Merge branch 'upstream-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/linvil...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / infiniband / hw / cxgb3 / iwch_cm.c
1 /*
2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38
39 #include <net/neighbour.h>
40 #include <net/netevent.h>
41 #include <net/route.h>
42
43 #include "tcb.h"
44 #include "cxgb3_offload.h"
45 #include "iwch.h"
46 #include "iwch_provider.h"
47 #include "iwch_cm.h"
48
49 static char *states[] = {
50 "idle",
51 "listen",
52 "connecting",
53 "mpa_wait_req",
54 "mpa_req_sent",
55 "mpa_req_rcvd",
56 "mpa_rep_sent",
57 "fpdu_mode",
58 "aborting",
59 "closing",
60 "moribund",
61 "dead",
62 NULL,
63 };
64
65 static int ep_timeout_secs = 10;
66 module_param(ep_timeout_secs, int, 0444);
67 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
68 "in seconds (default=10)");
69
70 static int mpa_rev = 1;
71 module_param(mpa_rev, int, 0444);
72 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
73 "1 is spec compliant. (default=1)");
74
75 static int markers_enabled = 0;
76 module_param(markers_enabled, int, 0444);
77 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
78
79 static int crc_enabled = 1;
80 module_param(crc_enabled, int, 0444);
81 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
82
83 static int rcv_win = 256 * 1024;
84 module_param(rcv_win, int, 0444);
85 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
86
87 static int snd_win = 32 * 1024;
88 module_param(snd_win, int, 0444);
89 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
90
91 static unsigned int nocong = 0;
92 module_param(nocong, uint, 0444);
93 MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
94
95 static unsigned int cong_flavor = 1;
96 module_param(cong_flavor, uint, 0444);
97 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
98
99 static void process_work(struct work_struct *work);
100 static struct workqueue_struct *workq;
101 static DECLARE_WORK(skb_work, process_work);
102
103 static struct sk_buff_head rxq;
104 static cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS];
105
106 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
107 static void ep_timeout(unsigned long arg);
108 static void connect_reply_upcall(struct iwch_ep *ep, int status);
109
110 static void start_ep_timer(struct iwch_ep *ep)
111 {
112 PDBG("%s ep %p\n", __FUNCTION__, ep);
113 if (timer_pending(&ep->timer)) {
114 PDBG("%s stopped / restarted timer ep %p\n", __FUNCTION__, ep);
115 del_timer_sync(&ep->timer);
116 } else
117 get_ep(&ep->com);
118 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
119 ep->timer.data = (unsigned long)ep;
120 ep->timer.function = ep_timeout;
121 add_timer(&ep->timer);
122 }
123
124 static void stop_ep_timer(struct iwch_ep *ep)
125 {
126 PDBG("%s ep %p\n", __FUNCTION__, ep);
127 del_timer_sync(&ep->timer);
128 put_ep(&ep->com);
129 }
130
131 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
132 {
133 struct cpl_tid_release *req;
134
135 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
136 if (!skb)
137 return;
138 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
139 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
140 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
141 skb->priority = CPL_PRIORITY_SETUP;
142 tdev->send(tdev, skb);
143 return;
144 }
145
146 int iwch_quiesce_tid(struct iwch_ep *ep)
147 {
148 struct cpl_set_tcb_field *req;
149 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
150
151 if (!skb)
152 return -ENOMEM;
153 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
154 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
155 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
156 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
157 req->reply = 0;
158 req->cpu_idx = 0;
159 req->word = htons(W_TCB_RX_QUIESCE);
160 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
161 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
162
163 skb->priority = CPL_PRIORITY_DATA;
164 ep->com.tdev->send(ep->com.tdev, skb);
165 return 0;
166 }
167
168 int iwch_resume_tid(struct iwch_ep *ep)
169 {
170 struct cpl_set_tcb_field *req;
171 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
172
173 if (!skb)
174 return -ENOMEM;
175 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
176 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
177 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
178 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
179 req->reply = 0;
180 req->cpu_idx = 0;
181 req->word = htons(W_TCB_RX_QUIESCE);
182 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
183 req->val = 0;
184
185 skb->priority = CPL_PRIORITY_DATA;
186 ep->com.tdev->send(ep->com.tdev, skb);
187 return 0;
188 }
189
190 static void set_emss(struct iwch_ep *ep, u16 opt)
191 {
192 PDBG("%s ep %p opt %u\n", __FUNCTION__, ep, opt);
193 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
194 if (G_TCPOPT_TSTAMP(opt))
195 ep->emss -= 12;
196 if (ep->emss < 128)
197 ep->emss = 128;
198 PDBG("emss=%d\n", ep->emss);
199 }
200
201 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
202 {
203 unsigned long flags;
204 enum iwch_ep_state state;
205
206 spin_lock_irqsave(&epc->lock, flags);
207 state = epc->state;
208 spin_unlock_irqrestore(&epc->lock, flags);
209 return state;
210 }
211
212 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
213 {
214 epc->state = new;
215 }
216
217 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
218 {
219 unsigned long flags;
220
221 spin_lock_irqsave(&epc->lock, flags);
222 PDBG("%s - %s -> %s\n", __FUNCTION__, states[epc->state], states[new]);
223 __state_set(epc, new);
224 spin_unlock_irqrestore(&epc->lock, flags);
225 return;
226 }
227
228 static void *alloc_ep(int size, gfp_t gfp)
229 {
230 struct iwch_ep_common *epc;
231
232 epc = kmalloc(size, gfp);
233 if (epc) {
234 memset(epc, 0, size);
235 kref_init(&epc->kref);
236 spin_lock_init(&epc->lock);
237 init_waitqueue_head(&epc->waitq);
238 }
239 PDBG("%s alloc ep %p\n", __FUNCTION__, epc);
240 return epc;
241 }
242
243 void __free_ep(struct kref *kref)
244 {
245 struct iwch_ep_common *epc;
246 epc = container_of(kref, struct iwch_ep_common, kref);
247 PDBG("%s ep %p state %s\n", __FUNCTION__, epc, states[state_read(epc)]);
248 kfree(epc);
249 }
250
251 static void release_ep_resources(struct iwch_ep *ep)
252 {
253 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
254 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
255 dst_release(ep->dst);
256 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
257 if (ep->com.tdev->type == T3B)
258 release_tid(ep->com.tdev, ep->hwtid, NULL);
259 put_ep(&ep->com);
260 }
261
262 static void process_work(struct work_struct *work)
263 {
264 struct sk_buff *skb = NULL;
265 void *ep;
266 struct t3cdev *tdev;
267 int ret;
268
269 while ((skb = skb_dequeue(&rxq))) {
270 ep = *((void **) (skb->cb));
271 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
272 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
273 if (ret & CPL_RET_BUF_DONE)
274 kfree_skb(skb);
275
276 /*
277 * ep was referenced in sched(), and is freed here.
278 */
279 put_ep((struct iwch_ep_common *)ep);
280 }
281 }
282
283 static int status2errno(int status)
284 {
285 switch (status) {
286 case CPL_ERR_NONE:
287 return 0;
288 case CPL_ERR_CONN_RESET:
289 return -ECONNRESET;
290 case CPL_ERR_ARP_MISS:
291 return -EHOSTUNREACH;
292 case CPL_ERR_CONN_TIMEDOUT:
293 return -ETIMEDOUT;
294 case CPL_ERR_TCAM_FULL:
295 return -ENOMEM;
296 case CPL_ERR_CONN_EXIST:
297 return -EADDRINUSE;
298 default:
299 return -EIO;
300 }
301 }
302
303 /*
304 * Try and reuse skbs already allocated...
