?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
PM / sleep: Asynchronous threads for suspend_noirq
[linux/fpc-iii.git] / drivers / infiniband / hw / cxgb3 / iwch_cm.c
blob095bb046e2c82eb8ded011eba72a486b376b444f
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 WARN(1, "%s timer stopped when its not running! ep %p state %u\n",
132 __func__, ep, ep->com.state);
133 return;
134 }
135 del_timer_sync(&ep->timer);
136 put_ep(&ep->com);
137 }
138
139 static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
140 {
141 int error = 0;
142 struct cxio_rdev *rdev;
143
144 rdev = (struct cxio_rdev *)tdev->ulp;
145 if (cxio_fatal_error(rdev)) {
146 kfree_skb(skb);
147 return -EIO;
148 }
149 error = l2t_send(tdev, skb, l2e);
150 if (error < 0)
151 kfree_skb(skb);
152 return error;
153 }
154
155 int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
156 {
157 int error = 0;
158 struct cxio_rdev *rdev;
159
160 rdev = (struct cxio_rdev *)tdev->ulp;
161 if (cxio_fatal_error(rdev)) {
162 kfree_skb(skb);
163 return -EIO;
164 }
165 error = cxgb3_ofld_send(tdev, skb);
166 if (error < 0)
167 kfree_skb(skb);
168 return error;
169 }
170
171 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
172 {
173 struct cpl_tid_release *req;
174
175 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
176 if (!skb)
177 return;
178 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
179 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
180 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
181 skb->priority = CPL_PRIORITY_SETUP;
182 iwch_cxgb3_ofld_send(tdev, skb);
183 return;
184 }
185
186 int iwch_quiesce_tid(struct iwch_ep *ep)
187 {
188 struct cpl_set_tcb_field *req;
189 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
190
191 if (!skb)
192 return -ENOMEM;
193 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
194 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
195 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
196 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
197 req->reply = 0;
198 req->cpu_idx = 0;
199 req->word = htons(W_TCB_RX_QUIESCE);
200 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
201 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
202
203 skb->priority = CPL_PRIORITY_DATA;
204 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
205 }
206
207 int iwch_resume_tid(struct iwch_ep *ep)
208 {
209 struct cpl_set_tcb_field *req;
210 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
211
212 if (!skb)
213 return -ENOMEM;
214 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
215 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
216 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
217 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
218 req->reply = 0;
219 req->cpu_idx = 0;
220 req->word = htons(W_TCB_RX_QUIESCE);
221 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
222 req->val = 0;
223
224 skb->priority = CPL_PRIORITY_DATA;
225 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
226 }
227
228 static void set_emss(struct iwch_ep *ep, u16 opt)
229 {
230 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
231 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
232 if (G_TCPOPT_TSTAMP(opt))
233 ep->emss -= 12;
234 if (ep->emss < 128)
235 ep->emss = 128;
236 PDBG("emss=%d\n", ep->emss);
237 }
238
239 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
240 {
241 unsigned long flags;
242 enum iwch_ep_state state;
243
244 spin_lock_irqsave(&epc->lock, flags);
245 state = epc->state;
246 spin_unlock_irqrestore(&epc->lock, flags);
247 return state;
248 }
249
250 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
251 {
252 epc->state = new;
253 }
254
255 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
256 {
257 unsigned long flags;
258
259 spin_lock_irqsave(&epc->lock, flags);
260 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
261 __state_set(epc, new);
262 spin_unlock_irqrestore(&epc->lock, flags);
263 return;
264 }
265
266 static void *alloc_ep(int size, gfp_t gfp)
267 {
268 struct iwch_ep_common *epc;
269
270 epc = kzalloc(size, gfp);
271 if (epc) {
272 kref_init(&epc->kref);
273 spin_lock_init(&epc->lock);
274 init_waitqueue_head(&epc->waitq);
275 }
276 PDBG("%s alloc ep %p\n", __func__, epc);
277 return epc;
278 }
279
280 void __free_ep(struct kref *kref)
281 {
282 struct iwch_ep *ep;
283 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
284 struct iwch_ep, com);
285 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
286 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
287 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
288 dst_release(ep->dst);
289 l2t_release(ep->com.tdev, ep->l2t);
290 }
291 kfree(ep);
292 }
293
294 static void release_ep_resources(struct iwch_ep *ep)
295 {
296 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
297 set_bit(RELEASE_RESOURCES, &ep->com.flags);
298 put_ep(&ep->com);
299 }
300
301 static int status2errno(int status)
302 {
303 switch (status) {
304 case CPL_ERR_NONE:
305 return 0;
306 case CPL_ERR_CONN_RESET:
307 return -ECONNRESET;
308 case CPL_ERR_ARP_MISS:
309 return -EHOSTUNREACH;
310 case CPL_ERR_CONN_TIMEDOUT:
311 return -ETIMEDOUT;
312 case CPL_ERR_TCAM_FULL:
313 return -ENOMEM;
314 case CPL_ERR_CONN_EXIST:
315 return -EADDRINUSE;
316 default:
317 return -EIO;
318 }
319 }
320
321 /*
322 * Try and reuse skbs already allocated...
323 */
324 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
325 {
326 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
327 skb_trim(skb, 0);
328 skb_get(skb);
329 } else {
330 skb = alloc_skb(len, gfp);
331 }
332 return skb;
333 }
334
335 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
336 __be32 peer_ip, __be16 local_port,
337 __be16 peer_port, u8 tos)
338 {
339 struct rtable *rt;
340 struct flowi4 fl4;
341
342 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
343 peer_port, local_port, IPPROTO_TCP,
344 tos, 0);
345 if (IS_ERR(rt))
346 return NULL;
347 return rt;
348 }
349
350 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
351 {
352 int i = 0;
353
354 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
355 ++i;
356 return i;
357 }
358
359 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
360 {
361 PDBG("%s t3cdev %p\n", __func__, dev);
362 kfree_skb(skb);
363 }
364
365 /*
366 * Handle an ARP failure for an active open.
367 */
368 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
369 {
370 printk(KERN_ERR MOD "ARP failure duing connect\n");
371 kfree_skb(skb);
372 }
373
374 /*
375 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
376 * and send it along.
