?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add linux-next specific files for 20110831
[linux-2.6/next.git] / drivers / infiniband / hw / cxgb3 / iwch_cm.c
blob17bf9d95463c222f31169e42e220c91a9619d034
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/slab.h>
35 #include <linux/workqueue.h>
36 #include <linux/skbuff.h>
37 #include <linux/timer.h>
38 #include <linux/notifier.h>
39 #include <linux/inetdevice.h>
40
41 #include <net/neighbour.h>
42 #include <net/netevent.h>
43 #include <net/route.h>
44
45 #include "tcb.h"
46 #include "cxgb3_offload.h"
47 #include "iwch.h"
48 #include "iwch_provider.h"
49 #include "iwch_cm.h"
50
51 static char *states[] = {
52 "idle",
53 "listen",
54 "connecting",
55 "mpa_wait_req",
56 "mpa_req_sent",
57 "mpa_req_rcvd",
58 "mpa_rep_sent",
59 "fpdu_mode",
60 "aborting",
61 "closing",
62 "moribund",
63 "dead",
64 NULL,
65 };
66
67 int peer2peer = 0;
68 module_param(peer2peer, int, 0644);
69 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
70
71 static int ep_timeout_secs = 60;
72 module_param(ep_timeout_secs, int, 0644);
73 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
74 "in seconds (default=60)");
75
76 static int mpa_rev = 1;
77 module_param(mpa_rev, int, 0644);
78 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
79 "1 is spec compliant. (default=1)");
80
81 static int markers_enabled = 0;
82 module_param(markers_enabled, int, 0644);
83 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
84
85 static int crc_enabled = 1;
86 module_param(crc_enabled, int, 0644);
87 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
88
89 static int rcv_win = 256 * 1024;
90 module_param(rcv_win, int, 0644);
91 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
92
93 static int snd_win = 32 * 1024;
94 module_param(snd_win, int, 0644);
95 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
96
97 static unsigned int nocong = 0;
98 module_param(nocong, uint, 0644);
99 MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
100
101 static unsigned int cong_flavor = 1;
102 module_param(cong_flavor, uint, 0644);
103 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
104
105 static struct workqueue_struct *workq;
106
107 static struct sk_buff_head rxq;
108
109 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
110 static void ep_timeout(unsigned long arg);
111 static void connect_reply_upcall(struct iwch_ep *ep, int status);
112
113 static void start_ep_timer(struct iwch_ep *ep)
114 {
115 PDBG("%s ep %p\n", __func__, ep);
116 if (timer_pending(&ep->timer)) {
117 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
118 del_timer_sync(&ep->timer);
119 } else
120 get_ep(&ep->com);
121 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
122 ep->timer.data = (unsigned long)ep;
123 ep->timer.function = ep_timeout;
124 add_timer(&ep->timer);
125 }
126
127 static void stop_ep_timer(struct iwch_ep *ep)
128 {
129 PDBG("%s ep %p\n", __func__, ep);
130 if (!timer_pending(&ep->timer)) {
131 printk(KERN_ERR "%s timer stopped when its not running! ep %p state %u\n",
132 __func__, ep, ep->com.state);
133 WARN_ON(1);
134 return;
135 }
136 del_timer_sync(&ep->timer);
137 put_ep(&ep->com);
138 }
139
140 static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
141 {
142 int error = 0;
143 struct cxio_rdev *rdev;
144
145 rdev = (struct cxio_rdev *)tdev->ulp;
146 if (cxio_fatal_error(rdev)) {
147 kfree_skb(skb);
148 return -EIO;
149 }
150 error = l2t_send(tdev, skb, l2e);
151 if (error < 0)
152 kfree_skb(skb);
153 return error;
154 }
155
156 int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
157 {
158 int error = 0;
159 struct cxio_rdev *rdev;
160
161 rdev = (struct cxio_rdev *)tdev->ulp;
162 if (cxio_fatal_error(rdev)) {
163 kfree_skb(skb);
164 return -EIO;
165 }
166 error = cxgb3_ofld_send(tdev, skb);
167 if (error < 0)
168 kfree_skb(skb);
169 return error;
170 }
171
172 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
173 {
174 struct cpl_tid_release *req;
175
176 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
177 if (!skb)
178 return;
179 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
180 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
181 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
182 skb->priority = CPL_PRIORITY_SETUP;
183 iwch_cxgb3_ofld_send(tdev, skb);
184 return;
185 }
186
187 int iwch_quiesce_tid(struct iwch_ep *ep)
188 {
189 struct cpl_set_tcb_field *req;
190 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
191
192 if (!skb)
193 return -ENOMEM;
194 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
195 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
196 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
197 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
198 req->reply = 0;
199 req->cpu_idx = 0;
200 req->word = htons(W_TCB_RX_QUIESCE);
201 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
202 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
203
204 skb->priority = CPL_PRIORITY_DATA;
205 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
206 }
207
208 int iwch_resume_tid(struct iwch_ep *ep)
209 {
210 struct cpl_set_tcb_field *req;
211 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
212
213 if (!skb)
214 return -ENOMEM;
215 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
216 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
217 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
218 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
219 req->reply = 0;
220 req->cpu_idx = 0;
221 req->word = htons(W_TCB_RX_QUIESCE);
222 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
223 req->val = 0;
224
225 skb->priority = CPL_PRIORITY_DATA;
226 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
227 }
228
229 static void set_emss(struct iwch_ep *ep, u16 opt)
230 {
231 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
232 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
233 if (G_TCPOPT_TSTAMP(opt))
234 ep->emss -= 12;
235 if (ep->emss < 128)
236 ep->emss = 128;
237 PDBG("emss=%d\n", ep->emss);
238 }
239
240 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
241 {
242 unsigned long flags;
243 enum iwch_ep_state state;
244
245 spin_lock_irqsave(&epc->lock, flags);
246 state = epc->state;
247 spin_unlock_irqrestore(&epc->lock, flags);
248 return state;
249 }
250
251 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
252 {
253 epc->state = new;
254 }
255
256 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
257 {
258 unsigned long flags;
259
260 spin_lock_irqsave(&epc->lock, flags);
261 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
262 __state_set(epc, new);
263 spin_unlock_irqrestore(&epc->lock, flags);
264 return;
265 }
266
267 static void *alloc_ep(int size, gfp_t gfp)
268 {
269 struct iwch_ep_common *epc;
270
271 epc = kzalloc(size, gfp);
272 if (epc) {
273 kref_init(&epc->kref);
274 spin_lock_init(&epc->lock);
275 init_waitqueue_head(&epc->waitq);
276 }
277 PDBG("%s alloc ep %p\n", __func__, epc);
278 return epc;
279 }
280
281 void __free_ep(struct kref *kref)
282 {
283 struct iwch_ep *ep;
284 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
285 struct iwch_ep, com);
286 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
287 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
288 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
289 dst_release(ep->dst);
290 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
291 }
292 kfree(ep);
293 }
294
295 static void release_ep_resources(struct iwch_ep *ep)
296 {
297 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
298 set_bit(RELEASE_RESOURCES, &ep->com.flags);
299 put_ep(&ep->com);
300 }
301
302 static int status2errno(int status)
303 {
304 switch (status) {
305 case CPL_ERR_NONE:
306 return 0;
307 case CPL_ERR_CONN_RESET:
308 return -ECONNRESET;
309 case CPL_ERR_ARP_MISS:
310 return -EHOSTUNREACH;
311 case CPL_ERR_CONN_TIMEDOUT:
312 return -ETIMEDOUT;
313 case CPL_ERR_TCAM_FULL:
314 return -ENOMEM;
315 case CPL_ERR_CONN_EXIST:
316 return -EADDRINUSE;
317 default:
318 return -EIO;
319 }
320 }
321
322 /*
323 * Try and reuse skbs already allocated...
