projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
RDMA/iw_cxgb4: allocate enough space for debugfs "qps" dump
[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
cfdda9d7 52#include "iw_cxgb4.h"
84cc6ac6 53#include "clip_tbl.h"
cfdda9d7
SW		 54
55static char *states[] = {
56 "idle",
57 "listen",
58 "connecting",
59 "mpa_wait_req",
60 "mpa_req_sent",
61 "mpa_req_rcvd",
62 "mpa_rep_sent",
63 "fpdu_mode",
64 "aborting",
65 "closing",
66 "moribund",
67 "dead",
68 NULL,
69};
70
5be78ee9
VP		 71static int nocong;
72module_param(nocong, int, 0644);
73MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
74
75static int enable_ecn;
76module_param(enable_ecn, int, 0644);
77MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
78
b52fe09e 79static int dack_mode = 1;
ba6d3925 80module_param(dack_mode, int, 0644);
b52fe09e 81MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
ba6d3925 82
4c2c5763 83uint c4iw_max_read_depth = 32;
be4c9bad 84module_param(c4iw_max_read_depth, int, 0644);
4c2c5763
HS		 85MODULE_PARM_DESC(c4iw_max_read_depth,
86 "Per-connection max ORD/IRD (default=32)");
be4c9bad 87
cfdda9d7
SW		 88static int enable_tcp_timestamps;
89module_param(enable_tcp_timestamps, int, 0644);
90MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
91
92static int enable_tcp_sack;
93module_param(enable_tcp_sack, int, 0644);
94MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
95
96static int enable_tcp_window_scaling = 1;
97module_param(enable_tcp_window_scaling, int, 0644);
98MODULE_PARM_DESC(enable_tcp_window_scaling,
99 "Enable tcp window scaling (default=1)");
100
101int c4iw_debug;
102module_param(c4iw_debug, int, 0644);
103MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
104
df2d5130 105static int peer2peer = 1;
cfdda9d7 106module_param(peer2peer, int, 0644);
df2d5130 107MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
cfdda9d7
SW		 108
109static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
110module_param(p2p_type, int, 0644);
111MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
112 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
113
114static int ep_timeout_secs = 60;
115module_param(ep_timeout_secs, int, 0644);
116MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
117 "in seconds (default=60)");
118
b8ac3112 119static int mpa_rev = 2;
cfdda9d7
SW		 120module_param(mpa_rev, int, 0644);
121MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
ccd2c30b 122 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
b8ac3112 123 " compliant (default=2)");
cfdda9d7
SW		 124
125static int markers_enabled;
126module_param(markers_enabled, int, 0644);
127MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
128
129static int crc_enabled = 1;
130module_param(crc_enabled, int, 0644);
131MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
132
133static int rcv_win = 256 * 1024;
134module_param(rcv_win, int, 0644);
135MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
136
98ae68b7 137static int snd_win = 128 * 1024;
cfdda9d7 138module_param(snd_win, int, 0644);
98ae68b7 139MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
cfdda9d7 140
cfdda9d7 141static struct workqueue_struct *workq;
cfdda9d7
SW		 142
143static struct sk_buff_head rxq;
cfdda9d7
SW		 144
145static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
146static void ep_timeout(unsigned long arg);
147static void connect_reply_upcall(struct c4iw_ep *ep, int status);
9dec900c 148static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
cfdda9d7 149
be4c9bad
RD		 150static LIST_HEAD(timeout_list);
151static spinlock_t timeout_lock;
152
9ca6f7cf
H		 153static void deref_cm_id(struct c4iw_ep_common *epc)
154{
155 epc->cm_id->rem_ref(epc->cm_id);
156 epc->cm_id = NULL;
157 set_bit(CM_ID_DEREFED, &epc->history);
158}
159
160static void ref_cm_id(struct c4iw_ep_common *epc)
161{
162 set_bit(CM_ID_REFED, &epc->history);
163 epc->cm_id->add_ref(epc->cm_id);
164}
165
325abead
VP		 166static void deref_qp(struct c4iw_ep *ep)
167{
168 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
169 clear_bit(QP_REFERENCED, &ep->com.flags);
9ca6f7cf 170 set_bit(QP_DEREFED, &ep->com.history);
325abead
VP		 171}
172
173static void ref_qp(struct c4iw_ep *ep)
174{
175 set_bit(QP_REFERENCED, &ep->com.flags);
9ca6f7cf 176 set_bit(QP_REFED, &ep->com.history);
325abead
VP		 177 c4iw_qp_add_ref(&ep->com.qp->ibqp);
178}
179
cfdda9d7
SW		 180static void start_ep_timer(struct c4iw_ep *ep)
181{
182 PDBG("%s ep %p\n", __func__, ep);
183 if (timer_pending(&ep->timer)) {
1ec779cc
VP		 184 pr_err("%s timer already started! ep %p\n",
185 __func__, ep);
186 return;
187 }
188 clear_bit(TIMEOUT, &ep->com.flags);
189 c4iw_get_ep(&ep->com);
cfdda9d7
SW		 190 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
191 ep->timer.data = (unsigned long)ep;
192 ep->timer.function = ep_timeout;
193 add_timer(&ep->timer);
194}
195
b33bd0cb 196static int stop_ep_timer(struct c4iw_ep *ep)
cfdda9d7 197{
1ec779cc 198 PDBG("%s ep %p stopping\n", __func__, ep);
cfdda9d7 199 del_timer_sync(&ep->timer);
b33bd0cb 200 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
1ec779cc 201 c4iw_put_ep(&ep->com);
b33bd0cb
SW		 202 return 0;
203 }
204 return 1;
cfdda9d7
SW		 205}
206
207static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
208 struct l2t_entry *l2e)
209{
210 int error = 0;
211
212 if (c4iw_fatal_error(rdev)) {
213 kfree_skb(skb);
214 PDBG("%s - device in error state - dropping\n", __func__);
215 return -EIO;
216 }
217 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
218 if (error < 0)
219 kfree_skb(skb);
caa6c9f2
H		 220 else if (error == NET_XMIT_DROP)
221 return -ENOMEM;
74594861 222 return error < 0 ? error : 0;
cfdda9d7
SW		 223}
224
225int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
226{
227 int error = 0;
228
229 if (c4iw_fatal_error(rdev)) {
230 kfree_skb(skb);
231 PDBG("%s - device in error state - dropping\n", __func__);
232 return -EIO;
233 }
234 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
235 if (error < 0)
236 kfree_skb(skb);
74594861 237 return error < 0 ? error : 0;
cfdda9d7
SW		 238}
239
240static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
241{
242 struct cpl_tid_release *req;
243
244 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
245 if (!skb)
246 return;
247 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
248 INIT_TP_WR(req, hwtid);
249 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
250 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
251 c4iw_ofld_send(rdev, skb);
252 return;
253}
254
255static void set_emss(struct c4iw_ep *ep, u16 opt)
256{
6c53e938 257 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
04524a47
H		 258 ((AF_INET == ep->com.remote_addr.ss_family) ?
259 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
260 sizeof(struct tcphdr);
cfdda9d7 261 ep->mss = ep->emss;
6c53e938 262 if (TCPOPT_TSTAMP_G(opt))
04524a47 263 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
cfdda9d7
SW		 264 if (ep->emss < 128)
265 ep->emss = 128;
92e7ae71
HS		 266 if (ep->emss & 7)
267 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
6c53e938
HS		 268 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
269 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
cfdda9d7
SW		 270 ep->mss, ep->emss);
271}
272
273static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
274{
cfdda9d7
SW		 275 enum c4iw_ep_state state;
276
2f5b48c3 277 mutex_lock(&epc->mutex);
cfdda9d7 278 state = epc->state;
2f5b48c3 279 mutex_unlock(&epc->mutex);
cfdda9d7
SW		 280 return state;
281}
282
283static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
284{
285 epc->state = new;
286}
287
288static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
289{
2f5b48c3 290 mutex_lock(&epc->mutex);
cfdda9d7
SW		 291 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
292 __state_set(epc, new);
2f5b48c3 293 mutex_unlock(&epc->mutex);
cfdda9d7
SW		 294 return;
295}
296
297static void *alloc_ep(int size, gfp_t gfp)
298{
299 struct c4iw_ep_common *epc;
300
301 epc = kzalloc(size, gfp);
302 if (epc) {
303 kref_init(&epc->kref);
2f5b48c3 304 mutex_init(&epc->mutex);
aadc4df3 305 c4iw_init_wr_wait(&epc->wr_wait);
cfdda9d7
SW		 306 }
307 PDBG("%s alloc ep %p\n", __func__, epc);
308 return epc;
309}
310
944661dd
H		 311static void remove_ep_tid(struct c4iw_ep *ep)
312{
313 unsigned long flags;
314
315 spin_lock_irqsave(&ep->com.dev->lock, flags);
316 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
317 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
318}
319
320static void insert_ep_tid(struct c4iw_ep *ep)
321{
322 unsigned long flags;
323
324 spin_lock_irqsave(&ep->com.dev->lock, flags);
325 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
326 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
327}
328
329/*
330 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
331 */
332static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
333{
334 struct c4iw_ep *ep;
335 unsigned long flags;
336
337 spin_lock_irqsave(&dev->lock, flags);
338 ep = idr_find(&dev->hwtid_idr, tid);
339 if (ep)
340 c4iw_get_ep(&ep->com);
341 spin_unlock_irqrestore(&dev->lock, flags);
342 return ep;
343}
344
f86fac79
H		 345/*
346 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
347 */
348static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
349 unsigned int stid)
350{
351 struct c4iw_listen_ep *ep;
352 unsigned long flags;
353
354 spin_lock_irqsave(&dev->lock, flags);
355 ep = idr_find(&dev->stid_idr, stid);
356 if (ep)
357 c4iw_get_ep(&ep->com);
358 spin_unlock_irqrestore(&dev->lock, flags);
359 return ep;
360}
361
cfdda9d7
SW		 362void _c4iw_free_ep(struct kref *kref)
363{
364 struct c4iw_ep *ep;
365
366 ep = container_of(kref, struct c4iw_ep, com.kref);
9dec900c 367 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
325abead
VP		 368 if (test_bit(QP_REFERENCED, &ep->com.flags))
369 deref_qp(ep);
cfdda9d7 370 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
84cc6ac6
H		 371 if (ep->com.remote_addr.ss_family == AF_INET6) {
372 struct sockaddr_in6 *sin6 =
373 (struct sockaddr_in6 *)
170003c8 374 &ep->com.local_addr;
84cc6ac6
H		 375
376 cxgb4_clip_release(
377 ep->com.dev->rdev.lldi.ports[0],
378 (const u32 *)&sin6->sin6_addr.s6_addr,
379 1);
380 }
cfdda9d7
SW		 381 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
382 dst_release(ep->dst);
383 cxgb4_l2t_release(ep->l2t);
c878b706
H		 384 if (ep->mpa_skb)
385 kfree_skb(ep->mpa_skb);
cfdda9d7
SW		 386 }
387 kfree(ep);
388}
389
390static void release_ep_resources(struct c4iw_ep *ep)
391{
392 set_bit(RELEASE_RESOURCES, &ep->com.flags);
944661dd
H		 393
394 /*
395 * If we have a hwtid, then remove it from the idr table
396 * so lookups will no longer find this endpoint. Otherwise
397 * we have a race where one thread finds the ep ptr just
398 * before the other thread is freeing the ep memory.
399 */
400 if (ep->hwtid != -1)
401 remove_ep_tid(ep);
cfdda9d7
SW		 402 c4iw_put_ep(&ep->com);
403}
404
cfdda9d7
SW		 405static int status2errno(int status)
406{
407 switch (status) {
408 case CPL_ERR_NONE:
409 return 0;
410 case CPL_ERR_CONN_RESET:
411 return -ECONNRESET;
412 case CPL_ERR_ARP_MISS:
413 return -EHOSTUNREACH;
414 case CPL_ERR_CONN_TIMEDOUT:
415 return -ETIMEDOUT;
416 case CPL_ERR_TCAM_FULL:
417 return -ENOMEM;
418 case CPL_ERR_CONN_EXIST:
419 return -EADDRINUSE;
420 default:
421 return -EIO;
422 }
423}
424
425/*
426 * Try and reuse skbs already allocated...
427 */
428static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
429{
430 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
431 skb_trim(skb, 0);
432 skb_get(skb);
433 skb_reset_transport_header(skb);
434 } else {
435 skb = alloc_skb(len, gfp);
436 }
b38a0ad8 437 t4_set_arp_err_handler(skb, NULL, NULL);
cfdda9d7
SW		 438 return skb;
439}
440
830662f6
VP		 441static struct net_device *get_real_dev(struct net_device *egress_dev)
442{
11b8e22d 443 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
830662f6
VP		 444}
445
446static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
447{
448 int i;
449
450 egress_dev = get_real_dev(egress_dev);
451 for (i = 0; i < dev->rdev.lldi.nports; i++)
452 if (dev->rdev.lldi.ports[i] == egress_dev)
453 return 1;
454 return 0;
455}
456
457static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
458 __u8 *peer_ip, __be16 local_port,
459 __be16 peer_port, u8 tos,
460 __u32 sin6_scope_id)
461{
462 struct dst_entry *dst = NULL;
463
464 if (IS_ENABLED(CONFIG_IPV6)) {
465 struct flowi6 fl6;
466
467 memset(&fl6, 0, sizeof(fl6));
468 memcpy(&fl6.daddr, peer_ip, 16);
469 memcpy(&fl6.saddr, local_ip, 16);
470 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
471 fl6.flowi6_oif = sin6_scope_id;
472 dst = ip6_route_output(&init_net, NULL, &fl6);
473 if (!dst)
474 goto out;
475 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
476 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
477 dst_release(dst);
478 dst = NULL;
479 }
480 }
481
482out:
483 return dst;
484}
485
486static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
cfdda9d7
SW		 487 __be32 peer_ip, __be16 local_port,
488 __be16 peer_port, u8 tos)
489{
490 struct rtable *rt;
31e4543d 491 struct flowi4 fl4;
830662f6 492 struct neighbour *n;
78fbfd8a 493
31e4543d 494 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
78fbfd8a
DM		 495 peer_port, local_port, IPPROTO_TCP,
496 tos, 0);
b23dd4fe 497 if (IS_ERR(rt))
cfdda9d7 498 return NULL;
830662f6
VP		 499 n = dst_neigh_lookup(&rt->dst, &peer_ip);
500 if (!n)
501 return NULL;
f8e81908
SW		 502 if (!our_interface(dev, n->dev) &&
503 !(n->dev->flags & IFF_LOOPBACK)) {
d480201b 504 neigh_release(n);
830662f6
VP		 505 dst_release(&rt->dst);
506 return NULL;
507 }
508 neigh_release(n);
509 return &rt->dst;
cfdda9d7
SW		 510}
511
512static void arp_failure_discard(void *handle, struct sk_buff *skb)
513{
9dec900c 514 pr_err(MOD "ARP failure\n");
cfdda9d7
SW		 515 kfree_skb(skb);
516}
517
64bec74a
H		 518static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
519{
520 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
521}
522
9dec900c 523enum {
8d1f1a6b 524 NUM_FAKE_CPLS = 2,
9dec900c 525 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
8d1f1a6b 526 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
9dec900c
H		 527};
528
529static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
530{
531 struct c4iw_ep *ep;
532
533 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
534 release_ep_resources(ep);
535 return 0;
536}
537
8d1f1a6b
H		 538static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
539{
540 struct c4iw_ep *ep;
541
542 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
543 c4iw_put_ep(&ep->parent_ep->com);
544 release_ep_resources(ep);
545 return 0;
546}
547
9dec900c
H		 548/*
549 * Fake up a special CPL opcode and call sched() so process_work() will call
550 * _put_ep_safe() in a safe context to free the ep resources. This is needed
551 * because ARP error handlers are called in an ATOMIC context, and
552 * _c4iw_free_ep() needs to block.
553 */
8d1f1a6b
H		 554static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
555 int cpl)
9dec900c
H		 556{
557 struct cpl_act_establish *rpl = cplhdr(skb);
558
559 /* Set our special ARP_FAILURE opcode */
8d1f1a6b 560 rpl->ot.opcode = cpl;
9dec900c
H		 561
562 /*
563 * Save ep in the skb->cb area, after where sched() will save the dev
564 * ptr.
565 */
566 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
567 sched(ep->com.dev, skb);
568}
569
570/* Handle an ARP failure for an accept */
571static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
572{
573 struct c4iw_ep *ep = handle;
574
575 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
576 ep->hwtid);
577
578 __state_set(&ep->com, DEAD);
8d1f1a6b 579 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
9dec900c
H		 580}
581
cfdda9d7
SW		 582/*
583 * Handle an ARP failure for an active open.
584 */
585static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
586{
5dab6d3a
H		 587 struct c4iw_ep *ep = handle;
588
e3d132d1 589 printk(KERN_ERR MOD "ARP failure during connect\n");
5dab6d3a 590 connect_reply_upcall(ep, -EHOSTUNREACH);
9dec900c 591 __state_set(&ep->com, DEAD);
84cc6ac6
H		 592 if (ep->com.remote_addr.ss_family == AF_INET6) {
593 struct sockaddr_in6 *sin6 =
170003c8 594 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 595 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
596 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
597 }
5dab6d3a
H		 598 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
599 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
8d1f1a6b 600 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
cfdda9d7
SW		 601}
602
603/*
604 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
605 * and send it along.