305 */
306 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
307 {
308 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
309 skb_trim(skb, 0);
310 skb_get(skb);
311 } else {
312 skb = alloc_skb(len, gfp);
313 }
314 return skb;
315 }
316
317 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
318 __be32 peer_ip, __be16 local_port,
319 __be16 peer_port, u8 tos)
320 {
321 struct rtable *rt;
322 struct flowi fl = {
323 .oif = 0,
324 .nl_u = {
325 .ip4_u = {
326 .daddr = peer_ip,
327 .saddr = local_ip,
328 .tos = tos}
329 },
330 .proto = IPPROTO_TCP,
331 .uli_u = {
332 .ports = {
333 .sport = local_port,
334 .dport = peer_port}
335 }
336 };
337
338 if (ip_route_output_flow(&rt, &fl, NULL, 0))
339 return NULL;
340 return rt;
341 }
342
343 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
344 {
345 int i = 0;
346
347 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
348 ++i;
349 return i;
350 }
351
352 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
353 {
354 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
355 kfree_skb(skb);
356 }
357
358 /*
359 * Handle an ARP failure for an active open.
360 */
361 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
362 {
363 printk(KERN_ERR MOD "ARP failure duing connect\n");
364 kfree_skb(skb);
365 }
366
367 /*
368 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
369 * and send it along.
370 */
371 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
372 {
373 struct cpl_abort_req *req = cplhdr(skb);
374
375 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
376 req->cmd = CPL_ABORT_NO_RST;
377 cxgb3_ofld_send(dev, skb);
378 }
379
380 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
381 {
382 struct cpl_close_con_req *req;
383 struct sk_buff *skb;
384
385 PDBG("%s ep %p\n", __FUNCTION__, ep);
386 skb = get_skb(NULL, sizeof(*req), gfp);
387 if (!skb) {
388 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
389 return -ENOMEM;
390 }
391 skb->priority = CPL_PRIORITY_DATA;
392 set_arp_failure_handler(skb, arp_failure_discard);
393 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
394 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
395 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
396 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
397 l2t_send(ep->com.tdev, skb, ep->l2t);
398 return 0;
399 }
400
401 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
402 {
403 struct cpl_abort_req *req;
404
405 PDBG("%s ep %p\n", __FUNCTION__, ep);
406 skb = get_skb(skb, sizeof(*req), gfp);
407 if (!skb) {
408 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
409 __FUNCTION__);
410 return -ENOMEM;
411 }
412 skb->priority = CPL_PRIORITY_DATA;
413 set_arp_failure_handler(skb, abort_arp_failure);
414 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
415 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
416 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
417 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
418 req->cmd = CPL_ABORT_SEND_RST;
419 l2t_send(ep->com.tdev, skb, ep->l2t);
420 return 0;
421 }
422
423 static int send_connect(struct iwch_ep *ep)
424 {
425 struct cpl_act_open_req *req;
426 struct sk_buff *skb;
427 u32 opt0h, opt0l, opt2;
428 unsigned int mtu_idx;
429 int wscale;
430
431 PDBG("%s ep %p\n", __FUNCTION__, ep);
432
433 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
434 if (!skb) {
435 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
436 __FUNCTION__);
437 return -ENOMEM;
438 }
439 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
440 wscale = compute_wscale(rcv_win);
441 opt0h = V_NAGLE(0) |
442 V_NO_CONG(nocong) |
443 V_KEEP_ALIVE(1) |
444 F_TCAM_BYPASS |
445 V_WND_SCALE(wscale) |
446 V_MSS_IDX(mtu_idx) |
447 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
448 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
449 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
450 skb->priority = CPL_PRIORITY_SETUP;
451 set_arp_failure_handler(skb, act_open_req_arp_failure);
452
453 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
454 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
455 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
456 req->local_port = ep->com.local_addr.sin_port;
457 req->peer_port = ep->com.remote_addr.sin_port;
458 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
459 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
460 req->opt0h = htonl(opt0h);
461 req->opt0l = htonl(opt0l);
462 req->params = 0;
463 req->opt2 = htonl(opt2);
464 l2t_send(ep->com.tdev, skb, ep->l2t);
465 return 0;
466 }
467
468 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
469 {
470 int mpalen;
471 struct tx_data_wr *req;
472 struct mpa_message *mpa;
473 int len;
474
475 PDBG("%s ep %p pd_len %d\n", __FUNCTION__, ep, ep->plen);
476
477 BUG_ON(skb_cloned(skb));
478
479 mpalen = sizeof(*mpa) + ep->plen;
480 if (skb->data + mpalen + sizeof(*req) > skb->end) {
481 kfree_skb(skb);
482 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
483 if (!skb) {
484 connect_reply_upcall(ep, -ENOMEM);
485 return;
486 }
487 }
488 skb_trim(skb, 0);
489 skb_reserve(skb, sizeof(*req));
490 skb_put(skb, mpalen);
491 skb->priority = CPL_PRIORITY_DATA;
492 mpa = (struct mpa_message *) skb->data;
493 memset(mpa, 0, sizeof(*mpa));
494 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
495 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
496 (markers_enabled ? MPA_MARKERS : 0);
497 mpa->private_data_size = htons(ep->plen);
498 mpa->revision = mpa_rev;
499
500 if (ep->plen)
501 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
502
503 /*
504 * Reference the mpa skb. This ensures the data area
505 * will remain in memory until the hw acks the tx.
506 * Function tx_ack() will deref it.
507 */
508 skb_get(skb);
509 set_arp_failure_handler(skb, arp_failure_discard);
510 skb->h.raw = skb->data;
511 len = skb->len;
512 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
513 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
514 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
515 req->len = htonl(len);
516 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
517 V_TX_SNDBUF(snd_win>>15));
518 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
519 req->sndseq = htonl(ep->snd_seq);
520 BUG_ON(ep->mpa_skb);
521 ep->mpa_skb = skb;
522 l2t_send(ep->com.tdev, skb, ep->l2t);
523 start_ep_timer(ep);
524 state_set(&ep->com, MPA_REQ_SENT);
525 return;
526 }
527
528 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
529 {
530 int mpalen;
531 struct tx_data_wr *req;
532 struct mpa_message *mpa;
533 struct sk_buff *skb;
534
535 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
536
537 mpalen = sizeof(*mpa) + plen;
538
539 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
540 if (!skb) {
541 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
542 return -ENOMEM;
543 }
544 skb_reserve(skb, sizeof(*req));
545 mpa = (struct mpa_message *) skb_put(skb, mpalen);
546 memset(mpa, 0, sizeof(*mpa));
547 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
548 mpa->flags = MPA_REJECT;
549 mpa->revision = mpa_rev;
550 mpa->private_data_size = htons(plen);
551 if (plen)
552 memcpy(mpa->private_data, pdata, plen);
553
554 /*
555 * Reference the mpa skb again. This ensures the data area
556 * will remain in memory until the hw acks the tx.
557 * Function tx_ack() will deref it.
558 */
559 skb_get(skb);
560 skb->priority = CPL_PRIORITY_DATA;
561 set_arp_failure_handler(skb, arp_failure_discard);
562 skb->h.raw = skb->data;
563 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
564 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
565 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
566 req->len = htonl(mpalen);
567 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
568 V_TX_SNDBUF(snd_win>>15));
569 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
570 req->sndseq = htonl(ep->snd_seq);
571 BUG_ON(ep->mpa_skb);
572 ep->mpa_skb = skb;
573 l2t_send(ep->com.tdev, skb, ep->l2t);
574 return 0;
575 }
576
577 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
578 {
579 int mpalen;
580 struct tx_data_wr *req;
581 struct mpa_message *mpa;
582 int len;
583 struct sk_buff *skb;
584
585 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
586
587 mpalen = sizeof(*mpa) + plen;
588
589 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
590 if (!skb) {
591 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
592 return -ENOMEM;
593 }
594 skb->priority = CPL_PRIORITY_DATA;
595 skb_reserve(skb, sizeof(*req));
596 mpa = (struct mpa_message *) skb_put(skb, mpalen);
597 memset(mpa, 0, sizeof(*mpa));
598 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
599 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
600 (markers_enabled ? MPA_MARKERS : 0);
601 mpa->revision = mpa_rev;
602 mpa->private_data_size = htons(plen);
603 if (plen)
604 memcpy(mpa->private_data, pdata, plen);
605
606 /*
607 * Reference the mpa skb. This ensures the data area
608 * will remain in memory until the hw acks the tx.