324 */
325 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
326 {
327 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
328 skb_trim(skb, 0);
329 skb_get(skb);
330 } else {
331 skb = alloc_skb(len, gfp);
332 }
333 return skb;
334 }
335
336 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
337 __be32 peer_ip, __be16 local_port,
338 __be16 peer_port, u8 tos)
339 {
340 struct rtable *rt;
341 struct flowi4 fl4;
342
343 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
344 peer_port, local_port, IPPROTO_TCP,
345 tos, 0);
346 if (IS_ERR(rt))
347 return NULL;
348 return rt;
349 }
350
351 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
352 {
353 int i = 0;
354
355 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
356 ++i;
357 return i;
358 }
359
360 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
361 {
362 PDBG("%s t3cdev %p\n", __func__, dev);
363 kfree_skb(skb);
364 }
365
366 /*
367 * Handle an ARP failure for an active open.
368 */
369 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
370 {
371 printk(KERN_ERR MOD "ARP failure duing connect\n");
372 kfree_skb(skb);
373 }
374
375 /*
376 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
377 * and send it along.
378 */
379 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
380 {
381 struct cpl_abort_req *req = cplhdr(skb);
382
383 PDBG("%s t3cdev %p\n", __func__, dev);
384 req->cmd = CPL_ABORT_NO_RST;
385 iwch_cxgb3_ofld_send(dev, skb);
386 }
387
388 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
389 {
390 struct cpl_close_con_req *req;
391 struct sk_buff *skb;
392
393 PDBG("%s ep %p\n", __func__, ep);
394 skb = get_skb(NULL, sizeof(*req), gfp);
395 if (!skb) {
396 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
397 return -ENOMEM;
398 }
399 skb->priority = CPL_PRIORITY_DATA;
400 set_arp_failure_handler(skb, arp_failure_discard);
401 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
402 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
403 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
404 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
405 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
406 }
407
408 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
409 {
410 struct cpl_abort_req *req;
411
412 PDBG("%s ep %p\n", __func__, ep);
413 skb = get_skb(skb, sizeof(*req), gfp);
414 if (!skb) {
415 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
416 __func__);
417 return -ENOMEM;
418 }
419 skb->priority = CPL_PRIORITY_DATA;
420 set_arp_failure_handler(skb, abort_arp_failure);
421 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
422 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
423 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
424 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
425 req->cmd = CPL_ABORT_SEND_RST;
426 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
427 }
428
429 static int send_connect(struct iwch_ep *ep)
430 {
431 struct cpl_act_open_req *req;
432 struct sk_buff *skb;
433 u32 opt0h, opt0l, opt2;
434 unsigned int mtu_idx;
435 int wscale;
436
437 PDBG("%s ep %p\n", __func__, ep);
438
439 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
440 if (!skb) {
441 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
442 __func__);
443 return -ENOMEM;
444 }
445 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
446 wscale = compute_wscale(rcv_win);
447 opt0h = V_NAGLE(0) |
448 V_NO_CONG(nocong) |
449 V_KEEP_ALIVE(1) |
450 F_TCAM_BYPASS |
451 V_WND_SCALE(wscale) |
452 V_MSS_IDX(mtu_idx) |
453 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
454 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
455 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
456 V_CONG_CONTROL_FLAVOR(cong_flavor);
457 skb->priority = CPL_PRIORITY_SETUP;
458 set_arp_failure_handler(skb, act_open_req_arp_failure);
459
460 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
461 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
462 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
463 req->local_port = ep->com.local_addr.sin_port;
464 req->peer_port = ep->com.remote_addr.sin_port;
465 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
466 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
467 req->opt0h = htonl(opt0h);
468 req->opt0l = htonl(opt0l);
469 req->params = 0;
470 req->opt2 = htonl(opt2);
471 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
472 }
473
474 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
475 {
476 int mpalen;
477 struct tx_data_wr *req;
478 struct mpa_message *mpa;
479 int len;
480
481 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
482
483 BUG_ON(skb_cloned(skb));
484
485 mpalen = sizeof(*mpa) + ep->plen;
486 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
487 kfree_skb(skb);
488 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
489 if (!skb) {
490 connect_reply_upcall(ep, -ENOMEM);
491 return;
492 }
493 }
494 skb_trim(skb, 0);
495 skb_reserve(skb, sizeof(*req));
496 skb_put(skb, mpalen);
497 skb->priority = CPL_PRIORITY_DATA;
498 mpa = (struct mpa_message *) skb->data;
499 memset(mpa, 0, sizeof(*mpa));
500 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
501 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
502 (markers_enabled ? MPA_MARKERS : 0);
503 mpa->private_data_size = htons(ep->plen);
504 mpa->revision = mpa_rev;
505
506 if (ep->plen)
507 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
508
509 /*
510 * Reference the mpa skb. This ensures the data area
511 * will remain in memory until the hw acks the tx.
512 * Function tx_ack() will deref it.
513 */
514 skb_get(skb);
515 set_arp_failure_handler(skb, arp_failure_discard);
516 skb_reset_transport_header(skb);
517 len = skb->len;
518 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
519 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
520 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
521 req->len = htonl(len);
522 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
523 V_TX_SNDBUF(snd_win>>15));
524 req->flags = htonl(F_TX_INIT);
525 req->sndseq = htonl(ep->snd_seq);
526 BUG_ON(ep->mpa_skb);
527 ep->mpa_skb = skb;
528 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
529 start_ep_timer(ep);
530 state_set(&ep->com, MPA_REQ_SENT);
531 return;
532 }
533
534 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
535 {
536 int mpalen;
537 struct tx_data_wr *req;
538 struct mpa_message *mpa;
539 struct sk_buff *skb;
540
541 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
542
543 mpalen = sizeof(*mpa) + plen;
544
545 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
546 if (!skb) {
547 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
548 return -ENOMEM;
549 }
550 skb_reserve(skb, sizeof(*req));
551 mpa = (struct mpa_message *) skb_put(skb, mpalen);
552 memset(mpa, 0, sizeof(*mpa));
553 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
554 mpa->flags = MPA_REJECT;
555 mpa->revision = mpa_rev;
556 mpa->private_data_size = htons(plen);
557 if (plen)
558 memcpy(mpa->private_data, pdata, plen);
559
560 /*
561 * Reference the mpa skb again. This ensures the data area
562 * will remain in memory until the hw acks the tx.
563 * Function tx_ack() will deref it.
564 */
565 skb_get(skb);
566 skb->priority = CPL_PRIORITY_DATA;
567 set_arp_failure_handler(skb, arp_failure_discard);
568 skb_reset_transport_header(skb);
569 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
570 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
571 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
572 req->len = htonl(mpalen);
573 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
574 V_TX_SNDBUF(snd_win>>15));
575 req->flags = htonl(F_TX_INIT);
576 req->sndseq = htonl(ep->snd_seq);
577 BUG_ON(ep->mpa_skb);
578 ep->mpa_skb = skb;
579 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
580 }
581
582 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
583 {
584 int mpalen;
585 struct tx_data_wr *req;
586 struct mpa_message *mpa;
587 int len;
588 struct sk_buff *skb;
589
590 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
591
592 mpalen = sizeof(*mpa) + plen;
593
594 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
595 if (!skb) {
596 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
597 return -ENOMEM;
598 }
599 skb->priority = CPL_PRIORITY_DATA;
600 skb_reserve(skb, sizeof(*req));
601 mpa = (struct mpa_message *) skb_put(skb, mpalen);
602 memset(mpa, 0, sizeof(*mpa));
603 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
604 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
605 (markers_enabled ? MPA_MARKERS : 0);
606 mpa->revision = mpa_rev;
607 mpa->private_data_size = htons(plen);
608 if (plen)
609 memcpy(mpa->private_data, pdata, plen);
610
611 /*
612 * Reference the mpa skb. This ensures the data area
613 * will remain in memory until the hw acks the tx.
614 * Function tx_ack() will deref it.
