RDMA/cxgb3: Don't free endpoints early
[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{
874d8df5
SW
285 struct iwch_ep *ep;
286 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
287 struct iwch_ep, com);
288 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
6e47fe43 289 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
874d8df5
SW
290 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
291 dst_release(ep->dst);
292 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
293 }
294 kfree(ep);
b038ced7
SW
295}
296
297static void release_ep_resources(struct iwch_ep *ep)
298{
33718363 299 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
6e47fe43 300 set_bit(RELEASE_RESOURCES, &ep->com.flags);
b038ced7
SW
301 put_ep(&ep->com);
302}
303
304static void process_work(struct work_struct *work)
305{
306 struct sk_buff *skb = NULL;
307 void *ep;
308 struct t3cdev *tdev;
309 int ret;
310
311 while ((skb = skb_dequeue(&rxq))) {
312 ep = *((void **) (skb->cb));
313 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
314 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
315 if (ret & CPL_RET_BUF_DONE)
316 kfree_skb(skb);
317
318 /*
319 * ep was referenced in sched(), and is freed here.
320 */
321 put_ep((struct iwch_ep_common *)ep);
322 }
323}
324
325static int status2errno(int status)
326{
327 switch (status) {
328 case CPL_ERR_NONE:
329 return 0;
330 case CPL_ERR_CONN_RESET:
331 return -ECONNRESET;
332 case CPL_ERR_ARP_MISS:
333 return -EHOSTUNREACH;
334 case CPL_ERR_CONN_TIMEDOUT:
335 return -ETIMEDOUT;
336 case CPL_ERR_TCAM_FULL:
337 return -ENOMEM;
338 case CPL_ERR_CONN_EXIST:
339 return -EADDRINUSE;
340 default:
341 return -EIO;
342 }
343}
344
345/*
346 * Try and reuse skbs already allocated...
347 */
348static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
349{
1f6a849b 350 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
b038ced7
SW
351 skb_trim(skb, 0);
352 skb_get(skb);
353 } else {
354 skb = alloc_skb(len, gfp);
355 }
356 return skb;
357}
358
359static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
360 __be32 peer_ip, __be16 local_port,
361 __be16 peer_port, u8 tos)
362{
363 struct rtable *rt;
364 struct flowi fl = {
365 .oif = 0,
366 .nl_u = {
367 .ip4_u = {
368 .daddr = peer_ip,
369 .saddr = local_ip,
370 .tos = tos}
371 },
372 .proto = IPPROTO_TCP,
373 .uli_u = {
374 .ports = {
375 .sport = local_port,
376 .dport = peer_port}
377 }
378 };
379
f1b050bf 380 if (ip_route_output_flow(&init_net, &rt, &fl, NULL, 0))
b038ced7
SW
381 return NULL;
382 return rt;
383}
384
385static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
386{
387 int i = 0;
388
389 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
390 ++i;
391 return i;
392}
393
394static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
395{
33718363 396 PDBG("%s t3cdev %p\n", __func__, dev);
b038ced7
SW
397 kfree_skb(skb);
398}
399
400/*
401 * Handle an ARP failure for an active open.
402 */
403static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
404{
405 printk(KERN_ERR MOD "ARP failure duing connect\n");
406 kfree_skb(skb);
407}
408
409/*
410 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
411 * and send it along.
412 */
413static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
414{
415 struct cpl_abort_req *req = cplhdr(skb);
416
33718363 417 PDBG("%s t3cdev %p\n", __func__, dev);
b038ced7 418 req->cmd = CPL_ABORT_NO_RST;
04b5d028 419 iwch_cxgb3_ofld_send(dev, skb);
b038ced7
SW
420}
421
422static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
423{
424 struct cpl_close_con_req *req;
425 struct sk_buff *skb;
426
33718363 427 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
428 skb = get_skb(NULL, sizeof(*req), gfp);
429 if (!skb) {
33718363 430 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
b038ced7
SW
431 return -ENOMEM;
432 }
433 skb->priority = CPL_PRIORITY_DATA;
434 set_arp_failure_handler(skb, arp_failure_discard);
435 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
436 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
437 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
438 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
04b5d028 439 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
440}
441
442static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
443{
444 struct cpl_abort_req *req;
445
33718363 446 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
447 skb = get_skb(skb, sizeof(*req), gfp);
448 if (!skb) {
449 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
33718363 450 __func__);
b038ced7
SW
451 return -ENOMEM;
452 }
453 skb->priority = CPL_PRIORITY_DATA;
454 set_arp_failure_handler(skb, abort_arp_failure);
455 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
456 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
457 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
458 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
459 req->cmd = CPL_ABORT_SEND_RST;
04b5d028 460 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
461}
462
463static int send_connect(struct iwch_ep *ep)
464{
465 struct cpl_act_open_req *req;
466 struct sk_buff *skb;
467 u32 opt0h, opt0l, opt2;
468 unsigned int mtu_idx;
469 int wscale;
470
33718363 471 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
472
473 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
474 if (!skb) {
475 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
33718363 476 __func__);
b038ced7
SW
477 return -ENOMEM;
478 }
479 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
480 wscale = compute_wscale(rcv_win);
481 opt0h = V_NAGLE(0) |
482 V_NO_CONG(nocong) |
483 V_KEEP_ALIVE(1) |
484 F_TCAM_BYPASS |
485 V_WND_SCALE(wscale) |
486 V_MSS_IDX(mtu_idx) |
487 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
488 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
489 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
490 skb->priority = CPL_PRIORITY_SETUP;
491 set_arp_failure_handler(skb, act_open_req_arp_failure);
492
493 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
494 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
495 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
496 req->local_port = ep->com.local_addr.sin_port;
497 req->peer_port = ep->com.remote_addr.sin_port;
498 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
499 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
500 req->opt0h = htonl(opt0h);
501 req->opt0l = htonl(opt0l);
502 req->params = 0;
503 req->opt2 = htonl(opt2);
04b5d028 504 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
505}
506
507static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
508{
509 int mpalen;
510 struct tx_data_wr *req;
511 struct mpa_message *mpa;
512 int len;
513
33718363 514 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
b038ced7
SW
515
516 BUG_ON(skb_cloned(skb));
517
518 mpalen = sizeof(*mpa) + ep->plen;
4305b541 519 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
b038ced7
SW
520 kfree_skb(skb);
521 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
522 if (!skb) {
523 connect_reply_upcall(ep, -ENOMEM);
524 return;
525 }
526 }
527 skb_trim(skb, 0);
528 skb_reserve(skb, sizeof(*req));
529 skb_put(skb, mpalen);
530 skb->priority = CPL_PRIORITY_DATA;
531 mpa = (struct mpa_message *) skb->data;
532 memset(mpa, 0, sizeof(*mpa));
533 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
534 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
535 (markers_enabled ? MPA_MARKERS : 0);
536 mpa->private_data_size = htons(ep->plen);
537 mpa->revision = mpa_rev;
538
539 if (ep->plen)
540 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
541
542 /*
543 * Reference the mpa skb. This ensures the data area
544 * will remain in memory until the hw acks the tx.
545 * Function tx_ack() will deref it.
