RDMA/cxgb3: Handle EEH events
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / infiniband / hw / cxgb3 / iwch_cm.c
CommitLineData
b038ced7
SW
1/*
2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
b038ced7
SW
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32#include <linux/module.h>
33#include <linux/list.h>
34#include <linux/workqueue.h>
35#include <linux/skbuff.h>
36#include <linux/timer.h>
37#include <linux/notifier.h>
8704e9a8 38#include <linux/inetdevice.h>
b038ced7
SW
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
f8b0dfd1
SW
66int peer2peer = 0;
67module_param(peer2peer, int, 0644);
68MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
69
77a8d574 70static int ep_timeout_secs = 60;
e54664c0 71module_param(ep_timeout_secs, int, 0644);
b038ced7 72MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
77a8d574 73 "in seconds (default=60)");
b038ced7
SW
74
75static int mpa_rev = 1;
e54664c0 76module_param(mpa_rev, int, 0644);
b038ced7
SW
77MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
78 "1 is spec compliant. (default=1)");
79
80static int markers_enabled = 0;
e54664c0 81module_param(markers_enabled, int, 0644);
b038ced7
SW
82MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
83
84static int crc_enabled = 1;
e54664c0 85module_param(crc_enabled, int, 0644);
b038ced7
SW
86MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
87
88static int rcv_win = 256 * 1024;
e54664c0 89module_param(rcv_win, int, 0644);
b038ced7
SW
90MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
91
92static int snd_win = 32 * 1024;
e54664c0 93module_param(snd_win, int, 0644);
b038ced7
SW
94MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
95
96static unsigned int nocong = 0;
e54664c0 97module_param(nocong, uint, 0644);
b038ced7
SW
98MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
99
100static unsigned int cong_flavor = 1;
e54664c0 101module_param(cong_flavor, uint, 0644);
b038ced7
SW
102MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
103
104static void process_work(struct work_struct *work);
105static struct workqueue_struct *workq;
106static DECLARE_WORK(skb_work, process_work);
107
108static struct sk_buff_head rxq;
109static cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS];
110
111static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
112static void ep_timeout(unsigned long arg);
113static void connect_reply_upcall(struct iwch_ep *ep, int status);
114
115static void start_ep_timer(struct iwch_ep *ep)
116{
33718363 117 PDBG("%s ep %p\n", __func__, ep);
b038ced7 118 if (timer_pending(&ep->timer)) {
33718363 119 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
b038ced7
SW
120 del_timer_sync(&ep->timer);
121 } else
122 get_ep(&ep->com);
123 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
124 ep->timer.data = (unsigned long)ep;
125 ep->timer.function = ep_timeout;
126 add_timer(&ep->timer);
127}
128
129static void stop_ep_timer(struct iwch_ep *ep)
130{
33718363 131 PDBG("%s ep %p\n", __func__, ep);
989a1780
SW
132 if (!timer_pending(&ep->timer)) {
133 printk(KERN_ERR "%s timer stopped when its not running! ep %p state %u\n",
134 __func__, ep, ep->com.state);
135 WARN_ON(1);
136 return;
137 }
b038ced7
SW
138 del_timer_sync(&ep->timer);
139 put_ep(&ep->com);
140}
141
04b5d028
SW
142int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
143{
144 int error = 0;
145 struct cxio_rdev *rdev;
146
147 rdev = (struct cxio_rdev *)tdev->ulp;
148 if (cxio_fatal_error(rdev)) {
149 kfree_skb(skb);
150 return -EIO;
151 }
152 error = l2t_send(tdev, skb, l2e);
153 if (error)
154 kfree_skb(skb);
155 return error;
156}
157
158int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
159{
160 int error = 0;
161 struct cxio_rdev *rdev;
162
163 rdev = (struct cxio_rdev *)tdev->ulp;
164 if (cxio_fatal_error(rdev)) {
165 kfree_skb(skb);
166 return -EIO;
167 }
168 error = cxgb3_ofld_send(tdev, skb);
169 if (error)
170 kfree_skb(skb);
171 return error;
172}
173
b038ced7
SW
174static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
175{
176 struct cpl_tid_release *req;
177
178 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
179 if (!skb)
180 return;
181 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
182 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
183 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
184 skb->priority = CPL_PRIORITY_SETUP;
04b5d028 185 iwch_cxgb3_ofld_send(tdev, skb);
b038ced7
SW
186 return;
187}
188
189int iwch_quiesce_tid(struct iwch_ep *ep)
190{
191 struct cpl_set_tcb_field *req;
192 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
193
194 if (!skb)
195 return -ENOMEM;
196 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
197 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
198 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
199 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
200 req->reply = 0;
201 req->cpu_idx = 0;
202 req->word = htons(W_TCB_RX_QUIESCE);
203 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
204 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
205
206 skb->priority = CPL_PRIORITY_DATA;
04b5d028 207 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
208}
209
210int iwch_resume_tid(struct iwch_ep *ep)
211{
212 struct cpl_set_tcb_field *req;
213 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
214
215 if (!skb)
216 return -ENOMEM;
217 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
218 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
219 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
220 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
221 req->reply = 0;
222 req->cpu_idx = 0;
223 req->word = htons(W_TCB_RX_QUIESCE);
224 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
225 req->val = 0;
226
227 skb->priority = CPL_PRIORITY_DATA;
04b5d028 228 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
229}
230
231static void set_emss(struct iwch_ep *ep, u16 opt)
232{
33718363 233 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
b038ced7
SW
234 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
235 if (G_TCPOPT_TSTAMP(opt))
236 ep->emss -= 12;
237 if (ep->emss < 128)
238 ep->emss = 128;
239 PDBG("emss=%d\n", ep->emss);
240}
241
242static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
243{
244 unsigned long flags;
245 enum iwch_ep_state state;
246
247 spin_lock_irqsave(&epc->lock, flags);
248 state = epc->state;
249 spin_unlock_irqrestore(&epc->lock, flags);
250 return state;
251}
252
2b540355 253static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
b038ced7
SW
254{
255 epc->state = new;
256}
257
258static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
259{
260 unsigned long flags;
261
262 spin_lock_irqsave(&epc->lock, flags);
33718363 263 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
b038ced7
SW
264 __state_set(epc, new);
265 spin_unlock_irqrestore(&epc->lock, flags);
266 return;
267}
268
269static void *alloc_ep(int size, gfp_t gfp)
270{
271 struct iwch_ep_common *epc;
272
dd00cc48 273 epc = kzalloc(size, gfp);
b038ced7 274 if (epc) {
b038ced7
SW
275 kref_init(&epc->kref);
276 spin_lock_init(&epc->lock);
277 init_waitqueue_head(&epc->waitq);
278 }
33718363 279 PDBG("%s alloc ep %p\n", __func__, epc);
b038ced7
SW
280 return epc;
281}
282
283void __free_ep(struct kref *kref)
284{
285 struct iwch_ep_common *epc;
286 epc = container_of(kref, struct iwch_ep_common, kref);
33718363 287 PDBG("%s ep %p state %s\n", __func__, epc, states[state_read(epc)]);
b038ced7
SW
288 kfree(epc);
289}
290
291static void release_ep_resources(struct iwch_ep *ep)
292{
33718363 293 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
b038ced7
SW
294 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
295 dst_release(ep->dst);
296 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
b038ced7
SW
297 put_ep(&ep->com);
298}
299
300static void process_work(struct work_struct *work)
301{
302 struct sk_buff *skb = NULL;
303 void *ep;
304 struct t3cdev *tdev;
305 int ret;
306
307 while ((skb = skb_dequeue(&rxq))) {
308 ep = *((void **) (skb->cb));
309 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
310 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
311 if (ret & CPL_RET_BUF_DONE)
312 kfree_skb(skb);
313
314 /*
315 * ep was referenced in sched(), and is freed here.
316 */
317 put_ep((struct iwch_ep_common *)ep);
318 }
319}
320
321static int status2errno(int status)
322{
323 switch (status) {
324 case CPL_ERR_NONE:
325 return 0;
326 case CPL_ERR_CONN_RESET:
327 return -ECONNRESET;
328 case CPL_ERR_ARP_MISS:
329 return -EHOSTUNREACH;
330 case CPL_ERR_CONN_TIMEDOUT:
331 return -ETIMEDOUT;
332 case CPL_ERR_TCAM_FULL:
333 return -ENOMEM;
334 case CPL_ERR_CONN_EXIST:
335 return -EADDRINUSE;
336 default:
337 return -EIO;
338 }
339}
340
341/*
342 * Try and reuse skbs already allocated...
