ipv4: Create and use route lookup helpers.
[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>
5a0e3ad6 34#include <linux/slab.h>
b038ced7
SW
35#include <linux/workqueue.h>
36#include <linux/skbuff.h>
37#include <linux/timer.h>
38#include <linux/notifier.h>
8704e9a8 39#include <linux/inetdevice.h>
b038ced7
SW
40
41#include <net/neighbour.h>
42#include <net/netevent.h>
43#include <net/route.h>
44
45#include "tcb.h"
46#include "cxgb3_offload.h"
47#include "iwch.h"
48#include "iwch_provider.h"
49#include "iwch_cm.h"
50
51static char *states[] = {
52 "idle",
53 "listen",
54 "connecting",
55 "mpa_wait_req",
56 "mpa_req_sent",
57 "mpa_req_rcvd",
58 "mpa_rep_sent",
59 "fpdu_mode",
60 "aborting",
61 "closing",
62 "moribund",
63 "dead",
64 NULL,
65};
66
f8b0dfd1
SW
67int peer2peer = 0;
68module_param(peer2peer, int, 0644);
69MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
70
77a8d574 71static int ep_timeout_secs = 60;
e54664c0 72module_param(ep_timeout_secs, int, 0644);
b038ced7 73MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
77a8d574 74 "in seconds (default=60)");
b038ced7
SW
75
76static int mpa_rev = 1;
e54664c0 77module_param(mpa_rev, int, 0644);
b038ced7
SW
78MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
79 "1 is spec compliant. (default=1)");
80
81static int markers_enabled = 0;
e54664c0 82module_param(markers_enabled, int, 0644);
b038ced7
SW
83MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
84
85static int crc_enabled = 1;
e54664c0 86module_param(crc_enabled, int, 0644);
b038ced7
SW
87MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
88
89static int rcv_win = 256 * 1024;
e54664c0 90module_param(rcv_win, int, 0644);
b038ced7
SW
91MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
92
93static int snd_win = 32 * 1024;
e54664c0 94module_param(snd_win, int, 0644);
b038ced7
SW
95MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
96
97static unsigned int nocong = 0;
e54664c0 98module_param(nocong, uint, 0644);
b038ced7
SW
99MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
100
101static unsigned int cong_flavor = 1;
e54664c0 102module_param(cong_flavor, uint, 0644);
b038ced7
SW
103MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
104
b038ced7 105static struct workqueue_struct *workq;
b038ced7
SW
106
107static struct sk_buff_head rxq;
b038ced7
SW
108
109static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
110static void ep_timeout(unsigned long arg);
111static void connect_reply_upcall(struct iwch_ep *ep, int status);
112
113static void start_ep_timer(struct iwch_ep *ep)
114{
33718363 115 PDBG("%s ep %p\n", __func__, ep);
b038ced7 116 if (timer_pending(&ep->timer)) {
33718363 117 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
b038ced7
SW
118 del_timer_sync(&ep->timer);
119 } else
120 get_ep(&ep->com);
121 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
122 ep->timer.data = (unsigned long)ep;
123 ep->timer.function = ep_timeout;
124 add_timer(&ep->timer);
125}
126
127static void stop_ep_timer(struct iwch_ep *ep)
128{
33718363 129 PDBG("%s ep %p\n", __func__, ep);
989a1780
SW
130 if (!timer_pending(&ep->timer)) {
131 printk(KERN_ERR "%s timer stopped when its not running! ep %p state %u\n",
132 __func__, ep, ep->com.state);
133 WARN_ON(1);
134 return;
135 }
b038ced7
SW
136 del_timer_sync(&ep->timer);
137 put_ep(&ep->com);
138}
139
18199f57 140static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
04b5d028
SW
141{
142 int error = 0;
143 struct cxio_rdev *rdev;
144
145 rdev = (struct cxio_rdev *)tdev->ulp;
146 if (cxio_fatal_error(rdev)) {
147 kfree_skb(skb);
148 return -EIO;
149 }
150 error = l2t_send(tdev, skb, l2e);
73a203d2 151 if (error < 0)
04b5d028
SW
152 kfree_skb(skb);
153 return error;
154}
155
156int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
157{
158 int error = 0;
159 struct cxio_rdev *rdev;
160
161 rdev = (struct cxio_rdev *)tdev->ulp;
162 if (cxio_fatal_error(rdev)) {
163 kfree_skb(skb);
164 return -EIO;
165 }
166 error = cxgb3_ofld_send(tdev, skb);
73a203d2 167 if (error < 0)
04b5d028
SW
168 kfree_skb(skb);
169 return error;
170}
171
b038ced7
SW
172static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
173{
174 struct cpl_tid_release *req;
175
176 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
177 if (!skb)
178 return;
179 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
180 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
181 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
182 skb->priority = CPL_PRIORITY_SETUP;
04b5d028 183 iwch_cxgb3_ofld_send(tdev, skb);
b038ced7
SW
184 return;
185}
186
187int iwch_quiesce_tid(struct iwch_ep *ep)
188{
189 struct cpl_set_tcb_field *req;
190 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
191
192 if (!skb)
193 return -ENOMEM;
194 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
195 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
196 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
197 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
198 req->reply = 0;
199 req->cpu_idx = 0;
200 req->word = htons(W_TCB_RX_QUIESCE);
201 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
202 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
203
204 skb->priority = CPL_PRIORITY_DATA;
04b5d028 205 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
206}
207
208int iwch_resume_tid(struct iwch_ep *ep)
209{
210 struct cpl_set_tcb_field *req;
211 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
212
213 if (!skb)
214 return -ENOMEM;
215 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
216 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
217 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
218 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
219 req->reply = 0;
220 req->cpu_idx = 0;
221 req->word = htons(W_TCB_RX_QUIESCE);
222 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
223 req->val = 0;
224
225 skb->priority = CPL_PRIORITY_DATA;
04b5d028 226 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
227}
228
229static void set_emss(struct iwch_ep *ep, u16 opt)
230{
33718363 231 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
b038ced7
SW
232 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
233 if (G_TCPOPT_TSTAMP(opt))
234 ep->emss -= 12;
235 if (ep->emss < 128)
236 ep->emss = 128;
237 PDBG("emss=%d\n", ep->emss);
238}
239
240static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
241{
242 unsigned long flags;
243 enum iwch_ep_state state;
244
245 spin_lock_irqsave(&epc->lock, flags);
246 state = epc->state;
247 spin_unlock_irqrestore(&epc->lock, flags);
248 return state;
249}
250
2b540355 251static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
b038ced7
SW
252{
253 epc->state = new;
254}
255
256static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
257{
258 unsigned long flags;
259
260 spin_lock_irqsave(&epc->lock, flags);
33718363 261 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
b038ced7
SW
262 __state_set(epc, new);
263 spin_unlock_irqrestore(&epc->lock, flags);
264 return;
265}
266
267static void *alloc_ep(int size, gfp_t gfp)
268{
269 struct iwch_ep_common *epc;
270
dd00cc48 271 epc = kzalloc(size, gfp);
b038ced7 272 if (epc) {
b038ced7
SW
273 kref_init(&epc->kref);
274 spin_lock_init(&epc->lock);
275 init_waitqueue_head(&epc->waitq);
276 }
33718363 277 PDBG("%s alloc ep %p\n", __func__, epc);
b038ced7
SW
278 return epc;
279}
280
281void __free_ep(struct kref *kref)
282{
874d8df5
SW
283 struct iwch_ep *ep;
284 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
285 struct iwch_ep, com);
286 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
6e47fe43 287 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
874d8df5
SW
288 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
289 dst_release(ep->dst);
290 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
291 }
292 kfree(ep);
b038ced7
SW
293}
294
295static void release_ep_resources(struct iwch_ep *ep)
296{
33718363 297 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
6e47fe43 298 set_bit(RELEASE_RESOURCES, &ep->com.flags);
b038ced7
SW
299 put_ep(&ep->com);
300}
301
b038ced7
SW
302static int status2errno(int status)
303{
304 switch (status) {
305 case CPL_ERR_NONE:
306 return 0;
307 case CPL_ERR_CONN_RESET:
308 return -ECONNRESET;
309 case CPL_ERR_ARP_MISS:
310 return -EHOSTUNREACH;
311 case CPL_ERR_CONN_TIMEDOUT:
312 return -ETIMEDOUT;
313 case CPL_ERR_TCAM_FULL:
314 return -ENOMEM;
315 case CPL_ERR_CONN_EXIST:
316 return -EADDRINUSE;
317 default:
318 return -EIO;
319 }
320}
321
322/*
323 * Try and reuse skbs already allocated...