546 */
547 skb_get(skb);
548 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 549 skb_reset_transport_header(skb);
b038ced7
SW
550 len = skb->len;
551 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 552 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
553 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
554 req->len = htonl(len);
555 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
556 V_TX_SNDBUF(snd_win>>15));
de3d3530 557 req->flags = htonl(F_TX_INIT);
b038ced7
SW
558 req->sndseq = htonl(ep->snd_seq);
559 BUG_ON(ep->mpa_skb);
560 ep->mpa_skb = skb;
04b5d028 561 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
562 start_ep_timer(ep);
563 state_set(&ep->com, MPA_REQ_SENT);
564 return;
565}
566
567static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
568{
569 int mpalen;
570 struct tx_data_wr *req;
571 struct mpa_message *mpa;
572 struct sk_buff *skb;
573
33718363 574 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
b038ced7
SW
575
576 mpalen = sizeof(*mpa) + plen;
577
578 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
579 if (!skb) {
33718363 580 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
b038ced7
SW
581 return -ENOMEM;
582 }
583 skb_reserve(skb, sizeof(*req));
584 mpa = (struct mpa_message *) skb_put(skb, mpalen);
585 memset(mpa, 0, sizeof(*mpa));
586 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
587 mpa->flags = MPA_REJECT;
588 mpa->revision = mpa_rev;
589 mpa->private_data_size = htons(plen);
590 if (plen)
591 memcpy(mpa->private_data, pdata, plen);
592
593 /*
594 * Reference the mpa skb again. This ensures the data area
595 * will remain in memory until the hw acks the tx.
596 * Function tx_ack() will deref it.
597 */
598 skb_get(skb);
599 skb->priority = CPL_PRIORITY_DATA;
600 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 601 skb_reset_transport_header(skb);
b038ced7 602 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 603 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
604 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
605 req->len = htonl(mpalen);
606 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
607 V_TX_SNDBUF(snd_win>>15));
de3d3530 608 req->flags = htonl(F_TX_INIT);
b038ced7
SW
609 req->sndseq = htonl(ep->snd_seq);
610 BUG_ON(ep->mpa_skb);
611 ep->mpa_skb = skb;
04b5d028 612 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
613}
614
615static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
616{
617 int mpalen;
618 struct tx_data_wr *req;
619 struct mpa_message *mpa;
620 int len;
621 struct sk_buff *skb;
622
33718363 623 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
b038ced7
SW
624
625 mpalen = sizeof(*mpa) + plen;
626
627 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
628 if (!skb) {
33718363 629 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
b038ced7
SW
630 return -ENOMEM;
631 }
632 skb->priority = CPL_PRIORITY_DATA;
633 skb_reserve(skb, sizeof(*req));
634 mpa = (struct mpa_message *) skb_put(skb, mpalen);
635 memset(mpa, 0, sizeof(*mpa));
636 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
637 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
638 (markers_enabled ? MPA_MARKERS : 0);
639 mpa->revision = mpa_rev;
640 mpa->private_data_size = htons(plen);
641 if (plen)
642 memcpy(mpa->private_data, pdata, plen);
643
644 /*
645 * Reference the mpa skb. This ensures the data area
646 * will remain in memory until the hw acks the tx.
647 * Function tx_ack() will deref it.
648 */
649 skb_get(skb);
650 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 651 skb_reset_transport_header(skb);
b038ced7
SW
652 len = skb->len;
653 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 654 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
655 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
656 req->len = htonl(len);
657 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
658 V_TX_SNDBUF(snd_win>>15));
de3d3530 659 req->flags = htonl(F_TX_INIT);
b038ced7
SW
660 req->sndseq = htonl(ep->snd_seq);
661 ep->mpa_skb = skb;
662 state_set(&ep->com, MPA_REP_SENT);
04b5d028 663 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
664}
665
666static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
667{
668 struct iwch_ep *ep = ctx;
669 struct cpl_act_establish *req = cplhdr(skb);
670 unsigned int tid = GET_TID(req);
671
33718363 672 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
b038ced7
SW
673
674 dst_confirm(ep->dst);
675
676 /* setup the hwtid for this connection */
677 ep->hwtid = tid;
678 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
679
680 ep->snd_seq = ntohl(req->snd_isn);
de3d3530 681 ep->rcv_seq = ntohl(req->rcv_isn);
b038ced7
SW
682
683 set_emss(ep, ntohs(req->tcp_opt));
684
685 /* dealloc the atid */
686 cxgb3_free_atid(ep->com.tdev, ep->atid);
687
688 /* start MPA negotiation */
689 send_mpa_req(ep, skb);
690
691 return 0;
692}
693
694static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
695{
696 PDBG("%s ep %p\n", __FILE__, ep);
697 state_set(&ep->com, ABORTING);
698 send_abort(ep, skb, gfp);
699}
700
701static void close_complete_upcall(struct iwch_ep *ep)
702{
703 struct iw_cm_event event;
704
33718363 705 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
706 memset(&event, 0, sizeof(event));
707 event.event = IW_CM_EVENT_CLOSE;
708 if (ep->com.cm_id) {
709 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
710 ep, ep->com.cm_id, ep->hwtid);
711 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
712 ep->com.cm_id->rem_ref(ep->com.cm_id);
713 ep->com.cm_id = NULL;
714 ep->com.qp = NULL;
715 }
716}
717
718static void peer_close_upcall(struct iwch_ep *ep)
719{
720 struct iw_cm_event event;
721
33718363 722 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
723 memset(&event, 0, sizeof(event));
724 event.event = IW_CM_EVENT_DISCONNECT;
725 if (ep->com.cm_id) {
726 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
727 ep, ep->com.cm_id, ep->hwtid);
728 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
729 }
730}
731
732static void peer_abort_upcall(struct iwch_ep *ep)
733{
734 struct iw_cm_event event;
735
33718363 736 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
737 memset(&event, 0, sizeof(event));
738 event.event = IW_CM_EVENT_CLOSE;
739 event.status = -ECONNRESET;
740 if (ep->com.cm_id) {
741 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
742 ep->com.cm_id, ep->hwtid);
743 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
744 ep->com.cm_id->rem_ref(ep->com.cm_id);
745 ep->com.cm_id = NULL;
746 ep->com.qp = NULL;
747 }
748}
749
750static void connect_reply_upcall(struct iwch_ep *ep, int status)
751{
752 struct iw_cm_event event;
753
33718363 754 PDBG("%s ep %p status %d\n", __func__, ep, status);
b038ced7
SW
755 memset(&event, 0, sizeof(event));
756 event.event = IW_CM_EVENT_CONNECT_REPLY;
757 event.status = status;
758 event.local_addr = ep->com.local_addr;
759 event.remote_addr = ep->com.remote_addr;
760
761 if ((status == 0) || (status == -ECONNREFUSED)) {
762 event.private_data_len = ep->plen;
763 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
764 }
765 if (ep->com.cm_id) {
33718363 766 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
b038ced7
SW
767 ep->hwtid, status);
768 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
769 }
770 if (status < 0) {
771 ep->com.cm_id->rem_ref(ep->com.cm_id);
772 ep->com.cm_id = NULL;
773 ep->com.qp = NULL;
774 }
775}
776
777static void connect_request_upcall(struct iwch_ep *ep)
778{
779 struct iw_cm_event event;
780
33718363 781 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
b038ced7
SW
782 memset(&event, 0, sizeof(event));
783 event.event = IW_CM_EVENT_CONNECT_REQUEST;
784 event.local_addr = ep->com.local_addr;
785 event.remote_addr = ep->com.remote_addr;
786 event.private_data_len = ep->plen;
787 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
788 event.provider_data = ep;
6e47fe43
SW
789 if (state_read(&ep->parent_ep->com) != DEAD) {
790 get_ep(&ep->com);
b038ced7
SW
791 ep->parent_ep->com.cm_id->event_handler(
792 ep->parent_ep->com.cm_id,
793 &event);
6e47fe43 794 }
b038ced7
SW
795 put_ep(&ep->parent_ep->com);
796 ep->parent_ep = NULL;
797}
798
799static void established_upcall(struct iwch_ep *ep)
800{
801 struct iw_cm_event event;
802
33718363 803 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
804 memset(&event, 0, sizeof(event));
805 event.event = IW_CM_EVENT_ESTABLISHED;
806 if (ep->com.cm_id) {
33718363 807 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
b038ced7
SW
808 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
809 }
810}
811
812static int update_rx_credits(struct iwch_ep *ep, u32 credits)
813{
814 struct cpl_rx_data_ack *req;
815 struct sk_buff *skb;
816
33718363 817 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
b038ced7
SW
818 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
819 if (!skb) {
820 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
821 return 0;
822 }
823
824 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
825 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
826 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
827 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
828 skb->priority = CPL_PRIORITY_ACK;
04b5d028 829 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
830 return credits;
831}
832
833static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
834{
835 struct mpa_message *mpa;
836 u16 plen;
837 struct iwch_qp_attributes attrs;
838 enum iwch_qp_attr_mask mask;
839 int err;
840
33718363 841 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
842
843 /*
844 * Stop mpa timer. If it expired, then the state has
845 * changed and we bail since ep_timeout already aborted
846 * the connection.
