RDMA/cxgb4: Add support for iWARP Port Mapper user space service
[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
SW
49
50#include "iw_cxgb4.h"
51
52static char *states[] = {
53 "idle",
54 "listen",
55 "connecting",
56 "mpa_wait_req",
57 "mpa_req_sent",
58 "mpa_req_rcvd",
59 "mpa_rep_sent",
60 "fpdu_mode",
61 "aborting",
62 "closing",
63 "moribund",
64 "dead",
65 NULL,
66};
67
5be78ee9
VP
68static int nocong;
69module_param(nocong, int, 0644);
70MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
71
72static int enable_ecn;
73module_param(enable_ecn, int, 0644);
74MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
75
b52fe09e 76static int dack_mode = 1;
ba6d3925 77module_param(dack_mode, int, 0644);
b52fe09e 78MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
ba6d3925 79
be4c9bad
RD
80int c4iw_max_read_depth = 8;
81module_param(c4iw_max_read_depth, int, 0644);
82MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
83
cfdda9d7
SW
84static int enable_tcp_timestamps;
85module_param(enable_tcp_timestamps, int, 0644);
86MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
87
88static int enable_tcp_sack;
89module_param(enable_tcp_sack, int, 0644);
90MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
91
92static int enable_tcp_window_scaling = 1;
93module_param(enable_tcp_window_scaling, int, 0644);
94MODULE_PARM_DESC(enable_tcp_window_scaling,
95 "Enable tcp window scaling (default=1)");
96
97int c4iw_debug;
98module_param(c4iw_debug, int, 0644);
99MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
100
df2d5130 101static int peer2peer = 1;
cfdda9d7 102module_param(peer2peer, int, 0644);
df2d5130 103MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
cfdda9d7
SW
104
105static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
106module_param(p2p_type, int, 0644);
107MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
108 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
109
110static int ep_timeout_secs = 60;
111module_param(ep_timeout_secs, int, 0644);
112MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
113 "in seconds (default=60)");
114
115static int mpa_rev = 1;
116module_param(mpa_rev, int, 0644);
117MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
d2fe99e8
KS
118 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
119 " compliant (default=1)");
cfdda9d7
SW
120
121static int markers_enabled;
122module_param(markers_enabled, int, 0644);
123MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
124
125static int crc_enabled = 1;
126module_param(crc_enabled, int, 0644);
127MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
128
129static int rcv_win = 256 * 1024;
130module_param(rcv_win, int, 0644);
131MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
132
98ae68b7 133static int snd_win = 128 * 1024;
cfdda9d7 134module_param(snd_win, int, 0644);
98ae68b7 135MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
cfdda9d7 136
cfdda9d7 137static struct workqueue_struct *workq;
cfdda9d7
SW
138
139static struct sk_buff_head rxq;
cfdda9d7
SW
140
141static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
142static void ep_timeout(unsigned long arg);
143static void connect_reply_upcall(struct c4iw_ep *ep, int status);
144
be4c9bad
RD
145static LIST_HEAD(timeout_list);
146static spinlock_t timeout_lock;
147
325abead
VP
148static void deref_qp(struct c4iw_ep *ep)
149{
150 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
151 clear_bit(QP_REFERENCED, &ep->com.flags);
152}
153
154static void ref_qp(struct c4iw_ep *ep)
155{
156 set_bit(QP_REFERENCED, &ep->com.flags);
157 c4iw_qp_add_ref(&ep->com.qp->ibqp);
158}
159
cfdda9d7
SW
160static void start_ep_timer(struct c4iw_ep *ep)
161{
162 PDBG("%s ep %p\n", __func__, ep);
163 if (timer_pending(&ep->timer)) {
1ec779cc
VP
164 pr_err("%s timer already started! ep %p\n",
165 __func__, ep);
166 return;
167 }
168 clear_bit(TIMEOUT, &ep->com.flags);
169 c4iw_get_ep(&ep->com);
cfdda9d7
SW
170 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
171 ep->timer.data = (unsigned long)ep;
172 ep->timer.function = ep_timeout;
173 add_timer(&ep->timer);
174}
175
b33bd0cb 176static int stop_ep_timer(struct c4iw_ep *ep)
cfdda9d7 177{
1ec779cc 178 PDBG("%s ep %p stopping\n", __func__, ep);
cfdda9d7 179 del_timer_sync(&ep->timer);
b33bd0cb 180 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
1ec779cc 181 c4iw_put_ep(&ep->com);
b33bd0cb
SW
182 return 0;
183 }
184 return 1;
cfdda9d7
SW
185}
186
187static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
188 struct l2t_entry *l2e)
189{
190 int error = 0;
191
192 if (c4iw_fatal_error(rdev)) {
193 kfree_skb(skb);
194 PDBG("%s - device in error state - dropping\n", __func__);
195 return -EIO;
196 }
197 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
198 if (error < 0)
199 kfree_skb(skb);
74594861 200 return error < 0 ? error : 0;
cfdda9d7
SW
201}
202
203int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
204{
205 int error = 0;
206
207 if (c4iw_fatal_error(rdev)) {
208 kfree_skb(skb);
209 PDBG("%s - device in error state - dropping\n", __func__);
210 return -EIO;
211 }
212 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
213 if (error < 0)
214 kfree_skb(skb);
74594861 215 return error < 0 ? error : 0;
cfdda9d7
SW
216}
217
218static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
219{
220 struct cpl_tid_release *req;
221
222 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
223 if (!skb)
224 return;
225 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
226 INIT_TP_WR(req, hwtid);
227 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
228 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
229 c4iw_ofld_send(rdev, skb);
230 return;
231}
232
233static void set_emss(struct c4iw_ep *ep, u16 opt)
234{
235 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
236 ep->mss = ep->emss;
237 if (GET_TCPOPT_TSTAMP(opt))
238 ep->emss -= 12;
239 if (ep->emss < 128)
240 ep->emss = 128;
241 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
242 ep->mss, ep->emss);
243}
244
245static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
246{
cfdda9d7
SW
247 enum c4iw_ep_state state;
248
2f5b48c3 249 mutex_lock(&epc->mutex);
cfdda9d7 250 state = epc->state;
2f5b48c3 251 mutex_unlock(&epc->mutex);
cfdda9d7
SW
252 return state;
253}
254
255static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
256{
257 epc->state = new;
258}
259
260static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
261{
2f5b48c3 262 mutex_lock(&epc->mutex);
cfdda9d7
SW
263 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
264 __state_set(epc, new);
2f5b48c3 265 mutex_unlock(&epc->mutex);
cfdda9d7
SW
266 return;
267}
268
269static void *alloc_ep(int size, gfp_t gfp)
270{
271 struct c4iw_ep_common *epc;
272
273 epc = kzalloc(size, gfp);
274 if (epc) {
275 kref_init(&epc->kref);
2f5b48c3 276 mutex_init(&epc->mutex);
aadc4df3 277 c4iw_init_wr_wait(&epc->wr_wait);
cfdda9d7
SW
278 }
279 PDBG("%s alloc ep %p\n", __func__, epc);
280 return epc;
281}
282
283void _c4iw_free_ep(struct kref *kref)
284{
285 struct c4iw_ep *ep;
286
287 ep = container_of(kref, struct c4iw_ep, com.kref);
288 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
325abead
VP
289 if (test_bit(QP_REFERENCED, &ep->com.flags))
290 deref_qp(ep);
cfdda9d7 291 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
fe7e0a4d 292 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
cfdda9d7
SW
293 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
294 dst_release(ep->dst);
295 cxgb4_l2t_release(ep->l2t);
296 }
9eccfe10
SW
297 if (test_bit(RELEASE_MAPINFO, &ep->com.flags)) {
298 print_addr(&ep->com, __func__, "remove_mapinfo/mapping");
299 iwpm_remove_mapinfo(&ep->com.local_addr,
300 &ep->com.mapped_local_addr);
301 iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
302 }
cfdda9d7
SW
303 kfree(ep);
304}
305
306static void release_ep_resources(struct c4iw_ep *ep)
307{
308 set_bit(RELEASE_RESOURCES, &ep->com.flags);
309 c4iw_put_ep(&ep->com);
310}
311
cfdda9d7
SW
312static int status2errno(int status)
313{
314 switch (status) {
315 case CPL_ERR_NONE:
316 return 0;
317 case CPL_ERR_CONN_RESET:
318 return -ECONNRESET;
319 case CPL_ERR_ARP_MISS:
320 return -EHOSTUNREACH;
321 case CPL_ERR_CONN_TIMEDOUT:
322 return -ETIMEDOUT;
323 case CPL_ERR_TCAM_FULL:
324 return -ENOMEM;
325 case CPL_ERR_CONN_EXIST:
326 return -EADDRINUSE;
327 default:
328 return -EIO;
329 }
330}
331
332/*
333 * Try and reuse skbs already allocated...
334 */
335static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
336{
337 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
338 skb_trim(skb, 0);
339 skb_get(skb);
340 skb_reset_transport_header(skb);
341 } else {
342 skb = alloc_skb(len, gfp);
343 }
b38a0ad8 344 t4_set_arp_err_handler(skb, NULL, NULL);
cfdda9d7
SW
345 return skb;
346}
347
830662f6
VP
348static struct net_device *get_real_dev(struct net_device *egress_dev)
349{
350 struct net_device *phys_dev = egress_dev;
351 if (egress_dev->priv_flags & IFF_802_1Q_VLAN)
352 phys_dev = vlan_dev_real_dev(egress_dev);
353 return phys_dev;
354}
355
356static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
357{
358 int i;
359
360 egress_dev = get_real_dev(egress_dev);
361 for (i = 0; i < dev->rdev.lldi.nports; i++)
362 if (dev->rdev.lldi.ports[i] == egress_dev)
363 return 1;
364 return 0;
365}
366
367static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
368 __u8 *peer_ip, __be16 local_port,
369 __be16 peer_port, u8 tos,
370 __u32 sin6_scope_id)
371{
372 struct dst_entry *dst = NULL;
373
374 if (IS_ENABLED(CONFIG_IPV6)) {
375 struct flowi6 fl6;
376
377 memset(&fl6, 0, sizeof(fl6));
378 memcpy(&fl6.daddr, peer_ip, 16);
379 memcpy(&fl6.saddr, local_ip, 16);
380 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
381 fl6.flowi6_oif = sin6_scope_id;
382 dst = ip6_route_output(&init_net, NULL, &fl6);
383 if (!dst)
384 goto out;
385 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
386 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
387 dst_release(dst);
388 dst = NULL;
389 }
390 }
391
392out:
393 return dst;
394}
395
396static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
cfdda9d7
SW
397 __be32 peer_ip, __be16 local_port,
398 __be16 peer_port, u8 tos)
399{
400 struct rtable *rt;
31e4543d 401 struct flowi4 fl4;
830662f6 402 struct neighbour *n;
78fbfd8a 403
31e4543d 404 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
78fbfd8a
DM
405 peer_port, local_port, IPPROTO_TCP,
406 tos, 0);
b23dd4fe 407 if (IS_ERR(rt))
cfdda9d7 408 return NULL;
830662f6
VP
409 n = dst_neigh_lookup(&rt->dst, &peer_ip);
410 if (!n)
411 return NULL;
f8e81908
SW
412 if (!our_interface(dev, n->dev) &&
413 !(n->dev->flags & IFF_LOOPBACK)) {
830662f6
VP
414 dst_release(&rt->dst);
415 return NULL;
416 }
417 neigh_release(n);
418 return &rt->dst;
cfdda9d7
SW
419}
420
421static void arp_failure_discard(void *handle, struct sk_buff *skb)
422{
423 PDBG("%s c4iw_dev %p\n", __func__, handle);
424 kfree_skb(skb);
425}
426
427/*
428 * Handle an ARP failure for an active open.
429 */
430static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
431{
432 printk(KERN_ERR MOD "ARP failure duing connect\n");
433 kfree_skb(skb);
434}
435
436/*
437 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
438 * and send it along.