606 */
607static void abort_arp_failure(void *handle, struct sk_buff *skb)
608{
761e19a5
H		 609 int ret;
610 struct c4iw_ep *ep = handle;
611 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
cfdda9d7
SW		 612 struct cpl_abort_req *req = cplhdr(skb);
613
614 PDBG("%s rdev %p\n", __func__, rdev);
615 req->cmd = CPL_ABORT_NO_RST;
761e19a5
H		 616 ret = c4iw_ofld_send(rdev, skb);
617 if (ret) {
618 __state_set(&ep->com, DEAD);
619 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
620 }
cfdda9d7
SW		 621}
622
fef4422d 623static int send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
cfdda9d7
SW		 624{
625 unsigned int flowclen = 80;
626 struct fw_flowc_wr *flowc;
627 int i;
ac8e4c69
H		 628 u16 vlan = ep->l2t->vlan;
629 int nparams;
630
631 if (vlan == CPL_L2T_VLAN_NONE)
632 nparams = 8;
633 else
634 nparams = 9;
cfdda9d7
SW		 635
636 skb = get_skb(skb, flowclen, GFP_KERNEL);
637 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
638
e2ac9628 639 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
ac8e4c69 640 FW_FLOWC_WR_NPARAMS_V(nparams));
e2ac9628
HS		 641 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(flowclen,
642 16)) | FW_WR_FLOWID_V(ep->hwtid));
cfdda9d7
SW		 643
644 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
5167865a 645 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
35b1de55 646 (ep->com.dev->rdev.lldi.pf));
cfdda9d7
SW		 647 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
648 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
649 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
650 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
651 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
652 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
653 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
654 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
655 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
656 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
657 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
b408ff28 658 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
cfdda9d7
SW		 659 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
660 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
ac8e4c69
H		 661 if (nparams == 9) {
662 u16 pri;
663
664 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
665 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
666 flowc->mnemval[8].val = cpu_to_be32(pri);
667 } else {
668 /* Pad WR to 16 byte boundary */
669 flowc->mnemval[8].mnemonic = 0;
670 flowc->mnemval[8].val = 0;
671 }
cfdda9d7
SW		 672 for (i = 0; i < 9; i++) {
673 flowc->mnemval[i].r4[0] = 0;
674 flowc->mnemval[i].r4[1] = 0;
675 flowc->mnemval[i].r4[2] = 0;
676 }
677
678 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
fef4422d 679 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
cfdda9d7
SW		 680}
681
682static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
683{
684 struct cpl_close_con_req *req;
685 struct sk_buff *skb;
686 int wrlen = roundup(sizeof *req, 16);
687
688 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
689 skb = get_skb(NULL, wrlen, gfp);
690 if (!skb) {
691 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
692 return -ENOMEM;
693 }
694 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
695 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
696 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
697 memset(req, 0, wrlen);
698 INIT_TP_WR(req, ep->hwtid);
699 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
700 ep->hwtid));
701 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
702}
703
704static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
705{
706 struct cpl_abort_req *req;
707 int wrlen = roundup(sizeof *req, 16);
708
709 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
710 skb = get_skb(skb, wrlen, gfp);
711 if (!skb) {
712 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
713 __func__);
714 return -ENOMEM;
715 }
716 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
761e19a5 717 t4_set_arp_err_handler(skb, ep, abort_arp_failure);
cfdda9d7
SW		 718 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
719 memset(req, 0, wrlen);
720 INIT_TP_WR(req, ep->hwtid);
721 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
722 req->cmd = CPL_ABORT_SEND_RST;
723 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
724}
725
92e7ae71 726static void best_mtu(const unsigned short *mtus, unsigned short mtu,
04524a47 727 unsigned int *idx, int use_ts, int ipv6)
92e7ae71 728{
04524a47
H		 729 unsigned short hdr_size = (ipv6 ?
730 sizeof(struct ipv6hdr) :
731 sizeof(struct iphdr)) +
92e7ae71 732 sizeof(struct tcphdr) +
04524a47
H		 733 (use_ts ?
734 round_up(TCPOLEN_TIMESTAMP, 4) : 0);
92e7ae71
HS		 735 unsigned short data_size = mtu - hdr_size;
736
737 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
738}
739
cfdda9d7
SW		 740static int send_connect(struct c4iw_ep *ep)
741{
963cab50
H		 742 struct cpl_act_open_req *req = NULL;
743 struct cpl_t5_act_open_req *t5req = NULL;
744 struct cpl_t6_act_open_req *t6req = NULL;
745 struct cpl_act_open_req6 *req6 = NULL;
746 struct cpl_t5_act_open_req6 *t5req6 = NULL;
747 struct cpl_t6_act_open_req6 *t6req6 = NULL;
cfdda9d7
SW		 748 struct sk_buff *skb;
749 u64 opt0;
750 u32 opt2;
751 unsigned int mtu_idx;
752 int wscale;
963cab50 753 int win, sizev4, sizev6, wrlen;
9eccfe10 754 struct sockaddr_in *la = (struct sockaddr_in *)
170003c8 755 &ep->com.local_addr;
9eccfe10 756 struct sockaddr_in *ra = (struct sockaddr_in *)
170003c8 757 &ep->com.remote_addr;
9eccfe10 758 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
170003c8 759 &ep->com.local_addr;
9eccfe10 760 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
170003c8 761 &ep->com.remote_addr;
84cc6ac6 762 int ret;
963cab50
H		 763 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
764 u32 isn = (prandom_u32() & ~7UL) - 1;
765
766 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
767 case CHELSIO_T4:
768 sizev4 = sizeof(struct cpl_act_open_req);
769 sizev6 = sizeof(struct cpl_act_open_req6);
770 break;
771 case CHELSIO_T5:
772 sizev4 = sizeof(struct cpl_t5_act_open_req);
773 sizev6 = sizeof(struct cpl_t5_act_open_req6);
774 break;
775 case CHELSIO_T6:
776 sizev4 = sizeof(struct cpl_t6_act_open_req);
777 sizev6 = sizeof(struct cpl_t6_act_open_req6);
778 break;
779 default:
780 pr_err("T%d Chip is not supported\n",
781 CHELSIO_CHIP_VERSION(adapter_type));
782 return -EINVAL;
783 }
830662f6
VP		 784
785 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
786 roundup(sizev4, 16) :
787 roundup(sizev6, 16);
cfdda9d7
SW		 788
789 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
790
791 skb = get_skb(NULL, wrlen, GFP_KERNEL);
792 if (!skb) {
793 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
794 __func__);
795 return -ENOMEM;
796 }
d4f1a5c6 797 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
cfdda9d7 798
92e7ae71 799 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
04524a47
H		 800 enable_tcp_timestamps,
801 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
cfdda9d7 802 wscale = compute_wscale(rcv_win);
b408ff28
HS		 803
804 /*
805 * Specify the largest window that will fit in opt0. The
806 * remainder will be specified in the rx_data_ack.
807 */
808 win = ep->rcv_win >> 10;
d7990b0c
AB		 809 if (win > RCV_BUFSIZ_M)
810 win = RCV_BUFSIZ_M;
b408ff28 811
6c53e938 812 opt0 = (nocong ? NO_CONG_F : 0) |
d7990b0c 813 KEEP_ALIVE_F |
6c53e938 814 DELACK_F |
d7990b0c
AB		 815 WND_SCALE_V(wscale) |
816 MSS_IDX_V(mtu_idx) |
817 L2T_IDX_V(ep->l2t->idx) |
818 TX_CHAN_V(ep->tx_chan) |
819 SMAC_SEL_V(ep->smac_idx) |
ac8e4c69 820 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 821 ULP_MODE_V(ULP_MODE_TCPDDP) |
822 RCV_BUFSIZ_V(win);
823 opt2 = RX_CHANNEL_V(0) |
6c53e938 824 CCTRL_ECN_V(enable_ecn) |
d7990b0c 825 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
cfdda9d7 826 if (enable_tcp_timestamps)
6c53e938 827 opt2 |= TSTAMPS_EN_F;
cfdda9d7 828 if (enable_tcp_sack)
6c53e938 829 opt2 |= SACK_EN_F;
cfdda9d7 830 if (wscale && enable_tcp_window_scaling)
d7990b0c 831 opt2 |= WND_SCALE_EN_F;
963cab50
H		 832 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
833 if (peer2peer)
834 isn += 4;
835
d7990b0c 836 opt2 |= T5_OPT_2_VALID_F;
cf7fe64a 837 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
0b741047 838 opt2 |= T5_ISS_F;
92e5011a 839 }
84cc6ac6
H		 840
841 if (ep->com.remote_addr.ss_family == AF_INET6)
842 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
843 (const u32 *)&la6->sin6_addr.s6_addr, 1);
844
5dab6d3a 845 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
cfdda9d7 846
963cab50
H		 847 if (ep->com.remote_addr.ss_family == AF_INET) {
848 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
849 case CHELSIO_T4:
850 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
830662f6 851 INIT_TP_WR(req, 0);
963cab50
H		 852 break;
853 case CHELSIO_T5:
854 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
855 wrlen);
856 INIT_TP_WR(t5req, 0);
857 req = (struct cpl_act_open_req *)t5req;
858 break;
859 case CHELSIO_T6:
860 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
861 wrlen);
862 INIT_TP_WR(t6req, 0);
863 req = (struct cpl_act_open_req *)t6req;
864 t5req = (struct cpl_t5_act_open_req *)t6req;
865 break;
866 default:
867 pr_err("T%d Chip is not supported\n",
868 CHELSIO_CHIP_VERSION(adapter_type));
869 ret = -EINVAL;
870 goto clip_release;
871 }
872
873 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
874 ((ep->rss_qid<<14) | ep->atid)));
875 req->local_port = la->sin_port;
876 req->peer_port = ra->sin_port;
877 req->local_ip = la->sin_addr.s_addr;
878 req->peer_ip = ra->sin_addr.s_addr;
879 req->opt0 = cpu_to_be64(opt0);
880
881 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
41b4f86c
KS		 882 req->params = cpu_to_be32(cxgb4_select_ntuple(
883 ep->com.dev->rdev.lldi.ports[0],
884 ep->l2t));
830662f6
VP		 885 req->opt2 = cpu_to_be32(opt2);
886 } else {
963cab50
H		 887 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
888 cxgb4_select_ntuple(
889 ep->com.dev->rdev.lldi.ports[0],
890 ep->l2t)));
891 t5req->rsvd = cpu_to_be32(isn);
892 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
893 t5req->opt2 = cpu_to_be32(opt2);
894 }
895 } else {
896 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
897 case CHELSIO_T4:
830662f6 898 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
830662f6 899 INIT_TP_WR(req6, 0);
963cab50
H		 900 break;
901 case CHELSIO_T5:
902 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
903 wrlen);
904 INIT_TP_WR(t5req6, 0);
905 req6 = (struct cpl_act_open_req6 *)t5req6;
906 break;
907 case CHELSIO_T6:
908 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
909 wrlen);
910 INIT_TP_WR(t6req6, 0);
911 req6 = (struct cpl_act_open_req6 *)t6req6;
912 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
913 break;
914 default:
915 pr_err("T%d Chip is not supported\n",
916 CHELSIO_CHIP_VERSION(adapter_type));
917 ret = -EINVAL;
918 goto clip_release;
919 }
920
921 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
922 ((ep->rss_qid<<14)|ep->atid)));
923 req6->local_port = la6->sin6_port;
924 req6->peer_port = ra6->sin6_port;
925 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
926 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
927 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
928 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
929 req6->opt0 = cpu_to_be64(opt0);
930
931 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
41b4f86c
KS		 932 req6->params = cpu_to_be32(cxgb4_select_ntuple(
933 ep->com.dev->rdev.lldi.ports[0],
934 ep->l2t));
830662f6 935 req6->opt2 = cpu_to_be32(opt2);
830662f6 936 } else {
963cab50
H		 937 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
938 cxgb4_select_ntuple(
41b4f86c 939 ep->com.dev->rdev.lldi.ports[0],
da22b896 940 ep->l2t)));
963cab50
H		 941 t5req6->rsvd = cpu_to_be32(isn);
942 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
943 t5req6->opt2 = cpu_to_be32(opt2);
830662f6 944 }
f079af7a
VP		 945 }
946
793dad94 947 set_bit(ACT_OPEN_REQ, &ep->com.history);
84cc6ac6 948 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
963cab50 949clip_release:
84cc6ac6
H		 950 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
951 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
952 (const u32 *)&la6->sin6_addr.s6_addr, 1);
953 return ret;
cfdda9d7
SW		 954}
955
caa6c9f2
H		 956static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
957 u8 mpa_rev_to_use)
cfdda9d7 958{
caa6c9f2 959 int mpalen, wrlen, ret;
cfdda9d7
SW		 960 struct fw_ofld_tx_data_wr *req;
961 struct mpa_message *mpa;
d2fe99e8 962 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW		 963
964 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
965
966 BUG_ON(skb_cloned(skb));
967
968 mpalen = sizeof(*mpa) + ep->plen;
d2fe99e8
KS		 969 if (mpa_rev_to_use == 2)
970 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 971 wrlen = roundup(mpalen + sizeof *req, 16);
972 skb = get_skb(skb, wrlen, GFP_KERNEL);
973 if (!skb) {
974 connect_reply_upcall(ep, -ENOMEM);
caa6c9f2 975 return -ENOMEM;
cfdda9d7
SW		 976 }
977 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
978
979 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
980 memset(req, 0, wrlen);
981 req->op_to_immdlen = cpu_to_be32(
e2ac9628
HS		 982 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
983 FW_WR_COMPL_F |
984 FW_WR_IMMDLEN_V(mpalen));
cfdda9d7 985 req->flowid_len16 = cpu_to_be32(
e2ac9628
HS		 986 FW_WR_FLOWID_V(ep->hwtid) |
987 FW_WR_LEN16_V(wrlen >> 4));
cfdda9d7
SW		 988 req->plen = cpu_to_be32(mpalen);
989 req->tunnel_to_proxy = cpu_to_be32(
e2ac9628
HS		 990 FW_OFLD_TX_DATA_WR_FLUSH_F |
991 FW_OFLD_TX_DATA_WR_SHOVE_F);
cfdda9d7
SW		 992
993 mpa = (struct mpa_message *)(req + 1);
994 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
3d4e7994
H		 995
996 mpa->flags = 0;
997 if (crc_enabled)
998 mpa->flags |= MPA_CRC;
999 if (markers_enabled) {
1000 mpa->flags |= MPA_MARKERS;
1001 ep->mpa_attr.recv_marker_enabled = 1;
1002 } else {
1003 ep->mpa_attr.recv_marker_enabled = 0;
1004 }
1005 if (mpa_rev_to_use == 2)
1006 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1007
cfdda9d7 1008 mpa->private_data_size = htons(ep->plen);
d2fe99e8 1009 mpa->revision = mpa_rev_to_use;
01b225e1 1010 if (mpa_rev_to_use == 1) {
d2fe99e8 1011 ep->tried_with_mpa_v1 = 1;
01b225e1
KS		 1012 ep->retry_with_mpa_v1 = 0;
1013 }
d2fe99e8
KS		 1014
1015 if (mpa_rev_to_use == 2) {
f747c34a
RD		 1016 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1017 sizeof (struct mpa_v2_conn_params));
4c2c5763
HS		 1018 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1019 ep->ord);
d2fe99e8
KS		 1020 mpa_v2_params.ird = htons((u16)ep->ird);
1021 mpa_v2_params.ord = htons((u16)ep->ord);
1022
1023 if (peer2peer) {
1024 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1025 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1026 mpa_v2_params.ord |=
1027 htons(MPA_V2_RDMA_WRITE_RTR);
1028 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1029 mpa_v2_params.ord |=
1030 htons(MPA_V2_RDMA_READ_RTR);
1031 }
1032 memcpy(mpa->private_data, &mpa_v2_params,
1033 sizeof(struct mpa_v2_conn_params));
cfdda9d7 1034
d2fe99e8
KS		 1035 if (ep->plen)
1036 memcpy(mpa->private_data +
1037 sizeof(struct mpa_v2_conn_params),
1038 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1039 } else
1040 if (ep->plen)
1041 memcpy(mpa->private_data,
1042 ep->mpa_pkt + sizeof(*mpa), ep->plen);
cfdda9d7
SW		 1043
1044 /*
1045 * Reference the mpa skb. This ensures the data area
1046 * will remain in memory until the hw acks the tx.
1047 * Function fw4_ack() will deref it.
1048 */
1049 skb_get(skb);
1050 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1051 BUG_ON(ep->mpa_skb);
1052 ep->mpa_skb = skb;
caa6c9f2
H		 1053 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1054 if (ret)
1055 return ret;
cfdda9d7 1056 start_ep_timer(ep);
a7db89eb 1057 __state_set(&ep->com, MPA_REQ_SENT);
cfdda9d7 1058 ep->mpa_attr.initiator = 1;
9c88aa00 1059 ep->snd_seq += mpalen;
caa6c9f2 1060 return ret;
cfdda9d7
SW		 1061}
1062
1063static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1064{
1065 int mpalen, wrlen;
1066 struct fw_ofld_tx_data_wr *req;
1067 struct mpa_message *mpa;
1068 struct sk_buff *skb;
d2fe99e8 1069 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW		 1070
1071 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1072
1073 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS		 1074 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1075 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 1076 wrlen = roundup(mpalen + sizeof *req, 16);
1077
1078 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1079 if (!skb) {
1080 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1081 return -ENOMEM;
1082 }
1083 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1084
1085 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1086 memset(req, 0, wrlen);
1087 req->op_to_immdlen = cpu_to_be32(
e2ac9628
HS		 1088 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1089 FW_WR_COMPL_F |
1090 FW_WR_IMMDLEN_V(mpalen));
cfdda9d7 1091 req->flowid_len16 = cpu_to_be32(
e2ac9628
HS		 1092 FW_WR_FLOWID_V(ep->hwtid) |
1093 FW_WR_LEN16_V(wrlen >> 4));
cfdda9d7
SW		 1094 req->plen = cpu_to_be32(mpalen);
1095 req->tunnel_to_proxy = cpu_to_be32(
e2ac9628
HS		 1096 FW_OFLD_TX_DATA_WR_FLUSH_F |
1097 FW_OFLD_TX_DATA_WR_SHOVE_F);
cfdda9d7
SW		 1098
1099 mpa = (struct mpa_message *)(req + 1);
1100 memset(mpa, 0, sizeof(*mpa));
1101 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1102 mpa->flags = MPA_REJECT;
fe7e0a4d 1103 mpa->revision = ep->mpa_attr.version;
cfdda9d7 1104 mpa->private_data_size = htons(plen);
d2fe99e8
KS		 1105
1106 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1107 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD		 1108 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1109 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS		 1110 mpa_v2_params.ird = htons(((u16)ep->ird) |
1111 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1112 0));
1113 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1114 (p2p_type ==
1115 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1116 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1117 FW_RI_INIT_P2PTYPE_READ_REQ ?
1118 MPA_V2_RDMA_READ_RTR : 0) : 0));
1119 memcpy(mpa->private_data, &mpa_v2_params,
1120 sizeof(struct mpa_v2_conn_params));
1121
1122 if (ep->plen)
1123 memcpy(mpa->private_data +
1124 sizeof(struct mpa_v2_conn_params), pdata, plen);
1125 } else
1126 if (plen)
1127 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW		 1128
1129 /*
1130 * Reference the mpa skb again. This ensures the data area
1131 * will remain in memory until the hw acks the tx.
1132 * Function fw4_ack() will deref it.