343 */
344static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
345{
1f6a849b 346 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
b038ced7
SW
347 skb_trim(skb, 0);
348 skb_get(skb);
349 } else {
350 skb = alloc_skb(len, gfp);
351 }
352 return skb;
353}
354
355static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
356 __be32 peer_ip, __be16 local_port,
357 __be16 peer_port, u8 tos)
358{
359 struct rtable *rt;
360 struct flowi fl = {
361 .oif = 0,
362 .nl_u = {
363 .ip4_u = {
364 .daddr = peer_ip,
365 .saddr = local_ip,
366 .tos = tos}
367 },
368 .proto = IPPROTO_TCP,
369 .uli_u = {
370 .ports = {
371 .sport = local_port,
372 .dport = peer_port}
373 }
374 };
375
f1b050bf 376 if (ip_route_output_flow(&init_net, &rt, &fl, NULL, 0))
b038ced7
SW
377 return NULL;
378 return rt;
379}
380
381static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
382{
383 int i = 0;
384
385 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
386 ++i;
387 return i;
388}
389
390static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
391{
33718363 392 PDBG("%s t3cdev %p\n", __func__, dev);
b038ced7
SW
393 kfree_skb(skb);
394}
395
396/*
397 * Handle an ARP failure for an active open.
398 */
399static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
400{
401 printk(KERN_ERR MOD "ARP failure duing connect\n");
402 kfree_skb(skb);
403}
404
405/*
406 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
407 * and send it along.
408 */
409static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
410{
411 struct cpl_abort_req *req = cplhdr(skb);
412
33718363 413 PDBG("%s t3cdev %p\n", __func__, dev);
b038ced7 414 req->cmd = CPL_ABORT_NO_RST;
04b5d028 415 iwch_cxgb3_ofld_send(dev, skb);
b038ced7
SW
416}
417
418static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
419{
420 struct cpl_close_con_req *req;
421 struct sk_buff *skb;
422
33718363 423 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
424 skb = get_skb(NULL, sizeof(*req), gfp);
425 if (!skb) {
33718363 426 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
b038ced7
SW
427 return -ENOMEM;
428 }
429 skb->priority = CPL_PRIORITY_DATA;
430 set_arp_failure_handler(skb, arp_failure_discard);
431 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
432 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
433 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
434 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
04b5d028 435 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
436}
437
438static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
439{
440 struct cpl_abort_req *req;
441
33718363 442 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
443 skb = get_skb(skb, sizeof(*req), gfp);
444 if (!skb) {
445 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
33718363 446 __func__);
b038ced7
SW
447 return -ENOMEM;
448 }
449 skb->priority = CPL_PRIORITY_DATA;
450 set_arp_failure_handler(skb, abort_arp_failure);
451 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
452 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
453 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
454 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
455 req->cmd = CPL_ABORT_SEND_RST;
04b5d028 456 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
457}
458
459static int send_connect(struct iwch_ep *ep)
460{
461 struct cpl_act_open_req *req;
462 struct sk_buff *skb;
463 u32 opt0h, opt0l, opt2;
464 unsigned int mtu_idx;
465 int wscale;
466
33718363 467 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
468
469 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
470 if (!skb) {
471 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
33718363 472 __func__);
b038ced7
SW
473 return -ENOMEM;
474 }
475 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
476 wscale = compute_wscale(rcv_win);
477 opt0h = V_NAGLE(0) |
478 V_NO_CONG(nocong) |
479 V_KEEP_ALIVE(1) |
480 F_TCAM_BYPASS |
481 V_WND_SCALE(wscale) |
482 V_MSS_IDX(mtu_idx) |
483 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
484 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
485 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
486 skb->priority = CPL_PRIORITY_SETUP;
487 set_arp_failure_handler(skb, act_open_req_arp_failure);
488
489 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
490 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
491 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
492 req->local_port = ep->com.local_addr.sin_port;
493 req->peer_port = ep->com.remote_addr.sin_port;
494 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
495 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
496 req->opt0h = htonl(opt0h);
497 req->opt0l = htonl(opt0l);
498 req->params = 0;
499 req->opt2 = htonl(opt2);
04b5d028 500 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
501}
502
503static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
504{
505 int mpalen;
506 struct tx_data_wr *req;
507 struct mpa_message *mpa;
508 int len;
509
33718363 510 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
b038ced7
SW
511
512 BUG_ON(skb_cloned(skb));
513
514 mpalen = sizeof(*mpa) + ep->plen;
4305b541 515 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
b038ced7
SW
516 kfree_skb(skb);
517 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
518 if (!skb) {
519 connect_reply_upcall(ep, -ENOMEM);
520 return;
521 }
522 }
523 skb_trim(skb, 0);
524 skb_reserve(skb, sizeof(*req));
525 skb_put(skb, mpalen);
526 skb->priority = CPL_PRIORITY_DATA;
527 mpa = (struct mpa_message *) skb->data;
528 memset(mpa, 0, sizeof(*mpa));
529 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
530 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
531 (markers_enabled ? MPA_MARKERS : 0);
532 mpa->private_data_size = htons(ep->plen);
533 mpa->revision = mpa_rev;
534
535 if (ep->plen)
536 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
537
538 /*
539 * Reference the mpa skb. This ensures the data area
540 * will remain in memory until the hw acks the tx.
541 * Function tx_ack() will deref it.
542 */
543 skb_get(skb);
544 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 545 skb_reset_transport_header(skb);
b038ced7
SW
546 len = skb->len;
547 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 548 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
549 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
550 req->len = htonl(len);
551 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
552 V_TX_SNDBUF(snd_win>>15));
de3d3530 553 req->flags = htonl(F_TX_INIT);
b038ced7
SW
554 req->sndseq = htonl(ep->snd_seq);
555 BUG_ON(ep->mpa_skb);
556 ep->mpa_skb = skb;
04b5d028 557 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
558 start_ep_timer(ep);
559 state_set(&ep->com, MPA_REQ_SENT);
560 return;
561}
562
563static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
564{
565 int mpalen;
566 struct tx_data_wr *req;
567 struct mpa_message *mpa;
568 struct sk_buff *skb;
569
33718363 570 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
b038ced7
SW
571
572 mpalen = sizeof(*mpa) + plen;
573
574 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
575 if (!skb) {
33718363 576 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
b038ced7
SW
577 return -ENOMEM;
578 }
579 skb_reserve(skb, sizeof(*req));
580 mpa = (struct mpa_message *) skb_put(skb, mpalen);
581 memset(mpa, 0, sizeof(*mpa));
582 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
583 mpa->flags = MPA_REJECT;
584 mpa->revision = mpa_rev;
585 mpa->private_data_size = htons(plen);
586 if (plen)
587 memcpy(mpa->private_data, pdata, plen);
588
589 /*
590 * Reference the mpa skb again. This ensures the data area
591 * will remain in memory until the hw acks the tx.
592 * Function tx_ack() will deref it.
593 */
594 skb_get(skb);
595 skb->priority = CPL_PRIORITY_DATA;
596 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 597 skb_reset_transport_header(skb);
b038ced7 598 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 599 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
600 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
601 req->len = htonl(mpalen);
602 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
603 V_TX_SNDBUF(snd_win>>15));
de3d3530 604 req->flags = htonl(F_TX_INIT);
b038ced7
SW
605 req->sndseq = htonl(ep->snd_seq);
606 BUG_ON(ep->mpa_skb);
607 ep->mpa_skb = skb;
04b5d028 608 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
609}
610
611static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
612{
613 int mpalen;
614 struct tx_data_wr *req;
615 struct mpa_message *mpa;
616 int len;
617 struct sk_buff *skb;
618
33718363 619 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
b038ced7
SW
620
621 mpalen = sizeof(*mpa) + plen;
622
623 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
624 if (!skb) {
33718363 625 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
b038ced7
SW
626 return -ENOMEM;
627 }
628 skb->priority = CPL_PRIORITY_DATA;
629 skb_reserve(skb, sizeof(*req));
630 mpa = (struct mpa_message *) skb_put(skb, mpalen);
631 memset(mpa, 0, sizeof(*mpa));
632 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
633 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
634 (markers_enabled ? MPA_MARKERS : 0);
635 mpa->revision = mpa_rev;
636 mpa->private_data_size = htons(plen);
637 if (plen)
638 memcpy(mpa->private_data, pdata, plen);
639
640 /*
641 * Reference the mpa skb. This ensures the data area
642 * will remain in memory until the hw acks the tx.