439 */
440static void abort_arp_failure(void *handle, struct sk_buff *skb)
441{
442 struct c4iw_rdev *rdev = handle;
443 struct cpl_abort_req *req = cplhdr(skb);
444
445 PDBG("%s rdev %p\n", __func__, rdev);
446 req->cmd = CPL_ABORT_NO_RST;
447 c4iw_ofld_send(rdev, skb);
448}
449
450static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
451{
452 unsigned int flowclen = 80;
453 struct fw_flowc_wr *flowc;
454 int i;
455
456 skb = get_skb(skb, flowclen, GFP_KERNEL);
457 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
458
459 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
460 FW_FLOWC_WR_NPARAMS(8));
461 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
462 16)) | FW_WR_FLOWID(ep->hwtid));
463
464 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
94788657 465 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
cfdda9d7
SW
466 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
467 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
468 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
469 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
470 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
471 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
472 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
473 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
474 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
475 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
476 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
477 flowc->mnemval[6].val = cpu_to_be32(snd_win);
478 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
479 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
480 /* Pad WR to 16 byte boundary */
481 flowc->mnemval[8].mnemonic = 0;
482 flowc->mnemval[8].val = 0;
483 for (i = 0; i < 9; i++) {
484 flowc->mnemval[i].r4[0] = 0;
485 flowc->mnemval[i].r4[1] = 0;
486 flowc->mnemval[i].r4[2] = 0;
487 }
488
489 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
490 c4iw_ofld_send(&ep->com.dev->rdev, skb);
491}
492
493static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
494{
495 struct cpl_close_con_req *req;
496 struct sk_buff *skb;
497 int wrlen = roundup(sizeof *req, 16);
498
499 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
500 skb = get_skb(NULL, wrlen, gfp);
501 if (!skb) {
502 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
503 return -ENOMEM;
504 }
505 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
506 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
507 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
508 memset(req, 0, wrlen);
509 INIT_TP_WR(req, ep->hwtid);
510 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
511 ep->hwtid));
512 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
513}
514
515static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
516{
517 struct cpl_abort_req *req;
518 int wrlen = roundup(sizeof *req, 16);
519
520 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
521 skb = get_skb(skb, wrlen, gfp);
522 if (!skb) {
523 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
524 __func__);
525 return -ENOMEM;
526 }
527 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
528 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
529 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
530 memset(req, 0, wrlen);
531 INIT_TP_WR(req, ep->hwtid);
532 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
533 req->cmd = CPL_ABORT_SEND_RST;
534 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
535}
536
9eccfe10
SW
537/*
538 * c4iw_form_pm_msg - Form a port mapper message with mapping info
539 */
540static void c4iw_form_pm_msg(struct c4iw_ep *ep,
541 struct iwpm_sa_data *pm_msg)
542{
543 memcpy(&pm_msg->loc_addr, &ep->com.local_addr,
544 sizeof(ep->com.local_addr));
545 memcpy(&pm_msg->rem_addr, &ep->com.remote_addr,
546 sizeof(ep->com.remote_addr));
547}
548
549/*
550 * c4iw_form_reg_msg - Form a port mapper message with dev info
551 */
552static void c4iw_form_reg_msg(struct c4iw_dev *dev,
553 struct iwpm_dev_data *pm_msg)
554{
555 memcpy(pm_msg->dev_name, dev->ibdev.name, IWPM_DEVNAME_SIZE);
556 memcpy(pm_msg->if_name, dev->rdev.lldi.ports[0]->name,
557 IWPM_IFNAME_SIZE);
558}
559
560static void c4iw_record_pm_msg(struct c4iw_ep *ep,
561 struct iwpm_sa_data *pm_msg)
562{
563 memcpy(&ep->com.mapped_local_addr, &pm_msg->mapped_loc_addr,
564 sizeof(ep->com.mapped_local_addr));
565 memcpy(&ep->com.mapped_remote_addr, &pm_msg->mapped_rem_addr,
566 sizeof(ep->com.mapped_remote_addr));
567}
568
cfdda9d7
SW
569static int send_connect(struct c4iw_ep *ep)
570{
571 struct cpl_act_open_req *req;
f079af7a 572 struct cpl_t5_act_open_req *t5_req;
830662f6
VP
573 struct cpl_act_open_req6 *req6;
574 struct cpl_t5_act_open_req6 *t5_req6;
cfdda9d7
SW
575 struct sk_buff *skb;
576 u64 opt0;
577 u32 opt2;
578 unsigned int mtu_idx;
579 int wscale;
830662f6
VP
580 int wrlen;
581 int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
582 sizeof(struct cpl_act_open_req) :
583 sizeof(struct cpl_t5_act_open_req);
584 int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
585 sizeof(struct cpl_act_open_req6) :
586 sizeof(struct cpl_t5_act_open_req6);
9eccfe10
SW
587 struct sockaddr_in *la = (struct sockaddr_in *)
588 &ep->com.mapped_local_addr;
589 struct sockaddr_in *ra = (struct sockaddr_in *)
590 &ep->com.mapped_remote_addr;
591 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
592 &ep->com.mapped_local_addr;
593 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
594 &ep->com.mapped_remote_addr;
830662f6
VP
595
596 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
597 roundup(sizev4, 16) :
598 roundup(sizev6, 16);
cfdda9d7
SW
599
600 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
601
602 skb = get_skb(NULL, wrlen, GFP_KERNEL);
603 if (!skb) {
604 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
605 __func__);
606 return -ENOMEM;
607 }
d4f1a5c6 608 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
cfdda9d7
SW
609
610 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
611 wscale = compute_wscale(rcv_win);
5be78ee9
VP
612 opt0 = (nocong ? NO_CONG(1) : 0) |
613 KEEP_ALIVE(1) |
ba6d3925 614 DELACK(1) |
cfdda9d7
SW
615 WND_SCALE(wscale) |
616 MSS_IDX(mtu_idx) |
617 L2T_IDX(ep->l2t->idx) |
618 TX_CHAN(ep->tx_chan) |
619 SMAC_SEL(ep->smac_idx) |
620 DSCP(ep->tos) |
b48f3b9c 621 ULP_MODE(ULP_MODE_TCPDDP) |
cfdda9d7
SW
622 RCV_BUFSIZ(rcv_win>>10);
623 opt2 = RX_CHANNEL(0) |
5be78ee9 624 CCTRL_ECN(enable_ecn) |
cfdda9d7
SW
625 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
626 if (enable_tcp_timestamps)
627 opt2 |= TSTAMPS_EN(1);
628 if (enable_tcp_sack)
629 opt2 |= SACK_EN(1);
630 if (wscale && enable_tcp_window_scaling)
631 opt2 |= WND_SCALE_EN(1);
92e5011a
SW
632 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
633 opt2 |= T5_OPT_2_VALID;
634 opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
635 }
cfdda9d7
SW
636 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
637
f079af7a 638 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
830662f6
VP
639 if (ep->com.remote_addr.ss_family == AF_INET) {
640 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
641 INIT_TP_WR(req, 0);
642 OPCODE_TID(req) = cpu_to_be32(
643 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
644 ((ep->rss_qid << 14) | ep->atid)));
645 req->local_port = la->sin_port;
646 req->peer_port = ra->sin_port;
647 req->local_ip = la->sin_addr.s_addr;
648 req->peer_ip = ra->sin_addr.s_addr;
649 req->opt0 = cpu_to_be64(opt0);
41b4f86c
KS
650 req->params = cpu_to_be32(cxgb4_select_ntuple(
651 ep->com.dev->rdev.lldi.ports[0],
652 ep->l2t));
830662f6
VP
653 req->opt2 = cpu_to_be32(opt2);
654 } else {
655 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
656
657 INIT_TP_WR(req6, 0);
658 OPCODE_TID(req6) = cpu_to_be32(
659 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
660 ((ep->rss_qid<<14)|ep->atid)));
661 req6->local_port = la6->sin6_port;
662 req6->peer_port = ra6->sin6_port;
663 req6->local_ip_hi = *((__be64 *)
664 (la6->sin6_addr.s6_addr));
665 req6->local_ip_lo = *((__be64 *)
666 (la6->sin6_addr.s6_addr + 8));
667 req6->peer_ip_hi = *((__be64 *)
668 (ra6->sin6_addr.s6_addr));
669 req6->peer_ip_lo = *((__be64 *)
670 (ra6->sin6_addr.s6_addr + 8));
671 req6->opt0 = cpu_to_be64(opt0);
41b4f86c
KS
672 req6->params = cpu_to_be32(cxgb4_select_ntuple(
673 ep->com.dev->rdev.lldi.ports[0],
674 ep->l2t));
830662f6
VP
675 req6->opt2 = cpu_to_be32(opt2);
676 }
f079af7a 677 } else {
830662f6
VP
678 if (ep->com.remote_addr.ss_family == AF_INET) {
679 t5_req = (struct cpl_t5_act_open_req *)
680 skb_put(skb, wrlen);
681 INIT_TP_WR(t5_req, 0);
682 OPCODE_TID(t5_req) = cpu_to_be32(
f079af7a
VP
683 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
684 ((ep->rss_qid << 14) | ep->atid)));
830662f6
VP
685 t5_req->local_port = la->sin_port;
686 t5_req->peer_port = ra->sin_port;
687 t5_req->local_ip = la->sin_addr.s_addr;
688 t5_req->peer_ip = ra->sin_addr.s_addr;
689 t5_req->opt0 = cpu_to_be64(opt0);
690 t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
41b4f86c
KS
691 cxgb4_select_ntuple(
692 ep->com.dev->rdev.lldi.ports[0],
693 ep->l2t)));
830662f6
VP
694 t5_req->opt2 = cpu_to_be32(opt2);
695 } else {
696 t5_req6 = (struct cpl_t5_act_open_req6 *)
697 skb_put(skb, wrlen);
698 INIT_TP_WR(t5_req6, 0);
699 OPCODE_TID(t5_req6) = cpu_to_be32(
700 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
701 ((ep->rss_qid<<14)|ep->atid)));
702 t5_req6->local_port = la6->sin6_port;
703 t5_req6->peer_port = ra6->sin6_port;
704 t5_req6->local_ip_hi = *((__be64 *)
705 (la6->sin6_addr.s6_addr));
706 t5_req6->local_ip_lo = *((__be64 *)
707 (la6->sin6_addr.s6_addr + 8));
708 t5_req6->peer_ip_hi = *((__be64 *)
709 (ra6->sin6_addr.s6_addr));
710 t5_req6->peer_ip_lo = *((__be64 *)
711 (ra6->sin6_addr.s6_addr + 8));
712 t5_req6->opt0 = cpu_to_be64(opt0);
713 t5_req6->params = (__force __be64)cpu_to_be32(
41b4f86c
KS
714 cxgb4_select_ntuple(
715 ep->com.dev->rdev.lldi.ports[0],
716 ep->l2t));
830662f6
VP
717 t5_req6->opt2 = cpu_to_be32(opt2);
718 }
f079af7a
VP
719 }
720
793dad94 721 set_bit(ACT_OPEN_REQ, &ep->com.history);
cfdda9d7
SW
722 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
723}
724
d2fe99e8
KS
725static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
726 u8 mpa_rev_to_use)
cfdda9d7
SW
727{
728 int mpalen, wrlen;
729 struct fw_ofld_tx_data_wr *req;
730 struct mpa_message *mpa;
d2fe99e8 731 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW
732
733 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
734
735 BUG_ON(skb_cloned(skb));
736
737 mpalen = sizeof(*mpa) + ep->plen;
d2fe99e8
KS
738 if (mpa_rev_to_use == 2)
739 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW
740 wrlen = roundup(mpalen + sizeof *req, 16);
741 skb = get_skb(skb, wrlen, GFP_KERNEL);
742 if (!skb) {
743 connect_reply_upcall(ep, -ENOMEM);
744 return;
745 }
746 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
747
748 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
749 memset(req, 0, wrlen);
750 req->op_to_immdlen = cpu_to_be32(
751 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
752 FW_WR_COMPL(1) |
753 FW_WR_IMMDLEN(mpalen));
754 req->flowid_len16 = cpu_to_be32(
755 FW_WR_FLOWID(ep->hwtid) |
756 FW_WR_LEN16(wrlen >> 4));
757 req->plen = cpu_to_be32(mpalen);
758 req->tunnel_to_proxy = cpu_to_be32(
759 FW_OFLD_TX_DATA_WR_FLUSH(1) |
760 FW_OFLD_TX_DATA_WR_SHOVE(1));
761
762 mpa = (struct mpa_message *)(req + 1);
763 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
764 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
d2fe99e8
KS
765 (markers_enabled ? MPA_MARKERS : 0) |
766 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
cfdda9d7 767 mpa->private_data_size = htons(ep->plen);
d2fe99e8 768 mpa->revision = mpa_rev_to_use;
01b225e1 769 if (mpa_rev_to_use == 1) {
d2fe99e8 770 ep->tried_with_mpa_v1 = 1;
01b225e1
KS
771 ep->retry_with_mpa_v1 = 0;
772 }
d2fe99e8
KS
773
774 if (mpa_rev_to_use == 2) {
f747c34a
RD
775 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
776 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS
777 mpa_v2_params.ird = htons((u16)ep->ird);
778 mpa_v2_params.ord = htons((u16)ep->ord);
779
780 if (peer2peer) {
781 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
782 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
783 mpa_v2_params.ord |=
784 htons(MPA_V2_RDMA_WRITE_RTR);
785 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
786 mpa_v2_params.ord |=
787 htons(MPA_V2_RDMA_READ_RTR);
788 }
789 memcpy(mpa->private_data, &mpa_v2_params,
790 sizeof(struct mpa_v2_conn_params));
cfdda9d7 791
d2fe99e8
KS
792 if (ep->plen)
793 memcpy(mpa->private_data +
794 sizeof(struct mpa_v2_conn_params),
795 ep->mpa_pkt + sizeof(*mpa), ep->plen);
796 } else
797 if (ep->plen)
798 memcpy(mpa->private_data,
799 ep->mpa_pkt + sizeof(*mpa), ep->plen);
cfdda9d7
SW
800
801 /*
802 * Reference the mpa skb. This ensures the data area
803 * will remain in memory until the hw acks the tx.
804 * Function fw4_ack() will deref it.
805 */
806 skb_get(skb);
807 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
808 BUG_ON(ep->mpa_skb);
809 ep->mpa_skb = skb;
810 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
811 start_ep_timer(ep);
a7db89eb 812 __state_set(&ep->com, MPA_REQ_SENT);
cfdda9d7 813 ep->mpa_attr.initiator = 1;
9c88aa00 814 ep->snd_seq += mpalen;
cfdda9d7
SW
815 return;
816}
817
818static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
819{
820 int mpalen, wrlen;
821 struct fw_ofld_tx_data_wr *req;
822 struct mpa_message *mpa;
823 struct sk_buff *skb;
d2fe99e8 824 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW
825
826 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
827
828 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS
829 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
830 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW
831 wrlen = roundup(mpalen + sizeof *req, 16);
832
833 skb = get_skb(NULL, wrlen, GFP_KERNEL);
834 if (!skb) {
835 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
836 return -ENOMEM;
837 }
838 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
839
840 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
841 memset(req, 0, wrlen);
842 req->op_to_immdlen = cpu_to_be32(
843 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
844 FW_WR_COMPL(1) |
845 FW_WR_IMMDLEN(mpalen));
846 req->flowid_len16 = cpu_to_be32(
847 FW_WR_FLOWID(ep->hwtid) |
848 FW_WR_LEN16(wrlen >> 4));
849 req->plen = cpu_to_be32(mpalen);
850 req->tunnel_to_proxy = cpu_to_be32(
851 FW_OFLD_TX_DATA_WR_FLUSH(1) |
852 FW_OFLD_TX_DATA_WR_SHOVE(1));
853
854 mpa = (struct mpa_message *)(req + 1);
855 memset(mpa, 0, sizeof(*mpa));
856 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
857 mpa->flags = MPA_REJECT;
fe7e0a4d 858 mpa->revision = ep->mpa_attr.version;
cfdda9d7 859 mpa->private_data_size = htons(plen);
d2fe99e8
KS
860
861 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
862 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD
863 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
864 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS
865 mpa_v2_params.ird = htons(((u16)ep->ird) |
866 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
867 0));
868 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
869 (p2p_type ==
870 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
871 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
872 FW_RI_INIT_P2PTYPE_READ_REQ ?
873 MPA_V2_RDMA_READ_RTR : 0) : 0));
874 memcpy(mpa->private_data, &mpa_v2_params,
875 sizeof(struct mpa_v2_conn_params));
876
877 if (ep->plen)
878 memcpy(mpa->private_data +
879 sizeof(struct mpa_v2_conn_params), pdata, plen);
880 } else
881 if (plen)
882 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW
883
884 /*
885 * Reference the mpa skb again. This ensures the data area
886 * will remain in memory until the hw acks the tx.
887 * Function fw4_ack() will deref it.
888 */
889 skb_get(skb);
890 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
891 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
892 BUG_ON(ep->mpa_skb);
893 ep->mpa_skb = skb;
9c88aa00 894 ep->snd_seq += mpalen;
cfdda9d7
SW
895 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
896}
897
898static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
899{
900 int mpalen, wrlen;
901 struct fw_ofld_tx_data_wr *req;
902 struct mpa_message *mpa;
903 struct sk_buff *skb;
d2fe99e8 904 struct mpa_v2_conn_params mpa_v2_params;
cfdda9d7
SW
905
906 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
907
908 mpalen = sizeof(*mpa) + plen;
d2fe99e8
KS
909 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
910 mpalen += sizeof(struct mpa_v2_conn_params);
cfdda9d7
SW
911 wrlen = roundup(mpalen + sizeof *req, 16);
912
913 skb = get_skb(NULL, wrlen, GFP_KERNEL);
914 if (!skb) {
915 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
916 return -ENOMEM;
917 }
918 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
919
920 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
921 memset(req, 0, wrlen);
922 req->op_to_immdlen = cpu_to_be32(
923 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
924 FW_WR_COMPL(1) |
925 FW_WR_IMMDLEN(mpalen));
926 req->flowid_len16 = cpu_to_be32(
927 FW_WR_FLOWID(ep->hwtid) |
928 FW_WR_LEN16(wrlen >> 4));
929 req->plen = cpu_to_be32(mpalen);
930 req->tunnel_to_proxy = cpu_to_be32(
931 FW_OFLD_TX_DATA_WR_FLUSH(1) |
932 FW_OFLD_TX_DATA_WR_SHOVE(1));
933
934 mpa = (struct mpa_message *)(req + 1);
935 memset(mpa, 0, sizeof(*mpa));
936 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
937 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
938 (markers_enabled ? MPA_MARKERS : 0);
d2fe99e8 939 mpa->revision = ep->mpa_attr.version;
cfdda9d7 940 mpa->private_data_size = htons(plen);
d2fe99e8
KS
941
942 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
943 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
f747c34a
RD
944 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
945 sizeof (struct mpa_v2_conn_params));
d2fe99e8
KS
946 mpa_v2_params.ird = htons((u16)ep->ird);
947 mpa_v2_params.ord = htons((u16)ep->ord);
948 if (peer2peer && (ep->mpa_attr.p2p_type !=
949 FW_RI_INIT_P2PTYPE_DISABLED)) {
950 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
951
952 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
953 mpa_v2_params.ord |=
954 htons(MPA_V2_RDMA_WRITE_RTR);
955 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
956 mpa_v2_params.ord |=
957 htons(MPA_V2_RDMA_READ_RTR);
958 }
959
960 memcpy(mpa->private_data, &mpa_v2_params,
961 sizeof(struct mpa_v2_conn_params));
962
963 if (ep->plen)
964 memcpy(mpa->private_data +
965 sizeof(struct mpa_v2_conn_params), pdata, plen);
966 } else
967 if (plen)
968 memcpy(mpa->private_data, pdata, plen);
cfdda9d7
SW
969
970 /*
971 * Reference the mpa skb. This ensures the data area
972 * will remain in memory until the hw acks the tx.
973 * Function fw4_ack() will deref it.