324 */
325static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
326{
1f6a849b 327 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
b038ced7
SW
328 skb_trim(skb, 0);
329 skb_get(skb);
330 } else {
331 skb = alloc_skb(len, gfp);
332 }
333 return skb;
334}
335
336static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
337 __be32 peer_ip, __be16 local_port,
338 __be16 peer_port, u8 tos)
339{
340 struct rtable *rt;
78fbfd8a
DM
341
342 rt = ip_route_output_ports(&init_net, NULL, peer_ip, local_ip,
343 peer_port, local_port, IPPROTO_TCP,
344 tos, 0);
b23dd4fe 345 if (IS_ERR(rt))
b038ced7
SW
346 return NULL;
347 return rt;
348}
349
350static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
351{
352 int i = 0;
353
354 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
355 ++i;
356 return i;
357}
358
359static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
360{
33718363 361 PDBG("%s t3cdev %p\n", __func__, dev);
b038ced7
SW
362 kfree_skb(skb);
363}
364
365/*
366 * Handle an ARP failure for an active open.
367 */
368static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
369{
370 printk(KERN_ERR MOD "ARP failure duing connect\n");
371 kfree_skb(skb);
372}
373
374/*
375 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
376 * and send it along.
377 */
378static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
379{
380 struct cpl_abort_req *req = cplhdr(skb);
381
33718363 382 PDBG("%s t3cdev %p\n", __func__, dev);
b038ced7 383 req->cmd = CPL_ABORT_NO_RST;
04b5d028 384 iwch_cxgb3_ofld_send(dev, skb);
b038ced7
SW
385}
386
387static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
388{
389 struct cpl_close_con_req *req;
390 struct sk_buff *skb;
391
33718363 392 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
393 skb = get_skb(NULL, sizeof(*req), gfp);
394 if (!skb) {
33718363 395 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
b038ced7
SW
396 return -ENOMEM;
397 }
398 skb->priority = CPL_PRIORITY_DATA;
399 set_arp_failure_handler(skb, arp_failure_discard);
400 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
401 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
402 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
403 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
04b5d028 404 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
405}
406
407static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
408{
409 struct cpl_abort_req *req;
410
33718363 411 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
412 skb = get_skb(skb, sizeof(*req), gfp);
413 if (!skb) {
414 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
33718363 415 __func__);
b038ced7
SW
416 return -ENOMEM;
417 }
418 skb->priority = CPL_PRIORITY_DATA;
419 set_arp_failure_handler(skb, abort_arp_failure);
420 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
421 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
422 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
423 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
424 req->cmd = CPL_ABORT_SEND_RST;
04b5d028 425 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
426}
427
428static int send_connect(struct iwch_ep *ep)
429{
430 struct cpl_act_open_req *req;
431 struct sk_buff *skb;
432 u32 opt0h, opt0l, opt2;
433 unsigned int mtu_idx;
434 int wscale;
435
33718363 436 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
437
438 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
439 if (!skb) {
440 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
33718363 441 __func__);
b038ced7
SW
442 return -ENOMEM;
443 }
444 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
445 wscale = compute_wscale(rcv_win);
446 opt0h = V_NAGLE(0) |
447 V_NO_CONG(nocong) |
448 V_KEEP_ALIVE(1) |
449 F_TCAM_BYPASS |
450 V_WND_SCALE(wscale) |
451 V_MSS_IDX(mtu_idx) |
452 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
453 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
bec658ff
SW
454 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
455 V_CONG_CONTROL_FLAVOR(cong_flavor);
b038ced7
SW
456 skb->priority = CPL_PRIORITY_SETUP;
457 set_arp_failure_handler(skb, act_open_req_arp_failure);
458
459 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
460 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
461 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
462 req->local_port = ep->com.local_addr.sin_port;
463 req->peer_port = ep->com.remote_addr.sin_port;
464 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
465 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
466 req->opt0h = htonl(opt0h);
467 req->opt0l = htonl(opt0l);
468 req->params = 0;
469 req->opt2 = htonl(opt2);
04b5d028 470 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
471}
472
473static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
474{
475 int mpalen;
476 struct tx_data_wr *req;
477 struct mpa_message *mpa;
478 int len;
479
33718363 480 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
b038ced7
SW
481
482 BUG_ON(skb_cloned(skb));
483
484 mpalen = sizeof(*mpa) + ep->plen;
4305b541 485 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
b038ced7
SW
486 kfree_skb(skb);
487 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
488 if (!skb) {
489 connect_reply_upcall(ep, -ENOMEM);
490 return;
491 }
492 }
493 skb_trim(skb, 0);
494 skb_reserve(skb, sizeof(*req));
495 skb_put(skb, mpalen);
496 skb->priority = CPL_PRIORITY_DATA;
497 mpa = (struct mpa_message *) skb->data;
498 memset(mpa, 0, sizeof(*mpa));
499 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
500 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
501 (markers_enabled ? MPA_MARKERS : 0);
502 mpa->private_data_size = htons(ep->plen);
503 mpa->revision = mpa_rev;
504
505 if (ep->plen)
506 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
507
508 /*
509 * Reference the mpa skb. This ensures the data area
510 * will remain in memory until the hw acks the tx.
511 * Function tx_ack() will deref it.
512 */
513 skb_get(skb);
514 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 515 skb_reset_transport_header(skb);
b038ced7
SW
516 len = skb->len;
517 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 518 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
519 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
520 req->len = htonl(len);
521 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
522 V_TX_SNDBUF(snd_win>>15));
de3d3530 523 req->flags = htonl(F_TX_INIT);
b038ced7
SW
524 req->sndseq = htonl(ep->snd_seq);
525 BUG_ON(ep->mpa_skb);
526 ep->mpa_skb = skb;
04b5d028 527 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
528 start_ep_timer(ep);
529 state_set(&ep->com, MPA_REQ_SENT);
530 return;
531}
532
533static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
534{
535 int mpalen;
536 struct tx_data_wr *req;
537 struct mpa_message *mpa;
538 struct sk_buff *skb;
539
33718363 540 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
b038ced7
SW
541
542 mpalen = sizeof(*mpa) + plen;
543
544 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
545 if (!skb) {
33718363 546 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
b038ced7
SW
547 return -ENOMEM;
548 }
549 skb_reserve(skb, sizeof(*req));
550 mpa = (struct mpa_message *) skb_put(skb, mpalen);
551 memset(mpa, 0, sizeof(*mpa));
552 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
553 mpa->flags = MPA_REJECT;
554 mpa->revision = mpa_rev;
555 mpa->private_data_size = htons(plen);
556 if (plen)
557 memcpy(mpa->private_data, pdata, plen);
558
559 /*
560 * Reference the mpa skb again. This ensures the data area
561 * will remain in memory until the hw acks the tx.
562 * Function tx_ack() will deref it.
563 */
564 skb_get(skb);
565 skb->priority = CPL_PRIORITY_DATA;
566 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 567 skb_reset_transport_header(skb);
b038ced7 568 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 569 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
570 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
571 req->len = htonl(mpalen);
572 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
573 V_TX_SNDBUF(snd_win>>15));
de3d3530 574 req->flags = htonl(F_TX_INIT);
b038ced7
SW
575 req->sndseq = htonl(ep->snd_seq);
576 BUG_ON(ep->mpa_skb);
577 ep->mpa_skb = skb;
04b5d028 578 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
579}
580
581static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
582{
583 int mpalen;
584 struct tx_data_wr *req;
585 struct mpa_message *mpa;
586 int len;
587 struct sk_buff *skb;
588
33718363 589 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
b038ced7
SW
590
591 mpalen = sizeof(*mpa) + plen;
592
593 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
594 if (!skb) {
33718363 595 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
b038ced7
SW
596 return -ENOMEM;
597 }
598 skb->priority = CPL_PRIORITY_DATA;
599 skb_reserve(skb, sizeof(*req));
600 mpa = (struct mpa_message *) skb_put(skb, mpalen);
601 memset(mpa, 0, sizeof(*mpa));
602 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
603 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
604 (markers_enabled ? MPA_MARKERS : 0);
605 mpa->revision = mpa_rev;
606 mpa->private_data_size = htons(plen);
607 if (plen)
608 memcpy(mpa->private_data, pdata, plen);
609
610 /*
611 * Reference the mpa skb. This ensures the data area
612 * will remain in memory until the hw acks the tx.
613 * Function tx_ack() will deref it.
