projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'mvebu-fixes-4.14-1' of git://git.infradead.org/linux-mvebu into fixes
[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);
de77b966 1903 req = __skb_put_zero(skb, sizeof(*req));
6c53e938 1904 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
e2ac9628 1905 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
41b4f86c
KS		 1906 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1907 ep->com.dev->rdev.lldi.ports[0],
5be78ee9 1908 ep->l2t));
170003c8 1909 sin = (struct sockaddr_in *)&ep->com.local_addr;
830662f6
VP		 1910 req->le.lport = sin->sin_port;
1911 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
170003c8 1912 sin = (struct sockaddr_in *)&ep->com.remote_addr;
830662f6
VP		 1913 req->le.pport = sin->sin_port;
1914 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
5be78ee9 1915 req->tcb.t_state_to_astid =
77a80e23
HS		 1916 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1917 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
5be78ee9 1918 req->tcb.cplrxdataack_cplpassacceptrpl =
77a80e23 1919 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
ef5d6355 1920 req->tcb.tx_max = (__force __be32) jiffies;
793dad94 1921 req->tcb.rcv_adv = htons(1);
44c6d069
VP		 1922 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1923 enable_tcp_timestamps,
1924 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
cc516700 1925 wscale = cxgb_compute_wscale(rcv_win);
b408ff28
HS		 1926
1927 /*
1928 * Specify the largest window that will fit in opt0. The
1929 * remainder will be specified in the rx_data_ack.
1930 */
1931 win = ep->rcv_win >> 10;
d7990b0c
AB		 1932 if (win > RCV_BUFSIZ_M)
1933 win = RCV_BUFSIZ_M;
b408ff28 1934
6c53e938
HS		 1935 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1936 (nocong ? NO_CONG_F : 0) |
d7990b0c 1937 KEEP_ALIVE_F |
6c53e938 1938 DELACK_F |
d7990b0c
AB		 1939 WND_SCALE_V(wscale) |
1940 MSS_IDX_V(mtu_idx) |
1941 L2T_IDX_V(ep->l2t->idx) |
1942 TX_CHAN_V(ep->tx_chan) |
1943 SMAC_SEL_V(ep->smac_idx) |
ac8e4c69 1944 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 1945 ULP_MODE_V(ULP_MODE_TCPDDP) |
1946 RCV_BUFSIZ_V(win));
6c53e938
HS		 1947 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1948 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
d7990b0c 1949 RX_CHANNEL_V(0) |
6c53e938 1950 CCTRL_ECN_V(enable_ecn) |
d7990b0c 1951 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
5be78ee9 1952 if (enable_tcp_timestamps)
6c53e938 1953 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
5be78ee9 1954 if (enable_tcp_sack)
6c53e938 1955 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
5be78ee9 1956 if (wscale && enable_tcp_window_scaling)
d7990b0c
AB		 1957 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1958 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1959 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
793dad94
VP		 1960 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1961 set_bit(ACT_OFLD_CONN, &ep->com.history);
caa6c9f2 1962 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
5be78ee9
VP		 1963}
1964
cfdda9d7 1965/*
4c72efef
H		 1966 * Some of the error codes above implicitly indicate that there is no TID
1967 * allocated with the result of an ACT_OPEN. We use this predicate to make
1968 * that explicit.
cfdda9d7
SW		 1969 */
1970static inline int act_open_has_tid(int status)
1971{
4c72efef
H		 1972 return (status != CPL_ERR_TCAM_PARITY &&
1973 status != CPL_ERR_TCAM_MISS &&
1974 status != CPL_ERR_TCAM_FULL &&
1975 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1976 status != CPL_ERR_CONN_EXIST);
cfdda9d7
SW		 1977}
1978
dd92b124
HS		 1979static char *neg_adv_str(unsigned int status)
1980{
1981 switch (status) {
1982 case CPL_ERR_RTX_NEG_ADVICE:
1983 return "Retransmit timeout";
1984 case CPL_ERR_PERSIST_NEG_ADVICE:
1985 return "Persist timeout";
1986 case CPL_ERR_KEEPALV_NEG_ADVICE:
1987 return "Keepalive timeout";
1988 default:
1989 return "Unknown";
1990 }
1991}
1992
b408ff28
HS		 1993static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1994{
1995 ep->snd_win = snd_win;
1996 ep->rcv_win = rcv_win;
a9a42886
JP		 1997 pr_debug("%s snd_win %d rcv_win %d\n",
1998 __func__, ep->snd_win, ep->rcv_win);
b408ff28
HS		 1999}
2000
793dad94
VP		 2001#define ACT_OPEN_RETRY_COUNT 2
2002
830662f6
VP		 2003static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2004 struct dst_entry *dst, struct c4iw_dev *cdev,
ac8e4c69 2005 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
830662f6
VP		 2006{
2007 struct neighbour *n;
2008 int err, step;
2009 struct net_device *pdev;
2010
2011 n = dst_neigh_lookup(dst, peer_ip);
2012 if (!n)
2013 return -ENODEV;
2014
2015 rcu_read_lock();
2016 err = -ENOMEM;
2017 if (n->dev->flags & IFF_LOOPBACK) {
2018 if (iptype == 4)
2019 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2020 else if (IS_ENABLED(CONFIG_IPV6))
2021 for_each_netdev(&init_net, pdev) {
2022 if (ipv6_chk_addr(&init_net,
2023 (struct in6_addr *)peer_ip,
2024 pdev, 1))
2025 break;
2026 }
2027 else
2028 pdev = NULL;
2029
2030 if (!pdev) {
2031 err = -ENODEV;
2032 goto out;
2033 }
2034 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
ac8e4c69 2035 n, pdev, rt_tos2priority(tos));
609e941a
SW		 2036 if (!ep->l2t) {
2037 dev_put(pdev);
830662f6 2038 goto out;
609e941a 2039 }
830662f6
VP		 2040 ep->mtu = pdev->mtu;
2041 ep->tx_chan = cxgb4_port_chan(pdev);
963cab50
H		 2042 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2043 cxgb4_port_viid(pdev));
830662f6
VP		 2044 step = cdev->rdev.lldi.ntxq /
2045 cdev->rdev.lldi.nchan;
2046 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2047 step = cdev->rdev.lldi.nrxq /
2048 cdev->rdev.lldi.nchan;
2049 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2050 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2051 cxgb4_port_idx(pdev) * step];
b408ff28 2052 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
830662f6
VP		 2053 dev_put(pdev);
2054 } else {
2055 pdev = get_real_dev(n->dev);
2056 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2057 n, pdev, 0);
2058 if (!ep->l2t)
2059 goto out;
2060 ep->mtu = dst_mtu(dst);
11b8e22d 2061 ep->tx_chan = cxgb4_port_chan(pdev);
963cab50
H		 2062 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2063 cxgb4_port_viid(pdev));
830662f6
VP		 2064 step = cdev->rdev.lldi.ntxq /
2065 cdev->rdev.lldi.nchan;
11b8e22d
SW		 2066 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2067 ep->ctrlq_idx = cxgb4_port_idx(pdev);
830662f6
VP		 2068 step = cdev->rdev.lldi.nrxq /
2069 cdev->rdev.lldi.nchan;
2070 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
11b8e22d 2071 cxgb4_port_idx(pdev) * step];
b408ff28 2072 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
830662f6
VP		 2073
2074 if (clear_mpa_v1) {
2075 ep->retry_with_mpa_v1 = 0;
2076 ep->tried_with_mpa_v1 = 0;
2077 }
2078 }
2079 err = 0;
2080out:
2081 rcu_read_unlock();
2082
2083 neigh_release(n);
2084
2085 return err;
2086}
2087
793dad94
VP		 2088static int c4iw_reconnect(struct c4iw_ep *ep)
2089{
2090 int err = 0;
4a740838 2091 int size = 0;
24d44a39 2092 struct sockaddr_in *laddr = (struct sockaddr_in *)
170003c8 2093 &ep->com.cm_id->m_local_addr;
24d44a39 2094 struct sockaddr_in *raddr = (struct sockaddr_in *)
170003c8 2095 &ep->com.cm_id->m_remote_addr;
830662f6 2096 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
170003c8 2097 &ep->com.cm_id->m_local_addr;
830662f6 2098 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
170003c8 2099 &ep->com.cm_id->m_remote_addr;
830662f6
VP		 2100 int iptype;
2101 __u8 *ra;
793dad94 2102
a9a42886 2103 pr_debug("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
793dad94 2104 init_timer(&ep->timer);
093108cb 2105 c4iw_init_wr_wait(&ep->com.wr_wait);
793dad94 2106
4a740838
H		 2107 /* When MPA revision is different on nodes, the node with MPA_rev=2
2108 * tries to reconnect with MPA_rev 1 for the same EP through
2109 * c4iw_reconnect(), where the same EP is assigned with new tid for
2110 * further connection establishment. As we are using the same EP pointer
2111 * for reconnect, few skbs are used during the previous c4iw_connect(),
2112 * which leaves the EP with inadequate skbs for further
2113 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2114 * skb_list() during peer_abort(). Allocate skbs which is already used.
2115 */
2116 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2117 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2118 err = -ENOMEM;
2119 goto fail1;
2120 }
2121
793dad94
VP		 2122 /*
2123 * Allocate an active TID to initiate a TCP connection.
2124 */
2125 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2126 if (ep->atid == -1) {
700456bd 2127 pr_err("%s - cannot alloc atid\n", __func__);
793dad94
VP		 2128 err = -ENOMEM;
2129 goto fail2;
2130 }
2131 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2132
2133 /* find a route */
170003c8 2134 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
804c2f3e
VP		 2135 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2136 laddr->sin_addr.s_addr,
2137 raddr->sin_addr.s_addr,
2138 laddr->sin_port,
2139 raddr->sin_port, ep->com.cm_id->tos);
830662f6
VP		 2140 iptype = 4;
2141 ra = (__u8 *)&raddr->sin_addr;
2142 } else {
95554761
VP		 2143 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2144 get_real_dev,
2145 laddr6->sin6_addr.s6_addr,
2146 raddr6->sin6_addr.s6_addr,
2147 laddr6->sin6_port,
2148 raddr6->sin6_port, 0,
2149 raddr6->sin6_scope_id);
830662f6
VP		 2150 iptype = 6;
2151 ra = (__u8 *)&raddr6->sin6_addr;
2152 }
2153 if (!ep->dst) {
700456bd 2154 pr_err("%s - cannot find route\n", __func__);
793dad94
VP		 2155 err = -EHOSTUNREACH;
2156 goto fail3;
2157 }
963cab50 2158 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
ac8e4c69
H		 2159 ep->com.dev->rdev.lldi.adapter_type,
2160 ep->com.cm_id->tos);
830662f6 2161 if (err) {
700456bd 2162 pr_err("%s - cannot alloc l2e\n", __func__);
793dad94
VP		 2163 goto fail4;
2164 }
2165
a9a42886
JP		 2166 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2167 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2168 ep->l2t->idx);
793dad94
VP		 2169
2170 state_set(&ep->com, CONNECTING);
ac8e4c69 2171 ep->tos = ep->com.cm_id->tos;
793dad94
VP		 2172
2173 /* send connect request to rnic */
2174 err = send_connect(ep);
2175 if (!err)
2176 goto out;
2177
2178 cxgb4_l2t_release(ep->l2t);
2179fail4:
2180 dst_release(ep->dst);
2181fail3:
2182 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2183 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2184fail2:
2185 /*
2186 * remember to send notification to upper layer.