1133 */
1134 skb_get(skb);
1135 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
64bec74a 1136 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
cfdda9d7
SW		 1137 BUG_ON(ep->mpa_skb);
1138 ep->mpa_skb = skb;
9c88aa00 1139 ep->snd_seq += mpalen;
cfdda9d7
SW		 1140 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1141}
1142
1143static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1144{
1145 int mpalen, wrlen;
1146 struct fw_ofld_tx_data_wr *req;
1147 struct mpa_message *mpa;
1148 struct sk_buff *skb;
d2fe99e8 1149 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW		 1150
1151 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1152
1153 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS		 1154 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1155 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW		 1156 wrlen = roundup(mpalen + sizeof *req, 16);
1157
1158 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1159 if (!skb) {
1160 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1161 return -ENOMEM;
1162 }
1163 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1164
1165 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1166 memset(req, 0, wrlen);
1167 req->op_to_immdlen = cpu_to_be32(
e2ac9628
HS		 1168 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1169 FW_WR_COMPL_F |
1170 FW_WR_IMMDLEN_V(mpalen));
cfdda9d7 1171 req->flowid_len16 = cpu_to_be32(
e2ac9628
HS		 1172 FW_WR_FLOWID_V(ep->hwtid) |
1173 FW_WR_LEN16_V(wrlen >> 4));
cfdda9d7
SW		 1174 req->plen = cpu_to_be32(mpalen);
1175 req->tunnel_to_proxy = cpu_to_be32(
e2ac9628
HS		 1176 FW_OFLD_TX_DATA_WR_FLUSH_F |
1177 FW_OFLD_TX_DATA_WR_SHOVE_F);
cfdda9d7
SW		 1178
1179 mpa = (struct mpa_message *)(req + 1);
1180 memset(mpa, 0, sizeof(*mpa));
1181 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
3d4e7994
H		 1182 mpa->flags = 0;
1183 if (ep->mpa_attr.crc_enabled)
1184 mpa->flags |= MPA_CRC;
1185 if (ep->mpa_attr.recv_marker_enabled)
1186 mpa->flags |= MPA_MARKERS;
d2fe99e8 1187 mpa->revision = ep->mpa_attr.version;
cfdda9d7 1188 mpa->private_data_size = htons(plen);
d2fe99e8
KS		 1189
1190 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1191 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD		 1192 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1193 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS		 1194 mpa_v2_params.ird = htons((u16)ep->ird);
1195 mpa_v2_params.ord = htons((u16)ep->ord);
1196 if (peer2peer && (ep->mpa_attr.p2p_type !=
1197 FW_RI_INIT_P2PTYPE_DISABLED)) {
1198 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1199
1200 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1201 mpa_v2_params.ord |=
1202 htons(MPA_V2_RDMA_WRITE_RTR);
1203 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1204 mpa_v2_params.ord |=
1205 htons(MPA_V2_RDMA_READ_RTR);
1206 }
1207
1208 memcpy(mpa->private_data, &mpa_v2_params,
1209 sizeof(struct mpa_v2_conn_params));
1210
1211 if (ep->plen)
1212 memcpy(mpa->private_data +
1213 sizeof(struct mpa_v2_conn_params), pdata, plen);
1214 } else
1215 if (plen)
1216 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW		 1217
1218 /*
1219 * Reference the mpa skb. This ensures the data area
1220 * will remain in memory until the hw acks the tx.
1221 * Function fw4_ack() will deref it.
1222 */
1223 skb_get(skb);
64bec74a 1224 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
cfdda9d7 1225 ep->mpa_skb = skb;
a7db89eb 1226 __state_set(&ep->com, MPA_REP_SENT);
9c88aa00 1227 ep->snd_seq += mpalen;
cfdda9d7
SW		 1228 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1229}
1230
1231static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1232{
1233 struct c4iw_ep *ep;
1234 struct cpl_act_establish *req = cplhdr(skb);
1235 unsigned int tid = GET_TID(req);
6c53e938 1236 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
cfdda9d7 1237 struct tid_info *t = dev->rdev.lldi.tids;
fef4422d 1238 int ret;
cfdda9d7
SW		 1239
1240 ep = lookup_atid(t, atid);
1241
1242 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1243 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1244
a7db89eb 1245 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 1246 dst_confirm(ep->dst);
1247
1248 /* setup the hwtid for this connection */
1249 ep->hwtid = tid;
1250 cxgb4_insert_tid(t, ep, tid);
944661dd 1251 insert_ep_tid(ep);
cfdda9d7
SW		 1252
1253 ep->snd_seq = be32_to_cpu(req->snd_isn);
1254 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1255
1256 set_emss(ep, ntohs(req->tcp_opt));
1257
1258 /* dealloc the atid */
793dad94 1259 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7 1260 cxgb4_free_atid(t, atid);
793dad94 1261 set_bit(ACT_ESTAB, &ep->com.history);
cfdda9d7
SW		 1262
1263 /* start MPA negotiation */
fef4422d
H		 1264 ret = send_flowc(ep, NULL);
1265 if (ret)
1266 goto err;
d2fe99e8 1267 if (ep->retry_with_mpa_v1)
caa6c9f2 1268 ret = send_mpa_req(ep, skb, 1);
d2fe99e8 1269 else
caa6c9f2
H		 1270 ret = send_mpa_req(ep, skb, mpa_rev);
1271 if (ret)
1272 goto err;
a7db89eb 1273 mutex_unlock(&ep->com.mutex);
cfdda9d7 1274 return 0;
fef4422d
H		 1275err:
1276 mutex_unlock(&ep->com.mutex);
1277 connect_reply_upcall(ep, -ENOMEM);
1278 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1279 return 0;
cfdda9d7
SW		 1280}
1281
be13b2df 1282static void close_complete_upcall(struct c4iw_ep *ep, int status)
cfdda9d7
SW		 1283{
1284 struct iw_cm_event event;
1285
1286 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1287 memset(&event, 0, sizeof(event));
1288 event.event = IW_CM_EVENT_CLOSE;
be13b2df 1289 event.status = status;
cfdda9d7
SW		 1290 if (ep->com.cm_id) {
1291 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1292 ep, ep->com.cm_id, ep->hwtid);
1293 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
9ca6f7cf 1294 deref_cm_id(&ep->com);
793dad94 1295 set_bit(CLOSE_UPCALL, &ep->com.history);
cfdda9d7
SW		 1296 }
1297}
1298
cfdda9d7
SW		 1299static void peer_close_upcall(struct c4iw_ep *ep)
1300{
1301 struct iw_cm_event event;
1302
1303 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1304 memset(&event, 0, sizeof(event));
1305 event.event = IW_CM_EVENT_DISCONNECT;
1306 if (ep->com.cm_id) {
1307 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1308 ep, ep->com.cm_id, ep->hwtid);
1309 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 1310 set_bit(DISCONN_UPCALL, &ep->com.history);
cfdda9d7
SW		 1311 }
1312}
1313
1314static void peer_abort_upcall(struct c4iw_ep *ep)
1315{
1316 struct iw_cm_event event;
1317
1318 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1319 memset(&event, 0, sizeof(event));
1320 event.event = IW_CM_EVENT_CLOSE;
1321 event.status = -ECONNRESET;
1322 if (ep->com.cm_id) {
1323 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1324 ep->com.cm_id, ep->hwtid);
1325 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
9ca6f7cf 1326 deref_cm_id(&ep->com);
793dad94 1327 set_bit(ABORT_UPCALL, &ep->com.history);
cfdda9d7
SW		 1328 }
1329}
1330
1331static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1332{
1333 struct iw_cm_event event;
1334
1335 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1336 memset(&event, 0, sizeof(event));
1337 event.event = IW_CM_EVENT_CONNECT_REPLY;
1338 event.status = status;
24d44a39
SW		 1339 memcpy(&event.local_addr, &ep->com.local_addr,
1340 sizeof(ep->com.local_addr));
1341 memcpy(&event.remote_addr, &ep->com.remote_addr,
1342 sizeof(ep->com.remote_addr));
cfdda9d7
SW		 1343
1344 if ((status == 0) || (status == -ECONNREFUSED)) {
d2fe99e8
KS		 1345 if (!ep->tried_with_mpa_v1) {
1346 /* this means MPA_v2 is used */
158c776d
H		 1347 event.ord = ep->ird;
1348 event.ird = ep->ord;
d2fe99e8
KS		 1349 event.private_data_len = ep->plen -
1350 sizeof(struct mpa_v2_conn_params);
1351 event.private_data = ep->mpa_pkt +
1352 sizeof(struct mpa_message) +
1353 sizeof(struct mpa_v2_conn_params);
1354 } else {
1355 /* this means MPA_v1 is used */
158c776d
H		 1356 event.ord = cur_max_read_depth(ep->com.dev);
1357 event.ird = cur_max_read_depth(ep->com.dev);
d2fe99e8
KS		 1358 event.private_data_len = ep->plen;
1359 event.private_data = ep->mpa_pkt +
1360 sizeof(struct mpa_message);
1361 }
cfdda9d7 1362 }
85963e4c
RD		 1363
1364 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1365 ep->hwtid, status);
793dad94 1366 set_bit(CONN_RPL_UPCALL, &ep->com.history);
85963e4c
RD		 1367 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1368
9ca6f7cf
H		 1369 if (status < 0)
1370 deref_cm_id(&ep->com);
cfdda9d7
SW		 1371}
1372
be13b2df 1373static int connect_request_upcall(struct c4iw_ep *ep)
cfdda9d7
SW		 1374{
1375 struct iw_cm_event event;
be13b2df 1376 int ret;
cfdda9d7
SW		 1377
1378 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1379 memset(&event, 0, sizeof(event));
1380 event.event = IW_CM_EVENT_CONNECT_REQUEST;
24d44a39
SW		 1381 memcpy(&event.local_addr, &ep->com.local_addr,
1382 sizeof(ep->com.local_addr));
1383 memcpy(&event.remote_addr, &ep->com.remote_addr,
1384 sizeof(ep->com.remote_addr));
cfdda9d7 1385 event.provider_data = ep;
d2fe99e8
KS		 1386 if (!ep->tried_with_mpa_v1) {
1387 /* this means MPA_v2 is used */
1388 event.ord = ep->ord;
1389 event.ird = ep->ird;
1390 event.private_data_len = ep->plen -
1391 sizeof(struct mpa_v2_conn_params);
1392 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1393 sizeof(struct mpa_v2_conn_params);
1394 } else {
1395 /* this means MPA_v1 is used. Send max supported */
4c2c5763
HS		 1396 event.ord = cur_max_read_depth(ep->com.dev);
1397 event.ird = cur_max_read_depth(ep->com.dev);
d2fe99e8
KS		 1398 event.private_data_len = ep->plen;
1399 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1400 }
be13b2df
SW		 1401 c4iw_get_ep(&ep->com);
1402 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1403 &event);
1404 if (ret)
1405 c4iw_put_ep(&ep->com);
793dad94 1406 set_bit(CONNREQ_UPCALL, &ep->com.history);
cfdda9d7 1407 c4iw_put_ep(&ep->parent_ep->com);
be13b2df 1408 return ret;
cfdda9d7
SW		 1409}
1410
1411static void established_upcall(struct c4iw_ep *ep)
1412{
1413 struct iw_cm_event event;
1414
1415 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1416 memset(&event, 0, sizeof(event));
1417 event.event = IW_CM_EVENT_ESTABLISHED;
3dd9a5dc
H		 1418 event.ird = ep->ord;
1419 event.ord = ep->ird;
cfdda9d7
SW		 1420 if (ep->com.cm_id) {
1421 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1422 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 1423 set_bit(ESTAB_UPCALL, &ep->com.history);
cfdda9d7
SW		 1424 }
1425}
1426
1427static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1428{
1429 struct cpl_rx_data_ack *req;
1430 struct sk_buff *skb;
1431 int wrlen = roundup(sizeof *req, 16);
1432
1433 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1434 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1435 if (!skb) {
1436 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1437 return 0;
1438 }
1439
b408ff28
HS		 1440 /*
1441 * If we couldn't specify the entire rcv window at connection setup
1442 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1443 * then add the overage in to the credits returned.
1444 */
d7990b0c
AB		 1445 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1446 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
b408ff28 1447
cfdda9d7
SW		 1448 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1449 memset(req, 0, wrlen);
1450 INIT_TP_WR(req, ep->hwtid);
1451 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1452 ep->hwtid));
d7990b0c 1453 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
cf7fe64a
HS		 1454 RX_DACK_CHANGE_F |
1455 RX_DACK_MODE_V(dack_mode));
d4f1a5c6 1456 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
cfdda9d7
SW		 1457 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1458 return credits;
1459}
1460
4c2c5763
HS		 1461#define RELAXED_IRD_NEGOTIATION 1
1462
f8e1e1d1
H		 1463/*
1464 * process_mpa_reply - process streaming mode MPA reply
1465 *
1466 * Returns:
1467 *
1468 * 0 upon success indicating a connect request was delivered to the ULP
1469 * or the mpa request is incomplete but valid so far.
1470 *
1471 * 1 if a failure requires the caller to close the connection.
1472 *
1473 * 2 if a failure requires the caller to abort the connection.
1474 */
cc18b939 1475static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
cfdda9d7
SW		 1476{
1477 struct mpa_message *mpa;
d2fe99e8 1478 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7 1479 u16 plen;
d2fe99e8
KS		 1480 u16 resp_ird, resp_ord;
1481 u8 rtr_mismatch = 0, insuff_ird = 0;
cfdda9d7
SW		 1482 struct c4iw_qp_attributes attrs;
1483 enum c4iw_qp_attr_mask mask;
1484 int err;
cc18b939 1485 int disconnect = 0;
cfdda9d7
SW		 1486
1487 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1488
cfdda9d7
SW		 1489 /*
1490 * If we get more than the supported amount of private data
1491 * then we must fail this connection.
1492 */
1493 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1494 err = -EINVAL;
da1cecdf 1495 goto err_stop_timer;
cfdda9d7
SW		 1496 }
1497
1498 /*
1499 * copy the new data into our accumulation buffer.
1500 */
1501 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1502 skb->len);
1503 ep->mpa_pkt_len += skb->len;
1504
1505 /*
1506 * if we don't even have the mpa message, then bail.
1507 */
1508 if (ep->mpa_pkt_len < sizeof(*mpa))
cc18b939 1509 return 0;
cfdda9d7
SW		 1510 mpa = (struct mpa_message *) ep->mpa_pkt;
1511
1512 /* Validate MPA header. */
d2fe99e8
KS		 1513 if (mpa->revision > mpa_rev) {
1514 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1515 " Received = %d\n", __func__, mpa_rev, mpa->revision);
cfdda9d7 1516 err = -EPROTO;
da1cecdf 1517 goto err_stop_timer;
cfdda9d7
SW		 1518 }
1519 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1520 err = -EPROTO;
da1cecdf 1521 goto err_stop_timer;
cfdda9d7
SW		 1522 }
1523
1524 plen = ntohs(mpa->private_data_size);
1525
1526 /*
1527 * Fail if there's too much private data.
1528 */
1529 if (plen > MPA_MAX_PRIVATE_DATA) {
1530 err = -EPROTO;
da1cecdf 1531 goto err_stop_timer;
cfdda9d7
SW		 1532 }
1533
1534 /*
1535 * If plen does not account for pkt size
1536 */
1537 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1538 err = -EPROTO;
da1cecdf 1539 goto err_stop_timer;
cfdda9d7
SW		 1540 }
1541
1542 ep->plen = (u8) plen;
1543
1544 /*
1545 * If we don't have all the pdata yet, then bail.
1546 * We'll continue process when more data arrives.
1547 */
1548 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
cc18b939 1549 return 0;
cfdda9d7
SW		 1550
1551 if (mpa->flags & MPA_REJECT) {
1552 err = -ECONNREFUSED;
da1cecdf 1553 goto err_stop_timer;
cfdda9d7
SW		 1554 }
1555
da1cecdf
H		 1556 /*
1557 * Stop mpa timer. If it expired, then
1558 * we ignore the MPA reply. process_timeout()
1559 * will abort the connection.
1560 */
1561 if (stop_ep_timer(ep))
1562 return 0;
1563
cfdda9d7
SW		 1564 /*
1565 * If we get here we have accumulated the entire mpa
1566 * start reply message including private data. And
1567 * the MPA header is valid.
1568 */
c529fb50 1569 __state_set(&ep->com, FPDU_MODE);
cfdda9d7 1570 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
cfdda9d7 1571 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS		 1572 ep->mpa_attr.version = mpa->revision;
1573 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1574
1575 if (mpa->revision == 2) {
1576 ep->mpa_attr.enhanced_rdma_conn =
1577 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1578 if (ep->mpa_attr.enhanced_rdma_conn) {
1579 mpa_v2_params = (struct mpa_v2_conn_params *)
1580 (ep->mpa_pkt + sizeof(*mpa));
1581 resp_ird = ntohs(mpa_v2_params->ird) &
1582 MPA_V2_IRD_ORD_MASK;
1583 resp_ord = ntohs(mpa_v2_params->ord) &
1584 MPA_V2_IRD_ORD_MASK;
4c2c5763
HS		 1585 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1586 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
d2fe99e8
KS		 1587
1588 /*
1589 * This is a double-check. Ideally, below checks are
1590 * not required since ird/ord stuff has been taken
1591 * care of in c4iw_accept_cr
1592 */
4c2c5763
HS		 1593 if (ep->ird < resp_ord) {
1594 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1595 ep->com.dev->rdev.lldi.max_ordird_qp)
1596 ep->ird = resp_ord;
1597 else
1598 insuff_ird = 1;
1599 } else if (ep->ird > resp_ord) {
1600 ep->ird = resp_ord;
1601 }
1602 if (ep->ord > resp_ird) {
1603 if (RELAXED_IRD_NEGOTIATION)
1604 ep->ord = resp_ird;
1605 else
1606 insuff_ird = 1;
1607 }
1608 if (insuff_ird) {
d2fe99e8
KS		 1609 err = -ENOMEM;
1610 ep->ird = resp_ord;
1611 ep->ord = resp_ird;
d2fe99e8
KS		 1612 }
1613
1614 if (ntohs(mpa_v2_params->ird) &
1615 MPA_V2_PEER2PEER_MODEL) {
1616 if (ntohs(mpa_v2_params->ord) &
1617 MPA_V2_RDMA_WRITE_RTR)
1618 ep->mpa_attr.p2p_type =
1619 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1620 else if (ntohs(mpa_v2_params->ord) &
1621 MPA_V2_RDMA_READ_RTR)
1622 ep->mpa_attr.p2p_type =
1623 FW_RI_INIT_P2PTYPE_READ_REQ;
1624 }
1625 }
1626 } else if (mpa->revision == 1)
1627 if (peer2peer)
1628 ep->mpa_attr.p2p_type = p2p_type;
1629
cfdda9d7 1630 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
d2fe99e8
KS		 1631 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1632 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1633 ep->mpa_attr.recv_marker_enabled,
1634 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1635 ep->mpa_attr.p2p_type, p2p_type);
1636
1637 /*
1638 * If responder's RTR does not match with that of initiator, assign
1639 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1640 * generated when moving QP to RTS state.
1641 * A TERM message will be sent after QP has moved to RTS state
1642 */
91018f86 1643 if ((ep->mpa_attr.version == 2) && peer2peer &&
d2fe99e8
KS		 1644 (ep->mpa_attr.p2p_type != p2p_type)) {
1645 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1646 rtr_mismatch = 1;
1647 }
cfdda9d7
SW		 1648
1649 attrs.mpa_attr = ep->mpa_attr;
1650 attrs.max_ird = ep->ird;
1651 attrs.max_ord = ep->ord;
1652 attrs.llp_stream_handle = ep;
1653 attrs.next_state = C4IW_QP_STATE_RTS;
1654
1655 mask = C4IW_QP_ATTR_NEXT_STATE |
1656 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1657 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1658
1659 /* bind QP and TID with INIT_WR */
1660 err = c4iw_modify_qp(ep->com.qp->rhp,
1661 ep->com.qp, mask, &attrs, 1);
1662 if (err)
1663 goto err;
d2fe99e8
KS		 1664
1665 /*
1666 * If responder's RTR requirement did not match with what initiator
1667 * supports, generate TERM message
1668 */
1669 if (rtr_mismatch) {
1670 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1671 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1672 attrs.ecode = MPA_NOMATCH_RTR;
1673 attrs.next_state = C4IW_QP_STATE_TERMINATE;
cc18b939 1674 attrs.send_term = 1;
d2fe99e8 1675 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939 1676 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
d2fe99e8 1677 err = -ENOMEM;
cc18b939 1678 disconnect = 1;
d2fe99e8
KS		 1679 goto out;
1680 }
1681
1682 /*
1683 * Generate TERM if initiator IRD is not sufficient for responder
1684 * provided ORD. Currently, we do the same behaviour even when
1685 * responder provided IRD is also not sufficient as regards to
1686 * initiator ORD.