847 */
848 stop_ep_timer(ep);
849 if (state_read(&ep->com) != MPA_REQ_SENT)
850 return;
851
852 /*
853 * If we get more than the supported amount of private data
854 * then we must fail this connection.
855 */
856 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
857 err = -EINVAL;
858 goto err;
859 }
860
861 /*
862 * copy the new data into our accumulation buffer.
863 */
d626f62b
ACM
864 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
865 skb->len);
b038ced7
SW
866 ep->mpa_pkt_len += skb->len;
867
868 /*
869 * if we don't even have the mpa message, then bail.
870 */
871 if (ep->mpa_pkt_len < sizeof(*mpa))
872 return;
873 mpa = (struct mpa_message *) ep->mpa_pkt;
874
875 /* Validate MPA header. */
876 if (mpa->revision != mpa_rev) {
877 err = -EPROTO;
878 goto err;
879 }
880 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
881 err = -EPROTO;
882 goto err;
883 }
884
885 plen = ntohs(mpa->private_data_size);
886
887 /*
888 * Fail if there's too much private data.
889 */
890 if (plen > MPA_MAX_PRIVATE_DATA) {
891 err = -EPROTO;
892 goto err;
893 }
894
895 /*
896 * If plen does not account for pkt size
897 */
898 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
899 err = -EPROTO;
900 goto err;
901 }
902
903 ep->plen = (u8) plen;
904
905 /*
906 * If we don't have all the pdata yet, then bail.
907 * We'll continue process when more data arrives.
908 */
909 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
910 return;
911
912 if (mpa->flags & MPA_REJECT) {
913 err = -ECONNREFUSED;
914 goto err;
915 }
916
917 /*
918 * If we get here we have accumulated the entire mpa
919 * start reply message including private data. And
920 * the MPA header is valid.
921 */
922 state_set(&ep->com, FPDU_MODE);
f8b0dfd1 923 ep->mpa_attr.initiator = 1;
b038ced7
SW
924 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
925 ep->mpa_attr.recv_marker_enabled = markers_enabled;
926 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
927 ep->mpa_attr.version = mpa_rev;
928 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
33718363 929 "xmit_marker_enabled=%d, version=%d\n", __func__,
b038ced7
SW
930 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
931 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
932
933 attrs.mpa_attr = ep->mpa_attr;
934 attrs.max_ird = ep->ird;
935 attrs.max_ord = ep->ord;
936 attrs.llp_stream_handle = ep;
937 attrs.next_state = IWCH_QP_STATE_RTS;
938
939 mask = IWCH_QP_ATTR_NEXT_STATE |
940 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
941 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
942
943 /* bind QP and TID with INIT_WR */
944 err = iwch_modify_qp(ep->com.qp->rhp,
945 ep->com.qp, mask, &attrs, 1);
f8b0dfd1
SW
946 if (err)
947 goto err;
948
949 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
950 iwch_post_zb_read(ep->com.qp);
951 }
952
953 goto out;
b038ced7
SW
954err:
955 abort_connection(ep, skb, GFP_KERNEL);
956out:
957 connect_reply_upcall(ep, err);
958 return;
959}
960
961static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
962{
963 struct mpa_message *mpa;
964 u16 plen;
965
33718363 966 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
967
968 /*
969 * Stop mpa timer. If it expired, then the state has
970 * changed and we bail since ep_timeout already aborted
971 * the connection.
972 */
973 stop_ep_timer(ep);
974 if (state_read(&ep->com) != MPA_REQ_WAIT)
975 return;
976
977 /*
978 * If we get more than the supported amount of private data
979 * then we must fail this connection.
980 */
981 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
982 abort_connection(ep, skb, GFP_KERNEL);
983 return;
984 }
985
33718363 986 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
b038ced7
SW
987
988 /*
989 * Copy the new data into our accumulation buffer.
990 */
d626f62b
ACM
991 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
992 skb->len);
b038ced7
SW
993 ep->mpa_pkt_len += skb->len;
994
995 /*
996 * If we don't even have the mpa message, then bail.
997 * We'll continue process when more data arrives.
998 */
999 if (ep->mpa_pkt_len < sizeof(*mpa))
1000 return;
33718363 1001 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
b038ced7
SW
1002 mpa = (struct mpa_message *) ep->mpa_pkt;
1003
1004 /*
1005 * Validate MPA Header.
1006 */
1007 if (mpa->revision != mpa_rev) {
1008 abort_connection(ep, skb, GFP_KERNEL);
1009 return;
1010 }
1011
1012 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1013 abort_connection(ep, skb, GFP_KERNEL);
1014 return;
1015 }
1016
1017 plen = ntohs(mpa->private_data_size);
1018
1019 /*
1020 * Fail if there's too much private data.
1021 */
1022 if (plen > MPA_MAX_PRIVATE_DATA) {
1023 abort_connection(ep, skb, GFP_KERNEL);
1024 return;
1025 }
1026
1027 /*
1028 * If plen does not account for pkt size
1029 */
1030 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1031 abort_connection(ep, skb, GFP_KERNEL);
1032 return;
1033 }
1034 ep->plen = (u8) plen;
1035
1036 /*
1037 * If we don't have all the pdata yet, then bail.
1038 */
1039 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1040 return;
1041
1042 /*
1043 * If we get here we have accumulated the entire mpa
1044 * start reply message including private data.
1045 */
f8b0dfd1 1046 ep->mpa_attr.initiator = 0;
b038ced7
SW
1047 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1048 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1049 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1050 ep->mpa_attr.version = mpa_rev;
1051 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
33718363 1052 "xmit_marker_enabled=%d, version=%d\n", __func__,
b038ced7
SW
1053 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1054 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1055
1056 state_set(&ep->com, MPA_REQ_RCVD);
1057
1058 /* drive upcall */
1059 connect_request_upcall(ep);
1060 return;
1061}
1062
1063static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1064{
1065 struct iwch_ep *ep = ctx;
1066 struct cpl_rx_data *hdr = cplhdr(skb);
1067 unsigned int dlen = ntohs(hdr->len);
1068
33718363 1069 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
b038ced7
SW
1070
1071 skb_pull(skb, sizeof(*hdr));
1072 skb_trim(skb, dlen);
1073
de3d3530
SW
1074 ep->rcv_seq += dlen;
1075 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1076
b038ced7
SW
1077 switch (state_read(&ep->com)) {
1078 case MPA_REQ_SENT:
1079 process_mpa_reply(ep, skb);
1080 break;
1081 case MPA_REQ_WAIT:
1082 process_mpa_request(ep, skb);
1083 break;
1084 case MPA_REP_SENT:
1085 break;
1086 default:
1087 printk(KERN_ERR MOD "%s Unexpected streaming data."