614 */
615 skb_get(skb);
616 set_arp_failure_handler(skb, arp_failure_discard);
badff6d0 617 skb_reset_transport_header(skb);
b038ced7
SW
618 len = skb->len;
619 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
f8b0dfd1 620 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
b038ced7
SW
621 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
622 req->len = htonl(len);
623 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
624 V_TX_SNDBUF(snd_win>>15));
de3d3530 625 req->flags = htonl(F_TX_INIT);
b038ced7
SW
626 req->sndseq = htonl(ep->snd_seq);
627 ep->mpa_skb = skb;
628 state_set(&ep->com, MPA_REP_SENT);
04b5d028 629 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
630}
631
632static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
633{
634 struct iwch_ep *ep = ctx;
635 struct cpl_act_establish *req = cplhdr(skb);
636 unsigned int tid = GET_TID(req);
637
33718363 638 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
b038ced7
SW
639
640 dst_confirm(ep->dst);
641
642 /* setup the hwtid for this connection */
643 ep->hwtid = tid;
644 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
645
646 ep->snd_seq = ntohl(req->snd_isn);
de3d3530 647 ep->rcv_seq = ntohl(req->rcv_isn);
b038ced7
SW
648
649 set_emss(ep, ntohs(req->tcp_opt));
650
651 /* dealloc the atid */
652 cxgb3_free_atid(ep->com.tdev, ep->atid);
653
654 /* start MPA negotiation */
655 send_mpa_req(ep, skb);
656
657 return 0;
658}
659
660static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
661{
662 PDBG("%s ep %p\n", __FILE__, ep);
663 state_set(&ep->com, ABORTING);
664 send_abort(ep, skb, gfp);
665}
666
667static void close_complete_upcall(struct iwch_ep *ep)
668{
669 struct iw_cm_event event;
670
33718363 671 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
672 memset(&event, 0, sizeof(event));
673 event.event = IW_CM_EVENT_CLOSE;
674 if (ep->com.cm_id) {
675 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
676 ep, ep->com.cm_id, ep->hwtid);
677 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
678 ep->com.cm_id->rem_ref(ep->com.cm_id);
679 ep->com.cm_id = NULL;
680 ep->com.qp = NULL;
681 }
682}
683
684static void peer_close_upcall(struct iwch_ep *ep)
685{
686 struct iw_cm_event event;
687
33718363 688 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
689 memset(&event, 0, sizeof(event));
690 event.event = IW_CM_EVENT_DISCONNECT;
691 if (ep->com.cm_id) {
692 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
693 ep, ep->com.cm_id, ep->hwtid);
694 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
695 }
696}
697
698static void peer_abort_upcall(struct iwch_ep *ep)
699{
700 struct iw_cm_event event;
701
33718363 702 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
703 memset(&event, 0, sizeof(event));
704 event.event = IW_CM_EVENT_CLOSE;
705 event.status = -ECONNRESET;
706 if (ep->com.cm_id) {
707 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
708 ep->com.cm_id, ep->hwtid);
709 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
710 ep->com.cm_id->rem_ref(ep->com.cm_id);
711 ep->com.cm_id = NULL;
712 ep->com.qp = NULL;
713 }
714}
715
716static void connect_reply_upcall(struct iwch_ep *ep, int status)
717{
718 struct iw_cm_event event;
719
33718363 720 PDBG("%s ep %p status %d\n", __func__, ep, status);
b038ced7
SW
721 memset(&event, 0, sizeof(event));
722 event.event = IW_CM_EVENT_CONNECT_REPLY;
723 event.status = status;
724 event.local_addr = ep->com.local_addr;
725 event.remote_addr = ep->com.remote_addr;
726
727 if ((status == 0) || (status == -ECONNREFUSED)) {
728 event.private_data_len = ep->plen;
729 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
730 }
731 if (ep->com.cm_id) {
33718363 732 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
b038ced7
SW
733 ep->hwtid, status);
734 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
735 }
736 if (status < 0) {
737 ep->com.cm_id->rem_ref(ep->com.cm_id);
738 ep->com.cm_id = NULL;
739 ep->com.qp = NULL;
740 }
741}
742
743static void connect_request_upcall(struct iwch_ep *ep)
744{
745 struct iw_cm_event event;
746
33718363 747 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
b038ced7
SW
748 memset(&event, 0, sizeof(event));
749 event.event = IW_CM_EVENT_CONNECT_REQUEST;
750 event.local_addr = ep->com.local_addr;
751 event.remote_addr = ep->com.remote_addr;
752 event.private_data_len = ep->plen;
753 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
754 event.provider_data = ep;
6e47fe43
SW
755 if (state_read(&ep->parent_ep->com) != DEAD) {
756 get_ep(&ep->com);
b038ced7
SW
757 ep->parent_ep->com.cm_id->event_handler(
758 ep->parent_ep->com.cm_id,
759 &event);
6e47fe43 760 }
b038ced7
SW
761 put_ep(&ep->parent_ep->com);
762 ep->parent_ep = NULL;
763}
764
765static void established_upcall(struct iwch_ep *ep)
766{
767 struct iw_cm_event event;
768
33718363 769 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
770 memset(&event, 0, sizeof(event));
771 event.event = IW_CM_EVENT_ESTABLISHED;
772 if (ep->com.cm_id) {
33718363 773 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
b038ced7
SW
774 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
775 }
776}
777
778static int update_rx_credits(struct iwch_ep *ep, u32 credits)
779{
780 struct cpl_rx_data_ack *req;
781 struct sk_buff *skb;
782
33718363 783 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
b038ced7
SW
784 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
785 if (!skb) {
786 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
787 return 0;
788 }
789
790 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
791 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
792 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
793 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
794 skb->priority = CPL_PRIORITY_ACK;
04b5d028 795 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
796 return credits;
797}
798
799static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
800{
801 struct mpa_message *mpa;
802 u16 plen;
803 struct iwch_qp_attributes attrs;
804 enum iwch_qp_attr_mask mask;
805 int err;
806
33718363 807 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
808
809 /*
810 * Stop mpa timer. If it expired, then the state has
811 * changed and we bail since ep_timeout already aborted
812 * the connection.
813 */
814 stop_ep_timer(ep);
815 if (state_read(&ep->com) != MPA_REQ_SENT)
816 return;
817
818 /*
819 * If we get more than the supported amount of private data
820 * then we must fail this connection.
821 */
822 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
823 err = -EINVAL;
824 goto err;
825 }
826
827 /*
828 * copy the new data into our accumulation buffer.
829 */
d626f62b
ACM
830 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
831 skb->len);
b038ced7
SW
832 ep->mpa_pkt_len += skb->len;
833
834 /*
835 * if we don't even have the mpa message, then bail.
836 */
837 if (ep->mpa_pkt_len < sizeof(*mpa))
838 return;
839 mpa = (struct mpa_message *) ep->mpa_pkt;
840
841 /* Validate MPA header. */
842 if (mpa->revision != mpa_rev) {
843 err = -EPROTO;
844 goto err;
845 }
846 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
847 err = -EPROTO;
848 goto err;
849 }
850
851 plen = ntohs(mpa->private_data_size);
852
853 /*
854 * Fail if there's too much private data.
855 */
856 if (plen > MPA_MAX_PRIVATE_DATA) {
857 err = -EPROTO;
858 goto err;
859 }
860
861 /*
862 * If plen does not account for pkt size
863 */
864 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
865 err = -EPROTO;
866 goto err;
867 }
868
869 ep->plen = (u8) plen;
870
871 /*
872 * If we don't have all the pdata yet, then bail.
873 * We'll continue process when more data arrives.
874 */
875 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
876 return;
877
878 if (mpa->flags & MPA_REJECT) {
879 err = -ECONNREFUSED;
880 goto err;
881 }
882
883 /*
884 * If we get here we have accumulated the entire mpa
885 * start reply message including private data. And
886 * the MPA header is valid.
887 */
888 state_set(&ep->com, FPDU_MODE);
f8b0dfd1 889 ep->mpa_attr.initiator = 1;
b038ced7
SW
890 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
891 ep->mpa_attr.recv_marker_enabled = markers_enabled;
892 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
893 ep->mpa_attr.version = mpa_rev;
894 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
33718363 895 "xmit_marker_enabled=%d, version=%d\n", __func__,
b038ced7
SW
896 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
897 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
898
899 attrs.mpa_attr = ep->mpa_attr;
900 attrs.max_ird = ep->ird;
901 attrs.max_ord = ep->ord;
902 attrs.llp_stream_handle = ep;
903 attrs.next_state = IWCH_QP_STATE_RTS;
904
905 mask = IWCH_QP_ATTR_NEXT_STATE |
906 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
907 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
908
909 /* bind QP and TID with INIT_WR */
910 err = iwch_modify_qp(ep->com.qp->rhp,
911 ep->com.qp, mask, &attrs, 1);
f8b0dfd1
SW
912 if (err)
913 goto err;
914
915 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
916 iwch_post_zb_read(ep->com.qp);
917 }
918
919 goto out;
b038ced7
SW
920err:
921 abort_connection(ep, skb, GFP_KERNEL);
922out:
923 connect_reply_upcall(ep, err);
924 return;
925}
926
927static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
928{
929 struct mpa_message *mpa;
930 u16 plen;
931
33718363 932 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
933
934 /*
935 * Stop mpa timer. If it expired, then the state has
936 * changed and we bail since ep_timeout already aborted
937 * the connection.