1687 */
1688 if (insuff_ird) {
1689 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1690 __func__);
1691 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1692 attrs.ecode = MPA_INSUFF_IRD;
1693 attrs.next_state = C4IW_QP_STATE_TERMINATE;
cc18b939 1694 attrs.send_term = 1;
d2fe99e8 1695 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939 1696 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
d2fe99e8 1697 err = -ENOMEM;
cc18b939 1698 disconnect = 1;
d2fe99e8
KS		 1699 goto out;
1700 }
cfdda9d7 1701 goto out;
da1cecdf
H		 1702err_stop_timer:
1703 stop_ep_timer(ep);
cfdda9d7 1704err:
f8e1e1d1 1705 disconnect = 2;
cfdda9d7
SW		 1706out:
1707 connect_reply_upcall(ep, err);
cc18b939 1708 return disconnect;
cfdda9d7
SW		 1709}
1710
fd6aabe4
H		 1711/*
1712 * process_mpa_request - process streaming mode MPA request
1713 *
1714 * Returns:
1715 *
1716 * 0 upon success indicating a connect request was delivered to the ULP
1717 * or the mpa request is incomplete but valid so far.
1718 *
1719 * 1 if a failure requires the caller to close the connection.
1720 *
1721 * 2 if a failure requires the caller to abort the connection.
1722 */
1723static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
cfdda9d7
SW		 1724{
1725 struct mpa_message *mpa;
d2fe99e8 1726 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7
SW		 1727 u16 plen;
1728
1729 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1730
cfdda9d7
SW		 1731 /*
1732 * If we get more than the supported amount of private data
1733 * then we must fail this connection.
1734 */
fd6aabe4
H		 1735 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1736 goto err_stop_timer;
cfdda9d7
SW		 1737
1738 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1739
1740 /*
1741 * Copy the new data into our accumulation buffer.
1742 */
1743 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1744 skb->len);
1745 ep->mpa_pkt_len += skb->len;
1746
1747 /*
1748 * If we don't even have the mpa message, then bail.
1749 * We'll continue process when more data arrives.
1750 */
1751 if (ep->mpa_pkt_len < sizeof(*mpa))
fd6aabe4 1752 return 0;
cfdda9d7
SW		 1753
1754 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
cfdda9d7
SW		 1755 mpa = (struct mpa_message *) ep->mpa_pkt;
1756
1757 /*
1758 * Validate MPA Header.
1759 */
d2fe99e8
KS		 1760 if (mpa->revision > mpa_rev) {
1761 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1762 " Received = %d\n", __func__, mpa_rev, mpa->revision);
fd6aabe4 1763 goto err_stop_timer;
cfdda9d7
SW		 1764 }
1765
fd6aabe4
H		 1766 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1767 goto err_stop_timer;
cfdda9d7
SW		 1768
1769 plen = ntohs(mpa->private_data_size);
1770
1771 /*
1772 * Fail if there's too much private data.
1773 */
fd6aabe4
H		 1774 if (plen > MPA_MAX_PRIVATE_DATA)
1775 goto err_stop_timer;
cfdda9d7
SW		 1776
1777 /*
1778 * If plen does not account for pkt size
1779 */
fd6aabe4
H		 1780 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1781 goto err_stop_timer;
cfdda9d7
SW		 1782 ep->plen = (u8) plen;
1783
1784 /*
1785 * If we don't have all the pdata yet, then bail.
1786 */
1787 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
fd6aabe4 1788 return 0;
cfdda9d7
SW		 1789
1790 /*
1791 * If we get here we have accumulated the entire mpa
1792 * start reply message including private data.
1793 */
1794 ep->mpa_attr.initiator = 0;
1795 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1796 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1797 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS		 1798 ep->mpa_attr.version = mpa->revision;
1799 if (mpa->revision == 1)
1800 ep->tried_with_mpa_v1 = 1;
1801 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1802
1803 if (mpa->revision == 2) {
1804 ep->mpa_attr.enhanced_rdma_conn =
1805 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1806 if (ep->mpa_attr.enhanced_rdma_conn) {
1807 mpa_v2_params = (struct mpa_v2_conn_params *)
1808 (ep->mpa_pkt + sizeof(*mpa));
1809 ep->ird = ntohs(mpa_v2_params->ird) &
1810 MPA_V2_IRD_ORD_MASK;
1811 ep->ord = ntohs(mpa_v2_params->ord) &
1812 MPA_V2_IRD_ORD_MASK;
4c2c5763
HS		 1813 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1814 ep->ord);
d2fe99e8
KS		 1815 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1816 if (peer2peer) {
1817 if (ntohs(mpa_v2_params->ord) &
1818 MPA_V2_RDMA_WRITE_RTR)
1819 ep->mpa_attr.p2p_type =
1820 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1821 else if (ntohs(mpa_v2_params->ord) &
1822 MPA_V2_RDMA_READ_RTR)
1823 ep->mpa_attr.p2p_type =
1824 FW_RI_INIT_P2PTYPE_READ_REQ;
1825 }
1826 }
1827 } else if (mpa->revision == 1)
1828 if (peer2peer)
1829 ep->mpa_attr.p2p_type = p2p_type;
1830
cfdda9d7
SW		 1831 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1832 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1833 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1834 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1835 ep->mpa_attr.p2p_type);
1836
e4b76a2a
H		 1837 __state_set(&ep->com, MPA_REQ_RCVD);
1838
1839 /* drive upcall */
1840 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1841 if (ep->parent_ep->com.state != DEAD) {
1842 if (connect_request_upcall(ep))
fd6aabe4 1843 goto err_unlock_parent;
e4b76a2a
H		 1844 } else {
1845 goto err_unlock_parent;
be13b2df 1846 }
e4b76a2a 1847 mutex_unlock(&ep->parent_ep->com.mutex);
fd6aabe4
H		 1848 return 0;
1849
1850err_unlock_parent:
1851 mutex_unlock(&ep->parent_ep->com.mutex);
1852 goto err_out;
1853err_stop_timer:
1854 (void)stop_ep_timer(ep);
1855err_out:
1856 return 2;
cfdda9d7
SW		 1857}
1858
1859static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1860{
1861 struct c4iw_ep *ep;
1862 struct cpl_rx_data *hdr = cplhdr(skb);
1863 unsigned int dlen = ntohs(hdr->len);
1864 unsigned int tid = GET_TID(hdr);
793dad94 1865 __u8 status = hdr->status;
cc18b939 1866 int disconnect = 0;
cfdda9d7 1867
944661dd 1868 ep = get_ep_from_tid(dev, tid);
977116c6
SW		 1869 if (!ep)
1870 return 0;
cfdda9d7
SW		 1871 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1872 skb_pull(skb, sizeof(*hdr));
1873 skb_trim(skb, dlen);
c529fb50 1874 mutex_lock(&ep->com.mutex);
cfdda9d7 1875
cfdda9d7
SW		 1876 /* update RX credits */
1877 update_rx_credits(ep, dlen);
1878
c529fb50 1879 switch (ep->com.state) {
cfdda9d7 1880 case MPA_REQ_SENT:
55abf8df 1881 ep->rcv_seq += dlen;
cc18b939 1882 disconnect = process_mpa_reply(ep, skb);
cfdda9d7
SW		 1883 break;
1884 case MPA_REQ_WAIT:
55abf8df 1885 ep->rcv_seq += dlen;
4a4dd8db 1886 disconnect = process_mpa_request(ep, skb);
cfdda9d7 1887 break;
1557967b
VP		 1888 case FPDU_MODE: {
1889 struct c4iw_qp_attributes attrs;
1890 BUG_ON(!ep->com.qp);
e8e5b927 1891 if (status)
1557967b 1892 pr_err("%s Unexpected streaming data." \
04236df2
VP		 1893 " qpid %u ep %p state %d tid %u status %d\n",
1894 __func__, ep->com.qp->wq.sq.qid, ep,
c529fb50 1895 ep->com.state, ep->hwtid, status);
97d7ec0c 1896 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1557967b 1897 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939
SW		 1898 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1899 disconnect = 1;
cfdda9d7
SW		 1900 break;
1901 }
1557967b
VP		 1902 default:
1903 break;
1904 }
c529fb50 1905 mutex_unlock(&ep->com.mutex);
cc18b939 1906 if (disconnect)
4a4dd8db 1907 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
944661dd 1908 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 1909 return 0;
1910}
1911
1912static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1913{
1914 struct c4iw_ep *ep;
1915 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
cfdda9d7
SW		 1916 int release = 0;
1917 unsigned int tid = GET_TID(rpl);
cfdda9d7 1918
944661dd 1919 ep = get_ep_from_tid(dev, tid);
4984037b
VP		 1920 if (!ep) {
1921 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1922 return 0;
1923 }
92dd6c3d 1924 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2f5b48c3 1925 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 1926 switch (ep->com.state) {
1927 case ABORTING:
91e9c071 1928 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 1929 __state_set(&ep->com, DEAD);
1930 release = 1;
1931 break;
1932 default:
1933 printk(KERN_ERR "%s ep %p state %d\n",
1934 __func__, ep, ep->com.state);
1935 break;
1936 }
2f5b48c3 1937 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 1938
1939 if (release)
1940 release_ep_resources(ep);
944661dd 1941 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 1942 return 0;
1943}
1944
caa6c9f2 1945static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
5be78ee9
VP		 1946{
1947 struct sk_buff *skb;
1948 struct fw_ofld_connection_wr *req;
1949 unsigned int mtu_idx;
1950 int wscale;
830662f6 1951 struct sockaddr_in *sin;
b408ff28 1952 int win;
5be78ee9
VP		 1953
1954 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1955 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1956 memset(req, 0, sizeof(*req));
6c53e938 1957 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
e2ac9628 1958 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
41b4f86c
KS		 1959 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1960 ep->com.dev->rdev.lldi.ports[0],
5be78ee9 1961 ep->l2t));
170003c8 1962 sin = (struct sockaddr_in *)&ep->com.local_addr;
830662f6
VP		 1963 req->le.lport = sin->sin_port;
1964 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
170003c8 1965 sin = (struct sockaddr_in *)&ep->com.remote_addr;
830662f6
VP		 1966 req->le.pport = sin->sin_port;
1967 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
5be78ee9 1968 req->tcb.t_state_to_astid =
77a80e23
HS		 1969 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1970 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
5be78ee9 1971 req->tcb.cplrxdataack_cplpassacceptrpl =
77a80e23 1972 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
ef5d6355 1973 req->tcb.tx_max = (__force __be32) jiffies;
793dad94 1974 req->tcb.rcv_adv = htons(1);
92e7ae71 1975 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
04524a47
H		 1976 enable_tcp_timestamps,
1977 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
5be78ee9 1978 wscale = compute_wscale(rcv_win);
b408ff28
HS		 1979
1980 /*
1981 * Specify the largest window that will fit in opt0. The
1982 * remainder will be specified in the rx_data_ack.
1983 */
1984 win = ep->rcv_win >> 10;
d7990b0c
AB		 1985 if (win > RCV_BUFSIZ_M)
1986 win = RCV_BUFSIZ_M;
b408ff28 1987
6c53e938
HS		 1988 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1989 (nocong ? NO_CONG_F : 0) |
d7990b0c 1990 KEEP_ALIVE_F |
6c53e938 1991 DELACK_F |
d7990b0c
AB		 1992 WND_SCALE_V(wscale) |
1993 MSS_IDX_V(mtu_idx) |
1994 L2T_IDX_V(ep->l2t->idx) |
1995 TX_CHAN_V(ep->tx_chan) |
1996 SMAC_SEL_V(ep->smac_idx) |
ac8e4c69 1997 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 1998 ULP_MODE_V(ULP_MODE_TCPDDP) |
1999 RCV_BUFSIZ_V(win));
6c53e938
HS		 2000 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2001 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
d7990b0c 2002 RX_CHANNEL_V(0) |
6c53e938 2003 CCTRL_ECN_V(enable_ecn) |
d7990b0c 2004 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
5be78ee9 2005 if (enable_tcp_timestamps)
6c53e938 2006 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
5be78ee9 2007 if (enable_tcp_sack)
6c53e938 2008 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
5be78ee9 2009 if (wscale && enable_tcp_window_scaling)
d7990b0c
AB		 2010 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2011 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2012 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
793dad94
VP		 2013 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2014 set_bit(ACT_OFLD_CONN, &ep->com.history);
caa6c9f2 2015 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
5be78ee9
VP		 2016}
2017
cfdda9d7
SW		 2018/*
2019 * Return whether a failed active open has allocated a TID
2020 */
2021static inline int act_open_has_tid(int status)
2022{
2023 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
2024 status != CPL_ERR_ARP_MISS;
2025}
2026
7a2cea2a
SW		 2027/* Returns whether a CPL status conveys negative advice.
2028 */
2029static int is_neg_adv(unsigned int status)
2030{
2031 return status == CPL_ERR_RTX_NEG_ADVICE ||
2032 status == CPL_ERR_PERSIST_NEG_ADVICE ||
2033 status == CPL_ERR_KEEPALV_NEG_ADVICE;
2034}
2035
dd92b124
HS		 2036static char *neg_adv_str(unsigned int status)
2037{
2038 switch (status) {
2039 case CPL_ERR_RTX_NEG_ADVICE:
2040 return "Retransmit timeout";
2041 case CPL_ERR_PERSIST_NEG_ADVICE:
2042 return "Persist timeout";
2043 case CPL_ERR_KEEPALV_NEG_ADVICE:
2044 return "Keepalive timeout";
2045 default:
2046 return "Unknown";
2047 }
2048}
2049
b408ff28
HS		 2050static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2051{
2052 ep->snd_win = snd_win;
2053 ep->rcv_win = rcv_win;
2054 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
2055}
2056
793dad94
VP		 2057#define ACT_OPEN_RETRY_COUNT 2
2058
830662f6
VP		 2059static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2060 struct dst_entry *dst, struct c4iw_dev *cdev,
ac8e4c69 2061 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
830662f6
VP		 2062{
2063 struct neighbour *n;
2064 int err, step;
2065 struct net_device *pdev;
2066
2067 n = dst_neigh_lookup(dst, peer_ip);
2068 if (!n)
2069 return -ENODEV;
2070
2071 rcu_read_lock();
2072 err = -ENOMEM;
2073 if (n->dev->flags & IFF_LOOPBACK) {
2074 if (iptype == 4)
2075 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2076 else if (IS_ENABLED(CONFIG_IPV6))
2077 for_each_netdev(&init_net, pdev) {
2078 if (ipv6_chk_addr(&init_net,
2079 (struct in6_addr *)peer_ip,
2080 pdev, 1))
2081 break;
2082 }
2083 else
2084 pdev = NULL;
2085
2086 if (!pdev) {
2087 err = -ENODEV;
2088 goto out;
2089 }
2090 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
ac8e4c69 2091 n, pdev, rt_tos2priority(tos));
830662f6
VP		 2092 if (!ep->l2t)
2093 goto out;
2094 ep->mtu = pdev->mtu;
2095 ep->tx_chan = cxgb4_port_chan(pdev);
963cab50
H		 2096 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2097 cxgb4_port_viid(pdev));
830662f6
VP		 2098 step = cdev->rdev.lldi.ntxq /
2099 cdev->rdev.lldi.nchan;
2100 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2101 step = cdev->rdev.lldi.nrxq /
2102 cdev->rdev.lldi.nchan;
2103 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2104 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2105 cxgb4_port_idx(pdev) * step];
b408ff28 2106 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
830662f6
VP		 2107 dev_put(pdev);
2108 } else {
2109 pdev = get_real_dev(n->dev);
2110 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2111 n, pdev, 0);
2112 if (!ep->l2t)
2113 goto out;
2114 ep->mtu = dst_mtu(dst);
11b8e22d 2115 ep->tx_chan = cxgb4_port_chan(pdev);
963cab50
H		 2116 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2117 cxgb4_port_viid(pdev));
830662f6
VP		 2118 step = cdev->rdev.lldi.ntxq /
2119 cdev->rdev.lldi.nchan;
11b8e22d
SW		 2120 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2121 ep->ctrlq_idx = cxgb4_port_idx(pdev);
830662f6
VP		 2122 step = cdev->rdev.lldi.nrxq /
2123 cdev->rdev.lldi.nchan;
2124 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
11b8e22d 2125 cxgb4_port_idx(pdev) * step];
b408ff28 2126 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
830662f6
VP		 2127
2128 if (clear_mpa_v1) {
2129 ep->retry_with_mpa_v1 = 0;
2130 ep->tried_with_mpa_v1 = 0;
2131 }
2132 }
2133 err = 0;
2134out:
2135 rcu_read_unlock();
2136
2137 neigh_release(n);
2138
2139 return err;
2140}
2141
793dad94
VP		 2142static int c4iw_reconnect(struct c4iw_ep *ep)
2143{
2144 int err = 0;
24d44a39 2145 struct sockaddr_in *laddr = (struct sockaddr_in *)
170003c8 2146 &ep->com.cm_id->m_local_addr;
24d44a39 2147 struct sockaddr_in *raddr = (struct sockaddr_in *)
170003c8 2148 &ep->com.cm_id->m_remote_addr;
830662f6 2149 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
170003c8 2150 &ep->com.cm_id->m_local_addr;
830662f6 2151 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
170003c8 2152 &ep->com.cm_id->m_remote_addr;
830662f6
VP		 2153 int iptype;
2154 __u8 *ra;
793dad94
VP		 2155
2156 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2157 init_timer(&ep->timer);
093108cb 2158 c4iw_init_wr_wait(&ep->com.wr_wait);
793dad94
VP		 2159
2160 /*
2161 * Allocate an active TID to initiate a TCP connection.
2162 */
2163 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2164 if (ep->atid == -1) {
2165 pr_err("%s - cannot alloc atid.\n", __func__);
2166 err = -ENOMEM;
2167 goto fail2;
2168 }
2169 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2170
2171 /* find a route */
170003c8 2172 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
830662f6
VP		 2173 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
2174 raddr->sin_addr.s_addr, laddr->sin_port,
ac8e4c69 2175 raddr->sin_port, ep->com.cm_id->tos);
830662f6
VP		 2176 iptype = 4;
2177 ra = (__u8 *)&raddr->sin_addr;
2178 } else {
2179 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
2180 raddr6->sin6_addr.s6_addr,
2181 laddr6->sin6_port, raddr6->sin6_port, 0,
2182 raddr6->sin6_scope_id);
2183 iptype = 6;
2184 ra = (__u8 *)&raddr6->sin6_addr;
2185 }
2186 if (!ep->dst) {
793dad94
VP		 2187 pr_err("%s - cannot find route.\n", __func__);
2188 err = -EHOSTUNREACH;
2189 goto fail3;
2190 }
963cab50 2191 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
ac8e4c69
H		 2192 ep->com.dev->rdev.lldi.adapter_type,
2193 ep->com.cm_id->tos);
830662f6 2194 if (err) {
793dad94 2195 pr_err("%s - cannot alloc l2e.\n", __func__);
793dad94
VP		 2196 goto fail4;
2197 }
2198
2199 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2200 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2201 ep->l2t->idx);
2202
2203 state_set(&ep->com, CONNECTING);
ac8e4c69 2204 ep->tos = ep->com.cm_id->tos;
793dad94
VP		 2205
2206 /* send connect request to rnic */
2207 err = send_connect(ep);
2208 if (!err)
2209 goto out;
2210
2211 cxgb4_l2t_release(ep->l2t);
2212fail4:
2213 dst_release(ep->dst);
2214fail3:
2215 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2216 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2217fail2:
2218 /*
2219 * remember to send notification to upper layer.