1088 " ep %p state %d tid %d\n",
33718363 1089 __func__, ep, state_read(&ep->com), ep->hwtid);
b038ced7
SW
1090
1091 /*
1092 * The ep will timeout and inform the ULP of the failure.
1093 * See ep_timeout().
1094 */
1095 break;
1096 }
1097
1098 /* update RX credits */
1099 update_rx_credits(ep, dlen);
1100
1101 return CPL_RET_BUF_DONE;
1102}
1103
1104/*
1105 * Upcall from the adapter indicating data has been transmitted.
1106 * For us its just the single MPA request or reply. We can now free
1107 * the skb holding the mpa message.
1108 */
1109static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1110{
1111 struct iwch_ep *ep = ctx;
1112 struct cpl_wr_ack *hdr = cplhdr(skb);
1113 unsigned int credits = ntohs(hdr->credits);
b038ced7 1114
33718363 1115 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
b038ced7 1116
f8b0dfd1
SW
1117 if (credits == 0) {
1118 PDBG(KERN_ERR "%s 0 credit ack ep %p state %u\n",
1119 __func__, ep, state_read(&ep->com));
b038ced7 1120 return CPL_RET_BUF_DONE;
f8b0dfd1
SW
1121 }
1122
b038ced7 1123 BUG_ON(credits != 1);
b038ced7 1124 dst_confirm(ep->dst);
f8b0dfd1
SW
1125 if (!ep->mpa_skb) {
1126 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1127 __func__, ep, state_read(&ep->com));
1128 if (ep->mpa_attr.initiator) {
1129 PDBG("%s initiator ep %p state %u\n",
1130 __func__, ep, state_read(&ep->com));
1131 if (peer2peer)
1132 iwch_post_zb_read(ep->com.qp);
1133 } else {
1134 PDBG("%s responder ep %p state %u\n",
1135 __func__, ep, state_read(&ep->com));
1136 ep->com.rpl_done = 1;
1137 wake_up(&ep->com.waitq);
1138 }
1139 } else {
1140 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1141 __func__, ep, state_read(&ep->com));
1142 kfree_skb(ep->mpa_skb);
1143 ep->mpa_skb = NULL;
b038ced7
SW
1144 }
1145 return CPL_RET_BUF_DONE;
1146}
1147
1148static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1149{
1150 struct iwch_ep *ep = ctx;
989a1780
SW
1151 unsigned long flags;
1152 int release = 0;
b038ced7 1153
33718363 1154 PDBG("%s ep %p\n", __func__, ep);
989a1780 1155 BUG_ON(!ep);
b038ced7 1156
aff9e39d
SW
1157 /*
1158 * We get 2 abort replies from the HW. The first one must
1159 * be ignored except for scribbling that we need one more.
1160 */
6e47fe43 1161 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
aff9e39d
SW
1162 return CPL_RET_BUF_DONE;
1163 }
1164
989a1780
SW
1165 spin_lock_irqsave(&ep->com.lock, flags);
1166 switch (ep->com.state) {
1167 case ABORTING:
1168 close_complete_upcall(ep);
1169 __state_set(&ep->com, DEAD);
1170 release = 1;
1171 break;
1172 default:
1173 printk(KERN_ERR "%s ep %p state %d\n",
1174 __func__, ep, ep->com.state);
1175 break;
1176 }
1177 spin_unlock_irqrestore(&ep->com.lock, flags);
1178
1179 if (release)
1180 release_ep_resources(ep);
b038ced7
SW
1181 return CPL_RET_BUF_DONE;
1182}
1183
96d0e493
SW
1184/*
1185 * Return whether a failed active open has allocated a TID
1186 */
1187static inline int act_open_has_tid(int status)
1188{
1189 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1190 status != CPL_ERR_ARP_MISS;
1191}
1192
b038ced7
SW
1193static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1194{
1195 struct iwch_ep *ep = ctx;
1196 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1197
33718363 1198 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
b038ced7
SW
1199 status2errno(rpl->status));
1200 connect_reply_upcall(ep, status2errno(rpl->status));
1201 state_set(&ep->com, DEAD);
8176d297 1202 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
b038ced7
SW
1203 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1204 cxgb3_free_atid(ep->com.tdev, ep->atid);
1205 dst_release(ep->dst);
1206 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1207 put_ep(&ep->com);
1208 return CPL_RET_BUF_DONE;
1209}
1210
1211static int listen_start(struct iwch_listen_ep *ep)
1212{
1213 struct sk_buff *skb;
1214 struct cpl_pass_open_req *req;
1215
33718363 1216 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1217 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1218 if (!skb) {
1219 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1220 return -ENOMEM;
1221 }
1222
1223 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1224 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1225 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1226 req->local_port = ep->com.local_addr.sin_port;
1227 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1228 req->peer_port = 0;
1229 req->peer_ip = 0;
1230 req->peer_netmask = 0;
1231 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1232 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1233 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1234
1235 skb->priority = 1;
04b5d028 1236 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
1237}
1238
1239static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1240{
1241 struct iwch_listen_ep *ep = ctx;
1242 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1243
33718363 1244 PDBG("%s ep %p status %d error %d\n", __func__, ep,
b038ced7
SW
1245 rpl->status, status2errno(rpl->status));
1246 ep->com.rpl_err = status2errno(rpl->status);
1247 ep->com.rpl_done = 1;
1248 wake_up(&ep->com.waitq);
1249
1250 return CPL_RET_BUF_DONE;
1251}
1252
1253static int listen_stop(struct iwch_listen_ep *ep)
1254{
1255 struct sk_buff *skb;
1256 struct cpl_close_listserv_req *req;
1257
33718363 1258 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1259 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1260 if (!skb) {
33718363 1261 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
b038ced7
SW
1262 return -ENOMEM;
1263 }
1264 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1265 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
60be4b59 1266 req->cpu_idx = 0;
b038ced7
SW
1267 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1268 skb->priority = 1;
04b5d028 1269 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
1270}
1271
1272static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1273 void *ctx)
1274{
1275 struct iwch_listen_ep *ep = ctx;
1276 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1277
33718363 1278 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1279 ep->com.rpl_err = status2errno(rpl->status);
1280 ep->com.rpl_done = 1;
1281 wake_up(&ep->com.waitq);
1282 return CPL_RET_BUF_DONE;
1283}
1284
1285static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1286{
1287 struct cpl_pass_accept_rpl *rpl;
1288 unsigned int mtu_idx;
1289 u32 opt0h, opt0l, opt2;
1290 int wscale;
1291
33718363 1292 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1293 BUG_ON(skb_cloned(skb));
1294 skb_trim(skb, sizeof(*rpl));
1295 skb_get(skb);
1296 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1297 wscale = compute_wscale(rcv_win);
1298 opt0h = V_NAGLE(0) |
1299 V_NO_CONG(nocong) |
1300 V_KEEP_ALIVE(1) |
1301 F_TCAM_BYPASS |
1302 V_WND_SCALE(wscale) |
1303 V_MSS_IDX(mtu_idx) |
1304 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1305 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1306 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1307
1308 rpl = cplhdr(skb);
1309 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1310 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1311 rpl->peer_ip = peer_ip;
1312 rpl->opt0h = htonl(opt0h);
1313 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1314 rpl->opt2 = htonl(opt2);
1315 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1316 skb->priority = CPL_PRIORITY_SETUP;
04b5d028 1317 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
1318
1319 return;
1320}
1321
1322static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1323 struct sk_buff *skb)
1324{
33718363 1325 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
b038ced7
SW
1326 peer_ip);
1327 BUG_ON(skb_cloned(skb));
1328 skb_trim(skb, sizeof(struct cpl_tid_release));
1329 skb_get(skb);
1330
8176d297 1331 if (tdev->type != T3A)
b038ced7
SW
1332 release_tid(tdev, hwtid, skb);
1333 else {
1334 struct cpl_pass_accept_rpl *rpl;
1335
1336 rpl = cplhdr(skb);
1337 skb->priority = CPL_PRIORITY_SETUP;
1338 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1339 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1340 hwtid));
1341 rpl->peer_ip = peer_ip;
1342 rpl->opt0h = htonl(F_TCAM_BYPASS);
1343 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1344 rpl->opt2 = 0;
1345 rpl->rsvd = rpl->opt2;
04b5d028 1346 iwch_cxgb3_ofld_send(tdev, skb);
b038ced7
SW
1347 }
1348}
1349
1350static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1351{
1352 struct iwch_ep *child_ep, *parent_ep = ctx;
1353 struct cpl_pass_accept_req *req = cplhdr(skb);
1354 unsigned int hwtid = GET_TID(req);
1355 struct dst_entry *dst;
1356 struct l2t_entry *l2t;
1357 struct rtable *rt;
1358 struct iff_mac tim;
1359
33718363 1360 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
b038ced7
SW
1361
1362 if (state_read(&parent_ep->com) != LISTEN) {
1363 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
33718363 1364 __func__);
b038ced7
SW
1365 goto reject;
1366 }
1367
1368 /*
1369 * Find the netdev for this connection request.