377 */
378 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
379 {
380 struct cpl_abort_req *req = cplhdr(skb);
381
382 PDBG("%s t3cdev %p\n", __func__, dev);
383 req->cmd = CPL_ABORT_NO_RST;
384 iwch_cxgb3_ofld_send(dev, skb);
385 }
386
387 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
388 {
389 struct cpl_close_con_req *req;
390 struct sk_buff *skb;
391
392 PDBG("%s ep %p\n", __func__, ep);
393 skb = get_skb(NULL, sizeof(*req), gfp);
394 if (!skb) {
395 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
396 return -ENOMEM;
397 }
398 skb->priority = CPL_PRIORITY_DATA;
399 set_arp_failure_handler(skb, arp_failure_discard);
400 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
401 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
402 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
403 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
404 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
405 }
406
407 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
408 {
409 struct cpl_abort_req *req;
410
411 PDBG("%s ep %p\n", __func__, ep);
412 skb = get_skb(skb, sizeof(*req), gfp);
413 if (!skb) {
414 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
415 __func__);
416 return -ENOMEM;
417 }
418 skb->priority = CPL_PRIORITY_DATA;
419 set_arp_failure_handler(skb, abort_arp_failure);
420 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
421 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
422 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
423 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
424 req->cmd = CPL_ABORT_SEND_RST;
425 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
426 }
427
428 static int send_connect(struct iwch_ep *ep)
429 {
430 struct cpl_act_open_req *req;
431 struct sk_buff *skb;
432 u32 opt0h, opt0l, opt2;
433 unsigned int mtu_idx;
434 int wscale;
435
436 PDBG("%s ep %p\n", __func__, ep);
437
438 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
439 if (!skb) {
440 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
441 __func__);
442 return -ENOMEM;
443 }
444 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
445 wscale = compute_wscale(rcv_win);
446 opt0h = V_NAGLE(0) |
447 V_NO_CONG(nocong) |
448 V_KEEP_ALIVE(1) |
449 F_TCAM_BYPASS |
450 V_WND_SCALE(wscale) |
451 V_MSS_IDX(mtu_idx) |
452 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
453 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
454 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
455 V_CONG_CONTROL_FLAVOR(cong_flavor);
456 skb->priority = CPL_PRIORITY_SETUP;
457 set_arp_failure_handler(skb, act_open_req_arp_failure);
458
459 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
460 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
461 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
462 req->local_port = ep->com.local_addr.sin_port;
463 req->peer_port = ep->com.remote_addr.sin_port;
464 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
465 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
466 req->opt0h = htonl(opt0h);
467 req->opt0l = htonl(opt0l);
468 req->params = 0;
469 req->opt2 = htonl(opt2);
470 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
471 }
472
473 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
474 {
475 int mpalen;
476 struct tx_data_wr *req;
477 struct mpa_message *mpa;
478 int len;
479
480 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
481
482 BUG_ON(skb_cloned(skb));
483
484 mpalen = sizeof(*mpa) + ep->plen;
485 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
486 kfree_skb(skb);
487 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
488 if (!skb) {
489 connect_reply_upcall(ep, -ENOMEM);
490 return;
491 }
492 }
493 skb_trim(skb, 0);
494 skb_reserve(skb, sizeof(*req));
495 skb_put(skb, mpalen);
496 skb->priority = CPL_PRIORITY_DATA;
497 mpa = (struct mpa_message *) skb->data;
498 memset(mpa, 0, sizeof(*mpa));
499 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
500 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
501 (markers_enabled ? MPA_MARKERS : 0);
502 mpa->private_data_size = htons(ep->plen);
503 mpa->revision = mpa_rev;
504
505 if (ep->plen)
506 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
507
508 /*
509 * Reference the mpa skb. This ensures the data area
510 * will remain in memory until the hw acks the tx.
511 * Function tx_ack() will deref it.
512 */
513 skb_get(skb);
514 set_arp_failure_handler(skb, arp_failure_discard);
515 skb_reset_transport_header(skb);
516 len = skb->len;
517 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
518 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
519 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
520 req->len = htonl(len);
521 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
522 V_TX_SNDBUF(snd_win>>15));
523 req->flags = htonl(F_TX_INIT);
524 req->sndseq = htonl(ep->snd_seq);
525 BUG_ON(ep->mpa_skb);
526 ep->mpa_skb = skb;
527 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
528 start_ep_timer(ep);
529 state_set(&ep->com, MPA_REQ_SENT);
530 return;
531 }
532
533 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
534 {
535 int mpalen;
536 struct tx_data_wr *req;
537 struct mpa_message *mpa;
538 struct sk_buff *skb;
539
540 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
541
542 mpalen = sizeof(*mpa) + plen;
543
544 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
545 if (!skb) {
546 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
547 return -ENOMEM;
548 }
549 skb_reserve(skb, sizeof(*req));
550 mpa = (struct mpa_message *) skb_put(skb, mpalen);
551 memset(mpa, 0, sizeof(*mpa));
552 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
553 mpa->flags = MPA_REJECT;
554 mpa->revision = mpa_rev;
555 mpa->private_data_size = htons(plen);
556 if (plen)
557 memcpy(mpa->private_data, pdata, plen);
558
559 /*
560 * Reference the mpa skb again. This ensures the data area
561 * will remain in memory until the hw acks the tx.
562 * Function tx_ack() will deref it.
563 */
564 skb_get(skb);
565 skb->priority = CPL_PRIORITY_DATA;
566 set_arp_failure_handler(skb, arp_failure_discard);
567 skb_reset_transport_header(skb);
568 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
569 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
570 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
571 req->len = htonl(mpalen);
572 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
573 V_TX_SNDBUF(snd_win>>15));
574 req->flags = htonl(F_TX_INIT);
575 req->sndseq = htonl(ep->snd_seq);
576 BUG_ON(ep->mpa_skb);
577 ep->mpa_skb = skb;
578 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
579 }
580
581 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
582 {
583 int mpalen;
584 struct tx_data_wr *req;
585 struct mpa_message *mpa;
586 int len;
587 struct sk_buff *skb;
588
589 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
590
591 mpalen = sizeof(*mpa) + plen;
592
593 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
594 if (!skb) {
595 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
596 return -ENOMEM;
597 }
598 skb->priority = CPL_PRIORITY_DATA;
599 skb_reserve(skb, sizeof(*req));
600 mpa = (struct mpa_message *) skb_put(skb, mpalen);
601 memset(mpa, 0, sizeof(*mpa));
602 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
603 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
604 (markers_enabled ? MPA_MARKERS : 0);
605 mpa->revision = mpa_rev;
606 mpa->private_data_size = htons(plen);
607 if (plen)
608 memcpy(mpa->private_data, pdata, plen);
609
610 /*
611 * Reference the mpa skb. This ensures the data area
612 * will remain in memory until the hw acks the tx.
613 * Function tx_ack() will deref it.
614 */
615 skb_get(skb);
616 set_arp_failure_handler(skb, arp_failure_discard);
617 skb_reset_transport_header(skb);
618 len = skb->len;
619 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
620 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
621 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
622 req->len = htonl(len);
623 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
624 V_TX_SNDBUF(snd_win>>15));
625 req->flags = htonl(F_TX_INIT);
626 req->sndseq = htonl(ep->snd_seq);
627 ep->mpa_skb = skb;
628 state_set(&ep->com, MPA_REP_SENT);
629 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
630 }
631
632 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
633 {
634 struct iwch_ep *ep = ctx;
635 struct cpl_act_establish *req = cplhdr(skb);
636 unsigned int tid = GET_TID(req);
637
638 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
639
640 dst_confirm(ep->dst);
641
642 /* setup the hwtid for this connection */
643 ep->hwtid = tid;
644 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
645
646 ep->snd_seq = ntohl(req->snd_isn);
647 ep->rcv_seq = ntohl(req->rcv_isn);
648
649 set_emss(ep, ntohs(req->tcp_opt));
650
651 /* dealloc the atid */
652 cxgb3_free_atid(ep->com.tdev, ep->atid);
653
654 /* start MPA negotiation */
655 send_mpa_req(ep, skb);
656
657 return 0;
658 }
659
660 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
661 {
662 PDBG("%s ep %p\n", __FILE__, ep);
663 state_set(&ep->com, ABORTING);
664 send_abort(ep, skb, gfp);
665 }
666