2187 * We are in here so the upper layer is not aware that this is
2188 * re-connect attempt and so, upper layer is still waiting for
2189 * response of 1st connect request.
2190 */
2191 connect_reply_upcall(ep, -ECONNRESET);
4a740838 2192fail1:
793dad94
VP		 2193 c4iw_put_ep(&ep->com);
2194out:
2195 return err;
2196}
2197
cfdda9d7
SW		 2198static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2199{
2200 struct c4iw_ep *ep;
2201 struct cpl_act_open_rpl *rpl = cplhdr(skb);
6c53e938
HS		 2202 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2203 ntohl(rpl->atid_status)));
cfdda9d7 2204 struct tid_info *t = dev->rdev.lldi.tids;
6c53e938 2205 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
830662f6
VP		 2206 struct sockaddr_in *la;
2207 struct sockaddr_in *ra;
2208 struct sockaddr_in6 *la6;
2209 struct sockaddr_in6 *ra6;
caa6c9f2 2210 int ret = 0;
cfdda9d7
SW		 2211
2212 ep = lookup_atid(t, atid);
170003c8
SW		 2213 la = (struct sockaddr_in *)&ep->com.local_addr;
2214 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2215 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2216 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
cfdda9d7 2217
a9a42886
JP		 2218 pr_debug("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2219 status, status2errno(status));
cfdda9d7 2220
b65eef0a 2221 if (cxgb_is_neg_adv(status)) {
a9a42886
JP		 2222 pr_debug("%s Connection problems for atid %u status %u (%s)\n",
2223 __func__, atid, status, neg_adv_str(status));
179d03bb
H		 2224 ep->stats.connect_neg_adv++;
2225 mutex_lock(&dev->rdev.stats.lock);
2226 dev->rdev.stats.neg_adv++;
2227 mutex_unlock(&dev->rdev.stats.lock);
cfdda9d7
SW		 2228 return 0;
2229 }
2230
793dad94
VP		 2231 set_bit(ACT_OPEN_RPL, &ep->com.history);
2232
d716a2a0
VP		 2233 /*
2234 * Log interesting failures.
2235 */
2236 switch (status) {
2237 case CPL_ERR_CONN_RESET:
2238 case CPL_ERR_CONN_TIMEDOUT:
2239 break;
5be78ee9 2240 case CPL_ERR_TCAM_FULL:
830662f6 2241 mutex_lock(&dev->rdev.stats.lock);
3b174d94 2242 dev->rdev.stats.tcam_full++;
830662f6
VP		 2243 mutex_unlock(&dev->rdev.stats.lock);
2244 if (ep->com.local_addr.ss_family == AF_INET &&
2245 dev->rdev.lldi.enable_fw_ofld_conn) {
caa6c9f2
H		 2246 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2247 ntohl(rpl->atid_status))));
2248 if (ret)
2249 goto fail;
793dad94
VP		 2250 return 0;
2251 }
2252 break;
2253 case CPL_ERR_CONN_EXIST:
2254 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2255 set_bit(ACT_RETRY_INUSE, &ep->com.history);
84cc6ac6
H		 2256 if (ep->com.remote_addr.ss_family == AF_INET6) {
2257 struct sockaddr_in6 *sin6 =
2258 (struct sockaddr_in6 *)
170003c8 2259 &ep->com.local_addr;
84cc6ac6
H		 2260 cxgb4_clip_release(
2261 ep->com.dev->rdev.lldi.ports[0],
2262 (const u32 *)
2263 &sin6->sin6_addr.s6_addr, 1);
2264 }
793dad94
VP		 2265 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2266 atid);
2267 cxgb4_free_atid(t, atid);
2268 dst_release(ep->dst);
2269 cxgb4_l2t_release(ep->l2t);
2270 c4iw_reconnect(ep);
2271 return 0;
2272 }
5be78ee9 2273 break;
d716a2a0 2274 default:
830662f6
VP		 2275 if (ep->com.local_addr.ss_family == AF_INET) {
2276 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2277 atid, status, status2errno(status),
2278 &la->sin_addr.s_addr, ntohs(la->sin_port),
2279 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2280 } else {
2281 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2282 atid, status, status2errno(status),
2283 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2284 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2285 }
d716a2a0
VP		 2286 break;
2287 }
2288
caa6c9f2 2289fail:
cfdda9d7
SW		 2290 connect_reply_upcall(ep, status2errno(status));
2291 state_set(&ep->com, DEAD);
2292
84cc6ac6
H		 2293 if (ep->com.remote_addr.ss_family == AF_INET6) {
2294 struct sockaddr_in6 *sin6 =
170003c8 2295 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 2296 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2297 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2298 }
cfdda9d7 2299 if (status && act_open_has_tid(status))
1dec4cec
GG		 2300 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2301 ep->com.local_addr.ss_family);
cfdda9d7 2302
793dad94 2303 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7
SW		 2304 cxgb4_free_atid(t, atid);
2305 dst_release(ep->dst);
2306 cxgb4_l2t_release(ep->l2t);
2307 c4iw_put_ep(&ep->com);
2308
2309 return 0;
2310}
2311
2312static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2313{
2314 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
cfdda9d7 2315 unsigned int stid = GET_TID(rpl);
f86fac79 2316 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
cfdda9d7
SW		 2317
2318 if (!ep) {
a9a42886 2319 pr_debug("%s stid %d lookup failure!\n", __func__, stid);
1cab775c 2320 goto out;
cfdda9d7 2321 }
a9a42886
JP		 2322 pr_debug("%s ep %p status %d error %d\n", __func__, ep,
2323 rpl->status, status2errno(rpl->status));
d9594d99 2324 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
f86fac79 2325 c4iw_put_ep(&ep->com);
1cab775c 2326out:
cfdda9d7
SW		 2327 return 0;
2328}
2329
cfdda9d7
SW		 2330static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2331{
2332 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
cfdda9d7 2333 unsigned int stid = GET_TID(rpl);
f86fac79 2334 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
cfdda9d7 2335
3c8415cc
SW		 2336 if (!ep) {
2337 pr_debug("%s stid %d lookup failure!\n", __func__, stid);
2338 goto out;
2339 }
a9a42886 2340 pr_debug("%s ep %p\n", __func__, ep);
d9594d99 2341 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
f86fac79 2342 c4iw_put_ep(&ep->com);
3c8415cc 2343out:
cfdda9d7
SW		 2344 return 0;
2345}
2346
9dec900c
H		 2347static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2348 struct cpl_pass_accept_req *req)
cfdda9d7
SW		 2349{
2350 struct cpl_pass_accept_rpl *rpl;
2351 unsigned int mtu_idx;
2352 u64 opt0;
2353 u32 opt2;
cc516700 2354 u32 wscale;
92e7ae71 2355 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
b408ff28 2356 int win;
963cab50 2357 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
cfdda9d7 2358
a9a42886 2359 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7 2360 BUG_ON(skb_cloned(skb));
92e7ae71 2361
cfdda9d7 2362 skb_get(skb);
92e7ae71 2363 rpl = cplhdr(skb);
963cab50 2364 if (!is_t4(adapter_type)) {
92e7ae71
HS		 2365 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2366 rpl5 = (void *)rpl;
2367 INIT_TP_WR(rpl5, ep->hwtid);
2368 } else {
2369 skb_trim(skb, sizeof(*rpl));
2370 INIT_TP_WR(rpl, ep->hwtid);
2371 }
2372 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2373 ep->hwtid));
2374
44c6d069
VP		 2375 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2376 enable_tcp_timestamps && req->tcpopt.tstamp,
2377 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
cc516700 2378 wscale = cxgb_compute_wscale(rcv_win);
b408ff28
HS		 2379
2380 /*
2381 * Specify the largest window that will fit in opt0. The
2382 * remainder will be specified in the rx_data_ack.
2383 */
2384 win = ep->rcv_win >> 10;
d7990b0c
AB		 2385 if (win > RCV_BUFSIZ_M)
2386 win = RCV_BUFSIZ_M;
6c53e938 2387 opt0 = (nocong ? NO_CONG_F : 0) |
d7990b0c 2388 KEEP_ALIVE_F |
6c53e938 2389 DELACK_F |
d7990b0c
AB		 2390 WND_SCALE_V(wscale) |
2391 MSS_IDX_V(mtu_idx) |
2392 L2T_IDX_V(ep->l2t->idx) |
2393 TX_CHAN_V(ep->tx_chan) |
2394 SMAC_SEL_V(ep->smac_idx) |
6c53e938 2395 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 2396 ULP_MODE_V(ULP_MODE_TCPDDP) |
2397 RCV_BUFSIZ_V(win);
2398 opt2 = RX_CHANNEL_V(0) |
2399 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
cfdda9d7
SW		 2400
2401 if (enable_tcp_timestamps && req->tcpopt.tstamp)
6c53e938 2402 opt2 |= TSTAMPS_EN_F;
cfdda9d7 2403 if (enable_tcp_sack && req->tcpopt.sack)
6c53e938 2404 opt2 |= SACK_EN_F;
cfdda9d7 2405 if (wscale && enable_tcp_window_scaling)
d7990b0c 2406 opt2 |= WND_SCALE_EN_F;
5be78ee9
VP		 2407 if (enable_ecn) {
2408 const struct tcphdr *tcph;
2409 u32 hlen = ntohl(req->hdr_len);
2410
963cab50
H		 2411 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2412 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2413 IP_HDR_LEN_G(hlen);
2414 else
2415 tcph = (const void *)(req + 1) +
2416 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
5be78ee9 2417 if (tcph->ece && tcph->cwr)
6c53e938 2418 opt2 |= CCTRL_ECN_V(1);
5be78ee9 2419 }
963cab50 2420 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
92e7ae71 2421 u32 isn = (prandom_u32() & ~7UL) - 1;
d7990b0c 2422 opt2 |= T5_OPT_2_VALID_F;
cf7fe64a 2423 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
0b741047 2424 opt2 |= T5_ISS_F;
92e7ae71
HS		 2425 rpl5 = (void *)rpl;
2426 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2427 if (peer2peer)
2428 isn += 4;
2429 rpl5->iss = cpu_to_be32(isn);
a9a42886 2430 pr_debug("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
92e5011a 2431 }
cfdda9d7 2432
cfdda9d7
SW		 2433 rpl->opt0 = cpu_to_be64(opt0);
2434 rpl->opt2 = cpu_to_be32(opt2);
d4f1a5c6 2435 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
9dec900c 2436 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
cfdda9d7 2437
9dec900c 2438 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
cfdda9d7
SW		 2439}
2440
830662f6 2441static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
cfdda9d7 2442{
a9a42886 2443 pr_debug("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
cfdda9d7
SW		 2444 BUG_ON(skb_cloned(skb));
2445 skb_trim(skb, sizeof(struct cpl_tid_release));
cfdda9d7
SW		 2446 release_tid(&dev->rdev, hwtid, skb);
2447 return;
2448}
2449
cfdda9d7
SW		 2450static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2451{
793dad94 2452 struct c4iw_ep *child_ep = NULL, *parent_ep;
cfdda9d7 2453 struct cpl_pass_accept_req *req = cplhdr(skb);
6c53e938 2454 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
cfdda9d7
SW		 2455 struct tid_info *t = dev->rdev.lldi.tids;
2456 unsigned int hwtid = GET_TID(req);
2457 struct dst_entry *dst;
830662f6 2458 __u8 local_ip[16], peer_ip[16];
cfdda9d7 2459 __be16 local_port, peer_port;
84cc6ac6 2460 struct sockaddr_in6 *sin6;
3786cf18 2461 int err;
1cab775c 2462 u16 peer_mss = ntohs(req->tcpopt.mss);
830662f6 2463 int iptype;
92e7ae71 2464 unsigned short hdrs;
ac8e4c69 2465 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
cfdda9d7 2466
f86fac79 2467 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
1cab775c 2468 if (!parent_ep) {
a9a42886
JP		 2469 pr_debug("%s connect request on invalid stid %d\n",
2470 __func__, stid);
1cab775c
VP		 2471 goto reject;
2472 }
1cab775c 2473
cfdda9d7 2474 if (state_read(&parent_ep->com) != LISTEN) {
a9a42886 2475 pr_debug("%s - listening ep not in LISTEN\n", __func__);
cfdda9d7
SW		 2476 goto reject;
2477 }
2478
85e42b04
VP		 2479 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2480 &iptype, local_ip, peer_ip, &local_port, &peer_port);
