hdr_len == ETH_HLEN) {
1cab775c
VP		 3918 eh = (struct ethhdr *)(req + 1);
3919 iph = (struct iphdr *)(eh + 1);
3920 } else {
3921 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3922 iph = (struct iphdr *)(vlan_eh + 1);
3923 skb->vlan_tci = ntohs(cpl->vlan);
3924 }
3925
3926 if (iph->version != 0x4)
3927 goto reject;
3928
3929 tcph = (struct tcphdr *)(iph + 1);
3930 skb_set_network_header(skb, (void *)iph - (void *)rss);
3931 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3932 skb_get(skb);
3933
a9a42886
JP		 3934 pr_debug("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3935 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3936 ntohs(tcph->source), iph->tos);
1cab775c 3937
804c2f3e
VP		 3938 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3939 iph->daddr, iph->saddr, tcph->dest,
3940 tcph->source, iph->tos);
830662f6 3941 if (!dst) {
1cab775c
VP		 3942 pr_err("%s - failed to find dst entry!\n",
3943 __func__);
3944 goto reject;
3945 }
1cab775c
VP		 3946 neigh = dst_neigh_lookup_skb(dst, skb);
3947
aaa0c23c
ZZ		 3948 if (!neigh) {
3949 pr_err("%s - failed to allocate neigh!\n",
3950 __func__);
3951 goto free_dst;
3952 }
3953
1cab775c
VP		 3954 if (neigh->dev->flags & IFF_LOOPBACK) {
3955 pdev = ip_dev_find(&init_net, iph->daddr);
3956 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3957 pdev, 0);
3958 pi = (struct port_info *)netdev_priv(pdev);
3959 tx_chan = cxgb4_port_chan(pdev);
3960 dev_put(pdev);
3961 } else {
830662f6 3962 pdev = get_real_dev(neigh->dev);
1cab775c 3963 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
830662f6
VP		 3964 pdev, 0);
3965 pi = (struct port_info *)netdev_priv(pdev);
3966 tx_chan = cxgb4_port_chan(pdev);
1cab775c 3967 }
ebf00060 3968 neigh_release(neigh);
1cab775c
VP		 3969 if (!e) {
3970 pr_err("%s - failed to allocate l2t entry!\n",
3971 __func__);
3972 goto free_dst;
3973 }
3974
3975 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3976 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
ef5d6355 3977 window = (__force u16) htons((__force u16)tcph->window);
1cab775c
VP		 3978
3979 /* Calcuate filter portion for LE region. */
41b4f86c
KS		 3980 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3981 dev->rdev.lldi.ports[0],
3982 e));
1cab775c
VP		 3983
3984 /*
3985 * Synthesize the cpl_pass_accept_req. We have everything except the
3986 * TID. Once firmware sends a reply with TID we update the TID field
3987 * in cpl and pass it through the regular cpl_pass_accept_req path.
3988 */
3989 build_cpl_pass_accept_req(skb, stid, iph->tos);
3990 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3991 tcph->source, ntohl(tcph->seq), filter, window,
3992 rss_qid, pi->port_id);
3993 cxgb4_l2t_release(e);
3994free_dst:
3995 dst_release(dst);
3996reject:
f86fac79
H		 3997 if (lep)
3998 c4iw_put_ep(&lep->com);
2f5b48c3
SW		 3999 return 0;
4000}
4001
be4c9bad
RD		 4002/*
4003 * These are the real handlers that are called from a
4004 * work queue.
