aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/cxgb3/iwch_cm.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/infiniband/hw/cxgb3/iwch_cm.c')
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_cm.c2081
1 files changed, 2081 insertions, 0 deletions
diff --git a/drivers/infiniband/hw/cxgb3/iwch_cm.c b/drivers/infiniband/hw/cxgb3/iwch_cm.c
new file mode 100644
index 000000000000..a522b1baa3b4
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb3/iwch_cm.c
@@ -0,0 +1,2081 @@
1/*
2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
3 * Copyright (c) 2006 Open Grid Computing, Inc. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33#include <linux/module.h>
34#include <linux/list.h>
35#include <linux/workqueue.h>
36#include <linux/skbuff.h>
37#include <linux/timer.h>
38#include <linux/notifier.h>
39
40#include <net/neighbour.h>
41#include <net/netevent.h>
42#include <net/route.h>
43
44#include "tcb.h"
45#include "cxgb3_offload.h"
46#include "iwch.h"
47#include "iwch_provider.h"
48#include "iwch_cm.h"
49
50static char *states[] = {
51 "idle",
52 "listen",
53 "connecting",
54 "mpa_wait_req",
55 "mpa_req_sent",
56 "mpa_req_rcvd",
57 "mpa_rep_sent",
58 "fpdu_mode",
59 "aborting",
60 "closing",
61 "moribund",
62 "dead",
63 NULL,
64};
65
66static int ep_timeout_secs = 10;
67module_param(ep_timeout_secs, int, 0444);
68MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
69 "in seconds (default=10)");
70
71static int mpa_rev = 1;
72module_param(mpa_rev, int, 0444);
73MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
74 "1 is spec compliant. (default=1)");
75
76static int markers_enabled = 0;
77module_param(markers_enabled, int, 0444);
78MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
79
80static int crc_enabled = 1;
81module_param(crc_enabled, int, 0444);
82MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
83
84static int rcv_win = 256 * 1024;
85module_param(rcv_win, int, 0444);
86MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
87
88static int snd_win = 32 * 1024;
89module_param(snd_win, int, 0444);
90MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
91
92static unsigned int nocong = 0;
93module_param(nocong, uint, 0444);
94MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
95
96static unsigned int cong_flavor = 1;
97module_param(cong_flavor, uint, 0444);
98MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
99
100static void process_work(struct work_struct *work);
101static struct workqueue_struct *workq;
102static DECLARE_WORK(skb_work, process_work);
103
104static struct sk_buff_head rxq;
105static cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS];
106
107static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
108static void ep_timeout(unsigned long arg);
109static void connect_reply_upcall(struct iwch_ep *ep, int status);
110
111static void start_ep_timer(struct iwch_ep *ep)
112{
113 PDBG("%s ep %p\n", __FUNCTION__, ep);
114 if (timer_pending(&ep->timer)) {
115 PDBG("%s stopped / restarted timer ep %p\n", __FUNCTION__, ep);
116 del_timer_sync(&ep->timer);
117 } else
118 get_ep(&ep->com);
119 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
120 ep->timer.data = (unsigned long)ep;
121 ep->timer.function = ep_timeout;
122 add_timer(&ep->timer);
123}
124
125static void stop_ep_timer(struct iwch_ep *ep)
126{
127 PDBG("%s ep %p\n", __FUNCTION__, ep);
128 del_timer_sync(&ep->timer);
129 put_ep(&ep->com);
130}
131
132static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
133{
134 struct cpl_tid_release *req;
135
136 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
137 if (!skb)
138 return;
139 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
140 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
141 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
142 skb->priority = CPL_PRIORITY_SETUP;
143 tdev->send(tdev, skb);
144 return;
145}
146
147int iwch_quiesce_tid(struct iwch_ep *ep)
148{
149 struct cpl_set_tcb_field *req;
150 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
151
152 if (!skb)
153 return -ENOMEM;
154 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
155 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
156 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
157 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
158 req->reply = 0;
159 req->cpu_idx = 0;
160 req->word = htons(W_TCB_RX_QUIESCE);
161 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
162 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
163
164 skb->priority = CPL_PRIORITY_DATA;
165 ep->com.tdev->send(ep->com.tdev, skb);
166 return 0;
167}
168
169int iwch_resume_tid(struct iwch_ep *ep)
170{
171 struct cpl_set_tcb_field *req;
172 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
173
174 if (!skb)
175 return -ENOMEM;
176 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
177 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
178 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
179 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
180 req->reply = 0;
181 req->cpu_idx = 0;
182 req->word = htons(W_TCB_RX_QUIESCE);
183 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
184 req->val = 0;
185
186 skb->priority = CPL_PRIORITY_DATA;
187 ep->com.tdev->send(ep->com.tdev, skb);
188 return 0;
189}
190
191static void set_emss(struct iwch_ep *ep, u16 opt)
192{
193 PDBG("%s ep %p opt %u\n", __FUNCTION__, ep, opt);
194 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
195 if (G_TCPOPT_TSTAMP(opt))
196 ep->emss -= 12;
197 if (ep->emss < 128)
198 ep->emss = 128;
199 PDBG("emss=%d\n", ep->emss);
200}
201
202static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
203{
204 unsigned long flags;
205 enum iwch_ep_state state;
206
207 spin_lock_irqsave(&epc->lock, flags);
208 state = epc->state;
209 spin_unlock_irqrestore(&epc->lock, flags);
210 return state;
211}
212
213static inline void __state_set(struct iwch_ep_common *epc,
214 enum iwch_ep_state new)
215{
216 epc->state = new;
217}
218
219static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
220{
221 unsigned long flags;
222
223 spin_lock_irqsave(&epc->lock, flags);
224 PDBG("%s - %s -> %s\n", __FUNCTION__, states[epc->state], states[new]);
225 __state_set(epc, new);
226 spin_unlock_irqrestore(&epc->lock, flags);
227 return;
228}
229
230static void *alloc_ep(int size, gfp_t gfp)
231{
232 struct iwch_ep_common *epc;
233
234 epc = kmalloc(size, gfp);
235 if (epc) {
236 memset(epc, 0, size);
237 kref_init(&epc->kref);
238 spin_lock_init(&epc->lock);
239 init_waitqueue_head(&epc->waitq);
240 }
241 PDBG("%s alloc ep %p\n", __FUNCTION__, epc);
242 return epc;
243}
244
245void __free_ep(struct kref *kref)
246{
247 struct iwch_ep_common *epc;
248 epc = container_of(kref, struct iwch_ep_common, kref);
249 PDBG("%s ep %p state %s\n", __FUNCTION__, epc, states[state_read(epc)]);
250 kfree(epc);
251}
252
253static void release_ep_resources(struct iwch_ep *ep)
254{
255 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
256 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
257 dst_release(ep->dst);
258 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
259 if (ep->com.tdev->type == T3B)
260 release_tid(ep->com.tdev, ep->hwtid, NULL);
261 put_ep(&ep->com);
262}
263
264static void process_work(struct work_struct *work)
265{
266 struct sk_buff *skb = NULL;
267 void *ep;
268 struct t3cdev *tdev;
269 int ret;
270
271 while ((skb = skb_dequeue(&rxq))) {
272 ep = *((void **) (skb->cb));
273 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
274 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
275 if (ret & CPL_RET_BUF_DONE)
276 kfree_skb(skb);
277
278 /*
279 * ep was referenced in sched(), and is freed here.