830662f6 2481
cfdda9d7 2482 /* Find output route */
830662f6 2483 if (iptype == 4) {
a9a42886
JP		 2484 pr_debug("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2485 , __func__, parent_ep, hwtid,
2486 local_ip, peer_ip, ntohs(local_port),
2487 ntohs(peer_port), peer_mss);
804c2f3e
VP		 2488 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2489 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2490 local_port, peer_port, tos);
830662f6 2491 } else {
a9a42886
JP		 2492 pr_debug("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2493 , __func__, parent_ep, hwtid,
2494 local_ip, peer_ip, ntohs(local_port),
2495 ntohs(peer_port), peer_mss);
95554761
VP		 2496 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2497 local_ip, peer_ip, local_port, peer_port,
2498 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2499 ((struct sockaddr_in6 *)
2500 &parent_ep->com.local_addr)->sin6_scope_id);
830662f6
VP		 2501 }
2502 if (!dst) {
700456bd 2503 pr_err("%s - failed to find dst entry!\n", __func__);
cfdda9d7
SW		 2504 goto reject;
2505 }
3786cf18
DM		 2506
2507 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2508 if (!child_ep) {
700456bd 2509 pr_err("%s - failed to allocate ep entry!\n", __func__);
cfdda9d7
SW		 2510 dst_release(dst);
2511 goto reject;
2512 }
2513
963cab50 2514 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
ac8e4c69 2515 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
3786cf18 2516 if (err) {
700456bd 2517 pr_err("%s - failed to allocate l2t entry!\n", __func__);
cfdda9d7 2518 dst_release(dst);
3786cf18 2519 kfree(child_ep);
cfdda9d7
SW		 2520 goto reject;
2521 }
3786cf18 2522
98b80a2a
RR		 2523 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2524 sizeof(struct tcphdr) +
92e7ae71
HS		 2525 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2526 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2527 child_ep->mtu = peer_mss + hdrs;
1cab775c 2528
4a740838
H		 2529 skb_queue_head_init(&child_ep->com.ep_skb_list);
2530 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2531 goto fail;
2532
cfdda9d7
SW		 2533 state_set(&child_ep->com, CONNECTING);
2534 child_ep->com.dev = dev;
2535 child_ep->com.cm_id = NULL;
5b6b8fe6 2536
830662f6
VP		 2537 if (iptype == 4) {
2538 struct sockaddr_in *sin = (struct sockaddr_in *)
170003c8 2539 &child_ep->com.local_addr;
5b6b8fe6 2540
b462b06e 2541 sin->sin_family = AF_INET;
830662f6
VP		 2542 sin->sin_port = local_port;
2543 sin->sin_addr.s_addr = *(__be32 *)local_ip;
5b6b8fe6
SW		 2544
2545 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
b462b06e 2546 sin->sin_family = AF_INET;
5b6b8fe6
SW		 2547 sin->sin_port = ((struct sockaddr_in *)
2548 &parent_ep->com.local_addr)->sin_port;
2549 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2550
170003c8 2551 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
b462b06e 2552 sin->sin_family = AF_INET;
830662f6
VP		 2553 sin->sin_port = peer_port;
2554 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2555 } else {
170003c8 2556 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
830662f6
VP		 2557 sin6->sin6_family = PF_INET6;
2558 sin6->sin6_port = local_port;
2559 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
5b6b8fe6
SW		 2560
2561 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2562 sin6->sin6_family = PF_INET6;
2563 sin6->sin6_port = ((struct sockaddr_in6 *)
2564 &parent_ep->com.local_addr)->sin6_port;
2565 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2566
170003c8 2567 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
830662f6
VP		 2568 sin6->sin6_family = PF_INET6;
2569 sin6->sin6_port = peer_port;
2570 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2571 }
5b6b8fe6 2572
cfdda9d7
SW		 2573 c4iw_get_ep(&parent_ep->com);
2574 child_ep->parent_ep = parent_ep;
ac8e4c69 2575 child_ep->tos = tos;
cfdda9d7
SW		 2576 child_ep->dst = dst;
2577 child_ep->hwtid = hwtid;
cfdda9d7 2578
a9a42886
JP		 2579 pr_debug("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2580 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
cfdda9d7
SW		 2581
2582 init_timer(&child_ep->timer);
1dec4cec
GG		 2583 cxgb4_insert_tid(t, child_ep, hwtid,
2584 child_ep->com.local_addr.ss_family);
944661dd 2585 insert_ep_tid(child_ep);
9dec900c
H		 2586 if (accept_cr(child_ep, skb, req)) {
2587 c4iw_put_ep(&parent_ep->com);
2588 release_ep_resources(child_ep);
2589 } else {
2590 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2591 }
84cc6ac6 2592 if (iptype == 6) {
170003c8 2593 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
84cc6ac6
H		 2594 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2595 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2596 }
cfdda9d7 2597 goto out;
4a740838
H		 2598fail:
2599 c4iw_put_ep(&child_ep->com);
cfdda9d7 2600reject:
830662f6 2601 reject_cr(dev, hwtid, skb);
3d318605 2602out:
f86fac79
H		 2603 if (parent_ep)
2604 c4iw_put_ep(&parent_ep->com);
cfdda9d7
SW		 2605 return 0;
2606}
2607
2608static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2609{
2610 struct c4iw_ep *ep;
2611 struct cpl_pass_establish *req = cplhdr(skb);
cfdda9d7 2612 unsigned int tid = GET_TID(req);
fef4422d 2613 int ret;
cfdda9d7 2614
944661dd 2615 ep = get_ep_from_tid(dev, tid);
a9a42886 2616 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 2617 ep->snd_seq = be32_to_cpu(req->snd_isn);
2618 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2619
a9a42886
JP		 2620 pr_debug("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2621 ntohs(req->tcp_opt));
1cab775c 2622
cfdda9d7
SW		 2623 set_emss(ep, ntohs(req->tcp_opt));
2624
2625 dst_confirm(ep->dst);
fef4422d
H		 2626 mutex_lock(&ep->com.mutex);
2627 ep->com.state = MPA_REQ_WAIT;
cfdda9d7 2628 start_ep_timer(ep);
793dad94 2629 set_bit(PASS_ESTAB, &ep->com.history);
4a740838 2630 ret = send_flowc(ep);
fef4422d
H		 2631 mutex_unlock(&ep->com.mutex);
2632 if (ret)
2633 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
944661dd 2634 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2635
2636 return 0;
2637}
2638
2639static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2640{
2641 struct cpl_peer_close *hdr = cplhdr(skb);
2642 struct c4iw_ep *ep;
2643 struct c4iw_qp_attributes attrs;
cfdda9d7
SW		 2644 int disconnect = 1;
2645 int release = 0;
cfdda9d7 2646 unsigned int tid = GET_TID(hdr);
8da7e7a5 2647 int ret;
cfdda9d7 2648
944661dd
H		 2649 ep = get_ep_from_tid(dev, tid);
2650 if (!ep)
2651 return 0;
2652
a9a42886 2653 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 2654 dst_confirm(ep->dst);
2655
793dad94 2656 set_bit(PEER_CLOSE, &ep->com.history);
2f5b48c3 2657 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2658 switch (ep->com.state) {
2659 case MPA_REQ_WAIT:
2660 __state_set(&ep->com, CLOSING);
2661 break;
2662 case MPA_REQ_SENT:
2663 __state_set(&ep->com, CLOSING);
2664 connect_reply_upcall(ep, -ECONNRESET);
2665 break;
2666 case MPA_REQ_RCVD:
2667
2668 /*
2669 * We're gonna mark this puppy DEAD, but keep
2670 * the reference on it until the ULP accepts or
2671 * rejects the CR. Also wake up anyone waiting
2672 * in rdma connection migration (see c4iw_accept_cr()).
2673 */
2674 __state_set(&ep->com, CLOSING);
a9a42886 2675 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2676 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2677 break;
2678 case MPA_REP_SENT:
2679 __state_set(&ep->com, CLOSING);
a9a42886 2680 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2681 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2682 break;
2683 case FPDU_MODE:
ca5a2202 2684 start_ep_timer(ep);
cfdda9d7 2685 __state_set(&ep->com, CLOSING);
30c95c2d 2686 attrs.next_state = C4IW_QP_STATE_CLOSING;
8da7e7a5 2687 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
30c95c2d 2688 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
8da7e7a5
SW		 2689 if (ret != -ECONNRESET) {
2690 peer_close_upcall(ep);
2691 disconnect = 1;
2692 }
cfdda9d7
SW		 2693 break;
2694 case ABORTING:
2695 disconnect = 0;
2696 break;
2697 case CLOSING:
2698 __state_set(&ep->com, MORIBUND);
2699 disconnect = 0;
2700 break;
2701 case MORIBUND:
b33bd0cb 2702 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2703 if (ep->com.cm_id && ep->com.qp) {
2704 attrs.next_state = C4IW_QP_STATE_IDLE;
2705 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2706 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2707 }
be13b2df 2708 close_complete_upcall(ep, 0);
cfdda9d7
SW		 2709 __state_set(&ep->com, DEAD);
2710 release = 1;
2711 disconnect = 0;
2712 break;
2713 case DEAD:
2714 disconnect = 0;
2715 break;
2716 default:
2717 BUG_ON(1);
2718 }
2f5b48c3 2719 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2720 if (disconnect)
2721 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2722 if (release)
2723 release_ep_resources(ep);
944661dd 2724 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2725 return 0;
2726}
2727
cfdda9d7
SW		 2728static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2729{
2730 struct cpl_abort_req_rss *req = cplhdr(skb);
2731 struct c4iw_ep *ep;
cfdda9d7
SW		 2732 struct sk_buff *rpl_skb;
2733 struct c4iw_qp_attributes attrs;
2734 int ret;
2735 int release = 0;
cfdda9d7 2736 unsigned int tid = GET_TID(req);
052f4731 2737 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
cfdda9d7 2738
944661dd
H		 2739 ep = get_ep_from_tid(dev, tid);
2740 if (!ep)
2741 return 0;
2742
b65eef0a 2743 if (cxgb_is_neg_adv(req->status)) {
a9a42886
JP		 2744 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
2745 __func__, ep->hwtid, req->status,
2746 neg_adv_str(req->status));
179d03bb
H		 2747 ep->stats.abort_neg_adv++;
2748 mutex_lock(&dev->rdev.stats.lock);
2749 dev->rdev.stats.neg_adv++;
2750 mutex_unlock(&dev->rdev.stats.lock);
944661dd 2751 goto deref_ep;
cfdda9d7 2752 }
a9a42886
JP		 2753 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2754 ep->com.state);
793dad94 2755 set_bit(PEER_ABORT, &ep->com.history);
2f5b48c3
SW		 2756
2757 /*
2758 * Wake up any threads in rdma_init() or rdma_fini().
d2fe99e8
KS		 2759 * However, this is not needed if com state is just
2760 * MPA_REQ_SENT
2f5b48c3 2761 */
d2fe99e8
KS		 2762 if (ep->com.state != MPA_REQ_SENT)
2763 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2f5b48c3
SW		 2764
2765 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2766 switch (ep->com.state) {
2767 case CONNECTING:
9dec900c 2768 c4iw_put_ep(&ep->parent_ep->com);
cfdda9d7
SW		 2769 break;
2770 case MPA_REQ_WAIT:
b33bd0cb 2771 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2772 break;
2773 case MPA_REQ_SENT:
b33bd0cb 2774 (void)stop_ep_timer(ep);
fe7e0a4d 2775 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
d2fe99e8
KS		 2776 connect_reply_upcall(ep, -ECONNRESET);
2777 else {
2778 /*
2779 * we just don't send notification upwards because we
2780 * want to retry with mpa_v1 without upper layers even
2781 * knowing it.