1370 */
1371 tim.mac_addr = req->dst_mac;
1372 tim.vlan_tag = ntohs(req->vlan_tag);
1373 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1374 printk(KERN_ERR
1375 "%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
33718363 1376 __func__,
b038ced7
SW
1377 req->dst_mac[0],
1378 req->dst_mac[1],
1379 req->dst_mac[2],
1380 req->dst_mac[3],
1381 req->dst_mac[4],
1382 req->dst_mac[5]);
1383 goto reject;
1384 }
1385
1386 /* Find output route */
1387 rt = find_route(tdev,
1388 req->local_ip,
1389 req->peer_ip,
1390 req->local_port,
1391 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1392 if (!rt) {
1393 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
33718363 1394 __func__);
b038ced7
SW
1395 goto reject;
1396 }
1397 dst = &rt->u.dst;
1398 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1399 if (!l2t) {
1400 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
33718363 1401 __func__);
b038ced7
SW
1402 dst_release(dst);
1403 goto reject;
1404 }
1405 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1406 if (!child_ep) {
1407 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
33718363 1408 __func__);
b038ced7
SW
1409 l2t_release(L2DATA(tdev), l2t);
1410 dst_release(dst);
1411 goto reject;
1412 }
1413 state_set(&child_ep->com, CONNECTING);
1414 child_ep->com.tdev = tdev;
1415 child_ep->com.cm_id = NULL;
1416 child_ep->com.local_addr.sin_family = PF_INET;
1417 child_ep->com.local_addr.sin_port = req->local_port;
1418 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1419 child_ep->com.remote_addr.sin_family = PF_INET;
1420 child_ep->com.remote_addr.sin_port = req->peer_port;
1421 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1422 get_ep(&parent_ep->com);
1423 child_ep->parent_ep = parent_ep;
1424 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1425 child_ep->l2t = l2t;
1426 child_ep->dst = dst;
1427 child_ep->hwtid = hwtid;
1428 init_timer(&child_ep->timer);
1429 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1430 accept_cr(child_ep, req->peer_ip, skb);
1431 goto out;
1432reject:
1433 reject_cr(tdev, hwtid, req->peer_ip, skb);
1434out:
1435 return CPL_RET_BUF_DONE;
1436}
1437
1438static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1439{
1440 struct iwch_ep *ep = ctx;
1441 struct cpl_pass_establish *req = cplhdr(skb);
1442
33718363 1443 PDBG("%s ep %p\n", __func__, ep);
b038ced7 1444 ep->snd_seq = ntohl(req->snd_isn);
de3d3530 1445 ep->rcv_seq = ntohl(req->rcv_isn);
b038ced7
SW
1446
1447 set_emss(ep, ntohs(req->tcp_opt));
1448
1449 dst_confirm(ep->dst);
1450 state_set(&ep->com, MPA_REQ_WAIT);
1451 start_ep_timer(ep);
1452
1453 return CPL_RET_BUF_DONE;
1454}
1455
1456static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1457{
1458 struct iwch_ep *ep = ctx;
1459 struct iwch_qp_attributes attrs;
1460 unsigned long flags;
1461 int disconnect = 1;
1462 int release = 0;
1463
33718363 1464 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1465 dst_confirm(ep->dst);
1466
1467 spin_lock_irqsave(&ep->com.lock, flags);
1468 switch (ep->com.state) {
1469 case MPA_REQ_WAIT:
1470 __state_set(&ep->com, CLOSING);
1471 break;
1472 case MPA_REQ_SENT:
1473 __state_set(&ep->com, CLOSING);
1474 connect_reply_upcall(ep, -ECONNRESET);
1475 break;
1476 case MPA_REQ_RCVD:
1477
1478 /*
1479 * We're gonna mark this puppy DEAD, but keep
1480 * the reference on it until the ULP accepts or
1481 * rejects the CR.
1482 */
1483 __state_set(&ep->com, CLOSING);
b038ced7
SW
1484 break;
1485 case MPA_REP_SENT:
1486 __state_set(&ep->com, CLOSING);
1487 ep->com.rpl_done = 1;
1488 ep->com.rpl_err = -ECONNRESET;
1489 PDBG("waking up ep %p\n", ep);
1490 wake_up(&ep->com.waitq);
1491 break;
1492 case FPDU_MODE:
42e31753 1493 start_ep_timer(ep);
b038ced7
SW
1494 __state_set(&ep->com, CLOSING);
1495 attrs.next_state = IWCH_QP_STATE_CLOSING;
1496 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1497 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1498 peer_close_upcall(ep);
1499 break;
1500 case ABORTING:
1501 disconnect = 0;
1502 break;
1503 case CLOSING:
b038ced7
SW
1504 __state_set(&ep->com, MORIBUND);
1505 disconnect = 0;
1506 break;
1507 case MORIBUND:
1508 stop_ep_timer(ep);
1509 if (ep->com.cm_id && ep->com.qp) {
1510 attrs.next_state = IWCH_QP_STATE_IDLE;
1511 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1512 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1513 }
1514 close_complete_upcall(ep);
1515 __state_set(&ep->com, DEAD);
1516 release = 1;
1517 disconnect = 0;
1518 break;
1519 case DEAD:
1520 disconnect = 0;
1521 break;
1522 default:
1523 BUG_ON(1);
1524 }
1525 spin_unlock_irqrestore(&ep->com.lock, flags);
1526 if (disconnect)
1527 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1528 if (release)
1529 release_ep_resources(ep);
1530 return CPL_RET_BUF_DONE;
1531}
1532
1533/*
1534 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1535 */
2b540355 1536static int is_neg_adv_abort(unsigned int status)
b038ced7
SW
1537{
1538 return status == CPL_ERR_RTX_NEG_ADVICE ||
1539 status == CPL_ERR_PERSIST_NEG_ADVICE;
1540}
1541
1542static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1543{
1544 struct cpl_abort_req_rss *req = cplhdr(skb);
1545 struct iwch_ep *ep = ctx;
1546 struct cpl_abort_rpl *rpl;
1547 struct sk_buff *rpl_skb;
1548 struct iwch_qp_attributes attrs;
1549 int ret;
989a1780
SW
1550 int release = 0;
1551 unsigned long flags;
b038ced7 1552
1580367e 1553 if (is_neg_adv_abort(req->status)) {
33718363 1554 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1580367e
SW
1555 ep->hwtid);
1556 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1557 return CPL_RET_BUF_DONE;
1558 }
1559
aff9e39d
SW
1560 /*
1561 * We get 2 peer aborts from the HW. The first one must
1562 * be ignored except for scribbling that we need one more.