643 * Function tx_ack() will deref it.
644 */
645 skb_get(skb);
646 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 647 skb_reset_transport_header(skb);
b038ced7
SW
648 len = skb->len;
649 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 650 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
651 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
652 req->len = htonl(len);
653 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
654 V_TX_SNDBUF(snd_win>>15));
de3d3530 655 req->flags = htonl(F_TX_INIT);
b038ced7
SW
656 req->sndseq = htonl(ep->snd_seq);
657 ep->mpa_skb = skb;
658 state_set(&ep->com, MPA_REP_SENT);
04b5d028 659 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
660}
661
662static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
663{
664 struct iwch_ep *ep = ctx;
665 struct cpl_act_establish *req = cplhdr(skb);
666 unsigned int tid = GET_TID(req);
667
33718363 668 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
b038ced7
SW
669
670 dst_confirm(ep->dst);
671
672 /* setup the hwtid for this connection */
673 ep->hwtid = tid;
674 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
675
676 ep->snd_seq = ntohl(req->snd_isn);
de3d3530 677 ep->rcv_seq = ntohl(req->rcv_isn);
b038ced7
SW
678
679 set_emss(ep, ntohs(req->tcp_opt));
680
681 /* dealloc the atid */
682 cxgb3_free_atid(ep->com.tdev, ep->atid);
683
684 /* start MPA negotiation */
685 send_mpa_req(ep, skb);
686
687 return 0;
688}
689
690static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
691{
692 PDBG("%s ep %p\n", __FILE__, ep);
693 state_set(&ep->com, ABORTING);
694 send_abort(ep, skb, gfp);
695}
696
697static void close_complete_upcall(struct iwch_ep *ep)
698{
699 struct iw_cm_event event;
700
33718363 701 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
702 memset(&event, 0, sizeof(event));
703 event.event = IW_CM_EVENT_CLOSE;
704 if (ep->com.cm_id) {
705 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
706 ep, ep->com.cm_id, ep->hwtid);
707 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
708 ep->com.cm_id->rem_ref(ep->com.cm_id);
709 ep->com.cm_id = NULL;
710 ep->com.qp = NULL;
711 }
712}
713
714static void peer_close_upcall(struct iwch_ep *ep)
715{
716 struct iw_cm_event event;
717
33718363 718 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
719 memset(&event, 0, sizeof(event));
720 event.event = IW_CM_EVENT_DISCONNECT;
721 if (ep->com.cm_id) {
722 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
723 ep, ep->com.cm_id, ep->hwtid);
724 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
725 }
726}
727
728static void peer_abort_upcall(struct iwch_ep *ep)
729{
730 struct iw_cm_event event;
731
33718363 732 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
733 memset(&event, 0, sizeof(event));
734 event.event = IW_CM_EVENT_CLOSE;
735 event.status = -ECONNRESET;
736 if (ep->com.cm_id) {
737 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
738 ep->com.cm_id, ep->hwtid);
739 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
740 ep->com.cm_id->rem_ref(ep->com.cm_id);
741 ep->com.cm_id = NULL;
742 ep->com.qp = NULL;
743 }
744}
745
746static void connect_reply_upcall(struct iwch_ep *ep, int status)
747{
748 struct iw_cm_event event;
749
33718363 750 PDBG("%s ep %p status %d\n", __func__, ep, status);
b038ced7
SW
751 memset(&event, 0, sizeof(event));
752 event.event = IW_CM_EVENT_CONNECT_REPLY;
753 event.status = status;
754 event.local_addr = ep->com.local_addr;
755 event.remote_addr = ep->com.remote_addr;
756
757 if ((status == 0) || (status == -ECONNREFUSED)) {
758 event.private_data_len = ep->plen;
759 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
760 }
761 if (ep->com.cm_id) {
33718363 762 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
b038ced7
SW
763 ep->hwtid, status);
764 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
765 }
766 if (status < 0) {
767 ep->com.cm_id->rem_ref(ep->com.cm_id);
768 ep->com.cm_id = NULL;
769 ep->com.qp = NULL;
770 }
771}
772
773static void connect_request_upcall(struct iwch_ep *ep)
774{
775 struct iw_cm_event event;
776
33718363 777 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
b038ced7
SW
778 memset(&event, 0, sizeof(event));
779 event.event = IW_CM_EVENT_CONNECT_REQUEST;
780 event.local_addr = ep->com.local_addr;
781 event.remote_addr = ep->com.remote_addr;
782 event.private_data_len = ep->plen;
783 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
784 event.provider_data = ep;
785 if (state_read(&ep->parent_ep->com) != DEAD)
786 ep->parent_ep->com.cm_id->event_handler(
787 ep->parent_ep->com.cm_id,
788 &event);
789 put_ep(&ep->parent_ep->com);
790 ep->parent_ep = NULL;
791}
792
793static void established_upcall(struct iwch_ep *ep)
794{
795 struct iw_cm_event event;
796
33718363 797 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
798 memset(&event, 0, sizeof(event));
799 event.event = IW_CM_EVENT_ESTABLISHED;
800 if (ep->com.cm_id) {
33718363 801 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
b038ced7
SW
802 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
803 }
804}
805
806static int update_rx_credits(struct iwch_ep *ep, u32 credits)
807{
808 struct cpl_rx_data_ack *req;
809 struct sk_buff *skb;
810
33718363 811 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
b038ced7
SW
812 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
813 if (!skb) {
814 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
815 return 0;
816 }
817
818 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
819 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
820 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
821 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
822 skb->priority = CPL_PRIORITY_ACK;
04b5d028 823 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
824 return credits;
825}
826
827static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
828{
829 struct mpa_message *mpa;
830 u16 plen;
831 struct iwch_qp_attributes attrs;
832 enum iwch_qp_attr_mask mask;
833 int err;
834
33718363 835 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
836
837 /*
838 * Stop mpa timer. If it expired, then the state has
839 * changed and we bail since ep_timeout already aborted
840 * the connection.
841 */
842 stop_ep_timer(ep);
843 if (state_read(&ep->com) != MPA_REQ_SENT)
844 return;
845
846 /*
847 * If we get more than the supported amount of private data
848 * then we must fail this connection.
849 */
850 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
851 err = -EINVAL;
852 goto err;
853 }
854
855 /*
856 * copy the new data into our accumulation buffer.
857 */
d626f62b
ACM
858 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
859 skb->len);
b038ced7
SW
860 ep->mpa_pkt_len += skb->len;
861
862 /*
863 * if we don't even have the mpa message, then bail.
864 */
865 if (ep->mpa_pkt_len < sizeof(*mpa))
866 return;
867 mpa = (struct mpa_message *) ep->mpa_pkt;
868
869 /* Validate MPA header. */
870 if (mpa->revision != mpa_rev) {
871 err = -EPROTO;
872 goto err;
873 }
874 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
875 err = -EPROTO;
876 goto err;
877 }
878
879 plen = ntohs(mpa->private_data_size);
880
881 /*
882 * Fail if there's too much private data.
883 */
884 if (plen > MPA_MAX_PRIVATE_DATA) {
885 err = -EPROTO;
886 goto err;
887 }
888
889 /*
890 * If plen does not account for pkt size
891 */
892 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
893 err = -EPROTO;
894 goto err;
895 }
896
897 ep->plen = (u8) plen;
898
899 /*
900 * If we don't have all the pdata yet, then bail.
901 * We'll continue process when more data arrives.
902 */
903 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
904 return;
905
906 if (mpa->flags & MPA_REJECT) {
907 err = -ECONNREFUSED;
908 goto err;
909 }
910
911 /*
912 * If we get here we have accumulated the entire mpa
913 * start reply message including private data. And
914 * the MPA header is valid.