938 */
939 stop_ep_timer(ep);
940 if (state_read(&ep->com) != MPA_REQ_WAIT)
941 return;
942
943 /*
944 * If we get more than the supported amount of private data
945 * then we must fail this connection.
946 */
947 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
948 abort_connection(ep, skb, GFP_KERNEL);
949 return;
950 }
951
33718363 952 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
b038ced7
SW
953
954 /*
955 * Copy the new data into our accumulation buffer.
956 */
d626f62b
ACM
957 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
958 skb->len);
b038ced7
SW
959 ep->mpa_pkt_len += skb->len;
960
961 /*
962 * If we don't even have the mpa message, then bail.
963 * We'll continue process when more data arrives.
964 */
965 if (ep->mpa_pkt_len < sizeof(*mpa))
966 return;
33718363 967 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
b038ced7
SW
968 mpa = (struct mpa_message *) ep->mpa_pkt;
969
970 /*
971 * Validate MPA Header.
972 */
973 if (mpa->revision != mpa_rev) {
974 abort_connection(ep, skb, GFP_KERNEL);
975 return;
976 }
977
978 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
979 abort_connection(ep, skb, GFP_KERNEL);
980 return;
981 }
982
983 plen = ntohs(mpa->private_data_size);
984
985 /*
986 * Fail if there's too much private data.
987 */
988 if (plen > MPA_MAX_PRIVATE_DATA) {
989 abort_connection(ep, skb, GFP_KERNEL);
990 return;
991 }
992
993 /*
994 * If plen does not account for pkt size
995 */
996 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
997 abort_connection(ep, skb, GFP_KERNEL);
998 return;
999 }
1000 ep->plen = (u8) plen;
1001
1002 /*
1003 * If we don't have all the pdata yet, then bail.
1004 */
1005 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1006 return;
1007
1008 /*
1009 * If we get here we have accumulated the entire mpa
1010 * start reply message including private data.
1011 */
f8b0dfd1 1012 ep->mpa_attr.initiator = 0;
b038ced7
SW
1013 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1014 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1015 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1016 ep->mpa_attr.version = mpa_rev;
1017 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
33718363 1018 "xmit_marker_enabled=%d, version=%d\n", __func__,
b038ced7
SW
1019 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1020 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1021
1022 state_set(&ep->com, MPA_REQ_RCVD);
1023
1024 /* drive upcall */
1025 connect_request_upcall(ep);
1026 return;
1027}
1028
1029static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1030{
1031 struct iwch_ep *ep = ctx;
1032 struct cpl_rx_data *hdr = cplhdr(skb);
1033 unsigned int dlen = ntohs(hdr->len);
1034
33718363 1035 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
b038ced7
SW
1036
1037 skb_pull(skb, sizeof(*hdr));
1038 skb_trim(skb, dlen);
1039
de3d3530
SW
1040 ep->rcv_seq += dlen;
1041 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1042
b038ced7
SW
1043 switch (state_read(&ep->com)) {
1044 case MPA_REQ_SENT:
1045 process_mpa_reply(ep, skb);
1046 break;
1047 case MPA_REQ_WAIT:
1048 process_mpa_request(ep, skb);
1049 break;
1050 case MPA_REP_SENT:
1051 break;
1052 default:
1053 printk(KERN_ERR MOD "%s Unexpected streaming data."
1054 " ep %p state %d tid %d\n",
33718363 1055 __func__, ep, state_read(&ep->com), ep->hwtid);
b038ced7
SW
1056
1057 /*
1058 * The ep will timeout and inform the ULP of the failure.
1059 * See ep_timeout().
1060 */
1061 break;
1062 }
1063
1064 /* update RX credits */
1065 update_rx_credits(ep, dlen);
1066
1067 return CPL_RET_BUF_DONE;
1068}
1069
1070/*
1071 * Upcall from the adapter indicating data has been transmitted.
1072 * For us its just the single MPA request or reply. We can now free
1073 * the skb holding the mpa message.
1074 */
1075static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1076{
1077 struct iwch_ep *ep = ctx;
1078 struct cpl_wr_ack *hdr = cplhdr(skb);
1079 unsigned int credits = ntohs(hdr->credits);
b038ced7 1080
33718363 1081 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
b038ced7 1082
f8b0dfd1 1083 if (credits == 0) {
ca7cf94f
JP
1084 PDBG("%s 0 credit ack ep %p state %u\n",
1085 __func__, ep, state_read(&ep->com));
b038ced7 1086 return CPL_RET_BUF_DONE;
f8b0dfd1
SW
1087 }
1088
b038ced7 1089 BUG_ON(credits != 1);
b038ced7 1090 dst_confirm(ep->dst);
f8b0dfd1
SW
1091 if (!ep->mpa_skb) {
1092 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1093 __func__, ep, state_read(&ep->com));
1094 if (ep->mpa_attr.initiator) {
1095 PDBG("%s initiator ep %p state %u\n",
1096 __func__, ep, state_read(&ep->com));
1097 if (peer2peer)
1098 iwch_post_zb_read(ep->com.qp);
1099 } else {
1100 PDBG("%s responder ep %p state %u\n",
1101 __func__, ep, state_read(&ep->com));
1102 ep->com.rpl_done = 1;
1103 wake_up(&ep->com.waitq);
1104 }
1105 } else {
1106 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1107 __func__, ep, state_read(&ep->com));
1108 kfree_skb(ep->mpa_skb);
1109 ep->mpa_skb = NULL;
b038ced7
SW
1110 }
1111 return CPL_RET_BUF_DONE;
1112}
1113
1114static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1115{
1116 struct iwch_ep *ep = ctx;
989a1780
SW
1117 unsigned long flags;
1118 int release = 0;
b038ced7 1119
33718363 1120 PDBG("%s ep %p\n", __func__, ep);
989a1780 1121 BUG_ON(!ep);
b038ced7 1122
aff9e39d
SW
1123 /*
1124 * We get 2 abort replies from the HW. The first one must
1125 * be ignored except for scribbling that we need one more.
1126 */
6e47fe43 1127 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
aff9e39d
SW
1128 return CPL_RET_BUF_DONE;
1129 }
1130
989a1780
SW
1131 spin_lock_irqsave(&ep->com.lock, flags);
1132 switch (ep->com.state) {
1133 case ABORTING:
1134 close_complete_upcall(ep);
1135 __state_set(&ep->com, DEAD);
1136 release = 1;
1137 break;
1138 default:
1139 printk(KERN_ERR "%s ep %p state %d\n",
1140 __func__, ep, ep->com.state);
1141 break;
1142 }
1143 spin_unlock_irqrestore(&ep->com.lock, flags);
1144
1145 if (release)
1146 release_ep_resources(ep);
b038ced7
SW
1147 return CPL_RET_BUF_DONE;
1148}
1149
96d0e493
SW
1150/*
1151 * Return whether a failed active open has allocated a TID
1152 */
1153static inline int act_open_has_tid(int status)
1154{
1155 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1156 status != CPL_ERR_ARP_MISS;
1157}
1158
b038ced7
SW
1159static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1160{
1161 struct iwch_ep *ep = ctx;
1162 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1163
33718363 1164 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
b038ced7
SW
1165 status2errno(rpl->status));
1166 connect_reply_upcall(ep, status2errno(rpl->status));
1167 state_set(&ep->com, DEAD);
8176d297 1168 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
b038ced7
SW
1169 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1170 cxgb3_free_atid(ep->com.tdev, ep->atid);
1171 dst_release(ep->dst);
1172 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1173 put_ep(&ep->com);
1174 return CPL_RET_BUF_DONE;
1175}
1176
1177static int listen_start(struct iwch_listen_ep *ep)
1178{
1179 struct sk_buff *skb;
1180 struct cpl_pass_open_req *req;
1181
33718363 1182 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1183 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1184 if (!skb) {
1185 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1186 return -ENOMEM;
1187 }
1188
1189 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1190 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1191 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1192 req->local_port = ep->com.local_addr.sin_port;
1193 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1194 req->peer_port = 0;
1195 req->peer_ip = 0;
1196 req->peer_netmask = 0;
1197 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1198 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1199 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1200