609 * Function tx_ack() will deref it.
610 */
611 skb_get(skb);
612 set_arp_failure_handler(skb, arp_failure_discard);
613 skb->h.raw = skb->data;
614 len = skb->len;
615 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
616 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
617 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
618 req->len = htonl(len);
619 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
620 V_TX_SNDBUF(snd_win>>15));
621 req->flags = htonl(F_TX_MORE | F_TX_IMM_ACK | F_TX_INIT);
622 req->sndseq = htonl(ep->snd_seq);
623 ep->mpa_skb = skb;
624 state_set(&ep->com, MPA_REP_SENT);
625 l2t_send(ep->com.tdev, skb, ep->l2t);
626 return 0;
627 }
628
629 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
630 {
631 struct iwch_ep *ep = ctx;
632 struct cpl_act_establish *req = cplhdr(skb);
633 unsigned int tid = GET_TID(req);
634
635 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, tid);
636
637 dst_confirm(ep->dst);
638
639 /* setup the hwtid for this connection */
640 ep->hwtid = tid;
641 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
642
643 ep->snd_seq = ntohl(req->snd_isn);
644
645 set_emss(ep, ntohs(req->tcp_opt));
646
647 /* dealloc the atid */
648 cxgb3_free_atid(ep->com.tdev, ep->atid);
649
650 /* start MPA negotiation */
651 send_mpa_req(ep, skb);
652
653 return 0;
654 }
655
656 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
657 {
658 PDBG("%s ep %p\n", __FILE__, ep);
659 state_set(&ep->com, ABORTING);
660 send_abort(ep, skb, gfp);
661 }
662
663 static void close_complete_upcall(struct iwch_ep *ep)
664 {
665 struct iw_cm_event event;
666
667 PDBG("%s ep %p\n", __FUNCTION__, ep);
668 memset(&event, 0, sizeof(event));
669 event.event = IW_CM_EVENT_CLOSE;
670 if (ep->com.cm_id) {
671 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
672 ep, ep->com.cm_id, ep->hwtid);
673 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
674 ep->com.cm_id->rem_ref(ep->com.cm_id);
675 ep->com.cm_id = NULL;
676 ep->com.qp = NULL;
677 }
678 }
679
680 static void peer_close_upcall(struct iwch_ep *ep)
681 {
682 struct iw_cm_event event;
683
684 PDBG("%s ep %p\n", __FUNCTION__, ep);
685 memset(&event, 0, sizeof(event));
686 event.event = IW_CM_EVENT_DISCONNECT;
687 if (ep->com.cm_id) {
688 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
689 ep, ep->com.cm_id, ep->hwtid);
690 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
691 }
692 }
693
694 static void peer_abort_upcall(struct iwch_ep *ep)
695 {
696 struct iw_cm_event event;
697
698 PDBG("%s ep %p\n", __FUNCTION__, ep);
699 memset(&event, 0, sizeof(event));
700 event.event = IW_CM_EVENT_CLOSE;
701 event.status = -ECONNRESET;
702 if (ep->com.cm_id) {
703 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
704 ep->com.cm_id, ep->hwtid);
705 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
706 ep->com.cm_id->rem_ref(ep->com.cm_id);
707 ep->com.cm_id = NULL;
708 ep->com.qp = NULL;
709 }
710 }
711
712 static void connect_reply_upcall(struct iwch_ep *ep, int status)
713 {
714 struct iw_cm_event event;
715
716 PDBG("%s ep %p status %d\n", __FUNCTION__, ep, status);
717 memset(&event, 0, sizeof(event));
718 event.event = IW_CM_EVENT_CONNECT_REPLY;
719 event.status = status;
720 event.local_addr = ep->com.local_addr;
721 event.remote_addr = ep->com.remote_addr;
722
723 if ((status == 0) || (status == -ECONNREFUSED)) {
724 event.private_data_len = ep->plen;
725 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
726 }
727 if (ep->com.cm_id) {
728 PDBG("%s ep %p tid %d status %d\n", __FUNCTION__, ep,
729 ep->hwtid, status);
730 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
731 }
732 if (status < 0) {
733 ep->com.cm_id->rem_ref(ep->com.cm_id);
734 ep->com.cm_id = NULL;
735 ep->com.qp = NULL;
736 }
737 }
738
739 static void connect_request_upcall(struct iwch_ep *ep)
740 {
741 struct iw_cm_event event;
742
743 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
744 memset(&event, 0, sizeof(event));
745 event.event = IW_CM_EVENT_CONNECT_REQUEST;
746 event.local_addr = ep->com.local_addr;
747 event.remote_addr = ep->com.remote_addr;
748 event.private_data_len = ep->plen;
749 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
750 event.provider_data = ep;
751 if (state_read(&ep->parent_ep->com) != DEAD)
752 ep->parent_ep->com.cm_id->event_handler(
753 ep->parent_ep->com.cm_id,
754 &event);
755 put_ep(&ep->parent_ep->com);
756 ep->parent_ep = NULL;
757 }
758
759 static void established_upcall(struct iwch_ep *ep)
760 {
761 struct iw_cm_event event;
762
763 PDBG("%s ep %p\n", __FUNCTION__, ep);
764 memset(&event, 0, sizeof(event));
765 event.event = IW_CM_EVENT_ESTABLISHED;
766 if (ep->com.cm_id) {
767 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
768 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
769 }
770 }
771
772 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
773 {
774 struct cpl_rx_data_ack *req;
775 struct sk_buff *skb;
776
777 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
778 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
779 if (!skb) {
780 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
781 return 0;
782 }
783
784 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
785 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
786 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
787 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
788 skb->priority = CPL_PRIORITY_ACK;
789 ep->com.tdev->send(ep->com.tdev, skb);
790 return credits;
791 }
792
793 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
794 {
795 struct mpa_message *mpa;
796 u16 plen;
797 struct iwch_qp_attributes attrs;
798 enum iwch_qp_attr_mask mask;
799 int err;
800
801 PDBG("%s ep %p\n", __FUNCTION__, ep);
802
803 /*
804 * Stop mpa timer. If it expired, then the state has
805 * changed and we bail since ep_timeout already aborted
806 * the connection.
807 */
808 stop_ep_timer(ep);
809 if (state_read(&ep->com) != MPA_REQ_SENT)
810 return;
811
812 /*
813 * If we get more than the supported amount of private data
814 * then we must fail this connection.
815 */
816 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
817 err = -EINVAL;
818 goto err;
819 }
820
821 /*
822 * copy the new data into our accumulation buffer.
823 */
824 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
825 ep->mpa_pkt_len += skb->len;
826
827 /*
828 * if we don't even have the mpa message, then bail.
829 */
830 if (ep->mpa_pkt_len < sizeof(*mpa))
831 return;
832 mpa = (struct mpa_message *) ep->mpa_pkt;
833
834 /* Validate MPA header. */
835 if (mpa->revision != mpa_rev) {
836 err = -EPROTO;
837 goto err;
838 }
839 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
840 err = -EPROTO;
841 goto err;
842 }
843
844 plen = ntohs(mpa->private_data_size);
845
846 /*
847 * Fail if there's too much private data.