915 */
916 state_set(&ep->com, FPDU_MODE);
f8b0dfd1 917 ep->mpa_attr.initiator = 1;
b038ced7
SW
918 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
919 ep->mpa_attr.recv_marker_enabled = markers_enabled;
920 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
921 ep->mpa_attr.version = mpa_rev;
922 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
33718363 923 "xmit_marker_enabled=%d, version=%d\n", __func__,
b038ced7
SW
924 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
925 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
926
927 attrs.mpa_attr = ep->mpa_attr;
928 attrs.max_ird = ep->ird;
929 attrs.max_ord = ep->ord;
930 attrs.llp_stream_handle = ep;
931 attrs.next_state = IWCH_QP_STATE_RTS;
932
933 mask = IWCH_QP_ATTR_NEXT_STATE |
934 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
935 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
936
937 /* bind QP and TID with INIT_WR */
938 err = iwch_modify_qp(ep->com.qp->rhp,
939 ep->com.qp, mask, &attrs, 1);
f8b0dfd1
SW
940 if (err)
941 goto err;
942
943 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
944 iwch_post_zb_read(ep->com.qp);
945 }
946
947 goto out;
b038ced7
SW
948err:
949 abort_connection(ep, skb, GFP_KERNEL);
950out:
951 connect_reply_upcall(ep, err);
952 return;
953}
954
955static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
956{
957 struct mpa_message *mpa;
958 u16 plen;
959
33718363 960 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
961
962 /*
963 * Stop mpa timer. If it expired, then the state has
964 * changed and we bail since ep_timeout already aborted
965 * the connection.
966 */
967 stop_ep_timer(ep);
968 if (state_read(&ep->com) != MPA_REQ_WAIT)
969 return;
970
971 /*
972 * If we get more than the supported amount of private data
973 * then we must fail this connection.
974 */
975 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
976 abort_connection(ep, skb, GFP_KERNEL);
977 return;
978 }
979
33718363 980 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
b038ced7
SW
981
982 /*
983 * Copy the new data into our accumulation buffer.
984 */
d626f62b
ACM
985 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
986 skb->len);
b038ced7
SW
987 ep->mpa_pkt_len += skb->len;
988
989 /*
990 * If we don't even have the mpa message, then bail.
991 * We'll continue process when more data arrives.
992 */
993 if (ep->mpa_pkt_len < sizeof(*mpa))
994 return;
33718363 995 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
b038ced7
SW
996 mpa = (struct mpa_message *) ep->mpa_pkt;
997
998 /*
999 * Validate MPA Header.
1000 */
1001 if (mpa->revision != mpa_rev) {
1002 abort_connection(ep, skb, GFP_KERNEL);
1003 return;
1004 }
1005
1006 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1007 abort_connection(ep, skb, GFP_KERNEL);
1008 return;
1009 }
1010
1011 plen = ntohs(mpa->private_data_size);
1012
1013 /*
1014 * Fail if there's too much private data.
1015 */
1016 if (plen > MPA_MAX_PRIVATE_DATA) {
1017 abort_connection(ep, skb, GFP_KERNEL);
1018 return;
1019 }
1020
1021 /*
1022 * If plen does not account for pkt size
1023 */
1024 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1025 abort_connection(ep, skb, GFP_KERNEL);
1026 return;
1027 }
1028 ep->plen = (u8) plen;
1029
1030 /*
1031 * If we don't have all the pdata yet, then bail.
1032 */
1033 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1034 return;
1035
1036 /*
1037 * If we get here we have accumulated the entire mpa
1038 * start reply message including private data.
1039 */
f8b0dfd1 1040 ep->mpa_attr.initiator = 0;
b038ced7
SW
1041 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1042 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1043 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1044 ep->mpa_attr.version = mpa_rev;
1045 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
33718363 1046 "xmit_marker_enabled=%d, version=%d\n", __func__,
b038ced7
SW
1047 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1048 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1049
1050 state_set(&ep->com, MPA_REQ_RCVD);
1051
1052 /* drive upcall */
1053 connect_request_upcall(ep);
1054 return;
1055}
1056
1057static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1058{
1059 struct iwch_ep *ep = ctx;
1060 struct cpl_rx_data *hdr = cplhdr(skb);
1061 unsigned int dlen = ntohs(hdr->len);
1062
33718363 1063 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
b038ced7
SW
1064
1065 skb_pull(skb, sizeof(*hdr));
1066 skb_trim(skb, dlen);
1067
de3d3530
SW
1068 ep->rcv_seq += dlen;
1069 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1070
b038ced7
SW
1071 switch (state_read(&ep->com)) {
1072 case MPA_REQ_SENT:
1073 process_mpa_reply(ep, skb);
1074 break;
1075 case MPA_REQ_WAIT:
1076 process_mpa_request(ep, skb);
1077 break;
1078 case MPA_REP_SENT:
1079 break;
1080 default:
1081 printk(KERN_ERR MOD "%s Unexpected streaming data."
1082 " ep %p state %d tid %d\n",
33718363 1083 __func__, ep, state_read(&ep->com), ep->hwtid);
b038ced7
SW
1084
1085 /*
1086 * The ep will timeout and inform the ULP of the failure.
1087 * See ep_timeout().
1088 */
1089 break;
1090 }
1091
1092 /* update RX credits */
1093 update_rx_credits(ep, dlen);
1094
1095 return CPL_RET_BUF_DONE;
1096}
1097
1098/*
1099 * Upcall from the adapter indicating data has been transmitted.
1100 * For us its just the single MPA request or reply. We can now free
1101 * the skb holding the mpa message.
1102 */
1103static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1104{
1105 struct iwch_ep *ep = ctx;
1106 struct cpl_wr_ack *hdr = cplhdr(skb);
1107 unsigned int credits = ntohs(hdr->credits);
b038ced7 1108
33718363 1109 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
b038ced7 1110
f8b0dfd1
SW
1111 if (credits == 0) {
1112 PDBG(KERN_ERR "%s 0 credit ack ep %p state %u\n",
1113 __func__, ep, state_read(&ep->com));
b038ced7 1114 return CPL_RET_BUF_DONE;
f8b0dfd1
SW
1115 }
1116
b038ced7 1117 BUG_ON(credits != 1);
b038ced7 1118 dst_confirm(ep->dst);
f8b0dfd1
SW
1119 if (!ep->mpa_skb) {
1120 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1121 __func__, ep, state_read(&ep->com));
1122 if (ep->mpa_attr.initiator) {
1123 PDBG("%s initiator ep %p state %u\n",
1124 __func__, ep, state_read(&ep->com));
1125 if (peer2peer)
1126 iwch_post_zb_read(ep->com.qp);
1127 } else {
1128 PDBG("%s responder ep %p state %u\n",
1129 __func__, ep, state_read(&ep->com));
1130 ep->com.rpl_done = 1;
1131 wake_up(&ep->com.waitq);
1132 }
1133 } else {
1134 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1135 __func__, ep, state_read(&ep->com));
1136 kfree_skb(ep->mpa_skb);
1137 ep->mpa_skb = NULL;
b038ced7
SW
1138 }
1139 return CPL_RET_BUF_DONE;
1140}
1141
1142static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1143{
1144 struct iwch_ep *ep = ctx;
989a1780
SW
1145 unsigned long flags;
1146 int release = 0;
b038ced7 1147
33718363 1148 PDBG("%s ep %p\n", __func__, ep);
989a1780 1149 BUG_ON(!ep);
b038ced7 1150
aff9e39d
SW
1151 /*
1152 * We get 2 abort replies from the HW. The first one must
1153 * be ignored except for scribbling that we need one more.