2782 *
2783 * do some housekeeping so as to re-initiate the
2784 * connection
2785 */
a9a42886
JP		 2786 pr_debug("%s: mpa_rev=%d. Retrying with mpav1\n",
2787 __func__, mpa_rev);
d2fe99e8
KS		 2788 ep->retry_with_mpa_v1 = 1;
2789 }
cfdda9d7
SW		 2790 break;
2791 case MPA_REP_SENT:
cfdda9d7
SW		 2792 break;
2793 case MPA_REQ_RCVD:
cfdda9d7
SW		 2794 break;
2795 case MORIBUND:
2796 case CLOSING:
ca5a2202 2797 stop_ep_timer(ep);
cfdda9d7
SW		 2798 /*FALLTHROUGH*/
2799 case FPDU_MODE:
2800 if (ep->com.cm_id && ep->com.qp) {
2801 attrs.next_state = C4IW_QP_STATE_ERROR;
2802 ret = c4iw_modify_qp(ep->com.qp->rhp,
2803 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2804 &attrs, 1);
2805 if (ret)
700456bd 2806 pr_err("%s - qp <- error failed!\n", __func__);
cfdda9d7
SW		 2807 }
2808 peer_abort_upcall(ep);
2809 break;
2810 case ABORTING:
2811 break;
2812 case DEAD:
a9a42886 2813 pr_debug("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2f5b48c3 2814 mutex_unlock(&ep->com.mutex);
944661dd 2815 goto deref_ep;
cfdda9d7
SW		 2816 default:
2817 BUG_ON(1);
2818 break;
2819 }
2820 dst_confirm(ep->dst);
2821 if (ep->com.state != ABORTING) {
2822 __state_set(&ep->com, DEAD);
d2fe99e8
KS		 2823 /* we don't release if we want to retry with mpa_v1 */
2824 if (!ep->retry_with_mpa_v1)
2825 release = 1;
cfdda9d7 2826 }
2f5b48c3 2827 mutex_unlock(&ep->com.mutex);
cfdda9d7 2828
4a740838
H		 2829 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2830 if (WARN_ON(!rpl_skb)) {
cfdda9d7
SW		 2831 release = 1;
2832 goto out;
2833 }
052f4731
VP		 2834
2835 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2836
cfdda9d7
SW		 2837 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2838out:
cfdda9d7
SW		 2839 if (release)
2840 release_ep_resources(ep);
fe7e0a4d 2841 else if (ep->retry_with_mpa_v1) {
84cc6ac6
H		 2842 if (ep->com.remote_addr.ss_family == AF_INET6) {
2843 struct sockaddr_in6 *sin6 =
2844 (struct sockaddr_in6 *)
170003c8 2845 &ep->com.local_addr;
84cc6ac6
H		 2846 cxgb4_clip_release(
2847 ep->com.dev->rdev.lldi.ports[0],
2848 (const u32 *)&sin6->sin6_addr.s6_addr,
2849 1);
2850 }
fe7e0a4d 2851 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
1dec4cec
GG		 2852 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2853 ep->com.local_addr.ss_family);
d2fe99e8
KS		 2854 dst_release(ep->dst);
2855 cxgb4_l2t_release(ep->l2t);
2856 c4iw_reconnect(ep);
2857 }
2858
944661dd
H		 2859deref_ep:
2860 c4iw_put_ep(&ep->com);
2861 /* Dereferencing ep, referenced in peer_abort_intr() */
2862 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2863 return 0;
2864}
2865
2866static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2867{
2868 struct c4iw_ep *ep;
2869 struct c4iw_qp_attributes attrs;
2870 struct cpl_close_con_rpl *rpl = cplhdr(skb);
cfdda9d7 2871 int release = 0;
cfdda9d7 2872 unsigned int tid = GET_TID(rpl);
cfdda9d7 2873
944661dd
H		 2874 ep = get_ep_from_tid(dev, tid);
2875 if (!ep)
2876 return 0;
cfdda9d7 2877
a9a42886 2878 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7
SW		 2879
2880 /* The cm_id may be null if we failed to connect */
2f5b48c3 2881 mutex_lock(&ep->com.mutex);
9ca6f7cf 2882 set_bit(CLOSE_CON_RPL, &ep->com.history);
cfdda9d7
SW		 2883 switch (ep->com.state) {
2884 case CLOSING:
2885 __state_set(&ep->com, MORIBUND);
2886 break;
2887 case MORIBUND:
b33bd0cb 2888 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2889 if ((ep->com.cm_id) && (ep->com.qp)) {
2890 attrs.next_state = C4IW_QP_STATE_IDLE;
2891 c4iw_modify_qp(ep->com.qp->rhp,
2892 ep->com.qp,
2893 C4IW_QP_ATTR_NEXT_STATE,
2894 &attrs, 1);
2895 }
be13b2df 2896 close_complete_upcall(ep, 0);
cfdda9d7
SW		 2897 __state_set(&ep->com, DEAD);
2898 release = 1;
2899 break;
2900 case ABORTING:
2901 case DEAD:
2902 break;
2903 default:
2904 BUG_ON(1);
2905 break;
2906 }
2f5b48c3 2907 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2908 if (release)
2909 release_ep_resources(ep);
944661dd 2910 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2911 return 0;
2912}
2913
2914static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2915{
0e42c1f4 2916 struct cpl_rdma_terminate *rpl = cplhdr(skb);
0e42c1f4
SW		 2917 unsigned int tid = GET_TID(rpl);
2918 struct c4iw_ep *ep;
2919 struct c4iw_qp_attributes attrs;
cfdda9d7 2920
944661dd 2921 ep = get_ep_from_tid(dev, tid);
0e42c1f4 2922 BUG_ON(!ep);
cfdda9d7 2923
30c95c2d 2924 if (ep && ep->com.qp) {
700456bd
JP		 2925 pr_warn("TERM received tid %u qpid %u\n",
2926 tid, ep->com.qp->wq.sq.qid);
0e42c1f4
SW		 2927 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2928 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2929 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2930 } else
700456bd 2931 pr_warn("TERM received tid %u no ep/qp\n", tid);
944661dd 2932 c4iw_put_ep(&ep->com);
cfdda9d7 2933
cfdda9d7
SW		 2934 return 0;
2935}
2936
2937/*
2938 * Upcall from the adapter indicating data has been transmitted.
2939 * For us its just the single MPA request or reply. We can now free
2940 * the skb holding the mpa message.
2941 */
2942static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2943{
2944 struct c4iw_ep *ep;
2945 struct cpl_fw4_ack *hdr = cplhdr(skb);
2946 u8 credits = hdr->credits;
2947 unsigned int tid = GET_TID(hdr);
cfdda9d7
SW		 2948
2949
944661dd
H		 2950 ep = get_ep_from_tid(dev, tid);
2951 if (!ep)
2952 return 0;
a9a42886
JP		 2953 pr_debug("%s ep %p tid %u credits %u\n",
2954 __func__, ep, ep->hwtid, credits);
cfdda9d7 2955 if (credits == 0) {
a9a42886
JP		 2956 pr_debug("%s 0 credit ack ep %p tid %u state %u\n",
2957 __func__, ep, ep->hwtid, state_read(&ep->com));
944661dd 2958 goto out;
cfdda9d7
SW		 2959 }
2960
2961 dst_confirm(ep->dst);
2962 if (ep->mpa_skb) {
a9a42886
JP		 2963 pr_debug("%s last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
2964 __func__, ep, ep->hwtid,
2965 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
12eb5137 2966 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2967 kfree_skb(ep->mpa_skb);
2968 ep->mpa_skb = NULL;
e4b76a2a
H		 2969 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2970 stop_ep_timer(ep);
2971 mutex_unlock(&ep->com.mutex);
cfdda9d7 2972 }
944661dd
H		 2973out:
2974 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2975 return 0;
2976}
2977
cfdda9d7
SW		 2978int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2979{
bce2841f 2980 int abort;
cfdda9d7 2981 struct c4iw_ep *ep = to_ep(cm_id);
bce2841f 2982
a9a42886 2983 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cfdda9d7 2984
a7db89eb 2985 mutex_lock(&ep->com.mutex);
e8667a9b 2986 if (ep->com.state != MPA_REQ_RCVD) {
a7db89eb 2987 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2988 c4iw_put_ep(&ep->com);
2989 return -ECONNRESET;
2990 }
793dad94 2991 set_bit(ULP_REJECT, &ep->com.history);
cfdda9d7 2992 if (mpa_rev == 0)
bce2841f
H		 2993 abort = 1;
2994 else
2995 abort = send_mpa_reject(ep, pdata, pdata_len);
a7db89eb 2996 mutex_unlock(&ep->com.mutex);
bce2841f
H		 2997
2998 stop_ep_timer(ep);
2999 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
cfdda9d7
SW		 3000 c4iw_put_ep(&ep->com);
3001 return 0;
3002}
3003
3004int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3005{
3006 int err;
3007 struct c4iw_qp_attributes attrs;
3008 enum c4iw_qp_attr_mask mask;
3009 struct c4iw_ep *ep = to_ep(cm_id);
3010 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3011 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
eaf4c6d4 3012 int abort = 0;
cfdda9d7 3013
a9a42886 3014 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
a7db89eb
SW		 3015
3016 mutex_lock(&ep->com.mutex);
e8667a9b 3017 if (ep->com.state != MPA_REQ_RCVD) {
cfdda9d7 3018 err = -ECONNRESET;
eaf4c6d4 3019 goto err_out;
cfdda9d7
SW		 3020 }
3021
cfdda9d7
SW		 3022 BUG_ON(!qp);
3023
793dad94 3024 set_bit(ULP_ACCEPT, &ep->com.history);
4c2c5763
HS		 3025 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3026 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
cfdda9d7 3027 err = -EINVAL;
eaf4c6d4 3028 goto err_abort;
cfdda9d7
SW		 3029 }
3030
d2fe99e8