848 */
849 if (plen > MPA_MAX_PRIVATE_DATA) {
850 err = -EPROTO;
851 goto err;
852 }
853
854 /*
855 * If plen does not account for pkt size
856 */
857 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
858 err = -EPROTO;
859 goto err;
860 }
861
862 ep->plen = (u8) plen;
863
864 /*
865 * If we don't have all the pdata yet, then bail.
866 * We'll continue process when more data arrives.
867 */
868 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
869 return;
870
871 if (mpa->flags & MPA_REJECT) {
872 err = -ECONNREFUSED;
873 goto err;
874 }
875
876 /*
877 * If we get here we have accumulated the entire mpa
878 * start reply message including private data. And
879 * the MPA header is valid.
880 */
881 state_set(&ep->com, FPDU_MODE);
882 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
883 ep->mpa_attr.recv_marker_enabled = markers_enabled;
884 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
885 ep->mpa_attr.version = mpa_rev;
886 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
887 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
888 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
889 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
890
891 attrs.mpa_attr = ep->mpa_attr;
892 attrs.max_ird = ep->ird;
893 attrs.max_ord = ep->ord;
894 attrs.llp_stream_handle = ep;
895 attrs.next_state = IWCH_QP_STATE_RTS;
896
897 mask = IWCH_QP_ATTR_NEXT_STATE |
898 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
899 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
900
901 /* bind QP and TID with INIT_WR */
902 err = iwch_modify_qp(ep->com.qp->rhp,
903 ep->com.qp, mask, &attrs, 1);
904 if (!err)
905 goto out;
906 err:
907 abort_connection(ep, skb, GFP_KERNEL);
908 out:
909 connect_reply_upcall(ep, err);
910 return;
911 }
912
913 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
914 {
915 struct mpa_message *mpa;
916 u16 plen;
917
918 PDBG("%s ep %p\n", __FUNCTION__, ep);
919
920 /*
921 * Stop mpa timer. If it expired, then the state has
922 * changed and we bail since ep_timeout already aborted
923 * the connection.
924 */
925 stop_ep_timer(ep);
926 if (state_read(&ep->com) != MPA_REQ_WAIT)
927 return;
928
929 /*
930 * If we get more than the supported amount of private data
931 * then we must fail this connection.
932 */
933 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
934 abort_connection(ep, skb, GFP_KERNEL);
935 return;
936 }
937
938 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
939
940 /*
941 * Copy the new data into our accumulation buffer.
942 */
943 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
944 ep->mpa_pkt_len += skb->len;
945
946 /*
947 * If we don't even have the mpa message, then bail.
948 * We'll continue process when more data arrives.
949 */
950 if (ep->mpa_pkt_len < sizeof(*mpa))
951 return;
952 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
953 mpa = (struct mpa_message *) ep->mpa_pkt;
954
955 /*
956 * Validate MPA Header.
957 */
958 if (mpa->revision != mpa_rev) {
959 abort_connection(ep, skb, GFP_KERNEL);
960 return;
961 }
962
963 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
964 abort_connection(ep, skb, GFP_KERNEL);
965 return;
966 }
967
968 plen = ntohs(mpa->private_data_size);
969
970 /*
971 * Fail if there's too much private data.
972 */
973 if (plen > MPA_MAX_PRIVATE_DATA) {
974 abort_connection(ep, skb, GFP_KERNEL);
975 return;
976 }
977
978 /*
979 * If plen does not account for pkt size
980 */
981 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
982 abort_connection(ep, skb, GFP_KERNEL);
983 return;
984 }
985 ep->plen = (u8) plen;
986
987 /*
988 * If we don't have all the pdata yet, then bail.
989 */
990 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
991 return;
992
993 /*
994 * If we get here we have accumulated the entire mpa
995 * start reply message including private data.
996 */
997 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
998 ep->mpa_attr.recv_marker_enabled = markers_enabled;
999 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1000 ep->mpa_attr.version = mpa_rev;
1001 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1002 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
1003 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1004 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1005
1006 state_set(&ep->com, MPA_REQ_RCVD);
1007
1008 /* drive upcall */
1009 connect_request_upcall(ep);
1010 return;
1011 }
1012
1013 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1014 {
1015 struct iwch_ep *ep = ctx;
1016 struct cpl_rx_data *hdr = cplhdr(skb);
1017 unsigned int dlen = ntohs(hdr->len);
1018
1019 PDBG("%s ep %p dlen %u\n", __FUNCTION__, ep, dlen);
1020
1021 skb_pull(skb, sizeof(*hdr));
1022 skb_trim(skb, dlen);
1023
1024 switch (state_read(&ep->com)) {
1025 case MPA_REQ_SENT:
1026 process_mpa_reply(ep, skb);
1027 break;
1028 case MPA_REQ_WAIT:
1029 process_mpa_request(ep, skb);
1030 break;
1031 case MPA_REP_SENT:
1032 break;
1033 default:
1034 printk(KERN_ERR MOD "%s Unexpected streaming data."
1035 " ep %p state %d tid %d\n",
1036 __FUNCTION__, ep, state_read(&ep->com), ep->hwtid);
1037
1038 /*
1039 * The ep will timeout and inform the ULP of the failure.
1040 * See ep_timeout().
1041 */
1042 break;
1043 }
1044
1045 /* update RX credits */
1046 update_rx_credits(ep, dlen);
1047
1048 return CPL_RET_BUF_DONE;
1049 }
1050
1051 /*
1052 * Upcall from the adapter indicating data has been transmitted.
1053 * For us its just the single MPA request or reply. We can now free
1054 * the skb holding the mpa message.