974 */
975 skb_get(skb);
976 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
977 ep->mpa_skb = skb;
a7db89eb 978 __state_set(&ep->com, MPA_REP_SENT);
9c88aa00 979 ep->snd_seq += mpalen;
cfdda9d7
SW
980 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
981}
982
983static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
984{
985 struct c4iw_ep *ep;
986 struct cpl_act_establish *req = cplhdr(skb);
987 unsigned int tid = GET_TID(req);
988 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
989 struct tid_info *t = dev->rdev.lldi.tids;
990
991 ep = lookup_atid(t, atid);
992
993 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
994 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
995
a7db89eb 996 mutex_lock(&ep->com.mutex);
cfdda9d7
SW
997 dst_confirm(ep->dst);
998
999 /* setup the hwtid for this connection */
1000 ep->hwtid = tid;
1001 cxgb4_insert_tid(t, ep, tid);
793dad94 1002 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
cfdda9d7
SW
1003
1004 ep->snd_seq = be32_to_cpu(req->snd_isn);
1005 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1006
1007 set_emss(ep, ntohs(req->tcp_opt));
1008
1009 /* dealloc the atid */
793dad94 1010 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7 1011 cxgb4_free_atid(t, atid);
793dad94 1012 set_bit(ACT_ESTAB, &ep->com.history);
cfdda9d7
SW
1013
1014 /* start MPA negotiation */
1015 send_flowc(ep, NULL);
d2fe99e8
KS
1016 if (ep->retry_with_mpa_v1)
1017 send_mpa_req(ep, skb, 1);
1018 else
1019 send_mpa_req(ep, skb, mpa_rev);
a7db89eb 1020 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
1021 return 0;
1022}
1023
be13b2df 1024static void close_complete_upcall(struct c4iw_ep *ep, int status)
cfdda9d7
SW
1025{
1026 struct iw_cm_event event;
1027
1028 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1029 memset(&event, 0, sizeof(event));
1030 event.event = IW_CM_EVENT_CLOSE;
be13b2df 1031 event.status = status;
cfdda9d7
SW
1032 if (ep->com.cm_id) {
1033 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1034 ep, ep->com.cm_id, ep->hwtid);
1035 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1036 ep->com.cm_id->rem_ref(ep->com.cm_id);
1037 ep->com.cm_id = NULL;
793dad94 1038 set_bit(CLOSE_UPCALL, &ep->com.history);
cfdda9d7
SW
1039 }
1040}
1041
1042static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1043{
1044 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
cc18b939 1045 __state_set(&ep->com, ABORTING);
793dad94 1046 set_bit(ABORT_CONN, &ep->com.history);
cfdda9d7
SW
1047 return send_abort(ep, skb, gfp);
1048}
1049
1050static void peer_close_upcall(struct c4iw_ep *ep)
1051{
1052 struct iw_cm_event event;
1053
1054 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1055 memset(&event, 0, sizeof(event));
1056 event.event = IW_CM_EVENT_DISCONNECT;
1057 if (ep->com.cm_id) {
1058 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1059 ep, ep->com.cm_id, ep->hwtid);
1060 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 1061 set_bit(DISCONN_UPCALL, &ep->com.history);
cfdda9d7
SW
1062 }
1063}
1064
1065static void peer_abort_upcall(struct c4iw_ep *ep)
1066{
1067 struct iw_cm_event event;
1068
1069 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1070 memset(&event, 0, sizeof(event));
1071 event.event = IW_CM_EVENT_CLOSE;
1072 event.status = -ECONNRESET;
1073 if (ep->com.cm_id) {
1074 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1075 ep->com.cm_id, ep->hwtid);
1076 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1077 ep->com.cm_id->rem_ref(ep->com.cm_id);
1078 ep->com.cm_id = NULL;
793dad94 1079 set_bit(ABORT_UPCALL, &ep->com.history);
cfdda9d7
SW
1080 }
1081}
1082
1083static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1084{
1085 struct iw_cm_event event;
1086
1087 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1088 memset(&event, 0, sizeof(event));
1089 event.event = IW_CM_EVENT_CONNECT_REPLY;
1090 event.status = status;
24d44a39
SW
1091 memcpy(&event.local_addr, &ep->com.local_addr,
1092 sizeof(ep->com.local_addr));
1093 memcpy(&event.remote_addr, &ep->com.remote_addr,
1094 sizeof(ep->com.remote_addr));
cfdda9d7
SW
1095
1096 if ((status == 0) || (status == -ECONNREFUSED)) {
d2fe99e8
KS
1097 if (!ep->tried_with_mpa_v1) {
1098 /* this means MPA_v2 is used */
1099 event.private_data_len = ep->plen -
1100 sizeof(struct mpa_v2_conn_params);
1101 event.private_data = ep->mpa_pkt +
1102 sizeof(struct mpa_message) +
1103 sizeof(struct mpa_v2_conn_params);
1104 } else {
1105 /* this means MPA_v1 is used */
1106 event.private_data_len = ep->plen;
1107 event.private_data = ep->mpa_pkt +
1108 sizeof(struct mpa_message);
1109 }
cfdda9d7 1110 }
85963e4c
RD
1111
1112 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1113 ep->hwtid, status);
793dad94 1114 set_bit(CONN_RPL_UPCALL, &ep->com.history);
85963e4c
RD
1115 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1116
cfdda9d7
SW
1117 if (status < 0) {
1118 ep->com.cm_id->rem_ref(ep->com.cm_id);
1119 ep->com.cm_id = NULL;
cfdda9d7
SW
1120 }
1121}
1122
be13b2df 1123static int connect_request_upcall(struct c4iw_ep *ep)
cfdda9d7
SW
1124{
1125 struct iw_cm_event event;
be13b2df 1126 int ret;
cfdda9d7
SW
1127
1128 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1129 memset(&event, 0, sizeof(event));
1130 event.event = IW_CM_EVENT_CONNECT_REQUEST;
24d44a39
SW
1131 memcpy(&event.local_addr, &ep->com.local_addr,
1132 sizeof(ep->com.local_addr));
1133 memcpy(&event.remote_addr, &ep->com.remote_addr,
1134 sizeof(ep->com.remote_addr));
cfdda9d7 1135 event.provider_data = ep;
d2fe99e8
KS
1136 if (!ep->tried_with_mpa_v1) {
1137 /* this means MPA_v2 is used */
1138 event.ord = ep->ord;
1139 event.ird = ep->ird;
1140 event.private_data_len = ep->plen -
1141 sizeof(struct mpa_v2_conn_params);
1142 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1143 sizeof(struct mpa_v2_conn_params);
1144 } else {
1145 /* this means MPA_v1 is used. Send max supported */
1146 event.ord = c4iw_max_read_depth;
1147 event.ird = c4iw_max_read_depth;
1148 event.private_data_len = ep->plen;
1149 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1150 }
be13b2df
SW
1151 c4iw_get_ep(&ep->com);
1152 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1153 &event);
1154 if (ret)
1155 c4iw_put_ep(&ep->com);
793dad94 1156 set_bit(CONNREQ_UPCALL, &ep->com.history);
cfdda9d7 1157 c4iw_put_ep(&ep->parent_ep->com);
be13b2df 1158 return ret;
cfdda9d7
SW
1159}
1160
1161static void established_upcall(struct c4iw_ep *ep)
1162{
1163 struct iw_cm_event event;
1164
1165 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1166 memset(&event, 0, sizeof(event));
1167 event.event = IW_CM_EVENT_ESTABLISHED;
d2fe99e8
KS
1168 event.ird = ep->ird;
1169 event.ord = ep->ord;
cfdda9d7
SW
1170 if (ep->com.cm_id) {
1171 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1172 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
793dad94 1173 set_bit(ESTAB_UPCALL, &ep->com.history);
cfdda9d7
SW
1174 }
1175}
1176
1177static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1178{
1179 struct cpl_rx_data_ack *req;
1180 struct sk_buff *skb;
1181 int wrlen = roundup(sizeof *req, 16);
1182
1183 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1184 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1185 if (!skb) {
1186 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1187 return 0;
1188 }
1189
1190 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1191 memset(req, 0, wrlen);
1192 INIT_TP_WR(req, ep->hwtid);
1193 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1194 ep->hwtid));
ba6d3925
SW
1195 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1196 F_RX_DACK_CHANGE |
1197 V_RX_DACK_MODE(dack_mode));
d4f1a5c6 1198 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
cfdda9d7
SW
1199 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1200 return credits;
1201}
1202
cc18b939 1203static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
cfdda9d7
SW
1204{
1205 struct mpa_message *mpa;
d2fe99e8 1206 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7 1207 u16 plen;
d2fe99e8
KS
1208 u16 resp_ird, resp_ord;
1209 u8 rtr_mismatch = 0, insuff_ird = 0;
cfdda9d7
SW
1210 struct c4iw_qp_attributes attrs;
1211 enum c4iw_qp_attr_mask mask;
1212 int err;
cc18b939 1213 int disconnect = 0;
cfdda9d7
SW
1214
1215 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1216
1217 /*
b33bd0cb
SW
1218 * Stop mpa timer. If it expired, then
1219 * we ignore the MPA reply. process_timeout()
1220 * will abort the connection.
cfdda9d7 1221 */
b33bd0cb 1222 if (stop_ep_timer(ep))
cc18b939 1223 return 0;
cfdda9d7
SW
1224
1225 /*
1226 * If we get more than the supported amount of private data
1227 * then we must fail this connection.
1228 */
1229 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1230 err = -EINVAL;
1231 goto err;
1232 }
1233
1234 /*
1235 * copy the new data into our accumulation buffer.
1236 */
1237 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1238 skb->len);
1239 ep->mpa_pkt_len += skb->len;
1240
1241 /*
1242 * if we don't even have the mpa message, then bail.
1243 */
1244 if (ep->mpa_pkt_len < sizeof(*mpa))
cc18b939 1245 return 0;
cfdda9d7
SW
1246 mpa = (struct mpa_message *) ep->mpa_pkt;
1247
1248 /* Validate MPA header. */
d2fe99e8
KS
1249 if (mpa->revision > mpa_rev) {
1250 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1251 " Received = %d\n", __func__, mpa_rev, mpa->revision);
cfdda9d7
SW
1252 err = -EPROTO;
1253 goto err;
1254 }
1255 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1256 err = -EPROTO;
1257 goto err;
1258 }
1259
1260 plen = ntohs(mpa->private_data_size);
1261
1262 /*
1263 * Fail if there's too much private data.
1264 */
1265 if (plen > MPA_MAX_PRIVATE_DATA) {
1266 err = -EPROTO;
1267 goto err;
1268 }
1269
1270 /*
1271 * If plen does not account for pkt size
1272 */
1273 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1274 err = -EPROTO;
1275 goto err;
1276 }
1277
1278 ep->plen = (u8) plen;
1279
1280 /*
1281 * If we don't have all the pdata yet, then bail.
1282 * We'll continue process when more data arrives.
1283 */
1284 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
cc18b939 1285 return 0;
cfdda9d7
SW
1286
1287 if (mpa->flags & MPA_REJECT) {
1288 err = -ECONNREFUSED;
1289 goto err;
1290 }
1291
1292 /*
1293 * If we get here we have accumulated the entire mpa
1294 * start reply message including private data. And
1295 * the MPA header is valid.
1296 */
c529fb50 1297 __state_set(&ep->com, FPDU_MODE);
cfdda9d7
SW
1298 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1299 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1300 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS
1301 ep->mpa_attr.version = mpa->revision;
1302 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1303
1304 if (mpa->revision == 2) {
1305 ep->mpa_attr.enhanced_rdma_conn =
1306 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1307 if (ep->mpa_attr.enhanced_rdma_conn) {
1308 mpa_v2_params = (struct mpa_v2_conn_params *)
1309 (ep->mpa_pkt + sizeof(*mpa));
1310 resp_ird = ntohs(mpa_v2_params->ird) &
1311 MPA_V2_IRD_ORD_MASK;
1312 resp_ord = ntohs(mpa_v2_params->ord) &
1313 MPA_V2_IRD_ORD_MASK;
1314
1315 /*
1316 * This is a double-check. Ideally, below checks are
1317 * not required since ird/ord stuff has been taken
1318 * care of in c4iw_accept_cr
1319 */
1320 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1321 err = -ENOMEM;
1322 ep->ird = resp_ord;
1323 ep->ord = resp_ird;
1324 insuff_ird = 1;
1325 }
1326
1327 if (ntohs(mpa_v2_params->ird) &
1328 MPA_V2_PEER2PEER_MODEL) {
1329 if (ntohs(mpa_v2_params->ord) &
1330 MPA_V2_RDMA_WRITE_RTR)
1331 ep->mpa_attr.p2p_type =
1332 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1333 else if (ntohs(mpa_v2_params->ord) &
1334 MPA_V2_RDMA_READ_RTR)
1335 ep->mpa_attr.p2p_type =
1336 FW_RI_INIT_P2PTYPE_READ_REQ;
1337 }
1338 }
1339 } else if (mpa->revision == 1)
1340 if (peer2peer)
1341 ep->mpa_attr.p2p_type = p2p_type;
1342
cfdda9d7 1343 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
d2fe99e8
KS
1344 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1345 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1346 ep->mpa_attr.recv_marker_enabled,
1347 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1348 ep->mpa_attr.p2p_type, p2p_type);
1349
1350 /*
1351 * If responder's RTR does not match with that of initiator, assign
1352 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1353 * generated when moving QP to RTS state.
1354 * A TERM message will be sent after QP has moved to RTS state
1355 */
91018f86 1356 if ((ep->mpa_attr.version == 2) && peer2peer &&
d2fe99e8
KS
1357 (ep->mpa_attr.p2p_type != p2p_type)) {
1358 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1359 rtr_mismatch = 1;
1360 }
cfdda9d7
SW
1361
1362 attrs.mpa_attr = ep->mpa_attr;
1363 attrs.max_ird = ep->ird;
1364 attrs.max_ord = ep->ord;
1365 attrs.llp_stream_handle = ep;
1366 attrs.next_state = C4IW_QP_STATE_RTS;
1367
1368 mask = C4IW_QP_ATTR_NEXT_STATE |
1369 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1370 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1371
1372 /* bind QP and TID with INIT_WR */
1373 err = c4iw_modify_qp(ep->com.qp->rhp,
1374 ep->com.qp, mask, &attrs, 1);
1375 if (err)
1376 goto err;
d2fe99e8
KS
1377
1378 /*
1379 * If responder's RTR requirement did not match with what initiator
1380 * supports, generate TERM message
1381 */
1382 if (rtr_mismatch) {
1383 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1384 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1385 attrs.ecode = MPA_NOMATCH_RTR;
1386 attrs.next_state = C4IW_QP_STATE_TERMINATE;
cc18b939 1387 attrs.send_term = 1;
d2fe99e8 1388 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939 1389 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
d2fe99e8 1390 err = -ENOMEM;
cc18b939 1391 disconnect = 1;
d2fe99e8
KS
1392 goto out;
1393 }
1394
1395 /*
1396 * Generate TERM if initiator IRD is not sufficient for responder
1397 * provided ORD. Currently, we do the same behaviour even when
1398 * responder provided IRD is also not sufficient as regards to
1399 * initiator ORD.
1400 */
1401 if (insuff_ird) {
1402 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1403 __func__);
1404 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1405 attrs.ecode = MPA_INSUFF_IRD;
1406 attrs.next_state = C4IW_QP_STATE_TERMINATE;
cc18b939 1407 attrs.send_term = 1;
d2fe99e8 1408 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939 1409 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
d2fe99e8 1410 err = -ENOMEM;
cc18b939 1411 disconnect = 1;
d2fe99e8
KS
1412 goto out;
1413 }
cfdda9d7
SW
1414 goto out;
1415err:
c529fb50 1416 __state_set(&ep->com, ABORTING);
b21ef16a 1417 send_abort(ep, skb, GFP_KERNEL);
cfdda9d7
SW
1418out:
1419 connect_reply_upcall(ep, err);
cc18b939 1420 return disconnect;
cfdda9d7
SW
1421}
1422
1423static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1424{
1425 struct mpa_message *mpa;
d2fe99e8 1426 struct mpa_v2_conn_params *mpa_v2_params;
cfdda9d7
SW
1427 u16 plen;
1428
1429 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1430
cfdda9d7
SW
1431 /*
1432 * If we get more than the supported amount of private data
1433 * then we must fail this connection.
1434 */
1435 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
b33bd0cb 1436 (void)stop_ep_timer(ep);
cfdda9d7
SW
1437 abort_connection(ep, skb, GFP_KERNEL);
1438 return;
1439 }
1440
1441 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1442
1443 /*
1444 * Copy the new data into our accumulation buffer.
1445 */
1446 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1447 skb->len);
1448 ep->mpa_pkt_len += skb->len;
1449
1450 /*
1451 * If we don't even have the mpa message, then bail.
1452 * We'll continue process when more data arrives.
1453 */
1454 if (ep->mpa_pkt_len < sizeof(*mpa))
1455 return;
1456
1457 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
cfdda9d7
SW
1458 mpa = (struct mpa_message *) ep->mpa_pkt;
1459
1460 /*
1461 * Validate MPA Header.
1462 */
d2fe99e8
KS
1463 if (mpa->revision > mpa_rev) {
1464 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1465 " Received = %d\n", __func__, mpa_rev, mpa->revision);
b33bd0cb 1466 (void)stop_ep_timer(ep);
cfdda9d7
SW
1467 abort_connection(ep, skb, GFP_KERNEL);
1468 return;
1469 }
1470
1471 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
b33bd0cb 1472 (void)stop_ep_timer(ep);
cfdda9d7
SW
1473 abort_connection(ep, skb, GFP_KERNEL);
1474 return;
1475 }
1476
1477 plen = ntohs(mpa->private_data_size);
1478
1479 /*
1480 * Fail if there's too much private data.