2220 * We are in here so the upper layer is not aware that this is
2221 * re-connect attempt and so, upper layer is still waiting for
2222 * response of 1st connect request.
2223 */
2224 connect_reply_upcall(ep, -ECONNRESET);
2225 c4iw_put_ep(&ep->com);
2226out:
2227 return err;
2228}
2229
cfdda9d7
SW		 2230static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2231{
2232 struct c4iw_ep *ep;
2233 struct cpl_act_open_rpl *rpl = cplhdr(skb);
6c53e938
HS		 2234 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2235 ntohl(rpl->atid_status)));
cfdda9d7 2236 struct tid_info *t = dev->rdev.lldi.tids;
6c53e938 2237 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
830662f6
VP		 2238 struct sockaddr_in *la;
2239 struct sockaddr_in *ra;
2240 struct sockaddr_in6 *la6;
2241 struct sockaddr_in6 *ra6;
caa6c9f2 2242 int ret = 0;
cfdda9d7
SW		 2243
2244 ep = lookup_atid(t, atid);
170003c8
SW		 2245 la = (struct sockaddr_in *)&ep->com.local_addr;
2246 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2247 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2248 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
cfdda9d7
SW		 2249
2250 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2251 status, status2errno(status));
2252
7a2cea2a 2253 if (is_neg_adv(status)) {
179d03bb
H		 2254 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2255 __func__, atid, status, neg_adv_str(status));
2256 ep->stats.connect_neg_adv++;
2257 mutex_lock(&dev->rdev.stats.lock);
2258 dev->rdev.stats.neg_adv++;
2259 mutex_unlock(&dev->rdev.stats.lock);
cfdda9d7
SW		 2260 return 0;
2261 }
2262
793dad94
VP		 2263 set_bit(ACT_OPEN_RPL, &ep->com.history);
2264
d716a2a0
VP		 2265 /*
2266 * Log interesting failures.
2267 */
2268 switch (status) {
2269 case CPL_ERR_CONN_RESET:
2270 case CPL_ERR_CONN_TIMEDOUT:
2271 break;
5be78ee9 2272 case CPL_ERR_TCAM_FULL:
830662f6 2273 mutex_lock(&dev->rdev.stats.lock);
3b174d94 2274 dev->rdev.stats.tcam_full++;
830662f6
VP		 2275 mutex_unlock(&dev->rdev.stats.lock);
2276 if (ep->com.local_addr.ss_family == AF_INET &&
2277 dev->rdev.lldi.enable_fw_ofld_conn) {
caa6c9f2
H		 2278 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2279 ntohl(rpl->atid_status))));
2280 if (ret)
2281 goto fail;
793dad94
VP		 2282 return 0;
2283 }
2284 break;
2285 case CPL_ERR_CONN_EXIST:
2286 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2287 set_bit(ACT_RETRY_INUSE, &ep->com.history);
84cc6ac6
H		 2288 if (ep->com.remote_addr.ss_family == AF_INET6) {
2289 struct sockaddr_in6 *sin6 =
2290 (struct sockaddr_in6 *)
170003c8 2291 &ep->com.local_addr;
84cc6ac6
H		 2292 cxgb4_clip_release(
2293 ep->com.dev->rdev.lldi.ports[0],
2294 (const u32 *)
2295 &sin6->sin6_addr.s6_addr, 1);
2296 }
793dad94
VP		 2297 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2298 atid);
2299 cxgb4_free_atid(t, atid);
2300 dst_release(ep->dst);
2301 cxgb4_l2t_release(ep->l2t);
2302 c4iw_reconnect(ep);
2303 return 0;
2304 }
5be78ee9 2305 break;
d716a2a0 2306 default:
830662f6
VP		 2307 if (ep->com.local_addr.ss_family == AF_INET) {
2308 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2309 atid, status, status2errno(status),
2310 &la->sin_addr.s_addr, ntohs(la->sin_port),
2311 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2312 } else {
2313 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2314 atid, status, status2errno(status),
2315 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2316 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2317 }
d716a2a0
VP		 2318 break;
2319 }
2320
caa6c9f2 2321fail:
cfdda9d7
SW		 2322 connect_reply_upcall(ep, status2errno(status));
2323 state_set(&ep->com, DEAD);
2324
84cc6ac6
H		 2325 if (ep->com.remote_addr.ss_family == AF_INET6) {
2326 struct sockaddr_in6 *sin6 =
170003c8 2327 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 2328 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2329 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2330 }
cfdda9d7
SW		 2331 if (status && act_open_has_tid(status))
2332 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2333
793dad94 2334 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7
SW		 2335 cxgb4_free_atid(t, atid);
2336 dst_release(ep->dst);
2337 cxgb4_l2t_release(ep->l2t);
2338 c4iw_put_ep(&ep->com);
2339
2340 return 0;
2341}
2342
2343static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2344{
2345 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
cfdda9d7 2346 unsigned int stid = GET_TID(rpl);
f86fac79 2347 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
cfdda9d7
SW		 2348
2349 if (!ep) {
1cab775c
VP		 2350 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2351 goto out;
cfdda9d7
SW		 2352 }
2353 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2354 rpl->status, status2errno(rpl->status));
d9594d99 2355 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
f86fac79 2356 c4iw_put_ep(&ep->com);
1cab775c 2357out:
cfdda9d7
SW		 2358 return 0;
2359}
2360
cfdda9d7
SW		 2361static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2362{
2363 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
cfdda9d7 2364 unsigned int stid = GET_TID(rpl);
f86fac79 2365 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
cfdda9d7
SW		 2366
2367 PDBG("%s ep %p\n", __func__, ep);
d9594d99 2368 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
f86fac79 2369 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2370 return 0;
2371}
2372
9dec900c
H		 2373static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2374 struct cpl_pass_accept_req *req)
cfdda9d7
SW		 2375{
2376 struct cpl_pass_accept_rpl *rpl;
2377 unsigned int mtu_idx;
2378 u64 opt0;
2379 u32 opt2;
2380 int wscale;
92e7ae71 2381 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
b408ff28 2382 int win;
963cab50 2383 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
cfdda9d7
SW		 2384
2385 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2386 BUG_ON(skb_cloned(skb));
92e7ae71 2387
cfdda9d7 2388 skb_get(skb);
92e7ae71 2389 rpl = cplhdr(skb);
963cab50 2390 if (!is_t4(adapter_type)) {
92e7ae71
HS		 2391 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2392 rpl5 = (void *)rpl;
2393 INIT_TP_WR(rpl5, ep->hwtid);
2394 } else {
2395 skb_trim(skb, sizeof(*rpl));
2396 INIT_TP_WR(rpl, ep->hwtid);
2397 }
2398 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2399 ep->hwtid));
2400
2401 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
04524a47
H		 2402 enable_tcp_timestamps && req->tcpopt.tstamp,
2403 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
cfdda9d7 2404 wscale = compute_wscale(rcv_win);
b408ff28
HS		 2405
2406 /*
2407 * Specify the largest window that will fit in opt0. The
2408 * remainder will be specified in the rx_data_ack.
2409 */
2410 win = ep->rcv_win >> 10;
d7990b0c
AB		 2411 if (win > RCV_BUFSIZ_M)
2412 win = RCV_BUFSIZ_M;
6c53e938 2413 opt0 = (nocong ? NO_CONG_F : 0) |
d7990b0c 2414 KEEP_ALIVE_F |
6c53e938 2415 DELACK_F |
d7990b0c
AB		 2416 WND_SCALE_V(wscale) |
2417 MSS_IDX_V(mtu_idx) |
2418 L2T_IDX_V(ep->l2t->idx) |
2419 TX_CHAN_V(ep->tx_chan) |
2420 SMAC_SEL_V(ep->smac_idx) |
6c53e938 2421 DSCP_V(ep->tos >> 2) |
d7990b0c
AB		 2422 ULP_MODE_V(ULP_MODE_TCPDDP) |
2423 RCV_BUFSIZ_V(win);
2424 opt2 = RX_CHANNEL_V(0) |
2425 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
cfdda9d7
SW		 2426
2427 if (enable_tcp_timestamps && req->tcpopt.tstamp)
6c53e938 2428 opt2 |= TSTAMPS_EN_F;
cfdda9d7 2429 if (enable_tcp_sack && req->tcpopt.sack)
6c53e938 2430 opt2 |= SACK_EN_F;
cfdda9d7 2431 if (wscale && enable_tcp_window_scaling)
d7990b0c 2432 opt2 |= WND_SCALE_EN_F;
5be78ee9
VP		 2433 if (enable_ecn) {
2434 const struct tcphdr *tcph;
2435 u32 hlen = ntohl(req->hdr_len);
2436
963cab50
H		 2437 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2438 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2439 IP_HDR_LEN_G(hlen);
2440 else
2441 tcph = (const void *)(req + 1) +
2442 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
5be78ee9 2443 if (tcph->ece && tcph->cwr)
6c53e938 2444 opt2 |= CCTRL_ECN_V(1);
5be78ee9 2445 }
963cab50 2446 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
92e7ae71 2447 u32 isn = (prandom_u32() & ~7UL) - 1;
d7990b0c 2448 opt2 |= T5_OPT_2_VALID_F;
cf7fe64a 2449 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
0b741047 2450 opt2 |= T5_ISS_F;
92e7ae71
HS		 2451 rpl5 = (void *)rpl;
2452 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2453 if (peer2peer)
2454 isn += 4;
2455 rpl5->iss = cpu_to_be32(isn);
2456 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
92e5011a 2457 }
cfdda9d7 2458
cfdda9d7
SW		 2459 rpl->opt0 = cpu_to_be64(opt0);
2460 rpl->opt2 = cpu_to_be32(opt2);
d4f1a5c6 2461 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
9dec900c 2462 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
cfdda9d7 2463
9dec900c 2464 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
cfdda9d7
SW		 2465}
2466
830662f6 2467static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
cfdda9d7 2468{
830662f6 2469 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
cfdda9d7
SW		 2470 BUG_ON(skb_cloned(skb));
2471 skb_trim(skb, sizeof(struct cpl_tid_release));
cfdda9d7
SW		 2472 release_tid(&dev->rdev, hwtid, skb);
2473 return;
2474}
2475
963cab50
H		 2476static void get_4tuple(struct cpl_pass_accept_req *req, enum chip_type type,
2477 int *iptype, __u8 *local_ip, __u8 *peer_ip,
cfdda9d7
SW		 2478 __be16 *local_port, __be16 *peer_port)
2479{
963cab50
H		 2480 int eth_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2481 ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2482 T6_ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2483 int ip_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2484 IP_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2485 T6_IP_HDR_LEN_G(be32_to_cpu(req->hdr_len));
cfdda9d7 2486 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
830662f6 2487 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
cfdda9d7
SW		 2488 struct tcphdr *tcp = (struct tcphdr *)
2489 ((u8 *)(req + 1) + eth_len + ip_len);
2490
830662f6
VP		 2491 if (ip->version == 4) {
2492 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2493 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2494 ntohs(tcp->dest));
2495 *iptype = 4;
2496 memcpy(peer_ip, &ip->saddr, 4);
2497 memcpy(local_ip, &ip->daddr, 4);
2498 } else {
2499 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2500 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2501 ntohs(tcp->dest));
2502 *iptype = 6;
2503 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2504 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2505 }
cfdda9d7
SW		 2506 *peer_port = tcp->source;
2507 *local_port = tcp->dest;
2508
2509 return;
2510}
2511
2512static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2513{
793dad94 2514 struct c4iw_ep *child_ep = NULL, *parent_ep;
cfdda9d7 2515 struct cpl_pass_accept_req *req = cplhdr(skb);
6c53e938 2516 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
cfdda9d7
SW		 2517 struct tid_info *t = dev->rdev.lldi.tids;
2518 unsigned int hwtid = GET_TID(req);
2519 struct dst_entry *dst;
830662f6 2520 __u8 local_ip[16], peer_ip[16];
cfdda9d7 2521 __be16 local_port, peer_port;
84cc6ac6 2522 struct sockaddr_in6 *sin6;
3786cf18 2523 int err;
1cab775c 2524 u16 peer_mss = ntohs(req->tcpopt.mss);
830662f6 2525 int iptype;
92e7ae71 2526 unsigned short hdrs;
ac8e4c69 2527 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
cfdda9d7 2528
f86fac79 2529 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
1cab775c
VP		 2530 if (!parent_ep) {
2531 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2532 goto reject;
2533 }
1cab775c 2534
cfdda9d7 2535 if (state_read(&parent_ep->com) != LISTEN) {
6812faef 2536 PDBG("%s - listening ep not in LISTEN\n", __func__);
cfdda9d7
SW		 2537 goto reject;
2538 }
2539
963cab50
H		 2540 get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, &iptype,
2541 local_ip, peer_ip, &local_port, &peer_port);
830662f6 2542
cfdda9d7 2543 /* Find output route */
830662f6
VP		 2544 if (iptype == 4) {
2545 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2546 , __func__, parent_ep, hwtid,
2547 local_ip, peer_ip, ntohs(local_port),
2548 ntohs(peer_port), peer_mss);
2549 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2550 local_port, peer_port,
ac8e4c69 2551 tos);
830662f6
VP		 2552 } else {
2553 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2554 , __func__, parent_ep, hwtid,
2555 local_ip, peer_ip, ntohs(local_port),
2556 ntohs(peer_port), peer_mss);
2557 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
6c53e938 2558 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
830662f6
VP		 2559 ((struct sockaddr_in6 *)
2560 &parent_ep->com.local_addr)->sin6_scope_id);
2561 }
2562 if (!dst) {
cfdda9d7
SW		 2563 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2564 __func__);
2565 goto reject;
2566 }
3786cf18
DM		 2567
2568 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2569 if (!child_ep) {
2570 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
cfdda9d7
SW		 2571 __func__);
2572 dst_release(dst);
2573 goto reject;
2574 }
2575
963cab50 2576 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
ac8e4c69 2577 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
3786cf18
DM		 2578 if (err) {
2579 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
cfdda9d7 2580 __func__);
cfdda9d7 2581 dst_release(dst);
3786cf18 2582 kfree(child_ep);
cfdda9d7
SW		 2583 goto reject;
2584 }
3786cf18 2585
92e7ae71
HS		 2586 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2587 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2588 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2589 child_ep->mtu = peer_mss + hdrs;
1cab775c 2590
cfdda9d7
SW		 2591 state_set(&child_ep->com, CONNECTING);
2592 child_ep->com.dev = dev;
2593 child_ep->com.cm_id = NULL;
5b6b8fe6 2594
830662f6
VP		 2595 if (iptype == 4) {
2596 struct sockaddr_in *sin = (struct sockaddr_in *)
170003c8 2597 &child_ep->com.local_addr;
5b6b8fe6 2598
830662f6
VP		 2599 sin->sin_family = PF_INET;
2600 sin->sin_port = local_port;
2601 sin->sin_addr.s_addr = *(__be32 *)local_ip;
5b6b8fe6
SW		 2602
2603 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2604 sin->sin_family = PF_INET;
2605 sin->sin_port = ((struct sockaddr_in *)
2606 &parent_ep->com.local_addr)->sin_port;
2607 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2608
170003c8 2609 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
830662f6
VP		 2610 sin->sin_family = PF_INET;
2611 sin->sin_port = peer_port;
2612 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2613 } else {
170003c8 2614 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
830662f6
VP		 2615 sin6->sin6_family = PF_INET6;
2616 sin6->sin6_port = local_port;
2617 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
5b6b8fe6
SW		 2618
2619 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2620 sin6->sin6_family = PF_INET6;
2621 sin6->sin6_port = ((struct sockaddr_in6 *)
2622 &parent_ep->com.local_addr)->sin6_port;
2623 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2624
170003c8 2625 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
830662f6
VP		 2626 sin6->sin6_family = PF_INET6;
2627 sin6->sin6_port = peer_port;
2628 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2629 }
5b6b8fe6 2630
cfdda9d7
SW		 2631 c4iw_get_ep(&parent_ep->com);
2632 child_ep->parent_ep = parent_ep;
ac8e4c69 2633 child_ep->tos = tos;
cfdda9d7
SW		 2634 child_ep->dst = dst;
2635 child_ep->hwtid = hwtid;
cfdda9d7
SW		 2636
2637 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
3786cf18 2638 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
cfdda9d7
SW		 2639
2640 init_timer(&child_ep->timer);
2641 cxgb4_insert_tid(t, child_ep, hwtid);
944661dd 2642 insert_ep_tid(child_ep);
9dec900c
H		 2643 if (accept_cr(child_ep, skb, req)) {
2644 c4iw_put_ep(&parent_ep->com);
2645 release_ep_resources(child_ep);
2646 } else {
2647 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2648 }
84cc6ac6 2649 if (iptype == 6) {
170003c8 2650 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
84cc6ac6
H		 2651 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2652 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2653 }
cfdda9d7
SW		 2654 goto out;
2655reject:
830662f6 2656 reject_cr(dev, hwtid, skb);
f86fac79
H		 2657 if (parent_ep)
2658 c4iw_put_ep(&parent_ep->com);
cfdda9d7
SW		 2659out:
2660 return 0;
2661}
2662
2663static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2664{
2665 struct c4iw_ep *ep;
2666 struct cpl_pass_establish *req = cplhdr(skb);
cfdda9d7 2667 unsigned int tid = GET_TID(req);
fef4422d 2668 int ret;
cfdda9d7 2669
944661dd 2670 ep = get_ep_from_tid(dev, tid);
cfdda9d7
SW		 2671 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2672 ep->snd_seq = be32_to_cpu(req->snd_isn);
2673 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2674
1cab775c
VP		 2675 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2676 ntohs(req->tcp_opt));
2677
cfdda9d7
SW		 2678 set_emss(ep, ntohs(req->tcp_opt));
2679
2680 dst_confirm(ep->dst);
fef4422d
H		 2681 mutex_lock(&ep->com.mutex);
2682 ep->com.state = MPA_REQ_WAIT;
cfdda9d7 2683 start_ep_timer(ep);
793dad94 2684 set_bit(PASS_ESTAB, &ep->com.history);
fef4422d
H		 2685 ret = send_flowc(ep, skb);
2686 mutex_unlock(&ep->com.mutex);
2687 if (ret)
2688 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
944661dd 2689 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2690
2691 return 0;
2692}
2693
2694static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2695{
2696 struct cpl_peer_close *hdr = cplhdr(skb);
2697 struct c4iw_ep *ep;
2698 struct c4iw_qp_attributes attrs;
cfdda9d7
SW		 2699 int disconnect = 1;
2700 int release = 0;
cfdda9d7 2701 unsigned int tid = GET_TID(hdr);
8da7e7a5 2702 int ret;
cfdda9d7 2703
944661dd
H		 2704 ep = get_ep_from_tid(dev, tid);
2705 if (!ep)
2706 return 0;
2707
cfdda9d7
SW		 2708 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2709 dst_confirm(ep->dst);
2710
793dad94 2711 set_bit(PEER_CLOSE, &ep->com.history);
2f5b48c3 2712 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2713 switch (ep->com.state) {
2714 case MPA_REQ_WAIT:
2715 __state_set(&ep->com, CLOSING);
2716 break;
2717 case MPA_REQ_SENT:
2718 __state_set(&ep->com, CLOSING);
2719 connect_reply_upcall(ep, -ECONNRESET);
2720 break;
2721 case MPA_REQ_RCVD:
2722
2723 /*
2724 * We're gonna mark this puppy DEAD, but keep
2725 * the reference on it until the ULP accepts or
2726 * rejects the CR. Also wake up anyone waiting
2727 * in rdma connection migration (see c4iw_accept_cr()).