280 */
281 put_ep((struct iwch_ep_common *)ep);
282 }
283}
284
285static int status2errno(int status)
286{
287 switch (status) {
288 case CPL_ERR_NONE:
289 return 0;
290 case CPL_ERR_CONN_RESET:
291 return -ECONNRESET;
292 case CPL_ERR_ARP_MISS:
293 return -EHOSTUNREACH;
294 case CPL_ERR_CONN_TIMEDOUT:
295 return -ETIMEDOUT;
296 case CPL_ERR_TCAM_FULL:
297 return -ENOMEM;
298 case CPL_ERR_CONN_EXIST:
299 return -EADDRINUSE;
300 default:
301 return -EIO;
302 }
303}
304
305/*
306 * Try and reuse skbs already allocated...
307 */
308static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
309{
310 if (skb) {
311 BUG_ON(skb_cloned(skb));
312 skb_trim(skb, 0);
313 skb_get(skb);
314 } else {
315 skb = alloc_skb(len, gfp);
316 }
317 return skb;
318}
319
320static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
321 __be32 peer_ip, __be16 local_port,
322 __be16 peer_port, u8 tos)
323{
324 struct rtable *rt;
325 struct flowi fl = {
326 .oif = 0,
327 .nl_u = {
328 .ip4_u = {
329 .daddr = peer_ip,
330 .saddr = local_ip,
331 .tos = tos}
332 },
333 .proto = IPPROTO_TCP,
334 .uli_u = {
335 .ports = {
336 .sport = local_port,
337 .dport = peer_port}
338 }
339 };
340
341 if (ip_route_output_flow(&rt, &fl, NULL, 0))
342 return NULL;
343 return rt;
344}
345
346static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
347{
348 int i = 0;
349
350 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
351 ++i;
352 return i;
353}
354
355static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
356{
357 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
358 kfree_skb(skb);
359}
360
361/*
362 * Handle an ARP failure for an active open.
363 */
364static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
365{
366 printk(KERN_ERR MOD "ARP failure duing connect\n");
367 kfree_skb(skb);
368}
369
370/*
371 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
372 * and send it along.
373 */
374static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
375{
376 struct cpl_abort_req *req = cplhdr(skb);
377
378 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
379 req->cmd = CPL_ABORT_NO_RST;
380 cxgb3_ofld_send(dev, skb);
381}
382
383static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
384{
385 struct cpl_close_con_req *req;
386 struct sk_buff *skb;
387
388 PDBG("%s ep %p\n", __FUNCTION__, ep);
389 skb = get_skb(NULL, sizeof(*req), gfp);
390 if (!skb) {
391 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
392 return -ENOMEM;
393 }
394 skb->priority = CPL_PRIORITY_DATA;
395 set_arp_failure_handler(skb, arp_failure_discard);
396 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
397 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
398 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
399 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
400 l2t_send(ep->com.tdev, skb, ep->l2t);
401 return 0;
402}
403
404static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
405{
406 struct cpl_abort_req *req;
407
408 PDBG("%s ep %p\n", __FUNCTION__, ep);
409 skb = get_skb(skb, sizeof(*req), gfp);
410 if (!skb) {
411 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
412 __FUNCTION__);
413 return -ENOMEM;
414 }
415 skb->priority = CPL_PRIORITY_DATA;
416 set_arp_failure_handler(skb, abort_arp_failure);
417 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
418 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
419 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
420 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
421 req->cmd = CPL_ABORT_SEND_RST;
422 l2t_send(ep->com.tdev, skb, ep->l2t);
423 return 0;
424}
425
426static int send_connect(struct iwch_ep *ep)
427{
428 struct cpl_act_open_req *req;
429 struct sk_buff *skb;
430 u32 opt0h, opt0l, opt2;
431 unsigned int mtu_idx;
432 int wscale;
433
434 PDBG("%s ep %p\n", __FUNCTION__, ep);
435
436 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
437 if (!skb) {
438 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
439 __FUNCTION__);
440 return -ENOMEM;
441 }
442 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
443 wscale = compute_wscale(rcv_win);
444 opt0h = V_NAGLE(0) |
445 V_NO_CONG(nocong) |
446 V_KEEP_ALIVE(1) |
447 F_TCAM_BYPASS |
448 V_WND_SCALE(wscale) |
449 V_MSS_IDX(mtu_idx) |
450 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
451 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
452 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
453 skb->priority = CPL_PRIORITY_SETUP;
454 set_arp_failure_handler(skb, act_open_req_arp_failure);
455
456 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
457 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
458 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
459 req->local_port = ep->com.local_addr.sin_port;
460 req->peer_port = ep->com.remote_addr.sin_port;
461 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
462 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
463 req->opt0h = htonl(opt0h);
464 req->opt0l = htonl(opt0l);
465 req->params = 0;
466 req->opt2 = htonl(opt2);
467 l2t_send(ep->com.tdev, skb, ep->l2t);
468 return 0;
469}
470
471static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
472{
473 int mpalen;
474 struct tx_data_wr *req;
475 struct mpa_message *mpa;
476 int len;
477
478 PDBG("%s ep %p pd_len %d\n", __FUNCTION__, ep, ep->plen);
479
480 BUG_ON(skb_cloned(skb));
481
482 mpalen = sizeof(*mpa) + ep->plen;
483 if (skb->data + mpalen + sizeof(*req) > skb->end) {
484 kfree_skb(skb);
485 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
486 if (!skb) {
487 connect_reply_upcall(ep, -ENOMEM);
488 return;
489 }
490 }
491 skb_trim(skb, 0);
492 skb_reserve(skb, sizeof(*req));
493 skb_put(skb, mpalen);
494 skb->priority = CPL_PRIORITY_DATA;
495 mpa = (struct mpa_message *) skb->data;
496 memset(mpa, 0, sizeof(*mpa));
497 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
498 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
499 (markers_enabled ? MPA_MARKERS : 0);
500 mpa->private_data_size = htons(ep->plen);
501 mpa->revision = mpa_rev;
502
503 if (ep->plen)
504 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
505
506 /*
507 * Reference the mpa skb. This ensures the data area
508 * will remain in memory until the hw acks the tx.
509 * Function tx_ack() will deref it.
510 */
511 skb_get(skb);
512 set_arp_failure_handler(skb, arp_failure_discard);
513 skb->h.raw = skb->data;
514 len = skb->len;
515 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
516 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
517 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
518 req->len = htonl(len);
519 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
520 V_TX_SNDBUF(snd_win>>15));
521 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
522 req->sndseq = htonl(ep->snd_seq);
523 BUG_ON(ep->mpa_skb);
524 ep->mpa_skb = skb;
525 l2t_send(ep->com.tdev, skb, ep->l2t);
526 start_ep_timer(ep);
527 state_set(&ep->com, MPA_REQ_SENT);
528 return;
529}
530
531static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
532{
533 int mpalen;
534 struct tx_data_wr *req;
535 struct mpa_message *mpa;
536 struct sk_buff *skb;
537
538 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
539
540 mpalen = sizeof(*mpa) + plen;
541
542 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
543 if (!skb) {
544 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
545 return -ENOMEM;
546 }
547 skb_reserve(skb, sizeof(*req));
548 mpa = (struct mpa_message *) skb_put(skb, mpalen);
549 memset(mpa, 0, sizeof(*mpa));
550 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
551 mpa->flags = MPA_REJECT;
552 mpa->revision = mpa_rev;
553 mpa->private_data_size = htons(plen);
554 if (plen)
555 memcpy(mpa->private_data, pdata, plen);
556
557 /*
558 * Reference the mpa skb again. This ensures the data area
559 * will remain in memory until the hw acks the tx.
560 * Function tx_ack() will deref it.