1201 skb->priority = 1;
04b5d028 1202 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
1203}
1204
1205static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1206{
1207 struct iwch_listen_ep *ep = ctx;
1208 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1209
33718363 1210 PDBG("%s ep %p status %d error %d\n", __func__, ep,
b038ced7
SW
1211 rpl->status, status2errno(rpl->status));
1212 ep->com.rpl_err = status2errno(rpl->status);
1213 ep->com.rpl_done = 1;
1214 wake_up(&ep->com.waitq);
1215
1216 return CPL_RET_BUF_DONE;
1217}
1218
1219static int listen_stop(struct iwch_listen_ep *ep)
1220{
1221 struct sk_buff *skb;
1222 struct cpl_close_listserv_req *req;
1223
33718363 1224 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1225 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1226 if (!skb) {
33718363 1227 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
b038ced7
SW
1228 return -ENOMEM;
1229 }
1230 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1231 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
60be4b59 1232 req->cpu_idx = 0;
b038ced7
SW
1233 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1234 skb->priority = 1;
04b5d028 1235 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
b038ced7
SW
1236}
1237
1238static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1239 void *ctx)
1240{
1241 struct iwch_listen_ep *ep = ctx;
1242 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1243
33718363 1244 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1245 ep->com.rpl_err = status2errno(rpl->status);
1246 ep->com.rpl_done = 1;
1247 wake_up(&ep->com.waitq);
1248 return CPL_RET_BUF_DONE;
1249}
1250
1251static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1252{
1253 struct cpl_pass_accept_rpl *rpl;
1254 unsigned int mtu_idx;
1255 u32 opt0h, opt0l, opt2;
1256 int wscale;
1257
33718363 1258 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1259 BUG_ON(skb_cloned(skb));
1260 skb_trim(skb, sizeof(*rpl));
1261 skb_get(skb);
1262 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1263 wscale = compute_wscale(rcv_win);
1264 opt0h = V_NAGLE(0) |
1265 V_NO_CONG(nocong) |
1266 V_KEEP_ALIVE(1) |
1267 F_TCAM_BYPASS |
1268 V_WND_SCALE(wscale) |
1269 V_MSS_IDX(mtu_idx) |
1270 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1271 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
bec658ff
SW
1272 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1273 V_CONG_CONTROL_FLAVOR(cong_flavor);
b038ced7
SW
1274
1275 rpl = cplhdr(skb);
1276 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1277 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1278 rpl->peer_ip = peer_ip;
1279 rpl->opt0h = htonl(opt0h);
1280 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1281 rpl->opt2 = htonl(opt2);
1282 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1283 skb->priority = CPL_PRIORITY_SETUP;
04b5d028 1284 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
b038ced7
SW
1285
1286 return;
1287}
1288
1289static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1290 struct sk_buff *skb)
1291{
33718363 1292 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
b038ced7
SW
1293 peer_ip);
1294 BUG_ON(skb_cloned(skb));
1295 skb_trim(skb, sizeof(struct cpl_tid_release));
1296 skb_get(skb);
1297
8176d297 1298 if (tdev->type != T3A)
b038ced7
SW
1299 release_tid(tdev, hwtid, skb);
1300 else {
1301 struct cpl_pass_accept_rpl *rpl;
1302
1303 rpl = cplhdr(skb);
1304 skb->priority = CPL_PRIORITY_SETUP;
1305 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1306 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1307 hwtid));
1308 rpl->peer_ip = peer_ip;
1309 rpl->opt0h = htonl(F_TCAM_BYPASS);
1310 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1311 rpl->opt2 = 0;
1312 rpl->rsvd = rpl->opt2;
04b5d028 1313 iwch_cxgb3_ofld_send(tdev, skb);
b038ced7
SW
1314 }
1315}
1316
1317static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1318{
1319 struct iwch_ep *child_ep, *parent_ep = ctx;
1320 struct cpl_pass_accept_req *req = cplhdr(skb);
1321 unsigned int hwtid = GET_TID(req);
1322 struct dst_entry *dst;
1323 struct l2t_entry *l2t;
1324 struct rtable *rt;
1325 struct iff_mac tim;
1326
33718363 1327 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
b038ced7
SW
1328
1329 if (state_read(&parent_ep->com) != LISTEN) {
1330 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
33718363 1331 __func__);
b038ced7
SW
1332 goto reject;
1333 }
1334
1335 /*
1336 * Find the netdev for this connection request.
1337 */
1338 tim.mac_addr = req->dst_mac;
1339 tim.vlan_tag = ntohs(req->vlan_tag);
1340 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
eacc4d6a
HS
1341 printk(KERN_ERR "%s bad dst mac %pM\n",
1342 __func__, req->dst_mac);
b038ced7
SW
1343 goto reject;
1344 }
1345
1346 /* Find output route */
1347 rt = find_route(tdev,
1348 req->local_ip,
1349 req->peer_ip,
1350 req->local_port,
1351 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1352 if (!rt) {
1353 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
33718363 1354 __func__);
b038ced7
SW
1355 goto reject;
1356 }
d8d1f30b 1357 dst = &rt->dst;
b038ced7
SW
1358 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1359 if (!l2t) {
1360 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
33718363 1361 __func__);
b038ced7
SW
1362 dst_release(dst);
1363 goto reject;
1364 }
1365 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1366 if (!child_ep) {
1367 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
33718363 1368 __func__);
b038ced7
SW
1369 l2t_release(L2DATA(tdev), l2t);
1370 dst_release(dst);
1371 goto reject;
1372 }
1373 state_set(&child_ep->com, CONNECTING);
1374 child_ep->com.tdev = tdev;
1375 child_ep->com.cm_id = NULL;
1376 child_ep->com.local_addr.sin_family = PF_INET;
1377 child_ep->com.local_addr.sin_port = req->local_port;
1378 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1379 child_ep->com.remote_addr.sin_family = PF_INET;
1380 child_ep->com.remote_addr.sin_port = req->peer_port;
1381 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1382 get_ep(&parent_ep->com);
1383 child_ep->parent_ep = parent_ep;
1384 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1385 child_ep->l2t = l2t;
1386 child_ep->dst = dst;
1387 child_ep->hwtid = hwtid;
1388 init_timer(&child_ep->timer);
1389 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1390 accept_cr(child_ep, req->peer_ip, skb);
1391 goto out;
1392reject:
1393 reject_cr(tdev, hwtid, req->peer_ip, skb);
1394out:
1395 return CPL_RET_BUF_DONE;
1396}
1397
1398static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1399{
1400 struct iwch_ep *ep = ctx;
1401 struct cpl_pass_establish *req = cplhdr(skb);
1402
33718363 1403 PDBG("%s ep %p\n", __func__, ep);
b038ced7 1404 ep->snd_seq = ntohl(req->snd_isn);
de3d3530 1405 ep->rcv_seq = ntohl(req->rcv_isn);
b038ced7
SW
1406
1407 set_emss(ep, ntohs(req->tcp_opt));
1408
1409 dst_confirm(ep->dst);
1410 state_set(&ep->com, MPA_REQ_WAIT);
1411 start_ep_timer(ep);
1412
1413 return CPL_RET_BUF_DONE;
1414}
1415
1416static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1417{
1418 struct iwch_ep *ep = ctx;
1419 struct iwch_qp_attributes attrs;
1420 unsigned long flags;
1421 int disconnect = 1;
1422 int release = 0;
1423
33718363 1424 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1425 dst_confirm(ep->dst);
1426
1427 spin_lock_irqsave(&ep->com.lock, flags);
1428 switch (ep->com.state) {
1429 case MPA_REQ_WAIT:
1430 __state_set(&ep->com, CLOSING);
1431 break;
1432 case MPA_REQ_SENT:
1433 __state_set(&ep->com, CLOSING);
1434 connect_reply_upcall(ep, -ECONNRESET);
1435 break;
1436 case MPA_REQ_RCVD:
1437
1438 /*
1439 * We're gonna mark this puppy DEAD, but keep
1440 * the reference on it until the ULP accepts or
a52bf98d
SW
1441 * rejects the CR. Also wake up anyone waiting
1442 * in rdma connection migration (see iwch_accept_cr()).