1154 */
1155 if (!(ep->flags & ABORT_REQ_IN_PROGRESS)) {
1156 ep->flags |= ABORT_REQ_IN_PROGRESS;
1157 return CPL_RET_BUF_DONE;
1158 }
1159
989a1780
SW
1160 spin_lock_irqsave(&ep->com.lock, flags);
1161 switch (ep->com.state) {
1162 case ABORTING:
1163 close_complete_upcall(ep);
1164 __state_set(&ep->com, DEAD);
1165 release = 1;
1166 break;
1167 default:
1168 printk(KERN_ERR "%s ep %p state %d\n",
1169 __func__, ep, ep->com.state);
1170 break;
1171 }
1172 spin_unlock_irqrestore(&ep->com.lock, flags);
1173
1174 if (release)
1175 release_ep_resources(ep);
b038ced7
SW
1176 return CPL_RET_BUF_DONE;
1177}
1178
96d0e493
SW
1179/*
1180 * Return whether a failed active open has allocated a TID
1181 */
1182static inline int act_open_has_tid(int status)
1183{
1184 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1185 status != CPL_ERR_ARP_MISS;
1186}
1187
b038ced7
SW
1188static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1189{
1190 struct iwch_ep *ep = ctx;
1191 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1192
33718363 1193 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
b038ced7
SW
1194 status2errno(rpl->status));
1195 connect_reply_upcall(ep, status2errno(rpl->status));
1196 state_set(&ep->com, DEAD);
8176d297 1197 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
b038ced7
SW
1198 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1199 cxgb3_free_atid(ep->com.tdev, ep->atid);
1200 dst_release(ep->dst);
1201 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1202 put_ep(&ep->com);
1203 return CPL_RET_BUF_DONE;
1204}
1205
1206static int listen_start(struct iwch_listen_ep *ep)
1207{
1208 struct sk_buff *skb;
1209 struct cpl_pass_open_req *req;
1210
33718363 1211 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1212 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1213 if (!skb) {
1214 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1215 return -ENOMEM;
1216 }
1217
1218 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1219 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1220 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1221 req->local_port = ep->com.local_addr.sin_port;
1222 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1223 req->peer_port = 0;
1224 req->peer_ip = 0;
1225 req->peer_netmask = 0;
1226 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1227 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1228 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1229
1230 skb->priority = 1;
04b5d028 1231 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
1232}
1233
1234static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1235{
1236 struct iwch_listen_ep *ep = ctx;
1237 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1238
33718363 1239 PDBG("%s ep %p status %d error %d\n", __func__, ep,
b038ced7
SW
1240 rpl->status, status2errno(rpl->status));
1241 ep->com.rpl_err = status2errno(rpl->status);
1242 ep->com.rpl_done = 1;
1243 wake_up(&ep->com.waitq);
1244
1245 return CPL_RET_BUF_DONE;
1246}
1247
1248static int listen_stop(struct iwch_listen_ep *ep)
1249{
1250 struct sk_buff *skb;
1251 struct cpl_close_listserv_req *req;
1252
33718363 1253 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1254 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1255 if (!skb) {
33718363 1256 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
b038ced7
SW
1257 return -ENOMEM;
1258 }
1259 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1260 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
60be4b59 1261 req->cpu_idx = 0;
b038ced7
SW
1262 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1263 skb->priority = 1;
04b5d028 1264 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
1265}
1266
1267static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1268 void *ctx)
1269{
1270 struct iwch_listen_ep *ep = ctx;
1271 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1272
33718363 1273 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1274 ep->com.rpl_err = status2errno(rpl->status);
1275 ep->com.rpl_done = 1;
1276 wake_up(&ep->com.waitq);
1277 return CPL_RET_BUF_DONE;
1278}
1279
1280static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1281{
1282 struct cpl_pass_accept_rpl *rpl;
1283 unsigned int mtu_idx;
1284 u32 opt0h, opt0l, opt2;
1285 int wscale;
1286
33718363 1287 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1288 BUG_ON(skb_cloned(skb));
1289 skb_trim(skb, sizeof(*rpl));
1290 skb_get(skb);
1291 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1292 wscale = compute_wscale(rcv_win);
1293 opt0h = V_NAGLE(0) |
1294 V_NO_CONG(nocong) |
1295 V_KEEP_ALIVE(1) |
1296 F_TCAM_BYPASS |
1297 V_WND_SCALE(wscale) |
1298 V_MSS_IDX(mtu_idx) |
1299 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1300 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1301 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1302
1303 rpl = cplhdr(skb);
1304 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1305 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1306 rpl->peer_ip = peer_ip;
1307 rpl->opt0h = htonl(opt0h);
1308 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1309 rpl->opt2 = htonl(opt2);
1310 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1311 skb->priority = CPL_PRIORITY_SETUP;
04b5d028 1312 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
1313
1314 return;
1315}
1316
1317static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1318 struct sk_buff *skb)
1319{
33718363 1320 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
b038ced7
SW
1321 peer_ip);
1322 BUG_ON(skb_cloned(skb));
1323 skb_trim(skb, sizeof(struct cpl_tid_release));
1324 skb_get(skb);
1325
8176d297 1326 if (tdev->type != T3A)
b038ced7
SW
1327 release_tid(tdev, hwtid, skb);
1328 else {
1329 struct cpl_pass_accept_rpl *rpl;
1330
1331 rpl = cplhdr(skb);
1332 skb->priority = CPL_PRIORITY_SETUP;
1333 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1334 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1335 hwtid));
1336 rpl->peer_ip = peer_ip;
1337 rpl->opt0h = htonl(F_TCAM_BYPASS);
1338 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1339 rpl->opt2 = 0;
1340 rpl->rsvd = rpl->opt2;
04b5d028 1341 iwch_cxgb3_ofld_send(tdev, skb);
b038ced7
SW
1342 }
1343}
1344
1345static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1346{
1347 struct iwch_ep *child_ep, *parent_ep = ctx;
1348 struct cpl_pass_accept_req *req = cplhdr(skb);
1349 unsigned int hwtid = GET_TID(req);
1350 struct dst_entry *dst;
1351 struct l2t_entry *l2t;
1352 struct rtable *rt;
1353 struct iff_mac tim;
1354
33718363 1355 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
b038ced7
SW
1356
1357 if (state_read(&parent_ep->com) != LISTEN) {
1358 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
33718363 1359 __func__);
b038ced7
SW
1360 goto reject;
1361 }
1362
1363 /*
1364 * Find the netdev for this connection request.
1365 */
1366 tim.mac_addr = req->dst_mac;
1367 tim.vlan_tag = ntohs(req->vlan_tag);
1368 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1369 printk(KERN_ERR
1370 "%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
33718363 1371 __func__,
b038ced7
SW
1372 req->dst_mac[0],
1373 req->dst_mac[1],
1374 req->dst_mac[2],
1375 req->dst_mac[3],
1376 req->dst_mac[4],
1377 req->dst_mac[5]);
1378 goto reject;
1379 }
1380
1381 /* Find output route */
1382 rt = find_route(tdev,
1383 req->local_ip,
1384 req->peer_ip,
1385 req->local_port,
1386 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1387 if (!rt) {
1388 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
33718363 1389 __func__);
b038ced7
SW
1390 goto reject;
1391 }
1392 dst = &rt->u.dst;
1393 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1394 if (!l2t) {
1395 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
33718363 1396 __func__);
b038ced7
SW
1397 dst_release(dst);
1398 goto reject;
1399 }
1400 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1401 if (!child_ep) {
1402 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
33718363 1403 __func__);
b038ced7
SW
1404 l2t_release(L2DATA(tdev), l2t);
1405 dst_release(dst);
1406 goto reject;
1407 }
1408 state_set(&child_ep->com, CONNECTING);
1409 child_ep->com.tdev = tdev;
1410 child_ep->com.cm_id = NULL;
1411 child_ep->com.local_addr.sin_family = PF_INET;
1412 child_ep->com.local_addr.sin_port = req->local_port;
1413 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1414 child_ep->com.remote_addr.sin_family = PF_INET;
1415 child_ep->com.remote_addr.sin_port = req->peer_port;
1416 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1417 get_ep(&parent_ep->com);
1418 child_ep->parent_ep = parent_ep;
1419 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1420 child_ep->l2t = l2t;
1421 child_ep->dst = dst;
1422 child_ep->hwtid = hwtid;
1423 init_timer(&child_ep->timer);
1424 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1425 accept_cr(child_ep, req->peer_ip, skb);
1426 goto out;
1427reject:
1428 reject_cr(tdev, hwtid, req->peer_ip, skb);
1429out:
1430 return CPL_RET_BUF_DONE;
1431}
1432
1433static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1434{
1435 struct iwch_ep *ep = ctx;
1436 struct cpl_pass_establish *req = cplhdr(skb);
1437
33718363 1438 PDBG("%s ep %p\n", __func__, ep);
b038ced7 1439 ep->snd_seq = ntohl(req->snd_isn);
de3d3530 1440 ep->rcv_seq = ntohl(req->rcv_isn);
b038ced7
SW
1441
1442 set_emss(ep, ntohs(req->tcp_opt));
1443
1444 dst_confirm(ep->dst);
1445 state_set(&ep->com, MPA_REQ_WAIT);
1446 start_ep_timer(ep);
1447
1448 return CPL_RET_BUF_DONE;
1449}
1450
1451static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1452{
1453 struct iwch_ep *ep = ctx;
1454 struct iwch_qp_attributes attrs;
1455 unsigned long flags;
1456 int disconnect = 1;
1457 int release = 0;
1458
33718363 1459 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1460 dst_confirm(ep->dst);
1461
1462 spin_lock_irqsave(&ep->com.lock, flags);
1463 switch (ep->com.state) {
1464 case MPA_REQ_WAIT:
1465 __state_set(&ep->com, CLOSING);
1466 break;
1467 case MPA_REQ_SENT:
1468 __state_set(&ep->com, CLOSING);
1469 connect_reply_upcall(ep, -ECONNRESET);
1470 break;
1471 case MPA_REQ_RCVD:
1472
1473 /*
1474 * We're gonna mark this puppy DEAD, but keep
1475 * the reference on it until the ULP accepts or
1476 * rejects the CR.