561 */
562 skb_get(skb);
563 skb->priority = CPL_PRIORITY_DATA;
564 set_arp_failure_handler(skb, arp_failure_discard);
565 skb->h.raw = skb->data;
566 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
567 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
568 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
569 req->len = htonl(mpalen);
570 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
571 V_TX_SNDBUF(snd_win>>15));
572 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
573 req->sndseq = htonl(ep->snd_seq);
574 BUG_ON(ep->mpa_skb);
575 ep->mpa_skb = skb;
576 l2t_send(ep->com.tdev, skb, ep->l2t);
577 return 0;
578}
579
580static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
581{
582 int mpalen;
583 struct tx_data_wr *req;
584 struct mpa_message *mpa;
585 int len;
586 struct sk_buff *skb;
587
588 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
589
590 mpalen = sizeof(*mpa) + plen;
591
592 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
593 if (!skb) {
594 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
595 return -ENOMEM;
596 }
597 skb->priority = CPL_PRIORITY_DATA;
598 skb_reserve(skb, sizeof(*req));
599 mpa = (struct mpa_message *) skb_put(skb, mpalen);
600 memset(mpa, 0, sizeof(*mpa));
601 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
602 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
603 (markers_enabled ? MPA_MARKERS : 0);
604 mpa->revision = mpa_rev;
605 mpa->private_data_size = htons(plen);
606 if (plen)
607 memcpy(mpa->private_data, pdata, plen);
608
609 /*
610 * Reference the mpa skb. This ensures the data area
611 * will remain in memory until the hw acks the tx.
612 * Function tx_ack() will deref it.
613 */
614 skb_get(skb);
615 set_arp_failure_handler(skb, arp_failure_discard);
616 skb->h.raw = skb->data;
617 len = skb->len;
618 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
619 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
620 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
621 req->len = htonl(len);
622 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
623 V_TX_SNDBUF(snd_win>>15));
624 req->flags = htonl(F_TX_MORE | F_TX_IMM_ACK | F_TX_INIT);
625 req->sndseq = htonl(ep->snd_seq);
626 ep->mpa_skb = skb;
627 state_set(&ep->com, MPA_REP_SENT);
628 l2t_send(ep->com.tdev, skb, ep->l2t);
629 return 0;
630}
631
632static 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", __FUNCTION__, 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
648 set_emss(ep, ntohs(req->tcp_opt));
649
650 /* dealloc the atid */
651 cxgb3_free_atid(ep->com.tdev, ep->atid);
652
653 /* start MPA negotiation */
654 send_mpa_req(ep, skb);
655
656 return 0;
657}
658
659static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
660{
661 PDBG("%s ep %p\n", __FILE__, ep);
662 state_set(&ep->com, ABORTING);
663 send_abort(ep, skb, gfp);
664}
665
666static void close_complete_upcall(struct iwch_ep *ep)
667{
668 struct iw_cm_event event;
669
670 PDBG("%s ep %p\n", __FUNCTION__, ep);
671 memset(&event, 0, sizeof(event));
672 event.event = IW_CM_EVENT_CLOSE;
673 if (ep->com.cm_id) {
674 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
675 ep, ep->com.cm_id, ep->hwtid);
676 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
677 ep->com.cm_id->rem_ref(ep->com.cm_id);
678 ep->com.cm_id = NULL;
679 ep->com.qp = NULL;
680 }
681}
682
683static void peer_close_upcall(struct iwch_ep *ep)
684{
685 struct iw_cm_event event;
686
687 PDBG("%s ep %p\n", __FUNCTION__, ep);
688 memset(&event, 0, sizeof(event));
689 event.event = IW_CM_EVENT_DISCONNECT;
690 if (ep->com.cm_id) {
691 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
692 ep, ep->com.cm_id, ep->hwtid);
693 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
694 }
695}
696
697static void peer_abort_upcall(struct iwch_ep *ep)
698{
699 struct iw_cm_event event;
700
701 PDBG("%s ep %p\n", __FUNCTION__, ep);
702 memset(&event, 0, sizeof(event));
703 event.event = IW_CM_EVENT_CLOSE;
704 event.status = -ECONNRESET;
705 if (ep->com.cm_id) {
706 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
707 ep->com.cm_id, ep->hwtid);
708 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
709 ep->com.cm_id->rem_ref(ep->com.cm_id);
710 ep->com.cm_id = NULL;
711 ep->com.qp = NULL;
712 }
713}
714
715static void connect_reply_upcall(struct iwch_ep *ep, int status)
716{
717 struct iw_cm_event event;
718
719 PDBG("%s ep %p status %d\n", __FUNCTION__, ep, status);
720 memset(&event, 0, sizeof(event));
721 event.event = IW_CM_EVENT_CONNECT_REPLY;
722 event.status = status;
723 event.local_addr = ep->com.local_addr;
724 event.remote_addr = ep->com.remote_addr;
725
726 if ((status == 0) || (status == -ECONNREFUSED)) {
727 event.private_data_len = ep->plen;
728 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
729 }
730 if (ep->com.cm_id) {
731 PDBG("%s ep %p tid %d status %d\n", __FUNCTION__, ep,
732 ep->hwtid, status);
733 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
734 }
735 if (status < 0) {
736 ep->com.cm_id->rem_ref(ep->com.cm_id);
737 ep->com.cm_id = NULL;
738 ep->com.qp = NULL;
739 }
740}
741
742static void connect_request_upcall(struct iwch_ep *ep)
743{
744 struct iw_cm_event event;
745
746 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
747 memset(&event, 0, sizeof(event));
748 event.event = IW_CM_EVENT_CONNECT_REQUEST;
749 event.local_addr = ep->com.local_addr;
750 event.remote_addr = ep->com.remote_addr;
751 event.private_data_len = ep->plen;
752 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
753 event.provider_data = ep;
754 if (state_read(&ep->parent_ep->com) != DEAD)
755 ep->parent_ep->com.cm_id->event_handler(
756 ep->parent_ep->com.cm_id,
757 &event);
758 put_ep(&ep->parent_ep->com);
759 ep->parent_ep = NULL;
760}
761
762static void established_upcall(struct iwch_ep *ep)
763{
764 struct iw_cm_event event;
765
766 PDBG("%s ep %p\n", __FUNCTION__, ep);
767 memset(&event, 0, sizeof(event));
768 event.event = IW_CM_EVENT_ESTABLISHED;
769 if (ep->com.cm_id) {
770 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
771 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
772 }
773}
774
775static int update_rx_credits(struct iwch_ep *ep, u32 credits)
776{
777 struct cpl_rx_data_ack *req;
778 struct sk_buff *skb;
779
780 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
781 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
782 if (!skb) {
783 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
784 return 0;
785 }
786
787 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
788 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
789 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
790 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
791 skb->priority = CPL_PRIORITY_ACK;
792 ep->com.tdev->send(ep->com.tdev, skb);
793 return credits;
794}
795
796static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
797{
798 struct mpa_message *mpa;
799 u16 plen;
800 struct iwch_qp_attributes attrs;
801 enum iwch_qp_attr_mask mask;
802 int err;
803
804 PDBG("%s ep %p\n", __FUNCTION__, ep);
805
806 /*
807 * Stop mpa timer. If it expired, then the state has
808 * changed and we bail since ep_timeout already aborted
809 * the connection.
810 */
811 stop_ep_timer(ep);
812 if (state_read(&ep->com) != MPA_REQ_SENT)
813 return;
814
815 /*
816 * If we get more than the supported amount of private data
817 * then we must fail this connection.
818 */
819 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
820 err = -EINVAL;
821 goto err;
822 }
823
824 /*
825 * copy the new data into our accumulation buffer.
826 */
827 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
828 ep->mpa_pkt_len += skb->len;
829
830 /*
831 * if we don't even have the mpa message, then bail.