b038ced7
SW
1443 */
1444 __state_set(&ep->com, CLOSING);
a52bf98d
SW
1445 ep->com.rpl_done = 1;
1446 ep->com.rpl_err = -ECONNRESET;
1447 PDBG("waking up ep %p\n", ep);
1448 wake_up(&ep->com.waitq);
b038ced7
SW
1449 break;
1450 case MPA_REP_SENT:
1451 __state_set(&ep->com, CLOSING);
1452 ep->com.rpl_done = 1;
1453 ep->com.rpl_err = -ECONNRESET;
1454 PDBG("waking up ep %p\n", ep);
1455 wake_up(&ep->com.waitq);
1456 break;
1457 case FPDU_MODE:
42e31753 1458 start_ep_timer(ep);
b038ced7
SW
1459 __state_set(&ep->com, CLOSING);
1460 attrs.next_state = IWCH_QP_STATE_CLOSING;
1461 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1462 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1463 peer_close_upcall(ep);
1464 break;
1465 case ABORTING:
1466 disconnect = 0;
1467 break;
1468 case CLOSING:
b038ced7
SW
1469 __state_set(&ep->com, MORIBUND);
1470 disconnect = 0;
1471 break;
1472 case MORIBUND:
1473 stop_ep_timer(ep);
1474 if (ep->com.cm_id && ep->com.qp) {
1475 attrs.next_state = IWCH_QP_STATE_IDLE;
1476 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1477 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1478 }
1479 close_complete_upcall(ep);
1480 __state_set(&ep->com, DEAD);
1481 release = 1;
1482 disconnect = 0;
1483 break;
1484 case DEAD:
1485 disconnect = 0;
1486 break;
1487 default:
1488 BUG_ON(1);
1489 }
1490 spin_unlock_irqrestore(&ep->com.lock, flags);
1491 if (disconnect)
1492 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1493 if (release)
1494 release_ep_resources(ep);
1495 return CPL_RET_BUF_DONE;
1496}
1497
1498/*
1499 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1500 */
2b540355 1501static int is_neg_adv_abort(unsigned int status)
b038ced7
SW
1502{
1503 return status == CPL_ERR_RTX_NEG_ADVICE ||
1504 status == CPL_ERR_PERSIST_NEG_ADVICE;
1505}
1506
1507static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1508{
1509 struct cpl_abort_req_rss *req = cplhdr(skb);
1510 struct iwch_ep *ep = ctx;
1511 struct cpl_abort_rpl *rpl;
1512 struct sk_buff *rpl_skb;
1513 struct iwch_qp_attributes attrs;
1514 int ret;
989a1780
SW
1515 int release = 0;
1516 unsigned long flags;
b038ced7 1517
1580367e 1518 if (is_neg_adv_abort(req->status)) {
33718363 1519 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1580367e
SW
1520 ep->hwtid);
1521 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1522 return CPL_RET_BUF_DONE;
1523 }
1524
aff9e39d
SW
1525 /*
1526 * We get 2 peer aborts from the HW. The first one must
1527 * be ignored except for scribbling that we need one more.
1528 */
6e47fe43 1529 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
aff9e39d
SW
1530 return CPL_RET_BUF_DONE;
1531 }
1532
989a1780
SW
1533 spin_lock_irqsave(&ep->com.lock, flags);
1534 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1535 switch (ep->com.state) {
b038ced7
SW
1536 case CONNECTING:
1537 break;
1538 case MPA_REQ_WAIT:
adf376b3 1539 stop_ep_timer(ep);
b038ced7
SW
1540 break;
1541 case MPA_REQ_SENT:
adf376b3 1542 stop_ep_timer(ep);
b038ced7
SW
1543 connect_reply_upcall(ep, -ECONNRESET);
1544 break;
1545 case MPA_REP_SENT:
1546 ep->com.rpl_done = 1;
1547 ep->com.rpl_err = -ECONNRESET;
1548 PDBG("waking up ep %p\n", ep);
1549 wake_up(&ep->com.waitq);
1550 break;
1551 case MPA_REQ_RCVD:
1552
1553 /*
1554 * We're gonna mark this puppy DEAD, but keep
1555 * the reference on it until the ULP accepts or
a52bf98d
SW
1556 * rejects the CR. Also wake up anyone waiting
1557 * in rdma connection migration (see iwch_accept_cr()).
b038ced7 1558 */
a52bf98d
SW
1559 ep->com.rpl_done = 1;
1560 ep->com.rpl_err = -ECONNRESET;
1561 PDBG("waking up ep %p\n", ep);
1562 wake_up(&ep->com.waitq);
b038ced7
SW
1563 break;
1564 case MORIBUND:
42e31753 1565 case CLOSING:
b038ced7 1566 stop_ep_timer(ep);
42e31753 1567 /*FALLTHROUGH*/
b038ced7 1568 case FPDU_MODE:
b038ced7
SW
1569 if (ep->com.cm_id && ep->com.qp) {
1570 attrs.next_state = IWCH_QP_STATE_ERROR;
1571 ret = iwch_modify_qp(ep->com.qp->rhp,
1572 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1573 &attrs, 1);
1574 if (ret)
1575 printk(KERN_ERR MOD
1576 "%s - qp <- error failed!\n",
33718363 1577 __func__);
b038ced7
SW
1578 }
1579 peer_abort_upcall(ep);
1580 break;
1581 case ABORTING:
1582 break;
1583 case DEAD:
33718363 1584 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
989a1780 1585 spin_unlock_irqrestore(&ep->com.lock, flags);
b038ced7
SW
1586 return CPL_RET_BUF_DONE;
1587 default:
1588 BUG_ON(1);
1589 break;
1590 }
1591 dst_confirm(ep->dst);
989a1780
SW
1592 if (ep->com.state != ABORTING) {
1593 __state_set(&ep->com, DEAD);
1594 release = 1;
1595 }
1596 spin_unlock_irqrestore(&ep->com.lock, flags);
b038ced7
SW
1597
1598 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1599 if (!rpl_skb) {
1600 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
33718363 1601 __func__);
989a1780
SW
1602 release = 1;
1603 goto out;
b038ced7
SW
1604 }
1605 rpl_skb->priority = CPL_PRIORITY_DATA;
1606 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1607 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1608 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1609 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1610 rpl->cmd = CPL_ABORT_NO_RST;
04b5d028 1611 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
989a1780
SW
1612out:
1613 if (release)
b038ced7 1614 release_ep_resources(ep);
b038ced7
SW
1615 return CPL_RET_BUF_DONE;
1616}
1617
1618static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1619{
1620 struct iwch_ep *ep = ctx;
1621 struct iwch_qp_attributes attrs;
1622 unsigned long flags;
1623 int release = 0;
1624
33718363 1625 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1626 BUG_ON(!ep);
1627
1628 /* The cm_id may be null if we failed to connect */
1629 spin_lock_irqsave(&ep->com.lock, flags);
1630 switch (ep->com.state) {
1631 case CLOSING:
b038ced7
SW
1632 __state_set(&ep->com, MORIBUND);
1633 break;
1634 case MORIBUND:
1635 stop_ep_timer(ep);
1636 if ((ep->com.cm_id) && (ep->com.qp)) {
1637 attrs.next_state = IWCH_QP_STATE_IDLE;
1638 iwch_modify_qp(ep->com.qp->rhp,
1639 ep->com.qp,
1640 IWCH_QP_ATTR_NEXT_STATE,
1641 &attrs, 1);
1642 }
1643 close_complete_upcall(ep);
1644 __state_set(&ep->com, DEAD);
1645 release = 1;
1646 break;
42e31753 1647 case ABORTING:
b038ced7 1648 case DEAD:
c4d49776 1649 break;
b038ced7
SW
1650 default:
1651 BUG_ON(1);
1652 break;
1653 }
1654 spin_unlock_irqrestore(&ep->com.lock, flags);
1655 if (release)
1656 release_ep_resources(ep);
1657 return CPL_RET_BUF_DONE;
1658}
1659
1660/*
1661 * T3A does 3 things when a TERM is received:
1662 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1663 * 2) generate an async event on the QP with the TERMINATE opcode
1664 * 3) post a TERMINATE opcde cqe into the associated CQ.
1665 *
1666 * For (1), we save the message in the qp for later consumer consumption.
1667 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1668 * For (3), we toss the CQE in cxio_poll_cq().
1669 *
1670 * terminate() handles case (1)...