KS		 3031 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3032 if (conn_param->ord > ep->ird) {
4c2c5763 3033 if (RELAXED_IRD_NEGOTIATION) {
30b03b15 3034 conn_param->ord = ep->ird;
4c2c5763
HS		 3035 } else {
3036 ep->ird = conn_param->ird;
3037 ep->ord = conn_param->ord;
3038 send_mpa_reject(ep, conn_param->private_data,
3039 conn_param->private_data_len);
4c2c5763 3040 err = -ENOMEM;
eaf4c6d4 3041 goto err_abort;
4c2c5763 3042 }
d2fe99e8 3043 }
4c2c5763
HS		 3044 if (conn_param->ird < ep->ord) {
3045 if (RELAXED_IRD_NEGOTIATION &&
3046 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3047 conn_param->ird = ep->ord;
3048 } else {
d2fe99e8 3049 err = -ENOMEM;
eaf4c6d4 3050 goto err_abort;
d2fe99e8
KS		 3051 }
3052 }
d2fe99e8 3053 }
cfdda9d7
SW		 3054 ep->ird = conn_param->ird;
3055 ep->ord = conn_param->ord;
3056
4c2c5763 3057 if (ep->mpa_attr.version == 1) {
d2fe99e8
KS		 3058 if (peer2peer && ep->ird == 0)
3059 ep->ird = 1;
4c2c5763
HS		 3060 } else {
3061 if (peer2peer &&
3062 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
f57b780c 3063 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
4c2c5763
HS		 3064 ep->ird = 1;
3065 }
cfdda9d7 3066
a9a42886 3067 pr_debug("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
cfdda9d7 3068
d2fe99e8 3069 ep->com.cm_id = cm_id;
9ca6f7cf 3070 ref_cm_id(&ep->com);
d2fe99e8 3071 ep->com.qp = qp;
325abead 3072 ref_qp(ep);
d2fe99e8 3073
cfdda9d7
SW		 3074 /* bind QP to EP and move to RTS */
3075 attrs.mpa_attr = ep->mpa_attr;
3076 attrs.max_ird = ep->ird;
3077 attrs.max_ord = ep->ord;
3078 attrs.llp_stream_handle = ep;
3079 attrs.next_state = C4IW_QP_STATE_RTS;
3080
3081 /* bind QP and TID with INIT_WR */
3082 mask = C4IW_QP_ATTR_NEXT_STATE |
3083 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3084 C4IW_QP_ATTR_MPA_ATTR |
3085 C4IW_QP_ATTR_MAX_IRD |
3086 C4IW_QP_ATTR_MAX_ORD;
3087
3088 err = c4iw_modify_qp(ep->com.qp->rhp,
3089 ep->com.qp, mask, &attrs, 1);
3090 if (err)
eaf4c6d4 3091 goto err_deref_cm_id;
e4b76a2a
H		 3092
3093 set_bit(STOP_MPA_TIMER, &ep->com.flags);
cfdda9d7
SW		 3094 err = send_mpa_reply(ep, conn_param->private_data,
3095 conn_param->private_data_len);
3096 if (err)
eaf4c6d4 3097 goto err_deref_cm_id;
cfdda9d7 3098
a7db89eb 3099 __state_set(&ep->com, FPDU_MODE);
cfdda9d7 3100 established_upcall(ep);
a7db89eb 3101 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 3102 c4iw_put_ep(&ep->com);
3103 return 0;
eaf4c6d4 3104err_deref_cm_id:
9ca6f7cf 3105 deref_cm_id(&ep->com);
eaf4c6d4
H		 3106err_abort:
3107 abort = 1;
3108err_out:
a7db89eb 3109 mutex_unlock(&ep->com.mutex);
eaf4c6d4
H		 3110 if (abort)
3111 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
cfdda9d7
SW		 3112 c4iw_put_ep(&ep->com);
3113 return err;
3114}
3115
830662f6
VP		 3116static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3117{
3118 struct in_device *ind;
3119 int found = 0;
170003c8
SW		 3120 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3121 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
830662f6
VP		 3122
3123 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3124 if (!ind)
3125 return -EADDRNOTAVAIL;
3126 for_primary_ifa(ind) {
3127 laddr->sin_addr.s_addr = ifa->ifa_address;
3128 raddr->sin_addr.s_addr = ifa->ifa_address;
3129 found = 1;
3130 break;
3131 }
3132 endfor_ifa(ind);
3133 in_dev_put(ind);
3134 return found ? 0 : -EADDRNOTAVAIL;
3135}
3136
3137static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3138 unsigned char banned_flags)
3139{
3140 struct inet6_dev *idev;
3141 int err = -EADDRNOTAVAIL;
3142
3143 rcu_read_lock();
3144 idev = __in6_dev_get(dev);
3145 if (idev != NULL) {
3146 struct inet6_ifaddr *ifp;
3147
3148 read_lock_bh(&idev->lock);
3149 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3150 if (ifp->scope == IFA_LINK &&
3151 !(ifp->flags & banned_flags)) {
3152 memcpy(addr, &ifp->addr, 16);
3153 err = 0;
3154 break;
3155 }
3156 }
3157 read_unlock_bh(&idev->lock);
3158 }
3159 rcu_read_unlock();
3160 return err;
3161}
3162
3163static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3164{
3165 struct in6_addr uninitialized_var(addr);
170003c8
SW		 3166 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3167 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
830662f6 3168
54b9a96f 3169 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
830662f6
VP		 3170 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3171 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3172 return 0;
3173 }
3174 return -EADDRNOTAVAIL;
3175}
3176
cfdda9d7
SW		 3177int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3178{
cfdda9d7
SW		 3179 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3180 struct c4iw_ep *ep;
3786cf18 3181 int err = 0;
9eccfe10
SW		 3182 struct sockaddr_in *laddr;
3183 struct sockaddr_in *raddr;
3184 struct sockaddr_in6 *laddr6;
3185 struct sockaddr_in6 *raddr6;
830662f6
VP		 3186 __u8 *ra;
3187 int iptype;
cfdda9d7 3188
4c2c5763
HS		 3189 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3190 (conn_param->ird > cur_max_read_depth(dev))) {
be4c9bad
RD		 3191 err = -EINVAL;
3192 goto out;
3193 }
cfdda9d7
SW		 3194 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3195 if (!ep) {
700456bd 3196 pr_err("%s - cannot alloc ep\n", __func__);
cfdda9d7
SW		 3197 err = -ENOMEM;
3198 goto out;
3199 }
4a740838
H		 3200
3201 skb_queue_head_init(&ep->com.ep_skb_list);
3202 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3203 err = -ENOMEM;
3204 goto fail1;
3205 }
3206
cfdda9d7
SW		 3207 init_timer(&ep->timer);
3208 ep->plen = conn_param->private_data_len;
3209 if (ep->plen)
3210 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3211 conn_param->private_data, ep->plen);
3212 ep->ird = conn_param->ird;
3213 ep->ord = conn_param->ord;
3214
3215 if (peer2peer && ep->ord == 0)
3216 ep->ord = 1;
3217
cfdda9d7 3218 ep->com.cm_id = cm_id;
9ca6f7cf
H		 3219 ref_cm_id(&ep->com);
3220 ep->com.dev = dev;
cfdda9d7 3221 ep->com.qp = get_qhp(dev, conn_param->qpn);
830662f6 3222 if (!ep->com.qp) {
a9a42886 3223 pr_debug("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
830662f6 3224 err = -EINVAL;
4a740838 3225 goto fail2;
830662f6 3226 }
325abead 3227 ref_qp(ep);
a9a42886
JP		 3228 pr_debug("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3229 ep->com.qp, cm_id);
cfdda9d7
SW		 3230
3231 /*
3232 * Allocate an active TID to initiate a TCP connection.
3233 */
3234 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3235 if (ep->atid == -1) {
700456bd 3236 pr_err("%s - cannot alloc atid\n", __func__);
cfdda9d7 3237 err = -ENOMEM;
4a740838 3238 goto fail2;
cfdda9d7 3239 }
793dad94 3240 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
cfdda9d7 3241
170003c8 3242 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
9eccfe10 3243 sizeof(ep->com.local_addr));
170003c8 3244 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
9eccfe10
SW		 3245 sizeof(ep->com.remote_addr));
3246
170003c8
SW		 3247 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3248 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3249 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3250 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
9eccfe10 3251
170003c8 3252 if (cm_id->m_remote_addr.ss_family == AF_INET) {
830662f6
VP		 3253 iptype = 4;
3254 ra = (__u8 *)&raddr->sin_addr;
cfdda9d7 3255
830662f6
VP		 3256 /*
3257 * Handle loopback requests to INADDR_ANY.
3258 */
ba987e51 3259 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
830662f6
VP		 3260 err = pick_local_ipaddrs(dev, cm_id);
3261 if (err)
4a740838 3262 goto fail2;
830662f6
VP		 3263 }
3264
3265 /* find a route */
a9a42886
JP		 3266 pr_debug("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3267 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3268 ra, ntohs(raddr->sin_port));
804c2f3e
VP		 3269 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3270 laddr->sin_addr.s_addr,
3271 raddr->sin_addr.s_addr,
3272 laddr->sin_port,
3273 raddr->sin_port, cm_id->tos);
830662f6
VP		 3274 } else {
3275 iptype = 6;
3276 ra = (__u8 *)&raddr6->sin6_addr;
3277
3278 /*
3279 * Handle loopback requests to INADDR_ANY.