832 */
833 if (ep->mpa_pkt_len < sizeof(*mpa))
834 return;
835 mpa = (struct mpa_message *) ep->mpa_pkt;
836
837 /* Validate MPA header. */
838 if (mpa->revision != mpa_rev) {
839 err = -EPROTO;
840 goto err;
841 }
842 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
843 err = -EPROTO;
844 goto err;
845 }
846
847 plen = ntohs(mpa->private_data_size);
848
849 /*
850 * Fail if there's too much private data.
851 */
852 if (plen > MPA_MAX_PRIVATE_DATA) {
853 err = -EPROTO;
854 goto err;
855 }
856
857 /*
858 * If plen does not account for pkt size
859 */
860 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
861 err = -EPROTO;
862 goto err;
863 }
864
865 ep->plen = (u8) plen;
866
867 /*
868 * If we don't have all the pdata yet, then bail.
869 * We'll continue process when more data arrives.
870 */
871 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
872 return;
873
874 if (mpa->flags & MPA_REJECT) {
875 err = -ECONNREFUSED;
876 goto err;
877 }
878
879 /*
880 * If we get here we have accumulated the entire mpa
881 * start reply message including private data. And
882 * the MPA header is valid.
883 */
884 state_set(&ep->com, FPDU_MODE);
885 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
886 ep->mpa_attr.recv_marker_enabled = markers_enabled;
887 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
888 ep->mpa_attr.version = mpa_rev;
889 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
890 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
891 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
892 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
893
894 attrs.mpa_attr = ep->mpa_attr;
895 attrs.max_ird = ep->ird;
896 attrs.max_ord = ep->ord;
897 attrs.llp_stream_handle = ep;
898 attrs.next_state = IWCH_QP_STATE_RTS;
899
900 mask = IWCH_QP_ATTR_NEXT_STATE |
901 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
902 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
903
904 /* bind QP and TID with INIT_WR */
905 err = iwch_modify_qp(ep->com.qp->rhp,
906 ep->com.qp, mask, &attrs, 1);
907 if (!err)
908 goto out;
909err:
910 abort_connection(ep, skb, GFP_KERNEL);
911out:
912 connect_reply_upcall(ep, err);
913 return;
914}
915
916static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
917{
918 struct mpa_message *mpa;
919 u16 plen;
920
921 PDBG("%s ep %p\n", __FUNCTION__, ep);
922
923 /*
924 * Stop mpa timer. If it expired, then the state has
925 * changed and we bail since ep_timeout already aborted
926 * the connection.
927 */
928 stop_ep_timer(ep);
929 if (state_read(&ep->com) != MPA_REQ_WAIT)
930 return;
931
932 /*
933 * If we get more than the supported amount of private data
934 * then we must fail this connection.
935 */
936 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
937 abort_connection(ep, skb, GFP_KERNEL);
938 return;
939 }
940
941 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
942
943 /*
944 * Copy the new data into our accumulation buffer.
945 */
946 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
947 ep->mpa_pkt_len += skb->len;
948
949 /*
950 * If we don't even have the mpa message, then bail.
951 * We'll continue process when more data arrives.
952 */
953 if (ep->mpa_pkt_len < sizeof(*mpa))
954 return;
955 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
956 mpa = (struct mpa_message *) ep->mpa_pkt;
957
958 /*
959 * Validate MPA Header.
960 */
961 if (mpa->revision != mpa_rev) {
962 abort_connection(ep, skb, GFP_KERNEL);
963 return;
964 }
965
966 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
967 abort_connection(ep, skb, GFP_KERNEL);
968 return;
969 }
970
971 plen = ntohs(mpa->private_data_size);
972
973 /*
974 * Fail if there's too much private data.
975 */
976 if (plen > MPA_MAX_PRIVATE_DATA) {
977 abort_connection(ep, skb, GFP_KERNEL);
978 return;
979 }
980
981 /*
982 * If plen does not account for pkt size
983 */
984 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
985 abort_connection(ep, skb, GFP_KERNEL);
986 return;
987 }
988 ep->plen = (u8) plen;
989
990 /*
991 * If we don't have all the pdata yet, then bail.
992 */
993 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
994 return;
995
996 /*
997 * If we get here we have accumulated the entire mpa
998 * start reply message including private data.
999 */
1000 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1001 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1002 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1003 ep->mpa_attr.version = mpa_rev;
1004 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1005 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
1006 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1007 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1008
1009 state_set(&ep->com, MPA_REQ_RCVD);
1010
1011 /* drive upcall */
1012 connect_request_upcall(ep);
1013 return;
1014}
1015
1016static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1017{
1018 struct iwch_ep *ep = ctx;
1019 struct cpl_rx_data *hdr = cplhdr(skb);
1020 unsigned int dlen = ntohs(hdr->len);
1021
1022 PDBG("%s ep %p dlen %u\n", __FUNCTION__, ep, dlen);
1023
1024 skb_pull(skb, sizeof(*hdr));
1025 skb_trim(skb, dlen);
1026
1027 switch (state_read(&ep->com)) {
1028 case MPA_REQ_SENT:
1029 process_mpa_reply(ep, skb);
1030 break;
1031 case MPA_REQ_WAIT:
1032 process_mpa_request(ep, skb);
1033 break;
1034 case MPA_REP_SENT:
1035 break;
1036 default:
1037 printk(KERN_ERR MOD "%s Unexpected streaming data."
1038 " ep %p state %d tid %d\n",
1039 __FUNCTION__, ep, state_read(&ep->com), ep->hwtid);
1040
1041 /*
1042 * The ep will timeout and inform the ULP of the failure.
1043 * See ep_timeout().
1044 */
1045 break;
1046 }
1047
1048 /* update RX credits */
1049 update_rx_credits(ep, dlen);
1050
1051 return CPL_RET_BUF_DONE;
1052}
1053
1054/*
1055 * Upcall from the adapter indicating data has been transmitted.
1056 * For us its just the single MPA request or reply. We can now free
1057 * the skb holding the mpa message.