1477 */
1478 __state_set(&ep->com, CLOSING);
1479 get_ep(&ep->com);
1480 break;
1481 case MPA_REP_SENT:
1482 __state_set(&ep->com, CLOSING);
1483 ep->com.rpl_done = 1;
1484 ep->com.rpl_err = -ECONNRESET;
1485 PDBG("waking up ep %p\n", ep);
1486 wake_up(&ep->com.waitq);
1487 break;
1488 case FPDU_MODE:
42e31753 1489 start_ep_timer(ep);
b038ced7
SW
1490 __state_set(&ep->com, CLOSING);
1491 attrs.next_state = IWCH_QP_STATE_CLOSING;
1492 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1493 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1494 peer_close_upcall(ep);
1495 break;
1496 case ABORTING:
1497 disconnect = 0;
1498 break;
1499 case CLOSING:
b038ced7
SW
1500 __state_set(&ep->com, MORIBUND);
1501 disconnect = 0;
1502 break;
1503 case MORIBUND:
1504 stop_ep_timer(ep);
1505 if (ep->com.cm_id && ep->com.qp) {
1506 attrs.next_state = IWCH_QP_STATE_IDLE;
1507 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1508 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1509 }
1510 close_complete_upcall(ep);
1511 __state_set(&ep->com, DEAD);
1512 release = 1;
1513 disconnect = 0;
1514 break;
1515 case DEAD:
1516 disconnect = 0;
1517 break;
1518 default:
1519 BUG_ON(1);
1520 }
1521 spin_unlock_irqrestore(&ep->com.lock, flags);
1522 if (disconnect)
1523 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1524 if (release)
1525 release_ep_resources(ep);
1526 return CPL_RET_BUF_DONE;
1527}
1528
1529/*
1530 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1531 */
2b540355 1532static int is_neg_adv_abort(unsigned int status)
b038ced7
SW
1533{
1534 return status == CPL_ERR_RTX_NEG_ADVICE ||
1535 status == CPL_ERR_PERSIST_NEG_ADVICE;
1536}
1537
1538static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1539{
1540 struct cpl_abort_req_rss *req = cplhdr(skb);
1541 struct iwch_ep *ep = ctx;
1542 struct cpl_abort_rpl *rpl;
1543 struct sk_buff *rpl_skb;
1544 struct iwch_qp_attributes attrs;
1545 int ret;
989a1780
SW
1546 int release = 0;
1547 unsigned long flags;
b038ced7 1548
1580367e 1549 if (is_neg_adv_abort(req->status)) {
33718363 1550 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1580367e
SW
1551 ep->hwtid);
1552 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1553 return CPL_RET_BUF_DONE;
1554 }
1555
aff9e39d
SW
1556 /*
1557 * We get 2 peer aborts from the HW. The first one must
1558 * be ignored except for scribbling that we need one more.
1559 */
1560 if (!(ep->flags & PEER_ABORT_IN_PROGRESS)) {
1561 ep->flags |= PEER_ABORT_IN_PROGRESS;
1562 return CPL_RET_BUF_DONE;
1563 }
1564
989a1780
SW
1565 spin_lock_irqsave(&ep->com.lock, flags);
1566 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1567 switch (ep->com.state) {
b038ced7
SW
1568 case CONNECTING:
1569 break;
1570 case MPA_REQ_WAIT:
adf376b3 1571 stop_ep_timer(ep);
b038ced7
SW
1572 break;
1573 case MPA_REQ_SENT:
adf376b3 1574 stop_ep_timer(ep);
b038ced7
SW
1575 connect_reply_upcall(ep, -ECONNRESET);
1576 break;
1577 case MPA_REP_SENT:
1578 ep->com.rpl_done = 1;
1579 ep->com.rpl_err = -ECONNRESET;
1580 PDBG("waking up ep %p\n", ep);
1581 wake_up(&ep->com.waitq);
1582 break;
1583 case MPA_REQ_RCVD:
1584
1585 /*
1586 * We're gonna mark this puppy DEAD, but keep
1587 * the reference on it until the ULP accepts or
1588 * rejects the CR.
1589 */
1590 get_ep(&ep->com);
1591 break;
1592 case MORIBUND:
42e31753 1593 case CLOSING:
b038ced7 1594 stop_ep_timer(ep);
42e31753 1595 /*FALLTHROUGH*/
b038ced7 1596 case FPDU_MODE:
b038ced7
SW
1597 if (ep->com.cm_id && ep->com.qp) {
1598 attrs.next_state = IWCH_QP_STATE_ERROR;
1599 ret = iwch_modify_qp(ep->com.qp->rhp,
1600 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1601 &attrs, 1);
1602 if (ret)
1603 printk(KERN_ERR MOD
1604 "%s - qp <- error failed!\n",
33718363 1605 __func__);
b038ced7
SW
1606 }
1607 peer_abort_upcall(ep);
1608 break;
1609 case ABORTING:
1610 break;
1611 case DEAD:
33718363 1612 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
989a1780 1613 spin_unlock_irqrestore(&ep->com.lock, flags);
b038ced7
SW
1614 return CPL_RET_BUF_DONE;
1615 default:
1616 BUG_ON(1);
1617 break;
1618 }
1619 dst_confirm(ep->dst);
989a1780
SW
1620 if (ep->com.state != ABORTING) {
1621 __state_set(&ep->com, DEAD);
1622 release = 1;
1623 }
1624 spin_unlock_irqrestore(&ep->com.lock, flags);
b038ced7
SW
1625
1626 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1627 if (!rpl_skb) {
1628 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
33718363 1629 __func__);
989a1780
SW
1630 release = 1;
1631 goto out;
b038ced7
SW
1632 }
1633 rpl_skb->priority = CPL_PRIORITY_DATA;
1634 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1635 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1636 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1637 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1638 rpl->cmd = CPL_ABORT_NO_RST;
04b5d028 1639 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
989a1780
SW
1640out:
1641 if (release)
b038ced7 1642 release_ep_resources(ep);
b038ced7
SW
1643 return CPL_RET_BUF_DONE;
1644}
1645
1646static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1647{
1648 struct iwch_ep *ep = ctx;
1649 struct iwch_qp_attributes attrs;
1650 unsigned long flags;
1651 int release = 0;
1652
33718363 1653 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1654 BUG_ON(!ep);
1655
1656 /* The cm_id may be null if we failed to connect */
1657 spin_lock_irqsave(&ep->com.lock, flags);
1658 switch (ep->com.state) {
1659 case CLOSING:
b038ced7
SW
1660 __state_set(&ep->com, MORIBUND);
1661 break;
1662 case MORIBUND:
1663 stop_ep_timer(ep);
1664 if ((ep->com.cm_id) && (ep->com.qp)) {
1665 attrs.next_state = IWCH_QP_STATE_IDLE;
1666 iwch_modify_qp(ep->com.qp->rhp,
1667 ep->com.qp,
1668 IWCH_QP_ATTR_NEXT_STATE,
1669 &attrs, 1);
1670 }
1671 close_complete_upcall(ep);
1672 __state_set(&ep->com, DEAD);
1673 release = 1;
1674 break;
42e31753 1675 case ABORTING:
b038ced7 1676 case DEAD:
c4d49776 1677 break;
b038ced7
SW
1678 default:
1679 BUG_ON(1);
1680 break;
1681 }
1682 spin_unlock_irqrestore(&ep->com.lock, flags);
1683 if (release)
1684 release_ep_resources(ep);
1685 return CPL_RET_BUF_DONE;
1686}
1687
1688/*
1689 * T3A does 3 things when a TERM is received:
1690 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1691 * 2) generate an async event on the QP with the TERMINATE opcode
1692 * 3) post a TERMINATE opcde cqe into the associated CQ.
1693 *
1694 * For (1), we save the message in the qp for later consumer consumption.
1695 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1696 * For (3), we toss the CQE in cxio_poll_cq().
1697 *
1698 * terminate() handles case (1)...