1671 */
1672static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1673{
1674 struct iwch_ep *ep = ctx;
1675
42fb61f0
SW
1676 if (state_read(&ep->com) != FPDU_MODE)
1677 return CPL_RET_BUF_DONE;
1678
33718363 1679 PDBG("%s ep %p\n", __func__, ep);
b038ced7 1680 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
33718363 1681 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
d626f62b
ACM
1682 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1683 skb->len);
b038ced7
SW
1684 ep->com.qp->attr.terminate_msg_len = skb->len;
1685 ep->com.qp->attr.is_terminate_local = 0;
1686 return CPL_RET_BUF_DONE;
1687}
1688
1689static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1690{
1691 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1692 struct iwch_ep *ep = ctx;
1693
33718363 1694 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
b038ced7
SW
1695 rep->status);
1696 if (rep->status) {
1697 struct iwch_qp_attributes attrs;
1698
1699 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
33718363 1700 __func__, ep->hwtid);
2f236735 1701 stop_ep_timer(ep);
b038ced7
SW
1702 attrs.next_state = IWCH_QP_STATE_ERROR;
1703 iwch_modify_qp(ep->com.qp->rhp,
1704 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1705 &attrs, 1);
1706 abort_connection(ep, NULL, GFP_KERNEL);
1707 }
1708 return CPL_RET_BUF_DONE;
1709}
1710
1711static void ep_timeout(unsigned long arg)
1712{
1713 struct iwch_ep *ep = (struct iwch_ep *)arg;
1714 struct iwch_qp_attributes attrs;
1715 unsigned long flags;
989a1780 1716 int abort = 1;
b038ced7
SW
1717
1718 spin_lock_irqsave(&ep->com.lock, flags);
33718363 1719 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
b038ced7
SW
1720 ep->com.state);
1721 switch (ep->com.state) {
1722 case MPA_REQ_SENT:
989a1780 1723 __state_set(&ep->com, ABORTING);
b038ced7
SW
1724 connect_reply_upcall(ep, -ETIMEDOUT);
1725 break;
1726 case MPA_REQ_WAIT:
989a1780 1727 __state_set(&ep->com, ABORTING);
b038ced7 1728 break;
42e31753 1729 case CLOSING:
b038ced7
SW
1730 case MORIBUND:
1731 if (ep->com.cm_id && ep->com.qp) {
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 }
989a1780 1737 __state_set(&ep->com, ABORTING);
b038ced7
SW
1738 break;
1739 default:
989a1780
SW
1740 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1741 __func__, ep, ep->com.state);
1742 WARN_ON(1);
1743 abort = 0;
b038ced7 1744 }
b038ced7 1745 spin_unlock_irqrestore(&ep->com.lock, flags);
989a1780
SW
1746 if (abort)
1747 abort_connection(ep, NULL, GFP_ATOMIC);
b038ced7
SW
1748 put_ep(&ep->com);
1749}
1750
1751int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1752{
1753 int err;
1754 struct iwch_ep *ep = to_ep(cm_id);
33718363 1755 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
b038ced7
SW
1756
1757 if (state_read(&ep->com) == DEAD) {
1758 put_ep(&ep->com);
1759 return -ECONNRESET;
1760 }
1761 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
b038ced7
SW
1762 if (mpa_rev == 0)
1763 abort_connection(ep, NULL, GFP_KERNEL);
1764 else {
1765 err = send_mpa_reject(ep, pdata, pdata_len);
7d526e6b 1766 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
b038ced7 1767 }
6e47fe43 1768 put_ep(&ep->com);
b038ced7
SW
1769 return 0;
1770}
1771
1772int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1773{
1774 int err;
1775 struct iwch_qp_attributes attrs;
1776 enum iwch_qp_attr_mask mask;
1777 struct iwch_ep *ep = to_ep(cm_id);
1778 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1779 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1780
33718363 1781 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
6e47fe43
SW
1782 if (state_read(&ep->com) == DEAD) {
1783 err = -ECONNRESET;
1784 goto err;
1785 }
b038ced7
SW
1786
1787 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1788 BUG_ON(!qp);
1789
1790 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1791 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1792 abort_connection(ep, NULL, GFP_KERNEL);
6e47fe43
SW
1793 err = -EINVAL;
1794 goto err;
b038ced7
SW
1795 }
1796
1797 cm_id->add_ref(cm_id);
1798 ep->com.cm_id = cm_id;
1799 ep->com.qp = qp;
1800
b038ced7
SW
1801 ep->ird = conn_param->ird;
1802 ep->ord = conn_param->ord;
96ac7e88
SW
1803
1804 if (peer2peer && ep->ird == 0)
1805 ep->ird = 1;
1806
33718363 1807 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
de3d3530 1808
b038ced7
SW
1809 /* bind QP to EP and move to RTS */
1810 attrs.mpa_attr = ep->mpa_attr;
1f71f503 1811 attrs.max_ird = ep->ird;
b038ced7
SW
1812 attrs.max_ord = ep->ord;
1813 attrs.llp_stream_handle = ep;
1814 attrs.next_state = IWCH_QP_STATE_RTS;
1815
1816 /* bind QP and TID with INIT_WR */
1817 mask = IWCH_QP_ATTR_NEXT_STATE |
1818 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1819 IWCH_QP_ATTR_MPA_ATTR |
1820 IWCH_QP_ATTR_MAX_IRD |
1821 IWCH_QP_ATTR_MAX_ORD;
1822
1823 err = iwch_modify_qp(ep->com.qp->rhp,
1824 ep->com.qp, mask, &attrs, 1);
de3d3530 1825 if (err)
6e47fe43 1826 goto err1;
b038ced7 1827
f8b0dfd1
SW
1828 /* if needed, wait for wr_ack */
1829 if (iwch_rqes_posted(qp)) {
1830 wait_event(ep->com.waitq, ep->com.rpl_done);
1831 err = ep->com.rpl_err;
1832 if (err)
6e47fe43 1833 goto err1;
f8b0dfd1
SW
1834 }
1835
de3d3530
SW
1836 err = send_mpa_reply(ep, conn_param->private_data,
1837 conn_param->private_data_len);
1838 if (err)
6e47fe43 1839 goto err1;
de3d3530 1840
de3d3530
SW
1841
1842 state_set(&ep->com, FPDU_MODE);
1843 established_upcall(ep);
1844 put_ep(&ep->com);
1845 return 0;
6e47fe43 1846err1:
de3d3530
SW
1847 ep->com.cm_id = NULL;
1848 ep->com.qp = NULL;
1849 cm_id->rem_ref(cm_id);
6e47fe43 1850err:
b038ced7
SW
1851 put_ep(&ep->com);
1852 return err;
1853}
1854
8704e9a8
SW
1855static int is_loopback_dst(struct iw_cm_id *cm_id)
1856{
1857 struct net_device *dev;
1858
1859 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1860 if (!dev)
1861 return 0;
1862 dev_put(dev);
1863 return 1;
1864}
1865
b038ced7
SW
1866int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1867{
1868 int err = 0;
1869 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1870 struct iwch_ep *ep;
1871 struct rtable *rt;
1872
8704e9a8
SW
1873 if (is_loopback_dst(cm_id)) {
1874 err = -ENOSYS;
1875 goto out;
1876 }
1877
b038ced7
SW
1878 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1879 if (!ep) {
33718363 1880 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
b038ced7
SW
1881 err = -ENOMEM;
1882 goto out;
1883 }
1884 init_timer(&ep->timer);
1885 ep->plen = conn_param->private_data_len;
1886 if (ep->plen)
1887 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1888 conn_param->private_data, ep->plen);
1889 ep->ird = conn_param->ird;
1890 ep->ord = conn_param->ord;
96ac7e88
SW
1891
1892 if (peer2peer && ep->ord == 0)
1893 ep->ord = 1;
1894
b038ced7
SW
1895 ep->com.tdev = h->rdev.t3cdev_p;
1896
1897 cm_id->add_ref(cm_id);
1898 ep->com.cm_id = cm_id;
1899 ep->com.qp = get_qhp(h, conn_param->qpn);
1900 BUG_ON(!ep->com.qp);
33718363 1901 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
b038ced7
SW
1902 ep->com.qp, cm_id);
1903
1904 /*
1905 * Allocate an active TID to initiate a TCP connection.
1906 */
1907 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1908 if (ep->atid == -1) {
33718363 1909 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
b038ced7
SW
1910 err = -ENOMEM;
1911 goto fail2;
1912 }
1913
1914 /* find a route */
1915 rt = find_route(h->rdev.t3cdev_p,
1916 cm_id->local_addr.sin_addr.s_addr,
1917 cm_id->remote_addr.sin_addr.s_addr,
1918 cm_id->local_addr.sin_port,
1919 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1920 if (!rt) {
33718363 1921 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
b038ced7
SW
1922 err = -EHOSTUNREACH;
1923 goto fail3;
1924 }
d8d1f30b 1925 ep->dst = &rt->dst;
b038ced7
SW
1926
1927 /* get a l2t entry */
1928 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1929 ep->dst->neighbour->dev);
1930 if (!ep->l2t) {
33718363 1931 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
b038ced7
SW
1932 err = -ENOMEM;
1933 goto fail4;
1934 }
1935
1936 state_set(&ep->com, CONNECTING);
1937 ep->tos = IPTOS_LOWDELAY;
1938 ep->com.local_addr = cm_id->local_addr;
1939 ep->com.remote_addr = cm_id->remote_addr;
1940
1941 /* send connect request to rnic */
1942 err = send_connect(ep);
1943 if (!err)
1944 goto out;
1945
1946 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1947fail4:
1948 dst_release(ep->dst);
1949fail3:
1950 cxgb3_free_atid(ep->com.tdev, ep->atid);
1951fail2:
dc35fac9 1952 cm_id->rem_ref(cm_id);
b038ced7
SW
1953 put_ep(&ep->com);
1954out:
1955 return err;
1956}
1957
1958int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1959{
1960 int err = 0;
1961 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1962 struct iwch_listen_ep *ep;
1963
1964
1965 might_sleep();
1966
1967 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1968 if (!ep) {
33718363 1969 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
b038ced7
SW
1970 err = -ENOMEM;
1971 goto fail1;
1972 }
33718363 1973 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
1974 ep->com.tdev = h->rdev.t3cdev_p;
1975 cm_id->add_ref(cm_id);
1976 ep->com.cm_id = cm_id;
1977 ep->backlog = backlog;
1978 ep->com.local_addr = cm_id->local_addr;
1979
1980 /*
1981 * Allocate a server TID.