1481 */
1482 if (plen > MPA_MAX_PRIVATE_DATA) {
b33bd0cb 1483 (void)stop_ep_timer(ep);
cfdda9d7
SW
1484 abort_connection(ep, skb, GFP_KERNEL);
1485 return;
1486 }
1487
1488 /*
1489 * If plen does not account for pkt size
1490 */
1491 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
b33bd0cb 1492 (void)stop_ep_timer(ep);
cfdda9d7
SW
1493 abort_connection(ep, skb, GFP_KERNEL);
1494 return;
1495 }
1496 ep->plen = (u8) plen;
1497
1498 /*
1499 * If we don't have all the pdata yet, then bail.
1500 */
1501 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1502 return;
1503
1504 /*
1505 * If we get here we have accumulated the entire mpa
1506 * start reply message including private data.
1507 */
1508 ep->mpa_attr.initiator = 0;
1509 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1510 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1511 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
d2fe99e8
KS
1512 ep->mpa_attr.version = mpa->revision;
1513 if (mpa->revision == 1)
1514 ep->tried_with_mpa_v1 = 1;
1515 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1516
1517 if (mpa->revision == 2) {
1518 ep->mpa_attr.enhanced_rdma_conn =
1519 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1520 if (ep->mpa_attr.enhanced_rdma_conn) {
1521 mpa_v2_params = (struct mpa_v2_conn_params *)
1522 (ep->mpa_pkt + sizeof(*mpa));
1523 ep->ird = ntohs(mpa_v2_params->ird) &
1524 MPA_V2_IRD_ORD_MASK;
1525 ep->ord = ntohs(mpa_v2_params->ord) &
1526 MPA_V2_IRD_ORD_MASK;
1527 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1528 if (peer2peer) {
1529 if (ntohs(mpa_v2_params->ord) &
1530 MPA_V2_RDMA_WRITE_RTR)
1531 ep->mpa_attr.p2p_type =
1532 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1533 else if (ntohs(mpa_v2_params->ord) &
1534 MPA_V2_RDMA_READ_RTR)
1535 ep->mpa_attr.p2p_type =
1536 FW_RI_INIT_P2PTYPE_READ_REQ;
1537 }
1538 }
1539 } else if (mpa->revision == 1)
1540 if (peer2peer)
1541 ep->mpa_attr.p2p_type = p2p_type;
1542
cfdda9d7
SW
1543 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1544 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1545 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1546 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1547 ep->mpa_attr.p2p_type);
1548
b33bd0cb
SW
1549 /*
1550 * If the endpoint timer already expired, then we ignore
1551 * the start request. process_timeout() will abort
1552 * the connection.
1553 */
1554 if (!stop_ep_timer(ep)) {
1555 __state_set(&ep->com, MPA_REQ_RCVD);
1556
1557 /* drive upcall */
1558 mutex_lock(&ep->parent_ep->com.mutex);
1559 if (ep->parent_ep->com.state != DEAD) {
1560 if (connect_request_upcall(ep))
1561 abort_connection(ep, skb, GFP_KERNEL);
1562 } else {
be13b2df 1563 abort_connection(ep, skb, GFP_KERNEL);
b33bd0cb
SW
1564 }
1565 mutex_unlock(&ep->parent_ep->com.mutex);
be13b2df 1566 }
cfdda9d7
SW
1567 return;
1568}
1569
1570static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1571{
1572 struct c4iw_ep *ep;
1573 struct cpl_rx_data *hdr = cplhdr(skb);
1574 unsigned int dlen = ntohs(hdr->len);
1575 unsigned int tid = GET_TID(hdr);
1576 struct tid_info *t = dev->rdev.lldi.tids;
793dad94 1577 __u8 status = hdr->status;
cc18b939 1578 int disconnect = 0;
cfdda9d7
SW
1579
1580 ep = lookup_tid(t, tid);
977116c6
SW
1581 if (!ep)
1582 return 0;
cfdda9d7
SW
1583 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1584 skb_pull(skb, sizeof(*hdr));
1585 skb_trim(skb, dlen);
c529fb50 1586 mutex_lock(&ep->com.mutex);
cfdda9d7 1587
cfdda9d7
SW
1588 /* update RX credits */
1589 update_rx_credits(ep, dlen);
1590
c529fb50 1591 switch (ep->com.state) {
cfdda9d7 1592 case MPA_REQ_SENT:
55abf8df 1593 ep->rcv_seq += dlen;
cc18b939 1594 disconnect = process_mpa_reply(ep, skb);
cfdda9d7
SW
1595 break;
1596 case MPA_REQ_WAIT:
55abf8df 1597 ep->rcv_seq += dlen;
cfdda9d7
SW
1598 process_mpa_request(ep, skb);
1599 break;
1557967b
VP
1600 case FPDU_MODE: {
1601 struct c4iw_qp_attributes attrs;
1602 BUG_ON(!ep->com.qp);
e8e5b927 1603 if (status)
1557967b 1604 pr_err("%s Unexpected streaming data." \
04236df2
VP
1605 " qpid %u ep %p state %d tid %u status %d\n",
1606 __func__, ep->com.qp->wq.sq.qid, ep,
c529fb50 1607 ep->com.state, ep->hwtid, status);
97d7ec0c 1608 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1557967b 1609 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
cc18b939
SW
1610 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1611 disconnect = 1;
cfdda9d7
SW
1612 break;
1613 }
1557967b
VP
1614 default:
1615 break;
1616 }
c529fb50 1617 mutex_unlock(&ep->com.mutex);
cc18b939
SW
1618 if (disconnect)
1619 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
cfdda9d7
SW
1620 return 0;
1621}
1622
1623static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1624{
1625 struct c4iw_ep *ep;
1626 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
cfdda9d7
SW
1627 int release = 0;
1628 unsigned int tid = GET_TID(rpl);
1629 struct tid_info *t = dev->rdev.lldi.tids;
1630
1631 ep = lookup_tid(t, tid);
4984037b
VP
1632 if (!ep) {
1633 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1634 return 0;
1635 }
92dd6c3d 1636 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2f5b48c3 1637 mutex_lock(&ep->com.mutex);
cfdda9d7
SW
1638 switch (ep->com.state) {
1639 case ABORTING:
91e9c071 1640 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW
1641 __state_set(&ep->com, DEAD);
1642 release = 1;
1643 break;
1644 default:
1645 printk(KERN_ERR "%s ep %p state %d\n",
1646 __func__, ep, ep->com.state);
1647 break;
1648 }
2f5b48c3 1649 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
1650
1651 if (release)
1652 release_ep_resources(ep);
1653 return 0;
1654}
1655
5be78ee9
VP
1656static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1657{
1658 struct sk_buff *skb;
1659 struct fw_ofld_connection_wr *req;
1660 unsigned int mtu_idx;
1661 int wscale;
830662f6 1662 struct sockaddr_in *sin;
5be78ee9
VP
1663
1664 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1665 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1666 memset(req, 0, sizeof(*req));
1667 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1668 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
41b4f86c
KS
1669 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1670 ep->com.dev->rdev.lldi.ports[0],
5be78ee9 1671 ep->l2t));
9eccfe10 1672 sin = (struct sockaddr_in *)&ep->com.mapped_local_addr;
830662f6
VP
1673 req->le.lport = sin->sin_port;
1674 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
9eccfe10 1675 sin = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
830662f6
VP
1676 req->le.pport = sin->sin_port;
1677 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
5be78ee9
VP
1678 req->tcb.t_state_to_astid =
1679 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1680 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1681 req->tcb.cplrxdataack_cplpassacceptrpl =
1682 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
ef5d6355 1683 req->tcb.tx_max = (__force __be32) jiffies;
793dad94 1684 req->tcb.rcv_adv = htons(1);
5be78ee9
VP
1685 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1686 wscale = compute_wscale(rcv_win);
ef5d6355 1687 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
5be78ee9
VP
1688 (nocong ? NO_CONG(1) : 0) |
1689 KEEP_ALIVE(1) |
1690 DELACK(1) |
1691 WND_SCALE(wscale) |
1692 MSS_IDX(mtu_idx) |
1693 L2T_IDX(ep->l2t->idx) |
1694 TX_CHAN(ep->tx_chan) |
1695 SMAC_SEL(ep->smac_idx) |
1696 DSCP(ep->tos) |
1697 ULP_MODE(ULP_MODE_TCPDDP) |
ef5d6355
VP
1698 RCV_BUFSIZ(rcv_win >> 10));
1699 req->tcb.opt2 = (__force __be32) (PACE(1) |
5be78ee9
VP
1700 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1701 RX_CHANNEL(0) |
1702 CCTRL_ECN(enable_ecn) |
ef5d6355 1703 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
5be78ee9 1704 if (enable_tcp_timestamps)
ef5d6355 1705 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
5be78ee9 1706 if (enable_tcp_sack)
ef5d6355 1707 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
5be78ee9 1708 if (wscale && enable_tcp_window_scaling)
ef5d6355
VP
1709 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1710 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1711 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
793dad94
VP
1712 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1713 set_bit(ACT_OFLD_CONN, &ep->com.history);
5be78ee9
VP
1714 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1715}
1716
cfdda9d7
SW
1717/*
1718 * Return whether a failed active open has allocated a TID
1719 */
1720static inline int act_open_has_tid(int status)
1721{
1722 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1723 status != CPL_ERR_ARP_MISS;
1724}
1725
7a2cea2a
SW
1726/* Returns whether a CPL status conveys negative advice.
1727 */
1728static int is_neg_adv(unsigned int status)
1729{
1730 return status == CPL_ERR_RTX_NEG_ADVICE ||
1731 status == CPL_ERR_PERSIST_NEG_ADVICE ||
1732 status == CPL_ERR_KEEPALV_NEG_ADVICE;
1733}
1734
793dad94
VP
1735#define ACT_OPEN_RETRY_COUNT 2
1736
830662f6
VP
1737static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1738 struct dst_entry *dst, struct c4iw_dev *cdev,
1739 bool clear_mpa_v1)
1740{
1741 struct neighbour *n;
1742 int err, step;
1743 struct net_device *pdev;
1744
1745 n = dst_neigh_lookup(dst, peer_ip);
1746 if (!n)
1747 return -ENODEV;
1748
1749 rcu_read_lock();
1750 err = -ENOMEM;
1751 if (n->dev->flags & IFF_LOOPBACK) {
1752 if (iptype == 4)
1753 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1754 else if (IS_ENABLED(CONFIG_IPV6))
1755 for_each_netdev(&init_net, pdev) {
1756 if (ipv6_chk_addr(&init_net,
1757 (struct in6_addr *)peer_ip,
1758 pdev, 1))
1759 break;
1760 }
1761 else
1762 pdev = NULL;
1763
1764 if (!pdev) {
1765 err = -ENODEV;
1766 goto out;
1767 }
1768 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1769 n, pdev, 0);
1770 if (!ep->l2t)
1771 goto out;
1772 ep->mtu = pdev->mtu;
1773 ep->tx_chan = cxgb4_port_chan(pdev);
1774 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1775 step = cdev->rdev.lldi.ntxq /
1776 cdev->rdev.lldi.nchan;
1777 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1778 step = cdev->rdev.lldi.nrxq /
1779 cdev->rdev.lldi.nchan;
1780 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1781 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1782 cxgb4_port_idx(pdev) * step];
1783 dev_put(pdev);
1784 } else {
1785 pdev = get_real_dev(n->dev);
1786 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1787 n, pdev, 0);
1788 if (!ep->l2t)
1789 goto out;
1790 ep->mtu = dst_mtu(dst);
1791 ep->tx_chan = cxgb4_port_chan(n->dev);
1792 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1793 step = cdev->rdev.lldi.ntxq /
1794 cdev->rdev.lldi.nchan;
1795 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1796 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1797 step = cdev->rdev.lldi.nrxq /
1798 cdev->rdev.lldi.nchan;
1799 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1800 cxgb4_port_idx(n->dev) * step];
1801
1802 if (clear_mpa_v1) {
1803 ep->retry_with_mpa_v1 = 0;
1804 ep->tried_with_mpa_v1 = 0;
1805 }
1806 }
1807 err = 0;
1808out:
1809 rcu_read_unlock();
1810
1811 neigh_release(n);
1812
1813 return err;
1814}
1815
793dad94
VP
1816static int c4iw_reconnect(struct c4iw_ep *ep)
1817{
1818 int err = 0;
24d44a39
SW
1819 struct sockaddr_in *laddr = (struct sockaddr_in *)
1820 &ep->com.cm_id->local_addr;
1821 struct sockaddr_in *raddr = (struct sockaddr_in *)
1822 &ep->com.cm_id->remote_addr;
830662f6
VP
1823 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1824 &ep->com.cm_id->local_addr;
1825 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1826 &ep->com.cm_id->remote_addr;
1827 int iptype;
1828 __u8 *ra;
793dad94
VP
1829
1830 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1831 init_timer(&ep->timer);
1832
1833 /*
1834 * Allocate an active TID to initiate a TCP connection.
1835 */
1836 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1837 if (ep->atid == -1) {
1838 pr_err("%s - cannot alloc atid.\n", __func__);
1839 err = -ENOMEM;
1840 goto fail2;
1841 }
1842 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1843
1844 /* find a route */
830662f6
VP
1845 if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1846 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1847 raddr->sin_addr.s_addr, laddr->sin_port,
1848 raddr->sin_port, 0);
1849 iptype = 4;
1850 ra = (__u8 *)&raddr->sin_addr;
1851 } else {
1852 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1853 raddr6->sin6_addr.s6_addr,
1854 laddr6->sin6_port, raddr6->sin6_port, 0,
1855 raddr6->sin6_scope_id);
1856 iptype = 6;
1857 ra = (__u8 *)&raddr6->sin6_addr;
1858 }
1859 if (!ep->dst) {
793dad94
VP
1860 pr_err("%s - cannot find route.\n", __func__);
1861 err = -EHOSTUNREACH;
1862 goto fail3;
1863 }
830662f6
VP
1864 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1865 if (err) {
793dad94 1866 pr_err("%s - cannot alloc l2e.\n", __func__);
793dad94
VP
1867 goto fail4;
1868 }
1869
1870 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1871 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1872 ep->l2t->idx);
1873
1874 state_set(&ep->com, CONNECTING);
1875 ep->tos = 0;
1876
1877 /* send connect request to rnic */
1878 err = send_connect(ep);
1879 if (!err)
1880 goto out;
1881
1882 cxgb4_l2t_release(ep->l2t);
1883fail4:
1884 dst_release(ep->dst);
1885fail3:
1886 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1887 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1888fail2:
1889 /*
1890 * remember to send notification to upper layer.
1891 * We are in here so the upper layer is not aware that this is
1892 * re-connect attempt and so, upper layer is still waiting for
1893 * response of 1st connect request.
1894 */
1895 connect_reply_upcall(ep, -ECONNRESET);
1896 c4iw_put_ep(&ep->com);
1897out:
1898 return err;
1899}
1900
cfdda9d7
SW
1901static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1902{
1903 struct c4iw_ep *ep;
1904 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1905 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1906 ntohl(rpl->atid_status)));
1907 struct tid_info *t = dev->rdev.lldi.tids;
1908 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
830662f6
VP
1909 struct sockaddr_in *la;
1910 struct sockaddr_in *ra;
1911 struct sockaddr_in6 *la6;
1912 struct sockaddr_in6 *ra6;
cfdda9d7
SW
1913
1914 ep = lookup_atid(t, atid);
9eccfe10
SW
1915 la = (struct sockaddr_in *)&ep->com.mapped_local_addr;
1916 ra = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
1917 la6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
1918 ra6 = (struct sockaddr_in6 *)&ep->com.mapped_remote_addr;
cfdda9d7
SW
1919
1920 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1921 status, status2errno(status));
1922
7a2cea2a 1923 if (is_neg_adv(status)) {
cfdda9d7
SW
1924 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1925 atid);
1926 return 0;
1927 }
1928
793dad94
VP
1929 set_bit(ACT_OPEN_RPL, &ep->com.history);
1930
d716a2a0
VP
1931 /*
1932 * Log interesting failures.