1055 */
1056 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1057 {
1058 struct iwch_ep *ep = ctx;
1059 struct cpl_wr_ack *hdr = cplhdr(skb);
1060 unsigned int credits = ntohs(hdr->credits);
1061 enum iwch_qp_attr_mask mask;
1062
1063 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
1064
1065 if (credits == 0)
1066 return CPL_RET_BUF_DONE;
1067 BUG_ON(credits != 1);
1068 BUG_ON(ep->mpa_skb == NULL);
1069 kfree_skb(ep->mpa_skb);
1070 ep->mpa_skb = NULL;
1071 dst_confirm(ep->dst);
1072 if (state_read(&ep->com) == MPA_REP_SENT) {
1073 struct iwch_qp_attributes attrs;
1074
1075 /* bind QP to EP and move to RTS */
1076 attrs.mpa_attr = ep->mpa_attr;
1077 attrs.max_ird = ep->ord;
1078 attrs.max_ord = ep->ord;
1079 attrs.llp_stream_handle = ep;
1080 attrs.next_state = IWCH_QP_STATE_RTS;
1081
1082 /* bind QP and TID with INIT_WR */
1083 mask = IWCH_QP_ATTR_NEXT_STATE |
1084 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1085 IWCH_QP_ATTR_MPA_ATTR |
1086 IWCH_QP_ATTR_MAX_IRD |
1087 IWCH_QP_ATTR_MAX_ORD;
1088
1089 ep->com.rpl_err = iwch_modify_qp(ep->com.qp->rhp,
1090 ep->com.qp, mask, &attrs, 1);
1091
1092 if (!ep->com.rpl_err) {
1093 state_set(&ep->com, FPDU_MODE);
1094 established_upcall(ep);
1095 }
1096
1097 ep->com.rpl_done = 1;
1098 PDBG("waking up ep %p\n", ep);
1099 wake_up(&ep->com.waitq);
1100 }
1101 return CPL_RET_BUF_DONE;
1102 }
1103
1104 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1105 {
1106 struct iwch_ep *ep = ctx;
1107
1108 PDBG("%s ep %p\n", __FUNCTION__, ep);
1109
1110 close_complete_upcall(ep);
1111 state_set(&ep->com, DEAD);
1112 release_ep_resources(ep);
1113 return CPL_RET_BUF_DONE;
1114 }
1115
1116 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1117 {
1118 struct iwch_ep *ep = ctx;
1119 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1120
1121 PDBG("%s ep %p status %u errno %d\n", __FUNCTION__, ep, rpl->status,
1122 status2errno(rpl->status));
1123 connect_reply_upcall(ep, status2errno(rpl->status));
1124 state_set(&ep->com, DEAD);
1125 if (ep->com.tdev->type == T3B)
1126 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1127 cxgb3_free_atid(ep->com.tdev, ep->atid);
1128 dst_release(ep->dst);
1129 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1130 put_ep(&ep->com);
1131 return CPL_RET_BUF_DONE;
1132 }
1133
1134 static int listen_start(struct iwch_listen_ep *ep)
1135 {
1136 struct sk_buff *skb;
1137 struct cpl_pass_open_req *req;
1138
1139 PDBG("%s ep %p\n", __FUNCTION__, ep);
1140 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1141 if (!skb) {
1142 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1143 return -ENOMEM;
1144 }
1145
1146 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1147 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1148 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1149 req->local_port = ep->com.local_addr.sin_port;
1150 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1151 req->peer_port = 0;
1152 req->peer_ip = 0;
1153 req->peer_netmask = 0;
1154 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1155 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1156 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1157
1158 skb->priority = 1;
1159 ep->com.tdev->send(ep->com.tdev, skb);
1160 return 0;
1161 }
1162
1163 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1164 {
1165 struct iwch_listen_ep *ep = ctx;
1166 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1167
1168 PDBG("%s ep %p status %d error %d\n", __FUNCTION__, ep,
1169 rpl->status, status2errno(rpl->status));
1170 ep->com.rpl_err = status2errno(rpl->status);
1171 ep->com.rpl_done = 1;
1172 wake_up(&ep->com.waitq);
1173
1174 return CPL_RET_BUF_DONE;
1175 }
1176
1177 static int listen_stop(struct iwch_listen_ep *ep)
1178 {
1179 struct sk_buff *skb;
1180 struct cpl_close_listserv_req *req;
1181
1182 PDBG("%s ep %p\n", __FUNCTION__, ep);
1183 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1184 if (!skb) {
1185 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
1186 return -ENOMEM;
1187 }
1188 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1189 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1190 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1191 skb->priority = 1;
1192 ep->com.tdev->send(ep->com.tdev, skb);
1193 return 0;
1194 }
1195
1196 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1197 void *ctx)
1198 {
1199 struct iwch_listen_ep *ep = ctx;
1200 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1201
1202 PDBG("%s ep %p\n", __FUNCTION__, ep);
1203 ep->com.rpl_err = status2errno(rpl->status);
1204 ep->com.rpl_done = 1;
1205 wake_up(&ep->com.waitq);
1206 return CPL_RET_BUF_DONE;
1207 }
1208
1209 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1210 {
1211 struct cpl_pass_accept_rpl *rpl;
1212 unsigned int mtu_idx;
1213 u32 opt0h, opt0l, opt2;
1214 int wscale;
1215
1216 PDBG("%s ep %p\n", __FUNCTION__, ep);
1217 BUG_ON(skb_cloned(skb));
1218 skb_trim(skb, sizeof(*rpl));
1219 skb_get(skb);
1220 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1221 wscale = compute_wscale(rcv_win);
1222 opt0h = V_NAGLE(0) |
1223 V_NO_CONG(nocong) |
1224 V_KEEP_ALIVE(1) |
1225 F_TCAM_BYPASS |
1226 V_WND_SCALE(wscale) |
1227 V_MSS_IDX(mtu_idx) |
1228 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1229 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1230 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1231
1232 rpl = cplhdr(skb);
1233 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1234 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1235 rpl->peer_ip = peer_ip;
1236 rpl->opt0h = htonl(opt0h);
1237 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1238 rpl->opt2 = htonl(opt2);
1239 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1240 skb->priority = CPL_PRIORITY_SETUP;
1241 l2t_send(ep->com.tdev, skb, ep->l2t);
1242
1243 return;
1244 }
1245
1246 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1247 struct sk_buff *skb)
1248 {
1249 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __FUNCTION__, tdev, hwtid,
1250 peer_ip);
1251 BUG_ON(skb_cloned(skb));
1252 skb_trim(skb, sizeof(struct cpl_tid_release));
1253 skb_get(skb);
1254
1255 if (tdev->type == T3B)
1256 release_tid(tdev, hwtid, skb);
1257 else {
1258 struct cpl_pass_accept_rpl *rpl;
1259
1260 rpl = cplhdr(skb);
1261 skb->priority = CPL_PRIORITY_SETUP;
1262 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1263 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1264 hwtid));
1265 rpl->peer_ip = peer_ip;
1266 rpl->opt0h = htonl(F_TCAM_BYPASS);
1267 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1268 rpl->opt2 = 0;
1269 rpl->rsvd = rpl->opt2;
1270 tdev->send(tdev, skb);
1271 }
1272 }
1273
1274 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1275 {
1276 struct iwch_ep *child_ep, *parent_ep = ctx;
1277 struct cpl_pass_accept_req *req = cplhdr(skb);
1278 unsigned int hwtid = GET_TID(req);
1279 struct dst_entry *dst;
1280 struct l2t_entry *l2t;
1281 struct rtable *rt;
1282 struct iff_mac tim;
1283
1284 PDBG("%s parent ep %p tid %u\n", __FUNCTION__, parent_ep, hwtid);
1285
1286 if (state_read(&parent_ep->com) != LISTEN) {
1287 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1288 __FUNCTION__);
1289 goto reject;
1290 }
1291
1292 /*
1293 * Find the netdev for this connection request.
1294 */
1295 tim.mac_addr = req->dst_mac;
1296 tim.vlan_tag = ntohs(req->vlan_tag);
1297 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1298 printk(KERN_ERR
1299 "%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
1300 __FUNCTION__,
1301 req->dst_mac[0],
1302 req->dst_mac[1],
1303 req->dst_mac[2],
1304 req->dst_mac[3],
1305 req->dst_mac[4],
1306 req->dst_mac[5]);
1307 goto reject;
1308 }
1309
1310 /* Find output route */
1311 rt = find_route(tdev,
1312 req->local_ip,
1313 req->peer_ip,
1314 req->local_port,
1315 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1316 if (!rt) {
1317 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1318 __FUNCTION__);
1319 goto reject;
1320 }
1321 dst = &rt->u.dst;
1322 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1323 if (!l2t) {
1324 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1325 __FUNCTION__);
1326 dst_release(dst);
1327 goto reject;
1328 }
1329 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1330 if (!child_ep) {
1331 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1332 __FUNCTION__);
1333 l2t_release(L2DATA(tdev), l2t);
1334 dst_release(dst);
1335 goto reject;
1336 }
1337 state_set(&child_ep->com, CONNECTING);
1338 child_ep->com.tdev = tdev;
1339 child_ep->com.cm_id = NULL;
1340 child_ep->com.local_addr.sin_family = PF_INET;
1341 child_ep->com.local_addr.sin_port = req->local_port;
1342 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1343 child_ep->com.remote_addr.sin_family = PF_INET;
1344 child_ep->com.remote_addr.sin_port = req->peer_port;
1345 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1346 get_ep(&parent_ep->com);
1347 child_ep->parent_ep = parent_ep;
1348 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1349 child_ep->l2t = l2t;
1350 child_ep->dst = dst;
1351 child_ep->hwtid = hwtid;
1352 init_timer(&child_ep->timer);
1353 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1354 accept_cr(child_ep, req->peer_ip, skb);
1355 goto out;
1356 reject:
1357 reject_cr(tdev, hwtid, req->peer_ip, skb);
1358 out:
1359 return CPL_RET_BUF_DONE;
1360 }
1361
1362 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1363 {
1364 struct iwch_ep *ep = ctx;
1365 struct cpl_pass_establish *req = cplhdr(skb);
1366
1367 PDBG("%s ep %p\n", __FUNCTION__, ep);
1368 ep->snd_seq = ntohl(req->snd_isn);
1369
1370 set_emss(ep, ntohs(req->tcp_opt));
1371
1372 dst_confirm(ep->dst);
1373 state_set(&ep->com, MPA_REQ_WAIT);
1374 start_ep_timer(ep);
1375
1376 return CPL_RET_BUF_DONE;
1377 }
1378
1379 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1380 {
1381 struct iwch_ep *ep = ctx;
1382 struct iwch_qp_attributes attrs;
1383 unsigned long flags;
1384 int disconnect = 1;
1385 int release = 0;
1386
1387 PDBG("%s ep %p\n", __FUNCTION__, ep);
1388 dst_confirm(ep->dst);
1389
1390 spin_lock_irqsave(&ep->com.lock, flags);
1391 switch (ep->com.state) {
1392 case MPA_REQ_WAIT:
1393 __state_set(&ep->com, CLOSING);
1394 break;
1395 case MPA_REQ_SENT:
1396 __state_set(&ep->com, CLOSING);
1397 connect_reply_upcall(ep, -ECONNRESET);
1398 break;
1399 case MPA_REQ_RCVD:
1400
1401 /*
1402 * We're gonna mark this puppy DEAD, but keep
1403 * the reference on it until the ULP accepts or
1404 * rejects the CR.