615 */
616 skb_get(skb);
617 set_arp_failure_handler(skb, arp_failure_discard);
618 skb_reset_transport_header(skb);
619 len = skb->len;
620 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
621 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
622 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
623 req->len = htonl(len);
624 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
625 V_TX_SNDBUF(snd_win>>15));
626 req->flags = htonl(F_TX_INIT);
627 req->sndseq = htonl(ep->snd_seq);
628 ep->mpa_skb = skb;
629 state_set(&ep->com, MPA_REP_SENT);
630 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
631 }
632
633 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
634 {
635 struct iwch_ep *ep = ctx;
636 struct cpl_act_establish *req = cplhdr(skb);
637 unsigned int tid = GET_TID(req);
638
639 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
640
641 dst_confirm(ep->dst);
642
643 /* setup the hwtid for this connection */
644 ep->hwtid = tid;
645 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
646
647 ep->snd_seq = ntohl(req->snd_isn);
648 ep->rcv_seq = ntohl(req->rcv_isn);
649
650 set_emss(ep, ntohs(req->tcp_opt));
651
652 /* dealloc the atid */
653 cxgb3_free_atid(ep->com.tdev, ep->atid);
654
655 /* start MPA negotiation */
656 send_mpa_req(ep, skb);
657
658 return 0;
659 }
660
661 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
662 {
663 PDBG("%s ep %p\n", __FILE__, ep);
664 state_set(&ep->com, ABORTING);
665 send_abort(ep, skb, gfp);
666 }
667
668 static void close_complete_upcall(struct iwch_ep *ep)
669 {
670 struct iw_cm_event event;
671
672 PDBG("%s ep %p\n", __func__, ep);
673 memset(&event, 0, sizeof(event));
674 event.event = IW_CM_EVENT_CLOSE;
675 if (ep->com.cm_id) {
676 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
677 ep, ep->com.cm_id, ep->hwtid);
678 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
679 ep->com.cm_id->rem_ref(ep->com.cm_id);
680 ep->com.cm_id = NULL;
681 ep->com.qp = NULL;
682 }
683 }
684
685 static void peer_close_upcall(struct iwch_ep *ep)
686 {
687 struct iw_cm_event event;
688
689 PDBG("%s ep %p\n", __func__, ep);
690 memset(&event, 0, sizeof(event));
691 event.event = IW_CM_EVENT_DISCONNECT;
692 if (ep->com.cm_id) {
693 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
694 ep, ep->com.cm_id, ep->hwtid);
695 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
696 }
697 }
698
699 static void peer_abort_upcall(struct iwch_ep *ep)
700 {
701 struct iw_cm_event event;
702
703 PDBG("%s ep %p\n", __func__, ep);
704 memset(&event, 0, sizeof(event));
705 event.event = IW_CM_EVENT_CLOSE;
706 event.status = -ECONNRESET;
707 if (ep->com.cm_id) {
708 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
709 ep->com.cm_id, ep->hwtid);
710 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
711 ep->com.cm_id->rem_ref(ep->com.cm_id);
712 ep->com.cm_id = NULL;
713 ep->com.qp = NULL;
714 }
715 }
716
717 static void connect_reply_upcall(struct iwch_ep *ep, int status)
718 {
719 struct iw_cm_event event;
720
721 PDBG("%s ep %p status %d\n", __func__, ep, status);
722 memset(&event, 0, sizeof(event));
723 event.event = IW_CM_EVENT_CONNECT_REPLY;
724 event.status = status;
725 event.local_addr = ep->com.local_addr;
726 event.remote_addr = ep->com.remote_addr;
727
728 if ((status == 0) || (status == -ECONNREFUSED)) {
729 event.private_data_len = ep->plen;
730 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
731 }
732 if (ep->com.cm_id) {
733 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
734 ep->hwtid, status);
735 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
736 }
737 if (status < 0) {
738 ep->com.cm_id->rem_ref(ep->com.cm_id);
739 ep->com.cm_id = NULL;
740 ep->com.qp = NULL;
741 }
742 }
743
744 static void connect_request_upcall(struct iwch_ep *ep)
745 {
746 struct iw_cm_event event;
747
748 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
749 memset(&event, 0, sizeof(event));
750 event.event = IW_CM_EVENT_CONNECT_REQUEST;
751 event.local_addr = ep->com.local_addr;
752 event.remote_addr = ep->com.remote_addr;
753 event.private_data_len = ep->plen;
754 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
755 event.provider_data = ep;
756 if (state_read(&ep->parent_ep->com) != DEAD) {
757 get_ep(&ep->com);
758 ep->parent_ep->com.cm_id->event_handler(
759 ep->parent_ep->com.cm_id,
760 &event);
761 }
762 put_ep(&ep->parent_ep->com);
763 ep->parent_ep = NULL;
764 }
765
766 static void established_upcall(struct iwch_ep *ep)
767 {
768 struct iw_cm_event event;
769
770 PDBG("%s ep %p\n", __func__, ep);
771 memset(&event, 0, sizeof(event));
772 event.event = IW_CM_EVENT_ESTABLISHED;
773 if (ep->com.cm_id) {
774 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
775 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
776 }
777 }
778
779 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
780 {
781 struct cpl_rx_data_ack *req;
782 struct sk_buff *skb;
783
784 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
785 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
786 if (!skb) {
787 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
788 return 0;
789 }
790
791 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
792 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
793 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
794 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
795 skb->priority = CPL_PRIORITY_ACK;
796 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
797 return credits;
798 }
799
800 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
801 {
802 struct mpa_message *mpa;
803 u16 plen;
804 struct iwch_qp_attributes attrs;
805 enum iwch_qp_attr_mask mask;
806 int err;
807
808 PDBG("%s ep %p\n", __func__, ep);
809
810 /*
811 * Stop mpa timer. If it expired, then the state has
812 * changed and we bail since ep_timeout already aborted
813 * the connection.
814 */
815 stop_ep_timer(ep);
816 if (state_read(&ep->com) != MPA_REQ_SENT)
817 return;
818
819 /*
820 * If we get more than the supported amount of private data
821 * then we must fail this connection.
822 */
823 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
824 err = -EINVAL;
825 goto err;
826 }
827
828 /*
829 * copy the new data into our accumulation buffer.
830 */
831 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
832 skb->len);
833 ep->mpa_pkt_len += skb->len;
834
835 /*
836 * if we don't even have the mpa message, then bail.
837 */
838 if (ep->mpa_pkt_len < sizeof(*mpa))
839 return;
840 mpa = (struct mpa_message *) ep->mpa_pkt;
841
842 /* Validate MPA header. */
843 if (mpa->revision != mpa_rev) {
844 err = -EPROTO;
845 goto err;
846 }
847 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
848 err = -EPROTO;
849 goto err;
850 }
851
852 plen = ntohs(mpa->private_data_size);
853
854 /*
855 * Fail if there's too much private data.
856 */
857 if (plen > MPA_MAX_PRIVATE_DATA) {
858 err = -EPROTO;
859 goto err;
860 }
861
862 /*
863 * If plen does not account for pkt size
864 */
865 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
866 err = -EPROTO;
867 goto err;
868 }
869
870 ep->plen = (u8) plen;
871
872 /*
873 * If we don't have all the pdata yet, then bail.
874 * We'll continue process when more data arrives.
875 */
876 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
877 return;
878
879 if (mpa->flags & MPA_REJECT) {
880 err = -ECONNREFUSED;
881 goto err;
882 }
883
884 /*
885 * If we get here we have accumulated the entire mpa
886 * start reply message including private data. And
887 * the MPA header is valid.
888 */
889 state_set(&ep->com, FPDU_MODE);
890 ep->mpa_attr.initiator = 1;
891 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
892 ep->mpa_attr.recv_marker_enabled = markers_enabled;
893 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
894 ep->mpa_attr.version = mpa_rev;
895 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
896 "xmit_marker_enabled=%d, version=%d\n", __func__,
897 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
898 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
899
900 attrs.mpa_attr = ep->mpa_attr;
901 attrs.max_ird = ep->ird;
902 attrs.max_ord = ep->ord;
903 attrs.llp_stream_handle = ep;
904 attrs.next_state = IWCH_QP_STATE_RTS;
905
906 mask = IWCH_QP_ATTR_NEXT_STATE |
907 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
908 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
909
910 /* bind QP and TID with INIT_WR */
911 err = iwch_modify_qp(ep->com.qp->rhp,
912 ep->com.qp, mask, &attrs, 1);
913 if (err)
914 goto err;
915
916 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
917 iwch_post_zb_read(ep);
918 }
919
920 goto out;
921 err:
922 abort_connection(ep, skb, GFP_KERNEL);
923 out:
924 connect_reply_upcall(ep, err);
925 return;
926 }
927
928 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
929 {
930 struct mpa_message *mpa;
931 u16 plen;
932
933 PDBG("%s ep %p\n", __func__, ep);
934
935 /*
936 * Stop mpa timer. If it expired, then the state has
937 * changed and we bail since ep_timeout already aborted
938 * the connection.