3280 */
3281 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3282 err = pick_local_ip6addrs(dev, cm_id);
3283 if (err)
4a740838 3284 goto fail2;
830662f6
VP		 3285 }
3286
3287 /* find a route */
a9a42886
JP		 3288 pr_debug("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3289 __func__, laddr6->sin6_addr.s6_addr,
3290 ntohs(laddr6->sin6_port),
3291 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
95554761
VP		 3292 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3293 laddr6->sin6_addr.s6_addr,
3294 raddr6->sin6_addr.s6_addr,
3295 laddr6->sin6_port,
3296 raddr6->sin6_port, 0,
3297 raddr6->sin6_scope_id);
830662f6
VP		 3298 }
3299 if (!ep->dst) {
700456bd 3300 pr_err("%s - cannot find route\n", __func__);
cfdda9d7 3301 err = -EHOSTUNREACH;
4a740838 3302 goto fail3;
cfdda9d7 3303 }
cfdda9d7 3304
963cab50 3305 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
ac8e4c69 3306 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3786cf18 3307 if (err) {
700456bd 3308 pr_err("%s - cannot alloc l2e\n", __func__);
4a740838 3309 goto fail4;
cfdda9d7
SW		 3310 }
3311
a9a42886
JP		 3312 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3313 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3314 ep->l2t->idx);
cfdda9d7
SW		 3315
3316 state_set(&ep->com, CONNECTING);
ac8e4c69 3317 ep->tos = cm_id->tos;
cfdda9d7
SW		 3318
3319 /* send connect request to rnic */
3320 err = send_connect(ep);
3321 if (!err)
3322 goto out;
3323
3324 cxgb4_l2t_release(ep->l2t);
4a740838 3325fail4:
9eccfe10 3326 dst_release(ep->dst);
4a740838 3327fail3:
793dad94 3328 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
cfdda9d7 3329 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
4a740838
H		 3330fail2:
3331 skb_queue_purge(&ep->com.ep_skb_list);
9ca6f7cf 3332 deref_cm_id(&ep->com);
4a740838 3333fail1:
cfdda9d7
SW		 3334 c4iw_put_ep(&ep->com);
3335out:
3336 return err;
3337}
3338
830662f6
VP		 3339static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3340{
3341 int err;
9eccfe10 3342 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
170003c8 3343 &ep->com.local_addr;
830662f6 3344
28de1f74
H		 3345 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3346 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3347 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3348 if (err)
3349 return err;
3350 }
830662f6
VP		 3351 c4iw_init_wr_wait(&ep->com.wr_wait);
3352 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3353 ep->stid, &sin6->sin6_addr,
3354 sin6->sin6_port,
3355 ep->com.dev->rdev.lldi.rxq_ids[0]);
3356 if (!err)
3357 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3358 &ep->com.wr_wait,
3359 0, 0, __func__);
e6b11163
H		 3360 else if (err > 0)
3361 err = net_xmit_errno(err);
28de1f74
H		 3362 if (err) {
3363 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3364 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
830662f6
VP		 3365 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3366 err, ep->stid,
3367 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
28de1f74 3368 }
830662f6
VP		 3369 return err;
3370}
3371
3372static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3373{
3374 int err;
9eccfe10 3375 struct sockaddr_in *sin = (struct sockaddr_in *)
170003c8 3376 &ep->com.local_addr;
830662f6
VP		 3377
3378 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3379 do {
3380 err = cxgb4_create_server_filter(
3381 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3382 sin->sin_addr.s_addr, sin->sin_port, 0,
3383 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3384 if (err == -EBUSY) {
99718e59
H		 3385 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3386 err = -EIO;
3387 break;
3388 }
830662f6
VP		 3389 set_current_state(TASK_UNINTERRUPTIBLE);
3390 schedule_timeout(usecs_to_jiffies(100));
3391 }
3392 } while (err == -EBUSY);
3393 } else {
3394 c4iw_init_wr_wait(&ep->com.wr_wait);
3395 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3396 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3397 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3398 if (!err)
3399 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3400 &ep->com.wr_wait,
3401 0, 0, __func__);
e6b11163
H		 3402 else if (err > 0)
3403 err = net_xmit_errno(err);
830662f6
VP		 3404 }
3405 if (err)
3406 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3407 , err, ep->stid,
3408 &sin->sin_addr, ntohs(sin->sin_port));
3409 return err;
3410}
3411
cfdda9d7
SW		 3412int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3413{
3414 int err = 0;
3415 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3416 struct c4iw_listen_ep *ep;
3417
cfdda9d7
SW		 3418 might_sleep();
3419
3420 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3421 if (!ep) {
700456bd 3422 pr_err("%s - cannot alloc ep\n", __func__);
cfdda9d7
SW		 3423 err = -ENOMEM;
3424 goto fail1;
3425 }
4a740838 3426 skb_queue_head_init(&ep->com.ep_skb_list);
a9a42886 3427 pr_debug("%s ep %p\n", __func__, ep);
cfdda9d7 3428 ep->com.cm_id = cm_id;
9ca6f7cf 3429 ref_cm_id(&ep->com);
cfdda9d7
SW		 3430 ep->com.dev = dev;
3431 ep->backlog = backlog;
170003c8 3432 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
24d44a39 3433 sizeof(ep->com.local_addr));
cfdda9d7
SW		 3434
3435 /*
3436 * Allocate a server TID.
3437 */
8c044690
KS		 3438 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3439 ep->com.local_addr.ss_family == AF_INET)
830662f6 3440 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
170003c8 3441 cm_id->m_local_addr.ss_family, ep);
1cab775c 3442 else
830662f6 3443 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
170003c8 3444 cm_id->m_local_addr.ss_family, ep);
1cab775c 3445
cfdda9d7 3446 if (ep->stid == -1) {
700456bd 3447 pr_err("%s - cannot alloc stid\n", __func__);
cfdda9d7
SW		 3448 err = -ENOMEM;
3449 goto fail2;
3450 }
793dad94 3451 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
9eccfe10 3452
170003c8
SW		 3453 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3454 sizeof(ep->com.local_addr));
9eccfe10 3455
cfdda9d7 3456 state_set(&ep->com, LISTEN);
830662f6
VP		 3457 if (ep->com.local_addr.ss_family == AF_INET)
3458 err = create_server4(dev, ep);
3459 else
3460 err = create_server6(dev, ep);
cfdda9d7
SW		 3461 if (!err) {
3462 cm_id->provider_data = ep;
3463 goto out;
3464 }
8b1bbf36 3465 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
830662f6
VP		 3466 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3467 ep->com.local_addr.ss_family);
cfdda9d7 3468fail2:
9ca6f7cf 3469 deref_cm_id(&ep->com);
cfdda9d7
SW		 3470 c4iw_put_ep(&ep->com);
3471fail1:
3472out:
3473 return err;
3474}
3475
3476int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3477{
3478 int err;
3479 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3480
a9a42886 3481 pr_debug("%s ep %p\n", __func__, ep);
cfdda9d7
SW		 3482
3483 might_sleep();
3484 state_set(&ep->com, DEAD);
830662f6
VP		 3485 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3486 ep->com.local_addr.ss_family == AF_INET) {
1cab775c
VP		 3487 err = cxgb4_remove_server_filter(
3488 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3489 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3490 } else {
84cc6ac6 3491 struct sockaddr_in6 *sin6;
1cab775c 3492 c4iw_init_wr_wait(&ep->com.wr_wait);
830662f6
VP		 3493 err = cxgb4_remove_server(
3494 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3495 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
1cab775c
VP		 3496 if (err)
3497 goto done;
3498 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3499 0, 0, __func__);
170003c8 3500 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 3501 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3502 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
1cab775c 3503 }
793dad94 3504 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
830662f6
VP		 3505 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3506 ep->com.local_addr.ss_family);
cfdda9d7 3507done:
9ca6f7cf 3508 deref_cm_id(&ep->com);
cfdda9d7
SW		 3509 c4iw_put_ep(&ep->com);
3510 return err;
3511}
3512
3513int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3514{
3515 int ret = 0;
cfdda9d7
SW		 3516 int close = 0;
3517 int fatal = 0;
3518 struct c4iw_rdev *rdev;
cfdda9d7 3519
2f5b48c3 3520 mutex_lock(&ep->com.mutex);
cfdda9d7 3521
a9a42886
JP		 3522 pr_debug("%s ep %p state %s, abrupt %d\n", __func__, ep,
3523 states[ep->com.state], abrupt);
cfdda9d7 3524
6e410d8f
H		 3525 /*
3526 * Ref the ep here in case we have fatal errors causing the
3527 * ep to be released and freed.
3528 */
3529 c4iw_get_ep(&ep->com);
3530
cfdda9d7
SW		 3531 rdev = &ep->com.dev->rdev;
3532 if (c4iw_fatal_error(rdev)) {
3533 fatal = 1;
be13b2df 3534 close_complete_upcall(ep, -EIO);
cfdda9d7
SW		 3535 ep->com.state = DEAD;
3536 }
3537 switch (ep->com.state) {
3538 case MPA_REQ_WAIT:
3539 case MPA_REQ_SENT:
3540 case MPA_REQ_RCVD:
3541 case MPA_REP_SENT:
3542 case FPDU_MODE:
4a740838 3543 case CONNECTING:
cfdda9d7
SW		 3544 close = 1;
3545 if (abrupt)
3546 ep->com.state = ABORTING;
3547 else {
3548 ep->com.state = CLOSING;
12eb5137
SW		 3549
3550 /*
3551 * if we close before we see the fw4_ack() then we fix
3552 * up the timer state since we're reusing it.
3553 */
3554 if (ep->mpa_skb &&
3555 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3556 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3557 stop_ep_timer(ep);
3558 }
ca5a2202 3559 start_ep_timer(ep);
cfdda9d7
SW		 3560 }
3561 set_bit(CLOSE_SENT, &ep->com.flags);
3562 break;
3563 case CLOSING:
3564 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3565 close = 1;
3566 if (abrupt) {
b33bd0cb 3567 (void)stop_ep_timer(ep);
cfdda9d7
SW		 3568 ep->com.state = ABORTING;
3569 } else
3570 ep->com.state = MORIBUND;
3571 }
3572 break;
3573 case MORIBUND:
3574 case ABORTING:
3575 case DEAD:
a9a42886
JP		 3576 pr_debug("%s ignoring disconnect ep %p state %u\n",
3577 __func__, ep, ep->com.state);
cfdda9d7
SW		 3578 break;
3579 default:
3580 BUG();
3581 break;
3582 }
3583
cfdda9d7 3584 if (close) {
8da7e7a5 3585 if (abrupt) {
793dad94 3586 set_bit(EP_DISC_ABORT, &ep->com.history);
be13b2df 3587 close_complete_upcall(ep, -ECONNRESET);
4a740838 3588 ret = send_abort(ep);
793dad94
VP		 3589 } else {
3590 set_bit(EP_DISC_CLOSE, &ep->com.history);
4a740838 3591 ret = send_halfclose(ep);
793dad94 3592 }
88bc230d 3593 if (ret) {
9ca6f7cf 3594 set_bit(EP_DISC_FAIL, &ep->com.history);
88bc230d
H		 3595 if (!abrupt) {
3596 stop_ep_timer(ep);
3597 close_complete_upcall(ep, -EIO);
3598 }
c00dcbaf
H		 3599 if (ep->com.qp) {
3600 struct c4iw_qp_attributes attrs;
3601
3602 attrs.next_state = C4IW_QP_STATE_ERROR;
3603 ret = c4iw_modify_qp(ep->com.qp->rhp,
3604 ep->com.qp,
3605 C4IW_QP_ATTR_NEXT_STATE,
3606 &attrs, 1);
3607 if (ret)
700456bd 3608 pr_err("%s - qp <- error failed!\n",
c00dcbaf
H		 3609 __func__);
3610 }
cfdda9d7 3611 fatal = 1;
88bc230d 3612 }
cfdda9d7 3613 }
8da7e7a5 3614 mutex_unlock(&ep->com.mutex);
6e410d8f 3615 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 3616 if (fatal)
3617 release_ep_resources(ep);
3618 return ret;
3619}
3620
1cab775c
VP		 3621static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3622 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3623{
3624 struct c4iw_ep *ep;
793dad94 3625 int atid = be32_to_cpu(req->tid);
1cab775c 3626
ef5d6355
VP		 3627 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3628 (__force u32) req->tid);
1cab775c
VP		 3629 if (!ep)
3630 return;
3631
3632 switch (req->retval) {
3633 case FW_ENOMEM:
793dad94
VP		 3634 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3635 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3636 send_fw_act_open_req(ep, atid);
3637 return;
3638 }
1cab775c 3639 case FW_EADDRINUSE:
793dad94
VP		 3640 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3641 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3642 send_fw_act_open_req(ep, atid);
3643 return;
3644 }
1cab775c
VP		 3645 break;
3646 default:
3647 pr_info("%s unexpected ofld conn wr retval %d\n",
3648 __func__, req->retval);
3649 break;
3650 }
793dad94
VP		 3651 pr_err("active ofld_connect_wr failure %d atid %d\n",
3652 req->retval, atid);
3653 mutex_lock(&dev->rdev.stats.lock);
3654 dev->rdev.stats.act_ofld_conn_fails++;
3655 mutex_unlock(&dev->rdev.stats.lock);
1cab775c 3656 connect_reply_upcall(ep, status2errno(req->retval));
793dad94 3657 state_set(&ep->com, DEAD);
84cc6ac6
H		 3658 if (ep->com.remote_addr.ss_family == AF_INET6) {
3659 struct sockaddr_in6 *sin6 =
170003c8 3660 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 3661 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3662 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3663 }
793dad94
VP		 3664 remove_handle(dev, &dev->atid_idr, atid);
3665 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3666 dst_release(ep->dst);
3667 cxgb4_l2t_release(ep->l2t);
3668 c4iw_put_ep(&ep->com);
1cab775c
VP		 3669}
3670
3671static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3672 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3673{
3674 struct sk_buff *rpl_skb;
3675 struct cpl_pass_accept_req *cpl;
3676 int ret;
3677
710a3110 3678 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
1cab775c
VP		 3679 BUG_ON(!rpl_skb);
3680 if (req->retval) {
a9a42886 3681 pr_debug("%s passive open failure %d\n", __func__, req->retval);
793dad94
VP		 3682 mutex_lock(&dev->rdev.stats.lock);
3683 dev->rdev.stats.pas_ofld_conn_fails++;
3684 mutex_unlock(&dev->rdev.stats.lock);
1cab775c
VP		 3685 kfree_skb(rpl_skb);
3686 } else {
3687 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3688 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
ef5d6355
VP		 3689 (__force u32) htonl(
3690 (__force u32) req->tid)));
1cab775c
VP		 3691 ret = pass_accept_req(dev, rpl_skb);
3692 if (!ret)
3693 kfree_skb(rpl_skb);
3694 }
3695 return;
3696}
3697
3698static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2f5b48c3
SW		 3699{
3700 struct cpl_fw6_msg *rpl = cplhdr(skb);
1cab775c
VP		 3701 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3702
3703 switch (rpl->type) {
3704 case FW6_TYPE_CQE:
3705 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3706 break;
3707 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3708 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3709 switch (req->t_state) {
3710 case TCP_SYN_SENT:
3711 active_ofld_conn_reply(dev, skb, req);
3712 break;
3713 case TCP_SYN_RECV:
3714 passive_ofld_conn_reply(dev, skb, req);
3715 break;
3716 default:
3717 pr_err("%s unexpected ofld conn wr state %d\n",
3718 __func__, req->t_state);
3719 break;
3720 }
3721 break;
3722 }
3723 return 0;
3724}
3725
3726static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3727{
963cab50
H		 3728 __be32 l2info;
3729 __be16 hdr_len, vlantag, len;
3730 u16 eth_hdr_len;
3731 int tcp_hdr_len, ip_hdr_len;
1cab775c
VP		 3732 u8 intf;
3733 struct cpl_rx_pkt *cpl = cplhdr(skb);
3734 struct cpl_pass_accept_req *req;
3735 struct tcp_options_received tmp_opt;
f079af7a 3736 struct c4iw_dev *dev;
963cab50 3737 enum chip_type type;
1cab775c 3738
f079af7a 3739 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
1cab775c 3740 /* Store values from cpl_rx_pkt in temporary location. */
963cab50
H		 3741 vlantag = cpl->vlan;
3742 len = cpl->len;
3743 l2info = cpl->l2info;
3744 hdr_len = cpl->hdr_len;
1cab775c
VP		 3745 intf = cpl->iff;
3746
3747 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3748
3749 /*
3750 * We need to parse the TCP options from SYN packet.