667 static void close_complete_upcall(struct iwch_ep *ep)
668 {
669 struct iw_cm_event event;
670
671 PDBG("%s ep %p\n", __func__, ep);
672 memset(&event, 0, sizeof(event));
673 event.event = IW_CM_EVENT_CLOSE;
674 if (ep->com.cm_id) {
675 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
676 ep, ep->com.cm_id, ep->hwtid);
677 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
678 ep->com.cm_id->rem_ref(ep->com.cm_id);
679 ep->com.cm_id = NULL;
680 ep->com.qp = NULL;
681 }
682 }
683
684 static void peer_close_upcall(struct iwch_ep *ep)
685 {
686 struct iw_cm_event event;
687
688 PDBG("%s ep %p\n", __func__, ep);
689 memset(&event, 0, sizeof(event));
690 event.event = IW_CM_EVENT_DISCONNECT;
691 if (ep->com.cm_id) {
692 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
693 ep, ep->com.cm_id, ep->hwtid);
694 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
695 }
696 }
697
698 static void peer_abort_upcall(struct iwch_ep *ep)
699 {
700 struct iw_cm_event event;
701
702 PDBG("%s ep %p\n", __func__, ep);
703 memset(&event, 0, sizeof(event));
704 event.event = IW_CM_EVENT_CLOSE;
705 event.status = -ECONNRESET;
706 if (ep->com.cm_id) {
707 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
708 ep->com.cm_id, ep->hwtid);
709 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
710 ep->com.cm_id->rem_ref(ep->com.cm_id);
711 ep->com.cm_id = NULL;
712 ep->com.qp = NULL;
713 }
714 }
715
716 static void connect_reply_upcall(struct iwch_ep *ep, int status)
717 {
718 struct iw_cm_event event;
719
720 PDBG("%s ep %p status %d\n", __func__, ep, status);
721 memset(&event, 0, sizeof(event));
722 event.event = IW_CM_EVENT_CONNECT_REPLY;
723 event.status = status;
724 memcpy(&event.local_addr, &ep->com.local_addr,
725 sizeof(ep->com.local_addr));
726 memcpy(&event.remote_addr, &ep->com.remote_addr,
727 sizeof(ep->com.remote_addr));
728
729 if ((status == 0) || (status == -ECONNREFUSED)) {
730 event.private_data_len = ep->plen;
731 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
732 }
733 if (ep->com.cm_id) {
734 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
735 ep->hwtid, status);
736 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
737 }
738 if (status < 0) {
739 ep->com.cm_id->rem_ref(ep->com.cm_id);
740 ep->com.cm_id = NULL;
741 ep->com.qp = NULL;
742 }
743 }
744
745 static void connect_request_upcall(struct iwch_ep *ep)
746 {
747 struct iw_cm_event event;
748
749 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
750 memset(&event, 0, sizeof(event));
751 event.event = IW_CM_EVENT_CONNECT_REQUEST;
752 memcpy(&event.local_addr, &ep->com.local_addr,
753 sizeof(ep->com.local_addr));
754 memcpy(&event.remote_addr, &ep->com.remote_addr,
755 sizeof(ep->com.local_addr));
756 event.private_data_len = ep->plen;
757 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
758 event.provider_data = ep;
759 /*
760 * Until ird/ord negotiation via MPAv2 support is added, send max
761 * supported values
762 */
763 event.ird = event.ord = 8;
764 if (state_read(&ep->parent_ep->com) != DEAD) {
765 get_ep(&ep->com);
766 ep->parent_ep->com.cm_id->event_handler(
767 ep->parent_ep->com.cm_id,
768 &event);
769 }
770 put_ep(&ep->parent_ep->com);
771 ep->parent_ep = NULL;
772 }
773
774 static void established_upcall(struct iwch_ep *ep)
775 {
776 struct iw_cm_event event;
777
778 PDBG("%s ep %p\n", __func__, ep);
779 memset(&event, 0, sizeof(event));
780 event.event = IW_CM_EVENT_ESTABLISHED;
781 /*
782 * Until ird/ord negotiation via MPAv2 support is added, send max
783 * supported values
784 */
785 event.ird = event.ord = 8;
786 if (ep->com.cm_id) {
787 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
788 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
789 }
790 }
791
792 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
793 {
794 struct cpl_rx_data_ack *req;
795 struct sk_buff *skb;
796
797 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
798 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
799 if (!skb) {
800 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
801 return 0;
802 }
803
804 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
805 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
806 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
807 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
808 skb->priority = CPL_PRIORITY_ACK;
809 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
810 return credits;
811 }
812
813 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
814 {
815 struct mpa_message *mpa;
816 u16 plen;
817 struct iwch_qp_attributes attrs;
818 enum iwch_qp_attr_mask mask;
819 int err;
820
821 PDBG("%s ep %p\n", __func__, ep);
822
823 /*
824 * Stop mpa timer. If it expired, then the state has
825 * changed and we bail since ep_timeout already aborted
826 * the connection.
827 */
828 stop_ep_timer(ep);
829 if (state_read(&ep->com) != MPA_REQ_SENT)
830 return;
831
832 /*
833 * If we get more than the supported amount of private data
834 * then we must fail this connection.
835 */
836 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
837 err = -EINVAL;
838 goto err;
839 }
840
841 /*
842 * copy the new data into our accumulation buffer.
843 */
844 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
845 skb->len);
846 ep->mpa_pkt_len += skb->len;
847
848 /*
849 * if we don't even have the mpa message, then bail.
850 */
851 if (ep->mpa_pkt_len < sizeof(*mpa))
852 return;
853 mpa = (struct mpa_message *) ep->mpa_pkt;
854
855 /* Validate MPA header. */
856 if (mpa->revision != mpa_rev) {
857 err = -EPROTO;
858 goto err;
859 }
860 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
861 err = -EPROTO;
862 goto err;
863 }
864
865 plen = ntohs(mpa->private_data_size);
866
867 /*
868 * Fail if there's too much private data.
869 */
870 if (plen > MPA_MAX_PRIVATE_DATA) {
871 err = -EPROTO;
872 goto err;
873 }
874
875 /*
876 * If plen does not account for pkt size
877 */
878 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
879 err = -EPROTO;
880 goto err;
881 }
882
883 ep->plen = (u8) plen;
884
885 /*
886 * If we don't have all the pdata yet, then bail.
887 * We'll continue process when more data arrives.
888 */
889 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
890 return;
891
892 if (mpa->flags & MPA_REJECT) {
893 err = -ECONNREFUSED;
894 goto err;
895 }
896
897 /*
898 * If we get here we have accumulated the entire mpa
899 * start reply message including private data. And
900 * the MPA header is valid.
901 */
902 state_set(&ep->com, FPDU_MODE);
903 ep->mpa_attr.initiator = 1;
904 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
905 ep->mpa_attr.recv_marker_enabled = markers_enabled;
906 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
907 ep->mpa_attr.version = mpa_rev;
908 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
909 "xmit_marker_enabled=%d, version=%d\n", __func__,
910 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
911 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
912
913 attrs.mpa_attr = ep->mpa_attr;
914 attrs.max_ird = ep->ird;
915 attrs.max_ord = ep->ord;
916 attrs.llp_stream_handle = ep;
917 attrs.next_state = IWCH_QP_STATE_RTS;
918
919 mask = IWCH_QP_ATTR_NEXT_STATE |
920 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
921 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
922
923 /* bind QP and TID with INIT_WR */
924 err = iwch_modify_qp(ep->com.qp->rhp,
925 ep->com.qp, mask, &attrs, 1);
926 if (err)
927 goto err;
928
929 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
930 iwch_post_zb_read(ep);
931 }
932
933 goto out;
934 err:
935 abort_connection(ep, skb, GFP_KERNEL);
936 out:
937 connect_reply_upcall(ep, err);
938 return;
939 }
940
941 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
942 {
943 struct mpa_message *mpa;
944 u16 plen;
945
946 PDBG("%s ep %p\n", __func__, ep);
947
948 /*
949 * Stop mpa timer. If it expired, then the state has
950 * changed and we bail since ep_timeout already aborted
951 * the connection.
952 */
953 stop_ep_timer(ep);
954 if (state_read(&ep->com) != MPA_REQ_WAIT)
955 return;
956
957 /*
958 * If we get more than the supported amount of private data
959 * then we must fail this connection.
960 */
961 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
962 abort_connection(ep, skb, GFP_KERNEL);
963 return;
964 }
965
966 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
967
968 /*
969 * Copy the new data into our accumulation buffer.
970 */
971 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
972 skb->len);
973 ep->mpa_pkt_len += skb->len;
974
975 /*
976 * If we don't even have the mpa message, then bail.
977 * We'll continue process when more data arrives.
978 */
979 if (ep->mpa_pkt_len < sizeof(*mpa))
980 return;
981 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
982 mpa = (struct mpa_message *) ep->mpa_pkt;
983
984 /*
985 * Validate MPA Header.
986 */
987 if (mpa->revision != mpa_rev) {
988 abort_connection(ep, skb, GFP_KERNEL);
989 return;
990 }
991
992 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
993 abort_connection(ep, skb, GFP_KERNEL);
994 return;
995 }
996
997 plen = ntohs(mpa->private_data_size);
998
999 /*
1000 * Fail if there's too much private data.