939 */
940 stop_ep_timer(ep);
941 if (state_read(&ep->com) != MPA_REQ_WAIT)
942 return;
943
944 /*
945 * If we get more than the supported amount of private data
946 * then we must fail this connection.
947 */
948 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
949 abort_connection(ep, skb, GFP_KERNEL);
950 return;
951 }
952
953 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
954
955 /*
956 * Copy the new data into our accumulation buffer.
957 */
958 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
959 skb->len);
960 ep->mpa_pkt_len += skb->len;
961
962 /*
963 * If we don't even have the mpa message, then bail.
964 * We'll continue process when more data arrives.
965 */
966 if (ep->mpa_pkt_len < sizeof(*mpa))
967 return;
968 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
969 mpa = (struct mpa_message *) ep->mpa_pkt;
970
971 /*
972 * Validate MPA Header.
973 */
974 if (mpa->revision != mpa_rev) {
975 abort_connection(ep, skb, GFP_KERNEL);
976 return;
977 }
978
979 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
980 abort_connection(ep, skb, GFP_KERNEL);
981 return;
982 }
983
984 plen = ntohs(mpa->private_data_size);
985
986 /*
987 * Fail if there's too much private data.
988 */
989 if (plen > MPA_MAX_PRIVATE_DATA) {
990 abort_connection(ep, skb, GFP_KERNEL);
991 return;
992 }
993
994 /*
995 * If plen does not account for pkt size
996 */
997 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
998 abort_connection(ep, skb, GFP_KERNEL);
999 return;
1000 }
1001 ep->plen = (u8) plen;
1002
1003 /*
1004 * If we don't have all the pdata yet, then bail.
1005 */
1006 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1007 return;
1008
1009 /*
1010 * If we get here we have accumulated the entire mpa
1011 * start reply message including private data.
1012 */
1013 ep->mpa_attr.initiator = 0;
1014 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1015 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1016 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1017 ep->mpa_attr.version = mpa_rev;
1018 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1019 "xmit_marker_enabled=%d, version=%d\n", __func__,
1020 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1021 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1022
1023 state_set(&ep->com, MPA_REQ_RCVD);
1024
1025 /* drive upcall */
1026 connect_request_upcall(ep);
1027 return;
1028 }
1029
1030 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1031 {
1032 struct iwch_ep *ep = ctx;
1033 struct cpl_rx_data *hdr = cplhdr(skb);
1034 unsigned int dlen = ntohs(hdr->len);
1035
1036 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1037
1038 skb_pull(skb, sizeof(*hdr));
1039 skb_trim(skb, dlen);
1040
1041 ep->rcv_seq += dlen;
1042 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1043
1044 switch (state_read(&ep->com)) {
1045 case MPA_REQ_SENT:
1046 process_mpa_reply(ep, skb);
1047 break;
1048 case MPA_REQ_WAIT:
1049 process_mpa_request(ep, skb);
1050 break;
1051 case MPA_REP_SENT:
1052 break;
1053 default:
1054 printk(KERN_ERR MOD "%s Unexpected streaming data."
1055 " ep %p state %d tid %d\n",
1056 __func__, ep, state_read(&ep->com), ep->hwtid);
1057
1058 /*
1059 * The ep will timeout and inform the ULP of the failure.
1060 * See ep_timeout().
1061 */
1062 break;
1063 }
1064
1065 /* update RX credits */
1066 update_rx_credits(ep, dlen);
1067
1068 return CPL_RET_BUF_DONE;
1069 }
1070
1071 /*
1072 * Upcall from the adapter indicating data has been transmitted.
1073 * For us its just the single MPA request or reply. We can now free
1074 * the skb holding the mpa message.
1075 */
1076 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1077 {
1078 struct iwch_ep *ep = ctx;
1079 struct cpl_wr_ack *hdr = cplhdr(skb);
1080 unsigned int credits = ntohs(hdr->credits);
1081 unsigned long flags;
1082 int post_zb = 0;
1083
1084 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1085
1086 if (credits == 0) {
1087 PDBG("%s 0 credit ack ep %p state %u\n",
1088 __func__, ep, state_read(&ep->com));
1089 return CPL_RET_BUF_DONE;
1090 }
1091
1092 spin_lock_irqsave(&ep->com.lock, flags);
1093 BUG_ON(credits != 1);
1094 dst_confirm(ep->dst);
1095 if (!ep->mpa_skb) {
1096 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1097 __func__, ep, ep->com.state);
1098 if (ep->mpa_attr.initiator) {
1099 PDBG("%s initiator ep %p state %u\n",
1100 __func__, ep, ep->com.state);
1101 if (peer2peer && ep->com.state == FPDU_MODE)
1102 post_zb = 1;
1103 } else {
1104 PDBG("%s responder ep %p state %u\n",
1105 __func__, ep, ep->com.state);
1106 if (ep->com.state == MPA_REQ_RCVD) {
1107 ep->com.rpl_done = 1;
1108 wake_up(&ep->com.waitq);
1109 }
1110 }
1111 } else {
1112 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1113 __func__, ep, ep->com.state);
1114 kfree_skb(ep->mpa_skb);
1115 ep->mpa_skb = NULL;
1116 }
1117 spin_unlock_irqrestore(&ep->com.lock, flags);
1118 if (post_zb)
1119 iwch_post_zb_read(ep);
1120 return CPL_RET_BUF_DONE;
1121 }
1122
1123 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1124 {
1125 struct iwch_ep *ep = ctx;
1126 unsigned long flags;
1127 int release = 0;
1128
1129 PDBG("%s ep %p\n", __func__, ep);
1130 BUG_ON(!ep);
1131
1132 /*
1133 * We get 2 abort replies from the HW. The first one must
1134 * be ignored except for scribbling that we need one more.