1933 */
1934 switch (status) {
1935 case CPL_ERR_CONN_RESET:
1936 case CPL_ERR_CONN_TIMEDOUT:
1937 break;
5be78ee9 1938 case CPL_ERR_TCAM_FULL:
830662f6 1939 mutex_lock(&dev->rdev.stats.lock);
3b174d94 1940 dev->rdev.stats.tcam_full++;
830662f6
VP
1941 mutex_unlock(&dev->rdev.stats.lock);
1942 if (ep->com.local_addr.ss_family == AF_INET &&
1943 dev->rdev.lldi.enable_fw_ofld_conn) {
793dad94
VP
1944 send_fw_act_open_req(ep,
1945 GET_TID_TID(GET_AOPEN_ATID(
1946 ntohl(rpl->atid_status))));
1947 return 0;
1948 }
1949 break;
1950 case CPL_ERR_CONN_EXIST:
1951 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1952 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1953 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1954 atid);
1955 cxgb4_free_atid(t, atid);
1956 dst_release(ep->dst);
1957 cxgb4_l2t_release(ep->l2t);
1958 c4iw_reconnect(ep);
1959 return 0;
1960 }
5be78ee9 1961 break;
d716a2a0 1962 default:
830662f6
VP
1963 if (ep->com.local_addr.ss_family == AF_INET) {
1964 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1965 atid, status, status2errno(status),
1966 &la->sin_addr.s_addr, ntohs(la->sin_port),
1967 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
1968 } else {
1969 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1970 atid, status, status2errno(status),
1971 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1972 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1973 }
d716a2a0
VP
1974 break;
1975 }
1976
cfdda9d7
SW
1977 connect_reply_upcall(ep, status2errno(status));
1978 state_set(&ep->com, DEAD);
1979
1980 if (status && act_open_has_tid(status))
1981 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1982
793dad94 1983 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
cfdda9d7
SW
1984 cxgb4_free_atid(t, atid);
1985 dst_release(ep->dst);
1986 cxgb4_l2t_release(ep->l2t);
1987 c4iw_put_ep(&ep->com);
1988
1989 return 0;
1990}
1991
1992static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1993{
1994 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1995 struct tid_info *t = dev->rdev.lldi.tids;
1996 unsigned int stid = GET_TID(rpl);
1997 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1998
1999 if (!ep) {
1cab775c
VP
2000 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2001 goto out;
cfdda9d7
SW
2002 }
2003 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2004 rpl->status, status2errno(rpl->status));
d9594d99 2005 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
cfdda9d7 2006
1cab775c 2007out:
cfdda9d7
SW
2008 return 0;
2009}
2010
cfdda9d7
SW
2011static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2012{
2013 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2014 struct tid_info *t = dev->rdev.lldi.tids;
2015 unsigned int stid = GET_TID(rpl);
2016 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
2017
2018 PDBG("%s ep %p\n", __func__, ep);
d9594d99 2019 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
cfdda9d7
SW
2020 return 0;
2021}
2022
830662f6 2023static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
cfdda9d7
SW
2024 struct cpl_pass_accept_req *req)
2025{
2026 struct cpl_pass_accept_rpl *rpl;
2027 unsigned int mtu_idx;
2028 u64 opt0;
2029 u32 opt2;
2030 int wscale;
2031
2032 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2033 BUG_ON(skb_cloned(skb));
2034 skb_trim(skb, sizeof(*rpl));
2035 skb_get(skb);
2036 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
2037 wscale = compute_wscale(rcv_win);
5be78ee9
VP
2038 opt0 = (nocong ? NO_CONG(1) : 0) |
2039 KEEP_ALIVE(1) |
ba6d3925 2040 DELACK(1) |
cfdda9d7
SW
2041 WND_SCALE(wscale) |
2042 MSS_IDX(mtu_idx) |
2043 L2T_IDX(ep->l2t->idx) |
2044 TX_CHAN(ep->tx_chan) |
2045 SMAC_SEL(ep->smac_idx) |
5be78ee9 2046 DSCP(ep->tos >> 2) |
b48f3b9c 2047 ULP_MODE(ULP_MODE_TCPDDP) |
cfdda9d7
SW
2048 RCV_BUFSIZ(rcv_win>>10);
2049 opt2 = RX_CHANNEL(0) |
2050 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
2051
2052 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2053 opt2 |= TSTAMPS_EN(1);
2054 if (enable_tcp_sack && req->tcpopt.sack)
2055 opt2 |= SACK_EN(1);
2056 if (wscale && enable_tcp_window_scaling)
2057 opt2 |= WND_SCALE_EN(1);
5be78ee9
VP
2058 if (enable_ecn) {
2059 const struct tcphdr *tcph;
2060 u32 hlen = ntohl(req->hdr_len);
2061
2062 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2063 G_IP_HDR_LEN(hlen);
2064 if (tcph->ece && tcph->cwr)
2065 opt2 |= CCTRL_ECN(1);
2066 }
92e5011a
SW
2067 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
2068 opt2 |= T5_OPT_2_VALID;
2069 opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
2070 }
cfdda9d7
SW
2071
2072 rpl = cplhdr(skb);
2073 INIT_TP_WR(rpl, ep->hwtid);
2074 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2075 ep->hwtid));
2076 rpl->opt0 = cpu_to_be64(opt0);
2077 rpl->opt2 = cpu_to_be32(opt2);
d4f1a5c6 2078 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
b38a0ad8 2079 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
cfdda9d7
SW
2080 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2081
2082 return;
2083}
2084
830662f6 2085static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
cfdda9d7 2086{
830662f6 2087 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
cfdda9d7
SW
2088 BUG_ON(skb_cloned(skb));
2089 skb_trim(skb, sizeof(struct cpl_tid_release));
2090 skb_get(skb);
2091 release_tid(&dev->rdev, hwtid, skb);
2092 return;
2093}
2094
830662f6
VP
2095static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2096 __u8 *local_ip, __u8 *peer_ip,
cfdda9d7
SW
2097 __be16 *local_port, __be16 *peer_port)
2098{
2099 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2100 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2101 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
830662f6 2102 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
cfdda9d7
SW
2103 struct tcphdr *tcp = (struct tcphdr *)
2104 ((u8 *)(req + 1) + eth_len + ip_len);
2105
830662f6
VP
2106 if (ip->version == 4) {
2107 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2108 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2109 ntohs(tcp->dest));
2110 *iptype = 4;
2111 memcpy(peer_ip, &ip->saddr, 4);
2112 memcpy(local_ip, &ip->daddr, 4);
2113 } else {
2114 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2115 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2116 ntohs(tcp->dest));
2117 *iptype = 6;
2118 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2119 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2120 }
cfdda9d7
SW
2121 *peer_port = tcp->source;
2122 *local_port = tcp->dest;
2123
2124 return;
2125}
2126
2127static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2128{
793dad94 2129 struct c4iw_ep *child_ep = NULL, *parent_ep;
cfdda9d7
SW
2130 struct cpl_pass_accept_req *req = cplhdr(skb);
2131 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2132 struct tid_info *t = dev->rdev.lldi.tids;
2133 unsigned int hwtid = GET_TID(req);
2134 struct dst_entry *dst;
830662f6 2135 __u8 local_ip[16], peer_ip[16];
cfdda9d7 2136 __be16 local_port, peer_port;
3786cf18 2137 int err;
1cab775c 2138 u16 peer_mss = ntohs(req->tcpopt.mss);
830662f6 2139 int iptype;
cfdda9d7
SW
2140
2141 parent_ep = lookup_stid(t, stid);
1cab775c
VP
2142 if (!parent_ep) {
2143 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2144 goto reject;
2145 }
1cab775c 2146
cfdda9d7
SW
2147 if (state_read(&parent_ep->com) != LISTEN) {
2148 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2149 __func__);
2150 goto reject;
2151 }
2152
830662f6
VP
2153 get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2154
cfdda9d7 2155 /* Find output route */
830662f6
VP
2156 if (iptype == 4) {
2157 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2158 , __func__, parent_ep, hwtid,
2159 local_ip, peer_ip, ntohs(local_port),
2160 ntohs(peer_port), peer_mss);
2161 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2162 local_port, peer_port,
2163 GET_POPEN_TOS(ntohl(req->tos_stid)));
2164 } else {
2165 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2166 , __func__, parent_ep, hwtid,
2167 local_ip, peer_ip, ntohs(local_port),
2168 ntohs(peer_port), peer_mss);
2169 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2170 PASS_OPEN_TOS(ntohl(req->tos_stid)),
2171 ((struct sockaddr_in6 *)
2172 &parent_ep->com.local_addr)->sin6_scope_id);
2173 }
2174 if (!dst) {
cfdda9d7
SW
2175 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2176 __func__);
2177 goto reject;
2178 }
3786cf18
DM
2179
2180 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2181 if (!child_ep) {
2182 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
cfdda9d7
SW
2183 __func__);
2184 dst_release(dst);
2185 goto reject;
2186 }
2187
830662f6 2188 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
3786cf18
DM
2189 if (err) {
2190 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
cfdda9d7 2191 __func__);
cfdda9d7 2192 dst_release(dst);
3786cf18 2193 kfree(child_ep);
cfdda9d7
SW
2194 goto reject;
2195 }
3786cf18 2196
1cab775c
VP
2197 if (peer_mss && child_ep->mtu > (peer_mss + 40))
2198 child_ep->mtu = peer_mss + 40;
2199
cfdda9d7
SW
2200 state_set(&child_ep->com, CONNECTING);
2201 child_ep->com.dev = dev;
2202 child_ep->com.cm_id = NULL;
830662f6
VP
2203 if (iptype == 4) {
2204 struct sockaddr_in *sin = (struct sockaddr_in *)
2205 &child_ep->com.local_addr;
2206 sin->sin_family = PF_INET;
2207 sin->sin_port = local_port;
2208 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2209 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2210 sin->sin_family = PF_INET;
2211 sin->sin_port = peer_port;
2212 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2213 } else {
2214 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2215 &child_ep->com.local_addr;
2216 sin6->sin6_family = PF_INET6;
2217 sin6->sin6_port = local_port;
2218 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2219 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2220 sin6->sin6_family = PF_INET6;
2221 sin6->sin6_port = peer_port;
2222 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2223 }
cfdda9d7
SW
2224 c4iw_get_ep(&parent_ep->com);
2225 child_ep->parent_ep = parent_ep;
2226 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
cfdda9d7
SW
2227 child_ep->dst = dst;
2228 child_ep->hwtid = hwtid;
cfdda9d7
SW
2229
2230 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
3786cf18 2231 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
cfdda9d7
SW
2232
2233 init_timer(&child_ep->timer);
2234 cxgb4_insert_tid(t, child_ep, hwtid);
b3de6cfe 2235 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
830662f6 2236 accept_cr(child_ep, skb, req);
793dad94 2237 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
cfdda9d7
SW
2238 goto out;
2239reject:
830662f6 2240 reject_cr(dev, hwtid, skb);
cfdda9d7
SW
2241out:
2242 return 0;
2243}
2244
2245static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2246{
2247 struct c4iw_ep *ep;
2248 struct cpl_pass_establish *req = cplhdr(skb);
2249 struct tid_info *t = dev->rdev.lldi.tids;
2250 unsigned int tid = GET_TID(req);
2251
2252 ep = lookup_tid(t, tid);
2253 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2254 ep->snd_seq = be32_to_cpu(req->snd_isn);
2255 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2256
1cab775c
VP
2257 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2258 ntohs(req->tcp_opt));
2259
cfdda9d7
SW
2260 set_emss(ep, ntohs(req->tcp_opt));
2261
2262 dst_confirm(ep->dst);
2263 state_set(&ep->com, MPA_REQ_WAIT);
2264 start_ep_timer(ep);
2265 send_flowc(ep, skb);
793dad94 2266 set_bit(PASS_ESTAB, &ep->com.history);
cfdda9d7
SW
2267
2268 return 0;
2269}
2270
2271static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2272{
2273 struct cpl_peer_close *hdr = cplhdr(skb);
2274 struct c4iw_ep *ep;
2275 struct c4iw_qp_attributes attrs;
cfdda9d7
SW
2276 int disconnect = 1;
2277 int release = 0;
cfdda9d7
SW
2278 struct tid_info *t = dev->rdev.lldi.tids;
2279 unsigned int tid = GET_TID(hdr);
8da7e7a5 2280 int ret;
cfdda9d7
SW
2281
2282 ep = lookup_tid(t, tid);
2283 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2284 dst_confirm(ep->dst);
2285
793dad94 2286 set_bit(PEER_CLOSE, &ep->com.history);
2f5b48c3 2287 mutex_lock(&ep->com.mutex);
cfdda9d7
SW
2288 switch (ep->com.state) {
2289 case MPA_REQ_WAIT:
2290 __state_set(&ep->com, CLOSING);
2291 break;
2292 case MPA_REQ_SENT:
2293 __state_set(&ep->com, CLOSING);
2294 connect_reply_upcall(ep, -ECONNRESET);
2295 break;
2296 case MPA_REQ_RCVD:
2297
2298 /*
2299 * We're gonna mark this puppy DEAD, but keep
2300 * the reference on it until the ULP accepts or
2301 * rejects the CR. Also wake up anyone waiting
2302 * in rdma connection migration (see c4iw_accept_cr()).
2303 */
2304 __state_set(&ep->com, CLOSING);
cfdda9d7 2305 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2306 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW
2307 break;
2308 case MPA_REP_SENT:
2309 __state_set(&ep->com, CLOSING);
cfdda9d7 2310 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
d9594d99 2311 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
cfdda9d7
SW
2312 break;
2313 case FPDU_MODE:
ca5a2202 2314 start_ep_timer(ep);
cfdda9d7 2315 __state_set(&ep->com, CLOSING);
30c95c2d 2316 attrs.next_state = C4IW_QP_STATE_CLOSING;
8da7e7a5 2317 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
30c95c2d 2318 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
8da7e7a5
SW
2319 if (ret != -ECONNRESET) {
2320 peer_close_upcall(ep);
2321 disconnect = 1;
2322 }
cfdda9d7
SW
2323 break;
2324 case ABORTING:
2325 disconnect = 0;
2326 break;
2327 case CLOSING:
2328 __state_set(&ep->com, MORIBUND);
2329 disconnect = 0;
2330 break;
2331 case MORIBUND:
b33bd0cb 2332 (void)stop_ep_timer(ep);
cfdda9d7
SW
2333 if (ep->com.cm_id && ep->com.qp) {
2334 attrs.next_state = C4IW_QP_STATE_IDLE;
2335 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2336 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2337 }
be13b2df 2338 close_complete_upcall(ep, 0);
cfdda9d7
SW
2339 __state_set(&ep->com, DEAD);
2340 release = 1;
2341 disconnect = 0;
2342 break;
2343 case DEAD:
2344 disconnect = 0;
2345 break;
2346 default:
2347 BUG_ON(1);
2348 }
2f5b48c3 2349 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
2350 if (disconnect)
2351 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2352 if (release)
2353 release_ep_resources(ep);
2354 return 0;
2355}
2356
cfdda9d7
SW
2357static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2358{
2359 struct cpl_abort_req_rss *req = cplhdr(skb);
2360 struct c4iw_ep *ep;
2361 struct cpl_abort_rpl *rpl;
2362 struct sk_buff *rpl_skb;
2363 struct c4iw_qp_attributes attrs;
2364 int ret;
2365 int release = 0;
cfdda9d7
SW
2366 struct tid_info *t = dev->rdev.lldi.tids;
2367 unsigned int tid = GET_TID(req);
cfdda9d7
SW
2368
2369 ep = lookup_tid(t, tid);
7a2cea2a 2370 if (is_neg_adv(req->status)) {
cfdda9d7
SW
2371 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2372 ep->hwtid);
2373 return 0;
2374 }
cfdda9d7
SW
2375 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2376 ep->com.state);
793dad94 2377 set_bit(PEER_ABORT, &ep->com.history);
2f5b48c3
SW
2378
2379 /*
2380 * Wake up any threads in rdma_init() or rdma_fini().
d2fe99e8
KS
2381 * However, this is not needed if com state is just
2382 * MPA_REQ_SENT
2f5b48c3 2383 */
d2fe99e8
KS
2384 if (ep->com.state != MPA_REQ_SENT)
2385 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2f5b48c3
SW
2386
2387 mutex_lock(&ep->com.mutex);
cfdda9d7
SW
2388 switch (ep->com.state) {
2389 case CONNECTING:
2390 break;
2391 case MPA_REQ_WAIT:
b33bd0cb 2392 (void)stop_ep_timer(ep);
cfdda9d7
SW
2393 break;
2394 case MPA_REQ_SENT:
b33bd0cb 2395 (void)stop_ep_timer(ep);
fe7e0a4d 2396 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
d2fe99e8
KS
2397 connect_reply_upcall(ep, -ECONNRESET);
2398 else {
2399 /*
2400 * we just don't send notification upwards because we
2401 * want to retry with mpa_v1 without upper layers even
2402 * knowing it.