1001 */
1002 if (plen > MPA_MAX_PRIVATE_DATA) {
1003 abort_connection(ep, skb, GFP_KERNEL);
1004 return;
1005 }
1006
1007 /*
1008 * If plen does not account for pkt size
1009 */
1010 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1011 abort_connection(ep, skb, GFP_KERNEL);
1012 return;
1013 }
1014 ep->plen = (u8) plen;
1015
1016 /*
1017 * If we don't have all the pdata yet, then bail.
1018 */
1019 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1020 return;
1021
1022 /*
1023 * If we get here we have accumulated the entire mpa
1024 * start reply message including private data.
1025 */
1026 ep->mpa_attr.initiator = 0;
1027 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1028 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1029 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1030 ep->mpa_attr.version = mpa_rev;
1031 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1032 "xmit_marker_enabled=%d, version=%d\n", __func__,
1033 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1034 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1035
1036 state_set(&ep->com, MPA_REQ_RCVD);
1037
1038 /* drive upcall */
1039 connect_request_upcall(ep);
1040 return;
1041 }
1042
1043 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1044 {
1045 struct iwch_ep *ep = ctx;
1046 struct cpl_rx_data *hdr = cplhdr(skb);
1047 unsigned int dlen = ntohs(hdr->len);
1048
1049 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1050
1051 skb_pull(skb, sizeof(*hdr));
1052 skb_trim(skb, dlen);
1053
1054 ep->rcv_seq += dlen;
1055 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1056
1057 switch (state_read(&ep->com)) {
1058 case MPA_REQ_SENT:
1059 process_mpa_reply(ep, skb);
1060 break;
1061 case MPA_REQ_WAIT:
1062 process_mpa_request(ep, skb);
1063 break;
1064 case MPA_REP_SENT:
1065 break;
1066 default:
1067 printk(KERN_ERR MOD "%s Unexpected streaming data."
1068 " ep %p state %d tid %d\n",
1069 __func__, ep, state_read(&ep->com), ep->hwtid);
1070
1071 /*
1072 * The ep will timeout and inform the ULP of the failure.
1073 * See ep_timeout().
1074 */
1075 break;
1076 }
1077
1078 /* update RX credits */
1079 update_rx_credits(ep, dlen);
1080
1081 return CPL_RET_BUF_DONE;
1082 }
1083
1084 /*
1085 * Upcall from the adapter indicating data has been transmitted.
1086 * For us its just the single MPA request or reply. We can now free
1087 * the skb holding the mpa message.
1088 */
1089 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1090 {
1091 struct iwch_ep *ep = ctx;
1092 struct cpl_wr_ack *hdr = cplhdr(skb);
1093 unsigned int credits = ntohs(hdr->credits);
1094 unsigned long flags;
1095 int post_zb = 0;
1096
1097 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1098
1099 if (credits == 0) {
1100 PDBG("%s 0 credit ack ep %p state %u\n",
1101 __func__, ep, state_read(&ep->com));
1102 return CPL_RET_BUF_DONE;
1103 }
1104
1105 spin_lock_irqsave(&ep->com.lock, flags);
1106 BUG_ON(credits != 1);
1107 dst_confirm(ep->dst);
1108 if (!ep->mpa_skb) {
1109 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1110 __func__, ep, ep->com.state);
1111 if (ep->mpa_attr.initiator) {
1112 PDBG("%s initiator ep %p state %u\n",
1113 __func__, ep, ep->com.state);
1114 if (peer2peer && ep->com.state == FPDU_MODE)
1115 post_zb = 1;
1116 } else {
1117 PDBG("%s responder ep %p state %u\n",
1118 __func__, ep, ep->com.state);
1119 if (ep->com.state == MPA_REQ_RCVD) {
1120 ep->com.rpl_done = 1;
1121 wake_up(&ep->com.waitq);
1122 }
1123 }
1124 } else {
1125 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1126 __func__, ep, ep->com.state);
1127 kfree_skb(ep->mpa_skb);
1128 ep->mpa_skb = NULL;
1129 }
1130 spin_unlock_irqrestore(&ep->com.lock, flags);
1131 if (post_zb)
1132 iwch_post_zb_read(ep);
1133 return CPL_RET_BUF_DONE;
1134 }
1135
1136 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1137 {
1138 struct iwch_ep *ep = ctx;
1139 unsigned long flags;
1140 int release = 0;
1141
1142 PDBG("%s ep %p\n", __func__, ep);
1143 BUG_ON(!ep);
1144
1145 /*
1146 * We get 2 abort replies from the HW. The first one must
1147 * be ignored except for scribbling that we need one more.
1148 */
1149 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1150 return CPL_RET_BUF_DONE;
1151 }
1152
1153 spin_lock_irqsave(&ep->com.lock, flags);
1154 switch (ep->com.state) {
1155 case ABORTING:
1156 close_complete_upcall(ep);
1157 __state_set(&ep->com, DEAD);
1158 release = 1;
1159 break;
1160 default:
1161 printk(KERN_ERR "%s ep %p state %d\n",
1162 __func__, ep, ep->com.state);
1163 break;
1164 }
1165 spin_unlock_irqrestore(&ep->com.lock, flags);
1166
1167 if (release)
1168 release_ep_resources(ep);
1169 return CPL_RET_BUF_DONE;
1170 }
1171
1172 /*
1173 * Return whether a failed active open has allocated a TID
1174 */
1175 static inline int act_open_has_tid(int status)
1176 {
1177 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1178 status != CPL_ERR_ARP_MISS;
1179 }
1180
1181 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1182 {
1183 struct iwch_ep *ep = ctx;
1184 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1185
1186 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1187 status2errno(rpl->status));
1188 connect_reply_upcall(ep, status2errno(rpl->status));
1189 state_set(&ep->com, DEAD);
1190 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1191 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1192 cxgb3_free_atid(ep->com.tdev, ep->atid);
1193 dst_release(ep->dst);
1194 l2t_release(ep->com.tdev, ep->l2t);
1195 put_ep(&ep->com);
1196 return CPL_RET_BUF_DONE;
1197 }
1198
1199 static int listen_start(struct iwch_listen_ep *ep)
1200 {
1201 struct sk_buff *skb;
1202 struct cpl_pass_open_req *req;
1203
1204 PDBG("%s ep %p\n", __func__, ep);
1205 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1206 if (!skb) {
1207 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1208 return -ENOMEM;
1209 }
1210
1211 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1212 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1213 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1214 req->local_port = ep->com.local_addr.sin_port;
1215 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1216 req->peer_port = 0;
1217 req->peer_ip = 0;
1218 req->peer_netmask = 0;
1219 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1220 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1221 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1222
1223 skb->priority = 1;
1224 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1225 }
1226
1227 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1228 {
1229 struct iwch_listen_ep *ep = ctx;
1230 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1231
1232 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1233 rpl->status, status2errno(rpl->status));
1234 ep->com.rpl_err = status2errno(rpl->status);
1235 ep->com.rpl_done = 1;
1236 wake_up(&ep->com.waitq);
1237
1238 return CPL_RET_BUF_DONE;
1239 }
1240
1241 static int listen_stop(struct iwch_listen_ep *ep)
1242 {
1243 struct sk_buff *skb;
1244 struct cpl_close_listserv_req *req;
1245
1246 PDBG("%s ep %p\n", __func__, ep);
1247 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1248 if (!skb) {
1249 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1250 return -ENOMEM;
1251 }
1252 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1253 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1254 req->cpu_idx = 0;
1255 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1256 skb->priority = 1;
1257 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1258 }
1259
1260 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1261 void *ctx)
1262 {
1263 struct iwch_listen_ep *ep = ctx;
1264 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1265
1266 PDBG("%s ep %p\n", __func__, ep);
1267 ep->com.rpl_err = status2errno(rpl->status);
1268 ep->com.rpl_done = 1;
1269 wake_up(&ep->com.waitq);
1270 return CPL_RET_BUF_DONE;
1271 }
1272
1273 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1274 {
1275 struct cpl_pass_accept_rpl *rpl;
1276 unsigned int mtu_idx;