1135 */
1136 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1137 return CPL_RET_BUF_DONE;
1138 }
1139
1140 spin_lock_irqsave(&ep->com.lock, flags);
1141 switch (ep->com.state) {
1142 case ABORTING:
1143 close_complete_upcall(ep);
1144 __state_set(&ep->com, DEAD);
1145 release = 1;
1146 break;
1147 default:
1148 printk(KERN_ERR "%s ep %p state %d\n",
1149 __func__, ep, ep->com.state);
1150 break;
1151 }
1152 spin_unlock_irqrestore(&ep->com.lock, flags);
1153
1154 if (release)
1155 release_ep_resources(ep);
1156 return CPL_RET_BUF_DONE;
1157 }
1158
1159 /*
1160 * Return whether a failed active open has allocated a TID
1161 */
1162 static inline int act_open_has_tid(int status)
1163 {
1164 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1165 status != CPL_ERR_ARP_MISS;
1166 }
1167
1168 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1169 {
1170 struct iwch_ep *ep = ctx;
1171 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1172
1173 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1174 status2errno(rpl->status));
1175 connect_reply_upcall(ep, status2errno(rpl->status));
1176 state_set(&ep->com, DEAD);
1177 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1178 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1179 cxgb3_free_atid(ep->com.tdev, ep->atid);
1180 dst_release(ep->dst);
1181 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1182 put_ep(&ep->com);
1183 return CPL_RET_BUF_DONE;
1184 }
1185
1186 static int listen_start(struct iwch_listen_ep *ep)
1187 {
1188 struct sk_buff *skb;
1189 struct cpl_pass_open_req *req;
1190
1191 PDBG("%s ep %p\n", __func__, ep);
1192 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1193 if (!skb) {
1194 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1195 return -ENOMEM;
1196 }
1197
1198 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1199 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1200 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1201 req->local_port = ep->com.local_addr.sin_port;
1202 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1203 req->peer_port = 0;
1204 req->peer_ip = 0;
1205 req->peer_netmask = 0;
1206 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1207 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1208 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1209
1210 skb->priority = 1;
1211 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1212 }
1213
1214 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1215 {
1216 struct iwch_listen_ep *ep = ctx;
1217 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1218
1219 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1220 rpl->status, status2errno(rpl->status));
1221 ep->com.rpl_err = status2errno(rpl->status);
1222 ep->com.rpl_done = 1;
1223 wake_up(&ep->com.waitq);
1224
1225 return CPL_RET_BUF_DONE;
1226 }
1227
1228 static int listen_stop(struct iwch_listen_ep *ep)
1229 {
1230 struct sk_buff *skb;
1231 struct cpl_close_listserv_req *req;
1232
1233 PDBG("%s ep %p\n", __func__, ep);
1234 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1235 if (!skb) {
1236 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1237 return -ENOMEM;
1238 }
1239 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1240 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1241 req->cpu_idx = 0;
1242 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1243 skb->priority = 1;
1244 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1245 }
1246
1247 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1248 void *ctx)
1249 {
1250 struct iwch_listen_ep *ep = ctx;
1251 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1252
1253 PDBG("%s ep %p\n", __func__, ep);
1254 ep->com.rpl_err = status2errno(rpl->status);
1255 ep->com.rpl_done = 1;
1256 wake_up(&ep->com.waitq);
1257 return CPL_RET_BUF_DONE;
1258 }
1259
1260 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1261 {
1262 struct cpl_pass_accept_rpl *rpl;
1263 unsigned int mtu_idx;
1264 u32 opt0h, opt0l, opt2;
1265 int wscale;
1266
1267 PDBG("%s ep %p\n", __func__, ep);
1268 BUG_ON(skb_cloned(skb));
1269 skb_trim(skb, sizeof(*rpl));
1270 skb_get(skb);
1271 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1272 wscale = compute_wscale(rcv_win);
1273 opt0h = V_NAGLE(0) |
1274 V_NO_CONG(nocong) |
1275 V_KEEP_ALIVE(1) |
1276 F_TCAM_BYPASS |
1277 V_WND_SCALE(wscale) |
1278 V_MSS_IDX(mtu_idx) |
1279 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1280 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1281 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1282 V_CONG_CONTROL_FLAVOR(cong_flavor);
1283
1284 rpl = cplhdr(skb);
1285 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1286 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1287 rpl->peer_ip = peer_ip;
1288 rpl->opt0h = htonl(opt0h);
1289 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1290 rpl->opt2 = htonl(opt2);
1291 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1292 skb->priority = CPL_PRIORITY_SETUP;
1293 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1294
1295 return;
1296 }
1297
1298 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1299 struct sk_buff *skb)
1300 {
1301 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1302 peer_ip);
1303 BUG_ON(skb_cloned(skb));
1304 skb_trim(skb, sizeof(struct cpl_tid_release));
1305 skb_get(skb);
1306
1307 if (tdev->type != T3A)
1308 release_tid(tdev, hwtid, skb);
1309 else {
1310 struct cpl_pass_accept_rpl *rpl;
1311
1312 rpl = cplhdr(skb);
1313 skb->priority = CPL_PRIORITY_SETUP;
1314 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1315 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1316 hwtid));
1317 rpl->peer_ip = peer_ip;
1318 rpl->opt0h = htonl(F_TCAM_BYPASS);
1319 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1320 rpl->opt2 = 0;
1321 rpl->rsvd = rpl->opt2;
1322 iwch_cxgb3_ofld_send(tdev, skb);
1323 }
1324 }
1325
1326 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1327 {
1328 struct iwch_ep *child_ep, *parent_ep = ctx;
1329 struct cpl_pass_accept_req *req = cplhdr(skb);
1330 unsigned int hwtid = GET_TID(req);
1331 struct neighbour *neigh;
1332 struct dst_entry *dst;
1333 struct l2t_entry *l2t;
1334 struct rtable *rt;
1335 struct iff_mac tim;
1336
1337 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1338
1339 if (state_read(&parent_ep->com) != LISTEN) {
1340 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1341 __func__);
1342 goto reject;
1343 }
1344
1345 /*
1346 * Find the netdev for this connection request.
1347 */
1348 tim.mac_addr = req->dst_mac;
1349 tim.vlan_tag = ntohs(req->vlan_tag);
1350 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1351 printk(KERN_ERR "%s bad dst mac %pM\n",
1352 __func__, req->dst_mac);
1353 goto reject;
1354 }
1355
1356 /* Find output route */
1357 rt = find_route(tdev,
1358 req->local_ip,
1359 req->peer_ip,
1360 req->local_port,
1361 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1362 if (!rt) {
1363 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1364 __func__);
1365 goto reject;
1366 }
1367 dst = &rt->dst;
1368 neigh = dst_get_neighbour(dst);
1369 l2t = t3_l2t_get(tdev, neigh, neigh->dev);
1370 if (!l2t) {
1371 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1372 __func__);
1373 dst_release(dst);
1374 goto reject;
1375 }
1376 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1377 if (!child_ep) {
1378 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1379 __func__);
1380 l2t_release(L2DATA(tdev), l2t);
1381 dst_release(dst);
1382 goto reject;
1383 }
1384 state_set(&child_ep->com, CONNECTING);
1385 child_ep->com.tdev = tdev;
1386 child_ep->com.cm_id = NULL;
1387 child_ep->com.local_addr.sin_family = PF_INET;
1388 child_ep->com.local_addr.sin_port = req->local_port;
1389 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1390 child_ep->com.remote_addr.sin_family = PF_INET;
1391 child_ep->com.remote_addr.sin_port = req->peer_port;
1392 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1393 get_ep(&parent_ep->com);
1394 child_ep->parent_ep = parent_ep;
1395 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1396 child_ep->l2t = l2t;
1397 child_ep->dst = dst;
1398 child_ep->hwtid = hwtid;
1399 init_timer(&child_ep->timer);
1400 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1401 accept_cr(child_ep, req->peer_ip, skb);
1402 goto out;
1403 reject:
1404 reject_cr(tdev, hwtid, req->peer_ip, skb);
1405 out:
1406 return CPL_RET_BUF_DONE;
1407 }
1408
1409 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1410 {
1411 struct iwch_ep *ep = ctx;
1412 struct cpl_pass_establish *req = cplhdr(skb);
1413
1414 PDBG("%s ep %p\n", __func__, ep);
1415 ep->snd_seq = ntohl(req->snd_isn);
1416 ep->rcv_seq = ntohl(req->rcv_isn);
1417
1418 set_emss(ep, ntohs(req->tcp_opt));
1419
1420 dst_confirm(ep->dst);
1421 state_set(&ep->com, MPA_REQ_WAIT);
1422 start_ep_timer(ep);
1423
1424 return CPL_RET_BUF_DONE;
1425 }
1426
1427 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1428 {
1429 struct iwch_ep *ep = ctx;
1430 struct iwch_qp_attributes attrs;
1431 unsigned long flags;
1432 int disconnect = 1;
1433 int release = 0;
1434
1435 PDBG("%s ep %p\n", __func__, ep);
1436 dst_confirm(ep->dst);
1437
1438 spin_lock_irqsave(&ep->com.lock, flags);
1439 switch (ep->com.state) {
1440 case MPA_REQ_WAIT:
1441 __state_set(&ep->com, CLOSING);
1442 break;
1443 case MPA_REQ_SENT:
1444 __state_set(&ep->com, CLOSING);
1445 connect_reply_upcall(ep, -ECONNRESET);
1446 break;
1447 case MPA_REQ_RCVD:
1448
1449 /*
1450 * We're gonna mark this puppy DEAD, but keep
1451 * the reference on it until the ULP accepts or
1452 * rejects the CR. Also wake up anyone waiting
1453 * in rdma connection migration (see iwch_accept_cr()).