2728 */
2729 __state_set(&ep->com, CLOSING);
cfdda9d7 2730 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2731 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2732 break;
2733 case MPA_REP_SENT:
2734 __state_set(&ep->com, CLOSING);
cfdda9d7 2735 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2736 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW		 2737 break;
2738 case FPDU_MODE:
ca5a2202 2739 start_ep_timer(ep);
cfdda9d7 2740 __state_set(&ep->com, CLOSING);
30c95c2d 2741 attrs.next_state = C4IW_QP_STATE_CLOSING;
8da7e7a5 2742 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
30c95c2d 2743 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
8da7e7a5
SW		 2744 if (ret != -ECONNRESET) {
2745 peer_close_upcall(ep);
2746 disconnect = 1;
2747 }
cfdda9d7
SW		 2748 break;
2749 case ABORTING:
2750 disconnect = 0;
2751 break;
2752 case CLOSING:
2753 __state_set(&ep->com, MORIBUND);
2754 disconnect = 0;
2755 break;
2756 case MORIBUND:
b33bd0cb 2757 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2758 if (ep->com.cm_id && ep->com.qp) {
2759 attrs.next_state = C4IW_QP_STATE_IDLE;
2760 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2761 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2762 }
be13b2df 2763 close_complete_upcall(ep, 0);
cfdda9d7
SW		 2764 __state_set(&ep->com, DEAD);
2765 release = 1;
2766 disconnect = 0;
2767 break;
2768 case DEAD:
2769 disconnect = 0;
2770 break;
2771 default:
2772 BUG_ON(1);
2773 }
2f5b48c3 2774 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2775 if (disconnect)
2776 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2777 if (release)
2778 release_ep_resources(ep);
944661dd 2779 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2780 return 0;
2781}
2782
cfdda9d7
SW		 2783static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2784{
2785 struct cpl_abort_req_rss *req = cplhdr(skb);
2786 struct c4iw_ep *ep;
2787 struct cpl_abort_rpl *rpl;
2788 struct sk_buff *rpl_skb;
2789 struct c4iw_qp_attributes attrs;
2790 int ret;
2791 int release = 0;
cfdda9d7 2792 unsigned int tid = GET_TID(req);
cfdda9d7 2793
944661dd
H		 2794 ep = get_ep_from_tid(dev, tid);
2795 if (!ep)
2796 return 0;
2797
7a2cea2a 2798 if (is_neg_adv(req->status)) {
179d03bb
H		 2799 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2800 __func__, ep->hwtid, req->status,
2801 neg_adv_str(req->status));
2802 ep->stats.abort_neg_adv++;
2803 mutex_lock(&dev->rdev.stats.lock);
2804 dev->rdev.stats.neg_adv++;
2805 mutex_unlock(&dev->rdev.stats.lock);
944661dd 2806 goto deref_ep;
cfdda9d7 2807 }
cfdda9d7
SW		 2808 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2809 ep->com.state);
793dad94 2810 set_bit(PEER_ABORT, &ep->com.history);
2f5b48c3
SW		 2811
2812 /*
2813 * Wake up any threads in rdma_init() or rdma_fini().
d2fe99e8
KS		 2814 * However, this is not needed if com state is just
2815 * MPA_REQ_SENT
2f5b48c3 2816 */
d2fe99e8
KS		 2817 if (ep->com.state != MPA_REQ_SENT)
2818 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2f5b48c3
SW		 2819
2820 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 2821 switch (ep->com.state) {
2822 case CONNECTING:
9dec900c 2823 c4iw_put_ep(&ep->parent_ep->com);
cfdda9d7
SW		 2824 break;
2825 case MPA_REQ_WAIT:
b33bd0cb 2826 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2827 break;
2828 case MPA_REQ_SENT:
b33bd0cb 2829 (void)stop_ep_timer(ep);
fe7e0a4d 2830 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
d2fe99e8
KS		 2831 connect_reply_upcall(ep, -ECONNRESET);
2832 else {
2833 /*
2834 * we just don't send notification upwards because we
2835 * want to retry with mpa_v1 without upper layers even
2836 * knowing it.
2837 *
2838 * do some housekeeping so as to re-initiate the
2839 * connection
2840 */
2841 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2842 mpa_rev);
2843 ep->retry_with_mpa_v1 = 1;
2844 }
cfdda9d7
SW		 2845 break;
2846 case MPA_REP_SENT:
cfdda9d7
SW		 2847 break;
2848 case MPA_REQ_RCVD:
cfdda9d7
SW		 2849 break;
2850 case MORIBUND:
2851 case CLOSING:
ca5a2202 2852 stop_ep_timer(ep);
cfdda9d7
SW		 2853 /*FALLTHROUGH*/
2854 case FPDU_MODE:
2855 if (ep->com.cm_id && ep->com.qp) {
2856 attrs.next_state = C4IW_QP_STATE_ERROR;
2857 ret = c4iw_modify_qp(ep->com.qp->rhp,
2858 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2859 &attrs, 1);
2860 if (ret)
2861 printk(KERN_ERR MOD
2862 "%s - qp <- error failed!\n",
2863 __func__);
2864 }
2865 peer_abort_upcall(ep);
2866 break;
2867 case ABORTING:
2868 break;
2869 case DEAD:
2870 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2f5b48c3 2871 mutex_unlock(&ep->com.mutex);
944661dd 2872 goto deref_ep;
cfdda9d7
SW		 2873 default:
2874 BUG_ON(1);
2875 break;
2876 }
2877 dst_confirm(ep->dst);
2878 if (ep->com.state != ABORTING) {
2879 __state_set(&ep->com, DEAD);
d2fe99e8
KS		 2880 /* we don't release if we want to retry with mpa_v1 */
2881 if (!ep->retry_with_mpa_v1)
2882 release = 1;
cfdda9d7 2883 }
2f5b48c3 2884 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2885
2886 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2887 if (!rpl_skb) {
2888 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2889 __func__);
2890 release = 1;
2891 goto out;
2892 }
2893 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2894 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2895 INIT_TP_WR(rpl, ep->hwtid);
2896 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2897 rpl->cmd = CPL_ABORT_NO_RST;
2898 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2899out:
cfdda9d7
SW		 2900 if (release)
2901 release_ep_resources(ep);
fe7e0a4d 2902 else if (ep->retry_with_mpa_v1) {
84cc6ac6
H		 2903 if (ep->com.remote_addr.ss_family == AF_INET6) {
2904 struct sockaddr_in6 *sin6 =
2905 (struct sockaddr_in6 *)
170003c8 2906 &ep->com.local_addr;
84cc6ac6
H		 2907 cxgb4_clip_release(
2908 ep->com.dev->rdev.lldi.ports[0],
2909 (const u32 *)&sin6->sin6_addr.s6_addr,
2910 1);
2911 }
fe7e0a4d 2912 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
d2fe99e8
KS		 2913 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2914 dst_release(ep->dst);
2915 cxgb4_l2t_release(ep->l2t);
2916 c4iw_reconnect(ep);
2917 }
2918
944661dd
H		 2919deref_ep:
2920 c4iw_put_ep(&ep->com);
2921 /* Dereferencing ep, referenced in peer_abort_intr() */
2922 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2923 return 0;
2924}
2925
2926static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2927{
2928 struct c4iw_ep *ep;
2929 struct c4iw_qp_attributes attrs;
2930 struct cpl_close_con_rpl *rpl = cplhdr(skb);
cfdda9d7 2931 int release = 0;
cfdda9d7 2932 unsigned int tid = GET_TID(rpl);
cfdda9d7 2933
944661dd
H		 2934 ep = get_ep_from_tid(dev, tid);
2935 if (!ep)
2936 return 0;
cfdda9d7
SW		 2937
2938 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2939 BUG_ON(!ep);
2940
2941 /* The cm_id may be null if we failed to connect */
2f5b48c3 2942 mutex_lock(&ep->com.mutex);
9ca6f7cf 2943 set_bit(CLOSE_CON_RPL, &ep->com.history);
cfdda9d7
SW		 2944 switch (ep->com.state) {
2945 case CLOSING:
2946 __state_set(&ep->com, MORIBUND);
2947 break;
2948 case MORIBUND:
b33bd0cb 2949 (void)stop_ep_timer(ep);
cfdda9d7
SW		 2950 if ((ep->com.cm_id) && (ep->com.qp)) {
2951 attrs.next_state = C4IW_QP_STATE_IDLE;
2952 c4iw_modify_qp(ep->com.qp->rhp,
2953 ep->com.qp,
2954 C4IW_QP_ATTR_NEXT_STATE,
2955 &attrs, 1);
2956 }
be13b2df 2957 close_complete_upcall(ep, 0);
cfdda9d7
SW		 2958 __state_set(&ep->com, DEAD);
2959 release = 1;
2960 break;
2961 case ABORTING:
2962 case DEAD:
2963 break;
2964 default:
2965 BUG_ON(1);
2966 break;
2967 }
2f5b48c3 2968 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 2969 if (release)
2970 release_ep_resources(ep);
944661dd 2971 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 2972 return 0;
2973}
2974
2975static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2976{
0e42c1f4 2977 struct cpl_rdma_terminate *rpl = cplhdr(skb);
0e42c1f4
SW		 2978 unsigned int tid = GET_TID(rpl);
2979 struct c4iw_ep *ep;
2980 struct c4iw_qp_attributes attrs;
cfdda9d7 2981
944661dd 2982 ep = get_ep_from_tid(dev, tid);
0e42c1f4 2983 BUG_ON(!ep);
cfdda9d7 2984
30c95c2d 2985 if (ep && ep->com.qp) {
0e42c1f4
SW		 2986 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2987 ep->com.qp->wq.sq.qid);
2988 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2989 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2990 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2991 } else
30c95c2d 2992 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
944661dd 2993 c4iw_put_ep(&ep->com);
cfdda9d7 2994
cfdda9d7
SW		 2995 return 0;
2996}
2997
2998/*
2999 * Upcall from the adapter indicating data has been transmitted.
3000 * For us its just the single MPA request or reply. We can now free
3001 * the skb holding the mpa message.
3002 */
3003static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3004{
3005 struct c4iw_ep *ep;
3006 struct cpl_fw4_ack *hdr = cplhdr(skb);
3007 u8 credits = hdr->credits;
3008 unsigned int tid = GET_TID(hdr);
cfdda9d7
SW		 3009
3010
944661dd
H		 3011 ep = get_ep_from_tid(dev, tid);
3012 if (!ep)
3013 return 0;
cfdda9d7
SW		 3014 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
3015 if (credits == 0) {
aa1ad260
JP		 3016 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
3017 __func__, ep, ep->hwtid, state_read(&ep->com));
944661dd 3018 goto out;
cfdda9d7
SW		 3019 }
3020
3021 dst_confirm(ep->dst);
3022 if (ep->mpa_skb) {
3023 PDBG("%s last streaming msg ack ep %p tid %u state %u "
3024 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
3025 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
3026 kfree_skb(ep->mpa_skb);
3027 ep->mpa_skb = NULL;
e4b76a2a
H		 3028 mutex_lock(&ep->com.mutex);
3029 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3030 stop_ep_timer(ep);
3031 mutex_unlock(&ep->com.mutex);
cfdda9d7 3032 }
944661dd
H		 3033out:
3034 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 3035 return 0;
3036}
3037
cfdda9d7
SW		 3038int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3039{
a7db89eb
SW		 3040 int err = 0;
3041 int disconnect = 0;
cfdda9d7
SW		 3042 struct c4iw_ep *ep = to_ep(cm_id);
3043 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3044
a7db89eb 3045 mutex_lock(&ep->com.mutex);
e8667a9b 3046 if (ep->com.state != MPA_REQ_RCVD) {
a7db89eb 3047 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 3048 c4iw_put_ep(&ep->com);
3049 return -ECONNRESET;
3050 }
793dad94 3051 set_bit(ULP_REJECT, &ep->com.history);
cfdda9d7 3052 if (mpa_rev == 0)
eaf4c6d4 3053 disconnect = 2;
cfdda9d7
SW		 3054 else {
3055 err = send_mpa_reject(ep, pdata, pdata_len);
a7db89eb 3056 disconnect = 1;
cfdda9d7 3057 }
a7db89eb 3058 mutex_unlock(&ep->com.mutex);
e4b76a2a
H		 3059 if (disconnect) {
3060 stop_ep_timer(ep);
eaf4c6d4 3061 err = c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
e4b76a2a 3062 }
cfdda9d7
SW		 3063 c4iw_put_ep(&ep->com);
3064 return 0;
3065}
3066
3067int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3068{
3069 int err;
3070 struct c4iw_qp_attributes attrs;
3071 enum c4iw_qp_attr_mask mask;
3072 struct c4iw_ep *ep = to_ep(cm_id);
3073 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3074 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
eaf4c6d4 3075 int abort = 0;
cfdda9d7
SW		 3076
3077 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
a7db89eb
SW		 3078
3079 mutex_lock(&ep->com.mutex);
e8667a9b 3080 if (ep->com.state != MPA_REQ_RCVD) {
cfdda9d7 3081 err = -ECONNRESET;
eaf4c6d4 3082 goto err_out;
cfdda9d7
SW		 3083 }
3084
cfdda9d7
SW		 3085 BUG_ON(!qp);
3086
793dad94 3087 set_bit(ULP_ACCEPT, &ep->com.history);
4c2c5763
HS		 3088 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3089 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
cfdda9d7 3090 err = -EINVAL;
eaf4c6d4 3091 goto err_abort;
cfdda9d7
SW		 3092 }
3093
d2fe99e8
KS		 3094 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3095 if (conn_param->ord > ep->ird) {
4c2c5763
HS		 3096 if (RELAXED_IRD_NEGOTIATION) {
3097 ep->ord = ep->ird;
3098 } else {
3099 ep->ird = conn_param->ird;
3100 ep->ord = conn_param->ord;
3101 send_mpa_reject(ep, conn_param->private_data,
3102 conn_param->private_data_len);
4c2c5763 3103 err = -ENOMEM;
eaf4c6d4 3104 goto err_abort;
4c2c5763 3105 }
d2fe99e8 3106 }
4c2c5763
HS		 3107 if (conn_param->ird < ep->ord) {
3108 if (RELAXED_IRD_NEGOTIATION &&
3109 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3110 conn_param->ird = ep->ord;
3111 } else {
d2fe99e8 3112 err = -ENOMEM;
eaf4c6d4 3113 goto err_abort;
d2fe99e8
KS		 3114 }
3115 }
d2fe99e8 3116 }
cfdda9d7
SW		 3117 ep->ird = conn_param->ird;
3118 ep->ord = conn_param->ord;
3119
4c2c5763 3120 if (ep->mpa_attr.version == 1) {
d2fe99e8
KS		 3121 if (peer2peer && ep->ird == 0)
3122 ep->ird = 1;
4c2c5763
HS		 3123 } else {
3124 if (peer2peer &&
3125 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
f57b780c 3126 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
4c2c5763
HS		 3127 ep->ird = 1;
3128 }
cfdda9d7
SW		 3129
3130 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3131
d2fe99e8 3132 ep->com.cm_id = cm_id;
9ca6f7cf 3133 ref_cm_id(&ep->com);
d2fe99e8 3134 ep->com.qp = qp;
325abead 3135 ref_qp(ep);
d2fe99e8 3136
cfdda9d7
SW		 3137 /* bind QP to EP and move to RTS */
3138 attrs.mpa_attr = ep->mpa_attr;
3139 attrs.max_ird = ep->ird;
3140 attrs.max_ord = ep->ord;
3141 attrs.llp_stream_handle = ep;
3142 attrs.next_state = C4IW_QP_STATE_RTS;
3143
3144 /* bind QP and TID with INIT_WR */
3145 mask = C4IW_QP_ATTR_NEXT_STATE |
3146 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3147 C4IW_QP_ATTR_MPA_ATTR |
3148 C4IW_QP_ATTR_MAX_IRD |
3149 C4IW_QP_ATTR_MAX_ORD;
3150
3151 err = c4iw_modify_qp(ep->com.qp->rhp,
3152 ep->com.qp, mask, &attrs, 1);
3153 if (err)
eaf4c6d4 3154 goto err_deref_cm_id;
e4b76a2a
H		 3155
3156 set_bit(STOP_MPA_TIMER, &ep->com.flags);
cfdda9d7
SW		 3157 err = send_mpa_reply(ep, conn_param->private_data,
3158 conn_param->private_data_len);
3159 if (err)
eaf4c6d4 3160 goto err_deref_cm_id;
cfdda9d7 3161
a7db89eb 3162 __state_set(&ep->com, FPDU_MODE);
cfdda9d7 3163 established_upcall(ep);
a7db89eb 3164 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW		 3165 c4iw_put_ep(&ep->com);
3166 return 0;
eaf4c6d4 3167err_deref_cm_id:
9ca6f7cf 3168 deref_cm_id(&ep->com);
eaf4c6d4
H		 3169err_abort:
3170 abort = 1;
3171err_out:
a7db89eb 3172 mutex_unlock(&ep->com.mutex);
eaf4c6d4
H		 3173 if (abort)
3174 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
cfdda9d7
SW		 3175 c4iw_put_ep(&ep->com);
3176 return err;
3177}
3178
830662f6
VP		 3179static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3180{
3181 struct in_device *ind;
3182 int found = 0;
170003c8
SW		 3183 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3184 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
830662f6
VP		 3185
3186 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3187 if (!ind)
3188 return -EADDRNOTAVAIL;
3189 for_primary_ifa(ind) {
3190 laddr->sin_addr.s_addr = ifa->ifa_address;
3191 raddr->sin_addr.s_addr = ifa->ifa_address;
3192 found = 1;
3193 break;
3194 }
3195 endfor_ifa(ind);
3196 in_dev_put(ind);
3197 return found ? 0 : -EADDRNOTAVAIL;
3198}
3199
3200static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3201 unsigned char banned_flags)
3202{
3203 struct inet6_dev *idev;
3204 int err = -EADDRNOTAVAIL;
3205
3206 rcu_read_lock();
3207 idev = __in6_dev_get(dev);
3208 if (idev != NULL) {
3209 struct inet6_ifaddr *ifp;
3210
3211 read_lock_bh(&idev->lock);
3212 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3213 if (ifp->scope == IFA_LINK &&
3214 !(ifp->flags & banned_flags)) {
3215 memcpy(addr, &ifp->addr, 16);
3216 err = 0;
3217 break;
3218 }
3219 }
3220 read_unlock_bh(&idev->lock);
3221 }
3222 rcu_read_unlock();
3223 return err;
3224}
3225
3226static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3227{
3228 struct in6_addr uninitialized_var(addr);
170003c8
SW		 3229 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3230 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
830662f6 3231
54b9a96f 3232 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
830662f6
VP		 3233 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3234 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3235 return 0;
3236 }
3237 return -EADDRNOTAVAIL;
3238}
3239
cfdda9d7
SW		 3240int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3241{
cfdda9d7
SW		 3242 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3243 struct c4iw_ep *ep;
3786cf18 3244 int err = 0;
9eccfe10
SW		 3245 struct sockaddr_in *laddr;
3246 struct sockaddr_in *raddr;
3247 struct sockaddr_in6 *laddr6;
3248 struct sockaddr_in6 *raddr6;
830662f6
VP		 3249 __u8 *ra;
3250 int iptype;
cfdda9d7 3251
4c2c5763
HS		 3252 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3253 (conn_param->ird > cur_max_read_depth(dev))) {
be4c9bad
RD		 3254 err = -EINVAL;
3255 goto out;
3256 }
cfdda9d7
SW		 3257 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3258 if (!ep) {
3259 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3260 err = -ENOMEM;
3261 goto out;
3262 }
3263 init_timer(&ep->timer);
3264 ep->plen = conn_param->private_data_len;
3265 if (ep->plen)
3266 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3267 conn_param->private_data, ep->plen);
3268 ep->ird = conn_param->ird;
3269 ep->ord = conn_param->ord;
3270
3271 if (peer2peer && ep->ord == 0)
3272 ep->ord = 1;
3273
cfdda9d7 3274 ep->com.cm_id = cm_id;
9ca6f7cf
H		 3275 ref_cm_id(&ep->com);
3276 ep->com.dev = dev;
cfdda9d7 3277 ep->com.qp = get_qhp(dev, conn_param->qpn);
830662f6
VP		 3278 if (!ep->com.qp) {
3279 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3280 err = -EINVAL;
9eccfe10 3281 goto fail1;
830662f6 3282 }
325abead 3283 ref_qp(ep);
cfdda9d7
SW		 3284 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3285 ep->com.qp, cm_id);
3286
3287 /*
3288 * Allocate an active TID to initiate a TCP connection.