2403 *
2404 * do some housekeeping so as to re-initiate the
2405 * connection
2406 */
2407 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2408 mpa_rev);
2409 ep->retry_with_mpa_v1 = 1;
2410 }
cfdda9d7
SW
2411 break;
2412 case MPA_REP_SENT:
cfdda9d7
SW
2413 break;
2414 case MPA_REQ_RCVD:
cfdda9d7
SW
2415 break;
2416 case MORIBUND:
2417 case CLOSING:
ca5a2202 2418 stop_ep_timer(ep);
cfdda9d7
SW
2419 /*FALLTHROUGH*/
2420 case FPDU_MODE:
2421 if (ep->com.cm_id && ep->com.qp) {
2422 attrs.next_state = C4IW_QP_STATE_ERROR;
2423 ret = c4iw_modify_qp(ep->com.qp->rhp,
2424 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2425 &attrs, 1);
2426 if (ret)
2427 printk(KERN_ERR MOD
2428 "%s - qp <- error failed!\n",
2429 __func__);
2430 }
2431 peer_abort_upcall(ep);
2432 break;
2433 case ABORTING:
2434 break;
2435 case DEAD:
2436 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2f5b48c3 2437 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
2438 return 0;
2439 default:
2440 BUG_ON(1);
2441 break;
2442 }
2443 dst_confirm(ep->dst);
2444 if (ep->com.state != ABORTING) {
2445 __state_set(&ep->com, DEAD);
d2fe99e8
KS
2446 /* we don't release if we want to retry with mpa_v1 */
2447 if (!ep->retry_with_mpa_v1)
2448 release = 1;
cfdda9d7 2449 }
2f5b48c3 2450 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
2451
2452 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2453 if (!rpl_skb) {
2454 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2455 __func__);
2456 release = 1;
2457 goto out;
2458 }
2459 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2460 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2461 INIT_TP_WR(rpl, ep->hwtid);
2462 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2463 rpl->cmd = CPL_ABORT_NO_RST;
2464 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2465out:
cfdda9d7
SW
2466 if (release)
2467 release_ep_resources(ep);
fe7e0a4d
VP
2468 else if (ep->retry_with_mpa_v1) {
2469 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
d2fe99e8
KS
2470 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2471 dst_release(ep->dst);
2472 cxgb4_l2t_release(ep->l2t);
2473 c4iw_reconnect(ep);
2474 }
2475
cfdda9d7
SW
2476 return 0;
2477}
2478
2479static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2480{
2481 struct c4iw_ep *ep;
2482 struct c4iw_qp_attributes attrs;
2483 struct cpl_close_con_rpl *rpl = cplhdr(skb);
cfdda9d7
SW
2484 int release = 0;
2485 struct tid_info *t = dev->rdev.lldi.tids;
2486 unsigned int tid = GET_TID(rpl);
cfdda9d7
SW
2487
2488 ep = lookup_tid(t, tid);
2489
2490 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2491 BUG_ON(!ep);
2492
2493 /* The cm_id may be null if we failed to connect */
2f5b48c3 2494 mutex_lock(&ep->com.mutex);
cfdda9d7
SW
2495 switch (ep->com.state) {
2496 case CLOSING:
2497 __state_set(&ep->com, MORIBUND);
2498 break;
2499 case MORIBUND:
b33bd0cb 2500 (void)stop_ep_timer(ep);
cfdda9d7
SW
2501 if ((ep->com.cm_id) && (ep->com.qp)) {
2502 attrs.next_state = C4IW_QP_STATE_IDLE;
2503 c4iw_modify_qp(ep->com.qp->rhp,
2504 ep->com.qp,
2505 C4IW_QP_ATTR_NEXT_STATE,
2506 &attrs, 1);
2507 }
be13b2df 2508 close_complete_upcall(ep, 0);
cfdda9d7
SW
2509 __state_set(&ep->com, DEAD);
2510 release = 1;
2511 break;
2512 case ABORTING:
2513 case DEAD:
2514 break;
2515 default:
2516 BUG_ON(1);
2517 break;
2518 }
2f5b48c3 2519 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
2520 if (release)
2521 release_ep_resources(ep);
2522 return 0;
2523}
2524
2525static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2526{
0e42c1f4 2527 struct cpl_rdma_terminate *rpl = cplhdr(skb);
cfdda9d7 2528 struct tid_info *t = dev->rdev.lldi.tids;
0e42c1f4
SW
2529 unsigned int tid = GET_TID(rpl);
2530 struct c4iw_ep *ep;
2531 struct c4iw_qp_attributes attrs;
cfdda9d7
SW
2532
2533 ep = lookup_tid(t, tid);
0e42c1f4 2534 BUG_ON(!ep);
cfdda9d7 2535
30c95c2d 2536 if (ep && ep->com.qp) {
0e42c1f4
SW
2537 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2538 ep->com.qp->wq.sq.qid);
2539 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2540 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2541 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2542 } else
30c95c2d 2543 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
cfdda9d7 2544
cfdda9d7
SW
2545 return 0;
2546}
2547
2548/*
2549 * Upcall from the adapter indicating data has been transmitted.
2550 * For us its just the single MPA request or reply. We can now free
2551 * the skb holding the mpa message.
2552 */
2553static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2554{
2555 struct c4iw_ep *ep;
2556 struct cpl_fw4_ack *hdr = cplhdr(skb);
2557 u8 credits = hdr->credits;
2558 unsigned int tid = GET_TID(hdr);
2559 struct tid_info *t = dev->rdev.lldi.tids;
2560
2561
2562 ep = lookup_tid(t, tid);
2563 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2564 if (credits == 0) {
aa1ad260
JP
2565 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2566 __func__, ep, ep->hwtid, state_read(&ep->com));
cfdda9d7
SW
2567 return 0;
2568 }
2569
2570 dst_confirm(ep->dst);
2571 if (ep->mpa_skb) {
2572 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2573 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2574 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2575 kfree_skb(ep->mpa_skb);
2576 ep->mpa_skb = NULL;
2577 }
2578 return 0;
2579}
2580
cfdda9d7
SW
2581int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2582{
a7db89eb
SW
2583 int err = 0;
2584 int disconnect = 0;
cfdda9d7
SW
2585 struct c4iw_ep *ep = to_ep(cm_id);
2586 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2587
a7db89eb
SW
2588 mutex_lock(&ep->com.mutex);
2589 if (ep->com.state == DEAD) {
2590 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
2591 c4iw_put_ep(&ep->com);
2592 return -ECONNRESET;
2593 }
793dad94 2594 set_bit(ULP_REJECT, &ep->com.history);
a7db89eb 2595 BUG_ON(ep->com.state != MPA_REQ_RCVD);
cfdda9d7
SW
2596 if (mpa_rev == 0)
2597 abort_connection(ep, NULL, GFP_KERNEL);
2598 else {
2599 err = send_mpa_reject(ep, pdata, pdata_len);
a7db89eb 2600 disconnect = 1;
cfdda9d7 2601 }
a7db89eb
SW
2602 mutex_unlock(&ep->com.mutex);
2603 if (disconnect)
2604 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
cfdda9d7
SW
2605 c4iw_put_ep(&ep->com);
2606 return 0;
2607}
2608
2609int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2610{
2611 int err;
2612 struct c4iw_qp_attributes attrs;
2613 enum c4iw_qp_attr_mask mask;
2614 struct c4iw_ep *ep = to_ep(cm_id);
2615 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2616 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2617
2618 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
a7db89eb
SW
2619
2620 mutex_lock(&ep->com.mutex);
2621 if (ep->com.state == DEAD) {
cfdda9d7
SW
2622 err = -ECONNRESET;
2623 goto err;
2624 }
2625
a7db89eb 2626 BUG_ON(ep->com.state != MPA_REQ_RCVD);
cfdda9d7
SW
2627 BUG_ON(!qp);
2628
793dad94 2629 set_bit(ULP_ACCEPT, &ep->com.history);
be4c9bad
RD
2630 if ((conn_param->ord > c4iw_max_read_depth) ||
2631 (conn_param->ird > c4iw_max_read_depth)) {
cfdda9d7
SW
2632 abort_connection(ep, NULL, GFP_KERNEL);
2633 err = -EINVAL;
2634 goto err;
2635 }
2636
d2fe99e8
KS
2637 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2638 if (conn_param->ord > ep->ird) {
2639 ep->ird = conn_param->ird;
2640 ep->ord = conn_param->ord;
2641 send_mpa_reject(ep, conn_param->private_data,
2642 conn_param->private_data_len);
2643 abort_connection(ep, NULL, GFP_KERNEL);
2644 err = -ENOMEM;
2645 goto err;
2646 }
2647 if (conn_param->ird > ep->ord) {
2648 if (!ep->ord)
2649 conn_param->ird = 1;
2650 else {
2651 abort_connection(ep, NULL, GFP_KERNEL);
2652 err = -ENOMEM;
2653 goto err;
2654 }
2655 }
cfdda9d7 2656
d2fe99e8 2657 }
cfdda9d7
SW
2658 ep->ird = conn_param->ird;
2659 ep->ord = conn_param->ord;
2660
d2fe99e8
KS
2661 if (ep->mpa_attr.version != 2)
2662 if (peer2peer && ep->ird == 0)
2663 ep->ird = 1;
cfdda9d7
SW
2664
2665 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2666
d2fe99e8
KS
2667 cm_id->add_ref(cm_id);
2668 ep->com.cm_id = cm_id;
2669 ep->com.qp = qp;
325abead 2670 ref_qp(ep);
d2fe99e8 2671
cfdda9d7
SW
2672 /* bind QP to EP and move to RTS */
2673 attrs.mpa_attr = ep->mpa_attr;
2674 attrs.max_ird = ep->ird;
2675 attrs.max_ord = ep->ord;
2676 attrs.llp_stream_handle = ep;
2677 attrs.next_state = C4IW_QP_STATE_RTS;
2678
2679 /* bind QP and TID with INIT_WR */
2680 mask = C4IW_QP_ATTR_NEXT_STATE |
2681 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2682 C4IW_QP_ATTR_MPA_ATTR |
2683 C4IW_QP_ATTR_MAX_IRD |
2684 C4IW_QP_ATTR_MAX_ORD;
2685
2686 err = c4iw_modify_qp(ep->com.qp->rhp,
2687 ep->com.qp, mask, &attrs, 1);
2688 if (err)
2689 goto err1;
2690 err = send_mpa_reply(ep, conn_param->private_data,
2691 conn_param->private_data_len);
2692 if (err)
2693 goto err1;
2694
a7db89eb 2695 __state_set(&ep->com, FPDU_MODE);
cfdda9d7 2696 established_upcall(ep);
a7db89eb 2697 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
2698 c4iw_put_ep(&ep->com);
2699 return 0;
2700err1:
2701 ep->com.cm_id = NULL;
cfdda9d7
SW
2702 cm_id->rem_ref(cm_id);
2703err:
a7db89eb 2704 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
2705 c4iw_put_ep(&ep->com);
2706 return err;
2707}
2708
830662f6
VP
2709static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2710{
2711 struct in_device *ind;
2712 int found = 0;
2713 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2714 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2715
2716 ind = in_dev_get(dev->rdev.lldi.ports[0]);
2717 if (!ind)
2718 return -EADDRNOTAVAIL;
2719 for_primary_ifa(ind) {
2720 laddr->sin_addr.s_addr = ifa->ifa_address;
2721 raddr->sin_addr.s_addr = ifa->ifa_address;
2722 found = 1;
2723 break;
2724 }
2725 endfor_ifa(ind);
2726 in_dev_put(ind);
2727 return found ? 0 : -EADDRNOTAVAIL;
2728}
2729
2730static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2731 unsigned char banned_flags)
2732{
2733 struct inet6_dev *idev;
2734 int err = -EADDRNOTAVAIL;
2735
2736 rcu_read_lock();
2737 idev = __in6_dev_get(dev);
2738 if (idev != NULL) {
2739 struct inet6_ifaddr *ifp;
2740
2741 read_lock_bh(&idev->lock);
2742 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2743 if (ifp->scope == IFA_LINK &&
2744 !(ifp->flags & banned_flags)) {
2745 memcpy(addr, &ifp->addr, 16);
2746 err = 0;
2747 break;
2748 }
2749 }
2750 read_unlock_bh(&idev->lock);
2751 }
2752 rcu_read_unlock();
2753 return err;
2754}
2755
2756static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2757{
2758 struct in6_addr uninitialized_var(addr);
2759 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2760 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2761
2762 if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2763 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2764 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2765 return 0;
2766 }
2767 return -EADDRNOTAVAIL;
2768}
2769
cfdda9d7
SW
2770int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2771{
cfdda9d7
SW
2772 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2773 struct c4iw_ep *ep;
3786cf18 2774 int err = 0;
9eccfe10
SW
2775 struct sockaddr_in *laddr;
2776 struct sockaddr_in *raddr;
2777 struct sockaddr_in6 *laddr6;
2778 struct sockaddr_in6 *raddr6;
2779 struct iwpm_dev_data pm_reg_msg;
2780 struct iwpm_sa_data pm_msg;
830662f6
VP
2781 __u8 *ra;
2782 int iptype;
9eccfe10 2783 int iwpm_err = 0;
cfdda9d7 2784
be4c9bad
RD
2785 if ((conn_param->ord > c4iw_max_read_depth) ||
2786 (conn_param->ird > c4iw_max_read_depth)) {
2787 err = -EINVAL;
2788 goto out;
2789 }
cfdda9d7
SW
2790 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2791 if (!ep) {
2792 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2793 err = -ENOMEM;
2794 goto out;
2795 }
2796 init_timer(&ep->timer);
2797 ep->plen = conn_param->private_data_len;
2798 if (ep->plen)
2799 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2800 conn_param->private_data, ep->plen);
2801 ep->ird = conn_param->ird;
2802 ep->ord = conn_param->ord;
2803
2804 if (peer2peer && ep->ord == 0)
2805 ep->ord = 1;
2806
2807 cm_id->add_ref(cm_id);
2808 ep->com.dev = dev;
2809 ep->com.cm_id = cm_id;
2810 ep->com.qp = get_qhp(dev, conn_param->qpn);
830662f6
VP
2811 if (!ep->com.qp) {
2812 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2813 err = -EINVAL;
9eccfe10 2814 goto fail1;
830662f6 2815 }
325abead 2816 ref_qp(ep);
cfdda9d7
SW
2817 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2818 ep->com.qp, cm_id);
2819
2820 /*
2821 * Allocate an active TID to initiate a TCP connection.
2822 */
2823 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2824 if (ep->atid == -1) {
2825 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2826 err = -ENOMEM;
9eccfe10 2827 goto fail1;
cfdda9d7 2828 }
793dad94 2829 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
cfdda9d7 2830
9eccfe10
SW
2831 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2832 sizeof(ep->com.local_addr));
2833 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2834 sizeof(ep->com.remote_addr));
2835
2836 /* No port mapper available, go with the specified peer information */
2837 memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
2838 sizeof(ep->com.mapped_local_addr));
2839 memcpy(&ep->com.mapped_remote_addr, &cm_id->remote_addr,
2840 sizeof(ep->com.mapped_remote_addr));
2841
2842 c4iw_form_reg_msg(dev, &pm_reg_msg);
2843 iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
2844 if (iwpm_err) {
2845 PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
2846 __func__, iwpm_err);
2847 }
2848 if (iwpm_valid_pid() && !iwpm_err) {
2849 c4iw_form_pm_msg(ep, &pm_msg);
2850 iwpm_err = iwpm_add_and_query_mapping(&pm_msg, RDMA_NL_C4IW);
2851 if (iwpm_err)
2852 PDBG("%s: Port Mapper query fail (err = %d).\n",
2853 __func__, iwpm_err);
2854 else
2855 c4iw_record_pm_msg(ep, &pm_msg);
2856 }
2857 if (iwpm_create_mapinfo(&ep->com.local_addr,
2858 &ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
2859 iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
2860 err = -ENOMEM;
2861 goto fail1;
2862 }
2863 print_addr(&ep->com, __func__, "add_query/create_mapinfo");
2864 set_bit(RELEASE_MAPINFO, &ep->com.flags);
2865
2866 laddr = (struct sockaddr_in *)&ep->com.mapped_local_addr;
2867 raddr = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
2868 laddr6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
2869 raddr6 = (struct sockaddr_in6 *) &ep->com.mapped_remote_addr;
2870
830662f6
VP
2871 if (cm_id->remote_addr.ss_family == AF_INET) {
2872 iptype = 4;
2873 ra = (__u8 *)&raddr->sin_addr;
cfdda9d7 2874
830662f6
VP
2875 /*
2876 * Handle loopback requests to INADDR_ANY.