1563 */
6e47fe43 1564 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
aff9e39d
SW
1565 return CPL_RET_BUF_DONE;
1566 }
1567
989a1780
SW
1568 spin_lock_irqsave(&ep->com.lock, flags);
1569 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1570 switch (ep->com.state) {
b038ced7
SW
1571 case CONNECTING:
1572 break;
1573 case MPA_REQ_WAIT:
adf376b3 1574 stop_ep_timer(ep);
b038ced7
SW
1575 break;
1576 case MPA_REQ_SENT:
adf376b3 1577 stop_ep_timer(ep);
b038ced7
SW
1578 connect_reply_upcall(ep, -ECONNRESET);
1579 break;
1580 case MPA_REP_SENT:
1581 ep->com.rpl_done = 1;
1582 ep->com.rpl_err = -ECONNRESET;
1583 PDBG("waking up ep %p\n", ep);
1584 wake_up(&ep->com.waitq);
1585 break;
1586 case MPA_REQ_RCVD:
1587
1588 /*
1589 * We're gonna mark this puppy DEAD, but keep
1590 * the reference on it until the ULP accepts or
1591 * rejects the CR.
1592 */
b038ced7
SW
1593 break;
1594 case MORIBUND:
42e31753 1595 case CLOSING:
b038ced7 1596 stop_ep_timer(ep);
42e31753 1597 /*FALLTHROUGH*/
b038ced7 1598 case FPDU_MODE:
b038ced7
SW
1599 if (ep->com.cm_id && ep->com.qp) {
1600 attrs.next_state = IWCH_QP_STATE_ERROR;
1601 ret = iwch_modify_qp(ep->com.qp->rhp,
1602 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1603 &attrs, 1);
1604 if (ret)
1605 printk(KERN_ERR MOD
1606 "%s - qp <- error failed!\n",
33718363 1607 __func__);
b038ced7
SW
1608 }
1609 peer_abort_upcall(ep);
1610 break;
1611 case ABORTING:
1612 break;
1613 case DEAD:
33718363 1614 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
989a1780 1615 spin_unlock_irqrestore(&ep->com.lock, flags);
b038ced7
SW
1616 return CPL_RET_BUF_DONE;
1617 default:
1618 BUG_ON(1);
1619 break;
1620 }
1621 dst_confirm(ep->dst);
989a1780
SW
1622 if (ep->com.state != ABORTING) {
1623 __state_set(&ep->com, DEAD);
1624 release = 1;
1625 }
1626 spin_unlock_irqrestore(&ep->com.lock, flags);
b038ced7
SW
1627
1628 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1629 if (!rpl_skb) {
1630 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
33718363 1631 __func__);
989a1780
SW
1632 release = 1;
1633 goto out;
b038ced7
SW
1634 }
1635 rpl_skb->priority = CPL_PRIORITY_DATA;
1636 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1637 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1638 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1639 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1640 rpl->cmd = CPL_ABORT_NO_RST;
04b5d028 1641 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
989a1780
SW
1642out:
1643 if (release)
b038ced7 1644 release_ep_resources(ep);
b038ced7
SW
1645 return CPL_RET_BUF_DONE;
1646}
1647
1648static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1649{
1650 struct iwch_ep *ep = ctx;
1651 struct iwch_qp_attributes attrs;
1652 unsigned long flags;
1653 int release = 0;
1654
33718363 1655 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1656 BUG_ON(!ep);
1657
1658 /* The cm_id may be null if we failed to connect */
1659 spin_lock_irqsave(&ep->com.lock, flags);
1660 switch (ep->com.state) {
1661 case CLOSING:
b038ced7
SW
1662 __state_set(&ep->com, MORIBUND);
1663 break;
1664 case MORIBUND:
1665 stop_ep_timer(ep);
1666 if ((ep->com.cm_id) && (ep->com.qp)) {
1667 attrs.next_state = IWCH_QP_STATE_IDLE;
1668 iwch_modify_qp(ep->com.qp->rhp,
1669 ep->com.qp,
1670 IWCH_QP_ATTR_NEXT_STATE,
1671 &attrs, 1);
1672 }
1673 close_complete_upcall(ep);
1674 __state_set(&ep->com, DEAD);
1675 release = 1;
1676 break;
42e31753 1677 case ABORTING:
b038ced7 1678 case DEAD:
c4d49776 1679 break;
b038ced7
SW
1680 default:
1681 BUG_ON(1);
1682 break;
1683 }
1684 spin_unlock_irqrestore(&ep->com.lock, flags);
1685 if (release)
1686 release_ep_resources(ep);
1687 return CPL_RET_BUF_DONE;
1688}
1689
1690/*
1691 * T3A does 3 things when a TERM is received:
1692 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1693 * 2) generate an async event on the QP with the TERMINATE opcode
1694 * 3) post a TERMINATE opcde cqe into the associated CQ.
1695 *
1696 * For (1), we save the message in the qp for later consumer consumption.
1697 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1698 * For (3), we toss the CQE in cxio_poll_cq().
1699 *
1700 * terminate() handles case (1)...
1701 */
1702static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1703{
1704 struct iwch_ep *ep = ctx;
1705
42fb61f0
SW
1706 if (state_read(&ep->com) != FPDU_MODE)
1707 return CPL_RET_BUF_DONE;
1708
33718363 1709 PDBG("%s ep %p\n", __func__, ep);
b038ced7 1710 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
33718363 1711 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
d626f62b
ACM
1712 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1713 skb->len);
b038ced7
SW
1714 ep->com.qp->attr.terminate_msg_len = skb->len;
1715 ep->com.qp->attr.is_terminate_local = 0;
1716 return CPL_RET_BUF_DONE;
1717}
1718
1719static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1720{
1721 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1722 struct iwch_ep *ep = ctx;
1723
33718363 1724 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
b038ced7
SW
1725 rep->status);
1726 if (rep->status) {
1727 struct iwch_qp_attributes attrs;
1728
1729 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
33718363 1730 __func__, ep->hwtid);
2f236735 1731 stop_ep_timer(ep);
b038ced7
SW
1732 attrs.next_state = IWCH_QP_STATE_ERROR;
1733 iwch_modify_qp(ep->com.qp->rhp,
1734 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1735 &attrs, 1);
1736 abort_connection(ep, NULL, GFP_KERNEL);
1737 }
1738 return CPL_RET_BUF_DONE;
1739}
1740
1741static void ep_timeout(unsigned long arg)
1742{
1743 struct iwch_ep *ep = (struct iwch_ep *)arg;
1744 struct iwch_qp_attributes attrs;
1745 unsigned long flags;
989a1780 1746 int abort = 1;
b038ced7
SW
1747
1748 spin_lock_irqsave(&ep->com.lock, flags);
33718363 1749 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
b038ced7
SW
1750 ep->com.state);
1751 switch (ep->com.state) {
1752 case MPA_REQ_SENT:
989a1780 1753 __state_set(&ep->com, ABORTING);
b038ced7
SW
1754 connect_reply_upcall(ep, -ETIMEDOUT);
1755 break;
1756 case MPA_REQ_WAIT:
989a1780 1757 __state_set(&ep->com, ABORTING);
b038ced7 1758 break;
42e31753 1759 case CLOSING:
b038ced7
SW
1760 case MORIBUND:
1761 if (ep->com.cm_id && ep->com.qp) {
1762 attrs.next_state = IWCH_QP_STATE_ERROR;
1763 iwch_modify_qp(ep->com.qp->rhp,
1764 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1765 &attrs, 1);
1766 }
989a1780 1767 __state_set(&ep->com, ABORTING);
b038ced7
SW
1768 break;
1769 default:
989a1780
SW
1770 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1771 __func__, ep, ep->com.state);
1772 WARN_ON(1);
1773 abort = 0;
b038ced7 1774 }
b038ced7 1775 spin_unlock_irqrestore(&ep->com.lock, flags);
989a1780
SW
1776 if (abort)