1058 */
1059static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1060{
1061 struct iwch_ep *ep = ctx;
1062 struct cpl_wr_ack *hdr = cplhdr(skb);
1063 unsigned int credits = ntohs(hdr->credits);
1064 enum iwch_qp_attr_mask mask;
1065
1066 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
1067
1068 if (credits == 0)
1069 return CPL_RET_BUF_DONE;
1070 BUG_ON(credits != 1);
1071 BUG_ON(ep->mpa_skb == NULL);
1072 kfree_skb(ep->mpa_skb);
1073 ep->mpa_skb = NULL;
1074 dst_confirm(ep->dst);
1075 if (state_read(&ep->com) == MPA_REP_SENT) {
1076 struct iwch_qp_attributes attrs;
1077
1078 /* bind QP to EP and move to RTS */
1079 attrs.mpa_attr = ep->mpa_attr;
1080 attrs.max_ird = ep->ord;
1081 attrs.max_ord = ep->ord;
1082 attrs.llp_stream_handle = ep;
1083 attrs.next_state = IWCH_QP_STATE_RTS;
1084
1085 /* bind QP and TID with INIT_WR */
1086 mask = IWCH_QP_ATTR_NEXT_STATE |
1087 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1088 IWCH_QP_ATTR_MPA_ATTR |
1089 IWCH_QP_ATTR_MAX_IRD |
1090 IWCH_QP_ATTR_MAX_ORD;
1091
1092 ep->com.rpl_err = iwch_modify_qp(ep->com.qp->rhp,
1093 ep->com.qp, mask, &attrs, 1);
1094
1095 if (!ep->com.rpl_err) {
1096 state_set(&ep->com, FPDU_MODE);
1097 established_upcall(ep);
1098 }
1099
1100 ep->com.rpl_done = 1;
1101 PDBG("waking up ep %p\n", ep);
1102 wake_up(&ep->com.waitq);
1103 }
1104 return CPL_RET_BUF_DONE;
1105}
1106
1107static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1108{
1109 struct iwch_ep *ep = ctx;
1110
1111 PDBG("%s ep %p\n", __FUNCTION__, ep);
1112
1113 close_complete_upcall(ep);
1114 state_set(&ep->com, DEAD);
1115 release_ep_resources(ep);
1116 return CPL_RET_BUF_DONE;
1117}
1118
1119static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1120{
1121 struct iwch_ep *ep = ctx;
1122 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1123
1124 PDBG("%s ep %p status %u errno %d\n", __FUNCTION__, ep, rpl->status,
1125 status2errno(rpl->status));
1126 connect_reply_upcall(ep, status2errno(rpl->status));
1127 state_set(&ep->com, DEAD);
1128 if (ep->com.tdev->type == T3B)
1129 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1130 cxgb3_free_atid(ep->com.tdev, ep->atid);
1131 dst_release(ep->dst);
1132 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1133 put_ep(&ep->com);
1134 return CPL_RET_BUF_DONE;
1135}
1136
1137static int listen_start(struct iwch_listen_ep *ep)
1138{
1139 struct sk_buff *skb;
1140 struct cpl_pass_open_req *req;
1141
1142 PDBG("%s ep %p\n", __FUNCTION__, ep);
1143 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1144 if (!skb) {
1145 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1146 return -ENOMEM;
1147 }
1148
1149 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1150 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1151 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1152 req->local_port = ep->com.local_addr.sin_port;
1153 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1154 req->peer_port = 0;
1155 req->peer_ip = 0;
1156 req->peer_netmask = 0;
1157 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1158 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1159 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1160
1161 skb->priority = 1;
1162 ep->com.tdev->send(ep->com.tdev, skb);
1163 return 0;
1164}
1165
1166static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1167{
1168 struct iwch_listen_ep *ep = ctx;
1169 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1170
1171 PDBG("%s ep %p status %d error %d\n", __FUNCTION__, ep,
1172 rpl->status, status2errno(rpl->status));
1173 ep->com.rpl_err = status2errno(rpl->status);
1174 ep->com.rpl_done = 1;
1175 wake_up(&ep->com.waitq);
1176
1177 return CPL_RET_BUF_DONE;
1178}
1179
1180static int listen_stop(struct iwch_listen_ep *ep)
1181{
1182 struct sk_buff *skb;
1183 struct cpl_close_listserv_req *req;
1184
1185 PDBG("%s ep %p\n", __FUNCTION__, ep);
1186 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1187 if (!skb) {
1188 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
1189 return -ENOMEM;
1190 }
1191 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1192 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1193 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1194 skb->priority = 1;
1195 ep->com.tdev->send(ep->com.tdev, skb);
1196 return 0;
1197}
1198
1199static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1200 void *ctx)
1201{
1202 struct iwch_listen_ep *ep = ctx;
1203 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1204
1205 PDBG("%s ep %p\n", __FUNCTION__, ep);
1206 ep->com.rpl_err = status2errno(rpl->status);
1207 ep->com.rpl_done = 1;
1208 wake_up(&ep->com.waitq);
1209 return CPL_RET_BUF_DONE;
1210}
1211
1212static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1213{
1214 struct cpl_pass_accept_rpl *rpl;
1215 unsigned int mtu_idx;
1216 u32 opt0h, opt0l, opt2;
1217 int wscale;
1218
1219 PDBG("%s ep %p\n", __FUNCTION__, ep);
1220 BUG_ON(skb_cloned(skb));
1221 skb_trim(skb, sizeof(*rpl));
1222 skb_get(skb);
1223 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1224 wscale = compute_wscale(rcv_win);
1225 opt0h = V_NAGLE(0) |
1226 V_NO_CONG(nocong) |
1227 V_KEEP_ALIVE(1) |
1228 F_TCAM_BYPASS |
1229 V_WND_SCALE(wscale) |
1230 V_MSS_IDX(mtu_idx) |
1231 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1232 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1233 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1234
1235 rpl = cplhdr(skb);
1236 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1237 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1238 rpl->peer_ip = peer_ip;
1239 rpl->opt0h = htonl(opt0h);
1240 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1241 rpl->opt2 = htonl(opt2);
1242 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1243 skb->priority = CPL_PRIORITY_SETUP;
1244 l2t_send(ep->com.tdev, skb, ep->l2t);
1245
1246 return;
1247}
1248
1249static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1250 struct sk_buff *skb)
1251{
1252 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __FUNCTION__, tdev, hwtid,
1253 peer_ip);
1254 BUG_ON(skb_cloned(skb));
1255 skb_trim(skb, sizeof(struct cpl_tid_release));
1256 skb_get(skb);
1257
1258 if (tdev->type == T3B)
1259 release_tid(tdev, hwtid, skb);
1260 else {
1261 struct cpl_pass_accept_rpl *rpl;
1262
1263 rpl = cplhdr(skb);
1264 skb->priority = CPL_PRIORITY_SETUP;
1265 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1266 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1267 hwtid));
1268 rpl->peer_ip = peer_ip;
1269 rpl->opt0h = htonl(F_TCAM_BYPASS);
1270 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1271 rpl->opt2 = 0;
1272 rpl->rsvd = rpl->opt2;
1273 tdev->send(tdev, skb);
1274 }
1275}
1276
1277static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1278{
1279 struct iwch_ep *child_ep, *parent_ep = ctx;
1280 struct cpl_pass_accept_req *req = cplhdr(skb);
1281 unsigned int hwtid = GET_TID(req);
1282 struct dst_entry *dst;
1283 struct l2t_entry *l2t;
1284 struct rtable *rt;
1285 struct iff_mac tim;
1286
1287 PDBG("%s parent ep %p tid %u\n", __FUNCTION__, parent_ep, hwtid);
1288
1289 if (state_read(&parent_ep->com) != LISTEN) {
1290 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1291 __FUNCTION__);
1292 goto reject;
1293 }
1294
1295 /*
1296 * Find the netdev for this connection request.