2877 */
2878 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2879 err = pick_local_ipaddrs(dev, cm_id);
2880 if (err)
9eccfe10 2881 goto fail1;
830662f6
VP
2882 }
2883
2884 /* find a route */
2885 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2886 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2887 ra, ntohs(raddr->sin_port));
2888 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2889 raddr->sin_addr.s_addr, laddr->sin_port,
2890 raddr->sin_port, 0);
2891 } else {
2892 iptype = 6;
2893 ra = (__u8 *)&raddr6->sin6_addr;
2894
2895 /*
2896 * Handle loopback requests to INADDR_ANY.
2897 */
2898 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2899 err = pick_local_ip6addrs(dev, cm_id);
2900 if (err)
9eccfe10 2901 goto fail1;
830662f6
VP
2902 }
2903
2904 /* find a route */
2905 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2906 __func__, laddr6->sin6_addr.s6_addr,
2907 ntohs(laddr6->sin6_port),
2908 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2909 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2910 raddr6->sin6_addr.s6_addr,
2911 laddr6->sin6_port, raddr6->sin6_port, 0,
2912 raddr6->sin6_scope_id);
2913 }
2914 if (!ep->dst) {
cfdda9d7
SW
2915 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2916 err = -EHOSTUNREACH;
9eccfe10 2917 goto fail2;
cfdda9d7 2918 }
cfdda9d7 2919
830662f6 2920 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
3786cf18 2921 if (err) {
cfdda9d7 2922 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
9eccfe10 2923 goto fail3;
cfdda9d7
SW
2924 }
2925
2926 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2927 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2928 ep->l2t->idx);
2929
2930 state_set(&ep->com, CONNECTING);
2931 ep->tos = 0;
cfdda9d7
SW
2932
2933 /* send connect request to rnic */
2934 err = send_connect(ep);
2935 if (!err)
2936 goto out;
2937
2938 cxgb4_l2t_release(ep->l2t);
cfdda9d7 2939fail3:
9eccfe10
SW
2940 dst_release(ep->dst);
2941fail2:
793dad94 2942 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
cfdda9d7 2943 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
9eccfe10 2944fail1:
cfdda9d7
SW
2945 cm_id->rem_ref(cm_id);
2946 c4iw_put_ep(&ep->com);
2947out:
2948 return err;
2949}
2950
830662f6
VP
2951static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2952{
2953 int err;
9eccfe10
SW
2954 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2955 &ep->com.mapped_local_addr;
830662f6
VP
2956
2957 c4iw_init_wr_wait(&ep->com.wr_wait);
2958 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2959 ep->stid, &sin6->sin6_addr,
2960 sin6->sin6_port,
2961 ep->com.dev->rdev.lldi.rxq_ids[0]);
2962 if (!err)
2963 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2964 &ep->com.wr_wait,
2965 0, 0, __func__);
2966 if (err)
2967 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2968 err, ep->stid,
2969 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2970 return err;
2971}
2972
2973static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2974{
2975 int err;
9eccfe10
SW
2976 struct sockaddr_in *sin = (struct sockaddr_in *)
2977 &ep->com.mapped_local_addr;
830662f6
VP
2978
2979 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2980 do {
2981 err = cxgb4_create_server_filter(
2982 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2983 sin->sin_addr.s_addr, sin->sin_port, 0,
2984 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2985 if (err == -EBUSY) {
2986 set_current_state(TASK_UNINTERRUPTIBLE);
2987 schedule_timeout(usecs_to_jiffies(100));
2988 }
2989 } while (err == -EBUSY);
2990 } else {
2991 c4iw_init_wr_wait(&ep->com.wr_wait);
2992 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2993 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2994 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2995 if (!err)
2996 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2997 &ep->com.wr_wait,
2998 0, 0, __func__);
2999 }
3000 if (err)
3001 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3002 , err, ep->stid,
3003 &sin->sin_addr, ntohs(sin->sin_port));
3004 return err;
3005}
3006
cfdda9d7
SW
3007int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3008{
3009 int err = 0;
3010 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3011 struct c4iw_listen_ep *ep;
9eccfe10
SW
3012 struct iwpm_dev_data pm_reg_msg;
3013 struct iwpm_sa_data pm_msg;
3014 int iwpm_err = 0;
cfdda9d7 3015
cfdda9d7
SW
3016 might_sleep();
3017
3018 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3019 if (!ep) {
3020 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3021 err = -ENOMEM;
3022 goto fail1;
3023 }
3024 PDBG("%s ep %p\n", __func__, ep);
3025 cm_id->add_ref(cm_id);
3026 ep->com.cm_id = cm_id;
3027 ep->com.dev = dev;
3028 ep->backlog = backlog;
24d44a39
SW
3029 memcpy(&ep->com.local_addr, &cm_id->local_addr,
3030 sizeof(ep->com.local_addr));
cfdda9d7
SW
3031
3032 /*
3033 * Allocate a server TID.
3034 */
8c044690
KS
3035 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3036 ep->com.local_addr.ss_family == AF_INET)
830662f6
VP
3037 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3038 cm_id->local_addr.ss_family, ep);
1cab775c 3039 else
830662f6
VP
3040 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3041 cm_id->local_addr.ss_family, ep);
1cab775c 3042
cfdda9d7 3043 if (ep->stid == -1) {
be4c9bad 3044 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
cfdda9d7
SW
3045 err = -ENOMEM;
3046 goto fail2;
3047 }
793dad94 3048 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
9eccfe10
SW
3049
3050 /* No port mapper available, go with the specified info */
3051 memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
3052 sizeof(ep->com.mapped_local_addr));
3053
3054 c4iw_form_reg_msg(dev, &pm_reg_msg);
3055 iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
3056 if (iwpm_err) {
3057 PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
3058 __func__, iwpm_err);
3059 }
3060 if (iwpm_valid_pid() && !iwpm_err) {
3061 memcpy(&pm_msg.loc_addr, &ep->com.local_addr,
3062 sizeof(ep->com.local_addr));
3063 iwpm_err = iwpm_add_mapping(&pm_msg, RDMA_NL_C4IW);
3064 if (iwpm_err)
3065 PDBG("%s: Port Mapper query fail (err = %d).\n",
3066 __func__, iwpm_err);
3067 else
3068 memcpy(&ep->com.mapped_local_addr,
3069 &pm_msg.mapped_loc_addr,
3070 sizeof(ep->com.mapped_local_addr));
3071 }
3072 if (iwpm_create_mapinfo(&ep->com.local_addr,
3073 &ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
3074 err = -ENOMEM;
3075 goto fail3;
3076 }
3077 print_addr(&ep->com, __func__, "add_mapping/create_mapinfo");
3078
3079 set_bit(RELEASE_MAPINFO, &ep->com.flags);
cfdda9d7 3080 state_set(&ep->com, LISTEN);
830662f6
VP
3081 if (ep->com.local_addr.ss_family == AF_INET)
3082 err = create_server4(dev, ep);
3083 else
3084 err = create_server6(dev, ep);
cfdda9d7
SW
3085 if (!err) {
3086 cm_id->provider_data = ep;
3087 goto out;
3088 }
9eccfe10
SW
3089
3090fail3:
830662f6
VP
3091 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3092 ep->com.local_addr.ss_family);
cfdda9d7
SW
3093fail2:
3094 cm_id->rem_ref(cm_id);
3095 c4iw_put_ep(&ep->com);
3096fail1:
3097out:
3098 return err;
3099}
3100
3101int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3102{
3103 int err;
3104 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3105
3106 PDBG("%s ep %p\n", __func__, ep);
3107
3108 might_sleep();
3109 state_set(&ep->com, DEAD);
830662f6
VP
3110 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3111 ep->com.local_addr.ss_family == AF_INET) {
1cab775c
VP
3112 err = cxgb4_remove_server_filter(
3113 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3114 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3115 } else {
3116 c4iw_init_wr_wait(&ep->com.wr_wait);
830662f6
VP
3117 err = cxgb4_remove_server(
3118 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3119 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
1cab775c
VP
3120 if (err)
3121 goto done;
3122 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3123 0, 0, __func__);
3124 }
793dad94 3125 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
830662f6
VP
3126 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3127 ep->com.local_addr.ss_family);
cfdda9d7 3128done:
cfdda9d7
SW
3129 cm_id->rem_ref(cm_id);
3130 c4iw_put_ep(&ep->com);
3131 return err;
3132}
3133
3134int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3135{
3136 int ret = 0;
cfdda9d7
SW
3137 int close = 0;
3138 int fatal = 0;
3139 struct c4iw_rdev *rdev;
cfdda9d7 3140
2f5b48c3 3141 mutex_lock(&ep->com.mutex);
cfdda9d7
SW
3142
3143 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3144 states[ep->com.state], abrupt);
3145
3146 rdev = &ep->com.dev->rdev;
3147 if (c4iw_fatal_error(rdev)) {
3148 fatal = 1;
be13b2df 3149 close_complete_upcall(ep, -EIO);
cfdda9d7
SW
3150 ep->com.state = DEAD;
3151 }
3152 switch (ep->com.state) {
3153 case MPA_REQ_WAIT:
3154 case MPA_REQ_SENT:
3155 case MPA_REQ_RCVD:
3156 case MPA_REP_SENT:
3157 case FPDU_MODE:
3158 close = 1;
3159 if (abrupt)
3160 ep->com.state = ABORTING;
3161 else {
3162 ep->com.state = CLOSING;
ca5a2202 3163 start_ep_timer(ep);
cfdda9d7
SW
3164 }
3165 set_bit(CLOSE_SENT, &ep->com.flags);
3166 break;
3167 case CLOSING:
3168 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3169 close = 1;
3170 if (abrupt) {
b33bd0cb 3171 (void)stop_ep_timer(ep);
cfdda9d7
SW
3172 ep->com.state = ABORTING;
3173 } else
3174 ep->com.state = MORIBUND;
3175 }
3176 break;
3177 case MORIBUND:
3178 case ABORTING:
3179 case DEAD:
3180 PDBG("%s ignoring disconnect ep %p state %u\n",
3181 __func__, ep, ep->com.state);
3182 break;
3183 default:
3184 BUG();
3185 break;
3186 }
3187
cfdda9d7 3188 if (close) {
8da7e7a5 3189 if (abrupt) {
793dad94 3190 set_bit(EP_DISC_ABORT, &ep->com.history);
be13b2df 3191 close_complete_upcall(ep, -ECONNRESET);
8da7e7a5 3192 ret = send_abort(ep, NULL, gfp);
793dad94
VP
3193 } else {
3194 set_bit(EP_DISC_CLOSE, &ep->com.history);
cfdda9d7 3195 ret = send_halfclose(ep, gfp);
793dad94 3196 }
cfdda9d7
SW
3197 if (ret)
3198 fatal = 1;
3199 }
8da7e7a5 3200 mutex_unlock(&ep->com.mutex);
cfdda9d7
SW
3201 if (fatal)
3202 release_ep_resources(ep);
3203 return ret;
3204}
3205
1cab775c
VP
3206static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3207 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3208{
3209 struct c4iw_ep *ep;
793dad94 3210 int atid = be32_to_cpu(req->tid);
1cab775c 3211
ef5d6355
VP
3212 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3213 (__force u32) req->tid);
1cab775c
VP
3214 if (!ep)
3215 return;
3216
3217 switch (req->retval) {
3218 case FW_ENOMEM:
793dad94
VP
3219 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3220 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3221 send_fw_act_open_req(ep, atid);
3222 return;
3223 }
1cab775c 3224 case FW_EADDRINUSE:
793dad94
VP
3225 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3226 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3227 send_fw_act_open_req(ep, atid);
3228 return;
3229 }
1cab775c
VP
3230 break;
3231 default:
3232 pr_info("%s unexpected ofld conn wr retval %d\n",
3233 __func__, req->retval);
3234 break;
3235 }
793dad94
VP
3236 pr_err("active ofld_connect_wr failure %d atid %d\n",
3237 req->retval, atid);
3238 mutex_lock(&dev->rdev.stats.lock);
3239 dev->rdev.stats.act_ofld_conn_fails++;
3240 mutex_unlock(&dev->rdev.stats.lock);
1cab775c 3241 connect_reply_upcall(ep, status2errno(req->retval));
793dad94
VP
3242 state_set(&ep->com, DEAD);
3243 remove_handle(dev, &dev->atid_idr, atid);
3244 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3245 dst_release(ep->dst);
3246 cxgb4_l2t_release(ep->l2t);
3247 c4iw_put_ep(&ep->com);
1cab775c
VP
3248}
3249
3250static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3251 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3252{
3253 struct sk_buff *rpl_skb;
3254 struct cpl_pass_accept_req *cpl;
3255 int ret;
3256
710a3110 3257 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
1cab775c
VP
3258 BUG_ON(!rpl_skb);
3259 if (req->retval) {
3260 PDBG("%s passive open failure %d\n", __func__, req->retval);
793dad94
VP
3261 mutex_lock(&dev->rdev.stats.lock);
3262 dev->rdev.stats.pas_ofld_conn_fails++;
3263 mutex_unlock(&dev->rdev.stats.lock);
1cab775c
VP
3264 kfree_skb(rpl_skb);
3265 } else {
3266 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3267 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
ef5d6355
VP
3268 (__force u32) htonl(
3269 (__force u32) req->tid)));
1cab775c
VP
3270 ret = pass_accept_req(dev, rpl_skb);
3271 if (!ret)
3272 kfree_skb(rpl_skb);
3273 }
3274 return;
3275}
3276
3277static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2f5b48c3
SW
3278{
3279 struct cpl_fw6_msg *rpl = cplhdr(skb);
1cab775c
VP
3280 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3281
3282 switch (rpl->type) {
3283 case FW6_TYPE_CQE:
3284 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3285 break;
3286 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3287 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3288 switch (req->t_state) {
3289 case TCP_SYN_SENT:
3290 active_ofld_conn_reply(dev, skb, req);
3291 break;
3292 case TCP_SYN_RECV:
3293 passive_ofld_conn_reply(dev, skb, req);
3294 break;
3295 default:
3296 pr_err("%s unexpected ofld conn wr state %d\n",
3297 __func__, req->t_state);
3298 break;
3299 }
3300 break;
3301 }
3302 return 0;
3303}
3304
3305static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3306{
3307 u32 l2info;
f079af7a 3308 u16 vlantag, len, hdr_len, eth_hdr_len;
1cab775c
VP
3309 u8 intf;
3310 struct cpl_rx_pkt *cpl = cplhdr(skb);
3311 struct cpl_pass_accept_req *req;
3312 struct tcp_options_received tmp_opt;
f079af7a 3313 struct c4iw_dev *dev;
1cab775c 3314
f079af7a 3315 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
1cab775c 3316 /* Store values from cpl_rx_pkt in temporary location. */
ef5d6355
VP
3317 vlantag = (__force u16) cpl->vlan;
3318 len = (__force u16) cpl->len;
3319 l2info = (__force u32) cpl->l2info;
3320 hdr_len = (__force u16) cpl->hdr_len;
1cab775c
VP
3321 intf = cpl->iff;
3322
3323 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3324
3325 /*
3326 * We need to parse the TCP options from SYN packet.
3327 * to generate cpl_pass_accept_req.