1454 */
1455 __state_set(&ep->com, CLOSING);
1456 ep->com.rpl_done = 1;
1457 ep->com.rpl_err = -ECONNRESET;
1458 PDBG("waking up ep %p\n", ep);
1459 wake_up(&ep->com.waitq);
1460 break;
1461 case MPA_REP_SENT:
1462 __state_set(&ep->com, CLOSING);
1463 ep->com.rpl_done = 1;
1464 ep->com.rpl_err = -ECONNRESET;
1465 PDBG("waking up ep %p\n", ep);
1466 wake_up(&ep->com.waitq);
1467 break;
1468 case FPDU_MODE:
1469 start_ep_timer(ep);
1470 __state_set(&ep->com, CLOSING);
1471 attrs.next_state = IWCH_QP_STATE_CLOSING;
1472 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1473 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1474 peer_close_upcall(ep);
1475 break;
1476 case ABORTING:
1477 disconnect = 0;
1478 break;
1479 case CLOSING:
1480 __state_set(&ep->com, MORIBUND);
1481 disconnect = 0;
1482 break;
1483 case MORIBUND:
1484 stop_ep_timer(ep);
1485 if (ep->com.cm_id && ep->com.qp) {
1486 attrs.next_state = IWCH_QP_STATE_IDLE;
1487 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1488 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1489 }
1490 close_complete_upcall(ep);
1491 __state_set(&ep->com, DEAD);
1492 release = 1;
1493 disconnect = 0;
1494 break;
1495 case DEAD:
1496 disconnect = 0;
1497 break;
1498 default:
1499 BUG_ON(1);
1500 }
1501 spin_unlock_irqrestore(&ep->com.lock, flags);
1502 if (disconnect)
1503 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1504 if (release)
1505 release_ep_resources(ep);
1506 return CPL_RET_BUF_DONE;
1507 }
1508
1509 /*
1510 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1511 */
1512 static int is_neg_adv_abort(unsigned int status)
1513 {
1514 return status == CPL_ERR_RTX_NEG_ADVICE ||
1515 status == CPL_ERR_PERSIST_NEG_ADVICE;
1516 }
1517
1518 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1519 {
1520 struct cpl_abort_req_rss *req = cplhdr(skb);
1521 struct iwch_ep *ep = ctx;
1522 struct cpl_abort_rpl *rpl;
1523 struct sk_buff *rpl_skb;
1524 struct iwch_qp_attributes attrs;
1525 int ret;
1526 int release = 0;
1527 unsigned long flags;
1528
1529 if (is_neg_adv_abort(req->status)) {
1530 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1531 ep->hwtid);
1532 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1533 return CPL_RET_BUF_DONE;
1534 }
1535
1536 /*
1537 * We get 2 peer aborts from the HW. The first one must
1538 * be ignored except for scribbling that we need one more.
1539 */
1540 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1541 return CPL_RET_BUF_DONE;
1542 }
1543
1544 spin_lock_irqsave(&ep->com.lock, flags);
1545 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1546 switch (ep->com.state) {
1547 case CONNECTING:
1548 break;
1549 case MPA_REQ_WAIT:
1550 stop_ep_timer(ep);
1551 break;
1552 case MPA_REQ_SENT:
1553 stop_ep_timer(ep);
1554 connect_reply_upcall(ep, -ECONNRESET);
1555 break;
1556 case MPA_REP_SENT:
1557 ep->com.rpl_done = 1;
1558 ep->com.rpl_err = -ECONNRESET;
1559 PDBG("waking up ep %p\n", ep);
1560 wake_up(&ep->com.waitq);
1561 break;
1562 case MPA_REQ_RCVD:
1563
1564 /*
1565 * We're gonna mark this puppy DEAD, but keep
1566 * the reference on it until the ULP accepts or
1567 * rejects the CR. Also wake up anyone waiting
1568 * in rdma connection migration (see iwch_accept_cr()).
1569 */
1570 ep->com.rpl_done = 1;
1571 ep->com.rpl_err = -ECONNRESET;
1572 PDBG("waking up ep %p\n", ep);
1573 wake_up(&ep->com.waitq);
1574 break;
1575 case MORIBUND:
1576 case CLOSING:
1577 stop_ep_timer(ep);
1578 /*FALLTHROUGH*/
1579 case FPDU_MODE:
1580 if (ep->com.cm_id && ep->com.qp) {
1581 attrs.next_state = IWCH_QP_STATE_ERROR;
1582 ret = iwch_modify_qp(ep->com.qp->rhp,
1583 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1584 &attrs, 1);
1585 if (ret)
1586 printk(KERN_ERR MOD
1587 "%s - qp <- error failed!\n",
1588 __func__);
1589 }
1590 peer_abort_upcall(ep);
1591 break;
1592 case ABORTING:
1593 break;
1594 case DEAD:
1595 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1596 spin_unlock_irqrestore(&ep->com.lock, flags);
1597 return CPL_RET_BUF_DONE;
1598 default:
1599 BUG_ON(1);
1600 break;
1601 }
1602 dst_confirm(ep->dst);
1603 if (ep->com.state != ABORTING) {
1604 __state_set(&ep->com, DEAD);
1605 release = 1;
1606 }
1607 spin_unlock_irqrestore(&ep->com.lock, flags);
1608
1609 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1610 if (!rpl_skb) {
1611 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1612 __func__);
1613 release = 1;
1614 goto out;
1615 }
1616 rpl_skb->priority = CPL_PRIORITY_DATA;
1617 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1618 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1619 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1620 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1621 rpl->cmd = CPL_ABORT_NO_RST;
1622 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1623 out:
1624 if (release)
1625 release_ep_resources(ep);
1626 return CPL_RET_BUF_DONE;
1627 }
1628
1629 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1630 {
1631 struct iwch_ep *ep = ctx;
1632 struct iwch_qp_attributes attrs;
1633 unsigned long flags;
1634 int release = 0;
1635
1636 PDBG("%s ep %p\n", __func__, ep);
1637 BUG_ON(!ep);
1638
1639 /* The cm_id may be null if we failed to connect */
1640 spin_lock_irqsave(&ep->com.lock, flags);
1641 switch (ep->com.state) {
1642 case CLOSING:
1643 __state_set(&ep->com, MORIBUND);
1644 break;
1645 case MORIBUND:
1646 stop_ep_timer(ep);
1647 if ((ep->com.cm_id) && (ep->com.qp)) {
1648 attrs.next_state = IWCH_QP_STATE_IDLE;
1649 iwch_modify_qp(ep->com.qp->rhp,
1650 ep->com.qp,
1651 IWCH_QP_ATTR_NEXT_STATE,
1652 &attrs, 1);
1653 }
1654 close_complete_upcall(ep);
1655 __state_set(&ep->com, DEAD);
1656 release = 1;
1657 break;
1658 case ABORTING:
1659 case DEAD:
1660 break;
1661 default:
1662 BUG_ON(1);
1663 break;
1664 }
1665 spin_unlock_irqrestore(&ep->com.lock, flags);
1666 if (release)
1667 release_ep_resources(ep);
1668 return CPL_RET_BUF_DONE;
1669 }
1670
1671 /*
1672 * T3A does 3 things when a TERM is received:
1673 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1674 * 2) generate an async event on the QP with the TERMINATE opcode
1675 * 3) post a TERMINATE opcde cqe into the associated CQ.
1676 *
1677 * For (1), we save the message in the qp for later consumer consumption.
1678 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1679 * For (3), we toss the CQE in cxio_poll_cq().
1680 *
1681 * terminate() handles case (1)...