1297 */
1298 tim.mac_addr = req->dst_mac;
1299 tim.vlan_tag = ntohs(req->vlan_tag);
1300 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1301 printk(KERN_ERR
1302 "%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
1303 __FUNCTION__,
1304 req->dst_mac[0],
1305 req->dst_mac[1],
1306 req->dst_mac[2],
1307 req->dst_mac[3],
1308 req->dst_mac[4],
1309 req->dst_mac[5]);
1310 goto reject;
1311 }
1312
1313 /* Find output route */
1314 rt = find_route(tdev,
1315 req->local_ip,
1316 req->peer_ip,
1317 req->local_port,
1318 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1319 if (!rt) {
1320 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1321 __FUNCTION__);
1322 goto reject;
1323 }
1324 dst = &rt->u.dst;
1325 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1326 if (!l2t) {
1327 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1328 __FUNCTION__);
1329 dst_release(dst);
1330 goto reject;
1331 }
1332 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1333 if (!child_ep) {
1334 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1335 __FUNCTION__);
1336 l2t_release(L2DATA(tdev), l2t);
1337 dst_release(dst);
1338 goto reject;
1339 }
1340 state_set(&child_ep->com, CONNECTING);
1341 child_ep->com.tdev = tdev;
1342 child_ep->com.cm_id = NULL;
1343 child_ep->com.local_addr.sin_family = PF_INET;
1344 child_ep->com.local_addr.sin_port = req->local_port;
1345 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1346 child_ep->com.remote_addr.sin_family = PF_INET;
1347 child_ep->com.remote_addr.sin_port = req->peer_port;
1348 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1349 get_ep(&parent_ep->com);
1350 child_ep->parent_ep = parent_ep;
1351 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1352 child_ep->l2t = l2t;
1353 child_ep->dst = dst;
1354 child_ep->hwtid = hwtid;
1355 init_timer(&child_ep->timer);
1356 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1357 accept_cr(child_ep, req->peer_ip, skb);
1358 goto out;
1359reject:
1360 reject_cr(tdev, hwtid, req->peer_ip, skb);
1361out:
1362 return CPL_RET_BUF_DONE;
1363}
1364
1365static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1366{
1367 struct iwch_ep *ep = ctx;
1368 struct cpl_pass_establish *req = cplhdr(skb);
1369
1370 PDBG("%s ep %p\n", __FUNCTION__, ep);
1371 ep->snd_seq = ntohl(req->snd_isn);
1372
1373 set_emss(ep, ntohs(req->tcp_opt));
1374
1375 dst_confirm(ep->dst);
1376 state_set(&ep->com, MPA_REQ_WAIT);
1377 start_ep_timer(ep);
1378
1379 return CPL_RET_BUF_DONE;
1380}
1381
1382static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1383{
1384 struct iwch_ep *ep = ctx;
1385 struct iwch_qp_attributes attrs;
1386 unsigned long flags;
1387 int disconnect = 1;
1388 int release = 0;
1389
1390 PDBG("%s ep %p\n", __FUNCTION__, ep);
1391 dst_confirm(ep->dst);
1392
1393 spin_lock_irqsave(&ep->com.lock, flags);
1394 switch (ep->com.state) {
1395 case MPA_REQ_WAIT:
1396 __state_set(&ep->com, CLOSING);
1397 break;
1398 case MPA_REQ_SENT:
1399 __state_set(&ep->com, CLOSING);
1400 connect_reply_upcall(ep, -ECONNRESET);
1401 break;
1402 case MPA_REQ_RCVD:
1403
1404 /*
1405 * We're gonna mark this puppy DEAD, but keep
1406 * the reference on it until the ULP accepts or
1407 * rejects the CR.
1408 */
1409 __state_set(&ep->com, CLOSING);
1410 get_ep(&ep->com);
1411 break;
1412 case MPA_REP_SENT:
1413 __state_set(&ep->com, CLOSING);
1414 ep->com.rpl_done = 1;
1415 ep->com.rpl_err = -ECONNRESET;
1416 PDBG("waking up ep %p\n", ep);
1417 wake_up(&ep->com.waitq);
1418 break;
1419 case FPDU_MODE:
1420 __state_set(&ep->com, CLOSING);
1421 attrs.next_state = IWCH_QP_STATE_CLOSING;
1422 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1423 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1424 peer_close_upcall(ep);
1425 break;
1426 case ABORTING:
1427 disconnect = 0;
1428 break;
1429 case CLOSING:
1430 start_ep_timer(ep);
1431 __state_set(&ep->com, MORIBUND);
1432 disconnect = 0;
1433 break;
1434 case MORIBUND:
1435 stop_ep_timer(ep);
1436 if (ep->com.cm_id && ep->com.qp) {
1437 attrs.next_state = IWCH_QP_STATE_IDLE;
1438 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1439 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1440 }
1441 close_complete_upcall(ep);
1442 __state_set(&ep->com, DEAD);
1443 release = 1;
1444 disconnect = 0;
1445 break;
1446 case DEAD:
1447 disconnect = 0;
1448 break;
1449 default:
1450 BUG_ON(1);
1451 }
1452 spin_unlock_irqrestore(&ep->com.lock, flags);
1453 if (disconnect)
1454 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1455 if (release)
1456 release_ep_resources(ep);
1457 return CPL_RET_BUF_DONE;
1458}
1459
1460/*
1461 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1462 */
1463static inline int is_neg_adv_abort(unsigned int status)
1464{
1465 return status == CPL_ERR_RTX_NEG_ADVICE ||
1466 status == CPL_ERR_PERSIST_NEG_ADVICE;
1467}
1468
1469static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1470{
1471 struct cpl_abort_req_rss *req = cplhdr(skb);
1472 struct iwch_ep *ep = ctx;
1473 struct cpl_abort_rpl *rpl;
1474 struct sk_buff *rpl_skb;
1475 struct iwch_qp_attributes attrs;
1476 int ret;
1477 int state;
1478
1479 if (is_neg_adv_abort(req->status)) {
1480 PDBG("%s neg_adv_abort ep %p tid %d\n", __FUNCTION__, ep,
1481 ep->hwtid);
1482 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1483 return CPL_RET_BUF_DONE;
1484 }
1485
1486 state = state_read(&ep->com);
1487 PDBG("%s ep %p state %u\n", __FUNCTION__, ep, state);
1488 switch (state) {
1489 case CONNECTING:
1490 break;
1491 case MPA_REQ_WAIT:
1492 break;
1493 case MPA_REQ_SENT:
1494 connect_reply_upcall(ep, -ECONNRESET);
1495 break;
1496 case MPA_REP_SENT:
1497 ep->com.rpl_done = 1;
1498 ep->com.rpl_err = -ECONNRESET;
1499 PDBG("waking up ep %p\n", ep);
1500 wake_up(&ep->com.waitq);
1501 break;
1502 case MPA_REQ_RCVD:
1503
1504 /*
1505 * We're gonna mark this puppy DEAD, but keep
1506 * the reference on it until the ULP accepts or
1507 * rejects the CR.
1508 */
1509 get_ep(&ep->com);
1510 break;
1511 case MORIBUND:
1512 stop_ep_timer(ep);
1513 case FPDU_MODE:
1514 case CLOSING:
1515 if (ep->com.cm_id && ep->com.qp) {
1516 attrs.next_state = IWCH_QP_STATE_ERROR;
1517 ret = iwch_modify_qp(ep->com.qp->rhp,
1518 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1519 &attrs, 1);
1520 if (ret)
1521 printk(KERN_ERR MOD
1522 "%s - qp <- error failed!\n",
1523 __FUNCTION__);
1524 }
1525 peer_abort_upcall(ep);
1526 break;
1527 case ABORTING:
1528 break;
1529 case DEAD:
1530 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __FUNCTION__);
1531 return CPL_RET_BUF_DONE;
1532 default:
1533 BUG_ON(1);
1534 break;
1535 }
1536 dst_confirm(ep->dst);
1537
1538 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1539 if (!rpl_skb) {
1540 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1541 __FUNCTION__);
1542 dst_release(ep->dst);
1543 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1544 put_ep(&ep->com);
1545 return CPL_RET_BUF_DONE;
1546 }
1547 rpl_skb->priority = CPL_PRIORITY_DATA;
1548 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1549 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1550 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1551 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1552 rpl->cmd = CPL_ABORT_NO_RST;
1553 ep->com.tdev->send(ep->com.tdev, rpl_skb);
1554 if (state != ABORTING) {
1555 state_set(&ep->com, DEAD);
1556 release_ep_resources(ep);
1557 }
1558 return CPL_RET_BUF_DONE;
1559}
1560
1561static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1562{
1563 struct iwch_ep *ep = ctx;
1564 struct iwch_qp_attributes attrs;
1565 unsigned long flags;
1566 int release = 0;
1567
1568 PDBG("%s ep %p\n", __FUNCTION__, ep);
1569 BUG_ON(!ep);
1570
1571 /* The cm_id may be null if we failed to connect */
1572 spin_lock_irqsave(&ep->com.lock, flags);
1573 switch (ep->com.state) {
1574 case CLOSING:
1575 start_ep_timer(ep);
1576 __state_set(&ep->com, MORIBUND);
1577 break;
1578 case MORIBUND:
1579 stop_ep_timer(ep);
1580 if ((ep->com.cm_id) && (ep->com.qp)) {
1581 attrs.next_state = IWCH_QP_STATE_IDLE;
1582 iwch_modify_qp(ep->com.qp->rhp,
1583 ep->com.qp,
1584 IWCH_QP_ATTR_NEXT_STATE,
1585 &attrs, 1);
1586 }
1587 close_complete_upcall(ep);
1588 __state_set(&ep->com, DEAD);
1589 release = 1;
1590 break;
1591 case DEAD:
1592 default:
1593 BUG_ON(1);
1594 break;
1595 }
1596 spin_unlock_irqrestore(&ep->com.lock, flags);
1597 if (release)
1598 release_ep_resources(ep);
1599 return CPL_RET_BUF_DONE;
1600}
1601
1602/*
1603 * T3A does 3 things when a TERM is received:
1604 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1605 * 2) generate an async event on the QP with the TERMINATE opcode
1606 * 3) post a TERMINATE opcde cqe into the associated CQ.