1405 */
1406 __state_set(&ep->com, CLOSING);
1407 get_ep(&ep->com);
1408 break;
1409 case MPA_REP_SENT:
1410 __state_set(&ep->com, CLOSING);
1411 ep->com.rpl_done = 1;
1412 ep->com.rpl_err = -ECONNRESET;
1413 PDBG("waking up ep %p\n", ep);
1414 wake_up(&ep->com.waitq);
1415 break;
1416 case FPDU_MODE:
1417 start_ep_timer(ep);
1418 __state_set(&ep->com, CLOSING);
1419 attrs.next_state = IWCH_QP_STATE_CLOSING;
1420 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1421 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1422 peer_close_upcall(ep);
1423 break;
1424 case ABORTING:
1425 disconnect = 0;
1426 break;
1427 case CLOSING:
1428 __state_set(&ep->com, MORIBUND);
1429 disconnect = 0;
1430 break;
1431 case MORIBUND:
1432 stop_ep_timer(ep);
1433 if (ep->com.cm_id && ep->com.qp) {
1434 attrs.next_state = IWCH_QP_STATE_IDLE;
1435 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1436 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1437 }
1438 close_complete_upcall(ep);
1439 __state_set(&ep->com, DEAD);
1440 release = 1;
1441 disconnect = 0;
1442 break;
1443 case DEAD:
1444 disconnect = 0;
1445 break;
1446 default:
1447 BUG_ON(1);
1448 }
1449 spin_unlock_irqrestore(&ep->com.lock, flags);
1450 if (disconnect)
1451 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1452 if (release)
1453 release_ep_resources(ep);
1454 return CPL_RET_BUF_DONE;
1455 }
1456
1457 /*
1458 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1459 */
1460 static int is_neg_adv_abort(unsigned int status)
1461 {
1462 return status == CPL_ERR_RTX_NEG_ADVICE ||
1463 status == CPL_ERR_PERSIST_NEG_ADVICE;
1464 }
1465
1466 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1467 {
1468 struct cpl_abort_req_rss *req = cplhdr(skb);
1469 struct iwch_ep *ep = ctx;
1470 struct cpl_abort_rpl *rpl;
1471 struct sk_buff *rpl_skb;
1472 struct iwch_qp_attributes attrs;
1473 int ret;
1474 int state;
1475
1476 if (is_neg_adv_abort(req->status)) {
1477 PDBG("%s neg_adv_abort ep %p tid %d\n", __FUNCTION__, ep,
1478 ep->hwtid);
1479 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1480 return CPL_RET_BUF_DONE;
1481 }
1482
1483 state = state_read(&ep->com);
1484 PDBG("%s ep %p state %u\n", __FUNCTION__, ep, state);
1485 switch (state) {
1486 case CONNECTING:
1487 break;
1488 case MPA_REQ_WAIT:
1489 stop_ep_timer(ep);
1490 break;
1491 case MPA_REQ_SENT:
1492 stop_ep_timer(ep);
1493 connect_reply_upcall(ep, -ECONNRESET);
1494 break;
1495 case MPA_REP_SENT:
1496 ep->com.rpl_done = 1;
1497 ep->com.rpl_err = -ECONNRESET;
1498 PDBG("waking up ep %p\n", ep);
1499 wake_up(&ep->com.waitq);
1500 break;
1501 case MPA_REQ_RCVD:
1502
1503 /*
1504 * We're gonna mark this puppy DEAD, but keep
1505 * the reference on it until the ULP accepts or
1506 * rejects the CR.
1507 */
1508 get_ep(&ep->com);
1509 break;
1510 case MORIBUND:
1511 case CLOSING:
1512 stop_ep_timer(ep);
1513 /*FALLTHROUGH*/
1514 case FPDU_MODE:
1515 if (ep->com.cm_id && ep->com.qp) {
1516 attrs.next_state = IWCH_QP_STATE_ERROR;
1517 ret = iwch_modify_qp(ep->com.qp->rhp,
1518 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1519 &attrs, 1);
1520 if (ret)
1521 printk(KERN_ERR MOD
1522 "%s - qp <- error failed!\n",
1523 __FUNCTION__);
1524 }
1525 peer_abort_upcall(ep);
1526 break;
1527 case ABORTING:
1528 break;
1529 case DEAD:
1530 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __FUNCTION__);
1531 return CPL_RET_BUF_DONE;
1532 default:
1533 BUG_ON(1);
1534 break;
1535 }
1536 dst_confirm(ep->dst);
1537
1538 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1539 if (!rpl_skb) {
1540 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1541 __FUNCTION__);
1542 dst_release(ep->dst);
1543 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1544 put_ep(&ep->com);
1545 return CPL_RET_BUF_DONE;
1546 }
1547 rpl_skb->priority = CPL_PRIORITY_DATA;
1548 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1549 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1550 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1551 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1552 rpl->cmd = CPL_ABORT_NO_RST;
1553 ep->com.tdev->send(ep->com.tdev, rpl_skb);
1554 if (state != ABORTING) {
1555 state_set(&ep->com, DEAD);
1556 release_ep_resources(ep);
1557 }
1558 return CPL_RET_BUF_DONE;
1559 }
1560
1561 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1562 {
1563 struct iwch_ep *ep = ctx;
1564 struct iwch_qp_attributes attrs;
1565 unsigned long flags;
1566 int release = 0;
1567
1568 PDBG("%s ep %p\n", __FUNCTION__, ep);
1569 BUG_ON(!ep);
1570
1571 /* The cm_id may be null if we failed to connect */
1572 spin_lock_irqsave(&ep->com.lock, flags);
1573 switch (ep->com.state) {
1574 case CLOSING:
1575 __state_set(&ep->com, MORIBUND);
1576 break;
1577 case MORIBUND:
1578 stop_ep_timer(ep);
1579 if ((ep->com.cm_id) && (ep->com.qp)) {
1580 attrs.next_state = IWCH_QP_STATE_IDLE;
1581 iwch_modify_qp(ep->com.qp->rhp,
1582 ep->com.qp,
1583 IWCH_QP_ATTR_NEXT_STATE,
1584 &attrs, 1);
1585 }
1586 close_complete_upcall(ep);
1587 __state_set(&ep->com, DEAD);
1588 release = 1;
1589 break;
1590 case ABORTING:
1591 break;
1592 case DEAD:
1593 default:
1594 BUG_ON(1);
1595 break;
1596 }
1597 spin_unlock_irqrestore(&ep->com.lock, flags);
1598 if (release)
1599 release_ep_resources(ep);
1600 return CPL_RET_BUF_DONE;
1601 }
1602
1603 /*
1604 * T3A does 3 things when a TERM is received:
1605 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1606 * 2) generate an async event on the QP with the TERMINATE opcode
1607 * 3) post a TERMINATE opcde cqe into the associated CQ.