3289 */
3290 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3291 if (ep->atid == -1) {
3292 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3293 err = -ENOMEM;
9eccfe10 3294 goto fail1;
cfdda9d7 3295 }
793dad94 3296 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
cfdda9d7 3297
170003c8 3298 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
9eccfe10 3299 sizeof(ep->com.local_addr));
170003c8 3300 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
9eccfe10
SW		 3301 sizeof(ep->com.remote_addr));
3302
170003c8
SW		 3303 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3304 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3305 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3306 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
9eccfe10 3307
170003c8 3308 if (cm_id->m_remote_addr.ss_family == AF_INET) {
830662f6
VP		 3309 iptype = 4;
3310 ra = (__u8 *)&raddr->sin_addr;
cfdda9d7 3311
830662f6
VP		 3312 /*
3313 * Handle loopback requests to INADDR_ANY.
3314 */
ba987e51 3315 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
830662f6
VP		 3316 err = pick_local_ipaddrs(dev, cm_id);
3317 if (err)
9eccfe10 3318 goto fail1;
830662f6
VP		 3319 }
3320
3321 /* find a route */
3322 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3323 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3324 ra, ntohs(raddr->sin_port));
3325 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3326 raddr->sin_addr.s_addr, laddr->sin_port,
ac8e4c69 3327 raddr->sin_port, cm_id->tos);
830662f6
VP		 3328 } else {
3329 iptype = 6;
3330 ra = (__u8 *)&raddr6->sin6_addr;
3331
3332 /*
3333 * Handle loopback requests to INADDR_ANY.
3334 */
3335 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3336 err = pick_local_ip6addrs(dev, cm_id);
3337 if (err)
9eccfe10 3338 goto fail1;
830662f6
VP		 3339 }
3340
3341 /* find a route */
3342 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3343 __func__, laddr6->sin6_addr.s6_addr,
3344 ntohs(laddr6->sin6_port),
3345 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3346 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3347 raddr6->sin6_addr.s6_addr,
3348 laddr6->sin6_port, raddr6->sin6_port, 0,
3349 raddr6->sin6_scope_id);
3350 }
3351 if (!ep->dst) {
cfdda9d7
SW		 3352 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3353 err = -EHOSTUNREACH;
9eccfe10 3354 goto fail2;
cfdda9d7 3355 }
cfdda9d7 3356
963cab50 3357 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
ac8e4c69 3358 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3786cf18 3359 if (err) {
cfdda9d7 3360 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
9eccfe10 3361 goto fail3;
cfdda9d7
SW		 3362 }
3363
3364 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3365 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3366 ep->l2t->idx);
3367
3368 state_set(&ep->com, CONNECTING);
ac8e4c69 3369 ep->tos = cm_id->tos;
cfdda9d7
SW		 3370
3371 /* send connect request to rnic */
3372 err = send_connect(ep);
3373 if (!err)
3374 goto out;
3375
3376 cxgb4_l2t_release(ep->l2t);
cfdda9d7 3377fail3:
9eccfe10
SW		 3378 dst_release(ep->dst);
3379fail2:
793dad94 3380 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
cfdda9d7 3381 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
9eccfe10 3382fail1:
9ca6f7cf 3383 deref_cm_id(&ep->com);
cfdda9d7
SW		 3384 c4iw_put_ep(&ep->com);
3385out:
3386 return err;
3387}
3388
830662f6
VP		 3389static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3390{
3391 int err;
9eccfe10 3392 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
170003c8 3393 &ep->com.local_addr;
830662f6 3394
28de1f74
H		 3395 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3396 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3397 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3398 if (err)
3399 return err;
3400 }
830662f6
VP		 3401 c4iw_init_wr_wait(&ep->com.wr_wait);
3402 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3403 ep->stid, &sin6->sin6_addr,
3404 sin6->sin6_port,
3405 ep->com.dev->rdev.lldi.rxq_ids[0]);
3406 if (!err)
3407 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3408 &ep->com.wr_wait,
3409 0, 0, __func__);
e6b11163
H		 3410 else if (err > 0)
3411 err = net_xmit_errno(err);
28de1f74
H		 3412 if (err) {
3413 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3414 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
830662f6
VP		 3415 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3416 err, ep->stid,
3417 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
28de1f74 3418 }
830662f6
VP		 3419 return err;
3420}
3421
3422static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3423{
3424 int err;
9eccfe10 3425 struct sockaddr_in *sin = (struct sockaddr_in *)
170003c8 3426 &ep->com.local_addr;
830662f6
VP		 3427
3428 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3429 do {
3430 err = cxgb4_create_server_filter(
3431 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3432 sin->sin_addr.s_addr, sin->sin_port, 0,
3433 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3434 if (err == -EBUSY) {
99718e59
H		 3435 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3436 err = -EIO;
3437 break;
3438 }
830662f6
VP		 3439 set_current_state(TASK_UNINTERRUPTIBLE);
3440 schedule_timeout(usecs_to_jiffies(100));
3441 }
3442 } while (err == -EBUSY);
3443 } else {
3444 c4iw_init_wr_wait(&ep->com.wr_wait);
3445 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3446 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3447 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3448 if (!err)
3449 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3450 &ep->com.wr_wait,
3451 0, 0, __func__);
e6b11163
H		 3452 else if (err > 0)
3453 err = net_xmit_errno(err);
830662f6
VP		 3454 }
3455 if (err)
3456 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3457 , err, ep->stid,
3458 &sin->sin_addr, ntohs(sin->sin_port));
3459 return err;
3460}
3461
cfdda9d7
SW		 3462int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3463{
3464 int err = 0;
3465 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3466 struct c4iw_listen_ep *ep;
3467
cfdda9d7
SW		 3468 might_sleep();
3469
3470 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3471 if (!ep) {
3472 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3473 err = -ENOMEM;
3474 goto fail1;
3475 }
3476 PDBG("%s ep %p\n", __func__, ep);
cfdda9d7 3477 ep->com.cm_id = cm_id;
9ca6f7cf 3478 ref_cm_id(&ep->com);
cfdda9d7
SW		 3479 ep->com.dev = dev;
3480 ep->backlog = backlog;
170003c8 3481 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
24d44a39 3482 sizeof(ep->com.local_addr));
cfdda9d7
SW		 3483
3484 /*
3485 * Allocate a server TID.
3486 */
8c044690
KS		 3487 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3488 ep->com.local_addr.ss_family == AF_INET)
830662f6 3489 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
170003c8 3490 cm_id->m_local_addr.ss_family, ep);
1cab775c 3491 else
830662f6 3492 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
170003c8 3493 cm_id->m_local_addr.ss_family, ep);
1cab775c 3494
cfdda9d7 3495 if (ep->stid == -1) {
be4c9bad 3496 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
cfdda9d7
SW		 3497 err = -ENOMEM;
3498 goto fail2;
3499 }
793dad94 3500 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
9eccfe10 3501
170003c8
SW		 3502 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3503 sizeof(ep->com.local_addr));
9eccfe10 3504
cfdda9d7 3505 state_set(&ep->com, LISTEN);
830662f6
VP		 3506 if (ep->com.local_addr.ss_family == AF_INET)
3507 err = create_server4(dev, ep);
3508 else
3509 err = create_server6(dev, ep);
cfdda9d7
SW		 3510 if (!err) {
3511 cm_id->provider_data = ep;
3512 goto out;
3513 }
9eccfe10 3514
830662f6
VP		 3515 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3516 ep->com.local_addr.ss_family);
cfdda9d7 3517fail2:
9ca6f7cf 3518 deref_cm_id(&ep->com);
cfdda9d7
SW		 3519 c4iw_put_ep(&ep->com);
3520fail1:
3521out:
3522 return err;
3523}
3524
3525int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3526{
3527 int err;
3528 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3529
3530 PDBG("%s ep %p\n", __func__, ep);
3531
3532 might_sleep();
3533 state_set(&ep->com, DEAD);
830662f6
VP		 3534 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3535 ep->com.local_addr.ss_family == AF_INET) {
1cab775c
VP		 3536 err = cxgb4_remove_server_filter(
3537 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3538 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3539 } else {
84cc6ac6 3540 struct sockaddr_in6 *sin6;
1cab775c 3541 c4iw_init_wr_wait(&ep->com.wr_wait);
830662f6
VP		 3542 err = cxgb4_remove_server(
3543 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3544 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
1cab775c
VP		 3545 if (err)
3546 goto done;
3547 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3548 0, 0, __func__);
170003c8 3549 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 3550 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3551 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
1cab775c 3552 }
793dad94 3553 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
830662f6
VP		 3554 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3555 ep->com.local_addr.ss_family);
cfdda9d7 3556done:
9ca6f7cf 3557 deref_cm_id(&ep->com);
cfdda9d7
SW		 3558 c4iw_put_ep(&ep->com);
3559 return err;
3560}
3561
3562int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3563{
3564 int ret = 0;
cfdda9d7
SW		 3565 int close = 0;
3566 int fatal = 0;
3567 struct c4iw_rdev *rdev;
cfdda9d7 3568
2f5b48c3 3569 mutex_lock(&ep->com.mutex);
cfdda9d7
SW		 3570
3571 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3572 states[ep->com.state], abrupt);
3573
6e410d8f
H		 3574 /*
3575 * Ref the ep here in case we have fatal errors causing the
3576 * ep to be released and freed.
3577 */
3578 c4iw_get_ep(&ep->com);
3579
cfdda9d7
SW		 3580 rdev = &ep->com.dev->rdev;
3581 if (c4iw_fatal_error(rdev)) {
3582 fatal = 1;
be13b2df 3583 close_complete_upcall(ep, -EIO);
cfdda9d7
SW		 3584 ep->com.state = DEAD;
3585 }
3586 switch (ep->com.state) {
3587 case MPA_REQ_WAIT:
3588 case MPA_REQ_SENT:
3589 case MPA_REQ_RCVD:
3590 case MPA_REP_SENT:
3591 case FPDU_MODE:
3592 close = 1;
3593 if (abrupt)
3594 ep->com.state = ABORTING;
3595 else {
3596 ep->com.state = CLOSING;
ca5a2202 3597 start_ep_timer(ep);
cfdda9d7
SW		 3598 }
3599 set_bit(CLOSE_SENT, &ep->com.flags);
3600 break;
3601 case CLOSING:
3602 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3603 close = 1;
3604 if (abrupt) {
b33bd0cb 3605 (void)stop_ep_timer(ep);
cfdda9d7
SW		 3606 ep->com.state = ABORTING;
3607 } else
3608 ep->com.state = MORIBUND;
3609 }
3610 break;
3611 case MORIBUND:
3612 case ABORTING:
3613 case DEAD:
3614 PDBG("%s ignoring disconnect ep %p state %u\n",
3615 __func__, ep, ep->com.state);
3616 break;
3617 default:
3618 BUG();
3619 break;
3620 }
3621
cfdda9d7 3622 if (close) {
8da7e7a5 3623 if (abrupt) {
793dad94 3624 set_bit(EP_DISC_ABORT, &ep->com.history);
be13b2df 3625 close_complete_upcall(ep, -ECONNRESET);
8da7e7a5 3626 ret = send_abort(ep, NULL, gfp);
793dad94
VP		 3627 } else {
3628 set_bit(EP_DISC_CLOSE, &ep->com.history);
cfdda9d7 3629 ret = send_halfclose(ep, gfp);
793dad94 3630 }
88bc230d 3631 if (ret) {
9ca6f7cf 3632 set_bit(EP_DISC_FAIL, &ep->com.history);
88bc230d
H		 3633 if (!abrupt) {
3634 stop_ep_timer(ep);
3635 close_complete_upcall(ep, -EIO);
3636 }
c00dcbaf
H		 3637 if (ep->com.qp) {
3638 struct c4iw_qp_attributes attrs;
3639
3640 attrs.next_state = C4IW_QP_STATE_ERROR;
3641 ret = c4iw_modify_qp(ep->com.qp->rhp,
3642 ep->com.qp,
3643 C4IW_QP_ATTR_NEXT_STATE,
3644 &attrs, 1);
3645 if (ret)
3646 pr_err(MOD
3647 "%s - qp <- error failed!\n",
3648 __func__);
3649 }
cfdda9d7 3650 fatal = 1;
88bc230d 3651 }
cfdda9d7 3652 }
8da7e7a5 3653 mutex_unlock(&ep->com.mutex);
6e410d8f 3654 c4iw_put_ep(&ep->com);
cfdda9d7
SW		 3655 if (fatal)
3656 release_ep_resources(ep);
3657 return ret;
3658}
3659
1cab775c
VP		 3660static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3661 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3662{
3663 struct c4iw_ep *ep;
793dad94 3664 int atid = be32_to_cpu(req->tid);
1cab775c 3665
ef5d6355
VP		 3666 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3667 (__force u32) req->tid);
1cab775c
VP		 3668 if (!ep)
3669 return;
3670
3671 switch (req->retval) {
3672 case FW_ENOMEM:
793dad94
VP		 3673 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3674 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3675 send_fw_act_open_req(ep, atid);
3676 return;
3677 }
1cab775c 3678 case FW_EADDRINUSE:
793dad94
VP		 3679 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3680 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3681 send_fw_act_open_req(ep, atid);
3682 return;
3683 }
1cab775c
VP		 3684 break;
3685 default:
3686 pr_info("%s unexpected ofld conn wr retval %d\n",
3687 __func__, req->retval);
3688 break;
3689 }
793dad94
VP		 3690 pr_err("active ofld_connect_wr failure %d atid %d\n",
3691 req->retval, atid);
3692 mutex_lock(&dev->rdev.stats.lock);
3693 dev->rdev.stats.act_ofld_conn_fails++;
3694 mutex_unlock(&dev->rdev.stats.lock);
1cab775c 3695 connect_reply_upcall(ep, status2errno(req->retval));
793dad94 3696 state_set(&ep->com, DEAD);
84cc6ac6
H		 3697 if (ep->com.remote_addr.ss_family == AF_INET6) {
3698 struct sockaddr_in6 *sin6 =
170003c8 3699 (struct sockaddr_in6 *)&ep->com.local_addr;
84cc6ac6
H		 3700 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3701 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3702 }
793dad94
VP		 3703 remove_handle(dev, &dev->atid_idr, atid);
3704 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3705 dst_release(ep->dst);
3706 cxgb4_l2t_release(ep->l2t);
3707 c4iw_put_ep(&ep->com);
1cab775c
VP		 3708}
3709
3710static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3711 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3712{
3713 struct sk_buff *rpl_skb;
3714 struct cpl_pass_accept_req *cpl;
3715 int ret;
3716
710a3110 3717 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
1cab775c
VP		 3718 BUG_ON(!rpl_skb);
3719 if (req->retval) {
3720 PDBG("%s passive open failure %d\n", __func__, req->retval);
793dad94
VP		 3721 mutex_lock(&dev->rdev.stats.lock);
3722 dev->rdev.stats.pas_ofld_conn_fails++;
3723 mutex_unlock(&dev->rdev.stats.lock);
1cab775c
VP		 3724 kfree_skb(rpl_skb);
3725 } else {
3726 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3727 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
ef5d6355
VP		 3728 (__force u32) htonl(
3729 (__force u32) req->tid)));
1cab775c
VP		 3730 ret = pass_accept_req(dev, rpl_skb);
3731 if (!ret)
3732 kfree_skb(rpl_skb);
3733 }
3734 return;
3735}
3736
3737static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2f5b48c3
SW		 3738{
3739 struct cpl_fw6_msg *rpl = cplhdr(skb);
1cab775c
VP		 3740 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3741
3742 switch (rpl->type) {
3743 case FW6_TYPE_CQE:
3744 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3745 break;
3746 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3747 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3748 switch (req->t_state) {
3749 case TCP_SYN_SENT:
3750 active_ofld_conn_reply(dev, skb, req);
3751 break;
3752 case TCP_SYN_RECV:
3753 passive_ofld_conn_reply(dev, skb, req);
3754 break;
3755 default:
3756 pr_err("%s unexpected ofld conn wr state %d\n",
3757 __func__, req->t_state);
3758 break;
3759 }
3760 break;
3761 }
3762 return 0;
3763}
3764
3765static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3766{
963cab50
H		 3767 __be32 l2info;
3768 __be16 hdr_len, vlantag, len;
3769 u16 eth_hdr_len;
3770 int tcp_hdr_len, ip_hdr_len;
1cab775c
VP		 3771 u8 intf;
3772 struct cpl_rx_pkt *cpl = cplhdr(skb);
3773 struct cpl_pass_accept_req *req;
3774 struct tcp_options_received tmp_opt;
f079af7a 3775 struct c4iw_dev *dev;
963cab50 3776 enum chip_type type;
1cab775c 3777
f079af7a 3778 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
1cab775c 3779 /* Store values from cpl_rx_pkt in temporary location. */
963cab50
H		 3780 vlantag = cpl->vlan;
3781 len = cpl->len;
3782 l2info = cpl->l2info;
3783 hdr_len = cpl->hdr_len;
1cab775c
VP		 3784 intf = cpl->iff;
3785
3786 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3787
3788 /*
3789 * We need to parse the TCP options from SYN packet.