1607 *
1608 * For (1), we save the message in the qp for later consumer consumption.
1609 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1610 * For (3), we toss the CQE in cxio_poll_cq().
1611 *
1612 * terminate() handles case (1)...
1613 */
1614static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1615{
1616 struct iwch_ep *ep = ctx;
1617
1618 PDBG("%s ep %p\n", __FUNCTION__, ep);
1619 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1620 PDBG("%s saving %d bytes of term msg\n", __FUNCTION__, skb->len);
1621 memcpy(ep->com.qp->attr.terminate_buffer, skb->data, skb->len);
1622 ep->com.qp->attr.terminate_msg_len = skb->len;
1623 ep->com.qp->attr.is_terminate_local = 0;
1624 return CPL_RET_BUF_DONE;
1625}
1626
1627static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1628{
1629 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1630 struct iwch_ep *ep = ctx;
1631
1632 PDBG("%s ep %p tid %u status %d\n", __FUNCTION__, ep, ep->hwtid,
1633 rep->status);
1634 if (rep->status) {
1635 struct iwch_qp_attributes attrs;
1636
1637 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1638 __FUNCTION__, ep->hwtid);
1639 attrs.next_state = IWCH_QP_STATE_ERROR;
1640 iwch_modify_qp(ep->com.qp->rhp,
1641 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1642 &attrs, 1);
1643 abort_connection(ep, NULL, GFP_KERNEL);
1644 }
1645 return CPL_RET_BUF_DONE;
1646}
1647
1648static void ep_timeout(unsigned long arg)
1649{
1650 struct iwch_ep *ep = (struct iwch_ep *)arg;
1651 struct iwch_qp_attributes attrs;
1652 unsigned long flags;
1653
1654 spin_lock_irqsave(&ep->com.lock, flags);
1655 PDBG("%s ep %p tid %u state %d\n", __FUNCTION__, ep, ep->hwtid,
1656 ep->com.state);
1657 switch (ep->com.state) {
1658 case MPA_REQ_SENT:
1659 connect_reply_upcall(ep, -ETIMEDOUT);
1660 break;
1661 case MPA_REQ_WAIT:
1662 break;
1663 case MORIBUND:
1664 if (ep->com.cm_id && ep->com.qp) {
1665 attrs.next_state = IWCH_QP_STATE_ERROR;
1666 iwch_modify_qp(ep->com.qp->rhp,
1667 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1668 &attrs, 1);
1669 }
1670 break;
1671 default:
1672 BUG();
1673 }
1674 __state_set(&ep->com, CLOSING);
1675 spin_unlock_irqrestore(&ep->com.lock, flags);
1676 abort_connection(ep, NULL, GFP_ATOMIC);
1677 put_ep(&ep->com);
1678}
1679
1680int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1681{
1682 int err;
1683 struct iwch_ep *ep = to_ep(cm_id);
1684 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1685
1686 if (state_read(&ep->com) == DEAD) {
1687 put_ep(&ep->com);
1688 return -ECONNRESET;
1689 }
1690 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1691 state_set(&ep->com, CLOSING);
1692 if (mpa_rev == 0)
1693 abort_connection(ep, NULL, GFP_KERNEL);
1694 else {
1695 err = send_mpa_reject(ep, pdata, pdata_len);
1696 err = send_halfclose(ep, GFP_KERNEL);
1697 }
1698 return 0;
1699}
1700
1701int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1702{
1703 int err;
1704 struct iwch_qp_attributes attrs;
1705 enum iwch_qp_attr_mask mask;
1706 struct iwch_ep *ep = to_ep(cm_id);
1707 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1708 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1709
1710 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1711 if (state_read(&ep->com) == DEAD) {
1712 put_ep(&ep->com);
1713 return -ECONNRESET;
1714 }
1715
1716 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1717 BUG_ON(!qp);
1718
1719 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1720 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1721 abort_connection(ep, NULL, GFP_KERNEL);
1722 return -EINVAL;
1723 }
1724
1725 cm_id->add_ref(cm_id);
1726 ep->com.cm_id = cm_id;
1727 ep->com.qp = qp;
1728
1729 ep->com.rpl_done = 0;
1730 ep->com.rpl_err = 0;
1731 ep->ird = conn_param->ird;
1732 ep->ord = conn_param->ord;
1733 PDBG("%s %d ird %d ord %d\n", __FUNCTION__, __LINE__, ep->ird, ep->ord);
1734 get_ep(&ep->com);
1735 err = send_mpa_reply(ep, conn_param->private_data,
1736 conn_param->private_data_len);
1737 if (err) {
1738 ep->com.cm_id = NULL;
1739 ep->com.qp = NULL;
1740 cm_id->rem_ref(cm_id);
1741 abort_connection(ep, NULL, GFP_KERNEL);
1742 put_ep(&ep->com);
1743 return err;
1744 }
1745
1746 /* bind QP to EP and move to RTS */
1747 attrs.mpa_attr = ep->mpa_attr;
1748 attrs.max_ird = ep->ord;
1749 attrs.max_ord = ep->ord;
1750 attrs.llp_stream_handle = ep;
1751 attrs.next_state = IWCH_QP_STATE_RTS;
1752
1753 /* bind QP and TID with INIT_WR */
1754 mask = IWCH_QP_ATTR_NEXT_STATE |
1755 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1756 IWCH_QP_ATTR_MPA_ATTR |
1757 IWCH_QP_ATTR_MAX_IRD |
1758 IWCH_QP_ATTR_MAX_ORD;
1759
1760 err = iwch_modify_qp(ep->com.qp->rhp,
1761 ep->com.qp, mask, &attrs, 1);
1762
1763 if (err) {
1764 ep->com.cm_id = NULL;
1765 ep->com.qp = NULL;
1766 cm_id->rem_ref(cm_id);
1767 abort_connection(ep, NULL, GFP_KERNEL);
1768 } else {
1769 state_set(&ep->com, FPDU_MODE);
1770 established_upcall(ep);
1771 }
1772 put_ep(&ep->com);
1773 return err;
1774}
1775
1776int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1777{
1778 int err = 0;
1779 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1780 struct iwch_ep *ep;
1781 struct rtable *rt;
1782
1783 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1784 if (!ep) {
1785 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1786 err = -ENOMEM;
1787 goto out;
1788 }
1789 init_timer(&ep->timer);
1790 ep->plen = conn_param->private_data_len;
1791 if (ep->plen)
1792 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1793 conn_param->private_data, ep->plen);
1794 ep->ird = conn_param->ird;
1795 ep->ord = conn_param->ord;
1796 ep->com.tdev = h->rdev.t3cdev_p;
1797
1798 cm_id->add_ref(cm_id);
1799 ep->com.cm_id = cm_id;
1800 ep->com.qp = get_qhp(h, conn_param->qpn);
1801 BUG_ON(!ep->com.qp);
1802 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __FUNCTION__, conn_param->qpn,
1803 ep->com.qp, cm_id);
1804
1805 /*
1806 * Allocate an active TID to initiate a TCP connection.