1608 *
1609 * For (1), we save the message in the qp for later consumer consumption.
1610 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1611 * For (3), we toss the CQE in cxio_poll_cq().
1612 *
1613 * terminate() handles case (1)...
1614 */
1615 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1616 {
1617 struct iwch_ep *ep = ctx;
1618
1619 PDBG("%s ep %p\n", __FUNCTION__, ep);
1620 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1621 PDBG("%s saving %d bytes of term msg\n", __FUNCTION__, skb->len);
1622 memcpy(ep->com.qp->attr.terminate_buffer, skb->data, skb->len);
1623 ep->com.qp->attr.terminate_msg_len = skb->len;
1624 ep->com.qp->attr.is_terminate_local = 0;
1625 return CPL_RET_BUF_DONE;
1626 }
1627
1628 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1629 {
1630 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1631 struct iwch_ep *ep = ctx;
1632
1633 PDBG("%s ep %p tid %u status %d\n", __FUNCTION__, ep, ep->hwtid,
1634 rep->status);
1635 if (rep->status) {
1636 struct iwch_qp_attributes attrs;
1637
1638 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1639 __FUNCTION__, ep->hwtid);
1640 stop_ep_timer(ep);
1641 attrs.next_state = IWCH_QP_STATE_ERROR;
1642 iwch_modify_qp(ep->com.qp->rhp,
1643 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1644 &attrs, 1);
1645 abort_connection(ep, NULL, GFP_KERNEL);
1646 }
1647 return CPL_RET_BUF_DONE;
1648 }
1649
1650 static void ep_timeout(unsigned long arg)
1651 {
1652 struct iwch_ep *ep = (struct iwch_ep *)arg;
1653 struct iwch_qp_attributes attrs;
1654 unsigned long flags;
1655
1656 spin_lock_irqsave(&ep->com.lock, flags);
1657 PDBG("%s ep %p tid %u state %d\n", __FUNCTION__, ep, ep->hwtid,
1658 ep->com.state);
1659 switch (ep->com.state) {
1660 case MPA_REQ_SENT:
1661 connect_reply_upcall(ep, -ETIMEDOUT);
1662 break;
1663 case MPA_REQ_WAIT:
1664 break;
1665 case CLOSING:
1666 case MORIBUND:
1667 if (ep->com.cm_id && ep->com.qp) {
1668 attrs.next_state = IWCH_QP_STATE_ERROR;
1669 iwch_modify_qp(ep->com.qp->rhp,
1670 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1671 &attrs, 1);
1672 }
1673 break;
1674 default:
1675 BUG();
1676 }
1677 __state_set(&ep->com, CLOSING);
1678 spin_unlock_irqrestore(&ep->com.lock, flags);
1679 abort_connection(ep, NULL, GFP_ATOMIC);
1680 put_ep(&ep->com);
1681 }
1682
1683 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1684 {
1685 int err;
1686 struct iwch_ep *ep = to_ep(cm_id);
1687 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1688
1689 if (state_read(&ep->com) == DEAD) {
1690 put_ep(&ep->com);
1691 return -ECONNRESET;
1692 }
1693 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1694 if (mpa_rev == 0)
1695 abort_connection(ep, NULL, GFP_KERNEL);
1696 else {
1697 err = send_mpa_reject(ep, pdata, pdata_len);
1698 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1699 }
1700 return 0;
1701 }
1702
1703 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1704 {
1705 int err;
1706 struct iwch_qp_attributes attrs;
1707 enum iwch_qp_attr_mask mask;
1708 struct iwch_ep *ep = to_ep(cm_id);
1709 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1710 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1711
1712 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1713 if (state_read(&ep->com) == DEAD) {
1714 put_ep(&ep->com);
1715 return -ECONNRESET;
1716 }
1717
1718 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1719 BUG_ON(!qp);
1720
1721 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1722 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1723 abort_connection(ep, NULL, GFP_KERNEL);
1724 return -EINVAL;
1725 }
1726
1727 cm_id->add_ref(cm_id);
1728 ep->com.cm_id = cm_id;
1729 ep->com.qp = qp;
1730
1731 ep->com.rpl_done = 0;
1732 ep->com.rpl_err = 0;
1733 ep->ird = conn_param->ird;
1734 ep->ord = conn_param->ord;
1735 PDBG("%s %d ird %d ord %d\n", __FUNCTION__, __LINE__, ep->ird, ep->ord);
1736 get_ep(&ep->com);
1737 err = send_mpa_reply(ep, conn_param->private_data,
1738 conn_param->private_data_len);
1739 if (err) {
1740 ep->com.cm_id = NULL;
1741 ep->com.qp = NULL;
1742 cm_id->rem_ref(cm_id);
1743 abort_connection(ep, NULL, GFP_KERNEL);
1744 put_ep(&ep->com);
1745 return err;
1746 }
1747
1748 /* bind QP to EP and move to RTS */
1749 attrs.mpa_attr = ep->mpa_attr;
1750 attrs.max_ird = ep->ord;
1751 attrs.max_ord = ep->ord;
1752 attrs.llp_stream_handle = ep;
1753 attrs.next_state = IWCH_QP_STATE_RTS;
1754
1755 /* bind QP and TID with INIT_WR */
1756 mask = IWCH_QP_ATTR_NEXT_STATE |
1757 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1758 IWCH_QP_ATTR_MPA_ATTR |
1759 IWCH_QP_ATTR_MAX_IRD |
1760 IWCH_QP_ATTR_MAX_ORD;
1761
1762 err = iwch_modify_qp(ep->com.qp->rhp,
1763 ep->com.qp, mask, &attrs, 1);
1764
1765 if (err) {
1766 ep->com.cm_id = NULL;
1767 ep->com.qp = NULL;
1768 cm_id->rem_ref(cm_id);
1769 abort_connection(ep, NULL, GFP_KERNEL);
1770 } else {
1771 state_set(&ep->com, FPDU_MODE);
1772 established_upcall(ep);
1773 }
1774 put_ep(&ep->com);
1775 return err;
1776 }
1777
1778 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1779 {
1780 int err = 0;
1781 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1782 struct iwch_ep *ep;
1783 struct rtable *rt;
1784
1785 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1786 if (!ep) {
1787 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1788 err = -ENOMEM;
1789 goto out;
1790 }
1791 init_timer(&ep->timer);
1792 ep->plen = conn_param->private_data_len;
1793 if (ep->plen)
1794 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1795 conn_param->private_data, ep->plen);
1796 ep->ird = conn_param->ird;
1797 ep->ord = conn_param->ord;
1798 ep->com.tdev = h->rdev.t3cdev_p;
1799
1800 cm_id->add_ref(cm_id);
1801 ep->com.cm_id = cm_id;
1802 ep->com.qp = get_qhp(h, conn_param->qpn);
1803 BUG_ON(!ep->com.qp);
1804 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __FUNCTION__, conn_param->qpn,
1805 ep->com.qp, cm_id);
1806
1807 /*
1808 * Allocate an active TID to initiate a TCP connection.