3751 * to generate cpl_pass_accept_req.
3752 */
3753 memset(&tmp_opt, 0, sizeof(tmp_opt));
3754 tcp_clear_options(&tmp_opt);
eed29f17 3755 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
1cab775c 3756
d58ff351 3757 req = __skb_push(skb, sizeof(*req));
1cab775c 3758 memset(req, 0, sizeof(*req));
cf7fe64a
HS		 3759 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3760 SYN_MAC_IDX_V(RX_MACIDX_G(
963cab50 3761 be32_to_cpu(l2info))) |
cf7fe64a 3762 SYN_XACT_MATCH_F);
963cab50
H		 3763 type = dev->rdev.lldi.adapter_type;
3764 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3765 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3766 req->hdr_len =
3767 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3768 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3769 eth_hdr_len = is_t4(type) ?
3770 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3771 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3772 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3773 IP_HDR_LEN_V(ip_hdr_len) |
3774 ETH_HDR_LEN_V(eth_hdr_len));
3775 } else { /* T6 and later */
3776 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3777 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3778 T6_IP_HDR_LEN_V(ip_hdr_len) |
3779 T6_ETH_HDR_LEN_V(eth_hdr_len));
3780 }
3781 req->vlan = vlantag;
3782 req->len = len;
6c53e938
HS		 3783 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3784 PASS_OPEN_TOS_V(tos));
1cab775c
VP		 3785 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3786 if (tmp_opt.wscale_ok)
3787 req->tcpopt.wsf = tmp_opt.snd_wscale;
3788 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3789 if (tmp_opt.sack_ok)
3790 req->tcpopt.sack = 1;
3791 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3792 return;
3793}
3794
3795static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3796 __be32 laddr, __be16 lport,
3797 __be32 raddr, __be16 rport,
3798 u32 rcv_isn, u32 filter, u16 window,
3799 u32 rss_qid, u8 port_id)
3800{
3801 struct sk_buff *req_skb;
3802 struct fw_ofld_connection_wr *req;
3803 struct cpl_pass_accept_req *cpl = cplhdr(skb);
1ce1d471 3804 int ret;
1cab775c
VP		 3805
3806 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
9ef63f31
PB		 3807 if (!req_skb)
3808 return;
de77b966 3809 req = __skb_put_zero(req_skb, sizeof(*req));
6c53e938 3810 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
e2ac9628 3811 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
77a80e23 3812 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
ef5d6355 3813 req->le.filter = (__force __be32) filter;
1cab775c
VP		 3814 req->le.lport = lport;
3815 req->le.pport = rport;
3816 req->le.u.ipv4.lip = laddr;
3817 req->le.u.ipv4.pip = raddr;
3818 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3819 req->tcb.rcv_adv = htons(window);
3820 req->tcb.t_state_to_astid =
77a80e23
HS		 3821 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3822 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3823 FW_OFLD_CONNECTION_WR_ASTID_V(
6c53e938 3824 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
1cab775c
VP		 3825
3826 /*
3827 * We store the qid in opt2 which will be used by the firmware
3828 * to send us the wr response.
3829 */
d7990b0c 3830 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
1cab775c
VP		 3831
3832 /*
3833 * We initialize the MSS index in TCB to 0xF.
3834 * So that when driver sends cpl_pass_accept_rpl
3835 * TCB picks up the correct value. If this was 0
3836 * TP will ignore any value > 0 for MSS index.
3837 */
d7990b0c 3838 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
6198dd8d 3839 req->cookie = (uintptr_t)skb;
1cab775c
VP		 3840
3841 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
1ce1d471
SW		 3842 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3843 if (ret < 0) {
3844 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3845 ret);
3846 kfree_skb(skb);
3847 kfree_skb(req_skb);
3848 }
1cab775c
VP		 3849}
3850
3851/*
3852 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3853 * messages when a filter is being used instead of server to
3854 * redirect a syn packet. When packets hit filter they are redirected
3855 * to the offload queue and driver tries to establish the connection
3856 * using firmware work request.
3857 */
3858static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3859{
3860 int stid;
3861 unsigned int filter;
3862 struct ethhdr *eh = NULL;
3863 struct vlan_ethhdr *vlan_eh = NULL;
3864 struct iphdr *iph;
3865 struct tcphdr *tcph;
3866 struct rss_header *rss = (void *)skb->data;
3867 struct cpl_rx_pkt *cpl = (void *)skb->data;
3868 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3869 struct l2t_entry *e;
3870 struct dst_entry *dst;
f86fac79 3871 struct c4iw_ep *lep = NULL;
1cab775c
VP		 3872 u16 window;
3873 struct port_info *pi;
3874 struct net_device *pdev;
f079af7a 3875 u16 rss_qid, eth_hdr_len;
1cab775c
VP		 3876 int step;
3877 u32 tx_chan;
3878 struct neighbour *neigh;
3879
3880 /* Drop all non-SYN packets */
bdc590b9 3881 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
1cab775c
VP		 3882 goto reject;
3883
3884 /*
3885 * Drop all packets which did not hit the filter.
3886 * Unlikely to happen.
3887 */
3888 if (!(rss->filter_hit && rss->filter_tid))
3889 goto reject;
3890
3891 /*
3892 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3893 */
a4ea025f 3894 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
1cab775c 3895
f86fac79 3896 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
1cab775c 3897 if (!lep) {
a9a42886
JP		 3898 pr_debug("%s connect request on invalid stid %d\n",
3899 __func__, stid);
1cab775c
VP		 3900 goto reject;
3901 }
3902
963cab50
H		 3903 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3904 case CHELSIO_T4:
3905 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3906 break;
3907 case CHELSIO_T5:
3908 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3909 break;
3910 case CHELSIO_T6:
3911 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3912 break;
3913 default:
3914 pr_err("T%d Chip is not supported\n",
3915 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3916 goto reject;
3917 }
3918
f079af7a 3919 if (eth_hdr_len == ETH_HLEN) {
1cab775c
VP		 3920 eh = (struct ethhdr *)(req + 1);
3921 iph = (struct iphdr *)(eh + 1);
3922 } else {
3923 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3924 iph = (struct iphdr *)(vlan_eh + 1);
3925 skb->vlan_tci = ntohs(cpl->vlan);
3926 }
3927
3928 if (iph->version != 0x4)
3929 goto reject;
3930
3931 tcph = (struct tcphdr *)(iph + 1);
3932 skb_set_network_header(skb, (void *)iph - (void *)rss);
3933 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3934 skb_get(skb);
3935
a9a42886
JP		 3936 pr_debug("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3937 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3938 ntohs(tcph->source), iph->tos);
1cab775c 3939
804c2f3e
VP		 3940 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3941 iph->daddr, iph->saddr, tcph->dest,
3942 tcph->source, iph->tos);
830662f6 3943 if (!dst) {
1cab775c
VP		 3944 pr_err("%s - failed to find dst entry!\n",
3945 __func__);
3946 goto reject;
3947 }
1cab775c
VP		 3948 neigh = dst_neigh_lookup_skb(dst, skb);
3949
aaa0c23c
ZZ		 3950 if (!neigh) {
3951 pr_err("%s - failed to allocate neigh!\n",
3952 __func__);
3953 goto free_dst;
3954 }
3955
1cab775c
VP		 3956 if (neigh->dev->flags & IFF_LOOPBACK) {
3957 pdev = ip_dev_find(&init_net, iph->daddr);
3958 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3959 pdev, 0);
3960 pi = (struct port_info *)netdev_priv(pdev);
3961 tx_chan = cxgb4_port_chan(pdev);
3962 dev_put(pdev);
3963 } else {
830662f6 3964 pdev = get_real_dev(neigh->dev);
1cab775c 3965 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
830662f6
VP		 3966 pdev, 0);
3967 pi = (struct port_info *)netdev_priv(pdev);
3968 tx_chan = cxgb4_port_chan(pdev);
1cab775c 3969 }
ebf00060 3970 neigh_release(neigh);
1cab775c
VP		 3971 if (!e) {
3972 pr_err("%s - failed to allocate l2t entry!\n",
3973 __func__);
3974 goto free_dst;
3975 }
3976
3977 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3978 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
ef5d6355 3979 window = (__force u16) htons((__force u16)tcph->window);
1cab775c
VP		 3980
3981 /* Calcuate filter portion for LE region. */
41b4f86c
KS		 3982 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3983 dev->rdev.lldi.ports[0],
3984 e));
1cab775c
VP		 3985
3986 /*
3987 * Synthesize the cpl_pass_accept_req. We have everything except the
3988 * TID. Once firmware sends a reply with TID we update the TID field
3989 * in cpl and pass it through the regular cpl_pass_accept_req path.