1682 */
1683 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1684 {
1685 struct iwch_ep *ep = ctx;
1686
1687 if (state_read(&ep->com) != FPDU_MODE)
1688 return CPL_RET_BUF_DONE;
1689
1690 PDBG("%s ep %p\n", __func__, ep);
1691 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1692 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1693 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1694 skb->len);
1695 ep->com.qp->attr.terminate_msg_len = skb->len;
1696 ep->com.qp->attr.is_terminate_local = 0;
1697 return CPL_RET_BUF_DONE;
1698 }
1699
1700 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1701 {
1702 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1703 struct iwch_ep *ep = ctx;
1704
1705 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1706 rep->status);
1707 if (rep->status) {
1708 struct iwch_qp_attributes attrs;
1709
1710 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1711 __func__, ep->hwtid);
1712 stop_ep_timer(ep);
1713 attrs.next_state = IWCH_QP_STATE_ERROR;
1714 iwch_modify_qp(ep->com.qp->rhp,
1715 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1716 &attrs, 1);
1717 abort_connection(ep, NULL, GFP_KERNEL);
1718 }
1719 return CPL_RET_BUF_DONE;
1720 }
1721
1722 static void ep_timeout(unsigned long arg)
1723 {
1724 struct iwch_ep *ep = (struct iwch_ep *)arg;
1725 struct iwch_qp_attributes attrs;
1726 unsigned long flags;
1727 int abort = 1;
1728
1729 spin_lock_irqsave(&ep->com.lock, flags);
1730 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1731 ep->com.state);
1732 switch (ep->com.state) {
1733 case MPA_REQ_SENT:
1734 __state_set(&ep->com, ABORTING);
1735 connect_reply_upcall(ep, -ETIMEDOUT);
1736 break;
1737 case MPA_REQ_WAIT:
1738 __state_set(&ep->com, ABORTING);
1739 break;
1740 case CLOSING:
1741 case MORIBUND:
1742 if (ep->com.cm_id && ep->com.qp) {
1743 attrs.next_state = IWCH_QP_STATE_ERROR;
1744 iwch_modify_qp(ep->com.qp->rhp,
1745 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1746 &attrs, 1);
1747 }
1748 __state_set(&ep->com, ABORTING);
1749 break;
1750 default:
1751 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1752 __func__, ep, ep->com.state);
1753 WARN_ON(1);
1754 abort = 0;
1755 }
1756 spin_unlock_irqrestore(&ep->com.lock, flags);
1757 if (abort)
1758 abort_connection(ep, NULL, GFP_ATOMIC);
1759 put_ep(&ep->com);
1760 }
1761
1762 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1763 {
1764 int err;
1765 struct iwch_ep *ep = to_ep(cm_id);
1766 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1767
1768 if (state_read(&ep->com) == DEAD) {
1769 put_ep(&ep->com);
1770 return -ECONNRESET;
1771 }
1772 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1773 if (mpa_rev == 0)
1774 abort_connection(ep, NULL, GFP_KERNEL);
1775 else {
1776 err = send_mpa_reject(ep, pdata, pdata_len);
1777 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1778 }
1779 put_ep(&ep->com);
1780 return 0;
1781 }
1782
1783 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1784 {
1785 int err;
1786 struct iwch_qp_attributes attrs;
1787 enum iwch_qp_attr_mask mask;
1788 struct iwch_ep *ep = to_ep(cm_id);
1789 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1790 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1791
1792 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1793 if (state_read(&ep->com) == DEAD) {
1794 err = -ECONNRESET;
1795 goto err;
1796 }
1797
1798 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1799 BUG_ON(!qp);
1800
1801 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1802 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1803 abort_connection(ep, NULL, GFP_KERNEL);
1804 err = -EINVAL;
1805 goto err;
1806 }
1807
1808 cm_id->add_ref(cm_id);
1809 ep->com.cm_id = cm_id;
1810 ep->com.qp = qp;
1811
1812 ep->ird = conn_param->ird;
1813 ep->ord = conn_param->ord;
1814
1815 if (peer2peer && ep->ird == 0)
1816 ep->ird = 1;
1817
1818 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1819
1820 /* bind QP to EP and move to RTS */
1821 attrs.mpa_attr = ep->mpa_attr;
1822 attrs.max_ird = ep->ird;
1823 attrs.max_ord = ep->ord;
1824 attrs.llp_stream_handle = ep;
1825 attrs.next_state = IWCH_QP_STATE_RTS;
1826
1827 /* bind QP and TID with INIT_WR */
1828 mask = IWCH_QP_ATTR_NEXT_STATE |
1829 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1830 IWCH_QP_ATTR_MPA_ATTR |
1831 IWCH_QP_ATTR_MAX_IRD |
1832 IWCH_QP_ATTR_MAX_ORD;
1833
1834 err = iwch_modify_qp(ep->com.qp->rhp,
1835 ep->com.qp, mask, &attrs, 1);
1836 if (err)
1837 goto err1;
1838
1839 /* if needed, wait for wr_ack */
1840 if (iwch_rqes_posted(qp)) {
1841 wait_event(ep->com.waitq, ep->com.rpl_done);
1842 err = ep->com.rpl_err;
1843 if (err)
1844 goto err1;
1845 }
1846
1847 err = send_mpa_reply(ep, conn_param->private_data,
1848 conn_param->private_data_len);
1849 if (err)
1850 goto err1;
1851
1852
1853 state_set(&ep->com, FPDU_MODE);
1854 established_upcall(ep);
1855 put_ep(&ep->com);
1856 return 0;
1857 err1:
1858 ep->com.cm_id = NULL;
1859 ep->com.qp = NULL;
1860 cm_id->rem_ref(cm_id);
1861 err:
1862 put_ep(&ep->com);
1863 return err;
1864 }
1865
1866 static int is_loopback_dst(struct iw_cm_id *cm_id)
1867 {
1868 struct net_device *dev;
1869
1870 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1871 if (!dev)
1872 return 0;
1873 dev_put(dev);
1874 return 1;
1875 }
1876
1877 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1878 {
1879 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1880 struct neighbour *neigh;
1881 struct iwch_ep *ep;
1882 struct rtable *rt;
1883 int err = 0;
1884
1885 if (is_loopback_dst(cm_id)) {
1886 err = -ENOSYS;
1887 goto out;
1888 }
1889
1890 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1891 if (!ep) {
1892 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1893 err = -ENOMEM;
1894 goto out;
1895 }
1896 init_timer(&ep->timer);
1897 ep->plen = conn_param->private_data_len;
1898 if (ep->plen)
1899 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1900 conn_param->private_data, ep->plen);
1901 ep->ird = conn_param->ird;
1902 ep->ord = conn_param->ord;
1903
1904 if (peer2peer && ep->ord == 0)
1905 ep->ord = 1;
1906
1907 ep->com.tdev = h->rdev.t3cdev_p;
1908
1909 cm_id->add_ref(cm_id);
1910 ep->com.cm_id = cm_id;
1911 ep->com.qp = get_qhp(h, conn_param->qpn);
1912 BUG_ON(!ep->com.qp);
1913 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1914 ep->com.qp, cm_id);
1915
1916 /*
1917 * Allocate an active TID to initiate a TCP connection.
1918 */
1919 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1920 if (ep->atid == -1) {
1921 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1922 err = -ENOMEM;
1923 goto fail2;
1924 }
1925
1926 /* find a route */
1927 rt = find_route(h->rdev.t3cdev_p,
1928 cm_id->local_addr.sin_addr.s_addr,
1929 cm_id->remote_addr.sin_addr.s_addr,
1930 cm_id->local_addr.sin_port,
1931 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1932 if (!rt) {
1933 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1934 err = -EHOSTUNREACH;
1935 goto fail3;
1936 }
1937 ep->dst = &rt->dst;
1938
1939 neigh = dst_get_neighbour(ep->dst);
1940
1941 /* get a l2t entry */
1942 ep->l2t = t3_l2t_get(ep->com.tdev, neigh, neigh->dev);
1943 if (!ep->l2t) {
1944 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1945 err = -ENOMEM;
1946 goto fail4;
1947 }
1948
1949 state_set(&ep->com, CONNECTING);
1950 ep->tos = IPTOS_LOWDELAY;
1951 ep->com.local_addr = cm_id->local_addr;
1952 ep->com.remote_addr = cm_id->remote_addr;
1953
1954 /* send connect request to rnic */
1955 err = send_connect(ep);
1956 if (!err)
1957 goto out;
1958
1959 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1960 fail4:
1961 dst_release(ep->dst);
1962 fail3:
1963 cxgb3_free_atid(ep->com.tdev, ep->atid);
1964 fail2:
1965 cm_id->rem_ref(cm_id);
1966 put_ep(&ep->com);
1967 out:
1968 return err;
1969 }
1970
1971 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1972 {
1973 int err = 0;
1974 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1975 struct iwch_listen_ep *ep;
1976
1977
1978 might_sleep();
1979
1980 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1981 if (!ep) {
1982 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1983 err = -ENOMEM;
1984 goto fail1;
1985 }
1986 PDBG("%s ep %p\n", __func__, ep);
1987 ep->com.tdev = h->rdev.t3cdev_p;
1988 cm_id->add_ref(cm_id);
1989 ep->com.cm_id = cm_id;
1990 ep->backlog = backlog;
1991 ep->com.local_addr = cm_id->local_addr;
1992
1993 /*
1994 * Allocate a server TID.