1982 */
1983 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1984 if (ep->stid == -1) {
33718363 1985 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
b038ced7
SW
1986 err = -ENOMEM;
1987 goto fail2;
1988 }
1989
1990 state_set(&ep->com, LISTEN);
1991 err = listen_start(ep);
1992 if (err)
1993 goto fail3;
1994
1995 /* wait for pass_open_rpl */
1996 wait_event(ep->com.waitq, ep->com.rpl_done);
1997 err = ep->com.rpl_err;
1998 if (!err) {
1999 cm_id->provider_data = ep;
2000 goto out;
2001 }
2002fail3:
2003 cxgb3_free_stid(ep->com.tdev, ep->stid);
2004fail2:
1b07db70 2005 cm_id->rem_ref(cm_id);
b038ced7
SW
2006 put_ep(&ep->com);
2007fail1:
2008out:
2009 return err;
2010}
2011
2012int iwch_destroy_listen(struct iw_cm_id *cm_id)
2013{
2014 int err;
2015 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2016
33718363 2017 PDBG("%s ep %p\n", __func__, ep);
b038ced7
SW
2018
2019 might_sleep();
2020 state_set(&ep->com, DEAD);
2021 ep->com.rpl_done = 0;
2022 ep->com.rpl_err = 0;
2023 err = listen_stop(ep);
04b5d028
SW
2024 if (err)
2025 goto done;
b038ced7
SW
2026 wait_event(ep->com.waitq, ep->com.rpl_done);
2027 cxgb3_free_stid(ep->com.tdev, ep->stid);
04b5d028 2028done:
b038ced7
SW
2029 err = ep->com.rpl_err;
2030 cm_id->rem_ref(cm_id);
2031 put_ep(&ep->com);
2032 return err;
2033}
2034
2035int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2036{
2037 int ret=0;
2038 unsigned long flags;
2039 int close = 0;
04b5d028
SW
2040 int fatal = 0;
2041 struct t3cdev *tdev;
2042 struct cxio_rdev *rdev;
b038ced7
SW
2043
2044 spin_lock_irqsave(&ep->com.lock, flags);
2045
33718363 2046 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
b038ced7
SW
2047 states[ep->com.state], abrupt);
2048
04b5d028
SW
2049 tdev = (struct t3cdev *)ep->com.tdev;
2050 rdev = (struct cxio_rdev *)tdev->ulp;
2051 if (cxio_fatal_error(rdev)) {
2052 fatal = 1;
2053 close_complete_upcall(ep);
2054 ep->com.state = DEAD;
2055 }
b038ced7
SW
2056 switch (ep->com.state) {
2057 case MPA_REQ_WAIT:
2058 case MPA_REQ_SENT:
2059 case MPA_REQ_RCVD:
2060 case MPA_REP_SENT:
2061 case FPDU_MODE:
b038ced7 2062 close = 1;
989a1780
SW
2063 if (abrupt)
2064 ep->com.state = ABORTING;
2065 else {
2066 ep->com.state = CLOSING;
2067 start_ep_timer(ep);
2068 }
6e47fe43 2069 set_bit(CLOSE_SENT, &ep->com.flags);
b038ced7
SW
2070 break;
2071 case CLOSING:
6e47fe43
SW
2072 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2073 close = 1;
2074 if (abrupt) {
2075 stop_ep_timer(ep);
2076 ep->com.state = ABORTING;
2077 } else
2078 ep->com.state = MORIBUND;
2079 }
b038ced7
SW
2080 break;
2081 case MORIBUND:
989a1780
SW
2082 case ABORTING:
2083 case DEAD:
2084 PDBG("%s ignoring disconnect ep %p state %u\n",
2085 __func__, ep, ep->com.state);
b038ced7
SW
2086 break;
2087 default:
2088 BUG();
2089 break;
2090 }
989a1780 2091
b038ced7
SW
2092 spin_unlock_irqrestore(&ep->com.lock, flags);
2093 if (close) {
2094 if (abrupt)
2095 ret = send_abort(ep, NULL, gfp);
2096 else
2097 ret = send_halfclose(ep, gfp);
04b5d028
SW
2098 if (ret)
2099 fatal = 1;
b038ced7 2100 }
04b5d028
SW
2101 if (fatal)
2102 release_ep_resources(ep);
b038ced7
SW
2103 return ret;
2104}
2105
2106int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2107 struct l2t_entry *l2t)
2108{
2109 struct iwch_ep *ep = ctx;
2110
2111 if (ep->dst != old)
2112 return 0;
2113
33718363 2114 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
b038ced7
SW
2115 l2t);
2116 dst_hold(new);
2117 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2118 ep->l2t = l2t;
2119 dst_release(old);
2120 ep->dst = new;
2121 return 1;
2122}
2123
2124/*
2125 * All the CM events are handled on a work queue to have a safe context.
617c9a7e 2126 * These are the real handlers that are called from the work queue.
b038ced7 2127 */
617c9a7e
RD
2128static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2129 [CPL_ACT_ESTABLISH] = act_establish,
2130 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2131 [CPL_RX_DATA] = rx_data,
2132 [CPL_TX_DMA_ACK] = tx_ack,
2133 [CPL_ABORT_RPL_RSS] = abort_rpl,
2134 [CPL_ABORT_RPL] = abort_rpl,
2135 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2136 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2137 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2138 [CPL_PASS_ESTABLISH] = pass_establish,
2139 [CPL_PEER_CLOSE] = peer_close,
2140 [CPL_ABORT_REQ_RSS] = peer_abort,
2141 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2142 [CPL_RDMA_TERMINATE] = terminate,
2143 [CPL_RDMA_EC_STATUS] = ec_status,
2144};
2145
2146static void process_work(struct work_struct *work)
2147{
2148 struct sk_buff *skb = NULL;
2149 void *ep;
2150 struct t3cdev *tdev;
2151 int ret;
2152
2153 while ((skb = skb_dequeue(&rxq))) {
2154 ep = *((void **) (skb->cb));
2155 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2156 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2157 if (ret & CPL_RET_BUF_DONE)
2158 kfree_skb(skb);
2159
2160 /*
2161 * ep was referenced in sched(), and is freed here.
2162 */
2163 put_ep((struct iwch_ep_common *)ep);
2164 }
2165}
2166
2167static DECLARE_WORK(skb_work, process_work);
2168
b038ced7
SW
2169static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2170{
2171 struct iwch_ep_common *epc = ctx;
2172
2173 get_ep(epc);
2174
2175 /*
2176 * Save ctx and tdev in the skb->cb area.
2177 */
2178 *((void **) skb->cb) = ctx;
2179 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2180
2181 /*
2182 * Queue the skb and schedule the worker thread.
2183 */
2184 skb_queue_tail(&rxq, skb);
2185 queue_work(workq, &skb_work);
2186 return 0;
2187}
2188
1ca19770
SW
2189static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2190{
2191 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2192
2193 if (rpl->status != CPL_ERR_NONE) {
2194 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2195 "for tid %u\n", rpl->status, GET_TID(rpl));
2196 }
2197 return CPL_RET_BUF_DONE;
2198}
2199
617c9a7e
RD
2200/*
2201 * All upcalls from the T3 Core go to sched() to schedule the
2202 * processing on a work queue.
2203 */
2204cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2205 [CPL_ACT_ESTABLISH] = sched,
2206 [CPL_ACT_OPEN_RPL] = sched,
2207 [CPL_RX_DATA] = sched,
2208 [CPL_TX_DMA_ACK] = sched,
2209 [CPL_ABORT_RPL_RSS] = sched,
2210 [CPL_ABORT_RPL] = sched,
2211 [CPL_PASS_OPEN_RPL] = sched,
2212 [CPL_CLOSE_LISTSRV_RPL] = sched,
2213 [CPL_PASS_ACCEPT_REQ] = sched,
2214 [CPL_PASS_ESTABLISH] = sched,
2215 [CPL_PEER_CLOSE] = sched,
2216 [CPL_CLOSE_CON_RPL] = sched,
2217 [CPL_ABORT_REQ_RSS] = sched,
2218 [CPL_RDMA_TERMINATE] = sched,
2219 [CPL_RDMA_EC_STATUS] = sched,
2220 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2221};
2222
b038ced7
SW
2223int __init iwch_cm_init(void)
2224{
2225 skb_queue_head_init(&rxq);
2226
2227 workq = create_singlethread_workqueue("iw_cxgb3");
2228 if (!workq)
2229 return -ENOMEM;
2230
b038ced7
SW
2231 return 0;
2232}
2233
2234void __exit iwch_cm_term(void)
2235{
2236 flush_workqueue(workq);
2237 destroy_workqueue(workq);
2238}