4005 */
9dec900c 4006static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
be4c9bad
RD		 4007 [CPL_ACT_ESTABLISH] = act_establish,
4008 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4009 [CPL_RX_DATA] = rx_data,
4010 [CPL_ABORT_RPL_RSS] = abort_rpl,
4011 [CPL_ABORT_RPL] = abort_rpl,
4012 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4013 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4014 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4015 [CPL_PASS_ESTABLISH] = pass_establish,
4016 [CPL_PEER_CLOSE] = peer_close,
4017 [CPL_ABORT_REQ_RSS] = peer_abort,
4018 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4019 [CPL_RDMA_TERMINATE] = terminate,
2f5b48c3 4020 [CPL_FW4_ACK] = fw4_ack,
1cab775c 4021 [CPL_FW6_MSG] = deferred_fw6_msg,
9dec900c 4022 [CPL_RX_PKT] = rx_pkt,
8d1f1a6b
H		 4023 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4024 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
be4c9bad
RD		 4025};
4026
4027static void process_timeout(struct c4iw_ep *ep)
4028{
4029 struct c4iw_qp_attributes attrs;
4030 int abort = 1;
4031
2f5b48c3 4032 mutex_lock(&ep->com.mutex);
a9a42886
JP		 4033 pr_debug("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4034 ep->com.state);
793dad94 4035 set_bit(TIMEDOUT, &ep->com.history);
be4c9bad
RD		 4036 switch (ep->com.state) {
4037 case MPA_REQ_SENT:
be4c9bad
RD		 4038 connect_reply_upcall(ep, -ETIMEDOUT);
4039 break;
4040 case MPA_REQ_WAIT:
ceb110a8 4041 case MPA_REQ_RCVD:
e4b76a2a 4042 case MPA_REP_SENT:
ceb110a8 4043 case FPDU_MODE:
be4c9bad
RD		 4044 break;
4045 case CLOSING:
4046 case MORIBUND:
4047 if (ep->com.cm_id && ep->com.qp) {
4048 attrs.next_state = C4IW_QP_STATE_ERROR;
4049 c4iw_modify_qp(ep->com.qp->rhp,
4050 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4051 &attrs, 1);
4052 }
be13b2df 4053 close_complete_upcall(ep, -ETIMEDOUT);
be4c9bad 4054 break;
b33bd0cb
SW		 4055 case ABORTING:
4056 case DEAD:
4057
4058 /*
4059 * These states are expected if the ep timed out at the same
4060 * time as another thread was calling stop_ep_timer().
4061 * So we silently do nothing for these states.
4062 */
4063 abort = 0;
4064 break;
be4c9bad 4065 default:
76f267b7 4066 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
be4c9bad 4067 __func__, ep, ep->hwtid, ep->com.state);
be4c9bad
RD		 4068 abort = 0;
4069 }
cc18b939 4070 mutex_unlock(&ep->com.mutex);
69736279
H		 4071 if (abort)
4072 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
be4c9bad
RD		 4073 c4iw_put_ep(&ep->com);
4074}
4075
4076static void process_timedout_eps(void)
4077{
4078 struct c4iw_ep *ep;
4079
4080 spin_lock_irq(&timeout_lock);
4081 while (!list_empty(&timeout_list)) {
4082 struct list_head *tmp;
4083
4084 tmp = timeout_list.next;
4085 list_del(tmp);
b33bd0cb
SW		 4086 tmp->next = NULL;
4087 tmp->prev = NULL;
be4c9bad
RD		 4088 spin_unlock_irq(&timeout_lock);
4089 ep = list_entry(tmp, struct c4iw_ep, entry);
4090 process_timeout(ep);
4091 spin_lock_irq(&timeout_lock);
4092 }
4093 spin_unlock_irq(&timeout_lock);
4094}
4095
4096static void process_work(struct work_struct *work)
4097{
4098 struct sk_buff *skb = NULL;
4099 struct c4iw_dev *dev;
c1d7356c 4100 struct cpl_act_establish *rpl;
be4c9bad
RD		 4101 unsigned int opcode;
4102 int ret;
4103
b33bd0cb 4104 process_timedout_eps();
be4c9bad
RD		 4105 while ((skb = skb_dequeue(&rxq))) {
4106 rpl = cplhdr(skb);
4107 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4108 opcode = rpl->ot.opcode;
4109
4110 BUG_ON(!work_handlers[opcode]);
4111 ret = work_handlers[opcode](dev, skb);
4112 if (!ret)
4113 kfree_skb(skb);
b33bd0cb 4114 process_timedout_eps();
be4c9bad 4115 }
be4c9bad
RD		 4116}
4117
4118static DECLARE_WORK(skb_work, process_work);
4119
4120static void ep_timeout(unsigned long arg)
4121{
4122 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
1ec779cc 4123 int kickit = 0;
be4c9bad
RD		 4124
4125 spin_lock(&timeout_lock);
1ec779cc 4126 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
b33bd0cb
SW		 4127 /*
4128 * Only insert if it is not already on the list.
4129 */
4130 if (!ep->entry.next) {
4131 list_add_tail(&ep->entry, &timeout_list);
4132 kickit = 1;
4133 }
1ec779cc 4134 }
be4c9bad 4135 spin_unlock(&timeout_lock);
1ec779cc
VP		 4136 if (kickit)
4137 queue_work(workq, &skb_work);
be4c9bad
RD		 4138}
4139
cfdda9d7
SW		 4140/*
4141 * All the CM events are handled on a work queue to have a safe context.
4142 */
4143static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4144{
4145
4146 /*
4147 * Save dev in the skb->cb area.
4148 */
4149 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4150
4151 /*
4152 * Queue the skb and schedule the worker thread.