1777 abort_connection(ep, NULL, GFP_ATOMIC);
b038ced7
SW
1778 put_ep(&ep->com);
1779}
1780
1781int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1782{
1783 int err;
1784 struct iwch_ep *ep = to_ep(cm_id);
33718363 1785 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
b038ced7
SW
1786
1787 if (state_read(&ep->com) == DEAD) {
1788 put_ep(&ep->com);
1789 return -ECONNRESET;
1790 }
1791 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
b038ced7
SW
1792 if (mpa_rev == 0)
1793 abort_connection(ep, NULL, GFP_KERNEL);
1794 else {
1795 err = send_mpa_reject(ep, pdata, pdata_len);
7d526e6b 1796 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
b038ced7 1797 }
6e47fe43 1798 put_ep(&ep->com);
b038ced7
SW
1799 return 0;
1800}
1801
1802int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1803{
1804 int err;
1805 struct iwch_qp_attributes attrs;
1806 enum iwch_qp_attr_mask mask;
1807 struct iwch_ep *ep = to_ep(cm_id);
1808 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1809 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1810
33718363 1811 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
6e47fe43
SW
1812 if (state_read(&ep->com) == DEAD) {
1813 err = -ECONNRESET;
1814 goto err;
1815 }
b038ced7
SW
1816
1817 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1818 BUG_ON(!qp);
1819
1820 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1821 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1822 abort_connection(ep, NULL, GFP_KERNEL);
6e47fe43
SW
1823 err = -EINVAL;
1824 goto err;
b038ced7
SW
1825 }
1826
1827 cm_id->add_ref(cm_id);
1828 ep->com.cm_id = cm_id;
1829 ep->com.qp = qp;
1830
1831 ep->com.rpl_done = 0;
1832 ep->com.rpl_err = 0;
1833 ep->ird = conn_param->ird;
1834 ep->ord = conn_param->ord;
96ac7e88
SW
1835
1836 if (peer2peer && ep->ird == 0)
1837 ep->ird = 1;
1838
33718363 1839 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
de3d3530 1840
b038ced7
SW
1841 /* bind QP to EP and move to RTS */
1842 attrs.mpa_attr = ep->mpa_attr;
1f71f503 1843 attrs.max_ird = ep->ird;
b038ced7
SW
1844 attrs.max_ord = ep->ord;
1845 attrs.llp_stream_handle = ep;
1846 attrs.next_state = IWCH_QP_STATE_RTS;
1847
1848 /* bind QP and TID with INIT_WR */
1849 mask = IWCH_QP_ATTR_NEXT_STATE |
1850 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1851 IWCH_QP_ATTR_MPA_ATTR |
1852 IWCH_QP_ATTR_MAX_IRD |
1853 IWCH_QP_ATTR_MAX_ORD;
1854
1855 err = iwch_modify_qp(ep->com.qp->rhp,
1856 ep->com.qp, mask, &attrs, 1);
de3d3530 1857 if (err)
6e47fe43 1858 goto err1;
b038ced7 1859
f8b0dfd1
SW
1860 /* if needed, wait for wr_ack */
1861 if (iwch_rqes_posted(qp)) {
1862 wait_event(ep->com.waitq, ep->com.rpl_done);
1863 err = ep->com.rpl_err;
1864 if (err)
6e47fe43 1865 goto err1;
f8b0dfd1
SW
1866 }
1867
de3d3530
SW
1868 err = send_mpa_reply(ep, conn_param->private_data,
1869 conn_param->private_data_len);
1870 if (err)
6e47fe43 1871 goto err1;
de3d3530 1872
de3d3530
SW
1873
1874 state_set(&ep->com, FPDU_MODE);
1875 established_upcall(ep);
1876 put_ep(&ep->com);
1877 return 0;
6e47fe43 1878err1:
de3d3530
SW
1879 ep->com.cm_id = NULL;
1880 ep->com.qp = NULL;
1881 cm_id->rem_ref(cm_id);
6e47fe43 1882err:
b038ced7
SW
1883 put_ep(&ep->com);
1884 return err;
1885}
1886
8704e9a8
SW
1887static int is_loopback_dst(struct iw_cm_id *cm_id)
1888{
1889 struct net_device *dev;
1890
1891 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1892 if (!dev)
1893 return 0;
1894 dev_put(dev);
1895 return 1;
1896}
1897
b038ced7
SW
1898int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1899{
1900 int err = 0;
1901 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1902 struct iwch_ep *ep;
1903 struct rtable *rt;
1904
8704e9a8
SW
1905 if (is_loopback_dst(cm_id)) {
1906 err = -ENOSYS;
1907 goto out;
1908 }
1909
b038ced7
SW
1910 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1911 if (!ep) {
33718363 1912 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
b038ced7
SW
1913 err = -ENOMEM;
1914 goto out;
1915 }
1916 init_timer(&ep->timer);
1917 ep->plen = conn_param->private_data_len;
1918 if (ep->plen)
1919 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1920 conn_param->private_data, ep->plen);
1921 ep->ird = conn_param->ird;
1922 ep->ord = conn_param->ord;
96ac7e88
SW
1923
1924 if (peer2peer && ep->ord == 0)
1925 ep->ord = 1;
1926
b038ced7
SW
1927 ep->com.tdev = h->rdev.t3cdev_p;
1928
1929 cm_id->add_ref(cm_id);
1930 ep->com.cm_id = cm_id;
1931 ep->com.qp = get_qhp(h, conn_param->qpn);
1932 BUG_ON(!ep->com.qp);
33718363 1933 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
b038ced7
SW
1934 ep->com.qp, cm_id);
1935
1936 /*
1937 * Allocate an active TID to initiate a TCP connection.
1938 */
1939 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1940 if (ep->atid == -1) {
33718363 1941 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
b038ced7
SW
1942 err = -ENOMEM;
1943 goto fail2;
1944 }
1945
1946 /* find a route */
1947 rt = find_route(h->rdev.t3cdev_p,
1948 cm_id->local_addr.sin_addr.s_addr,
1949 cm_id->remote_addr.sin_addr.s_addr,
1950 cm_id->local_addr.sin_port,
1951 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1952 if (!rt) {
33718363 1953 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
b038ced7
SW
1954 err = -EHOSTUNREACH;
1955 goto fail3;
1956 }
1957 ep->dst = &rt->u.dst;
1958
1959 /* get a l2t entry */
1960 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1961 ep->dst->neighbour->dev);
1962 if (!ep->l2t) {
33718363 1963 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
b038ced7
SW
1964 err = -ENOMEM;
1965 goto fail4;
1966 }
1967
1968 state_set(&ep->com, CONNECTING);
1969 ep->tos = IPTOS_LOWDELAY;
1970 ep->com.local_addr = cm_id->local_addr;
1971 ep->com.remote_addr = cm_id->remote_addr;
1972
1973 /* send connect request to rnic */
1974 err = send_connect(ep);
1975 if (!err)
1976 goto out;
1977
1978 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1979fail4:
1980 dst_release(ep->dst);
1981fail3:
1982 cxgb3_free_atid(ep->com.tdev, ep->atid);
1983fail2:
dc35fac9 1984 cm_id->rem_ref(cm_id);
b038ced7
SW
1985 put_ep(&ep->com);
1986out:
1987 return err;
1988}
1989
1990int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1991{
1992 int err = 0;
1993 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1994 struct iwch_listen_ep *ep;
1995
1996
1997 might_sleep();
1998
1999 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2000 if (!ep) {
33718363 2001 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
b038ced7
SW
2002 err = -ENOMEM;
2003 goto fail1;
2004 }
33718363 2005 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
2006 ep->com.tdev = h->rdev.t3cdev_p;
2007 cm_id->add_ref(cm_id);
2008 ep->com.cm_id = cm_id;
2009 ep->backlog = backlog;
2010 ep->com.local_addr = cm_id->local_addr;
2011
2012 /*
2013 * Allocate a server TID.