3790 * to generate cpl_pass_accept_req.
3791 */
3792 memset(&tmp_opt, 0, sizeof(tmp_opt));
3793 tcp_clear_options(&tmp_opt);
1a2c6181 3794 tcp_parse_options(skb, &tmp_opt, 0, NULL);
1cab775c
VP		 3795
3796 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3797 memset(req, 0, sizeof(*req));
cf7fe64a
HS		 3798 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3799 SYN_MAC_IDX_V(RX_MACIDX_G(
963cab50 3800 be32_to_cpu(l2info))) |
cf7fe64a 3801 SYN_XACT_MATCH_F);
963cab50
H		 3802 type = dev->rdev.lldi.adapter_type;
3803 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3804 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3805 req->hdr_len =
3806 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3807 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3808 eth_hdr_len = is_t4(type) ?
3809 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3810 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3811 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3812 IP_HDR_LEN_V(ip_hdr_len) |
3813 ETH_HDR_LEN_V(eth_hdr_len));
3814 } else { /* T6 and later */
3815 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3816 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3817 T6_IP_HDR_LEN_V(ip_hdr_len) |
3818 T6_ETH_HDR_LEN_V(eth_hdr_len));
3819 }
3820 req->vlan = vlantag;
3821 req->len = len;
6c53e938
HS		 3822 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3823 PASS_OPEN_TOS_V(tos));
1cab775c
VP		 3824 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3825 if (tmp_opt.wscale_ok)
3826 req->tcpopt.wsf = tmp_opt.snd_wscale;
3827 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3828 if (tmp_opt.sack_ok)
3829 req->tcpopt.sack = 1;
3830 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3831 return;
3832}
3833
3834static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3835 __be32 laddr, __be16 lport,
3836 __be32 raddr, __be16 rport,
3837 u32 rcv_isn, u32 filter, u16 window,
3838 u32 rss_qid, u8 port_id)
3839{
3840 struct sk_buff *req_skb;
3841 struct fw_ofld_connection_wr *req;
3842 struct cpl_pass_accept_req *cpl = cplhdr(skb);
1ce1d471 3843 int ret;
1cab775c
VP		 3844
3845 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3846 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3847 memset(req, 0, sizeof(*req));
6c53e938 3848 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
e2ac9628 3849 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
77a80e23 3850 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
ef5d6355 3851 req->le.filter = (__force __be32) filter;
1cab775c
VP		 3852 req->le.lport = lport;
3853 req->le.pport = rport;
3854 req->le.u.ipv4.lip = laddr;
3855 req->le.u.ipv4.pip = raddr;
3856 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3857 req->tcb.rcv_adv = htons(window);
3858 req->tcb.t_state_to_astid =
77a80e23
HS		 3859 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3860 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3861 FW_OFLD_CONNECTION_WR_ASTID_V(
6c53e938 3862 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
1cab775c
VP		 3863
3864 /*
3865 * We store the qid in opt2 which will be used by the firmware
3866 * to send us the wr response.
3867 */
d7990b0c 3868 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
1cab775c
VP		 3869
3870 /*
3871 * We initialize the MSS index in TCB to 0xF.
3872 * So that when driver sends cpl_pass_accept_rpl
3873 * TCB picks up the correct value. If this was 0
3874 * TP will ignore any value > 0 for MSS index.
3875 */
d7990b0c 3876 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
6198dd8d 3877 req->cookie = (uintptr_t)skb;
1cab775c
VP		 3878
3879 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
1ce1d471
SW		 3880 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3881 if (ret < 0) {
3882 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3883 ret);
3884 kfree_skb(skb);
3885 kfree_skb(req_skb);
3886 }
1cab775c
VP		 3887}
3888
3889/*
3890 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3891 * messages when a filter is being used instead of server to
3892 * redirect a syn packet. When packets hit filter they are redirected
3893 * to the offload queue and driver tries to establish the connection
3894 * using firmware work request.
3895 */
3896static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3897{
3898 int stid;
3899 unsigned int filter;
3900 struct ethhdr *eh = NULL;
3901 struct vlan_ethhdr *vlan_eh = NULL;
3902 struct iphdr *iph;
3903 struct tcphdr *tcph;
3904 struct rss_header *rss = (void *)skb->data;
3905 struct cpl_rx_pkt *cpl = (void *)skb->data;
3906 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3907 struct l2t_entry *e;
3908 struct dst_entry *dst;
f86fac79 3909 struct c4iw_ep *lep = NULL;
1cab775c
VP		 3910 u16 window;
3911 struct port_info *pi;
3912 struct net_device *pdev;
f079af7a 3913 u16 rss_qid, eth_hdr_len;
1cab775c
VP		 3914 int step;
3915 u32 tx_chan;
3916 struct neighbour *neigh;
3917
3918 /* Drop all non-SYN packets */
bdc590b9 3919 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
1cab775c
VP		 3920 goto reject;
3921
3922 /*
3923 * Drop all packets which did not hit the filter.
3924 * Unlikely to happen.
3925 */
3926 if (!(rss->filter_hit && rss->filter_tid))
3927 goto reject;
3928
3929 /*
3930 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3931 */
a4ea025f 3932 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
1cab775c 3933
f86fac79 3934 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
1cab775c
VP		 3935 if (!lep) {
3936 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3937 goto reject;
3938 }
3939
963cab50
H		 3940 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3941 case CHELSIO_T4:
3942 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3943 break;
3944 case CHELSIO_T5:
3945 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3946 break;
3947 case CHELSIO_T6:
3948 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3949 break;
3950 default:
3951 pr_err("T%d Chip is not supported\n",
3952 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3953 goto reject;
3954 }
3955
f079af7a 3956 if (eth_hdr_len == ETH_HLEN) {
1cab775c
VP		 3957 eh = (struct ethhdr *)(req + 1);
3958 iph = (struct iphdr *)(eh + 1);
3959 } else {
3960 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3961 iph = (struct iphdr *)(vlan_eh + 1);
3962 skb->vlan_tci = ntohs(cpl->vlan);
3963 }
3964
3965 if (iph->version != 0x4)
3966 goto reject;
3967
3968 tcph = (struct tcphdr *)(iph + 1);
3969 skb_set_network_header(skb, (void *)iph - (void *)rss);
3970 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3971 skb_get(skb);
3972
3973 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3974 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3975 ntohs(tcph->source), iph->tos);
3976
830662f6
VP		 3977 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3978 iph->tos);
3979 if (!dst) {
1cab775c
VP		 3980 pr_err("%s - failed to find dst entry!\n",
3981 __func__);
3982 goto reject;
3983 }
1cab775c
VP		 3984 neigh = dst_neigh_lookup_skb(dst, skb);
3985
aaa0c23c
ZZ		 3986 if (!neigh) {
3987 pr_err("%s - failed to allocate neigh!\n",
3988 __func__);
3989 goto free_dst;
3990 }
3991
1cab775c
VP		 3992 if (neigh->dev->flags & IFF_LOOPBACK) {
3993 pdev = ip_dev_find(&init_net, iph->daddr);
3994 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3995 pdev, 0);
3996 pi = (struct port_info *)netdev_priv(pdev);
3997 tx_chan = cxgb4_port_chan(pdev);
3998 dev_put(pdev);
3999 } else {
830662f6 4000 pdev = get_real_dev(neigh->dev);
1cab775c 4001 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
830662f6
VP		 4002 pdev, 0);
4003 pi = (struct port_info *)netdev_priv(pdev);
4004 tx_chan = cxgb4_port_chan(pdev);
1cab775c 4005 }
ebf00060 4006 neigh_release(neigh);
1cab775c
VP		 4007 if (!e) {
4008 pr_err("%s - failed to allocate l2t entry!\n",
4009 __func__);
4010 goto free_dst;
4011 }
4012
4013 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4014 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
ef5d6355 4015 window = (__force u16) htons((__force u16)tcph->window);
1cab775c
VP		 4016
4017 /* Calcuate filter portion for LE region. */
41b4f86c
KS		 4018 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4019 dev->rdev.lldi.ports[0],
4020 e));
1cab775c
VP		 4021
4022 /*
4023 * Synthesize the cpl_pass_accept_req. We have everything except the
4024 * TID. Once firmware sends a reply with TID we update the TID field
4025 * in cpl and pass it through the regular cpl_pass_accept_req path.
4026 */
4027 build_cpl_pass_accept_req(skb, stid, iph->tos);
4028 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4029 tcph->source, ntohl(tcph->seq), filter, window,
4030 rss_qid, pi->port_id);
4031 cxgb4_l2t_release(e);
4032free_dst:
4033 dst_release(dst);
4034reject:
f86fac79
H		 4035 if (lep)
4036 c4iw_put_ep(&lep->com);
2f5b48c3
SW		 4037 return 0;
4038}
4039
be4c9bad
RD		 4040/*
4041 * These are the real handlers that are called from a
4042 * work queue.
4043 */
9dec900c 4044static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
be4c9bad
RD		 4045 [CPL_ACT_ESTABLISH] = act_establish,
4046 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4047 [CPL_RX_DATA] = rx_data,
4048 [CPL_ABORT_RPL_RSS] = abort_rpl,
4049 [CPL_ABORT_RPL] = abort_rpl,
4050 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4051 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4052 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4053 [CPL_PASS_ESTABLISH] = pass_establish,
4054 [CPL_PEER_CLOSE] = peer_close,
4055 [CPL_ABORT_REQ_RSS] = peer_abort,
4056 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4057 [CPL_RDMA_TERMINATE] = terminate,
2f5b48c3 4058 [CPL_FW4_ACK] = fw4_ack,
1cab775c 4059 [CPL_FW6_MSG] = deferred_fw6_msg,
9dec900c 4060 [CPL_RX_PKT] = rx_pkt,
8d1f1a6b
H		 4061 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4062 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
be4c9bad
RD		 4063};
4064
4065static void process_timeout(struct c4iw_ep *ep)
4066{
4067 struct c4iw_qp_attributes attrs;
4068 int abort = 1;
4069
2f5b48c3 4070 mutex_lock(&ep->com.mutex);
be4c9bad
RD		 4071 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4072 ep->com.state);
793dad94 4073 set_bit(TIMEDOUT, &ep->com.history);
be4c9bad
RD		 4074 switch (ep->com.state) {
4075 case MPA_REQ_SENT:
be4c9bad
RD		 4076 connect_reply_upcall(ep, -ETIMEDOUT);
4077 break;
4078 case MPA_REQ_WAIT:
ceb110a8 4079 case MPA_REQ_RCVD:
e4b76a2a 4080 case MPA_REP_SENT:
ceb110a8 4081 case FPDU_MODE:
be4c9bad
RD		 4082 break;
4083 case CLOSING:
4084 case MORIBUND:
4085 if (ep->com.cm_id && ep->com.qp) {
4086 attrs.next_state = C4IW_QP_STATE_ERROR;
4087 c4iw_modify_qp(ep->com.qp->rhp,
4088 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4089 &attrs, 1);
4090 }
be13b2df 4091 close_complete_upcall(ep, -ETIMEDOUT);
be4c9bad 4092 break;
b33bd0cb
SW		 4093 case ABORTING:
4094 case DEAD:
4095
4096 /*
4097 * These states are expected if the ep timed out at the same
4098 * time as another thread was calling stop_ep_timer().
4099 * So we silently do nothing for these states.
4100 */
4101 abort = 0;
4102 break;
be4c9bad 4103 default:
76f267b7 4104 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
be4c9bad 4105 __func__, ep, ep->hwtid, ep->com.state);
be4c9bad
RD		 4106 abort = 0;
4107 }
cc18b939 4108 mutex_unlock(&ep->com.mutex);
69736279
H		 4109 if (abort)
4110 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
be4c9bad
RD		 4111 c4iw_put_ep(&ep->com);
4112}
4113
4114static void process_timedout_eps(void)
4115{
4116 struct c4iw_ep *ep;
4117
4118 spin_lock_irq(&timeout_lock);
4119 while (!list_empty(&timeout_list)) {
4120 struct list_head *tmp;
4121
4122 tmp = timeout_list.next;
4123 list_del(tmp);
b33bd0cb
SW		 4124 tmp->next = NULL;
4125 tmp->prev = NULL;
be4c9bad
RD		 4126 spin_unlock_irq(&timeout_lock);
4127 ep = list_entry(tmp, struct c4iw_ep, entry);
4128 process_timeout(ep);
4129 spin_lock_irq(&timeout_lock);
4130 }
4131 spin_unlock_irq(&timeout_lock);
4132}
4133
4134static void process_work(struct work_struct *work)
4135{
4136 struct sk_buff *skb = NULL;
4137 struct c4iw_dev *dev;
c1d7356c 4138 struct cpl_act_establish *rpl;
be4c9bad
RD		 4139 unsigned int opcode;
4140 int ret;
4141
b33bd0cb 4142 process_timedout_eps();
be4c9bad
RD		 4143 while ((skb = skb_dequeue(&rxq))) {
4144 rpl = cplhdr(skb);
4145 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4146 opcode = rpl->ot.opcode;
4147
4148 BUG_ON(!work_handlers[opcode]);
4149 ret = work_handlers[opcode](dev, skb);
4150 if (!ret)
4151 kfree_skb(skb);
b33bd0cb 4152 process_timedout_eps();
be4c9bad 4153 }
be4c9bad
RD		 4154}
4155
4156static DECLARE_WORK(skb_work, process_work);
4157
4158static void ep_timeout(unsigned long arg)
4159{
4160 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
1ec779cc 4161 int kickit = 0;
be4c9bad
RD		 4162
4163 spin_lock(&timeout_lock);
1ec779cc 4164 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
b33bd0cb
SW		 4165 /*
4166 * Only insert if it is not already on the list.
4167 */
4168 if (!ep->entry.next) {
4169 list_add_tail(&ep->entry, &timeout_list);
4170 kickit = 1;
4171 }
1ec779cc 4172 }
be4c9bad 4173 spin_unlock(&timeout_lock);
1ec779cc
VP		 4174 if (kickit)
4175 queue_work(workq, &skb_work);
be4c9bad
RD		 4176}
4177
cfdda9d7
SW		 4178/*
4179 * All the CM events are handled on a work queue to have a safe context.
4180 */
4181static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4182{
4183
4184 /*
4185 * Save dev in the skb->cb area.
4186 */
4187 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4188
4189 /*
4190 * Queue the skb and schedule the worker thread.
4191 */
4192 skb_queue_tail(&rxq, skb);
4193 queue_work(workq, &skb_work);
4194 return 0;
4195}
4196
4197static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4198{
4199 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4200
4201 if (rpl->status != CPL_ERR_NONE) {
4202 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4203 "for tid %u\n", rpl->status, GET_TID(rpl));
4204 }
2f5b48c3 4205 kfree_skb(skb);
cfdda9d7
SW		 4206 return 0;
4207}
4208
be4c9bad
RD		 4209static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4210{
4211 struct cpl_fw6_msg *rpl = cplhdr(skb);
4212 struct c4iw_wr_wait *wr_waitp;
4213 int ret;
4214
4215 PDBG("%s type %u\n", __func__, rpl->type);
4216
4217 switch (rpl->type) {
5be78ee9 4218 case FW6_TYPE_WR_RPL:
be4c9bad 4219 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
c8e081a1 4220 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
be4c9bad 4221 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
d9594d99
SW		 4222 if (wr_waitp)
4223 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2f5b48c3 4224 kfree_skb(skb);
be4c9bad 4225 break;
5be78ee9 4226 case FW6_TYPE_CQE:
5be78ee9 4227 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
1cab775c 4228 sched(dev, skb);
5be78ee9 4229 break;
be4c9bad
RD		 4230 default:
4231 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4232 rpl->type);
2f5b48c3 4233 kfree_skb(skb);
be4c9bad
RD		 4234 break;
4235 }
4236 return 0;
4237}
4238
8da7e7a5
SW		 4239static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4240{
4241 struct cpl_abort_req_rss *req = cplhdr(skb);
4242 struct c4iw_ep *ep;
8da7e7a5
SW		 4243 unsigned int tid = GET_TID(req);
4244
944661dd
H		 4245 ep = get_ep_from_tid(dev, tid);
4246 /* This EP will be dereferenced in peer_abort() */
14b92228
SW		 4247 if (!ep) {
4248 printk(KERN_WARNING MOD
4249 "Abort on non-existent endpoint, tid %d\n", tid);
4250 kfree_skb(skb);
4251 return 0;
4252 }
7a2cea2a 4253 if (is_neg_adv(req->status)) {
179d03bb
H		 4254 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4255 __func__, ep->hwtid, req->status,
4256 neg_adv_str(req->status));
944661dd 4257 goto out;
8da7e7a5
SW		 4258 }
4259 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4260 ep->com.state);
4261
093108cb 4262 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
944661dd 4263out:
8da7e7a5
SW		 4264 sched(dev, skb);
4265 return 0;
4266}
4267
be4c9bad
RD		 4268/*
4269 * Most upcalls from the T4 Core go to sched() to
4270 * schedule the processing on a work queue.
4271 */
4272c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4273 [CPL_ACT_ESTABLISH] = sched,
4274 [CPL_ACT_OPEN_RPL] = sched,
4275 [CPL_RX_DATA] = sched,
4276 [CPL_ABORT_RPL_RSS] = sched,
4277 [CPL_ABORT_RPL] = sched,
4278 [CPL_PASS_OPEN_RPL] = sched,
4279 [CPL_CLOSE_LISTSRV_RPL] = sched,
4280 [CPL_PASS_ACCEPT_REQ] = sched,
4281 [CPL_PASS_ESTABLISH] = sched,
4282 [CPL_PEER_CLOSE] = sched,
4283 [CPL_CLOSE_CON_RPL] = sched,
8da7e7a5 4284 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
be4c9bad
RD		 4285 [CPL_RDMA_TERMINATE] = sched,
4286 [CPL_FW4_ACK] = sched,
4287 [CPL_SET_TCB_RPL] = set_tcb_rpl,
1cab775c
VP		 4288 [CPL_FW6_MSG] = fw6_msg,
4289 [CPL_RX_PKT] = sched
be4c9bad
RD		 4290};
4291
cfdda9d7
SW		 4292int __init c4iw_cm_init(void)
4293{
be4c9bad 4294 spin_lock_init(&timeout_lock);
cfdda9d7
SW		 4295 skb_queue_head_init(&rxq);
4296
4297 workq = create_singlethread_workqueue("iw_cxgb4");
4298 if (!workq)
4299 return -ENOMEM;
4300
cfdda9d7
SW		 4301 return 0;
4302}
4303
46c1376d 4304void c4iw_cm_term(void)
cfdda9d7 4305{
be4c9bad 4306 WARN_ON(!list_empty(&timeout_list));
cfdda9d7
SW		 4307 flush_workqueue(workq);
4308 destroy_workqueue(workq);
4309}
Linux 6.x block layer and io_uring tree(s)