3328 */
3329 memset(&tmp_opt, 0, sizeof(tmp_opt));
3330 tcp_clear_options(&tmp_opt);
1a2c6181 3331 tcp_parse_options(skb, &tmp_opt, 0, NULL);
1cab775c
VP
3332
3333 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3334 memset(req, 0, sizeof(*req));
3335 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
ef5d6355
VP
3336 V_SYN_MAC_IDX(G_RX_MACIDX(
3337 (__force int) htonl(l2info))) |
1cab775c 3338 F_SYN_XACT_MATCH);
f079af7a
VP
3339 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3340 G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3341 G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
ef5d6355
VP
3342 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3343 (__force int) htonl(l2info))) |
3344 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3345 (__force int) htons(hdr_len))) |
3346 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3347 (__force int) htons(hdr_len))) |
f079af7a 3348 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
ef5d6355
VP
3349 req->vlan = (__force __be16) vlantag;
3350 req->len = (__force __be16) len;
1cab775c
VP
3351 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3352 PASS_OPEN_TOS(tos));
3353 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3354 if (tmp_opt.wscale_ok)
3355 req->tcpopt.wsf = tmp_opt.snd_wscale;
3356 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3357 if (tmp_opt.sack_ok)
3358 req->tcpopt.sack = 1;
3359 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3360 return;
3361}
3362
3363static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3364 __be32 laddr, __be16 lport,
3365 __be32 raddr, __be16 rport,
3366 u32 rcv_isn, u32 filter, u16 window,
3367 u32 rss_qid, u8 port_id)
3368{
3369 struct sk_buff *req_skb;
3370 struct fw_ofld_connection_wr *req;
3371 struct cpl_pass_accept_req *cpl = cplhdr(skb);
1ce1d471 3372 int ret;
1cab775c
VP
3373
3374 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3375 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3376 memset(req, 0, sizeof(*req));
3377 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3378 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3379 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
ef5d6355 3380 req->le.filter = (__force __be32) filter;
1cab775c
VP
3381 req->le.lport = lport;
3382 req->le.pport = rport;
3383 req->le.u.ipv4.lip = laddr;
3384 req->le.u.ipv4.pip = raddr;
3385 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3386 req->tcb.rcv_adv = htons(window);
3387 req->tcb.t_state_to_astid =
3388 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3389 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3390 V_FW_OFLD_CONNECTION_WR_ASTID(
3391 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3392
3393 /*
3394 * We store the qid in opt2 which will be used by the firmware
3395 * to send us the wr response.
3396 */
3397 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3398
3399 /*
3400 * We initialize the MSS index in TCB to 0xF.
3401 * So that when driver sends cpl_pass_accept_rpl
3402 * TCB picks up the correct value. If this was 0
3403 * TP will ignore any value > 0 for MSS index.
3404 */
3405 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
710a3110 3406 req->cookie = (unsigned long)skb;
1cab775c
VP
3407
3408 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
1ce1d471
SW
3409 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3410 if (ret < 0) {
3411 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3412 ret);
3413 kfree_skb(skb);
3414 kfree_skb(req_skb);
3415 }
1cab775c
VP
3416}
3417
3418/*
3419 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3420 * messages when a filter is being used instead of server to
3421 * redirect a syn packet. When packets hit filter they are redirected
3422 * to the offload queue and driver tries to establish the connection
3423 * using firmware work request.
3424 */
3425static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3426{
3427 int stid;
3428 unsigned int filter;
3429 struct ethhdr *eh = NULL;
3430 struct vlan_ethhdr *vlan_eh = NULL;
3431 struct iphdr *iph;
3432 struct tcphdr *tcph;
3433 struct rss_header *rss = (void *)skb->data;
3434 struct cpl_rx_pkt *cpl = (void *)skb->data;
3435 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3436 struct l2t_entry *e;
3437 struct dst_entry *dst;
1cab775c
VP
3438 struct c4iw_ep *lep;
3439 u16 window;
3440 struct port_info *pi;
3441 struct net_device *pdev;
f079af7a 3442 u16 rss_qid, eth_hdr_len;
1cab775c
VP
3443 int step;
3444 u32 tx_chan;
3445 struct neighbour *neigh;
3446
3447 /* Drop all non-SYN packets */
3448 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3449 goto reject;
3450
3451 /*
3452 * Drop all packets which did not hit the filter.
3453 * Unlikely to happen.
3454 */
3455 if (!(rss->filter_hit && rss->filter_tid))
3456 goto reject;
3457
3458 /*
3459 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3460 */
a4ea025f 3461 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
1cab775c
VP
3462
3463 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3464 if (!lep) {
3465 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3466 goto reject;
3467 }
3468
f079af7a
VP
3469 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3470 G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3471 G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3472 if (eth_hdr_len == ETH_HLEN) {
1cab775c
VP
3473 eh = (struct ethhdr *)(req + 1);
3474 iph = (struct iphdr *)(eh + 1);
3475 } else {
3476 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3477 iph = (struct iphdr *)(vlan_eh + 1);
3478 skb->vlan_tci = ntohs(cpl->vlan);
3479 }
3480
3481 if (iph->version != 0x4)
3482 goto reject;
3483
3484 tcph = (struct tcphdr *)(iph + 1);
3485 skb_set_network_header(skb, (void *)iph - (void *)rss);
3486 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3487 skb_get(skb);
3488
3489 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3490 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3491 ntohs(tcph->source), iph->tos);
3492
830662f6
VP
3493 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3494 iph->tos);
3495 if (!dst) {
1cab775c
VP
3496 pr_err("%s - failed to find dst entry!\n",
3497 __func__);
3498 goto reject;
3499 }
1cab775c
VP
3500 neigh = dst_neigh_lookup_skb(dst, skb);
3501
aaa0c23c
ZZ
3502 if (!neigh) {
3503 pr_err("%s - failed to allocate neigh!\n",
3504 __func__);
3505 goto free_dst;
3506 }
3507
1cab775c
VP
3508 if (neigh->dev->flags & IFF_LOOPBACK) {
3509 pdev = ip_dev_find(&init_net, iph->daddr);
3510 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3511 pdev, 0);
3512 pi = (struct port_info *)netdev_priv(pdev);
3513 tx_chan = cxgb4_port_chan(pdev);
3514 dev_put(pdev);
3515 } else {
830662f6 3516 pdev = get_real_dev(neigh->dev);
1cab775c 3517 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
830662f6
VP
3518 pdev, 0);
3519 pi = (struct port_info *)netdev_priv(pdev);
3520 tx_chan = cxgb4_port_chan(pdev);
1cab775c 3521 }
ebf00060 3522 neigh_release(neigh);
1cab775c
VP
3523 if (!e) {
3524 pr_err("%s - failed to allocate l2t entry!\n",
3525 __func__);
3526 goto free_dst;
3527 }
3528
3529 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3530 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
ef5d6355 3531 window = (__force u16) htons((__force u16)tcph->window);
1cab775c
VP
3532
3533 /* Calcuate filter portion for LE region. */
41b4f86c
KS
3534 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3535 dev->rdev.lldi.ports[0],
3536 e));
1cab775c
VP
3537
3538 /*
3539 * Synthesize the cpl_pass_accept_req. We have everything except the
3540 * TID. Once firmware sends a reply with TID we update the TID field
3541 * in cpl and pass it through the regular cpl_pass_accept_req path.
3542 */
3543 build_cpl_pass_accept_req(skb, stid, iph->tos);
3544 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3545 tcph->source, ntohl(tcph->seq), filter, window,
3546 rss_qid, pi->port_id);
3547 cxgb4_l2t_release(e);
3548free_dst:
3549 dst_release(dst);
3550reject:
2f5b48c3
SW
3551 return 0;
3552}
3553
be4c9bad
RD
3554/*
3555 * These are the real handlers that are called from a
3556 * work queue.
3557 */
3558static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3559 [CPL_ACT_ESTABLISH] = act_establish,
3560 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3561 [CPL_RX_DATA] = rx_data,
3562 [CPL_ABORT_RPL_RSS] = abort_rpl,
3563 [CPL_ABORT_RPL] = abort_rpl,
3564 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3565 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3566 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3567 [CPL_PASS_ESTABLISH] = pass_establish,
3568 [CPL_PEER_CLOSE] = peer_close,
3569 [CPL_ABORT_REQ_RSS] = peer_abort,
3570 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3571 [CPL_RDMA_TERMINATE] = terminate,
2f5b48c3 3572 [CPL_FW4_ACK] = fw4_ack,
1cab775c
VP
3573 [CPL_FW6_MSG] = deferred_fw6_msg,
3574 [CPL_RX_PKT] = rx_pkt
be4c9bad
RD
3575};
3576
3577static void process_timeout(struct c4iw_ep *ep)
3578{
3579 struct c4iw_qp_attributes attrs;
3580 int abort = 1;
3581
2f5b48c3 3582 mutex_lock(&ep->com.mutex);
be4c9bad
RD
3583 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3584 ep->com.state);
793dad94 3585 set_bit(TIMEDOUT, &ep->com.history);
be4c9bad
RD
3586 switch (ep->com.state) {
3587 case MPA_REQ_SENT:
3588 __state_set(&ep->com, ABORTING);
3589 connect_reply_upcall(ep, -ETIMEDOUT);
3590 break;
3591 case MPA_REQ_WAIT:
3592 __state_set(&ep->com, ABORTING);
3593 break;
3594 case CLOSING:
3595 case MORIBUND:
3596 if (ep->com.cm_id && ep->com.qp) {
3597 attrs.next_state = C4IW_QP_STATE_ERROR;
3598 c4iw_modify_qp(ep->com.qp->rhp,
3599 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3600 &attrs, 1);
3601 }
3602 __state_set(&ep->com, ABORTING);
be13b2df 3603 close_complete_upcall(ep, -ETIMEDOUT);
be4c9bad 3604 break;
b33bd0cb
SW
3605 case ABORTING:
3606 case DEAD:
3607
3608 /*
3609 * These states are expected if the ep timed out at the same
3610 * time as another thread was calling stop_ep_timer().
3611 * So we silently do nothing for these states.
3612 */
3613 abort = 0;
3614 break;
be4c9bad 3615 default:
76f267b7 3616 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
be4c9bad 3617 __func__, ep, ep->hwtid, ep->com.state);
be4c9bad
RD
3618 abort = 0;
3619 }
be4c9bad
RD
3620 if (abort)
3621 abort_connection(ep, NULL, GFP_KERNEL);
cc18b939 3622 mutex_unlock(&ep->com.mutex);
be4c9bad
RD
3623 c4iw_put_ep(&ep->com);
3624}
3625
3626static void process_timedout_eps(void)
3627{
3628 struct c4iw_ep *ep;
3629
3630 spin_lock_irq(&timeout_lock);
3631 while (!list_empty(&timeout_list)) {
3632 struct list_head *tmp;
3633
3634 tmp = timeout_list.next;
3635 list_del(tmp);
b33bd0cb
SW
3636 tmp->next = NULL;
3637 tmp->prev = NULL;
be4c9bad
RD
3638 spin_unlock_irq(&timeout_lock);
3639 ep = list_entry(tmp, struct c4iw_ep, entry);
3640 process_timeout(ep);
3641 spin_lock_irq(&timeout_lock);
3642 }
3643 spin_unlock_irq(&timeout_lock);
3644}
3645
3646static void process_work(struct work_struct *work)
3647{
3648 struct sk_buff *skb = NULL;
3649 struct c4iw_dev *dev;
c1d7356c 3650 struct cpl_act_establish *rpl;
be4c9bad
RD
3651 unsigned int opcode;
3652 int ret;
3653
b33bd0cb 3654 process_timedout_eps();
be4c9bad
RD
3655 while ((skb = skb_dequeue(&rxq))) {
3656 rpl = cplhdr(skb);
3657 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3658 opcode = rpl->ot.opcode;
3659
3660 BUG_ON(!work_handlers[opcode]);
3661 ret = work_handlers[opcode](dev, skb);
3662 if (!ret)
3663 kfree_skb(skb);
b33bd0cb 3664 process_timedout_eps();
be4c9bad 3665 }
be4c9bad
RD
3666}
3667
3668static DECLARE_WORK(skb_work, process_work);
3669
3670static void ep_timeout(unsigned long arg)
3671{
3672 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
1ec779cc 3673 int kickit = 0;
be4c9bad
RD
3674
3675 spin_lock(&timeout_lock);
1ec779cc 3676 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
b33bd0cb
SW
3677 /*
3678 * Only insert if it is not already on the list.
3679 */
3680 if (!ep->entry.next) {
3681 list_add_tail(&ep->entry, &timeout_list);
3682 kickit = 1;
3683 }
1ec779cc 3684 }
be4c9bad 3685 spin_unlock(&timeout_lock);
1ec779cc
VP
3686 if (kickit)
3687 queue_work(workq, &skb_work);
be4c9bad
RD
3688}
3689
cfdda9d7
SW
3690/*
3691 * All the CM events are handled on a work queue to have a safe context.
3692 */
3693static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3694{
3695
3696 /*
3697 * Save dev in the skb->cb area.
3698 */
3699 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3700
3701 /*
3702 * Queue the skb and schedule the worker thread.
3703 */
3704 skb_queue_tail(&rxq, skb);
3705 queue_work(workq, &skb_work);
3706 return 0;
3707}
3708
3709static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3710{
3711 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3712
3713 if (rpl->status != CPL_ERR_NONE) {
3714 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3715 "for tid %u\n", rpl->status, GET_TID(rpl));
3716 }
2f5b48c3 3717 kfree_skb(skb);
cfdda9d7
SW
3718 return 0;
3719}
3720
be4c9bad
RD
3721static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3722{
3723 struct cpl_fw6_msg *rpl = cplhdr(skb);
3724 struct c4iw_wr_wait *wr_waitp;
3725 int ret;
3726
3727 PDBG("%s type %u\n", __func__, rpl->type);
3728
3729 switch (rpl->type) {
5be78ee9 3730 case FW6_TYPE_WR_RPL:
be4c9bad 3731 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
c8e081a1 3732 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
be4c9bad 3733 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
d9594d99
SW
3734 if (wr_waitp)
3735 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2f5b48c3 3736 kfree_skb(skb);
be4c9bad 3737 break;
5be78ee9 3738 case FW6_TYPE_CQE:
5be78ee9 3739 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
1cab775c 3740 sched(dev, skb);
5be78ee9 3741 break;
be4c9bad
RD
3742 default:
3743 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3744 rpl->type);
2f5b48c3 3745 kfree_skb(skb);
be4c9bad
RD
3746 break;
3747 }
3748 return 0;
3749}
3750
8da7e7a5
SW
3751static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3752{
3753 struct cpl_abort_req_rss *req = cplhdr(skb);
3754 struct c4iw_ep *ep;
3755 struct tid_info *t = dev->rdev.lldi.tids;
3756 unsigned int tid = GET_TID(req);
3757
3758 ep = lookup_tid(t, tid);
14b92228
SW
3759 if (!ep) {
3760 printk(KERN_WARNING MOD
3761 "Abort on non-existent endpoint, tid %d\n", tid);
3762 kfree_skb(skb);
3763 return 0;
3764 }
7a2cea2a 3765 if (is_neg_adv(req->status)) {
8da7e7a5
SW
3766 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3767 ep->hwtid);
3768 kfree_skb(skb);
3769 return 0;
3770 }
3771 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3772 ep->com.state);
3773
3774 /*
3775 * Wake up any threads in rdma_init() or rdma_fini().
7c0a33d6
VP
3776 * However, if we are on MPAv2 and want to retry with MPAv1
3777 * then, don't wake up yet.
8da7e7a5 3778 */
7c0a33d6
VP
3779 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3780 if (ep->com.state != MPA_REQ_SENT)
3781 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3782 } else
3783 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
8da7e7a5
SW
3784 sched(dev, skb);
3785 return 0;
3786}
3787
be4c9bad
RD
3788/*
3789 * Most upcalls from the T4 Core go to sched() to
3790 * schedule the processing on a work queue.
3791 */
3792c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3793 [CPL_ACT_ESTABLISH] = sched,
3794 [CPL_ACT_OPEN_RPL] = sched,
3795 [CPL_RX_DATA] = sched,
3796 [CPL_ABORT_RPL_RSS] = sched,
3797 [CPL_ABORT_RPL] = sched,
3798 [CPL_PASS_OPEN_RPL] = sched,
3799 [CPL_CLOSE_LISTSRV_RPL] = sched,
3800 [CPL_PASS_ACCEPT_REQ] = sched,
3801 [CPL_PASS_ESTABLISH] = sched,
3802 [CPL_PEER_CLOSE] = sched,
3803 [CPL_CLOSE_CON_RPL] = sched,
8da7e7a5 3804 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
be4c9bad
RD
3805 [CPL_RDMA_TERMINATE] = sched,
3806 [CPL_FW4_ACK] = sched,
3807 [CPL_SET_TCB_RPL] = set_tcb_rpl,
1cab775c
VP
3808 [CPL_FW6_MSG] = fw6_msg,
3809 [CPL_RX_PKT] = sched
be4c9bad
RD
3810};
3811
cfdda9d7
SW
3812int __init c4iw_cm_init(void)
3813{
be4c9bad 3814 spin_lock_init(&timeout_lock);
cfdda9d7
SW
3815 skb_queue_head_init(&rxq);
3816
3817 workq = create_singlethread_workqueue("iw_cxgb4");
3818 if (!workq)
3819 return -ENOMEM;
3820
cfdda9d7
SW
3821 return 0;
3822}
3823
3824void __exit c4iw_cm_term(void)
3825{
be4c9bad 3826 WARN_ON(!list_empty(&timeout_list));
cfdda9d7
SW
3827 flush_workqueue(workq);
3828 destroy_workqueue(workq);
3829}