2014 */
2015 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2016 if (ep->stid == -1) {
33718363 2017 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
b038ced7
SW
2018 err = -ENOMEM;
2019 goto fail2;
2020 }
2021
2022 state_set(&ep->com, LISTEN);
2023 err = listen_start(ep);
2024 if (err)
2025 goto fail3;
2026
2027 /* wait for pass_open_rpl */
2028 wait_event(ep->com.waitq, ep->com.rpl_done);
2029 err = ep->com.rpl_err;
2030 if (!err) {
2031 cm_id->provider_data = ep;
2032 goto out;
2033 }
2034fail3:
2035 cxgb3_free_stid(ep->com.tdev, ep->stid);
2036fail2:
1b07db70 2037 cm_id->rem_ref(cm_id);
b038ced7
SW
2038 put_ep(&ep->com);
2039fail1:
2040out:
2041 return err;
2042}
2043
2044int iwch_destroy_listen(struct iw_cm_id *cm_id)
2045{
2046 int err;
2047 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2048
33718363 2049 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
2050
2051 might_sleep();
2052 state_set(&ep->com, DEAD);
2053 ep->com.rpl_done = 0;
2054 ep->com.rpl_err = 0;
2055 err = listen_stop(ep);
04b5d028
SW
2056 if (err)
2057 goto done;
b038ced7
SW
2058 wait_event(ep->com.waitq, ep->com.rpl_done);
2059 cxgb3_free_stid(ep->com.tdev, ep->stid);
04b5d028 2060done:
b038ced7
SW
2061 err = ep->com.rpl_err;
2062 cm_id->rem_ref(cm_id);
2063 put_ep(&ep->com);
2064 return err;
2065}
2066
2067int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2068{
2069 int ret=0;
2070 unsigned long flags;
2071 int close = 0;
04b5d028
SW
2072 int fatal = 0;
2073 struct t3cdev *tdev;
2074 struct cxio_rdev *rdev;
b038ced7
SW
2075
2076 spin_lock_irqsave(&ep->com.lock, flags);
2077
33718363 2078 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
b038ced7
SW
2079 states[ep->com.state], abrupt);
2080
04b5d028
SW
2081 tdev = (struct t3cdev *)ep->com.tdev;
2082 rdev = (struct cxio_rdev *)tdev->ulp;
2083 if (cxio_fatal_error(rdev)) {
2084 fatal = 1;
2085 close_complete_upcall(ep);
2086 ep->com.state = DEAD;
2087 }
b038ced7
SW
2088 switch (ep->com.state) {
2089 case MPA_REQ_WAIT:
2090 case MPA_REQ_SENT:
2091 case MPA_REQ_RCVD:
2092 case MPA_REP_SENT:
2093 case FPDU_MODE:
b038ced7 2094 close = 1;
989a1780
SW
2095 if (abrupt)
2096 ep->com.state = ABORTING;
2097 else {
2098 ep->com.state = CLOSING;
2099 start_ep_timer(ep);
2100 }
6e47fe43 2101 set_bit(CLOSE_SENT, &ep->com.flags);
b038ced7
SW
2102 break;
2103 case CLOSING:
6e47fe43
SW
2104 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2105 close = 1;
2106 if (abrupt) {
2107 stop_ep_timer(ep);
2108 ep->com.state = ABORTING;
2109 } else
2110 ep->com.state = MORIBUND;
2111 }
b038ced7
SW
2112 break;
2113 case MORIBUND:
989a1780
SW
2114 case ABORTING:
2115 case DEAD:
2116 PDBG("%s ignoring disconnect ep %p state %u\n",
2117 __func__, ep, ep->com.state);
b038ced7
SW
2118 break;
2119 default:
2120 BUG();
2121 break;
2122 }
989a1780 2123
b038ced7
SW
2124 spin_unlock_irqrestore(&ep->com.lock, flags);
2125 if (close) {
2126 if (abrupt)
2127 ret = send_abort(ep, NULL, gfp);
2128 else
2129 ret = send_halfclose(ep, gfp);
04b5d028
SW
2130 if (ret)
2131 fatal = 1;
b038ced7 2132 }
04b5d028
SW
2133 if (fatal)
2134 release_ep_resources(ep);
b038ced7
SW
2135 return ret;
2136}
2137
2138int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2139 struct l2t_entry *l2t)
2140{
2141 struct iwch_ep *ep = ctx;
2142
2143 if (ep->dst != old)
2144 return 0;
2145
33718363 2146 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
b038ced7
SW
2147 l2t);
2148 dst_hold(new);
2149 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2150 ep->l2t = l2t;
2151 dst_release(old);
2152 ep->dst = new;
2153 return 1;
2154}
2155
2156/*
2157 * All the CM events are handled on a work queue to have a safe context.
2158 */
2159static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2160{
2161 struct iwch_ep_common *epc = ctx;
2162
2163 get_ep(epc);
2164
2165 /*
2166 * Save ctx and tdev in the skb->cb area.
2167 */
2168 *((void **) skb->cb) = ctx;
2169 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2170
2171 /*
2172 * Queue the skb and schedule the worker thread.
2173 */
2174 skb_queue_tail(&rxq, skb);
2175 queue_work(workq, &skb_work);
2176 return 0;
2177}
2178
1ca19770
SW
2179static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2180{
2181 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2182
2183 if (rpl->status != CPL_ERR_NONE) {
2184 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2185 "for tid %u\n", rpl->status, GET_TID(rpl));
2186 }
2187 return CPL_RET_BUF_DONE;
2188}
2189
b038ced7
SW
2190int __init iwch_cm_init(void)
2191{
2192 skb_queue_head_init(&rxq);
2193
2194 workq = create_singlethread_workqueue("iw_cxgb3");
2195 if (!workq)
2196 return -ENOMEM;
2197
2198 /*
2199 * All upcalls from the T3 Core go to sched() to
2200 * schedule the processing on a work queue.
2201 */
2202 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2203 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2204 t3c_handlers[CPL_RX_DATA] = sched;
2205 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2206 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2207 t3c_handlers[CPL_ABORT_RPL] = sched;
2208 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2209 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2210 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2211 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2212 t3c_handlers[CPL_PEER_CLOSE] = sched;
2213 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2214 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2215 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2216 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
1ca19770 2217 t3c_handlers[CPL_SET_TCB_RPL] = set_tcb_rpl;
b038ced7
SW
2218
2219 /*
2220 * These are the real handlers that are called from a
2221 * work queue.
2222 */
2223 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2224 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2225 work_handlers[CPL_RX_DATA] = rx_data;
2226 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2227 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2228 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2229 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2230 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2231 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2232 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2233 work_handlers[CPL_PEER_CLOSE] = peer_close;
2234 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2235 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2236 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2237 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2238 return 0;
2239}
2240
2241void __exit iwch_cm_term(void)
2242{
2243 flush_workqueue(workq);
2244 destroy_workqueue(workq);
2245}