1699 */
1700static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1701{
1702 struct iwch_ep *ep = ctx;
1703
42fb61f0
SW
1704 if (state_read(&ep->com) != FPDU_MODE)
1705 return CPL_RET_BUF_DONE;
1706
33718363 1707 PDBG("%s ep %p\n", __func__, ep);
b038ced7 1708 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
33718363 1709 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
d626f62b
ACM
1710 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1711 skb->len);
b038ced7
SW
1712 ep->com.qp->attr.terminate_msg_len = skb->len;
1713 ep->com.qp->attr.is_terminate_local = 0;
1714 return CPL_RET_BUF_DONE;
1715}
1716
1717static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1718{
1719 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1720 struct iwch_ep *ep = ctx;
1721
33718363 1722 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
b038ced7
SW
1723 rep->status);
1724 if (rep->status) {
1725 struct iwch_qp_attributes attrs;
1726
1727 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
33718363 1728 __func__, ep->hwtid);
2f236735 1729 stop_ep_timer(ep);
b038ced7
SW
1730 attrs.next_state = IWCH_QP_STATE_ERROR;
1731 iwch_modify_qp(ep->com.qp->rhp,
1732 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1733 &attrs, 1);
1734 abort_connection(ep, NULL, GFP_KERNEL);
1735 }
1736 return CPL_RET_BUF_DONE;
1737}
1738
1739static void ep_timeout(unsigned long arg)
1740{
1741 struct iwch_ep *ep = (struct iwch_ep *)arg;
1742 struct iwch_qp_attributes attrs;
1743 unsigned long flags;
989a1780 1744 int abort = 1;
b038ced7
SW
1745
1746 spin_lock_irqsave(&ep->com.lock, flags);
33718363 1747 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
b038ced7
SW
1748 ep->com.state);
1749 switch (ep->com.state) {
1750 case MPA_REQ_SENT:
989a1780 1751 __state_set(&ep->com, ABORTING);
b038ced7
SW
1752 connect_reply_upcall(ep, -ETIMEDOUT);
1753 break;
1754 case MPA_REQ_WAIT:
989a1780 1755 __state_set(&ep->com, ABORTING);
b038ced7 1756 break;
42e31753 1757 case CLOSING:
b038ced7
SW
1758 case MORIBUND:
1759 if (ep->com.cm_id && ep->com.qp) {
1760 attrs.next_state = IWCH_QP_STATE_ERROR;
1761 iwch_modify_qp(ep->com.qp->rhp,
1762 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1763 &attrs, 1);
1764 }
989a1780 1765 __state_set(&ep->com, ABORTING);
b038ced7
SW
1766 break;
1767 default:
989a1780
SW
1768 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1769 __func__, ep, ep->com.state);
1770 WARN_ON(1);
1771 abort = 0;
b038ced7 1772 }
b038ced7 1773 spin_unlock_irqrestore(&ep->com.lock, flags);
989a1780
SW
1774 if (abort)
1775 abort_connection(ep, NULL, GFP_ATOMIC);
b038ced7
SW
1776 put_ep(&ep->com);
1777}
1778
1779int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1780{
1781 int err;
1782 struct iwch_ep *ep = to_ep(cm_id);
33718363 1783 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
b038ced7
SW
1784
1785 if (state_read(&ep->com) == DEAD) {
1786 put_ep(&ep->com);
1787 return -ECONNRESET;
1788 }
1789 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
b038ced7
SW
1790 if (mpa_rev == 0)
1791 abort_connection(ep, NULL, GFP_KERNEL);
1792 else {
1793 err = send_mpa_reject(ep, pdata, pdata_len);
7d526e6b 1794 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
b038ced7
SW
1795 }
1796 return 0;
1797}
1798
1799int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1800{
1801 int err;
1802 struct iwch_qp_attributes attrs;
1803 enum iwch_qp_attr_mask mask;
1804 struct iwch_ep *ep = to_ep(cm_id);
1805 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1806 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1807
33718363 1808 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
de3d3530 1809 if (state_read(&ep->com) == DEAD)
b038ced7 1810 return -ECONNRESET;
b038ced7
SW
1811
1812 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1813 BUG_ON(!qp);
1814
1815 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1816 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1817 abort_connection(ep, NULL, GFP_KERNEL);
1818 return -EINVAL;
1819 }
1820
1821 cm_id->add_ref(cm_id);
1822 ep->com.cm_id = cm_id;
1823 ep->com.qp = qp;
1824
1825 ep->com.rpl_done = 0;
1826 ep->com.rpl_err = 0;
1827 ep->ird = conn_param->ird;
1828 ep->ord = conn_param->ord;
33718363 1829 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
de3d3530 1830
b038ced7 1831 get_ep(&ep->com);
b038ced7
SW
1832
1833 /* bind QP to EP and move to RTS */
1834 attrs.mpa_attr = ep->mpa_attr;
1f71f503 1835 attrs.max_ird = ep->ird;
b038ced7
SW
1836 attrs.max_ord = ep->ord;
1837 attrs.llp_stream_handle = ep;
1838 attrs.next_state = IWCH_QP_STATE_RTS;
1839
1840 /* bind QP and TID with INIT_WR */
1841 mask = IWCH_QP_ATTR_NEXT_STATE |
1842 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1843 IWCH_QP_ATTR_MPA_ATTR |
1844 IWCH_QP_ATTR_MAX_IRD |
1845 IWCH_QP_ATTR_MAX_ORD;
1846
1847 err = iwch_modify_qp(ep->com.qp->rhp,
1848 ep->com.qp, mask, &attrs, 1);
de3d3530
SW
1849 if (err)
1850 goto err;
b038ced7 1851
f8b0dfd1
SW
1852 /* if needed, wait for wr_ack */
1853 if (iwch_rqes_posted(qp)) {
1854 wait_event(ep->com.waitq, ep->com.rpl_done);
1855 err = ep->com.rpl_err;
1856 if (err)
1857 goto err;
1858 }
1859
de3d3530
SW
1860 err = send_mpa_reply(ep, conn_param->private_data,
1861 conn_param->private_data_len);
1862 if (err)
1863 goto err;
1864
de3d3530
SW
1865
1866 state_set(&ep->com, FPDU_MODE);
1867 established_upcall(ep);
1868 put_ep(&ep->com);
1869 return 0;
1870err:
1871 ep->com.cm_id = NULL;
1872 ep->com.qp = NULL;
1873 cm_id->rem_ref(cm_id);
b038ced7
SW
1874 put_ep(&ep->com);
1875 return err;
1876}
1877
8704e9a8
SW
1878static int is_loopback_dst(struct iw_cm_id *cm_id)
1879{
1880 struct net_device *dev;
1881
1882 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1883 if (!dev)
1884 return 0;
1885 dev_put(dev);
1886 return 1;
1887}
1888
b038ced7
SW
1889int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1890{
1891 int err = 0;
1892 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1893 struct iwch_ep *ep;
1894 struct rtable *rt;
1895
8704e9a8
SW
1896 if (is_loopback_dst(cm_id)) {
1897 err = -ENOSYS;
1898 goto out;
1899 }
1900
b038ced7
SW
1901 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1902 if (!ep) {
33718363 1903 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
b038ced7
SW
1904 err = -ENOMEM;
1905 goto out;
1906 }
1907 init_timer(&ep->timer);
1908 ep->plen = conn_param->private_data_len;
1909 if (ep->plen)
1910 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1911 conn_param->private_data, ep->plen);
1912 ep->ird = conn_param->ird;
1913 ep->ord = conn_param->ord;
1914 ep->com.tdev = h->rdev.t3cdev_p;
1915
1916 cm_id->add_ref(cm_id);
1917 ep->com.cm_id = cm_id;
1918 ep->com.qp = get_qhp(h, conn_param->qpn);
1919 BUG_ON(!ep->com.qp);
33718363 1920 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
b038ced7
SW
1921 ep->com.qp, cm_id);
1922
1923 /*
1924 * Allocate an active TID to initiate a TCP connection.