1277 u32 opt0h, opt0l, opt2;
1278 int wscale;
1279
1280 PDBG("%s ep %p\n", __func__, ep);
1281 BUG_ON(skb_cloned(skb));
1282 skb_trim(skb, sizeof(*rpl));
1283 skb_get(skb);
1284 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1285 wscale = compute_wscale(rcv_win);
1286 opt0h = V_NAGLE(0) |
1287 V_NO_CONG(nocong) |
1288 V_KEEP_ALIVE(1) |
1289 F_TCAM_BYPASS |
1290 V_WND_SCALE(wscale) |
1291 V_MSS_IDX(mtu_idx) |
1292 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1293 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1294 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1295 V_CONG_CONTROL_FLAVOR(cong_flavor);
1296
1297 rpl = cplhdr(skb);
1298 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1299 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1300 rpl->peer_ip = peer_ip;
1301 rpl->opt0h = htonl(opt0h);
1302 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1303 rpl->opt2 = htonl(opt2);
1304 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1305 skb->priority = CPL_PRIORITY_SETUP;
1306 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1307
1308 return;
1309 }
1310
1311 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1312 struct sk_buff *skb)
1313 {
1314 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1315 peer_ip);
1316 BUG_ON(skb_cloned(skb));
1317 skb_trim(skb, sizeof(struct cpl_tid_release));
1318 skb_get(skb);
1319
1320 if (tdev->type != T3A)
1321 release_tid(tdev, hwtid, skb);
1322 else {
1323 struct cpl_pass_accept_rpl *rpl;
1324
1325 rpl = cplhdr(skb);
1326 skb->priority = CPL_PRIORITY_SETUP;
1327 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1328 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1329 hwtid));
1330 rpl->peer_ip = peer_ip;
1331 rpl->opt0h = htonl(F_TCAM_BYPASS);
1332 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1333 rpl->opt2 = 0;
1334 rpl->rsvd = rpl->opt2;
1335 iwch_cxgb3_ofld_send(tdev, skb);
1336 }
1337 }
1338
1339 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1340 {
1341 struct iwch_ep *child_ep, *parent_ep = ctx;
1342 struct cpl_pass_accept_req *req = cplhdr(skb);
1343 unsigned int hwtid = GET_TID(req);
1344 struct dst_entry *dst;
1345 struct l2t_entry *l2t;
1346 struct rtable *rt;
1347 struct iff_mac tim;
1348
1349 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1350
1351 if (state_read(&parent_ep->com) != LISTEN) {
1352 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1353 __func__);
1354 goto reject;
1355 }
1356
1357 /*
1358 * Find the netdev for this connection request.
1359 */
1360 tim.mac_addr = req->dst_mac;
1361 tim.vlan_tag = ntohs(req->vlan_tag);
1362 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1363 printk(KERN_ERR "%s bad dst mac %pM\n",
1364 __func__, req->dst_mac);
1365 goto reject;
1366 }
1367
1368 /* Find output route */
1369 rt = find_route(tdev,
1370 req->local_ip,
1371 req->peer_ip,
1372 req->local_port,
1373 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1374 if (!rt) {
1375 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1376 __func__);
1377 goto reject;
1378 }
1379 dst = &rt->dst;
1380 l2t = t3_l2t_get(tdev, dst, NULL, &req->peer_ip);
1381 if (!l2t) {
1382 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1383 __func__);
1384 dst_release(dst);
1385 goto reject;
1386 }
1387 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1388 if (!child_ep) {
1389 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1390 __func__);
1391 l2t_release(tdev, l2t);
1392 dst_release(dst);
1393 goto reject;
1394 }
1395 state_set(&child_ep->com, CONNECTING);
1396 child_ep->com.tdev = tdev;
1397 child_ep->com.cm_id = NULL;
1398 child_ep->com.local_addr.sin_family = PF_INET;
1399 child_ep->com.local_addr.sin_port = req->local_port;
1400 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1401 child_ep->com.remote_addr.sin_family = PF_INET;
1402 child_ep->com.remote_addr.sin_port = req->peer_port;
1403 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1404 get_ep(&parent_ep->com);
1405 child_ep->parent_ep = parent_ep;
1406 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1407 child_ep->l2t = l2t;
1408 child_ep->dst = dst;
1409 child_ep->hwtid = hwtid;
1410 init_timer(&child_ep->timer);
1411 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1412 accept_cr(child_ep, req->peer_ip, skb);
1413 goto out;
1414 reject:
1415 reject_cr(tdev, hwtid, req->peer_ip, skb);
1416 out:
1417 return CPL_RET_BUF_DONE;
1418 }
1419
1420 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1421 {
1422 struct iwch_ep *ep = ctx;
1423 struct cpl_pass_establish *req = cplhdr(skb);
1424
1425 PDBG("%s ep %p\n", __func__, ep);
1426 ep->snd_seq = ntohl(req->snd_isn);
1427 ep->rcv_seq = ntohl(req->rcv_isn);
1428
1429 set_emss(ep, ntohs(req->tcp_opt));
1430
1431 dst_confirm(ep->dst);
1432 state_set(&ep->com, MPA_REQ_WAIT);
1433 start_ep_timer(ep);
1434
1435 return CPL_RET_BUF_DONE;
1436 }
1437
1438 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1439 {
1440 struct iwch_ep *ep = ctx;
1441 struct iwch_qp_attributes attrs;
1442 unsigned long flags;
1443 int disconnect = 1;
1444 int release = 0;
1445
1446 PDBG("%s ep %p\n", __func__, ep);
1447 dst_confirm(ep->dst);
1448
1449 spin_lock_irqsave(&ep->com.lock, flags);
1450 switch (ep->com.state) {
1451 case MPA_REQ_WAIT:
1452 __state_set(&ep->com, CLOSING);
1453 break;
1454 case MPA_REQ_SENT:
1455 __state_set(&ep->com, CLOSING);
1456 connect_reply_upcall(ep, -ECONNRESET);
1457 break;
1458 case MPA_REQ_RCVD:
1459
1460 /*
1461 * We're gonna mark this puppy DEAD, but keep
1462 * the reference on it until the ULP accepts or
1463 * rejects the CR. Also wake up anyone waiting
1464 * in rdma connection migration (see iwch_accept_cr()).
1465 */
1466 __state_set(&ep->com, CLOSING);
1467 ep->com.rpl_done = 1;
1468 ep->com.rpl_err = -ECONNRESET;
1469 PDBG("waking up ep %p\n", ep);
1470 wake_up(&ep->com.waitq);
1471 break;
1472 case MPA_REP_SENT:
1473 __state_set(&ep->com, CLOSING);
1474 ep->com.rpl_done = 1;
1475 ep->com.rpl_err = -ECONNRESET;
1476 PDBG("waking up ep %p\n", ep);
1477 wake_up(&ep->com.waitq);
1478 break;
1479 case FPDU_MODE:
1480 start_ep_timer(ep);
1481 __state_set(&ep->com, CLOSING);
1482 attrs.next_state = IWCH_QP_STATE_CLOSING;
1483 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1484 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1485 peer_close_upcall(ep);
1486 break;
1487 case ABORTING:
1488 disconnect = 0;
1489 break;
1490 case CLOSING:
1491 __state_set(&ep->com, MORIBUND);
1492 disconnect = 0;
1493 break;
1494 case MORIBUND:
1495 stop_ep_timer(ep);
1496 if (ep->com.cm_id && ep->com.qp) {
1497 attrs.next_state = IWCH_QP_STATE_IDLE;
1498 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1499 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1500 }
1501 close_complete_upcall(ep);
1502 __state_set(&ep->com, DEAD);
1503 release = 1;
1504 disconnect = 0;
1505 break;
1506 case DEAD:
1507 disconnect = 0;
1508 break;
1509 default:
1510 BUG_ON(1);
1511 }
1512 spin_unlock_irqrestore(&ep->com.lock, flags);
1513 if (disconnect)
1514 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1515 if (release)
1516 release_ep_resources(ep);
1517 return CPL_RET_BUF_DONE;
1518 }
1519
1520 /*
1521 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1522 */
1523 static int is_neg_adv_abort(unsigned int status)
1524 {
1525 return status == CPL_ERR_RTX_NEG_ADVICE ||
1526 status == CPL_ERR_PERSIST_NEG_ADVICE;
1527 }
1528
1529 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1530 {
1531 struct cpl_abort_req_rss *req = cplhdr(skb);
1532 struct iwch_ep *ep = ctx;
1533 struct cpl_abort_rpl *rpl;
1534 struct sk_buff *rpl_skb;
1535 struct iwch_qp_attributes attrs;
1536 int ret;
1537 int release = 0;
1538 unsigned long flags;
1539
1540 if (is_neg_adv_abort(req->status)) {
1541 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1542 ep->hwtid);
1543 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1544 return CPL_RET_BUF_DONE;
1545 }
1546
1547 /*
1548 * We get 2 peer aborts from the HW. The first one must
1549 * be ignored except for scribbling that we need one more.