3990 */
3991 build_cpl_pass_accept_req(skb, stid, iph->tos);
3992 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3993 tcph->source, ntohl(tcph->seq), filter, window,
3994 rss_qid, pi->port_id);
3995 cxgb4_l2t_release(e);
3996free_dst:
3997 dst_release(dst);
3998reject:
f86fac79
H		 3999 if (lep)
4000 c4iw_put_ep(&lep->com);
2f5b48c3
SW		 4001 return 0;
4002}
4003
be4c9bad
RD		 4004/*
4005 * These are the real handlers that are called from a
4006 * work queue.
4007 */
9dec900c 4008static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
be4c9bad
RD		 4009 [CPL_ACT_ESTABLISH] = act_establish,
4010 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4011 [CPL_RX_DATA] = rx_data,
4012 [CPL_ABORT_RPL_RSS] = abort_rpl,
4013 [CPL_ABORT_RPL] = abort_rpl,
4014 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4015 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4016 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4017 [CPL_PASS_ESTABLISH] = pass_establish,
4018 [CPL_PEER_CLOSE] = peer_close,
4019 [CPL_ABORT_REQ_RSS] = peer_abort,
4020 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4021 [CPL_RDMA_TERMINATE] = terminate,
2f5b48c3 4022 [CPL_FW4_ACK] = fw4_ack,
1cab775c 4023 [CPL_FW6_MSG] = deferred_fw6_msg,
9dec900c 4024 [CPL_RX_PKT] = rx_pkt,
8d1f1a6b
H		 4025 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4026 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
be4c9bad
RD		 4027};
4028
4029static void process_timeout(struct c4iw_ep *ep)
4030{
4031 struct c4iw_qp_attributes attrs;
4032 int abort = 1;
4033
2f5b48c3 4034 mutex_lock(&ep->com.mutex);
a9a42886
JP		 4035 pr_debug("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4036 ep->com.state);
793dad94 4037 set_bit(TIMEDOUT, &ep->com.history);
be4c9bad
RD		 4038 switch (ep->com.state) {
4039 case MPA_REQ_SENT:
be4c9bad
RD		 4040 connect_reply_upcall(ep, -ETIMEDOUT);
4041 break;
4042 case MPA_REQ_WAIT:
ceb110a8 4043 case MPA_REQ_RCVD:
e4b76a2a 4044 case MPA_REP_SENT:
ceb110a8 4045 case FPDU_MODE:
be4c9bad
RD		 4046 break;
4047 case CLOSING:
4048 case MORIBUND:
4049 if (ep->com.cm_id && ep->com.qp) {
4050 attrs.next_state = C4IW_QP_STATE_ERROR;
4051 c4iw_modify_qp(ep->com.qp->rhp,
4052 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4053 &attrs, 1);
4054 }
be13b2df 4055 close_complete_upcall(ep, -ETIMEDOUT);
be4c9bad 4056 break;
b33bd0cb
SW		 4057 case ABORTING:
4058 case DEAD:
4059
4060 /*
4061 * These states are expected if the ep timed out at the same
4062 * time as another thread was calling stop_ep_timer().
4063 * So we silently do nothing for these states.
4064 */
4065 abort = 0;
4066 break;
be4c9bad 4067 default:
76f267b7 4068 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
be4c9bad 4069 __func__, ep, ep->hwtid, ep->com.state);
be4c9bad
RD		 4070 abort = 0;
4071 }
cc18b939 4072 mutex_unlock(&ep->com.mutex);
69736279
H		 4073 if (abort)
4074 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
be4c9bad
RD		 4075 c4iw_put_ep(&ep->com);
4076}
4077
4078static void process_timedout_eps(void)
4079{
4080 struct c4iw_ep *ep;
4081
4082 spin_lock_irq(&timeout_lock);
4083 while (!list_empty(&timeout_list)) {
4084 struct list_head *tmp;
4085
4086 tmp = timeout_list.next;
4087 list_del(tmp);
b33bd0cb
SW		 4088 tmp->next = NULL;
4089 tmp->prev = NULL;
be4c9bad
RD		 4090 spin_unlock_irq(&timeout_lock);
4091 ep = list_entry(tmp, struct c4iw_ep, entry);
4092 process_timeout(ep);
4093 spin_lock_irq(&timeout_lock);
4094 }
4095 spin_unlock_irq(&timeout_lock);
4096}
4097
4098static void process_work(struct work_struct *work)
4099{
4100 struct sk_buff *skb = NULL;
4101 struct c4iw_dev *dev;
c1d7356c 4102 struct cpl_act_establish *rpl;
be4c9bad
RD		 4103 unsigned int opcode;
4104 int ret;
4105
b33bd0cb 4106 process_timedout_eps();
be4c9bad
RD		 4107 while ((skb = skb_dequeue(&rxq))) {
4108 rpl = cplhdr(skb);
4109 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4110 opcode = rpl->ot.opcode;
4111
4112 BUG_ON(!work_handlers[opcode]);
4113 ret = work_handlers[opcode](dev, skb);
4114 if (!ret)
4115 kfree_skb(skb);
b33bd0cb 4116 process_timedout_eps();
be4c9bad 4117 }
be4c9bad
RD		 4118}
4119
4120static DECLARE_WORK(skb_work, process_work);
4121
4122static void ep_timeout(unsigned long arg)
4123{
4124 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
1ec779cc 4125 int kickit = 0;
be4c9bad
RD		 4126
4127 spin_lock(&timeout_lock);
1ec779cc 4128 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
b33bd0cb
SW		 4129 /*
4130 * Only insert if it is not already on the list.
4131 */
4132 if (!ep->entry.next) {
4133 list_add_tail(&ep->entry, &timeout_list);
4134 kickit = 1;
4135 }
1ec779cc 4136 }
be4c9bad 4137 spin_unlock(&timeout_lock);
1ec779cc
VP		 4138 if (kickit)
4139 queue_work(workq, &skb_work);
be4c9bad
RD		 4140}
4141
cfdda9d7
SW		 4142/*
4143 * All the CM events are handled on a work queue to have a safe context.
4144 */
4145static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4146{
4147
4148 /*
4149 * Save dev in the skb->cb area.
4150 */
4151 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4152
4153 /*
4154 * Queue the skb and schedule the worker thread.
4155 */
4156 skb_queue_tail(&rxq, skb);
4157 queue_work(workq, &skb_work);
4158 return 0;
4159}
4160
4161static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4162{
4163 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4164
4165 if (rpl->status != CPL_ERR_NONE) {
700456bd
JP		 4166 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4167 rpl->status, GET_TID(rpl));
cfdda9d7 4168 }
2f5b48c3 4169 kfree_skb(skb);
cfdda9d7
SW		 4170 return 0;
4171}
4172
be4c9bad
RD		 4173static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4174{
4175 struct cpl_fw6_msg *rpl = cplhdr(skb);
4176 struct c4iw_wr_wait *wr_waitp;
4177 int ret;
4178
a9a42886 4179 pr_debug("%s type %u\n", __func__, rpl->type);
be4c9bad
RD		 4180
4181 switch (rpl->type) {
5be78ee9 4182 case FW6_TYPE_WR_RPL:
be4c9bad 4183 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
c8e081a1 4184 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
a9a42886 4185 pr_debug("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
d9594d99
SW		 4186 if (wr_waitp)
4187 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2f5b48c3 4188 kfree_skb(skb);
be4c9bad 4189 break;
5be78ee9 4190 case FW6_TYPE_CQE:
5be78ee9 4191 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
1cab775c 4192 sched(dev, skb);
5be78ee9 4193 break;
be4c9bad 4194 default:
700456bd
JP		 4195 pr_err("%s unexpected fw6 msg type %u\n",
4196 __func__, rpl->type);
2f5b48c3 4197 kfree_skb(skb);
be4c9bad
RD		 4198 break;
4199 }
4200 return 0;
4201}
4202
8da7e7a5
SW		 4203static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4204{
4205 struct cpl_abort_req_rss *req = cplhdr(skb);
4206 struct c4iw_ep *ep;
8da7e7a5
SW		 4207 unsigned int tid = GET_TID(req);
4208
944661dd
H		 4209 ep = get_ep_from_tid(dev, tid);
4210 /* This EP will be dereferenced in peer_abort() */
14b92228 4211 if (!ep) {
700456bd 4212 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
14b92228
SW		 4213 kfree_skb(skb);
4214 return 0;
4215 }
b65eef0a 4216 if (cxgb_is_neg_adv(req->status)) {
a9a42886
JP		 4217 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
4218 __func__, ep->hwtid, req->status,
4219 neg_adv_str(req->status));
944661dd 4220 goto out;
8da7e7a5 4221 }
a9a42886
JP		 4222 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4223 ep->com.state);
8da7e7a5 4224
093108cb 4225 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
944661dd 4226out:
8da7e7a5
SW		 4227 sched(dev, skb);
4228 return 0;
4229}
4230
be4c9bad
RD		 4231/*
4232 * Most upcalls from the T4 Core go to sched() to
4233 * schedule the processing on a work queue.
4234 */
4235c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4236 [CPL_ACT_ESTABLISH] = sched,
4237 [CPL_ACT_OPEN_RPL] = sched,
4238 [CPL_RX_DATA] = sched,
4239 [CPL_ABORT_RPL_RSS] = sched,
4240 [CPL_ABORT_RPL] = sched,
4241 [CPL_PASS_OPEN_RPL] = sched,
4242 [CPL_CLOSE_LISTSRV_RPL] = sched,
4243 [CPL_PASS_ACCEPT_REQ] = sched,
4244 [CPL_PASS_ESTABLISH] = sched,
4245 [CPL_PEER_CLOSE] = sched,
4246 [CPL_CLOSE_CON_RPL] = sched,
8da7e7a5 4247 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
be4c9bad
RD		 4248 [CPL_RDMA_TERMINATE] = sched,
4249 [CPL_FW4_ACK] = sched,
4250 [CPL_SET_TCB_RPL] = set_tcb_rpl,
1cab775c
VP		 4251 [CPL_FW6_MSG] = fw6_msg,
4252 [CPL_RX_PKT] = sched
be4c9bad
RD		 4253};
4254
cfdda9d7
SW		 4255int __init c4iw_cm_init(void)
4256{
be4c9bad 4257 spin_lock_init(&timeout_lock);
cfdda9d7
SW		 4258 skb_queue_head_init(&rxq);
4259
52ee1a05 4260 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
cfdda9d7
SW		 4261 if (!workq)
4262 return -ENOMEM;
4263
cfdda9d7
SW		 4264 return 0;
4265}
4266
46c1376d 4267void c4iw_cm_term(void)
cfdda9d7 4268{
be4c9bad 4269 WARN_ON(!list_empty(&timeout_list));
cfdda9d7
SW		 4270 flush_workqueue(workq);
4271 destroy_workqueue(workq);
4272}
Linux 6.x block layer and io_uring tree(s)