1925 */
1926 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1927 if (ep->atid == -1) {
33718363 1928 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
b038ced7
SW
1929 err = -ENOMEM;
1930 goto fail2;
1931 }
1932
1933 /* find a route */
1934 rt = find_route(h->rdev.t3cdev_p,
1935 cm_id->local_addr.sin_addr.s_addr,
1936 cm_id->remote_addr.sin_addr.s_addr,
1937 cm_id->local_addr.sin_port,
1938 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1939 if (!rt) {
33718363 1940 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
b038ced7
SW
1941 err = -EHOSTUNREACH;
1942 goto fail3;
1943 }
1944 ep->dst = &rt->u.dst;
1945
1946 /* get a l2t entry */
1947 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1948 ep->dst->neighbour->dev);
1949 if (!ep->l2t) {
33718363 1950 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
b038ced7
SW
1951 err = -ENOMEM;
1952 goto fail4;
1953 }
1954
1955 state_set(&ep->com, CONNECTING);
1956 ep->tos = IPTOS_LOWDELAY;
1957 ep->com.local_addr = cm_id->local_addr;
1958 ep->com.remote_addr = cm_id->remote_addr;
1959
1960 /* send connect request to rnic */
1961 err = send_connect(ep);
1962 if (!err)
1963 goto out;
1964
1965 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1966fail4:
1967 dst_release(ep->dst);
1968fail3:
1969 cxgb3_free_atid(ep->com.tdev, ep->atid);
1970fail2:
dc35fac9 1971 cm_id->rem_ref(cm_id);
b038ced7
SW
1972 put_ep(&ep->com);
1973out:
1974 return err;
1975}
1976
1977int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1978{
1979 int err = 0;
1980 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1981 struct iwch_listen_ep *ep;
1982
1983
1984 might_sleep();
1985
1986 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1987 if (!ep) {
33718363 1988 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
b038ced7
SW
1989 err = -ENOMEM;
1990 goto fail1;
1991 }
33718363 1992 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1993 ep->com.tdev = h->rdev.t3cdev_p;
1994 cm_id->add_ref(cm_id);
1995 ep->com.cm_id = cm_id;
1996 ep->backlog = backlog;
1997 ep->com.local_addr = cm_id->local_addr;
1998
1999 /*
2000 * Allocate a server TID.
2001 */
2002 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2003 if (ep->stid == -1) {
33718363 2004 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
b038ced7
SW
2005 err = -ENOMEM;
2006 goto fail2;
2007 }
2008
2009 state_set(&ep->com, LISTEN);
2010 err = listen_start(ep);
2011 if (err)
2012 goto fail3;
2013
2014 /* wait for pass_open_rpl */
2015 wait_event(ep->com.waitq, ep->com.rpl_done);
2016 err = ep->com.rpl_err;
2017 if (!err) {
2018 cm_id->provider_data = ep;
2019 goto out;
2020 }
2021fail3:
2022 cxgb3_free_stid(ep->com.tdev, ep->stid);
2023fail2:
1b07db70 2024 cm_id->rem_ref(cm_id);
b038ced7
SW
2025 put_ep(&ep->com);
2026fail1:
2027out:
2028 return err;
2029}
2030
2031int iwch_destroy_listen(struct iw_cm_id *cm_id)
2032{
2033 int err;
2034 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2035
33718363 2036 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
2037
2038 might_sleep();
2039 state_set(&ep->com, DEAD);
2040 ep->com.rpl_done = 0;
2041 ep->com.rpl_err = 0;
2042 err = listen_stop(ep);
04b5d028
SW
2043 if (err)
2044 goto done;
b038ced7
SW
2045 wait_event(ep->com.waitq, ep->com.rpl_done);
2046 cxgb3_free_stid(ep->com.tdev, ep->stid);
04b5d028 2047done:
b038ced7
SW
2048 err = ep->com.rpl_err;
2049 cm_id->rem_ref(cm_id);
2050 put_ep(&ep->com);
2051 return err;
2052}
2053
2054int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2055{
2056 int ret=0;
2057 unsigned long flags;
2058 int close = 0;
04b5d028
SW
2059 int fatal = 0;
2060 struct t3cdev *tdev;
2061 struct cxio_rdev *rdev;
b038ced7
SW
2062
2063 spin_lock_irqsave(&ep->com.lock, flags);
2064
33718363 2065 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
b038ced7
SW
2066 states[ep->com.state], abrupt);
2067
04b5d028
SW
2068 tdev = (struct t3cdev *)ep->com.tdev;
2069 rdev = (struct cxio_rdev *)tdev->ulp;
2070 if (cxio_fatal_error(rdev)) {
2071 fatal = 1;
2072 close_complete_upcall(ep);
2073 ep->com.state = DEAD;
2074 }
b038ced7
SW
2075 switch (ep->com.state) {
2076 case MPA_REQ_WAIT:
2077 case MPA_REQ_SENT:
2078 case MPA_REQ_RCVD:
2079 case MPA_REP_SENT:
2080 case FPDU_MODE:
b038ced7 2081 close = 1;
989a1780
SW
2082 if (abrupt)
2083 ep->com.state = ABORTING;
2084 else {
2085 ep->com.state = CLOSING;
2086 start_ep_timer(ep);
2087 }
b038ced7
SW
2088 break;
2089 case CLOSING:
b038ced7 2090 close = 1;
989a1780
SW
2091 if (abrupt) {
2092 stop_ep_timer(ep);
2093 ep->com.state = ABORTING;
2094 } else
2095 ep->com.state = MORIBUND;
b038ced7
SW
2096 break;
2097 case MORIBUND:
989a1780
SW
2098 case ABORTING:
2099 case DEAD:
2100 PDBG("%s ignoring disconnect ep %p state %u\n",
2101 __func__, ep, ep->com.state);
b038ced7
SW
2102 break;
2103 default:
2104 BUG();
2105 break;
2106 }
989a1780 2107
b038ced7
SW
2108 spin_unlock_irqrestore(&ep->com.lock, flags);
2109 if (close) {
2110 if (abrupt)
2111 ret = send_abort(ep, NULL, gfp);
2112 else
2113 ret = send_halfclose(ep, gfp);
04b5d028
SW
2114 if (ret)
2115 fatal = 1;
b038ced7 2116 }
04b5d028
SW
2117 if (fatal)
2118 release_ep_resources(ep);
b038ced7
SW
2119 return ret;
2120}
2121
2122int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2123 struct l2t_entry *l2t)
2124{
2125 struct iwch_ep *ep = ctx;
2126
2127 if (ep->dst != old)
2128 return 0;
2129
33718363 2130 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
b038ced7
SW
2131 l2t);
2132 dst_hold(new);
2133 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2134 ep->l2t = l2t;
2135 dst_release(old);
2136 ep->dst = new;
2137 return 1;
2138}
2139
2140/*
2141 * All the CM events are handled on a work queue to have a safe context.
2142 */
2143static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2144{
2145 struct iwch_ep_common *epc = ctx;
2146
2147 get_ep(epc);
2148
2149 /*
2150 * Save ctx and tdev in the skb->cb area.
2151 */
2152 *((void **) skb->cb) = ctx;
2153 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2154
2155 /*
2156 * Queue the skb and schedule the worker thread.
2157 */
2158 skb_queue_tail(&rxq, skb);
2159 queue_work(workq, &skb_work);
2160 return 0;
2161}
2162
1ca19770
SW
2163static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2164{
2165 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2166
2167 if (rpl->status != CPL_ERR_NONE) {
2168 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2169 "for tid %u\n", rpl->status, GET_TID(rpl));
2170 }
2171 return CPL_RET_BUF_DONE;
2172}
2173
b038ced7
SW
2174int __init iwch_cm_init(void)
2175{
2176 skb_queue_head_init(&rxq);
2177
2178 workq = create_singlethread_workqueue("iw_cxgb3");
2179 if (!workq)
2180 return -ENOMEM;
2181
2182 /*
2183 * All upcalls from the T3 Core go to sched() to
2184 * schedule the processing on a work queue.
2185 */
2186 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2187 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2188 t3c_handlers[CPL_RX_DATA] = sched;
2189 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2190 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2191 t3c_handlers[CPL_ABORT_RPL] = sched;
2192 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2193 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2194 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2195 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2196 t3c_handlers[CPL_PEER_CLOSE] = sched;
2197 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2198 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2199 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2200 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
1ca19770 2201 t3c_handlers[CPL_SET_TCB_RPL] = set_tcb_rpl;
b038ced7
SW
2202
2203 /*
2204 * These are the real handlers that are called from a
2205 * work queue.
2206 */
2207 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2208 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2209 work_handlers[CPL_RX_DATA] = rx_data;
2210 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2211 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2212 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2213 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2214 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2215 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2216 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2217 work_handlers[CPL_PEER_CLOSE] = peer_close;
2218 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2219 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2220 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2221 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2222 return 0;
2223}
2224
2225void __exit iwch_cm_term(void)
2226{
2227 flush_workqueue(workq);
2228 destroy_workqueue(workq);
2229}