1550 */
1551 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1552 return CPL_RET_BUF_DONE;
1553 }
1554
1555 spin_lock_irqsave(&ep->com.lock, flags);
1556 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1557 switch (ep->com.state) {
1558 case CONNECTING:
1559 break;
1560 case MPA_REQ_WAIT:
1561 stop_ep_timer(ep);
1562 break;
1563 case MPA_REQ_SENT:
1564 stop_ep_timer(ep);
1565 connect_reply_upcall(ep, -ECONNRESET);
1566 break;
1567 case MPA_REP_SENT:
1568 ep->com.rpl_done = 1;
1569 ep->com.rpl_err = -ECONNRESET;
1570 PDBG("waking up ep %p\n", ep);
1571 wake_up(&ep->com.waitq);
1572 break;
1573 case MPA_REQ_RCVD:
1574
1575 /*
1576 * We're gonna mark this puppy DEAD, but keep
1577 * the reference on it until the ULP accepts or
1578 * rejects the CR. Also wake up anyone waiting
1579 * in rdma connection migration (see iwch_accept_cr()).
1580 */
1581 ep->com.rpl_done = 1;
1582 ep->com.rpl_err = -ECONNRESET;
1583 PDBG("waking up ep %p\n", ep);
1584 wake_up(&ep->com.waitq);
1585 break;
1586 case MORIBUND:
1587 case CLOSING:
1588 stop_ep_timer(ep);
1589 /*FALLTHROUGH*/
1590 case FPDU_MODE:
1591 if (ep->com.cm_id && ep->com.qp) {
1592 attrs.next_state = IWCH_QP_STATE_ERROR;
1593 ret = iwch_modify_qp(ep->com.qp->rhp,
1594 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1595 &attrs, 1);
1596 if (ret)
1597 printk(KERN_ERR MOD
1598 "%s - qp <- error failed!\n",
1599 __func__);
1600 }
1601 peer_abort_upcall(ep);
1602 break;
1603 case ABORTING:
1604 break;
1605 case DEAD:
1606 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1607 spin_unlock_irqrestore(&ep->com.lock, flags);
1608 return CPL_RET_BUF_DONE;
1609 default:
1610 BUG_ON(1);
1611 break;
1612 }
1613 dst_confirm(ep->dst);
1614 if (ep->com.state != ABORTING) {
1615 __state_set(&ep->com, DEAD);
1616 release = 1;
1617 }
1618 spin_unlock_irqrestore(&ep->com.lock, flags);
1619
1620 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1621 if (!rpl_skb) {
1622 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1623 __func__);
1624 release = 1;
1625 goto out;
1626 }
1627 rpl_skb->priority = CPL_PRIORITY_DATA;
1628 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1629 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1630 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1631 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1632 rpl->cmd = CPL_ABORT_NO_RST;
1633 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1634 out:
1635 if (release)
1636 release_ep_resources(ep);
1637 return CPL_RET_BUF_DONE;
1638 }
1639
1640 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1641 {
1642 struct iwch_ep *ep = ctx;
1643 struct iwch_qp_attributes attrs;
1644 unsigned long flags;
1645 int release = 0;
1646
1647 PDBG("%s ep %p\n", __func__, ep);
1648 BUG_ON(!ep);
1649
1650 /* The cm_id may be null if we failed to connect */
1651 spin_lock_irqsave(&ep->com.lock, flags);
1652 switch (ep->com.state) {
1653 case CLOSING:
1654 __state_set(&ep->com, MORIBUND);
1655 break;
1656 case MORIBUND:
1657 stop_ep_timer(ep);
1658 if ((ep->com.cm_id) && (ep->com.qp)) {
1659 attrs.next_state = IWCH_QP_STATE_IDLE;
1660 iwch_modify_qp(ep->com.qp->rhp,
1661 ep->com.qp,
1662 IWCH_QP_ATTR_NEXT_STATE,
1663 &attrs, 1);
1664 }
1665 close_complete_upcall(ep);
1666 __state_set(&ep->com, DEAD);
1667 release = 1;
1668 break;
1669 case ABORTING:
1670 case DEAD:
1671 break;
1672 default:
1673 BUG_ON(1);
1674 break;
1675 }
1676 spin_unlock_irqrestore(&ep->com.lock, flags);
1677 if (release)
1678 release_ep_resources(ep);
1679 return CPL_RET_BUF_DONE;
1680 }
1681
1682 /*
1683 * T3A does 3 things when a TERM is received:
1684 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1685 * 2) generate an async event on the QP with the TERMINATE opcode
1686 * 3) post a TERMINATE opcode cqe into the associated CQ.
1687 *
1688 * For (1), we save the message in the qp for later consumer consumption.
1689 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1690 * For (3), we toss the CQE in cxio_poll_cq().
1691 *
1692 * terminate() handles case (1)...