1995 */
1996 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1997 if (ep->stid == -1) {
1998 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1999 err = -ENOMEM;
2000 goto fail2;
2001 }
2002
2003 state_set(&ep->com, LISTEN);
2004 err = listen_start(ep);
2005 if (err)
2006 goto fail3;
2007
2008 /* wait for pass_open_rpl */
2009 wait_event(ep->com.waitq, ep->com.rpl_done);
2010 err = ep->com.rpl_err;
2011 if (!err) {
2012 cm_id->provider_data = ep;
2013 goto out;
2014 }
2015 fail3:
2016 cxgb3_free_stid(ep->com.tdev, ep->stid);
2017 fail2:
2018 cm_id->rem_ref(cm_id);
2019 put_ep(&ep->com);
2020 fail1:
2021 out:
2022 return err;
2023 }
2024
2025 int iwch_destroy_listen(struct iw_cm_id *cm_id)
2026 {
2027 int err;
2028 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2029
2030 PDBG("%s ep %p\n", __func__, ep);
2031
2032 might_sleep();
2033 state_set(&ep->com, DEAD);
2034 ep->com.rpl_done = 0;
2035 ep->com.rpl_err = 0;
2036 err = listen_stop(ep);
2037 if (err)
2038 goto done;
2039 wait_event(ep->com.waitq, ep->com.rpl_done);
2040 cxgb3_free_stid(ep->com.tdev, ep->stid);
2041 done:
2042 err = ep->com.rpl_err;
2043 cm_id->rem_ref(cm_id);
2044 put_ep(&ep->com);
2045 return err;
2046 }
2047
2048 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2049 {
2050 int ret=0;
2051 unsigned long flags;
2052 int close = 0;
2053 int fatal = 0;
2054 struct t3cdev *tdev;
2055 struct cxio_rdev *rdev;
2056
2057 spin_lock_irqsave(&ep->com.lock, flags);
2058
2059 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2060 states[ep->com.state], abrupt);
2061
2062 tdev = (struct t3cdev *)ep->com.tdev;
2063 rdev = (struct cxio_rdev *)tdev->ulp;
2064 if (cxio_fatal_error(rdev)) {
2065 fatal = 1;
2066 close_complete_upcall(ep);
2067 ep->com.state = DEAD;
2068 }
2069 switch (ep->com.state) {
2070 case MPA_REQ_WAIT:
2071 case MPA_REQ_SENT:
2072 case MPA_REQ_RCVD:
2073 case MPA_REP_SENT:
2074 case FPDU_MODE:
2075 close = 1;
2076 if (abrupt)
2077 ep->com.state = ABORTING;
2078 else {
2079 ep->com.state = CLOSING;
2080 start_ep_timer(ep);
2081 }
2082 set_bit(CLOSE_SENT, &ep->com.flags);
2083 break;
2084 case CLOSING:
2085 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2086 close = 1;
2087 if (abrupt) {
2088 stop_ep_timer(ep);
2089 ep->com.state = ABORTING;
2090 } else
2091 ep->com.state = MORIBUND;
2092 }
2093 break;
2094 case MORIBUND:
2095 case ABORTING:
2096 case DEAD:
2097 PDBG("%s ignoring disconnect ep %p state %u\n",
2098 __func__, ep, ep->com.state);
2099 break;
2100 default:
2101 BUG();
2102 break;
2103 }
2104
2105 spin_unlock_irqrestore(&ep->com.lock, flags);
2106 if (close) {
2107 if (abrupt)
2108 ret = send_abort(ep, NULL, gfp);
2109 else
2110 ret = send_halfclose(ep, gfp);
2111 if (ret)
2112 fatal = 1;
2113 }
2114 if (fatal)
2115 release_ep_resources(ep);
2116 return ret;
2117 }
2118
2119 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2120 struct l2t_entry *l2t)
2121 {
2122 struct iwch_ep *ep = ctx;
2123
2124 if (ep->dst != old)
2125 return 0;
2126
2127 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2128 l2t);
2129 dst_hold(new);
2130 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2131 ep->l2t = l2t;
2132 dst_release(old);
2133 ep->dst = new;
2134 return 1;
2135 }
2136
2137 /*
2138 * All the CM events are handled on a work queue to have a safe context.
2139 * These are the real handlers that are called from the work queue.
2140 */
2141 static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2142 [CPL_ACT_ESTABLISH] = act_establish,
2143 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2144 [CPL_RX_DATA] = rx_data,
2145 [CPL_TX_DMA_ACK] = tx_ack,
2146 [CPL_ABORT_RPL_RSS] = abort_rpl,
2147 [CPL_ABORT_RPL] = abort_rpl,
2148 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2149 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2150 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2151 [CPL_PASS_ESTABLISH] = pass_establish,
2152 [CPL_PEER_CLOSE] = peer_close,
2153 [CPL_ABORT_REQ_RSS] = peer_abort,
2154 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2155 [CPL_RDMA_TERMINATE] = terminate,
2156 [CPL_RDMA_EC_STATUS] = ec_status,
2157 };
2158
2159 static void process_work(struct work_struct *work)
2160 {
2161 struct sk_buff *skb = NULL;
2162 void *ep;
2163 struct t3cdev *tdev;
2164 int ret;
2165
2166 while ((skb = skb_dequeue(&rxq))) {
2167 ep = *((void **) (skb->cb));
2168 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2169 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2170 if (ret & CPL_RET_BUF_DONE)
2171 kfree_skb(skb);
2172
2173 /*
2174 * ep was referenced in sched(), and is freed here.
2175 */
2176 put_ep((struct iwch_ep_common *)ep);
2177 }
2178 }
2179
2180 static DECLARE_WORK(skb_work, process_work);
2181
2182 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2183 {
2184 struct iwch_ep_common *epc = ctx;
2185
2186 get_ep(epc);
2187
2188 /*
2189 * Save ctx and tdev in the skb->cb area.
2190 */
2191 *((void **) skb->cb) = ctx;
2192 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2193
2194 /*
2195 * Queue the skb and schedule the worker thread.
2196 */
2197 skb_queue_tail(&rxq, skb);
2198 queue_work(workq, &skb_work);
2199 return 0;
2200 }
2201
2202 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2203 {
2204 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2205
2206 if (rpl->status != CPL_ERR_NONE) {
2207 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2208 "for tid %u\n", rpl->status, GET_TID(rpl));
2209 }
2210 return CPL_RET_BUF_DONE;
2211 }
2212
2213 /*
2214 * All upcalls from the T3 Core go to sched() to schedule the
2215 * processing on a work queue.
2216 */
2217 cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2218 [CPL_ACT_ESTABLISH] = sched,
2219 [CPL_ACT_OPEN_RPL] = sched,
2220 [CPL_RX_DATA] = sched,
2221 [CPL_TX_DMA_ACK] = sched,
2222 [CPL_ABORT_RPL_RSS] = sched,
2223 [CPL_ABORT_RPL] = sched,
2224 [CPL_PASS_OPEN_RPL] = sched,
2225 [CPL_CLOSE_LISTSRV_RPL] = sched,
2226 [CPL_PASS_ACCEPT_REQ] = sched,
2227 [CPL_PASS_ESTABLISH] = sched,
2228 [CPL_PEER_CLOSE] = sched,
2229 [CPL_CLOSE_CON_RPL] = sched,
2230 [CPL_ABORT_REQ_RSS] = sched,
2231 [CPL_RDMA_TERMINATE] = sched,
2232 [CPL_RDMA_EC_STATUS] = sched,
2233 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2234 };
2235
2236 int __init iwch_cm_init(void)
2237 {
2238 skb_queue_head_init(&rxq);
2239
2240 workq = create_singlethread_workqueue("iw_cxgb3");
2241 if (!workq)
2242 return -ENOMEM;
2243
2244 return 0;
2245 }
2246
2247 void __exit iwch_cm_term(void)
2248 {
2249 flush_workqueue(workq);
2250 destroy_workqueue(workq);
2251 }
linux-next