1807 */
1808 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1809 if (ep->atid == -1) {
1810 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1811 err = -ENOMEM;
1812 goto fail2;
1813 }
1814
1815 /* find a route */
1816 rt = find_route(h->rdev.t3cdev_p,
1817 cm_id->local_addr.sin_addr.s_addr,
1818 cm_id->remote_addr.sin_addr.s_addr,
1819 cm_id->local_addr.sin_port,
1820 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1821 if (!rt) {
1822 printk(KERN_ERR MOD "%s - cannot find route.\n", __FUNCTION__);
1823 err = -EHOSTUNREACH;
1824 goto fail3;
1825 }
1826 ep->dst = &rt->u.dst;
1827
1828 /* get a l2t entry */
1829 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1830 ep->dst->neighbour->dev);
1831 if (!ep->l2t) {
1832 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __FUNCTION__);
1833 err = -ENOMEM;
1834 goto fail4;
1835 }
1836
1837 state_set(&ep->com, CONNECTING);
1838 ep->tos = IPTOS_LOWDELAY;
1839 ep->com.local_addr = cm_id->local_addr;
1840 ep->com.remote_addr = cm_id->remote_addr;
1841
1842 /* send connect request to rnic */
1843 err = send_connect(ep);
1844 if (!err)
1845 goto out;
1846
1847 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1848fail4:
1849 dst_release(ep->dst);
1850fail3:
1851 cxgb3_free_atid(ep->com.tdev, ep->atid);
1852fail2:
1853 put_ep(&ep->com);
1854out:
1855 return err;
1856}
1857
1858int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1859{
1860 int err = 0;
1861 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1862 struct iwch_listen_ep *ep;
1863
1864
1865 might_sleep();
1866
1867 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1868 if (!ep) {
1869 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1870 err = -ENOMEM;
1871 goto fail1;
1872 }
1873 PDBG("%s ep %p\n", __FUNCTION__, ep);
1874 ep->com.tdev = h->rdev.t3cdev_p;
1875 cm_id->add_ref(cm_id);
1876 ep->com.cm_id = cm_id;
1877 ep->backlog = backlog;
1878 ep->com.local_addr = cm_id->local_addr;
1879
1880 /*
1881 * Allocate a server TID.
1882 */
1883 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1884 if (ep->stid == -1) {
1885 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1886 err = -ENOMEM;
1887 goto fail2;
1888 }
1889
1890 state_set(&ep->com, LISTEN);
1891 err = listen_start(ep);
1892 if (err)
1893 goto fail3;
1894
1895 /* wait for pass_open_rpl */
1896 wait_event(ep->com.waitq, ep->com.rpl_done);
1897 err = ep->com.rpl_err;
1898 if (!err) {
1899 cm_id->provider_data = ep;
1900 goto out;
1901 }
1902fail3:
1903 cxgb3_free_stid(ep->com.tdev, ep->stid);
1904fail2:
1905 put_ep(&ep->com);
1906fail1:
1907out:
1908 return err;
1909}
1910
1911int iwch_destroy_listen(struct iw_cm_id *cm_id)
1912{
1913 int err;
1914 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
1915
1916 PDBG("%s ep %p\n", __FUNCTION__, ep);
1917
1918 might_sleep();
1919 state_set(&ep->com, DEAD);
1920 ep->com.rpl_done = 0;
1921 ep->com.rpl_err = 0;
1922 err = listen_stop(ep);
1923 wait_event(ep->com.waitq, ep->com.rpl_done);
1924 cxgb3_free_stid(ep->com.tdev, ep->stid);
1925 err = ep->com.rpl_err;
1926 cm_id->rem_ref(cm_id);
1927 put_ep(&ep->com);
1928 return err;
1929}
1930
1931int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
1932{
1933 int ret=0;
1934 unsigned long flags;
1935 int close = 0;
1936
1937 spin_lock_irqsave(&ep->com.lock, flags);
1938
1939 PDBG("%s ep %p state %s, abrupt %d\n", __FUNCTION__, ep,
1940 states[ep->com.state], abrupt);
1941
1942 if (ep->com.state == DEAD) {
1943 PDBG("%s already dead ep %p\n", __FUNCTION__, ep);
1944 goto out;
1945 }
1946
1947 if (abrupt) {
1948 if (ep->com.state != ABORTING) {
1949 ep->com.state = ABORTING;
1950 close = 1;
1951 }
1952 goto out;
1953 }
1954
1955 switch (ep->com.state) {
1956 case MPA_REQ_WAIT:
1957 case MPA_REQ_SENT:
1958 case MPA_REQ_RCVD:
1959 case MPA_REP_SENT:
1960 case FPDU_MODE:
1961 ep->com.state = CLOSING;
1962 close = 1;
1963 break;
1964 case CLOSING:
1965 start_ep_timer(ep);
1966 ep->com.state = MORIBUND;
1967 close = 1;
1968 break;
1969 case MORIBUND:
1970 break;
1971 default:
1972 BUG();
1973 break;
1974 }
1975out:
1976 spin_unlock_irqrestore(&ep->com.lock, flags);
1977 if (close) {
1978 if (abrupt)
1979 ret = send_abort(ep, NULL, gfp);
1980 else
1981 ret = send_halfclose(ep, gfp);
1982 }
1983 return ret;
1984}
1985
1986int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
1987 struct l2t_entry *l2t)
1988{
1989 struct iwch_ep *ep = ctx;
1990
1991 if (ep->dst != old)
1992 return 0;
1993
1994 PDBG("%s ep %p redirect to dst %p l2t %p\n", __FUNCTION__, ep, new,
1995 l2t);
1996 dst_hold(new);
1997 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1998 ep->l2t = l2t;
1999 dst_release(old);
2000 ep->dst = new;
2001 return 1;
2002}
2003
2004/*
2005 * All the CM events are handled on a work queue to have a safe context.
2006 */
2007static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2008{
2009 struct iwch_ep_common *epc = ctx;
2010
2011 get_ep(epc);
2012
2013 /*
2014 * Save ctx and tdev in the skb->cb area.
2015 */
2016 *((void **) skb->cb) = ctx;
2017 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2018
2019 /*
2020 * Queue the skb and schedule the worker thread.
2021 */
2022 skb_queue_tail(&rxq, skb);
2023 queue_work(workq, &skb_work);
2024 return 0;
2025}
2026
2027int __init iwch_cm_init(void)
2028{
2029 skb_queue_head_init(&rxq);
2030
2031 workq = create_singlethread_workqueue("iw_cxgb3");
2032 if (!workq)
2033 return -ENOMEM;
2034
2035 /*
2036 * All upcalls from the T3 Core go to sched() to
2037 * schedule the processing on a work queue.
2038 */
2039 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2040 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2041 t3c_handlers[CPL_RX_DATA] = sched;
2042 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2043 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2044 t3c_handlers[CPL_ABORT_RPL] = sched;
2045 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2046 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2047 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2048 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2049 t3c_handlers[CPL_PEER_CLOSE] = sched;
2050 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2051 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2052 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2053 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
2054
2055 /*
2056 * These are the real handlers that are called from a
2057 * work queue.
2058 */
2059 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2060 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2061 work_handlers[CPL_RX_DATA] = rx_data;
2062 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2063 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2064 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2065 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2066 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2067 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2068 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2069 work_handlers[CPL_PEER_CLOSE] = peer_close;
2070 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2071 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2072 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2073 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2074 return 0;
2075}
2076
2077void __exit iwch_cm_term(void)
2078{
2079 flush_workqueue(workq);
2080 destroy_workqueue(workq);
2081}