1693 */
1694 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1695 {
1696 struct iwch_ep *ep = ctx;
1697
1698 if (state_read(&ep->com) != FPDU_MODE)
1699 return CPL_RET_BUF_DONE;
1700
1701 PDBG("%s ep %p\n", __func__, ep);
1702 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1703 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1704 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1705 skb->len);
1706 ep->com.qp->attr.terminate_msg_len = skb->len;
1707 ep->com.qp->attr.is_terminate_local = 0;
1708 return CPL_RET_BUF_DONE;
1709 }
1710
1711 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1712 {
1713 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1714 struct iwch_ep *ep = ctx;
1715
1716 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1717 rep->status);
1718 if (rep->status) {
1719 struct iwch_qp_attributes attrs;
1720
1721 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1722 __func__, ep->hwtid);
1723 stop_ep_timer(ep);
1724 attrs.next_state = IWCH_QP_STATE_ERROR;
1725 iwch_modify_qp(ep->com.qp->rhp,
1726 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1727 &attrs, 1);
1728 abort_connection(ep, NULL, GFP_KERNEL);
1729 }
1730 return CPL_RET_BUF_DONE;
1731 }
1732
1733 static void ep_timeout(unsigned long arg)
1734 {
1735 struct iwch_ep *ep = (struct iwch_ep *)arg;
1736 struct iwch_qp_attributes attrs;
1737 unsigned long flags;
1738 int abort = 1;
1739
1740 spin_lock_irqsave(&ep->com.lock, flags);
1741 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1742 ep->com.state);
1743 switch (ep->com.state) {
1744 case MPA_REQ_SENT:
1745 __state_set(&ep->com, ABORTING);
1746 connect_reply_upcall(ep, -ETIMEDOUT);
1747 break;
1748 case MPA_REQ_WAIT:
1749 __state_set(&ep->com, ABORTING);
1750 break;
1751 case CLOSING:
1752 case MORIBUND:
1753 if (ep->com.cm_id && ep->com.qp) {
1754 attrs.next_state = IWCH_QP_STATE_ERROR;
1755 iwch_modify_qp(ep->com.qp->rhp,
1756 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1757 &attrs, 1);
1758 }
1759 __state_set(&ep->com, ABORTING);
1760 break;
1761 default:
1762 WARN(1, "%s unexpected state ep %p state %u\n",
1763 __func__, ep, ep->com.state);
1764 abort = 0;
1765 }
1766 spin_unlock_irqrestore(&ep->com.lock, flags);
1767 if (abort)
1768 abort_connection(ep, NULL, GFP_ATOMIC);
1769 put_ep(&ep->com);
1770 }
1771
1772 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1773 {
1774 int err;
1775 struct iwch_ep *ep = to_ep(cm_id);
1776 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1777
1778 if (state_read(&ep->com) == DEAD) {
1779 put_ep(&ep->com);
1780 return -ECONNRESET;
1781 }
1782 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1783 if (mpa_rev == 0)
1784 abort_connection(ep, NULL, GFP_KERNEL);
1785 else {
1786 err = send_mpa_reject(ep, pdata, pdata_len);
1787 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1788 }
1789 put_ep(&ep->com);
1790 return 0;
1791 }
1792
1793 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1794 {
1795 int err;
1796 struct iwch_qp_attributes attrs;
1797 enum iwch_qp_attr_mask mask;
1798 struct iwch_ep *ep = to_ep(cm_id);
1799 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1800 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1801
1802 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1803 if (state_read(&ep->com) == DEAD) {
1804 err = -ECONNRESET;
1805 goto err;
1806 }
1807
1808 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1809 BUG_ON(!qp);
1810
1811 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1812 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1813 abort_connection(ep, NULL, GFP_KERNEL);
1814 err = -EINVAL;
1815 goto err;
1816 }
1817
1818 cm_id->add_ref(cm_id);
1819 ep->com.cm_id = cm_id;
1820 ep->com.qp = qp;
1821
1822 ep->ird = conn_param->ird;
1823 ep->ord = conn_param->ord;
1824
1825 if (peer2peer && ep->ird == 0)
1826 ep->ird = 1;
1827
1828 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1829
1830 /* bind QP to EP and move to RTS */
1831 attrs.mpa_attr = ep->mpa_attr;
1832 attrs.max_ird = ep->ird;
1833 attrs.max_ord = ep->ord;
1834 attrs.llp_stream_handle = ep;
1835 attrs.next_state = IWCH_QP_STATE_RTS;
1836
1837 /* bind QP and TID with INIT_WR */
1838 mask = IWCH_QP_ATTR_NEXT_STATE |
1839 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1840 IWCH_QP_ATTR_MPA_ATTR |
1841 IWCH_QP_ATTR_MAX_IRD |
1842 IWCH_QP_ATTR_MAX_ORD;
1843
1844 err = iwch_modify_qp(ep->com.qp->rhp,
1845 ep->com.qp, mask, &attrs, 1);
1846 if (err)
1847 goto err1;
1848
1849 /* if needed, wait for wr_ack */
1850 if (iwch_rqes_posted(qp)) {
1851 wait_event(ep->com.waitq, ep->com.rpl_done);
1852 err = ep->com.rpl_err;
1853 if (err)
1854 goto err1;
1855 }
1856
1857 err = send_mpa_reply(ep, conn_param->private_data,
1858 conn_param->private_data_len);
1859 if (err)
1860 goto err1;
1861
1862
1863 state_set(&ep->com, FPDU_MODE);
1864 established_upcall(ep);
1865 put_ep(&ep->com);
1866 return 0;
1867 err1:
1868 ep->com.cm_id = NULL;
1869 ep->com.qp = NULL;
1870 cm_id->rem_ref(cm_id);
1871 err:
1872 put_ep(&ep->com);
1873 return err;
1874 }
1875
1876 static int is_loopback_dst(struct iw_cm_id *cm_id)
1877 {
1878 struct net_device *dev;
1879 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
1880
1881 dev = ip_dev_find(&init_net, raddr->sin_addr.s_addr);
1882 if (!dev)
1883 return 0;
1884 dev_put(dev);
1885 return 1;
1886 }
1887
1888 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1889 {
1890 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1891 struct iwch_ep *ep;
1892 struct rtable *rt;
1893 int err = 0;
1894 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
1895 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
1896
1897 if (cm_id->remote_addr.ss_family != PF_INET) {
1898 err = -ENOSYS;
1899 goto out;
1900 }
1901
1902 if (is_loopback_dst(cm_id)) {
1903 err = -ENOSYS;
1904 goto out;
1905 }
1906
1907 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1908 if (!ep) {
1909 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1910 err = -ENOMEM;
1911 goto out;
1912 }
1913 init_timer(&ep->timer);
1914 ep->plen = conn_param->private_data_len;
1915 if (ep->plen)
1916 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1917 conn_param->private_data, ep->plen);
1918 ep->ird = conn_param->ird;
1919 ep->ord = conn_param->ord;
1920
1921 if (peer2peer && ep->ord == 0)
1922 ep->ord = 1;
1923
1924 ep->com.tdev = h->rdev.t3cdev_p;
1925
1926 cm_id->add_ref(cm_id);
1927 ep->com.cm_id = cm_id;
1928 ep->com.qp = get_qhp(h, conn_param->qpn);
1929 BUG_ON(!ep->com.qp);
1930 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1931 ep->com.qp, cm_id);
1932
1933 /*
1934 * Allocate an active TID to initiate a TCP connection.
1935 */
1936 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1937 if (ep->atid == -1) {
1938 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1939 err = -ENOMEM;
1940 goto fail2;
1941 }
1942
1943 /* find a route */
1944 rt = find_route(h->rdev.t3cdev_p, laddr->sin_addr.s_addr,
1945 raddr->sin_addr.s_addr, laddr->sin_port,
1946 raddr->sin_port, IPTOS_LOWDELAY);
1947 if (!rt) {
1948 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1949 err = -EHOSTUNREACH;
1950 goto fail3;
1951 }
1952 ep->dst = &rt->dst;
1953 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst, NULL,
1954 &raddr->sin_addr.s_addr);
1955 if (!ep->l2t) {
1956 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1957 err = -ENOMEM;
1958 goto fail4;
1959 }
1960
1961 state_set(&ep->com, CONNECTING);
1962 ep->tos = IPTOS_LOWDELAY;
1963 memcpy(&ep->com.local_addr, &cm_id->local_addr,
1964 sizeof(ep->com.local_addr));
1965 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
1966 sizeof(ep->com.remote_addr));
1967
1968 /* send connect request to rnic */
1969 err = send_connect(ep);
1970 if (!err)
1971 goto out;
1972
1973 l2t_release(h->rdev.t3cdev_p, ep->l2t);
1974 fail4:
1975 dst_release(ep->dst);
1976 fail3:
1977 cxgb3_free_atid(ep->com.tdev, ep->atid);
1978 fail2:
1979 cm_id->rem_ref(cm_id);
1980 put_ep(&ep->com);
1981 out:
1982 return err;
1983 }
1984
1985 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1986 {
1987 int err = 0;
1988 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1989 struct iwch_listen_ep *ep;
1990
1991
1992 might_sleep();
1993
1994 if (cm_id->local_addr.ss_family != PF_INET) {
1995 err = -ENOSYS;
1996 goto fail1;
1997 }
1998
1999 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2000 if (!ep) {
2001 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2002 err = -ENOMEM;
2003 goto fail1;
2004 }
2005 PDBG("%s ep %p\n", __func__, ep);
2006 ep->com.tdev = h->rdev.t3cdev_p;
2007 cm_id->add_ref(cm_id);
2008 ep->com.cm_id = cm_id;
2009 ep->backlog = backlog;
2010 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2011 sizeof(ep->com.local_addr));
2012
2013 /*
2014 * Allocate a server TID.