1809 */
1810 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1811 if (ep->atid == -1) {
1812 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1813 err = -ENOMEM;
1814 goto fail2;
1815 }
1816
1817 /* find a route */
1818 rt = find_route(h->rdev.t3cdev_p,
1819 cm_id->local_addr.sin_addr.s_addr,
1820 cm_id->remote_addr.sin_addr.s_addr,
1821 cm_id->local_addr.sin_port,
1822 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1823 if (!rt) {
1824 printk(KERN_ERR MOD "%s - cannot find route.\n", __FUNCTION__);
1825 err = -EHOSTUNREACH;
1826 goto fail3;
1827 }
1828 ep->dst = &rt->u.dst;
1829
1830 /* get a l2t entry */
1831 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1832 ep->dst->neighbour->dev);
1833 if (!ep->l2t) {
1834 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __FUNCTION__);
1835 err = -ENOMEM;
1836 goto fail4;
1837 }
1838
1839 state_set(&ep->com, CONNECTING);
1840 ep->tos = IPTOS_LOWDELAY;
1841 ep->com.local_addr = cm_id->local_addr;
1842 ep->com.remote_addr = cm_id->remote_addr;
1843
1844 /* send connect request to rnic */
1845 err = send_connect(ep);
1846 if (!err)
1847 goto out;
1848
1849 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1850 fail4:
1851 dst_release(ep->dst);
1852 fail3:
1853 cxgb3_free_atid(ep->com.tdev, ep->atid);
1854 fail2:
1855 put_ep(&ep->com);
1856 out:
1857 return err;
1858 }
1859
1860 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1861 {
1862 int err = 0;
1863 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1864 struct iwch_listen_ep *ep;
1865
1866
1867 might_sleep();
1868
1869 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1870 if (!ep) {
1871 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1872 err = -ENOMEM;
1873 goto fail1;
1874 }
1875 PDBG("%s ep %p\n", __FUNCTION__, ep);
1876 ep->com.tdev = h->rdev.t3cdev_p;
1877 cm_id->add_ref(cm_id);
1878 ep->com.cm_id = cm_id;
1879 ep->backlog = backlog;
1880 ep->com.local_addr = cm_id->local_addr;
1881
1882 /*
1883 * Allocate a server TID.
1884 */
1885 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1886 if (ep->stid == -1) {
1887 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1888 err = -ENOMEM;
1889 goto fail2;
1890 }
1891
1892 state_set(&ep->com, LISTEN);
1893 err = listen_start(ep);
1894 if (err)
1895 goto fail3;
1896
1897 /* wait for pass_open_rpl */
1898 wait_event(ep->com.waitq, ep->com.rpl_done);
1899 err = ep->com.rpl_err;
1900 if (!err) {
1901 cm_id->provider_data = ep;
1902 goto out;
1903 }
1904 fail3:
1905 cxgb3_free_stid(ep->com.tdev, ep->stid);
1906 fail2:
1907 put_ep(&ep->com);
1908 fail1:
1909 out:
1910 return err;
1911 }
1912
1913 int iwch_destroy_listen(struct iw_cm_id *cm_id)
1914 {
1915 int err;
1916 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
1917
1918 PDBG("%s ep %p\n", __FUNCTION__, ep);
1919
1920 might_sleep();
1921 state_set(&ep->com, DEAD);
1922 ep->com.rpl_done = 0;
1923 ep->com.rpl_err = 0;
1924 err = listen_stop(ep);
1925 wait_event(ep->com.waitq, ep->com.rpl_done);
1926 cxgb3_free_stid(ep->com.tdev, ep->stid);
1927 err = ep->com.rpl_err;
1928 cm_id->rem_ref(cm_id);
1929 put_ep(&ep->com);
1930 return err;
1931 }
1932
1933 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
1934 {
1935 int ret=0;
1936 unsigned long flags;
1937 int close = 0;
1938
1939 spin_lock_irqsave(&ep->com.lock, flags);
1940
1941 PDBG("%s ep %p state %s, abrupt %d\n", __FUNCTION__, ep,
1942 states[ep->com.state], abrupt);
1943
1944 if (ep->com.state == DEAD) {
1945 PDBG("%s already dead ep %p\n", __FUNCTION__, ep);
1946 goto out;
1947 }
1948
1949 if (abrupt) {
1950 if (ep->com.state != ABORTING) {
1951 ep->com.state = ABORTING;
1952 close = 1;
1953 }
1954 goto out;
1955 }
1956
1957 switch (ep->com.state) {
1958 case MPA_REQ_WAIT:
1959 case MPA_REQ_SENT:
1960 case MPA_REQ_RCVD:
1961 case MPA_REP_SENT:
1962 case FPDU_MODE:
1963 start_ep_timer(ep);
1964 ep->com.state = CLOSING;
1965 close = 1;
1966 break;
1967 case CLOSING:
1968 ep->com.state = MORIBUND;
1969 close = 1;
1970 break;
1971 case MORIBUND:
1972 break;
1973 default:
1974 BUG();
1975 break;
1976 }
1977 out:
1978 spin_unlock_irqrestore(&ep->com.lock, flags);
1979 if (close) {
1980 if (abrupt)
1981 ret = send_abort(ep, NULL, gfp);
1982 else
1983 ret = send_halfclose(ep, gfp);
1984 }
1985 return ret;
1986 }
1987
1988 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
1989 struct l2t_entry *l2t)
1990 {
1991 struct iwch_ep *ep = ctx;
1992
1993 if (ep->dst != old)
1994 return 0;
1995
1996 PDBG("%s ep %p redirect to dst %p l2t %p\n", __FUNCTION__, ep, new,
1997 l2t);
1998 dst_hold(new);
1999 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2000 ep->l2t = l2t;
2001 dst_release(old);
2002 ep->dst = new;
2003 return 1;
2004 }
2005
2006 /*
2007 * All the CM events are handled on a work queue to have a safe context.
2008 */
2009 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2010 {
2011 struct iwch_ep_common *epc = ctx;
2012
2013 get_ep(epc);
2014
2015 /*
2016 * Save ctx and tdev in the skb->cb area.
2017 */
2018 *((void **) skb->cb) = ctx;
2019 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2020
2021 /*
2022 * Queue the skb and schedule the worker thread.
2023 */
2024 skb_queue_tail(&rxq, skb);
2025 queue_work(workq, &skb_work);
2026 return 0;
2027 }
2028
2029 int __init iwch_cm_init(void)
2030 {
2031 skb_queue_head_init(&rxq);
2032
2033 workq = create_singlethread_workqueue("iw_cxgb3");
2034 if (!workq)
2035 return -ENOMEM;
2036
2037 /*
2038 * All upcalls from the T3 Core go to sched() to
2039 * schedule the processing on a work queue.
2040 */
2041 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2042 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2043 t3c_handlers[CPL_RX_DATA] = sched;
2044 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2045 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2046 t3c_handlers[CPL_ABORT_RPL] = sched;
2047 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2048 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2049 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2050 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2051 t3c_handlers[CPL_PEER_CLOSE] = sched;
2052 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2053 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2054 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2055 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
2056
2057 /*
2058 * These are the real handlers that are called from a
2059 * work queue.
2060 */
2061 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2062 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2063 work_handlers[CPL_RX_DATA] = rx_data;
2064 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2065 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2066 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2067 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2068 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2069 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2070 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2071 work_handlers[CPL_PEER_CLOSE] = peer_close;
2072 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2073 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2074 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2075 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2076 return 0;
2077 }
2078
2079 void __exit iwch_cm_term(void)
2080 {
2081 flush_workqueue(workq);
2082 destroy_workqueue(workq);
2083 }