4153 */
4154 skb_queue_tail(&rxq, skb);
4155 queue_work(workq, &skb_work);
4156 return 0;
4157}
4158
4159static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4160{
4161 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4162
4163 if (rpl->status != CPL_ERR_NONE) {
700456bd
JP		 4164 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4165 rpl->status, GET_TID(rpl));
cfdda9d7 4166 }
2f5b48c3 4167 kfree_skb(skb);
cfdda9d7
SW		 4168 return 0;
4169}
4170
be4c9bad
RD		 4171static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4172{
4173 struct cpl_fw6_msg *rpl = cplhdr(skb);
4174 struct c4iw_wr_wait *wr_waitp;
4175 int ret;
4176
a9a42886 4177 pr_debug("%s type %u\n", __func__, rpl->type);
be4c9bad
RD		 4178
4179 switch (rpl->type) {
5be78ee9 4180 case FW6_TYPE_WR_RPL:
be4c9bad 4181 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
c8e081a1 4182 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
a9a42886 4183 pr_debug("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
d9594d99
SW		 4184 if (wr_waitp)
4185 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2f5b48c3 4186 kfree_skb(skb);
be4c9bad 4187 break;
5be78ee9 4188 case FW6_TYPE_CQE:
5be78ee9 4189 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
1cab775c 4190 sched(dev, skb);
5be78ee9 4191 break;
be4c9bad 4192 default:
700456bd
JP		 4193 pr_err("%s unexpected fw6 msg type %u\n",
4194 __func__, rpl->type);
2f5b48c3 4195 kfree_skb(skb);
be4c9bad
RD		 4196 break;
4197 }
4198 return 0;
4199}
4200
8da7e7a5
SW		 4201static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4202{
4203 struct cpl_abort_req_rss *req = cplhdr(skb);
4204 struct c4iw_ep *ep;
8da7e7a5
SW		 4205 unsigned int tid = GET_TID(req);
4206
944661dd
H		 4207 ep = get_ep_from_tid(dev, tid);
4208 /* This EP will be dereferenced in peer_abort() */
14b92228 4209 if (!ep) {
700456bd 4210 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
14b92228
SW		 4211 kfree_skb(skb);
4212 return 0;
4213 }
b65eef0a 4214 if (cxgb_is_neg_adv(req->status)) {
a9a42886
JP		 4215 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
4216 __func__, ep->hwtid, req->status,
4217 neg_adv_str(req->status));
944661dd 4218 goto out;
8da7e7a5 4219 }
a9a42886
JP		 4220 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4221 ep->com.state);
8da7e7a5 4222
093108cb 4223 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
944661dd 4224out:
8da7e7a5
SW		 4225 sched(dev, skb);
4226 return 0;
4227}
4228
be4c9bad
RD		 4229/*
4230 * Most upcalls from the T4 Core go to sched() to
4231 * schedule the processing on a work queue.
4232 */
4233c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4234 [CPL_ACT_ESTABLISH] = sched,
4235 [CPL_ACT_OPEN_RPL] = sched,
4236 [CPL_RX_DATA] = sched,
4237 [CPL_ABORT_RPL_RSS] = sched,
4238 [CPL_ABORT_RPL] = sched,
4239 [CPL_PASS_OPEN_RPL] = sched,
4240 [CPL_CLOSE_LISTSRV_RPL] = sched,
4241 [CPL_PASS_ACCEPT_REQ] = sched,
4242 [CPL_PASS_ESTABLISH] = sched,
4243 [CPL_PEER_CLOSE] = sched,
4244 [CPL_CLOSE_CON_RPL] = sched,
8da7e7a5 4245 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
be4c9bad
RD		 4246 [CPL_RDMA_TERMINATE] = sched,
4247 [CPL_FW4_ACK] = sched,
4248 [CPL_SET_TCB_RPL] = set_tcb_rpl,
1cab775c
VP		 4249 [CPL_FW6_MSG] = fw6_msg,
4250 [CPL_RX_PKT] = sched
be4c9bad
RD		 4251};
4252
cfdda9d7
SW		 4253int __init c4iw_cm_init(void)
4254{
be4c9bad 4255 spin_lock_init(&timeout_lock);
cfdda9d7
SW		 4256 skb_queue_head_init(&rxq);
4257
52ee1a05 4258 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
cfdda9d7
SW		 4259 if (!workq)
4260 return -ENOMEM;
4261
cfdda9d7
SW		 4262 return 0;
4263}
4264
46c1376d 4265void c4iw_cm_term(void)
cfdda9d7 4266{
be4c9bad 4267 WARN_ON(!list_empty(&timeout_list));
cfdda9d7
SW		 4268 flush_workqueue(workq);
4269 destroy_workqueue(workq);
4270}
Linux 6.x block layer and io_uring tree(s)