2015 */
2016 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2017 if (ep->stid == -1) {
2018 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2019 err = -ENOMEM;
2020 goto fail2;
2021 }
2022
2023 state_set(&ep->com, LISTEN);
2024 err = listen_start(ep);
2025 if (err)
2026 goto fail3;
2027
2028 /* wait for pass_open_rpl */
2029 wait_event(ep->com.waitq, ep->com.rpl_done);
2030 err = ep->com.rpl_err;
2031 if (!err) {
2032 cm_id->provider_data = ep;
2033 goto out;
2034 }
2035 fail3:
2036 cxgb3_free_stid(ep->com.tdev, ep->stid);
2037 fail2:
2038 cm_id->rem_ref(cm_id);
2039 put_ep(&ep->com);
2040 fail1:
2041 out:
2042 return err;
2043 }
2044
2045 int iwch_destroy_listen(struct iw_cm_id *cm_id)
2046 {
2047 int err;
2048 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2049
2050 PDBG("%s ep %p\n", __func__, ep);
2051
2052 might_sleep();
2053 state_set(&ep->com, DEAD);
2054 ep->com.rpl_done = 0;
2055 ep->com.rpl_err = 0;
2056 err = listen_stop(ep);
2057 if (err)
2058 goto done;
2059 wait_event(ep->com.waitq, ep->com.rpl_done);
2060 cxgb3_free_stid(ep->com.tdev, ep->stid);
2061 done:
2062 err = ep->com.rpl_err;
2063 cm_id->rem_ref(cm_id);
2064 put_ep(&ep->com);
2065 return err;
2066 }
2067
2068 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2069 {
2070 int ret=0;
2071 unsigned long flags;
2072 int close = 0;
2073 int fatal = 0;
2074 struct t3cdev *tdev;
2075 struct cxio_rdev *rdev;
2076
2077 spin_lock_irqsave(&ep->com.lock, flags);
2078
2079 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2080 states[ep->com.state], abrupt);
2081
2082 tdev = (struct t3cdev *)ep->com.tdev;
2083 rdev = (struct cxio_rdev *)tdev->ulp;
2084 if (cxio_fatal_error(rdev)) {
2085 fatal = 1;
2086 close_complete_upcall(ep);
2087 ep->com.state = DEAD;
2088 }
2089 switch (ep->com.state) {
2090 case MPA_REQ_WAIT:
2091 case MPA_REQ_SENT:
2092 case MPA_REQ_RCVD:
2093 case MPA_REP_SENT:
2094 case FPDU_MODE:
2095 close = 1;
2096 if (abrupt)
2097 ep->com.state = ABORTING;
2098 else {
2099 ep->com.state = CLOSING;
2100 start_ep_timer(ep);
2101 }
2102 set_bit(CLOSE_SENT, &ep->com.flags);
2103 break;
2104 case CLOSING:
2105 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2106 close = 1;
2107 if (abrupt) {
2108 stop_ep_timer(ep);
2109 ep->com.state = ABORTING;
2110 } else
2111 ep->com.state = MORIBUND;
2112 }
2113 break;
2114 case MORIBUND:
2115 case ABORTING:
2116 case DEAD:
2117 PDBG("%s ignoring disconnect ep %p state %u\n",
2118 __func__, ep, ep->com.state);
2119 break;
2120 default:
2121 BUG();
2122 break;
2123 }
2124
2125 spin_unlock_irqrestore(&ep->com.lock, flags);
2126 if (close) {
2127 if (abrupt)
2128 ret = send_abort(ep, NULL, gfp);
2129 else
2130 ret = send_halfclose(ep, gfp);
2131 if (ret)
2132 fatal = 1;
2133 }
2134 if (fatal)
2135 release_ep_resources(ep);
2136 return ret;
2137 }
2138
2139 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2140 struct l2t_entry *l2t)
2141 {
2142 struct iwch_ep *ep = ctx;
2143
2144 if (ep->dst != old)
2145 return 0;
2146
2147 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2148 l2t);
2149 dst_hold(new);
2150 l2t_release(ep->com.tdev, ep->l2t);
2151 ep->l2t = l2t;
2152 dst_release(old);
2153 ep->dst = new;
2154 return 1;
2155 }
2156
2157 /*
2158 * All the CM events are handled on a work queue to have a safe context.
2159 * These are the real handlers that are called from the work queue.
2160 */
2161 static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2162 [CPL_ACT_ESTABLISH] = act_establish,
2163 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2164 [CPL_RX_DATA] = rx_data,
2165 [CPL_TX_DMA_ACK] = tx_ack,
2166 [CPL_ABORT_RPL_RSS] = abort_rpl,
2167 [CPL_ABORT_RPL] = abort_rpl,
2168 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2169 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2170 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2171 [CPL_PASS_ESTABLISH] = pass_establish,
2172 [CPL_PEER_CLOSE] = peer_close,
2173 [CPL_ABORT_REQ_RSS] = peer_abort,
2174 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2175 [CPL_RDMA_TERMINATE] = terminate,
2176 [CPL_RDMA_EC_STATUS] = ec_status,
2177 };
2178
2179 static void process_work(struct work_struct *work)
2180 {
2181 struct sk_buff *skb = NULL;
2182 void *ep;
2183 struct t3cdev *tdev;
2184 int ret;
2185
2186 while ((skb = skb_dequeue(&rxq))) {
2187 ep = *((void **) (skb->cb));
2188 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2189 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2190 if (ret & CPL_RET_BUF_DONE)
2191 kfree_skb(skb);
2192
2193 /*
2194 * ep was referenced in sched(), and is freed here.
2195 */
2196 put_ep((struct iwch_ep_common *)ep);
2197 }
2198 }
2199
2200 static DECLARE_WORK(skb_work, process_work);
2201
2202 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2203 {
2204 struct iwch_ep_common *epc = ctx;
2205
2206 get_ep(epc);
2207
2208 /*
2209 * Save ctx and tdev in the skb->cb area.
2210 */
2211 *((void **) skb->cb) = ctx;
2212 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2213
2214 /*
2215 * Queue the skb and schedule the worker thread.
2216 */
2217 skb_queue_tail(&rxq, skb);
2218 queue_work(workq, &skb_work);
2219 return 0;
2220 }
2221
2222 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2223 {
2224 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2225
2226 if (rpl->status != CPL_ERR_NONE) {
2227 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2228 "for tid %u\n", rpl->status, GET_TID(rpl));
2229 }
2230 return CPL_RET_BUF_DONE;
2231 }
2232
2233 /*
2234 * All upcalls from the T3 Core go to sched() to schedule the
2235 * processing on a work queue.
2236 */
2237 cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2238 [CPL_ACT_ESTABLISH] = sched,
2239 [CPL_ACT_OPEN_RPL] = sched,
2240 [CPL_RX_DATA] = sched,
2241 [CPL_TX_DMA_ACK] = sched,
2242 [CPL_ABORT_RPL_RSS] = sched,
2243 [CPL_ABORT_RPL] = sched,
2244 [CPL_PASS_OPEN_RPL] = sched,
2245 [CPL_CLOSE_LISTSRV_RPL] = sched,
2246 [CPL_PASS_ACCEPT_REQ] = sched,
2247 [CPL_PASS_ESTABLISH] = sched,
2248 [CPL_PEER_CLOSE] = sched,
2249 [CPL_CLOSE_CON_RPL] = sched,
2250 [CPL_ABORT_REQ_RSS] = sched,
2251 [CPL_RDMA_TERMINATE] = sched,
2252 [CPL_RDMA_EC_STATUS] = sched,
2253 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2254 };
2255
2256 int __init iwch_cm_init(void)
2257 {
2258 skb_queue_head_init(&rxq);
2259
2260 workq = create_singlethread_workqueue("iw_cxgb3");
2261 if (!workq)
2262 return -ENOMEM;
2263
2264 return 0;
2265 }
2266
2267 void __exit iwch_cm_term(void)
2268 {
2269 flush_workqueue(workq);
2270 destroy_workqueue(workq);
2271 }
Linux with added FPC-III support